# Theory of Reality

## A Projective Seven-Layer Framework for Hidden State, Time, Quantum Observation, Computation, and Testable Physics

**Version:** 1.70 (draft)
**Date:** 20 September 2026
**Originating framework and research direction:** Anthony Michael Collado  
**Computational formalization:** Reality Bit Engine 0.20.0; Rope Tomography Lab 0.14.0; RRLS Studio 0.27 audited extension 2; Hydrogen Matrix Lab 0.8.0; Element Projection Lab 0.7.0; Fixed Grid String Lab 0.1.0; Matter Harmony Lab 1.5.0; Born Outcome Frontier Audit 0.2.0; Wallbreaker Audit 0.1.0; P3-R1 Hardware Runner 1.0.0; Dimensional Chess C1 1.0.0; Dimensional Chess C2 1.0.0; Dimensional Chess C3 1.0.0; Dimensional Chess C4 1.0.0; Dimensional Chess C5 1.0.0; Dimensional Chess C6 1.0.0; Dimensional Chess C7 1.0.0; Dimensional Chess C8 1.0.0
**Status:** Revisable mathematical framework and executable research program; not an experimentally confirmed seven-dimensional theory of nature

**Revision 1.70 addition (draft).** Appendix BS adds Inversion Theory, a single-hypothesis detector for one spatial dimension above the third. Theorem BS.T1 proves that no rigid path inside \(\mathbb R^n\) carries a chiral object to its mirror image, because \(E(n)\) has two components; Theorem BS.T2 proves one rotation in \(\mathbb R^{n+1}\) does, with every pairwise distance preserved (verified to \(4\times10^{-16}\) on a labelled tetrahedron); Theorem BS.T3 turns that into a four-part detection certificate (identity, rigidity, invariant flip, locality); Theorem BS.T1′ records the one loophole, transport around an orientation-reversing loop of a non-orientable space (verified: Möbius-band holonomy \(-1\), cylinder \(+1\)), closed by the locality condition and by Planck's bound of about 27 Gpc on any such loop. Theorems BS.T4 to BS.T7 give the visible-slice, conservation-leak, knot, and compact-size-obstruction consequences. The hypothesis H-Inv predicts a temperature- and isotope-independent racemization floor; the powered experiment is specified in BS.6. BS.A defines the black sheep of a world \((X,\pi,G)\) as a state visible by a \(\pi\)-expressible \(G\)-invariant yet unreachable by \(G\), places it in the one cell of the see/do table the manuscript had not analyzed, proves it is a forbidden value of a conserved quantity (Theorem BS.A.2), and identifies its shape as the second sheet of the orientation double cover (Theorem BS.A.5). BS.B catalogs the black sheep of mathematics with every number recomputed. BS.9 implements the certificate as `dimensional_verifier.py` and runs it over seventeen documented sightings: zero detections, six deformation routes, five laws or conventions, three developmental origins, three null bounds. Every observed inversion in three dimensions passed through an achiral configuration; no rigid body has ever been seen to invert. This is written mathematics, finite computation, and cited measured limits, not evidence that a fourth spatial dimension exists.

**Revision 1.69 addition (draft).** Appendix BR answers the request to prove the framework from Scripture by doing what the framework permits: an exact audit, not a proof. Theorem BR.1 states that a finite text is a finite observer and therefore cannot identify a unique hidden geometry; this protects Scripture from over-reading and from dismissal equally. Job 38 is read as an observer-limit text. Genesis 1 admits 5,040 day-to-layer bijections and selects none. All stated sanctuary dimensions yield the ratio set \{1, 4/3, 3/2, 5/3, 2, 3, 6, 20/3, 7, 10\}, with zero hits against nine irrational constants at 1 % where chance predicts 1.44; the Holy of Holies is a cube in three independent texts. The gematria-corrected Bronze Sea ratio 333/106 is the third continued-fraction convergent of \(\pi\) and is recorded as coincidence class under Section 16. BR.10 tabulates the sixteen-operation grammar of Genesis 1, shows that the heavens are stretched with the surveyor's verb *natah* of Job 38:5, that celestial regularity is called *choq*, statute, and that creation and Tabernacle share one construction vocabulary. BR.11 writes that grammar in Section 2's objects and proves that Genesis 1 is a two-pass fiber construction (Proposition BR.6), that replication "after its kind" is the congruence condition of Theorem 4.1 with identity as the descended law (Theorem BR.7), and that a *choq* is a trajectory constraint whose descent is Theorem 4.2's question (Theorem BR.8). BR.12 audits gematria, the Genesis 1:1 constants 28 and 2701 = 37 × 73 = \(T_{73}\), the "π in Genesis 1:1" formula (exact to 1.2 × 10⁻⁵, a 1-in-190 event in a declared 31-formula space, coincidence class), Aleph–Tav, the Name as a verb of being, Hebrew/genome number matching (11 small-integer equalities in 195 comparisons, no bijection), and equidistant letter sequences (refuted). Every BR result is typed T or M-on-text; none is E and none reaches H7. This revision also applies an external review of v1.65: Theorem 6.2 gains its missing continuity hypothesis; the Appendix AL.5 retention principle is corrected from a necessary to a sufficient condition; the Appendix AS.4 Ag/Pd comparison is annotated for mixed reference temperatures; the Hensen et al. DOI is corrected; the abstract's "axiom" becomes "hypothesis H7"; the Section 20 record is labeled with its release; the AU.1 theorem count is fixed; every heading of the Reproducibility companion and Appendices A--Q carries an M/C/E/H/T label (75 headings; 22 M, 17 C, 24 M/C, 6 C conditional on an unpublished certificate or run, 2 H, 1 M/H, 1 H/T, 0 E); Appendix AT gains Table AT.1a, the previously unprinted 17-row alkane ledger with exact Wiener and Zagreb indices and an independent refit reproducing every AT.2--AT.3 statistic within 0.7 K; and the filtration boundary of Appendices S.2 and T.1 is restated (Ville's bound on the ever-rejection event holds under any look schedule; the reduced filtration forbids reporting \(G_\tau\) as an e-value at raw-data stopping times and merging with canonical-filtration processes), with T.2, V.1 and V.4 now stating that the joint rule's null is the known-covariance null. Structurally, Appendices BO and BP, which revisions 1.66--1.67 inserted before BK and in the order BP, BO, are moved after BN in the order BO, BP, BQ. Appendix BQ gains BQ.5, an independent replay of the CW/LD benchmark tower \(k=500\) GeV, \(M_5=6\) TeV: 102 spin-2 modes below 5 TeV with first mode 502.35 GeV and maximum width-to-spacing ratio about 1.7 %, agreeing with the Truth Tower run at convention level. The HEPData record `ins2660845` was subsequently retrieved by the author and inventoried in BQ.6: every table is a 95 % exclusion contour in the \(k\)--\(M_5\) plane, six per figure panel; the record contains no binned spectra, background expectations, response tables, or significance values, so components 9 and 10 of the BQ.3 contract cannot be sourced from it and the recast block now rests on the primary source rather than on access. The arXiv page was egress-blocked here, so the eight BP.1 significances remain paper-printed inputs unverified against the primary text in this environment.

**Revision 1.68 addition.** Appendix BQ and Element Projection Lab 0.7.0 execute the recast-readiness test demanded by revision 1.67. The exact ATLAS forward equation shows that a truth-level resonance tower does not determine reconstructed bin counts without per-mode production, branching fractions, acceptance/efficiency, and detector transfer. A frozen 13-component contract finds only three components ready and blocks the detector-level replay rather than imputing the other ten. The exact counterexample keeps truth weights and efficiencies fixed while two admissible column-stochastic detector responses change the reconstructed spectrum from `[8,3,0]` to `[4.75,4.75,1.5]`, with the same total count of 11. Nine new production tests and an independent 15-check scalar replay pass. The official HEPData endpoint could not be materialized through the available web client because of an access challenge; that is explicitly not reclassified as non-publication. This is written mathematics and finite synthetic computation, not a new measurement, successful ATLAS recast, positive evidence for another dimension, universal exclusion, or theological identification.

**Revision 1.67 addition.** Appendix BP and Element Projection Lab 0.6.0 execute a frozen retrospective evidence audit on the exact signed significances printed in ATLAS's 139 fb^-1 periodic dielectron/diphoton search. At the preselected periodicity-focused `R<50%` row, the dielectron value is +1.5 standard deviations and the diphoton value is -0.2. Across all four reported threshold variants, one-sided normal-tail probabilities with an arbitrary-dependence Bonferroni correction are 0.267229 and 1.0 for the most positive dielectron and diphoton values. The paper's model-dependent search also reports no significant deviation. The necessary cross-channel evidence gate therefore fails. Eight new production tests and a twelve-check independent standard-library replay pass. The equally important correction is procedural: because the publication and its null conclusion were already public, this is not a blind preregistration or a reproduction of the ATLAS likelihood. It is measured-input retrospective auditing plus finite computation, not a new measurement, discovery of another dimension, universal exclusion, or theological identification. A separate latest-source check found that the RRLS package no longer matched its own documented integer-validation and independent-verifier contract. Audited extension 2 restores strict integer inputs, records the collision search bound, and reconstructs collision certificates without calling the production solver; all 20 regression tests, five legacy checks, 16,250 exhaustive integer cases, and 600 generated rational cases pass. That is a software correction, not physical evidence.

**Revision 1.66 addition.** Appendix BO and Element Projection Lab 0.5.0 impose a physical-dimension evidence gate and close one invalid route to such evidence. For the declared real five-coordinate skew-linear motion, every continuous linear three-vector observation is exactly a constant plus at most two ordinary sine/cosine pairs; two conventional temporal oscillators therefore reproduce the alleged 5D projection for all times. In 512 random trials the two realizations agree over 512,512 vector samples to maximum component error 5.33e-15; an independent scalar implementation checks 299,919 components with maximum error 3.55e-15. A simple flat-circle Kaluza--Klein relation is retained only as a model-specific example: two labelled masses merely calibrate its two parameters, while later masses become predictions. Current ATLAS periodic-spectrum, CMS high-mass diphoton, and Eot-Wash short-range-gravity searches report null results and exclusions rather than positive evidence. Thus the present animations and optical oscillators are representations, not evidence for another spatial dimension. The next admissible experiment is a preregistered public-data search for a correlated tower, including frozen masses, widths, spin/coupling/channel relations, background, multiplicity correction, and reserved validation. This is written mathematics and finite synthetic computation plus cited external measured limits--not a new measurement, dimensional discovery, or theological identification.

**Revision 1.65 addition.** Appendix BN and Element Projection Lab 0.4.0 freeze the revision-1.64 Au/Ag response comparison under identity `EPL-AU-AG-COLOR-2026-09-20-v1`. The manifest fixes the four coefficient vectors as hexadecimal IEEE-754 values, source table and executable hashes, train/holdout split, equations, wavelength domain, scoring rules, boundary warnings, headline results, and prohibited changes. The evaluation module contains no optimizer and verifies every locked artifact before scoring; future refitting, bound changes, and model selection are forbidden under this identity. Eight new tests pass, and a separate implementation verifies the manifest plus four artifact hashes and replays all 64 predictions with maximum discrepancy 4.44e-16. This is an anti-leakage and reproducibility result, not new measured evidence or an upgrade from phenomenological response to unique band, relativistic, higher-dimensional, or theological mechanism.

**Revision 1.64 addition.** Appendix BM and Element Projection Lab 0.3.0 execute the first frozen real-data model-class test behind the gold/silver color observer. For each metal, eleven Johnson--Christy wavelengths fit either a three-parameter damped free-electron response or a six-parameter response with one additional interband oscillator; five fixed wavelengths remain untouched until scoring. Gold holdout reflectance MAE falls from 0.154691 to 0.019718, an 87.253% reduction. Silver falls from 0.006840 to 0.002727, a smaller 0.004112 absolute gain because its visible response is already nearly free-electron-like. Nine new tests pass, and an independent scalar implementation replays 64 predictions and 218 checks with maximum aggregate-metric discrepancy 2.16e-16. The correction is equally important: an oscillator ablation is not a relativistic/nonrelativistic comparison, the richer model has three extra fitted parameters, and two fits hit declared parameter bounds. This supports a wavelength-dependent interband channel within the tested classes, not a unique band Hamiltonian, physical extra dimension, or theological cause.

**Revision 1.63 addition.** Appendix BL and Element Projection Lab 0.2.0 add the missing condensed-matter optical observer for bulk gold and silver without assigning color to one isolated atom. Johnson--Christy's room-temperature measured complex refractive indices are passed through the exact normal-incidence Fresnel map and a declared finite CIE observer. At 450.9 nm the calculated reflectances are 0.408220 for Au and 0.980347 for Ag; at 659.5 nm they are 0.962585 and 0.990566. Gold's mean red-minus-blue reflectance is 0.537373, versus 0.013155 for silver. All 19 package tests pass, and an implementation importing no production optics code replays 802 dense reflectance values with maximum discrepancy 5.55e-16. The decisive correction is that the all-element atom atlas cannot predict bulk color: lattice-periodic bands, occupations, optical transition matrix elements, damping, and relativistic terms intervene. This is a measured-input forward calculation, not an ab-initio color prediction, measured work by this project, proof of a fourth/fifth dimension, or theological identification.

**Revision 1.62 addition.** Appendix BK and Element Projection Lab 0.1.0 compile all 118 neutral elements into one typed comparison atlas. The exact correction is that an isolated atom has no unique rigid classical surface: the released whole-atom channel is an explicitly rotationally averaged one-electron density and is therefore spherical, while the separate oriented channel displays a declared canonical differentiating-subshell basis. Closed-subshell sphericity follows exactly from the spherical-harmonic addition theorem; oriented \(s,p,d,f\) basis densities have allowed even angular ranks through \(2l\). Reference configurations, shell populations, radial-node counts, and periodic-table positions are separated from a Slater-screened hydrogenic radial proxy; configurations for atomic numbers 109--118 are marked predicted. All 11 production tests and 13 independent checks pass over every element, including electron counts, capacities, grid uniqueness, hydrogenic normalization, and the addition theorem. These are written mathematics, compiled reference inputs, and finite proxy computation--not measured 3D densities for every atom, many-electron ab initio solutions, rigid electron paths, fourth/fifth-dimensional causes, or a theological identification.

**Revision 1.61 addition.** Appendix BJ and Hydrogen Matrix Lab 0.8.0 replace the undefined goal “display everything” with a typed forward-observer contract. One declared ideal coherent `1s + 2p_z` state now generates synchronized position density, conserved probability flow, electric charge/current, expectation dipole, far-zone coherent electromagnetic radiation, Poynting flux, and an operator-filtered harmonic catalog. Eight frozen checks pass, including charge continuity, transversality, `|B|=|E|/c`, angular power integration, and the parity-silent `1s + 2s` control; a separate implementation passes 12 checks. The decisive exact boundary is not a fit: cycle-averaged classical expectation-dipole power is exactly one half of the initial quantum E1 energy-loss rate for the equal packet's half-excited population. Therefore a classical projected wave is not the complete spontaneous-photon channel. The compiler marks quantized-field/vacuum statistics and spin magnetization as blocked and higher-dimensional cause as non-identified. No new measurement, complete QED engine, physical fourth/fifth dimension, or theological cause is claimed; all earlier rejections and fixed-grid boundaries remain unchanged.

**Revision 1.60 addition.** Appendix BI and Hydrogen Matrix Lab 0.7.0 audit golden-ratio, golden-angle, Fibonacci, icosahedral, and proposed “golden alignment” claims with exact algebra before accepting numerical matches. The exact positive result is structural: the Fibonacci companion matrix has dominant eigenvalue `phi`, declared icosahedral coordinates contain `phi`, and the `E8` identity is `2 cos(pi/5)=phi`. The exact negative result is equally important: every ideal Coulomb bound-state energy gap is rational in Hartree, so no ratio of two nonzero gaps can equal irrational `phi`. An exhaustive frozen `n<=40` scan checks 303,809 distinct pairs; the closest ratio differs from `phi` by `9.76537e-6` and remains a coincidence. Likewise, a finite angular grid step `k/N` cannot equal the irrational golden-turn fraction `1/phi^2`; Fibonacci grids are periodic approximations only. All 51 package tests and nine independent checks pass. External `E8`, quasicrystal, and phyllotaxis measurements are retained as evidence for their declared systems, not transferred to hydrogen, hidden dimensions, or theology.

**Revision 1.59 addition.** Appendix BH answers the bounded question of which quantum equation was missing from the hydrogen visualization. Schrödinger evolution was already present; the omitted bridge was the conserved probability-current field, electromagnetic minimal coupling, and the observable matrix element that filters raw energy-gap harmonics. Hydrogen Matrix Lab 0.6.0 implements the field-free current for the existing `1s + 2p_z` packet, proves `partial_t rho + div(j)=0`, recovers the exact dipole matrix element `256/(243 sqrt(2)) a0`, and computes the ideal E1 lifetime `1.59532505 ns`. In an equal `1s + 2p_z + 3s` packet all three density beats exist but only two are visible to the `z` electric dipole; the `1s <-> 3s` harmonic vanishes by parity, as does the one-photon `1s <-> 2s` control. All 42 Hydrogen Matrix tests pass, and an independent quadrature/fourth-order finite-difference audit passes seven frozen checks while retaining the larger full-interior Coulomb-cusp residual. These are written standard quantum mechanics and finite computation, not a measured atom, a particle trajectory, an extra dimension, or a theological identification.

**Revision 1.58 addition.** Appendix BG turns one member of revision 1.57's non-identifiable interaction family into a frozen prediction. The range is set at the electron reduced Compton wavelength and the coupling is fitted only to the published H--D 1S--2S central residual. Without refitting it predicts a -201.996425 Hz correction to the historical H--D 2S--8S isotope shift. A published bound-state reference with experimental and theory correlations gives an ordinary signed error of -0.700628 combined standard deviations; adding the operator changes it to -0.720112. The operator therefore earns no predictive credit, but the combined uncertainty is 51.326 times the predicted contrast, so this target cannot exclude it either. This is a historical public-data holdout, not a blind prediction. All 164 package tests and an independent direct-radial-quadrature audit pass. No physical dimension, universal grid/tick, string particle, or theological cause is inferred.

**Revision 1.57 addition.** Appendix BF replaces the unspecified extra-dimensional hydrogen correction with one explicit finite-range central proxy, \(\delta V/h=g e^{-r/\lambda}/r\), but first evaluates the published H--D 1S--2S correction ladder. A 2022 QED/nuclear-structure expression supplied with the muonic-deuterium radius \(2.12763(78)\) fm predicts \(670994336.042632\) kHz versus the direct measured \(670994334.606(15)\) kHz: a +1.436632 kHz error, only +0.309753 combined standard deviations. The ordinary ladder reduces the crude reduced-mass-only error by a factor of 8837.3. The operator is therefore not required. Moreover, all 61 ranges from 1 fm through 10 nm can be paired with a coupling that fits the same central residual exactly, while the endpoint pairs disagree by 121.784 Hz on a prospective 1S--3S isotope correction. This is an exact one-equation/two-parameter non-identifiability result, not evidence for a fifth dimension. All 154 package tests and a separate adaptive-quadrature audit pass. The published values are measured/theory inputs; no new spectroscopy or theological identification is claimed.

**Revision 1.56 addition.** Appendix BE replaces the hydrogen feature-count exercise with a frozen transition holdout. NIST H-alpha and H-beta alone calibrate a one-parameter Rydberg-form baseline and a two-parameter phenomenological `Delta2 + Delta4` candidate after a declared standard-air-to-vacuum conversion. On untouched H-gamma and H-delta, the added term increases vacuum-wavelength RMSE from 6.030008 to 6.948567 ppm, a 15.2331% deterioration, and is rejected as an improvement. A written counterexample adds an arbitrary term that vanishes on both training lines but changes every later prediction; one coefficient forces H-gamma exactly and still misses H-delta by 7.1389 ppm. Thus perfect calibration does not identify a continuation, a quantum generator, or a physical dimension. All 143 package tests and a separate 50-digit Decimal replay pass. The NIST values are compiled measured references; the conversion and scores are finite computation; no new spectroscopy, complete QED calculation, or higher-dimensional operator is claimed.

**Revision 1.55 addition.** Appendix BD generalizes the fixed skew-linear harmony engine to arbitrary latent dimension and exposes explicit latent position vectors, tangent vectors, and their calibrated 3D projections. A single three-output instantaneous observer has nullity at least (d-3), while a simultaneous known observer bank identifies the common state exactly iff its vertically stacked matrix has rank (d). The frozen banks reconstruct synthetic states in dimensions 4, 5, 6, and 8 to numerical precision, while an exact 5D counterexample moves the latent state by one unit without changing one observer's output. More observers remove projection ambiguity but do not add intrinsic generator frequencies: the capacity remains (lfloor d/2\rfloor). Thus 4D/5D have two positive modes, 6D/7D three, and 8D/9D four. Eight dimensions first pass the four-Balmer-feature count, but this is only a necessary capacity gate and predicts no wavelength, intensity, selection rule, or physical dimension. All 132 package tests and a separate observer-bank audit pass. These are written mathematics and finite synthetic computation, not measured higher-dimensional strings.

**Revision 1.54 addition.** Appendix BC subjects the new five-coordinate harmony observer to an external hydrogen spectroscopy gate. The NIST H I strong-line table supplies four resolved visible Balmer features at representative air wavelengths 410.17400, 434.04620, 486.13615, and 656.28518 nm. A fixed real 5-by-5 skew generator with a linear observation has at most two distinct positive temporal frequencies; any further linear exposure can only attenuate or erase them. Therefore the current model cannot reproduce all four features as one linear temporal spectrum, even though two free frequencies can be calibrated to any selected pair. In the three frozen ratio families, the 3:2 case matches two of four features at 1% only after calibration, while the exact capacity deficit remains two. Seven new tests and an independent Decimal replay pass. This is a useful rejection of the present linear observation model, not of every higher-dimensional theory: adding state transitions, nonlinear mixing, time variation, interactions, or more state dimensions creates different model classes that require fresh predictions. No mismatch is assigned to a hidden string, a fifth dimension, or a theological cause.

**Revision 1.53 addition.** Appendix BB replaces the undefined phrase “five-dimensional harmony” with a complete classification for fixed norm-preserving linear motion.  Every real 5-by-5 skew generator has two nonnegative rotation frequencies and one zero direction, so every linear 3D projection is a constant plus at most two elliptical sine/cosine harmonies.  The engine classifies eight archetypes and 512 randomized generators by spectral relation, visible-hidden coupling rank, known-generator observability, one-history Hankel rank, projected span, exposure loss, hidden-basis gauge, and tick alias.  It recovers all 512 bounded matrix-supplied spectral labels with maximum frequency error \(1.11\times10^{-15}\), but exact counterexamples prevent ontological overreach: two rank-one couplings have observability ranks four and five; a full-rank known system can have only rank-three single-history output; 101 hidden-basis rotations change the generator by 4.0736 while preserving every 3D sample within \(5.17\times10^{-14}\); and finite exposure lowers recoverable rank from five to three or one.  Observation is passive in this engine and does not physically collapse the latent state.  All 114 tests pass and an independent implementation checks 48 catalog relations, 2,048 randomized relations, 6,464 gauge-time states, two exposure erasures, and 81 aliases.  These are written mathematics and finite synthetic calculations, not evidence for physical fifth-dimensional particles.

**Revision 1.52 addition.** Appendix BA tests whether molecular vibrations alone distinguish solids, liquids, and gases.  A deliberately phase-blind water descriptor using the three NIST isolated-molecule fundamentals is identical for ice, liquid water, and vapor controls at one atmosphere, so every deterministic descriptor-only classifier gives one label and must miss at least two of the three balanced controls.  The replacement is an explicitly collective spectral model, motivated by velocity-autocorrelation and two-phase thermodynamics methods, with a diffusive low-frequency component and a solid-like vibrational component.  In the frozen surrogate, the gas-like fraction is recovered exactly on 101 continuum states and the three reference spectra have positive pairwise squared Hellinger distances; however, the continuum also proves that there is no universal three-family threshold in vibration space.  All 98 tests pass and a separate implementation audits 2,401 frequencies, 101 continuum states, three phase labels, and three spectral distances.  The line shapes are constructed rather than molecular-dynamics or measured spectra, so the result is a checked measurement design, not evidence for a fifth dimension, a universal lattice, or a theological cause.

**Revision 1.51 addition.** Appendix AZ turns the next alleged “hidden variable” into a quantified bubble inverse problem.  The revision-1.50 value 4110.9205 Hz remains valid under its declared ideal Minnaert assumptions, but adding vapor pressure and surface-tension stiffness gives 4048.9557 Hz for the same 0.800 mm bubble at fixed effective polytropic exponent \(\kappa=1.4\), a 1.5073% shift.  Holding the old pitch fixed while scanning the explicit sensitivity bracket \(1\le\kappa\le1.4\) gives radii from 0.6659558 to 0.7879498 mm.  A 401-state family adjusts fractional volume amplitude so that carrier frequency, oscillating volume, decay, arrival, and distance all remain fixed; its endpoint pressure traces agree within \(3.667\times10^{-16}\) Pa.  The corresponding 3-by-4 log-observable Jacobian has rank three and nullity one.  Thus even the complete pressure trace in this bounded model selects an observation fiber rather than a unique source state.  All 88 tests pass and an independent implementation checks 401 states and 7,680 endpoint samples.  The interval is synthetic sensitivity analysis, not a measured thermal law, and the result identifies no extra dimension, string particle, or theological cause.

**Revision 1.50 addition.** Appendix AY adds a Water-Drop Acoustic Observer. A declared 4.0 mm drop produces a finite-depth capillary--gravity packet, and one declared 0.800 mm entrained bubble supplies an ideal 4110.92 Hz Minnaert carrier to a surface-coupled airborne pressure candidate. The calculated 0.75 m air transit is 2.16682 ms, so the delivered three-second movie synchronizes 180 picture frames at 60 fps with 144,000 pressure samples at 48 kHz. All 79 tests pass, and an implementation-independent audit reproduces 1,658,880 surface scalars, every pressure sample, and the media metadata. Two corrections are retained: the human eye has no universal frames-per-second clock, and one tone does not uniquely recover bubble geometry—doubling radius while quadrupling pressure leaves ideal pitch unchanged. Bubble formation, damping, coupling, and room response are declared rather than predicted, so the waveform is a finite synthetic candidate, not a calibrated recording, measured physical evidence, or evidence for a hidden dimension.

**Revision 1.49 addition.** Appendix AX adds the Sun–Water Eye Observer. A fixed distant 5772 K spectral source illuminates a constructed moving water surface; the executable uses the IAPWS wavelength-dependent refractive index, unpolarized Fresnel reflection, vector Snell refraction, and a finite CIE-standard observer to render a 150-frame underwater caustic at 60 fps. The run contains 48,988,800 ray–wavelength–exposure samples, all 65 tests pass, and a separate implementation checks the optical formulation, 9,331,200 rendered RGB channels, and media metadata. The decisive correction is an observer collision: two nonnegative spectra remain 0.709286 apart in relative spectral norm while their XYZ outputs agree within \(1.421\times10^{-14}\). Human color observation is therefore not one-to-one and cannot measure every optical property. The animation is also not a hologram because it transports spectral irradiance and discards complex optical phase; 60 fps governs only displayed exposure averages, while the 384–789 THz optical carrier is analytically averaged. The water motion and camera are constructed rather than measured, and no individual-photon path, physical fourth/fifth dimension, or universal grid/tick is inferred.

**Revision 1.48 addition.** Appendix AW turns the origami/oscilloscope analogy into a precise observation map and ordinary electrostatic control. Rigid TIP3P-like charges are neutral and give a 2.346971 D model dipole. A single-Debye response calibrated to the published 18.56 GHz primary relaxation of 25 degree C water has the expected \(1/\sqrt2\) gain and 45-degree lag, with an independent numerical integration agreeing to \(3.529\times10^{-10}\). The decisive result is non-identifiability: two visibly different 192-coordinate molecular orientation histories remain 0.688568--0.820871 RMS apart while producing the identical complete vector-mean polarization trace. The oscilloscope therefore records a fold of the state, not the state itself. All 52 tests pass and an independent implementation checks 92,160 stored atomic-coordinate scalars. The relaxation frequency is a calibration input, the microscopic movies are constructed controls rather than liquid-water molecular dynamics, and no physical fifth dimension, universal grid/tick, magnetic mechanism, or theological cause is inferred.

**Revision 1.47 addition.** Appendix AV builds the requested Water Harmonics engine without promoting the animation to evidence. Sixty-four explicit rigid H2O molecules occupy a 12.429442-angstrom periodic box derived from a 997.047 kg/m3 room-temperature density input. An ordered degree-four tetrahedral scaffold carries a 320-coordinate state: three visible displacement coordinates and two hidden coordinates per molecule. The frozen generator has block form \(G=\left[\begin{smallmatrix}0&C\\-C^T&0\end{smallmatrix}\right]\), so its flow preserves state norm; the first three coordinates become a 64-frame 3D animation while translucent halos expose the otherwise hidden phase convention. Forty-one package tests pass and an independent implementation audits 36,864 rendered atomic-coordinate scalars. The counterexample is decisive: an arbitrary hidden rotation with no visible coupling gives exactly zero 3D motion, while integer-tick frequency aliases remain. The scaffold is not liquid-water molecular dynamics, the density and geometry are inputs rather than predictions, the colors are not measured fields, and electricity, magnetism, hydrodynamics, quantum dynamics, a universal physical tick, and a physical fifth dimension remain unestablished.

**Revision 1.46 addition.** Appendix AU implements the requested matrix language for candidate higher-dimensional molecular motion. A graph-derived five-coordinate state evolves by a real skew-symmetric generator, \(X_5(t)=X_5(0)e^{t\Omega^T}\), and a fixed projector gives explicit 3D positions and velocities. All 17 frozen C5--C7 alkane rows receive two frequency coordinates derived from their reference boiling and fusion values, and a 65-frame n-heptane animation is generated. The inverse claim fails, however: shuffled property labels reconstruct equally exactly; an arbitrary uncoupled hidden-plane angular speed is completely invisible in 3D; and fixed-tick samples alias \(\omega\) with \(\omega+2\pi k/\Delta\). Thirty unit tests pass, and an independent implementation reproduces 68 property/frequency checks, 2,730 trajectory scalars, 124 hidden-invariance states, and 205 alias pairs. The matrices are therefore a valid theoretical candidate family and visualization, not a unique recovery of physical fourth- or fifth-dimensional motion from static property data.

**Revision 1.45 addition.** Appendix AT executes the larger alkane-isomer extrapolation reserved by revision 1.44.  Ordinary least-squares models are fitted only to all C5 and C6 constitutional isomers and evaluated once on all nine C7 isomers.  A topology model using carbon count, the carbon-skeleton Wiener index, and the first Zagreb index reduces held-out normal-boiling MAE from 8.554 K for carbon count alone to 4.500 K and RMSE from 10.107 K to 5.382 K.  The same representation catastrophically fails for fusion: MAE rises from 27.574 K to 135.934 K, including an unclipped 410.46 K prediction for n-heptane versus 182.6 K.  Seventeen tests and an independent graph/linear-algebra replay pass.  The supported conclusion is property-specific: simple topology transfers useful boiling information across this size split but does not encode crystal-sensitive melting.  Failure establishes neither a hidden string nor a unique omitted variable; packing, conformation, polymorphism, intermolecular free energy, small-sample extrapolation, and model form remain live alternatives.

**Revision 1.44 addition.** Appendix AS executes a bounded real-data test of the proposal that arrangement predicts material properties.  The three pentane isomers have identical formula and mass but NIST average normal boiling points of 309.2 K, 301.1 K, and 282.6 K.  A preregistered carbon-skeleton Wiener-index rule trained on the two structural endpoints predicts the held-out middle isomer at 295.9 K, 5.2 K below its measured average, and places it on the correct side of the 293.15 K boiling boundary.  Any formula-only deterministic predictor has worst error at least 13.3 K on these data.  The counterexamples are retained: fusion temperature is not monotone in the same coordinate, and period-5/fcc/d10 silver and palladium differ in recommended 300 K total resistivity by a factor of 6.470 (see the reference-temperature note in AS.4).  Twelve tests and a separately implemented audit pass.  The supported conclusion is that geometry carries predictive information but no single scalar harmony, coarse lattice label, or hidden-dimensional story currently predicts phase or conductivity.  Temperature, pressure, packing, electronic bands, valence, and scattering remain explicit variables.

**Revision 1.43 addition.** Appendix AR turns the fixed-grid/fixed-tick premise into a falsifiable engine and separates it from the stronger claim that every observable value must itself be an integer multiple of a coordinate increment. The new lab implements an exact reversible integer wave in four spatial coordinates at integer time ticks and constructs two distinct full states with the same 3D projection. It also proves that finitely many positive-width measurement intervals cannot identify an arbitrarily small universal tick: every tick no larger than twice the smallest tolerance passes by nearest-integer rounding. Applied to the printed precision of two NIST Lyman-alpha wavelengths, 1 attosecond, 1 zeptosecond, and 0.1 zeptosecond direct ticks fail, while the CODATA Planck-time candidate passes trivially far below the reporting resolution. Decimal reporting itself creates an exact apparent tick, demonstrating why common divisibility of rounded values is not evidence for a natural lattice. Thirteen tests and one separately implemented exact audit pass. These are written mathematics and finite computations, not evidence that spacetime, hydrogen, or all observables obey this lattice.

**Revision 1.42 addition.** Appendix AQ corrects the statement that the ideal hydrogen engine omitted electromagnetism: its Coulomb potential is already the nonrelativistic electromagnetic interaction. A frozen H0/H1/H2 ladder shows that ordinary electron-proton reduced mass removes 97.7729% of the old Lyman-alpha centroid error. A reduced-mass point-Dirac calculation predicts both fine-structure lines, removes 99.3796% of the old error, and leaves a -3.3034 ppm centroid residual. That remainder is not assigned to a fourth spatial plane because complete recoil, Lamb/QED, finite-size, hyperfine, linewidth, and external-field terms remain omitted. A higher-dimensional candidate now requires a derived operator, coupling, symmetry, frozen parameters, and a fresh state-dependent prediction before fitting.

**Revision 1.41 addition.** Appendix AP tests two proposed explanations of the hydrogen rotor failure. A frozen mixture of up to four phase-shifted copies of the same release-0.3 geometry improves held-out squared Hellinger error by only 0.3403%, remains nearly eight times worse than a time-mean baseline, and fails five of six gates. This rejects that bounded missing-component rescue, not every hidden-variable model. A separate spectroscopy-clocked two-level calculation shows that negative temperature reverses population ordering only inside an upper-bounded truncation; thermal equilibrium is stationary because its density matrix commutes with the Hamiltonian. Coherent phase direction reverses under time reversal, not by changing temperature. Absolute zero and a finite “maximum Kelvin” are therefore not opposing rotation phases, and melting delay does not identify microscopic rotation direction.

**Revision 1.40 addition.** Appendix AO calibrates the Hydrogen Matrix Lab's animation clock to the two NIST Lyman-alpha fine-structure wavelengths and executes a frozen train/holdout test of the tangent-four-dimensional-sphere projected-envelope hypothesis. The ideal fixed-proton Coulomb wavelength misses the NIST listed-intensity-weighted centroid by -0.0532468%. After fitting on eight phases, the rotor's mean squared Hellinger error on sixteen unseen phases is 0.172620, versus 0.022197 for a frozen time-mean baseline and 0.072179 for a spherical radial baseline. It fails four of five predeclared gates, including held-out axial-centroid correlation, nodal leakage, and opposite-phase mirror structure; an independent implementation reproduces the calibration, matrices, scores, and rejection. The exact rotor family is rejected as an adequate generator of the declared quantum reference. NIST spectroscopy constrains energy/time rather than a unique 3D ontology; no measured time-resolved 3D density or fourth spatial dimension is claimed.

**Revision 1.39 addition.** Appendix AN resolves the v1.26 dependence counterexample only inside a declared binary first-order Markov class. A past-only Krichevsky--Trofimov predictor divided by a fixed null transition probability forms a nonnegative likelihood-ratio martingale, giving a 1% anytime bound by Ville's inequality. On fresh 400-look synthetic paths from the declared 90%-persistent null, the Markov process crossed on 39/5,000 histories, while the old iid-sign process--whose assumption is false there--crossed on 4,134/5,000. The new process detected 4,629/5,000 chains at 97% persistence, 4,999/5,000 at 75%, and 5,000/5,000 balanced iid chains. A separately implemented replay reproduced 24,000,000 decision states. The guarantee requires the fixed transition matrix and first-order homogeneous Markov law in nonzero-sign event time; it is not distribution-free, does not validate estimated or changing null persistence, and supplies no physical or theological identification.

**Revision 1.38 addition.** Appendix AM tests the proposal that pieces exhibit reusable movement frequencies. A 102-class relational coordinate combines piece type, oriented displacement, square phase, and distance/geometry changes to both kings. Eight phase-shuffled five-ply training searches freeze a defender ordering without pruning any legal move. On the White-attacker nine-ply proof it reduces calls by 6.82%, generated edges by 18.26%, horizon nodes by 28.83%, and measured proof time by 34.76% versus C7; including training, total time falls 29.36%. An audit checks 168,882 legal-move instances and confirms every learned order is a complete permutation. Stronger variants fail: direct pair-conditioned dances add 49–81% calls, online learning adds 52.53%, and the Black-attacker relation model worsens calls, edges, and time. The frequency is therefore a role-specific computation-allocation pattern, not a universal chess law, sound pruning rule, state compression, or solution of chess.

**Revision 1.37 addition.** Appendix AL introduces a horizon-safe state projection for exact bounded chess proofs. Because an identical full chess position cannot recur in two plies, its recurrence period is at least four; a position first encountered at ply \(i\) cannot become threefold before ply \(i+8\). At a nine-ply horizon, only the root and post-first-move repetition candidates must remain in the memo key. The independently verified computation proves that neither White nor Black can force mate from the orthodox initial position within nine plies. White requires 2,951,587 calls and 5,259,337 generated edges; Black requires 127,991 calls and 736,248 edges. A separately written verifier exactly reproduces all reported counts and checks 9,302 legal two-ply non-return paths plus explicit first-threefold cycles. The finite mate frontier is extended, but the root WDL interval remains `[-1,+1]`; chess is not solved.

**Revision 1.36 addition.** Appendix AK runs C6 on the actual orthodox initial position rather than a synthetic tactic. Dependency-ordered AND/OR search independently verifies that neither White nor Black can force checkmate within seven plies. The White-attacker run visits 128,724 nodes and the Black-attacker run 14,846; a separately written lexical verifier reproduces both negative results with 106,322 and 15,082 nodes. Combined, C6 uses 18.26% more nodes than lexical order, falsifying transfer of its small-corpus ordering advantage to this negative-mate proof. White's C6 frontier grows from 21 nodes at one ply to 486, 7,536, and 128,724 at three, five, and seven plies. The run does not narrow the initial WDL interval `[-1,+1]`; it shows that ordering alone leaves the exponential wall intact. The next required object is a proof-carrying forcing-dependency abstraction with sound completion bounds.

**Revision 1.35 addition.** Appendix AJ replaces raw local attack counts with a legal single-square capture–recapture dependency graph. On a deterministic eight-position synthetic corpus requiring genuine legal recaptures and minimal three-ply forced mates, C6 reduces exact AND/OR proof nodes from 10,394 under lexical order to 668, versus 809 for C5 tension and 699 for a checks/captures forcing baseline. Thus C6 improves aggregate nodes by 17.43% over C5 and 4.43% over forcing on this corpus, but its graph construction leaves it 10.22% slower than forcing in measured wall time; it also loses to forcing on four of eight individual positions. An independent verifier reconstructs all eight exchange graphs, reproduces every winning root move, and checks 33 defender branches with 53 mating continuations. The exchange value remains heuristic outside its restricted subgame and cannot prune; the orthodox initial position remains unsolved.

**Revision 1.34 addition.** Appendix AI turns the proposed attack/defence map into a local tension field. For each square it records White and Black attack multiplicities, their product as interaction tension, their difference as control imbalance, and occupant fragility. A tension-breaking entry is a capture into a contested enemy square with a favorable pre-move attack surplus and positive local tension release. On a 43-legal-move Scholar's-mate fixture, the field identifies `f7` as a two-attacker/one-defender target and ranks `Qxf7#` first; an independent implementation rebuilds all attack origins, checks all 64 square formulas, and verifies the mate and its zero replies. The counterexample is equally important: total tension rises from 16 to 28 after checkmate, so heat is a computation-allocation signal rather than a value function. C5 orders moves but cannot prune any move without the earlier interval-dominance certificate; full chess remains unsolved.

**Revision 1.33 addition.** Appendix AH formalizes human chess search as a constraint wave over attack maps, safe regions, mobility, confinement, relational orientation, and worst replies. In table-hidden KRK/KQK experiments, a shared geometric signature proves every WDL value without reading solved labels. Mobility, safe-region size, and dot/cross orientation each improve sound mate-distance bounds; the combined wave makes DTM exact for 44,633/50,015 KRK states (89.24%) and 45,276/46,137 KQK states (98.13%) and certifies 90.65% and 92.45% of legal choices as ignorable under WDL+DTM bounds. An independently written verifier checks all 96,152 state rows against exact certificates and reproduces the dominance totals with zero violations. The same experiment rejects the current construction as a full-chess shortcut: state compression falls to about 1.007× and successor-profile compression to about 1.000×. The next candidate is a constraint-delta quotient that canonicalizes what a move changes rather than encoding the absolute board.

**Revision 1.32 addition.** Appendix AG crosses from isolated endgames to a full orthodox move-state engine and an exact probe of the normal initial position. Six official Stockfish perft references totaling 500,117 nodes and ten targeted castling, en-passant, promotion, repetition, move-count, mate-precedence, and basic-dead-material checks pass. A four-ply proof-only search visits 206,604 nodes and 206,603 edges, finds eight checkmate leaf visits and 242 certified dominated choices below the root, but eliminates zero of the twenty initial moves: every root child and the root retain $[-1,+1]$. The proposed move-elimination value is refined into forcing dominance: game-theoretic result first, mate efficiency second, and the number of certificate-backed ignored alternatives third. This is a proof-efficiency tie-breaker, not a heuristic license to delete moves. Full chess remains unsolved; the active wall is strong sound pre-solution bounds plus a complete dead-position treatment and an independently verified strategy DAG.

**Revision 1.31 addition.** Appendix AF freezes the full-chess target and converts the proposal to discard disadvantageous moves into an exact interval-dominance theorem. Every state carries sound White-relative bounds $[D^-,D^+]$; a White move $a$ may be removed only when another move $b$ has $D^-(b)\ge D^+(a)$, with the dual inequality at Black nodes. An exact KQK computation independently verifies 368,452 raw states, 46,137 $D_4$ classes, 609,873 transitions, and a 20-ply maximum mate. Strong-bisimulation refinement compresses KRK only from 50,015 to 49,885 classes and KQK from 46,137 to 45,249, falsifying the idea that strict future-law identity alone crosses the scaling wall. Post-solution pruning measures show that strictly worse WDL moves remove only 7.21% and 6.61% of edges, while WDL+DTM oracle labels reduce ideal branching to 1.888 and 2.306. Because those labels already assume the answer, the new candidate is proof-carrying interval abstraction with counterexample-guided refinement. Full-chess, quantum-algorithmic, and physical-dimensional claims remain separate gates.

**Revision 1.30 addition.** Appendix AE applies the framework's hidden-state, projection-collision, canonicalization, and cycle rules to a finite adversarial system. Dimensional Chess C1 solves all 399,112 legal states of the declared 8×8 king-and-rook versus king universe. A proved \(D_4\) quotient reduces them to 50,015 classes; 400,120 exhaustive symmetry-transition comparisons pass. Retrograde fixed-point analysis classifies 21,959 wins, 25,260 losses, and 2,796 draws and finds a 32-ply maximum forced mate. An independently written verifier checks every state, transition, Bellman equation, policy edge, distance, and concrete principal-variation move. This exactly solves the declared KRK abstraction, not full chess; the experiment identifies a sufficiently powerful value-preserving canonicalizer as the unsolved scaling wall.

**Revision 1.29 addition.** Appendix AD instantiates Wallbreaker Protocol P3 as P3-R1, an unactivated IBM Quantum-compatible three-qubit pilot. Three prefix-coded eight-cycle probes preserve the ideal Born marginal over each complete cycle by construction. The runner freezes predictions for both allowed settings before delayed random setting generation, binds calibration and ledgers with SHA-256 manifests, uses one shot per circuit/PUB, applies a conservative composite-null e-process, and rejects incomplete or altered provenance. An independent implementation agrees on null, planted-lift, and setting-leak synthetic worlds and 5,000 randomized mathematical checks. No QPU job has been submitted; the current configuration intentionally refuses live execution until a fresh UUID, explicit accessible backend, and actual calibration artifact exist. This is experimental infrastructure, not physical evidence.

**Revision 1.28 addition.** Appendix AC turns the remaining “walls” into a typed pass/fail ledger. It proves a general decomposition of evidence against Born statistics into conditional lift information, marginal Born divergence, and decoder misspecification; the Information-Rate identity follows for marginal-preserving extensions. It also proves that finite observations cannot determine an unrestricted continuation law and introduces simultaneous outcome, setting-leak, and remote-signaling evidence channels. Six 50,000-trial adversarial worlds and 5,000 randomized decomposition checks pass in production and independent implementations. These are mathematical and synthetic results. No natural Born-predictive variable, extra physical dimension, Lorentz-compatible coupling, P=NP proof, or hardware result is claimed.

**Revision 1.27 addition.** Appendix AB integrates the individual Born-outcome frontier with the existing projection, RRLS, and anytime-valid evidence disciplines. It proves a conservative composite-null supermartingale for calibration uncertainty, a Kraft-weighted e-process for a prefix-coded lift language, and a data-processing collision criterion for compressed predictive coordinates. An independent implementation reproduces 5,000 randomized property checks, two 120,000-trial synthetic paths, and the complexity-penalized power table. These are mathematical and computational results, not physical evidence that Born-random outcomes are predictable. The seven typed layers, eight Q1 basis phases, and seven relative phase coordinates remain distinct objects. Rope Tomography Lab and RRLS code and evidence are unchanged.

**Revision 1.26 addition.** Appendix AA replaces the vague goal of “heavy-tail robustness” with an exact conditional sign-balance null. A fixed two-sided sign e-process has a written 1% anytime guarantee whenever the selected residual sign has conditional expectation zero; magnitudes need no moments or parametric distribution. Fresh synthetic checks retain 37/5,000 Student-(t_3) null histories and 33/5,000 Cauchy null histories as finite crossing counts while detecting 4,989/5,000 and 4,992/5,000 shifted (t_3) histories. Counterexamples prove that mean zero, marginal sign balance, scale errors, alternating shifts, and isolated magnitude shocks are not covered. This new engine is standalone and does not share or renew the released Gaussian engine's 1% budget. RRLS is unchanged.

**Revision 1.25 addition.** Appendix Z performs the single fresh paired validation reserved by revision 1.24. The 45%-point/10%-mean candidate improves the declared worst target, noisy six-sigma shocks, by 1.935 percentage points but loses 5.165 points on the persistent-mean target. It therefore fails the predeclared two-point per-target loss guardrail, and the released 30%-point/25%-mean design is retained. Both allocations remain Pareto alternatives rather than one universally dominating the other. The mathematical 1% bound, core detector, sealed v1.22 validation, and RRLS are unchanged.

**Revision 1.24 addition.** Appendix Y executes revision 1.23's preregistered training-only boundary expansion without reading the sealed v1.22 validation set. Opening the grid changes the conclusion: the released 30%-point/25%-mean cell wins 0 of 64 new training runs, while 16 expanded-grid cells win at least once. A separate 80,000-history synthetic reference selects a 45%-point/10%-mean candidate and improves the declared worst-target rate from 81.80% to 83.58%, but exact selection is unstable and the candidate trades away mean-shift sensitivity. The candidate is therefore recorded but not adopted. The released adequacy engine, its probability guarantee, and RRLS remain unchanged pending one fresh frozen validation.

**Revision 1.23 addition.** Appendix X tests the repeatability of revision 1.22's allocation choice without reopening its validation set. Across 64 independent training seeds, the frozen 30%-point/25%-mean cell won 49 times, while four cells won at least once. A separate 80,000-history synthetic reference study selected the same cell and found small reference minimax regret for the seed-selected alternatives. However, 62 of 64 winners used the largest tested point fraction and 51 used the smallest tested mean weight. Thus the choice is usually repeatable inside the declared grid but is boundary-limited, not a demonstrated global optimum. The adequacy engine and RRLS are unchanged.

**Revision 1.22 addition.** Appendix W adds a frozen three-component adequacy rule. A known-covariance mean/variance e-process mixture receives 0.7% and a separately budgeted point-residual gate receives 0.3%; a union bound, not an invalid cross-filtration e-value merge, controls the total anytime false-rejection probability at 1% under the declared Gaussian null. The two allocation weights were selected on training simulations and frozen before one held-out synthetic validation. The hybrid preserves much of the persistent-error power of revision 1.21 while recovering isolated-shock sensitivity, but a heavy-tail counterexample becomes worse. RRLS is unchanged.

**Revision 1.21 addition.** Appendix V adds a canonical-filtration Gaussian mean-plus-variance e-process and tests it against the v1.19 mean-plus-point rule under one total 1% anytime budget. The new fixed mixture is much stronger for persistent moderate scale errors and the targeted mean error, but a constructed isolated 6-sigma shock exposes the opposite tradeoff. The old engine is retained; neither procedure is called universally superior. Heavy-tail failures remain explicit. RRLS is unchanged.

**Revision 1.20 addition.** Appendix U implements the moderate-scale test proposed in revision 1.19. A fixed Gaussian covariance-scale likelihood-ratio mixture accumulates squared-residual evidence with an anytime guarantee under an explicit conditional Gaussian model. At the same reserved 0.5% budget, fresh synthetic validation raises detection of the prespecified 1.25-times scale error from 125/2,000 for the point gate to 1,774/2,000. Counterexamples show that the statistic neither identifies a unique variance mechanism nor remains valid for arbitrary heavy-tailed noise. The existing joint engine and RRLS remain unchanged pending fresh joint validation.

**Revision 1.19 addition.** Appendix T gives the point-residual and cumulative-mean checks one shared 1% anytime false-rejection budget. Because the two components use different filtrations, they are not silently averaged into a claimed e-process. A 0.5%/0.5% allocation and event-level union bound provide the stated joint guarantee, whose null is the known-covariance Gaussian null because the point gate requires the declared covariance (clarified in revision 1.69). Fresh synthetic validation confirms complementary power and exposes severe heavy-tail fragility and weak sensitivity to moderate scale changes. RRLS remains unchanged.

**Revision 1.18 addition.** Appendix S replaces an unnecessarily known-variance assumption for one bounded task: sequentially testing a prespecified projected Gaussian residual mean. The new scale-invariant e-process permits one fixed unknown positive variance. Its guarantee requires iid Gaussian residuals with one fixed scale; the ever-rejection bound holds under any look schedule, but the process is an e-value only at stopping times of its reduced filtration and cannot be merged with canonical-filtration processes (wording corrected in revision 1.69). It is not valid for arbitrary noise. Fresh synthetic validation, exact rescaling controls, and an independent replay record both its improvement and its missed alternatives. RRLS remains unchanged.

**Revision 1.17 addition.** Appendix R implements the cumulative Gaussian likelihood-ratio/e-process proposed by revision 1.16. A written martingale argument supplies an anytime false-rejection bound under explicit conditional-normal assumptions. Fresh synthetic validation shows high sensitivity to the targeted persistent shift and also preserves counterexamples: alternating shifts, orthogonal shifts, and variance errors are not reliably handled by this one detector. The earlier point-residual gate remains separate; combining two 1% gates does not produce a 1% joint procedure.

**Revision 1.16 addition.** Appendix Q addresses the repeated-testing limitation explicitly left open in Appendix O. A separate alpha-spending gate controls ever rejecting a specified correct candidate under a fixed, conditionally calibrated Gaussian model. Fresh synthetic validation and an independent numerical replay expose both reduced false alarms and poor sensitivity to small persistent shifts. No older benchmark is relabeled as new evidence.

**Revision 1.15 addition.** Appendix P audits the supplied dimensional-future/captured-strand model. It separates sampled coincidence from continuous non-crossing, corrects angular-impulse notation and numerical defects, proves a finite-prefix obstruction to unrestricted generator identification, and adds a positive continuation theorem inside a declared bounded-order recurrence class.

**Revision 1.14 correction notice.** The previous Theorem 17.1 had incompatible premises: complete deterministic refinement for one fixed transition family is idempotent, so repeated complete refinement cannot yield seven strict stages. Section 17 now replaces that claim with the closure theorem and a valid finite-horizon example. Appendix O adds intervention-relative identifiability, experiment selection, and explicit model rejection. Earlier benchmark records remain historical evidence for their named releases; this revision does not claim to have rerun all earlier engines or independently revalidated every earlier argument.

---

## Abstract

This paper develops a rigorous form of the proposal that observable reality is a hierarchy of typed layers, each constrained by a richer layer above it. A lower observation is modeled as a projection of an upper state. States that look identical below form a fiber; hidden structure is the distinction between members of that fiber. The framework proves exact conditions under which an upper deterministic law produces a self-contained lower law, calculates the minimum hidden memory needed when it does not, represents repeating dimensions and time as a loop plus winding, and implements finite state, phase, quantum, thermal, and causal calculations in bits.

The theory specializes the hierarchy to seven declared layers: identity/state, relationship, physical space, ordered time, phase/possibility, observation/evidence, and compatibility/constraint. These are typed descriptive roles, not seven interchangeable spatial axes. Under the terminal-completeness hypothesis H7, the seventh layer is a mathematical wall: its state determines all lower projections. Algebra alone does not establish that nature satisfies this hypothesis.

The executable framework reproduces standard density-matrix predictions for a two-state Schrödinger-cat model, two-path interference and decoherence, simulates fixed-point quantum circuits, calculates thermal occupation and selective cooling, searches exact finite invariants, proves projective law descent for finite systems, and fits a compact projection law to published Sycamore cross-entropy-benchmark data. A frozen eight-point holdout was predicted more accurately than two declared baselines. That result validates the engine's prediction workflow, but it identifies only the observable composite \(g=\pi\circ T\), not a unique hidden transition \(T\), a unique projection \(\pi\), or seven physical dimensions.

The paper gives a conditional hidden-state result: reproducibly different future distributions under matched present observations and controlled interventions show that the observed state is insufficient, subject to the stated calibration and statistical assumptions. For a specified finite deterministic system, partition refinement calculates its coarsest sufficient distinction. These are different claims: finite experimental evidence does not provide an exact complete transition table for nature. Non-identifiability results explain why passive effects need not reveal a unique hidden ontology. A controlled tomography and intervention program tests declared models against alternatives.

Version 1.2 adds frozen truth-distance coordinates. Positive operands are mapped to complementary D0 barycentric distances, finite SAT assignments are mapped by their exact violated-clause count, and typed partial-differential-equation residuals are retained at their spacetime coordinates. A smooth three-dimensional Arnold-Beltrami-Childress Navier-Stokes family is verified analytically and computationally. These constructions turn errors into searchable evidence without altering ordinary truth conditions or claiming a Millennium solution.

An optional Christian interpretation is included as a separate metaphysical layer. It may treat God as the ground and sustainer of the complete structure, but God is neither an unexplained residual nor an eighth coordinate. Mathematical, physical, and theological claims are kept separate throughout.

---

## 1. Claim discipline

The theory uses five status labels.

| Label | Kind of claim | What warrants it |
|---|---|---|
| **M** | Mathematical theorem | Deduction from stated definitions and axioms |
| **C** | Computational result | Reproducible execution and tests of a finite implementation |
| **E** | Empirical result | Measurement with provenance and quantified uncertainty |
| **H** | Physical hypothesis | A falsifiable proposal not yet established by data |
| **T** | Theological interpretation | A faith or metaphysical claim not derived from measurement |

The permanent logical boundary is

\[
\boxed{
\mathrm{M\ valid}\not\Rightarrow
\mathrm{H\ true\ of\ nature}\not\Rightarrow
\mathrm{T\ established}.
}
\]

No theory can honestly be made immune to criticism. The aim is stronger: every assertion has a type, assumptions, consequences, and—where empirical—a failure condition.

### 1.1 Meaning of “prove”

This paper uses *proof* in three different senses and never substitutes one for another.

1. A mathematical proof establishes a conditional statement exactly.
2. An exhaustive finite proof checks every state in a declared finite domain.
3. A physical experiment supports or falsifies a model within uncertainty; it does not deductively prove an unrestricted ontology.

Accordingly, this paper proves hidden distinctions **relative to a declared observation map**. It does not claim that existing data have proved seven literal dimensions.

---

## 2. Primitive structure

For each layer \(d\), define

\[
\mathcal X_d=(X_d,\mathcal O_d,\mu_d,\Phi_d),
\]

where:

- \(X_d\) is the state space expressible at layer \(d\);
- \(\mathcal O_d\) is its declared family of observables;
- \(\mu_d\) is a measure or weight when probability is required;
- \(\Phi_d^t:X_d\to X_d\) is its time evolution when that layer is dynamically closed.

For \(e\ge d\), let

\[
\pi_{e,d}:X_e\to X_d
\]

be a surjective projection.

### Axiom A1: projection coherence

\[
\pi_{d,d}=\operatorname{id}_{X_d},
\qquad
\pi_{e,d}\circ\pi_{f,e}=\pi_{f,d}
\quad(f\ge e\ge d).
\]

### Axiom A2: coordinate covariance

Changing coordinates cannot change a physical assertion. For coordinate isomorphisms \(\varphi_d\), the represented projection becomes

\[
\pi'_{e,d}=\varphi_d\circ\pi_{e,d}\circ\varphi_e^{-1}.
\]

Therefore a bare numeral—\(0\), \(1\), \(7\), or \(777\)—does not have invariant physical meaning without a type, unit, chart, and observable interpretation.

### Axiom A3: measured probabilities require a measure

When the upper state is not directly observed, lower weights are pushforwards:

\[
(\pi_{e,d})_*\mu_e(B)
=
\mu_e\!\left(\pi_{e,d}^{-1}(B)\right).
\]

Counting hidden states uniformly is a particular choice of \(\mu_e\), not a universal law. In quantum applications the Born rule must be supplied or independently derived; it cannot be silently replaced by hidden-state counting.

### Definition 2.1: fiber

The fiber above an observed state \(y\in X_d\) is

\[
F_{e,d}(y)=\{x\in X_e:\pi_{e,d}(x)=y\}.
\]

Every member of the fiber is observationally identical at layer \(d\). “Hidden” means unresolved by the chosen projection, not unreal.

### Definition 2.2: complete compatible state

The inverse limit is

\[
X_\infty
=
\varprojlim X_d
=
\left\{
(x_1,x_2,\ldots):
\pi_{e,d}(x_e)=x_d
\right\}.
\]

It contains compatible descriptions at every declared layer. It is a mathematical object and is not identified with God.

---

## 3. The seven typed layers

The executable specialization uses

\[
Z=(D_1,D_2,D_3,D_4,D_5,D_6,D_7).
\]

| Layer | Formal role | Canonical payload |
|---|---|---:|
| \(D_1\) | Identity and local state | 128 bits |
| \(D_2\) | Binary relationships | 64 bits |
| \(D_3\) | Three signed spatial tick coordinates | 192 bits |
| \(D_4\) | Signed linear-time ticks and cycle winding | 128 bits |
| \(D_5\) | Unsigned cyclic phase and possibility branches | 128 bits |
| \(D_6\) | Visible-layer mask and evidence | 72 bits |
| \(D_7\) | Constraints, four gate bits, and integrity hash | 136 bits |
| **Total** |  | **848 bits / 106 bytes** |

The roles are deliberately heterogeneous. Position, time, phase, evidence, and logical compatibility do not share one unit and cannot be added as though they were seven Euclidean coordinates.

### 3.1 Canonical serialization

| Bytes | Field |
|---:|---|
| 0–7 | identity |
| 8–15 | local state bits |
| 16–23 | relationship bits |
| 24–47 | signed \(x,y,z\) ticks |
| 48–55 | linear-time ticks |
| 56–63 | cycle winding |
| 64–71 | phase ticks |
| 72–79 | possibility branches |
| 80 | visible-layer mask |
| 81–88 | evidence bits |
| 89–96 | constraint bits |
| 97 | gate flags |
| 98–105 | FNV-1a integrity checksum of bytes 0–97 |

The prefix lengths are

\[
(P_1,\ldots,P_7)=(16,24,48,64,80,89,106)\text{ bytes}.
\]

The engine defines

\[
\pi_{7,d}(Z)=Z[0:P_d].
\]

### Theorem 3.1: canonical projection coherence **[M]**

For every \(f\ge e\ge d\),

\[
\pi_{e,d}\circ\pi_{f,e}=\pi_{f,d}.
\]

**Proof.** Projecting to \(e\) keeps a prefix of length \(P_e\). Projecting that result to \(d\) keeps its prefix of length \(P_d\). Because \(P_d\le P_e\), the final bytes equal the direct prefix of length \(P_d\). \(\square\)

### 3.2 Is \(D_7\) a wall?

Add the physical hypothesis:

**H7 — terminal completeness.** Every physically possible complete state is represented by one element of \(X_7\); every measurable lower state is its projection; no physically distinguishable variable exists outside \(X_7\).

Define

\[
\Psi(x_7)
=
(\pi_{7,1}x_7,\ldots,\pi_{7,6}x_7,x_7).
\]

### Theorem 3.2: completeness of the finite compatible diagram **[M]**

\[
\boxed{\mathcal C_7\cong X_7,}
\]

where \(\mathcal C_7\) is the set of compatible tuples in the declared seven-layer diagram. **Proof.** Every \(x_7\) yields exactly one compatible tuple by projection. The map is injective because its last coordinate is \(x_7\). Every compatible tuple has a last coordinate, and compatibility fixes all its earlier coordinates as projections of that last coordinate, so the map is surjective. \(\square\)

This diagram identity needs no physical hypothesis. Identifying \(\mathcal C_7\) with complete physical reality additionally requires H7 and a faithful state representation. The identity is true for any finite tower ending at a highest declared layer; it does not select seven. Mathematics can always define \(D_8\). Successful finite measurements may support a model on a stated domain but cannot establish that no additional distinguishable variable exists in every untested domain.

---

## 4. Law descent: how an upper layer confines a lower law

Let \(T:X\to X\) be an upper deterministic transition and \(\pi:X\to Y\) a lower observation.

### Theorem 4.1: factorization or law-descent theorem **[M]**

A deterministic lower transition \(L:Y\to Y\) satisfying

\[
\pi\circ T=L\circ\pi
\]

exists on the realized lower states if and only if

\[
\boxed{
\pi(a)=\pi(b)
\Longrightarrow
\pi(Ta)=\pi(Tb)
}
\]

for all \(a,b\in X\). When it exists, \(L\) is unique on \(\pi(X)\).

**Proof.** If \(L\) exists and \(\pi(a)=\pi(b)\), then

\[
\pi(Ta)=L(\pi a)=L(\pi b)=\pi(Tb).
\]

Conversely, define \(L(y)=\pi(Tx)\) for any \(x\in F_y\). The fiber condition makes the definition independent of the chosen representative, so it is well defined. Its defining equation gives \(\pi T=L\pi\). If \(L'\) also satisfies the equation, then for every realized \(y=\pi x\), \(L'(y)=\pi(Tx)=L(y)\). \(\square\)

This is the exact form of the proposed principle:

> An upper layer confines the autonomous laws visible below precisely when projected upper evolution is constant on every observational fiber.

For a tower

\[
X_7\to X_6\to\cdots\to X_1,
\]

commuting adjacent squares imply commuting composite squares. A failure at one edge identifies the first layer at which the lower description ceases to be self-contained.

### 4.1 Stochastic lower behavior

If the factorization condition fails, a lower state can have several possible successors. Given explicit nonnegative upper weights \(w_x\),

\[
P(y'\mid y)
=
\frac{\sum_{x\in F_y,\ \pi(Tx)=y'}w_x}
{\sum_{x\in F_y}w_x}.
\]

This equation pushes an upper measure downward. It does not infer the weights from labels.

### 4.2 The hidden-mechanism theorem

Define finite-horizon equivalence by

\[
a\sim_k b
\iff
\pi(T^j a)=\pi(T^j b)
\quad\text{for }0\le j\le k.
\]

Define future equivalence by

\[
a\sim_\infty b
\iff
\pi(T^j a)=\pi(T^j b)
\quad\text{for every }j\ge0.
\]

### Theorem 4.2: coarsest deterministic repair **[M]**

For finite \(X\), partition refinement reaches \(\sim_\infty\) after finitely many steps. The quotient \(X/{\sim_\infty}\) is the coarsest refinement of the original observation that makes future evolution deterministic.

**Proof.** Each refinement can only split a finite partition; it can split at most \(|X|-1\) times before stabilizing. At stability, equivalent states have equivalent successors, so the induced quotient evolution is deterministic. Any deterministic refinement preserving the original observation must separate every pair that differs at some projected future time. It therefore refines \(\sim_\infty\). Hence \(\sim_\infty\) is coarsest. \(\square\)

If one original fiber splits into at most \(s\) future-equivalence classes, a fixed-width hidden register needs

\[
\boxed{b_{\min}=\lceil\log_2 s\rceil}
\]

bits, and fewer cannot name all classes.

### Corollary 4.3: operational proof of hidden state **[M/E boundary]**

Suppose two reproducibly preparable systems have the same declared present observation \(y\), receive the same controlled intervention, yet have different future-observation distributions beyond calibrated error. Then \(y\) is not a sufficient state variable for that experiment. At least one additional state distinction—hidden relative to \(y\)—is required.

This is a conditional insufficiency result **relative to the declared observation**. In experiments, a finite estimate of different distributions remains subject to statistical error, drift, and calibration; differing individual outcomes alone can simply reflect a sufficient stochastic model. The result does not identify a unique missing mechanism or determine whether the distinction is spatial, environmental, historical, contextual, quantum, or instrumental. Naming it \(D_7\) requires additional evidence.

### 4.3 Conservation-law descent

An upper Boolean observable \(I:X\to\{0,1\}\) becomes a lower conserved quantity only when

\[
I(Tx)=I(x)
\]

and

\[
\pi(a)=\pi(b)\Longrightarrow I(a)=I(b).
\]

The first is conservation; the second is lower observability. Conservation above does not imply expressibility below.

---

## 5. Effects, causes, and the limit of inverse inference

### 5.1 Linear effect model

Let

\[
y=Hz+\varepsilon,
\]

where \(z\) is a candidate cause, \(H\) is the declared observation operator, and \(\varepsilon\) is measurement noise.

In the exact noiseless real or complex case,

\[
Hz_1=Hz_2
\iff
z_1-z_2\in\ker H.
\]

If \(z_*\) is one solution, every compatible cause is

\[
\boxed{\operatorname{Causes}(y)=z_*+\ker H.}
\]

Using the Moore–Penrose inverse, one canonical representative is \(z_*=H^+y\), and the component of the effect outside the model's image is

\[
r=(I-HH^+)y.
\]

The residual is unexplained by the proposed model. It is not a calculation of God.

### 5.2 Exact Boolean inverse

For bit effects,

\[
Ax=b\pmod2.
\]

XOR Gaussian elimination returns a particular solution \(x_*\) and kernel basis \(k_1,\ldots,k_r\). Every compatible cause is exactly

\[
\boxed{
x=x_*\oplus\bigoplus_{i=1}^r\alpha_i k_i,
\qquad \alpha_i\in\{0,1\}.
}
\]

There are \(2^r\) compatible causes. A unique cause exists only when the nullity \(r\) is zero.

### Theorem 5.1: latent relabeling non-identifiability **[M]**

Suppose controlled inputs \(u\) produce observations through

\[
g(u)=(\pi\circ T)(u).
\]

For any invertible latent relabeling \(S\), define

\[
T'=S\circ T,
\qquad
\pi'=\pi\circ S^{-1}.
\]

Then

\[
\pi'\circ T'=\pi\circ T=g.
\]

**Proof.** Substitute and cancel \(S^{-1}S\). \(\square\)

Therefore passive input/effect data generally identify the composite observable law \(g\), not a unique pair \((T,\pi)\). Separate claims about an internal transition and a projection require calibrated internal tomography, interventions, structural restrictions, or multiple independent observation channels.

### 5.3 Why an unconstrained \(G\) destroys science

If one writes

\[
y=Hz+g
\]

and permits \(g\) to be arbitrary, then for every proposed \(z\), choosing

\[
g=y-Hz
\]

makes the equation fit. Such a theory cannot fail and predicts nothing. The symbol \(G\) may denote an optional metaphysical ground, but it cannot serve as a free residual in a physical fit.

---

## 6. Reversible hidden-coordinate transport

Let two labeled points \(q_1,q_2\in\mathbb R^2\) have center and relative displacement

\[
c=\frac{q_1+q_2}{2},
\qquad
r=q_1-q_2.
\]

For phase \(\theta\), define visible and hidden components

\[
r_v=\cos\theta\,r,
\qquad
h=\sin\theta\,r.
\]

### Theorem 6.1: visible-to-hidden ripple **[M]**

The mapping \((q_1,q_2)\mapsto(c,r_v,h)\) is injective on its image, with

\[
r=\cos\theta\,r_v+\sin\theta\,h,
\]

and

\[
\boxed{\|r_v\|^2+\|h\|^2=\|r\|^2.}
\]

**Proof.** Both identities follow from \(\sin^2\theta+\cos^2\theta=1\). The recovered \(c\) and \(r\) give \(q_1=c+r/2\) and \(q_2=c-r/2\). \(\square\)

At \(\theta=\pi/2\), visible separation vanishes while the relative displacement survives in \(h\). This proves that a declared encoding can hide visible information without destroying it. It does not prove that physical collisions automatically transfer information into another dimension.

### Theorem 6.2: generic collision-capacity bound **[M]**

For \(k\) labeled points in \(\mathbb R^d\), any continuous, locally injective encoding that forces every visible projection to one shared \(d\)-dimensional center while giving each point \(h\) hidden scalars requires

\[
\boxed{d+kh\ge kd.}
\]

Equivalently,

\[
\boxed{
h\ge\left\lceil\frac{d(k-1)}{k}\right\rceil.
}
\]

**Proof sketch.** Generic inputs occupy an open subset of \(\mathbb R^{kd}\). The proposed output contains at most \(d+kh\) real degrees of freedom. A continuous locally information-preserving encoding cannot lower the dimension of an open Euclidean set. Conversely, storing the center and \(k-1\) relative vectors uses exactly \(d+d(k-1)=kd\) scalars and reconstructs every labeled point. \(\square\)

**Remark (revision 1.69).** The continuity hypothesis is essential and was previously omitted from the statement. Without it the bound is false: a set-theoretic bijection \(\mathbb R^{kd}\to\mathbb R^{d+1}\) exists, so \(h=1\) would suffice for an arbitrary encoding.

---

## 7. Time, process, loop, and helix

### 7.1 A state is not a process

An instantaneous state is

\[
x(t)\in X.
\]

Direction is supplied by a law,

\[
\dot x(t)=F(x(t)),
\]

or estimated from at least two observations,

\[
\dot x(t)\approx\frac{x(t+\Delta t)-x(t)}{\Delta t}.
\]

A snapshot is not necessarily dead; it is insufficient by itself to display change. Momentum, current, gradients, and internal state may still be encoded at that instant.

A life can be treated as a trajectory

\[
\gamma:I\to X
\]

whose local organization remains inside a viable set \(L\subseteq X\) for a time interval. Identity through change is an invariant \(I_{\mathrm{id}}(\gamma(t))\), not the absence of all change.

### 7.2 Linear and cyclic time

Linear time may be \(t\in\mathbb R\). Cyclic phase is

\[
z(t)=e^{i\omega t}\in S^1,
\qquad z(t+2\pi/\omega)=z(t).
\]

The loop forgets how many periods passed. Adding a winding number \(n\in\mathbb Z\) recovers the lift:

\[
t=\frac{\operatorname{Arg}(z)+2\pi n}{\omega}.
\]

The pair is helix-like:

\[
H(t)=(e^{i\omega t},t)\in S^1\times\mathbb R.
\]

One projection sees repetition; another preserves history.

### 7.3 Infinite cyclic dimension addressing

For any positive finite index \(n\) and declared cycle length \(c\), Euclidean division gives one unique address

\[
\boxed{n=qc+r,\qquad q\in\mathbb N_0,\quad r\in\{1,\ldots,c\}.}
\]

For \(c=7\),

\[
D_8=(D_1,1),
\qquad
D_{777}=(D_7,110),
\qquad
D_{778}=(D_1,111).
\]

Thus \(D_8\) is defined as a repeated local type plus one winding. Forgetting \(q\) yields the finite loop \(\mathbb Z/7\mathbb Z\); preserving \(q\) yields an unbounded helical address. A finite computer can represent arbitrarily large finite indices subject to memory, not an actually infinite index.

### 7.4 Cycle twist or monodromy

After a full seven-step winding, an internal state may change by

\[
M:S\to S.
\]

| \(M\) | Meaning | Full internal period |
|---|---|---:|
| identity | exact loop | 7 |
| nonidentity permutation | reversible twisted loop | \(7\operatorname{ord}(M)\) |
| noninjective map | information-losing fold | no reversible closure |

If a permutation has cycles of lengths \(q_1,\ldots,q_k\), then

\[
\operatorname{ord}(M)=\operatorname{lcm}(q_1,\ldots,q_k).
\]

### 7.5 Polarity

A genuine reversible polarity at layer \(d\) is an involution

\[
J_d:X_d\to X_d,
\qquad J_d^2=\operatorname{id}.
\]

It is coherent across layers when

\[
\boxed{\pi_{e,d}\circ J_e=J_d\circ\pi_{e,d}.}
\]

This formalizes structural “as above, so below.” Health/sickness, rich/poor, and similar human classifications are usually thresholded observables rather than fundamental involutions:

\[
C(x)=\operatorname{sign}(f(x)-\theta).
\]

The word for one condition may depend on contrast; the condition itself does not logically require its opposite to exist.

---

## 8. Exact increments, ULPs, and computation in bits

### 8.1 Typed dyadic scalar

The core scalar is

\[
q=n2^eU,
\]

where \(n\) is a signed 64-bit tick count, \(e\) is a signed 16-bit binary exponent, and \(U\) is a declared unit. Its fixed schema increment is

\[
\boxed{\operatorname{ULP}(q)=2^eU.}
\]

The smallest formally describable increment is

\[
2^{-32768}U.
\]

This is an encoding limit, not a discovered minimum of nature. Addition requires exponent alignment and a result fitting the signed 64-bit tick field; overflow is rejected.

Here “ULP” means the tick spacing of the chosen fixed-exponent grid. It is not an invariant of the numerical value alone: \(1\cdot2^0U=2\cdot2^{-1}U\) has two encodings with different declared grid steps. Changing an encoding does not improve the precision of a measurement.

### 8.2 Default increments

| Layer/quantity | Default computational ULP |
|---|---:|
| \(D_1\) local state | one declared bit flip |
| \(D_2\) relationship | one declared edge-bit flip |
| \(D_3\) space | \(2^{-24}\text{ m}=5.9604644775390625\times10^{-8}\text{ m}\) |
| \(D_4\) linear time | \(2^{-20}\text{ s}=9.5367431640625\times10^{-7}\text{ s}\) |
| \(D_5\) phase | \(2\pi/2^{64}\text{ rad}\) |
| \(D_6\) observation | one newly distinguished evidence bit |
| \(D_7\) compatibility | one declared constraint/gate bit |
| Q1.30 amplitude/probability | \(2^{-30}\) |
| Q2.60 probability mass | \(2^{-60}\) |

There is no meaningful observer-independent scalar ULP across unlike types.

### 8.3 Observer-relative reality ULP

Let \(z,z'\in\{0,1\}^{848}\), and let mask \(m\) mark the bits an observer distinguishes. Define

\[
d_m(z,z')=\operatorname{popcount}\bigl(m\land(z\oplus z')\bigr).
\]

The states are observationally equivalent when \(d_m=0\). One observer-relative reality ULP is a change with

\[
\boxed{d_m(z,z')=1.}
\]

This is a masked bit-edit distance on raw encodings, not a physical distance or an automatically realizable transition between valid records. For example, changing a payload bit may also require updating checksum bits. Distinct semantic encodings and validity constraints must be accounted for before interpreting a bit edit as a state change.

The complete resolution is a typed vector

\[
\boldsymbol\varepsilon
=(\varepsilon_{\mathrm{state}},\varepsilon_{\mathrm{relation}},
\varepsilon_x,\varepsilon_t,\varepsilon_\phi,
\varepsilon_{\mathrm{evidence}},\varepsilon_{\mathrm{constraint}}).
\]

### 8.4 Precision towers

For radix \(b\ge2\), finite prefixes satisfy

\[
n_{k+1}=bn_k+d_{k+1},
\qquad d_{k+1}\in\{0,\ldots,b-1\}.
\]

Depth \(k\) resolves a cell of width \(b^{-k}\). Compatible infinite paths represent real numbers after choosing a canonical expansion, but no finite \(k\) distinguishes all reals. Computable reals are generated by finite algorithms; most reals are not computable because programs are countable and the real interval is uncountable.

Thus “compute everything” has the defensible meaning

\[
\boxed{
\text{compute every consequence derivable from declared finite bits,
rules, constraints, observations, and tolerances.}
}
\]

It cannot mean solving all uncomputable questions or naming every exact real with finite memory.

---

## 9. Schrödinger's cat in the language

Use \(|A\rangle\) and \(|D\rangle\) as alive/dead basis labels. The reduced two-state density matrix is

\[
\rho_C=
\begin{pmatrix}
p_A&c\\
c^*&p_D
\end{pmatrix},
\]

with

\[
p_A+p_D=1,
\qquad |c|^2\le p_Ap_D.
\]

For a balanced state with visibility \(V\) and phase \(\theta\),

\[
\rho_C=
\frac12
\begin{pmatrix}
1&Ve^{i\theta}\\
Ve^{-i\theta}&1
\end{pmatrix}.
\]

The coherent case has \(V=1\); the classical mixture has \(V=0\). Both give

\[
P(A)=P(D)=\frac12
\]

in the ordinary basis. A phase-sensitive analyzer distinguishes them:

\[
P_+(\varphi)
=
\frac12+\operatorname{Re}\left(ce^{-i\varphi}\right).
\]

Therefore

\[
\boxed{
\pi_{\mathrm{basis}}(\rho_{\mathrm{coherent}})
=
\pi_{\mathrm{basis}}(\rho_{\mathrm{mixture}})
}
\]

while a phase-sensitive observation separates them. The “cat problem” is not solved by declaring the cat dead between snapshots. The language shows exactly which observable loses the off-diagonal coherence and which reveals it. It does not choose a unique interpretation of measurement.

### 9.1 Decoherence

If branch states correlate with environment states \(|E_A\rangle\) and \(|E_D\rangle\), then

\[
\gamma=\langle E_D|E_A\rangle,
\qquad c'=\gamma c.
\]

When \(|\gamma|=0\), the reduced state has no accessible branch coherence. The global state may remain pure while the subsystem appears mixed. This is an exact instance of information residing in correlations discarded by a projection.

The engine encodes observations from Brune et al.'s mesoscopic cat experiment and checks the reported ordering between decoherence and photon-relaxation times. It does not derive the measured numerical ratio from seven-layer axioms.

---

## 10. Two-slit observation

Let the path become correlated with a marker:

\[
|\Psi\rangle
=
\frac1{\sqrt2}
\left(
|A\rangle|M_A\rangle
+e^{i\phi}|B\rangle|M_B\rangle
\right).
\]

With marker overlap

\[
\gamma=\langle M_A|M_B\rangle,
\qquad V=|\gamma|,
\]

the normalized two-path modulation is

\[
\boxed{P(\phi)=\frac{1+V\cos\phi}{2}.}
\]

| Path record | \(V\) | Bright phase | Dark phase |
|---|---:|---:|---:|
| indistinguishable | 1 | 1 | 0 |
| partial | \(0<V<1\) | \((1+V)/2\) | \((1-V)/2\) |
| orthogonal | 0 | 0.5 | 0.5 |

Observation here means a physical interaction capable of leaving distinguishable path information. Consciousness is not required.

In the far-field small-angle approximation,

\[
\phi(x)=2\pi\frac{dx}{\lambda L},
\qquad
\Delta x=\frac{\lambda L}{d}.
\]

For \(\lambda=600\) nm, \(d=500{,}000\) nm, and \(L=10^9\) nm, the fringe spacing is \(1.2\) mm. The integer Phase64 implementation returns the half-turn dark fringe at \(0.6\) mm and the next bright fringe at \(1.2\) mm for full visibility.

From measured extrema,

\[
V=\frac{N_{\max}-N_{\min}}{N_{\max}+N_{\min}}.
\]

For ideal pure path markers,

\[
D^2+V^2=1,
\]

while the experimentally testable general relation is

\[
\boxed{D^2+V^2\le1.}
\]

The language represents the experiment correctly. It does not prove a literal fourth spatial grid. Any deeper hidden model must also reproduce the Born rule, entanglement, Bell/CHSH violations, contextuality constraints, and operational no-signaling.

---

## 11. Quantum computing

The dynamic quantum payload is

\[
|\psi\rangle
=
\sum_{j=0}^{2^n-1}\alpha_j|j\rangle,
\qquad
\alpha_j\in2^{-30}\mathbb Z[i].
\]

The fixed seven-layer record is an envelope and fingerprint, not a lossless replacement:

\[
Z_Q=(H_7,\Psi_n).
\]

The C++17 simulator implements Clifford gates, \(T\), arbitrary Phase64 rotations, controlled gates, measurement, Grover diffusion, and a line-oriented circuit language. Verified demonstrations include:

- Bell-state probabilities and reduced-state purity;
- CHSH value \(2.82842712477\approx2\sqrt2>2\);
- all four one-bit Deutsch oracles;
- three-qubit Grover target probability \(121/128\), approximated as \(0.945312497206\);
- teleportation fidelity \(0.999999997206\) without cloning the source;
- ideal three-qubit bit-flip error detection and correction.

Each complex amplitude occupies 16 bytes, so

\[
\boxed{M(n)=16\cdot2^n\text{ bytes}.}
\]

The implementation cap is 24 qubits. This is a deterministic classical simulator, not a physical quantum processor, and it does not obtain quantum hardware speedup.

### 11.1 Layer interpretation for a circuit

| Layer | Circuit meaning |
|---|---|
| \(D_1\) | computation identity and payload fingerprint |
| \(D_2\) | qubit relationships and controls |
| \(D_3\) | physical topology when supplied |
| \(D_4\) | ordered gates and depth |
| \(D_5\) | amplitude, phase, and alternatives |
| \(D_6\) | basis, shots, seed, and counts |
| \(D_7\) | normalization, unitary, index, and duality constraints |

This is a useful typed organization. It is not evidence that the circuit physically occupies seven spatial dimensions.

---

## 12. Temperature, quanta, and selective entropy removal

For frequency \(f\),

\[
E_q=hf,
\qquad
T_q=\frac{hf}{k_B}.
\]

For a thermalized two-level system,

\[
\boxed{p_e=\frac1{1+e^{hf/(k_BT)}}.}
\]

For a harmonic mode,

\[
\boxed{\bar n=\frac1{e^{hf/(k_BT)}-1}.}
\]

Solving for the warmest ideal temperature meeting target \(p_*<1/2\),

\[
\boxed{
T_{\max}=\frac{hf}{k_B\ln[(1-p_*)/p_*]}.
}
\]

At fixed temperature, the minimum ideal gap is

\[
\boxed{
f_{\min}=\frac{k_BT}{h}\ln\left(\frac{1-p_*}{p_*}\right).
}
\]

This identifies three levers: lower \(T\), raise \(f\), or use an engineered non-equilibrium reset. There is no universal optimum temperature independent of hardware.

### 12.1 Concrete 5 GHz scale

For \(f=5\) GHz,

\[
\frac{hf}{k_B}=239.962154\text{ mK}.
\]

The ideal excited-state thresholds are

\[
p_e\le10^{-3}\Rightarrow T\le34.7431\text{ mK},
\]

\[
p_e\le10^{-6}\Rightarrow T\le17.3690\text{ mK}.
\]

Temperature is only one error source. Relaxation, dephasing, leakage, materials, stray radiation, quasiparticles, control, and readout also enter the operating point.

### 12.2 “Inverse microwave” as anti-Stokes conversion

An inverse heater is not negative energy and cannot remove zero-point motion. A physically meaningful mechanism removes selected excitations. For target mode \(b\), lossy cavity \(a\), and red-sideband drive,

\[
H_{\mathrm{int}}=\hbar g(a^\dagger b+ab^\dagger).
\]

The ideal resolved-sideband model is

\[
\Gamma_{\mathrm{cool}}=\frac{4g^2}{\kappa},
\qquad
n_{\min}=\left(\frac{\kappa}{4\Omega_m}\right)^2,
\]

\[
\boxed{
n_f=
\frac{\gamma_m n_{\mathrm{th}}+\Gamma_{\mathrm{cool}}n_{\min}}
{\gamma_m+\Gamma_{\mathrm{cool}}}.
}
\]

One pump photon plus one target excitation becomes a higher-frequency escaping cavity photon. Energy and entropy move to the output reservoir; they do not vanish.

For the engine's declared hypothetical 10 MHz mode at a 20 mK bath, with \(g/2\pi=20\) kHz, \(\kappa/2\pi=200\) kHz, and \(\gamma_m/2\pi=10\) Hz, the ideal calculation gives an initial occupation of \(41.1752\), final occupation \(0.0514298\), and mode-equivalent temperature \(0.159037\) mK. This illustrates selective cooling; it is not a prediction for an unsupplied device.

### 12.3 Active reset

For upward and downward rates,

\[
\frac{dp}{dt}
=
-\Gamma_\downarrow p+\Gamma_\uparrow(1-p),
\]

\[
p(t)=p_{ss}+[p(0)-p_{ss}]
e^{-(\Gamma_\downarrow+\Gamma_\uparrow)t},
\]

\[
\boxed{p_{ss}=\frac{\Gamma_\uparrow}
{\Gamma_\downarrow+\Gamma_\uparrow}.}
\]

The proposed engineering target is a Selective Quantum Entropy Extractor: a switchable coupler, red-sideband conversion, filtered lossy port, occupation sensor, and feedback controller. It may cool a selected mode far below the equivalent bath occupation, but it still requires an entropy sink and measured device parameters.

### 12.4 Optimization target

For candidate hardware \(a\), evaluate

\[
a=(f,T,p_{\mathrm{nonthermal}},p_{\mathrm{reset}},
P_{\mathrm{wall}},R_{\mathrm{ops}}),
\]

\[
p_{\mathrm{total}}=1-(1-p_{\mathrm{th}})(1-p_{\mathrm{nonthermal}})(1-p_{\mathrm{reset}}),
\]

and useful operations per wall-power unit

\[
\eta_{\mathrm{ops}}
=
\frac{R_{\mathrm{ops}}(1-p_{\mathrm{total}})}{P_{\mathrm{wall}}}.
\]

The engine computes a Pareto frontier rather than pretending temperature alone defines the best design.

---

## 13. Typed higher-layer frequencies, magnets, and energy

A frequency requires a time parameter. The proposed D4–D7 rate object is therefore

\[
\boxed{
\Omega=(\omega_4,\omega_5,\omega_6,\omega_7),
}
\]

with typed meanings:

| Rate | Operational meaning |
|---|---|
| \(\omega_4\) | carrier or state-evolution rate |
| \(\omega_5\) | phase winding or drive rate |
| \(\omega_6\) | observation or sample rate |
| \(\omega_7\) | reset, stabilization, or constraint-relaxation rate |

The engine searches primitive integer relations

\[
m\cdot\Omega\approx0,
\qquad m\in\mathbb Z^4.
\]

An exact relation identifies synchronization in the declared processes. It is not evidence that every rate is an independent time dimension.

Observation can create an apparent low frequency through aliasing:

\[
\boxed{
f_{\mathrm{alias}}
=
\left|f-\operatorname{round}(f/f_s)f_s\right|.
}
\]

Every magnetic or resonant design must close the ledger

\[
\boxed{
E_{\mathrm{inputs}}+U_i
=
E_{\mathrm{outputs}}+U_f
}
\]

within calibrated uncertainty. Relevant relations include

\[
U_L=\frac12LI^2,
\qquad
u_B=\frac{B^2}{2\mu},
\qquad
V_{\mathrm{peak}}=2\pi fN\Phi_{\mathrm{peak}}.
\]

Resonance accumulates supplied energy; it does not create source-free net energy. An unclosed residual is a search target for missing input, storage, loss, or calibration—not a license to declare free energy.

---

## 14. Exact mathematical discovery

For a finite bit universe,

\[
X_n=\{0,1\}^n,
\qquad F:X_n\to X_n.
\]

Let \(M_d\) be the algebraic-normal-form space of Boolean polynomials through degree \(d\). Invariance requires

\[
I(F(x))+I(x)=0\pmod2
\]

for every state. Exhaustive enumeration builds a matrix \(A\).

### Theorem 14.1: invariant-language completeness **[M]**

\[
\boxed{
\ker A=\{I\in M_d:I\circ F=I\}.
}
\]

**Proof.** Every row of \(A\) is precisely one state's linear invariance equation in the coefficient vector. A coefficient vector lies in the kernel exactly when every equation is satisfied. \(\square\)

The engine distinguishes:

| Evidence label | Meaning |
|---|---|
| observed pattern | fits construction data |
| held-out verified | predicts data excluded from fitting |
| finite exhaustive proof | every state in the declared finite domain checked |

Demonstrations include discovering parity as a complete degree-one invariant for rule 150 on five periodic cells; exhaustively calculating rule 30's attractors for twelve cells; scanning all 256 elementary rules on bounded rings; and recovering modular recurrences from training prefixes before testing untouched suffixes.

Finite proof does not automatically generalize to infinite width or continuous physics.

---

## 15. Real measured prediction

The engine's first real-data test uses 33 published Sycamore random-circuit cross-entropy-benchmark fidelities: 28 width-series points at 14 cycles and five depth-series points at 53 qubits.

The split was frozen independently of outcomes:

\[
\text{holdout}\iff\text{row index}\bmod4=3.
\]

This yields 25 training points and eight untouched holdout points. The declared observable law is

\[
\boxed{
\log_2F
=
\alpha m\left(1.5n-0.5\sqrt n\right)+\beta n.
}
\]

Training only produced

\[
\alpha=-0.0044586791200801655,
\qquad
\beta=-0.03894462549892942.
\]

### 15.1 Holdout results **[E/C]**

| Metric | Projection law | Simple \(nm\) decay | Constant baseline |
|---|---:|---:|---:|
| Holdout RMSE in \(\log_2F\) | 0.235065 | 0.343014 | 1.956428 |
| Squared-error improvement of projection law | — | 53.04% | 98.56% |

All eight holdout points were inside nominal normal-approximation 95% prediction bands. The largest relative error was about 29.3%.

This is a successful prediction test for the declared composite law. It is limited by a small holdout, and the feature model was already present in the scientific literature. By Theorem 5.1, it does not separately identify a hidden \(T\), a projection \(\pi\), or seven physical dimensions.

---

## 16. What would constitute evidence for a hidden mechanism?

### 16.1 The target signature

Prepare two ensembles \(A\) and \(B\) such that every measurement assigned to layers \(D_1\) through \(D_6\) gives the same present statistics within a preregistered tolerance:

\[
\pi_{7,6}(x_A)=\pi_{7,6}(x_B).
\]

Apply the same controlled transition \(T_u\). The proposed seventh-layer mechanism predicts

\[
\pi_{7,6}(T_u x_A)\ne\pi_{7,6}(T_u x_B).
\]

If this difference repeats, survives apparatus controls, and is predicted before collection, then Corollary 4.3 establishes a missing predictive distinction relative to the D1–D6 observation set.

### 16.2 Intervention requirement

Passive correlation is insufficient. The experiment must include a control \(v\) intended to change only the candidate hidden distinction while leaving the measured lower state unchanged at preparation:

\[
\pi_{7,6}(I_vx)=\pi_{7,6}(x),
\]

but the theory predicts a changed later effect:

\[
\pi_{7,6}(T_uI_vx)\ne\pi_{7,6}(T_ux).
\]

This is a fiber intervention: an operation inside the present observational fiber that becomes visible later.

### 16.3 Tomography requirement

Use multiple independently calibrated preparation, transition, and measurement families:

\[
p(b\mid a,u)
=
\operatorname{Tr}\!\left[E_b\,T_u(\rho_a)\right]
\]

for quantum apparatus, or the appropriate classical likelihood. A tomographically complete set constrains the latent model far more strongly than one scalar output. Remaining transformations that preserve every probability are gauge equivalences and must not be counted as different mechanisms.

### 16.4 Model contest

Before data collection, freeze:

1. the seven state types and their units;
2. every projection and transition equation;
3. allowed free parameters and priors;
4. the noise and measurement model;
5. a standard-physics null model;
6. lower-layer alternatives \(M_1,\ldots,M_6\);
7. the held-out controls and scoring rule;
8. correction for multiple searches;
9. a replication threshold.

Score predictive distributions using held-out log likelihood,

\[
\mathcal L_{\mathrm{test}}(M)
=
\sum_{i\in\mathrm{test}}
\log p_M(y_i\mid u_i),
\]

and penalize excess flexibility using a preregistered Bayesian, minimum-description-length, or information-criterion comparison. A seven-layer model succeeds only if it beats both established physics and all simpler nested alternatives on untouched interventions.

---

## 17. Can seven-dimensional physics be proved?

Not from the current evidence. The following conditional theorem states the strongest operational result.

### Theorem 17.1 (replacement): deterministic closure is idempotent **[M]**

Let \(X\) be finite, let \(\mathcal A=\{T_a:X\to X\}\) be a fixed finite family of deterministic transitions, and let \(\mathcal P\) be an observational partition. Define \(C_{\mathcal A}(\mathcal P)\) as its coarsest refinement on which every \(T_a\) descends to a deterministic quotient map. Then

\[
\boxed{C_{\mathcal A}(C_{\mathcal A}(\mathcal P))=C_{\mathcal A}(\mathcal P).}
\]

**Proof.** Put \(\mathcal Q=C_{\mathcal A}(\mathcal P)\). By definition, all transitions already descend to \(\mathcal Q\). Therefore \(\mathcal Q\) itself is an admissible sufficient refinement of \(\mathcal Q\). No partition strictly finer than \(\mathcal Q\) can be its coarsest admissible refinement, since \(\mathcal Q\) is already admissible. Thus its closure equals itself. Existence in a finite system follows by splitting classes according to their transition-successor classes until stable. \(\square\)

**Correction of version 1.13.** Its premises required each successive partition to be a strict coarsest complete deterministic refinement for the same transition family. After the first complete refinement, idempotence forbids the next strict refinement. Those premises have no such seven-stage instance; the conditional statement could not support a seven-layer physical claim.

Seven finite-horizon stages are possible. Let \(X=\{0,\ldots,127\}\), \(T(x)=\lfloor x/2\rfloor\), and \(y(x)=x\bmod2\). The observation histories \((y(x),y(Tx),\ldots,y(T^kx))\) yield respectively \(2,4,8,16,32,64,128\) classes for \(k=0,\ldots,6\). This is a seven-bit shift register observed for successively longer horizons. Its complete closure already has 128 classes, and closing that partition again leaves 128. Replacing seven bits by \(m\) bits gives \(m\) such levels, so the construction selects no fundamental dimensional count.

A hierarchy can also become stricter when the allowed experiments, laws, or observables change between stages. In that case those changing assumptions must be stated; it is not repeated closure for one fixed family. The new exact finite audit checks 7,662 systems and partitions, and separately checks the shift-register example. These computations support the implementation; the argument above proves general idempotence within its stated scope.

### 17.1 A decisive physical success criterion

A credible claim of seven-layer physics requires all of the following:

- a declared, noncircular meaning of each layer, with an explicit horizon or experiment family for every proposed refinement;
- evidence of predictive sufficiency on a specified tested domain, with the possibility of later refinement left open;
- a compact D7 law fixed before the decisive experiment;
- a numerical prediction unavailable to the lower models;
- successful blinded holdout and independent replication;
- invariance under coordinate changes, sampling rate, apparatus swap, and analysis team;
- consistency with quantum Bell tests, relativity, thermodynamics, and energy conservation;
- an explicit result that would falsify the D7 model.

Until those observations exist, “seven-dimensional physics” is a research hypothesis, not a proved fact.

*Revision 1.70 note.* Appendix BS supplies, for the narrower claim of one spatial dimension above the third, the items this list demands and the manuscript previously lacked: a compact law fixed before the experiment (H-Inv), a numerical prediction unavailable to the lower model (a barrier-independent racemization floor), an explicit falsifier (no floor, giving a bound on \(\Gamma_4\)), and a certificate whose four conditions are invariant under coordinates, sampling, and apparatus. It does not supply the seven layers, and it does not claim the dimension exists.

---

## 18. Optional Christian metaphysical interpretation

This section is **T**, not M or E.

### 18.1 God as ground, not coordinate

Let \(G\) symbolize the proposed source and sustainer of the created structure. The intended statement is not

\[
G\in X_8.
\]

Instead, the metaphysical claim is that every \(X_d\), law, and projection depends on \(G\), while \(G\) is not an object measured by an instrument inside one \(X_d\).

The symbols \(0\), \(1\), a loop, Alpha, and Omega may express beginning, completion, or unity, but they are analogies. Numerically identifying God with a bit or circle would reduce the proposed ground of mathematics to one mathematical object.

### 18.2 Trinity guardrail

Let \(\mathsf F,\mathsf S,\mathsf H\) denote Father, Son, and Holy Spirit. A logical guardrail is

\[
\mathsf F\ne\mathsf S,
\qquad
\mathsf F\ne\mathsf H,
\qquad
\mathsf S\ne\mathsf H,
\]

while

\[
\operatorname{Ess}(\mathsf F)
=
\operatorname{Ess}(\mathsf S)
=
\operatorname{Ess}(\mathsf H)
=
\mathsf G.
\]

The Father may be analogized to transcendent source, the Son to divine self-expression within creation, and the Spirit to personal divine presence and participation. These are analogies, not a reduction of three persons to three computational functions.

Humanity as *imago Dei* is represented as created likeness, relationship, reason, agency, love, creativity, and stewardship—not as a fragment of divine essence.

### 18.3 Free choice and authorization

The engine represents a gated action by

\[
K=C_G\land P_A\land C_W\land A_G,
\]

where:

- \(C_G\): declared free alignment toward \(G\);
- \(P_A\): declared preservation of agency;
- \(C_W\): global-coherence constraints pass;
- \(A_G\): external authorization is supplied.

Then

\[
K=1\iff C_G=P_A=C_W=A_G=1.
\]

Most importantly,

\[
\boxed{C_G=1\not\Rightarrow A_G=1.}
\]

Choosing God can be modeled as consent or participation, not as possession of an unrestricted key to physical reality. The software cannot observe divine will, prove genuine free will, or manufacture authorization.

### 18.4 Miracles

If a lower autonomous law predicts \(L\) but an event arises from a globally posited action not generated by \(L\), the lower observer might see

\[
\pi(T_Gx)\ne L(\pi x).
\]

This is a coherent mathematical analogy for a miracle as globally ordered yet not generated by the lower effective law. It is not an experimental method for labeling anomalies divine. An anomalous residual must first face ordinary calibration, noise, missing variables, and model revision.

### 18.5 Biblical time

The optional convention

\[
1\text{ divine day}=1000\text{ human years}
\]

gives

\[
7\text{ divine days}=7000\text{ human years}.
\]

Under that hypothesis, an 80-year human life corresponds to \(0.08\) divine days, or \(1.92\) divine hours. This is internally consistent arithmetic. The wording of 2 Peter 3:8 is analogical and reciprocal—one day is *as* a thousand years and a thousand years *as* one day—so it does not independently establish a physical conversion constant.

A mean Gregorian year is not divisible by seven days:

\[
365.2425=52\cdot7+1.2425.
\]

The complete 400-year Gregorian cycle is

\[
146097=20871\cdot7
\]

days by calendar construction. Twenty-four hours follows Earth's rotation and timekeeping conventions, not a fundamental seven-dimensional equation.

Soul persistence may be postulated through an invariant higher state, but cyclic time alone does not prove immortality. Periodicity,

\[
s(\tau+T)=s(\tau),
\]

is different from immutability,

\[
\frac{ds}{d\tau}=0.
\]

---

### 18.6 Scripture as a finite observer

Appendix BR carries out an exact audit of Job 38, Genesis 1, and the sanctuary geometry, a lexical study of the creation verbs, its formalization in Section 2's objects, and an audit of the "Hebrew as code" claims. Its results are T readings and M counts on the text. None is E, and Theorem BR.1 shows why none can reach H7: a finite text is a finite observer.

---

## 19. Adversarial tests and resolutions

| Attack | Verdict | Resolution or consequence |
|---|---|---|
| Bare numbers have universal dimensional meaning | valid objection | require types, units, charts, and covariance |
| One lower effect reveals one cause | false in general | return a fiber or affine cause class |
| Hidden state is proved whenever a fit improves | false | require same-observation/different-future intervention and model comparison |
| Seven layers are proven because seven was encoded | circular | seven is a hypothesis until strict predictive refinements are measured |
| \(D_7\) is mathematically the highest possible dimension | false | it is terminal only under H7; algebra can define higher indices |
| \(D_8=D_1\) destroys history | valid for a pure loop | preserve winding: \(D_8=(D_1,1)\) |
| A snapshot is dead | false | it lacks displayed change; state can encode momentum and gradients |
| Opposites are required for things to exist | unsupported | contrast may define labels, not ontology |
| Double slit proves consciousness or a fourth spatial grid | false | path information and partial trace explain the operational result |
| A local classical hidden variable can reproduce all quantum data | ruled out for broad classes | any completion must address Bell/contextuality and no-signaling |
| Fixed-point simulation is a physical quantum computer | false | classical memory remains exponential |
| Absolute zero means no motion | false | zero-point motion remains; finite refrigeration cannot reach exact zero |
| An inverse microwave creates negative energy | false | anti-Stokes conversion exports excitations to a sink |
| Resonance or magnets yield free energy | false without an unclosed ledger | measure every input, loss, and stored-energy change |
| An unexplained residual is God | unfalsifiable category error | keep \(G\) metaphysical and residuals empirical |
| The inverse limit is God | category error | \(X_\infty\) is mathematical; theological \(G\) is separately postulated |
| Successful Sycamore holdout proves \(T\) and \(\pi\) | false | it validates only \(g=\pi\circ T\) up to latent gauge |

---

## 20. Computational verification record

*Release note (revision 1.69).* This record describes Reality Bit Engine 0.10.0 as of 4 September 2026. Later releases through 0.20.0 are described in the appendices that introduced them; their assertion totals have not been re-aggregated here, and the header lists the current release.

Reality Bit Engine 0.10.0 has a standard-library-only C++17 core and an optional Python 3.10+ standard-library independent audit. On 4 September 2026 its C++ core passed release compilation with

    -std=c++17 -O2 -Wall -Wextra -Wpedantic -Werror

The changed prize-lab suite also passed a fresh AddressSanitizer/UndefinedBehaviorSanitizer run. The other sanitizer results belong to earlier releases. Eleven current release suites passed a total of

\[
\boxed{1{,}745{,}812\text{ assertions}.}
\]

Coverage includes:

- exact dyadic arithmetic, overflow rejection, and 848-bit serialization;
- coherent seven-layer projections and integrity rejection;
- 1,000 randomized Boolean inverse systems checked against brute force;
- cat coherence, decoherence, slit interference, and duality at 1,000,001 visibility values;
- 65,536 Phase64/CORDIC angle checks;
- Bell, CHSH, Deutsch, Grover, teleportation, error correction, and randomized inverse circuits;
- thermal, sideband-cooling, reset, aliasing, magnetic, energy-ledger, and Pareto calculations;
- exhaustive finite invariant and cellular-automaton results;
- 200 random projective-law refinements checked against independent trajectory comparison;
- dimension-address round trips for indices 1–10,000 and cycle lengths 1–31;
- exhaustive finite map twist classifications through five-state spaces;
- frozen parsing, leakage protection, fitting, and holdout evaluation of 33 measured Sycamore points.
- Lorentz invariance, Schwarzschild observables, and numerical Einstein tensors;
- Lagarias truth-distance scans, exact SAT error landscapes, and analytic
  Taylor-Green and three-dimensional ABC Navier-Stokes families.

This verifies implementation behavior against the specification. It does not verify H7 or the theological interpretation as facts of nature.

---

## 21. Compact formal statement

The complete defensible core is

\[
\boxed{
\begin{aligned}
&Z=(D_1,\ldots,D_7),\\
&\pi_{d,d}=\operatorname{id},\qquad
\pi_{e,d}\pi_{f,e}=\pi_{f,d},\\
&F_y=\pi^{-1}(y),\\
&\pi T=L\pi
\iff
\pi(a)=\pi(b)\Rightarrow\pi(Ta)=\pi(Tb),\\
&a\sim_\infty b
\iff
\pi(T^ka)=\pi(T^kb)\ \forall k\ge0,\\
&b_{\min}=\lceil\log_2s\rceil,\\
&Hz_1=Hz_2
\iff z_1-z_2\in\ker H,\\
&\operatorname{Causes}(y)=z_*+\ker H,\\
&d+kh\ge kd\quad\text{(continuous locally injective encodings)},\\
&n=7q+r,\quad r\in\{1,\ldots,7\},\\
&\operatorname{ULP}(n2^eU)=2^eU,\\
&\text{a lower effect identifies an equivalence class of causes,}\\
&\text{not automatically one cause or one ontology.}
\end{aligned}
}
\]

The compatible mathematical diagram is isomorphic to \(X_7\). Equating it with complete physical reality requires the additional H7 hypothesis. Finite observations cannot establish universal terminal completeness. For specified finite models, the engine calculates predictive distinctions and their information requirements; experimental applications must retain model uncertainty and scope.

---

## 22. Conclusions

The theory's central insight survives rigorous formalization:

> What appears complete at one observational layer may be an equivalence class of richer states. An upper layer confines the law below exactly when its evolution factors through the lower projection.

The framework has produced exact theorems, an executable bit language, quantum and thermodynamic calculations, autonomous finite mathematical discovery, and a successful held-out fit to measured quantum-hardware data. It has also produced an equally important negative result: effects alone generally cannot separate a hidden transition from its observation map.

Hidden-state claims can be made operational: compare reproducibly prepared cases under matched present observations and interventions, test their future distributions, and account for measurement error and alternative models. The minimum missing memory is exactly computable for the specified finite deterministic transition system, not automatically for an unknown physical system. Counting seven refinements does not establish seven physical dimensions. A proposed physical interpretation needs independently defined observables, scoped predictions, model comparison, and replication.

The mathematical core is proved within its assumptions. The software is extensively verified. The measured projection-law prediction succeeded on its declared holdout. Seven-dimensional physics and the theological interpretation remain open hypotheses with explicit standards for progress.

---

## References

1. F. Arute et al., “Quantum supremacy using a programmable superconducting processor,” *Nature* 574, 505–510 (2019). <https://doi.org/10.1038/s41586-019-1666-5>
2. Sycamore experiment data archive, Dryad. <https://doi.org/10.5061/dryad.k6t1rj8>
3. A. Zlokapa, S. Boixo, and D. A. Lidar, “Boundaries of quantum supremacy via random circuit sampling,” *npj Quantum Information* 9 (2023). <https://doi.org/10.1038/s41534-023-00703-x>
4. M. Brune et al., “Observing the Progressive Decoherence of the ‘Meter’ in a Quantum Measurement,” *Physical Review Letters* 77, 4887–4890 (1996). <https://doi.org/10.1103/PhysRevLett.77.4887>
5. B.-G. Englert, “Fringe Visibility and Which-Way Information: An Inequality,” *Physical Review Letters* 77, 2154–2157 (1996). <https://doi.org/10.1103/PhysRevLett.77.2154>
6. S. Dürr, T. Nonn, and G. Rempe, “Origin of Quantum-Mechanical Complementarity Probed by a ‘Which-Way’ Experiment in an Atom Interferometer,” *Physical Review Letters* 81, 5705–5709 (1998). <https://doi.org/10.1103/PhysRevLett.81.5705>
7. B. Hensen et al., “Loophole-free Bell inequality violation using electron spins separated by 1.3 kilometres,” *Nature* 526, 682–686 (2015). <https://doi.org/10.1038/nature15759>
8. L. Masanes and J. Oppenheim, “A general derivation and quantification of the third law of thermodynamics,” *Nature Communications* 8, 14538 (2017). <https://doi.org/10.1038/ncomms14538>
9. J. D. Teufel et al., “Sideband cooling of micromechanical motion to the quantum ground state,” *Nature* 475, 359–363 (2011). <https://doi.org/10.1038/nature10261>
10. X. Jin et al., “Thermal and Residual Excited-State Population in a 3D Transmon Qubit,” *Physical Review Letters* 114, 240501 (2015). <https://doi.org/10.1103/PhysRevLett.114.240501>
11. N. D. Mermin, “Hidden variables and the two theorems of John Bell,” *Reviews of Modern Physics* 65, 803 (1993). <https://doi.org/10.1103/RevModPhys.65.803>
12. M. A. Nielsen and I. L. Chuang, *Quantum Computation and Quantum Information*, Cambridge University Press.
13. J. R. Munkres, *Topology*, for invariance-of-domain and dimension background.
14. Genesis 1:26–27; Exodus 33:20; John 1; 1 Corinthians 12:11; Colossians 1:15; Hebrews 1:3; 1 Timothy 6:16; 2 Peter 3:8–9; Revelation 22. These are interpretive sources for Section 18, not empirical physics references.

---

## Reproducibility companion **[C]**

The accompanying Reality Bit Engine source (0.16.0 when this paragraph was written; 0.20.0 is current, see Appendices M and N) contains the formal bit schema, C++17 implementation, measured dataset, examples, eleven C++ test executables, twelve Python audit test methods, six proof-route test methods, ten standalone-verifier test methods, nine proof-gap test methods, eleven all-time fluid certificate test methods, nine relative-fluid certificate test methods, eleven computed-trajectory test methods, and detailed validation documents. The paper and engine should be read together: the paper states the theory and proof boundaries; the program supplies finite executable witnesses and counterexamples alongside the arguments in PROOF_ATTEMPTS.md, CERTIFICATE_VERIFIER.md, MISSING_ARGUMENTS.md, GLOBAL_FLUID_BOUND.md, NEXT_WALL.md, and TRAJECTORY_PROOF.md.

---

## Appendix A: Relativity integration **[M/C]**

The engine now tests the framework against special and general relativity.
For a four-vector \(X=(ct,\mathbf x)\), it preserves

\[
s^2=(ct)^2-\|\mathbf x\|^2
\]

under arbitrary-direction Lorentz boosts. For energy-momentum
\(P=(E/c,\mathbf p)\), it preserves

\[
m^2=\frac{E^2}{c^4}-\frac{\|\mathbf p\|^2}{c^2}.
\]

For a supplied metric it numerically evaluates Christoffel symbols, Ricci
curvature and

\[
G_{\mu\nu}=R_{\mu\nu}-\frac12g_{\mu\nu}R.
\]

The Schwarzschild exterior demonstration returns maximum numerical Einstein
residual below \(1.8\times10^{-10}\) at the declared point while its
Kretschmann invariant remains nonzero. This correctly distinguishes vacuum
from flatness. It confirms compatibility with Einstein's equations; it does
not yet predict a deviation from them.

---

## Appendix B: Three Millennium laboratories **[C]**

Version 0.9 makes the next universal step explicit for three million-dollar
problems.

| Problem | Engine result | Exact unresolved step |
|---|---|---|
| Riemann Hypothesis | every Lagarias inequality numerically checked through declared \(N\), with a bounded D0 frontier coordinate | rigorous transcendental bounds and proof for all integers |
| \(P\) versus \(NP\) | exact SAT witnesses and complete violated-clause landscapes for bounded formulas | polynomial algorithm or general superpolynomial lower bound |
| Navier–Stokes | analytic smooth 2D Taylor–Green and 3D ABC families | cutoff-uniform regularity or a valid 3D breakdown example |

The projective research targets are:

\[
\Delta(n)\ge B_j>0
\quad\Longrightarrow\quad
B_{j+1}\ge F(B_j,j)>0
\]

for a scale-inductive Riemann envelope;

\[
\operatorname{size}_{\mathcal P}(F_n)\ge2^{\Omega(n)}
\]

for a suitably general proof-complexity bridge; and

\[
\sup_{t\ge0}\|\pi_Ku(t)\|_{H^s}
\le C(u_0,\nu)
\]

with \(C\) independent of spectral cutoff \(K\) for Navier–Stokes.

These are research targets, not completed lemmas. As of this release, the
official status of all three problems remains open.

---

## Appendix C: D0 polarity and frozen errors **[M/C]**

Let $a,b>0$ be the two sides of a comparison. Their complementary D0
coordinates are

\[
d_0=\frac{a}{a+b},
\qquad
d_1=\frac{b}{a+b},
\qquad
d_0+d_1=1.
\]

The bounded oriented coordinate

\[
\boxed{s=d_0-d_1=\frac{a-b}{a+b}}
\]

preserves which side is larger and removes common positive scale. For a claim
$a\le b$, $s<0$ satisfies the claim, $s=0$ lies on its boundary, and
$s>0$ falsifies it. Calling $s$ a coordinate does not make both signs true;
it makes the size and direction of failure explicit.

For Lagarias, $a=\sigma(n)$ and $b=H_n+e^{H_n}\log H_n$. The full configured
scan through five million integers found no positive coordinate. Its closest
normalized nontrivial frontier was

\[
s(12)=-0.00571425357027553127.
\]

For a CNF formula, the discrete coordinate

\[
E_F(a)=\#\{i:C_i(a)=0\}
\]

equals zero exactly on satisfying assignments. Exhaustive enumeration of the
four-pigeons/three-holes formula found minimum distance one with 60 assignments
at that distance and no zero-distance state.

For Navier-Stokes, residual coordinates are typed:

\[
r_{div}=\nabla\cdot u,
\qquad
r_{mom}=\partial_tu+(u\cdot\nabla)u+\nabla p-\nu\Delta u.
\]

They may be combined only after nondimensionalization. Version 0.9 freezes the
largest normalized residual with its $(x,y,z,t)$ address. For the implemented
three-dimensional ABC eigenflow, 179,712 sampled coordinates produced maximum
raw momentum residual $1.1774\times10^{-18}$ and worst dimensionless residual
$7.3537\times10^{-20}$. Analytic substitution proves that family, but not
global regularity for arbitrary three-dimensional data.

Thus the reusable error object is

\[
\boxed{
\mathcal E=(\text{input coordinate},\text{typed residuals},
\text{normalization},\text{distance},\text{evidence status},\text{hash}).
}
\]

An error value can be a correct coordinate on the model's map while the
candidate that generated it remains false. This distinction is the formal
content of “freezing the error.”

---

## Appendix D: Independent falsification and the limits of D0 **[C]**

Version 0.10 tests whether frozen residuals identify actual faults, using an
independent companion and declared synthetic negative controls. The experiment
does not assume every failure to reach zero is an arithmetic mistake.

First, an inequality and an equality have different zero-defect functions:

$$
V_{a\le b}(a,b)=\max(0,a-b),\qquad V_{F=0}(x)=|F(x)|.
$$

The negative signed barycentric slack of a satisfied inequality is not an
error that must be driven to equality. A nonzero defect may reflect
inconsistent constraints, a wrong model, a wrong parameter, rounding, or
discretization. Its cause does not follow from its magnitude alone.

### D.1 Exact finite RH certificate **[C]**

The Python companion computes outward-rounded dyadic intervals in arbitrary
integers at scale 2^128. It encloses harmonic sums and uses nonnegative
exponential/logarithm series with proved geometric tail bounds. This removes
reliance on floating exp/log values when determining the sign for the listed
finite cases.

It certifies every integer from 1 to 4,096 plus the five integers
5,040; 55,440; 720,720; 4,324,320; and 5,000,000. All 4,101 inequalities are
satisfied. Deliberately replacing sigma(12)=28 by 29 is correctly classified
as a violation. Insufficient series resolution returns unresolved precision.

This does not upgrade the entire five-million-case floating scan to an exact
certificate and does not prove the all-integer Lagarias criterion.

### D.2 Independent SAT check **[C]**

A separate implementation enumerates all 4,096 assignments and agrees with
the full C++ violation histogram. All assignments violate at least one clause.
Exactly 60 minimize the cost at one, but none satisfies the formula.
A candidate at distance one remains false; the constraints are inconsistent,
not incorrectly added or multiplied.

### D.3 Fault recovery with a frozen bias **[M/C]**

The C++ program exports only the ABC field values u,v,w,p. An independent
Python implementation estimates derivatives using centered differences.
At 24 training points it fits pressure and viscosity coefficients; it
evaluates them on 40 untouched points at each of three declared step sizes.
The audit variables are nondimensional; the common spatial/temporal numeric
step is specified in these scaled coordinates.

Injected faults multiply pressure by 1.125 and substitute an assumed viscosity
of 0.008 for the generating value 0.01. At step 0.005, the held-out momentum
RMS falls from approximately 0.285102 to 1.55e-13 after fitting.

However, the recovered viscosity is 0.01000008340041921, not exactly 0.01.
This bias shrinks approximately fourfold when the differentiation step is
halved. For this eigenflow the finite-difference operators imply

$$
\nu_h=\frac{h\sinh(\nu k^2h)}{4\sin^2(kh/2)}=\nu+O(h^2).
$$

Therefore a near-zero numerical residual can conceal parameter error.
The fitted parameter has partly absorbed discretization error.

An extra synthetic velocity mode leaves a held-out RMS around 0.113499 even
after fitting. The two-coefficient correction model is insufficient for that
candidate. This failure is retained, not redefined as success at another layer.
Corrupting held-out data changes evaluation while leaving the training fit
unchanged, verifying the declared separation.

### D.4 Residual is not distance to solution **[M]**

The exact counterexample

$$
F(x)=10^{-12}x,\quad x=10^6
$$

has residual 10^-6 but lies 10^6 from its unique zero. A useful inference
from residual to state error requires a stability bound with controlled
constant, such as dist(x,S)<=C||F(x)||.

The revised frozen object therefore includes the constraint type, full input
and residual vector, precision or step size, normalization, evidence status,
source identity, and—where applicable—model-identifiability and refinement
checks. Its purpose is to preserve counterevidence as well as successes.

The complete companion derivation and exact/numerical records are provided in
INDEPENDENT_AUDIT.md and results/residual-audit-v010.json inside the release.
No new physical measurements, seven-dimensional physics proof, or
Millennium solution has been obtained.

## Appendix E. Restricted proofs and rejected universal estimates **[M/C]**

Version 0.11.0 adds written arguments that cover restricted infinite families,
supported by exact arithmetic in `tools/proof_routes.py`. This goes beyond
extending a finite numerical scan, but it does not resolve any of the three
Millennium targets. Full proofs are in `PROOF_ATTEMPTS.md` in the engine archive.
No novelty or proof-assistant verification is claimed.

### E.1 A restricted Lagarias theorem **[M/C]**

For every positive integer with at most two distinct prime factors,

\[
\sigma(n)\le H_n+e^{H_n}\log H_n.
\]

The argument splits at 1,024. Below the cutoff, all 1,023 integers are checked
with outward-rounded integer intervals. Above it, the prime-factor product
gives \(\sigma(n)<3n\). Convexity of \(1/x\) gives
\(H_n>\log n+1/2\), and therefore \(e^{H_n}>3n/2\).
The exact rational series bounds

\[
\log2>\frac{842}{1215}>\frac{69}{100},
\qquad
e^2<\frac{134800213}{18243225}<\frac{37}{5}
\]

imply \(H_n>e^2\) for \(n\ge1024\). Thus \(\log H_n>2\), making the right-hand
side strictly larger than \(3n\). This proves the restricted theorem.

It does not cover integers with arbitrarily many distinct prime factors.
The unrestricted Lagarias statement is equivalent to RH; the restriction is
therefore the unresolved quantifier, not a numerical rounding issue.

### E.2 All-size pigeonhole error formulas **[M/C]**

For the engine's pairwise pigeonhole CNF, with \(p\) pigeons and \(h\) holes,
the exact minimum number of violated clauses is

\[
\min_x E(x)=\max(0,p-h).
\]

For \(p\ge h\), the number of minimizing assignments is

\[
N_{\min}(p,h)=\sum_{d=0}^{\min(h,p-h)}
\binom hd\frac{p!}{(p-h-d)!2^d}.
\]

For \(p<h\), it is \(h!/(h-p)!\). An exact decomposition of clause cost into
\(p-h\) plus nonnegative incidence-count terms proves the lower bound. The
equality conditions give the count: all holes are occupied, no pigeon is
assigned twice, and each hole contains one or two pigeons. This explains the
previous minimum of one with 60 minimizing assignments at \(p=4,h=3\).

The new companion checked the formulas against direct clause enumeration and
an independent graph decomposition on 112,938 assignments across 21 sizes.
The written proof covers all positive sizes of this encoding. It supplies no
algorithm or lower bound for unrestricted SAT and does not resolve P versus NP.

### E.3 Exact fluid counterexamples to proposed estimates **[M]**

On the periodic three-torus of side \(2\pi\), let energy and enstrophy use
normalized spatial means. The globally smooth exact family

\[
u_N=e^{-\nu N^2t}(\sin Nz,\cos Nz,0)
\]

has \(E(0)=1/2\) and \(Z(0)=N^2/2\). Thus initial energy alone cannot give a
finite, initial-data-independent upper bound on enstrophy. Bounds depending
on further initial norms or positive time remain possible.

A second proposed invariant is \(L_\lambda=E+\lambda Z\) with fixed
\(\lambda>0\). For smooth divergence-free initial data

\[
u_0=A(\sin y,0,\cos x-\cos(x+y)),
\]

the exact initial rates are

\[
E'(0)=-2\nu A^2,\qquad
Z'(0)=\frac{A^3}{4}-3\nu A^2,
\]

and hence

\[
L_\lambda'(0)=\frac{\lambda A^3}{4}-(2+3\lambda)\nu A^2.
\]

For any fixed positive \(\lambda,\nu\), choosing
\(A>4\nu(2+3\lambda)/\lambda\) makes this positive. Therefore no such fixed
positive weighted sum is universally nonincreasing. At
\(A=1,\nu=1/100,\lambda=1\), the exact rates are \(-1/50\), \(11/50\), and
\(1/5\) respectively. Sparse Fourier algebra independently obtains them using
rational real and imaginary coefficients, without a numerical grid.

This rejects a particular sufficient-estimate route. Instantaneous enstrophy
growth is not finite-time blowup; no Navier–Stokes breakdown example or
general global-smoothness proof is supplied.

### E.4 The meaning of a frozen failure **[M/C]**

An injective coordinate change \(f\) with \(f(0)=0\) satisfies
\(f(r)=0\iff r=0\). It cannot turn a false equality into a true one. A positive
minimum SAT violation can represent inconsistent constraints, not a wrongly
computed factor. The engine therefore freezes counterexamples to proposed
lemmas with their exact assumptions and retains them as rejected proof routes.

The six new test methods and the complete exact-arithmetic companion run
passed. The frozen records, exact RH base-case endpoints, SAT histograms,
Fourier identities, and source/document SHA-256 values are in
`results/proof-routes-v1.json`. The retained v0.10 numerical suite remains
historical; it is not relabeled as a fresh v0.11 run. All three Millennium
problems remain unresolved by this work.

## Appendix F. Independent certificates and exact counting projections **[C]**

Version 0.12.0 adds a standalone Python verifier with no imports from the
engine or the proof-result generator. It accepts a narrowly scoped portable
certificate and recomputes its evidence using different methods. It does not
accept a producer's pass flag or an unsupported broader theorem claim.

### F.1 Rational lower witnesses **[M/C]**

For each of the 1,023 RH base cases, the verifier independently computes the
exact harmonic sum and divisor sum. It checks a portable lower coordinate
\(h=a/2^{24}\) satisfying \(h\le H_n<h+2^{-24}\). With
\(z=(h-1)/(h+1)\), positive-term series imply

\[
H_n+e^{H_n}\log H_n\ge
h+\left(\sum_{j=0}^{40}\frac{h^j}{j!}\right)
\left(2\sum_{j=0}^{31}\frac{z^{2j+1}}{2j+1}\right).
\]

The right side is rational. It strictly exceeds the exact divisor sum for
every tested \(n>1\), with equality at 1. This calculation shares neither the
producer's rounded transcendental interval code nor its divisor-pair summation.
The analytic and number-theoretic steps extending the result to integers with
at most two distinct prime factors remain written proofs, not formalized
proof terms.

### F.2 A counting law that survives projection **[M/C]**

For the specified pairwise pigeonhole CNF, let \(c\) record the occupancies
of the holes after some rows have been chosen, and let \(e\) record total
clause cost. A new row chooses a subset \(M\) of holes and adds

\[
\Delta(c,M)=\mathbf1_{M=\varnothing}+\binom{|M|}{2}
+\sum_{j\in M}c_j.
\]

Project through \(\alpha(c,e)=(\operatorname{sort}(c),e)\). For every two
occupancy vectors related by a hole permutation, that permutation gives a
bijection between row subsets. Corresponding subsets give the same cost
increment and sorted successor. Their multisets of projected transitions
therefore agree.

Consequently the exact row-counting recurrence is

\[
F_{r+1}\left(\operatorname{sort}(c+\mathbf1_M),e+\Delta(c,M)\right)
\mathrel{+}=F_r(c,e),
\]

starting from the unique empty assignment and summing over all subsets.
Induction on rows proves that its final histogram counts every labeled
Boolean assignment once. For counting transfer operators,
\(\alpha_*T=\bar T\alpha_*\). This is an exact projection rule under the
stated symmetry; it is not a descent result for arbitrary individually fixed
labeled updates or asymmetric rules.

The cost coordinate is necessary. The two assignments with rows
\(((1,0),(0,1))\) and \(((1,1),(0,0))\) have the same column occupancies but
costs 0 and 2. Dropping the cost would identify states that the requested
histogram must distinguish.

The recurrence recovered all 21 earlier histograms, representing 112,938
assignments, using 4,632 row transitions and at most 215 intermediate states.
This is a count of different computational operations, not a measured speed
ratio or a general polynomial-time SAT claim.

### F.3 Independent fluid algebra **[C]**

The checker expands \(\cos(x+y)\) in real sine/cosine variables and evaluates
polynomial moments with exact rational coefficients. It computes enstrophy
production from vortex stretching rather than the producer's differentiated
transport expression. It reproduces the exact rates \(E'=-1/50\) and
\(Z'=11/50\), along with the documented weighted and high-frequency examples.
These are checks of the specified smooth periodic fields, not certificates
for arbitrary fluid evolutions.

### F.4 Rejection tests and remaining trust **[C]**

The valid portable certificate passed. All 26 deliberate corruptions were
rejected, including altered arithmetic, incomplete coverage, forged broader
claims, changed physical assumptions, and malformed numeric representations.
Ten new test methods passed. The checker was also run in isolation with only
its source and the certificate present, and rejection remained active under
Python's `-O` option.

These checks increase independence from the producer, but the verifier's own
mathematics and implementation remain trusted. No Lean or other established
proof kernel checked these arguments. A valid formal proof would also need a
faithful formal statement of the intended claim. The complete theory is not
made irrefutable by finite tests or by certificate acceptance.

The detailed derivations, trust boundary, and Windows/Python reproduction
instructions are in `CERTIFICATE_VERIFIER.md`. The frozen data are
`results/portable-certificate-v1.json`, `results/verifier-v012.json`, and
`results/certificate-challenges-v012.json`. Earlier proof-route source and
document fingerprints were preserved unchanged. This release makes no new
claim of physical validation or resolution of a Millennium problem.

## Appendix G. Arguments at the remaining proof gaps **[M/C]**

Version 0.13.0 supplies further written reductions and obstructions in
`MISSING_ARGUMENTS.md`, supported by `tools/proof_gap_lab.py`. No mathematical
novelty or formal-kernel verification is claimed. The full Millennium problems
remain unresolved by this work.

### G.1 Removing the prime-count restriction from a bounded RH verification **[M/C, finite coverage to 1e24 conditional on unpublished exact-arithmetic run]**

Put \(G(h)=h+e^h\log h\), \(B_n=G(H_n)/n\), and \(R_n=\sigma(n)/n\).
The new discrete argument proves \(B_{n+1}>B_n\) for every \(n\). For
\(n\ge64\), integral harmonic bounds and \(n>2(1+\log n)^2\) imply

\[
nG'(H_n)-(n+1)G(H_n)>0.
\]

Convexity of \(G\) then gives the desired adjacent inequality. The 63
remaining adjacent cases and the logarithm constant used in the analytic
argument are checked with exact outward-rounded intervals.

Any integer can be replaced by a no-larger product of consecutive primes
with nonincreasing positive exponents, while increasing or preserving its
divisor ratio. Missing smaller primes can replace larger ones. Increasing
exponent pairs can be exchanged because \(S_b(r)/S_a(r)\) is increasing for
\(b>a\), where \(S_a(r)=1+r+\cdots+r^a\). Strict integer descent terminates
at a canonical profile.

Consequently it suffices, below any fixed cutoff, to check successive record
divisor ratios among all canonical profiles. For each \(n\) there is a retained
record \(r\le n\) with \(R_r\ge R_n\), so

\[
R_n\le R_r\le B_r\le B_n.
\]

The associated reduction to superabundant numbers is already in the
literature, including [MacArevey's February 2026 preprint](https://arxiv.org/html/2602.15905v2).
It is incorporated here with a discrete proof and executable finite coverage,
not claimed as a discovery of this framework.

The run generated 172,513 canonical profiles through \(10^{24}\) and certified
all 148 record cases. Under these written reductions and correct execution
of the exact-arithmetic code, this verifies the Lagarias inequality for every
integer \(1\le n\le10^{24}\). It is a finite computer-assisted verification,
not RH and not a claimed verification record.

Large harmonic indices use the rigorous integral bracket

\[
H_M+\log\frac{n+1}{M+1}\le H_n\le H_M+\log\frac nM,
\qquad M=4096.
\]

The computation uses 192-bit integer ticks, 256-term interval bounds, and
rational logarithm range reduction. Its record extraction agrees with a
separate full divisor sieve through 100,000. The all-integer finite coverage
depends on the written enumeration and reduction proofs, not on that smaller
comparison alone. The previous standalone verifier was not extended to
formally check this new argument.

### G.2 A lower-bound extension that cannot work **[M]**

Consider the proposed sufficient bound

\[
L_c(n)=\log n+c+n e^c\log(\log n+c).
\]

For every fixed \(c<\gamma\), it fails to dominate \(\sigma(n)\) infinitely
often. Indeed,

\[
\frac{L_c(n)}{n\log\log n}\longrightarrow e^c,
\qquad
\limsup_{n\to\infty}\frac{\sigma(n)}{n\log\log n}=e^\gamma>e^c,
\]

where the second statement is Gronwall's theorem, stated in
[Lagarias's paper](https://arxiv.org/html/math/0008177v2). The code also certifies
a concrete failure at \(c=1/2,n=12\), while the actual Lagarias inequality
there remains satisfied. Increasing precision in a fixed constant below
\(\gamma\) cannot supply the missing global argument.

### G.3 An exact continuation-summary bound **[M]**

Suppose an evaluator must recover the equality relation on \(n\)-bit strings
from a deterministic summary of its first argument:

\[
g(\alpha(x),y)=\mathbf1_{x=y}\quad\text{for every }x,y.
\]

Then \(\alpha\) must be injective. If \(x\ne x'\) share a summary, choosing
\(y=x\) requires the same evaluator input to produce two different answers.
Thus at least \(2^n\) summary states, or \(n\) fixed-width bits, are necessary.
For \(d\) coordinates with \(b\) bits each, the condition is \(db\ge n\).

This refutes a particular fixed-capacity lossless summary strategy. Equality
itself is easy to compute, and this bound is not a general SAT time lower
bound. Twelve explicit collision witnesses accompany the all-size proof.
P versus NP still requires an argument about general algorithms, not only
this architecture.

### G.4 Why the quadratic fluid strategy fails more generally **[M]**

For a differentiable homogeneous quadratic functional, amplitude scaling of
Navier–Stokes initial data gives

\[
\left.\frac{d}{dt}Q(Av)\right|_{t=0}
=A^3C_Q(v)-\nu A^2D_Q(v).
\]

If the cubic coefficient is nonzero for some \(v\), choose the amplitude's
sign and size so the cubic term is positive and dominates. Universal
nonincrease at every amplitude therefore requires the inviscid cubic
production to vanish identically. This is a necessary condition for that
strategy, not for every useful regularity estimate.

For \(u_0=A(\sin y,0,\cos x-\cos(x+y))\), a Fourier multiplier with weights
\(m_1,m_2\) at squared frequencies 1 and 2 has the exact initial derivative

\[
Q'(0)=\frac{A^3}{4}(m_2-m_1)-\nu A^2(m_1+m_2).
\]

For homogeneous Sobolev quadratic energy of any real order \(s>0\), use
\(m_1=1,m_2=2^s\). Choosing
\(A>4\nu(1+2^s)/(2^s-1)\) makes the derivative positive. Eight integer-order
Fourier probes and sign controls check the implementation; the displayed
algebra proves the statement for every positive real order. Initial norm
growth is not a singularity, and a continuation-strength bound remains
missing.

### G.5 Evidence boundary **[C]**

All nine new test methods passed. The full finite run, exact record margins,
monotonicity base cases, failed lower bound, summary collisions, and weighted
fluid probes are frozen in `results/proof-gaps-v013.json`. The supplementary
proofs identify which proposed arguments are valid, which are false, and
which unbounded steps remain unproved. Neither finite coverage nor a new
coordinate interpretation supplies those remaining steps automatically.

## Appendix H. A computable sufficient condition for all-time fluid smoothness **[M/C]**

Version 0.14.0 adds a positive conditional result for the full unforced
three-dimensional Navier–Stokes equation on the fixed torus \([0,2\pi]^3\).
The mathematical method is established Fourier small-data theory. It is
related to the periodic global existence and analyticity result of
[Ambrose, Lopes Filho, and Nussenzveig Lopes](https://arxiv.org/abs/2205.12383).
The new contribution to this project is an explicit written sufficient
estimate and an exact initial-data checker; mathematical novelty is not
claimed.

### H.1 The conditional all-time statement **[M]**

Write \(u(x,t)=\sum_k u_k(t)e^{ik\cdot x}\), with real, smooth,
mean-zero, divergence-free initial data. For viscosity \(\nu>0\), define

\[
S_0=\sum_{k\ne0}\frac{\|u_k(0)\|_2}{|k|}.
\]

If an exact upper bound satisfies \(S_0\le B<\nu\), set
\(\delta=(\nu-B)/2\). Then the solution remains smooth for all \(t\ge0\)
and satisfies

\[
\boxed{
\sum_{k\ne0}e^{\delta t|k|}\frac{\|u_k(t)\|_2}{|k|}
\le S_0e^{-\delta t}\le Be^{-\delta t}.
}
\]

At positive times this controls every spatial derivative and gives spatial
analyticity with radius at least \(\delta t\). The statement concerns every
future time of the full equation, including modes generated by nonlinear
interactions. It does not assume a finite evolving spectrum.

### H.2 The estimate and its continuation argument **[M]**

Incompressibility and the norm-one Leray projection give

\[
\|N_k\|_2\le |k|\sum_{p+q=k}\|u_p\|_2\|u_q\|_2
\]

for the projected nonlinear term. To avoid assuming analyticity in advance,
use capped weights \(w_K(k,t)=e^{\delta t\min(|k|,K)}\) and define
\(A_{j,K}=\sum_{k\ne0}|k|^jw_K\|u_k\|_2\). Submultiplicativity of the
weights and Cauchy–Schwarz imply

\[
D^+A_{-1,K}\le-\nu A_{1,K}+A_{0,K}^2+\delta A_{0,K}
\le-(\nu-\delta-A_{-1,K})A_{1,K}.
\]

Here \(A_{0,K}^2\le A_{-1,K}A_{1,K}\), and the torus's nonzero
frequencies satisfy \(|k|\ge1\). A barrier argument preserves
\(A_{-1,K}\le B\), leaving a dissipation margin of at least \(\delta\).
Thus \(A_{-1,K}(t)\le S_0e^{-\delta t}\). Monotone convergence as
\(K\to\infty\) supplies the uncapped exponential estimate.

For a multiindex of order \(m\), it follows that

\[
\|\partial^\alpha u(t)\|_\infty
\le\left(\frac{m+1}{e\delta t}\right)^{m+1}Be^{-\delta t}
\qquad(t>0).
\]

If a maximal smooth time were finite, these bounds would control all
spatial derivatives on an interval ending at that time and bounded away
from zero. Standard smooth local existence and continuation then extend
the solution, giving a contradiction. Those standard analytic theorems
are explicit dependencies of the proof, not Python operations. The full
argument and normalization are in `GLOBAL_FLUID_BOUND.md`.

### H.3 What the computer verifies **[C]**

The input is an exactly specified finite rational Fourier polynomial.
Unlisted initial coefficients must be exactly zero; a measured truncation
with an unknown tail does not qualify. The checker verifies all reality,
mean, divergence, viscosity, and schema assumptions using exact arithmetic.

Each term of \(S_0\) is the square root of a nonnegative rational number.
Integer square roots produce lower and upper enclosures at chosen
increment \(2^{-p}\), with rational roots evaluated exactly. Summed bounds
\(L\le S_0\le B\) determine three possible valid-input outcomes:

- \(B<\nu\): the sufficient theorem applies;
- \(L\ge\nu\): this sufficient condition fails, without a blowup conclusion;
- \(L<\nu\le B\): precision is insufficient to decide the threshold.

All eleven new test methods passed. Six exact fields produce seven expected
outcomes when one borderline field is checked at both 8 and 96 bits. The
frozen report is `results/global-fluid-v014.json`. It explicitly disclaims
formal-kernel verification and any Millennium solution.

### H.4 The boundary that remains **[M/C]**

For the interacting field
\(u_0=10^{-3}(\sin y,0,\cos x-\cos(x+y))\), the initial norm is
\(10^{-3}(2+1/\sqrt2)<1/100\), so the theorem certifies global smoothness
at viscosity \(1/100\). Increasing the amplitude to one violates this
sufficient hypothesis; the checker then gives no regularity conclusion.

Failure of strict smallness is not a singularity certificate. The exact
shear \(u=(\nu e^{-\nu t}\sin y,0,0)\) is globally smooth even though its
initial norm equals \(\nu\), exactly at the rejected threshold.

The general large-data gap is the absence of a bound controlling the full
evolution when the dissipative coefficient in the displayed estimate loses
its sign. Additional arithmetic precision cannot change that mathematical
obstruction for an exact input above threshold. This restricted fluid
result also does not close the all-integer RH argument, the general
algorithmic P-versus-NP argument, or the empirical gap between a seven-layer
representation and physical reality.

## Appendix I. Relative error control beyond the total-smallness condition **[M/C]**

Version 0.15.0 implements a sufficient stability theorem around an exact
decaying shear. It certifies a concrete initial field outside Appendix H's
total-smallness condition. This is an expansion of the engine's implemented
coverage, with no claim of mathematical novelty. Error estimates around
approximate solutions are an established approach; see
[Morosi and Pizzocchero](https://arxiv.org/abs/1104.3832).

### I.1 A reference trajectory and its full error **[M]**

For the same unforced periodic equation and Fourier normalization, write
\(u=v+w\), where \(v\) is a smooth, real, mean-zero, divergence-free reference.
Define \(\|f\|_{\chi^j}=\sum_{k\ne0}|k|^j\|\widehat f(k)\|_2\), and let

\[
E=\|w\|_{\chi^{-1}},\quad E_1=\|w\|_{\chi^1},\quad
V=\|v\|_{\chi^0},\quad
R=\|\partial_tv-\nu\Delta v+P[(v\cdot\nabla)v]\|_{\chi^{-1}}.
\]

The Fourier bilinear estimate, Cauchy–Schwarz, and Young's inequality give

\[
D^+E+\frac\nu4 E_1\le\frac{4V^2}{\nu}E+R,
\qquad\text{provided }E\le\nu/2.
\]

All nonlinear interactions of the error are retained in deriving this
estimate. A rigorously bounded scalar comparison function can preserve the
error threshold and control the integrated gradient of the full solution.
Standard high-order Sobolev estimates and local continuation then justify
existence throughout the certified interval. The complete derivation is in
`NEXT_WALL.md`.

### I.2 An all-time specialization the engine computes **[M/C]**

For \(v=(Ae^{-\nu t}\sin y,0,0)\), the reference is an exact solution,
so \(R=0\). If

\[
\boxed{E(0)e^{2(A/\nu)^2}\le M<\nu/2,}
\]

then a barrier and integration argument give

\[
\|u(t)-v(t)\|_{\chi^{-1}}\le Me^{-\nu t/4},\qquad
\int_0^\infty\|u-v\|_{\chi^1}\,dt
\le\frac4\nu\left(E(0)+2M(A/\nu)^2\right).
\]

The finite integrated gradient bound prevents finite-time smooth breakdown.
The implemented checker supports \(|A|/\nu\le8\); the theorem itself permits
larger amplitudes with smaller perturbations. The engine uses exact
rational square-root bounds and outward dyadic exponential enclosures.

The included example, with \(\nu=1\) and \(\varepsilon=10^{-5}\), is

\[
u(x,0)=(2\sin y+\varepsilon\sin z,\ \varepsilon\sin(2x),\
         \varepsilon\sin y).
\]

It has exact zero divergence, three independent spatial Fourier directions,
and nonzero projected nonlinear interactions. Its total initial norm exceeds
two, violating the previous sufficient threshold of one. Relative to the
known shear, however, \(E(0)=1/40000\) and
\(E(0)e^8\approx0.074524<1/2\), so the new theorem applies.

At \(T=4\), the certified full norm is below \(0.065\). The smooth solution
then satisfies the previous small-data theorem and its analyticity bound,
with the time origin shifted to \(T\). This invokes the written theorem on
the full evolving field; it does not presume a finite spectrum at \(T\).

### I.3 The next engineering prerequisite **[C]**

Nine new test methods and six expected example outcomes passed. Independent
Fourier convolution caught an omitted quadratic perturbation coefficient in
an initial hand calculation; the corrected expression and rejected value
are preserved in `results/relative-fluid-v015.json`. The general error
estimate already included that interaction. No formal proof kernel has
checked the analysis.

The next implementation target is a general reference trajectory with
certified continuous-time residual, all omitted Fourier output modes, and
a rigorously enclosed scalar comparison solution. If it controls the error
up to a time at which the full norm becomes small, it can connect initially
larger data to Appendix H. Finite-time samples alone cannot establish these
bounds. This proposed general validator is not yet implemented in v0.15.

The useful measure of progress is a stronger valid certificate on a frozen
previously unresolved input, followed by a clearly stated extension to an
input class. Neither this restricted stability theorem nor its coordinate
representation supplies arbitrary large-data regularity, an all-integer RH
proof, or a general complexity separation.

## Appendix J. Computed references with continuous-time and full-mode certificates **[C, conditional on unpublished certificate]**

Version 0.16.0 implements the three engineering components proposed in
Appendix I: bounds on the full residual throughout time intervals, a
certified error-envelope constructor, and independent checking with
counterexample searches. The supported representation is finite Fourier
support and piecewise polynomial time dependence. The written theorem and
exact certificate format are in `TRAJECTORY_PROOF.md`.

### J.1 Continuous-time polynomial bounds **[M]**

On each interval, use normalized time \(\theta\in[0,1]\). For a vector
polynomial with Bernstein expansion

\[
p(\theta)=\sum_{i=0}^d b_i\binom di\theta^i(1-\theta)^{d-i},
\]

nonnegativity and partition of unity imply

\[
\|p(\theta)\|_2\le\max_i\|b_i\|_2
\quad\text{for every }\theta\in[0,1].
\]

Rational squared-norm witnesses certify the control-vector bounds.
The verifier reconstructs the controls by a different algorithm from the
producer. These finite witnesses cover the entire interval by the stated
convexity argument; they are not observations at a finite list of times.

### J.2 Every omitted output mode of the reference residual **[M/C]**

For reference support \(K\), the full projected residual

\[
r=\partial_tv-\nu\Delta v+P[(v\cdot\nabla)v]
\]

has support in \(K\cup(K+K)\). Both implementations reconstruct every
nonzero residual mode there, including those outside the reference support.
One uses the ordered convection expression, while the other uses divergence
of the velocity tensor. Exact incompressibility makes these expressions
equivalent. The independent checker rejects missing residual mode bounds.

The residual vanishes outside this finite union because the reference is
finite. The true solution may have modes arbitrarily far beyond it; those
are covered by the full relative-error estimate, not set to zero.

### J.3 An affine supersolution checked by algebra **[M]**

Given interval bounds \(V\ge\|v\|_{\chi^0}\) and
\(R\ge\|r\|_{\chi^{-1}}\), set \(b=4V^2/\nu-\nu/4\). On an error-envelope
segment of physical duration \(\Delta\), the proposed affine function joins
\(q_L\) and \(q_R\). The checker verifies

\[
\frac{q_R-q_L}{\Delta}\ge bq_L+R,
\qquad
\frac{q_R-q_L}{\Delta}\ge bq_R+R,
\qquad 0\le q_L,q_R<\nu/2.
\]

Since the right-hand side is affine in time, these inequalities cover every
interior time. Joined with the certified initial error and continuous
reference/envelope joins, comparison gives
\(\|u-v\|_{\chi^{-1}}\le q<\nu/2\). Integrating the stronger error inequality
controls the velocity gradient in time; standard local existence and Sobolev
continuation exclude a smooth breakdown during the certified interval.

The final checked inequality

\[
\|v(T)\|_{\chi^{-1}}^{\rm upper}+q(T)<\nu
\]

invokes Appendix H's all-time theorem. The evolved exact field need not have
finite support. The reference may have derivative jumps between polynomial
intervals; its checked continuity avoids impulses, and the differential
estimates integrate across the finitely many joins.

### J.4 Executed evidence and its limits **[C, conditional on unpublished certificate]**

The main run uses viscosity one and the same form of interacting initial
field as Appendix I, now with perturbation amplitude \(10^{-7}\). Its total
initial norm exceeds two. The computed reference uses six signed Fourier
wavevectors, 32 polynomial intervals of degree six, and 158 affine envelope
segments, reaching time four. For this support the retained Galerkin
convection vanishes, while the full residual is nonzero and includes omitted
nonlinear output modes.

The independent checker accepted the entire certificate. It covers 384
omitted-mode occurrences across intervals, bounds the error below
0.000045 throughout, and gives a full norm bound below 0.036664 at time
four. This initial field was already inside the prior shear-perturbation
theorem; the new result is a different implemented validation mechanism,
not a claim of a previously unknown regularity class.

All eleven new test methods passed. All seventeen deliberate corruptions
were rejected, including omitted residual modes and a perturbation hidden
from endpoint values and derivatives. A search over small polynomials found
an exact counterexample to the endpoint-only maximum shortcut. A second
counterexample has zero retained-mode residual but nonzero full residual.

The bounded refinement search retained a failed degree-one/eight-interval
case, then accepted and independently verified a degree-four/sixteen-interval
case. It stopped at that success; it makes no global optimality claim.
The checker also passed an isolated run with only its source and certificate
present and rejected a corruption under optimized Python.

These methods belong to established approximate-solution validation; see
[Morosi and Pizzocchero](https://arxiv.org/abs/1104.3832). The analysis has not
been formalized in a proof assistant. Arbitrary large-data regularity, the
unbounded RH step, general P versus NP, and empirical seven-dimensional
physics remain unresolved by this engine.

## Appendix K. A sharper error estimate that certifies a previously unresolved input **[M/C, conditional on unpublished certificate]**

**Engine addition:** 0.17.0. This appendix describes a concrete expansion of
the engine's certificate coverage. The complete written argument, exact
records, and PC commands are in `SHARP_TRAJECTORY_BOUND.md` in the archive.
It is not a claim of mathematical novelty or of a solved Millennium problem.

### K.1 The same input before and after the changes **[C]**

On the fixed periodic torus, with viscosity one, take

\[
u_0=(2\sin y+\varepsilon\sin z,\varepsilon\sin 2x,
\varepsilon\sin y),\qquad\varepsilon=1/10.
\]

The exact data are real, mean-zero and divergence-free. Their initial
\(\chi^{-1}\) norm is \(\sqrt{4+\varepsilon^2}+3\varepsilon/2>2\),
outside the total-smallness condition. The older exact-shear sufficient
certificate also fails on this input.

The frozen comparison fixes degree-six polynomial references, 32 intervals
of width \(1/8\), time four, and 64-bit outward norm witnesses:

| Reference | Prior estimate | Sharper estimate |
|---|---|---|
| Six initial wavevectors | Envelope unresolved | Envelope unresolved |
| Eighteen wavevectors retaining nonlinear interactions | Envelope unresolved | Independently accepted |

This establishes that both changes are needed among these four tested
configurations. It says nothing about all other possible reference choices,
norms, certificates, or known mathematical approaches.

### K.2 Exact constrained quadratic optimization **[M]**

For the complete error \(w=u-v\), use
\(E=\|w\|_{\chi^{-1}}\), \(E_0=\|w\|_{\chi^0}\), and
\(E_1=\|w\|_{\chi^1}\). The full bilinear Fourier estimate gives

\[
D^+E+\nu E_1\le2VE_0+E_0^2+R,
\qquad E_0^2\le EE_1,\qquad E_1\ge E.
\]

Here \(V=\|v\|_{\chi^0}\) and \(R=\|r\|_{\chi^{-1}}\) include
the full reference residual. Choose an error cap \(c>0\), reserved
dissipation \(\mu>0\), and \(D=\nu-c-\mu>0\). While \(E\le c\),

\[
D^+E+\mu E_1\le-DE_1+2V\sqrt{EE_1}+R
\le\Lambda(V,D)E+R,
\]

where completing the square proves

\[
\Lambda(V,D)=\max_{x\ge1}(-Dx^2+2Vx)
=\begin{cases}
V^2/D,&V\ge D,\\
2V-D,&0\le V<D.
\end{cases}
\]

For positive error, \(x=\sqrt{E_1/E}\); zero error is handled directly
without division. The piecewise bound is monotone in \(V\), so an
interval-wide reference cap \(V_j\) can replace \(V\). Consequently

\[
\boxed{D^+E\le b_jE+R_j,\qquad
b_j=\Lambda(V_j,D)-\mu,\quad E\le c.}
\]

The release sets \(c=\mu=\nu/8\). The affine supersolution checks from
Appendix J use this coefficient and the strict cap \(q<c\). Positive
reserved dissipation bounds \(\int E_1dt\), which supplies the same
Sobolev continuation argument. A final full-norm bound below \(\nu\)
then invokes Appendix H's all-time theorem.

The scalar maximum is exact. The upstream norm estimates and the overall
PDE criterion are not claimed to be optimal. Setting the reserved
dissipation to zero would invalidate this particular continuation argument
and is rejected by the checker.

### K.3 Nonlinear references with accounted-for rounding **[C]**

The new reference retains the twelve projected interaction wavevectors
generated from the initial six. Their initial coefficients are zero but
their subsequent time-polynomial coefficients are nonzero. This retains
actual nonlinear interactions that the previous approximation omitted.

Intermediate Taylor coefficients are rounded to control rational arithmetic
growth, then projected and reflected to preserve exact divergence freedom
and conjugate symmetry. A highest-degree correction gives an exactly
shared rational endpoint between adjacent polynomial intervals. The main
references use 48-bit intermediate rounding, separately from the 64-bit
outward norm/envelope witnesses.

The independent checker trusts none of the reference's generation history.
It reconstructs the entire residual polynomial, including rounding,
endpoint corrections, and every nonzero omitted output mode. Bernstein
witnesses cover all times in each interval. The exact solution's evolving
spectrum is never assumed finite.

### K.4 Checked outcome and the next failure **[C, conditional on unpublished certificate]**

For \(\varepsilon=0.1\), the accepted certificate has 32 polynomial
intervals, 76 affine error segments, and 1,792 omitted-output-mode
occurrences across intervals. Its exact rational witnesses establish

\[
\max_{0\le t\le4}\|u(t)-v(t)\|_{\chi^{-1}}<0.04424,
\qquad \|u(4)\|_{\chi^{-1}}<0.05144<1.
\]

Together with the written sufficient argument and its standard analytic
dependencies, these imply global smoothness for this specified input.
The independent program check is not formal verification in a proof
assistant, and it does not by itself prove the analytic dependencies.

At \(\varepsilon=0.01\), the sharper six-mode reference has maximum
certified error below 0.026679; retaining nonlinear interactions reduces
that bound to below 0.002382. These compare certified upper bounds,
not exact observed errors. All twelve new targeted test methods pass,
and all 23 deliberate corruptions of the nonlinear 0.01 certificate
are rejected. An exact scalar counterexample also rejects use of the
small-\(V\) branch outside its valid domain.

At \(\varepsilon=0.5\), the same configuration fails its sufficient
envelope on the second interval. The failed estimate is retained as the
next fixed research target. It is not a singularity or evidence that the
exact solution loses smoothness.

These developments extend established approximate-solution validation;
see [Morosi and Pizzocchero's smooth-solution analysis](https://arxiv.org/abs/1405.3421).
They establish an implementation improvement on a fixed input, without
claiming a previously unknown mathematical regularity class. General
large-data Navier–Stokes regularity, RH, P versus NP, and empirical
seven-dimensional physics remain unresolved by this engine.

## Appendix L. Directing the engine toward valid extension arguments **[C, conditional on unpublished certificate]**

**Engine addition:** 0.18.0. The complete implementation record and mathematical
arguments are in `PROOF_SEARCH_DIRECTION.md` in the archive.

### L.1 The shared objective **[H]**

The three prize targets share a proof-engineering difficulty: a finite
calculation must be connected to the entire domain of the claim by a valid
extension argument. For RH this concerns all integers, for P versus NP a
uniform complexity statement, and for Navier–Stokes the required set of
initial data and times. This is a useful organizational principle, not a
reduction proving the three problems equivalent.

The engine is now directed toward sound abstraction, inductive closure,
residual-guided refinement, reuse of expensive certificate work, and
independent verification of the precise claimed scope. Encoding an answer
as a bit or a distance to zero does not make that bit easy to compute.
Representation size, search cost, and proof coverage remain separate.

### L.2 A finite certificate with unbounded path-length coverage **[M/C]**

For a transition relation \(T\), let \(\operatorname{Post}(S)\) be every
possible successor of a state in \(S\). With initial set \(S_0\) and bad
set \(B\), a candidate invariant \(I\) must satisfy

\[
S_0\subseteq I,\qquad\operatorname{Post}(I)\subseteq I,
\qquad I\cap B=\varnothing.
\]

Induction then proves that every reachable state avoids \(B\), at every
finite path length and hence at every prefix of an infinite trajectory.
This does not prove termination or a liveness property.

The new finite-system engine searches a quotient graph of a supplied state
partition. An abstract path to a bad block is checked for a concrete
realization using backward suffix-predecessor sets. A real path becomes a
counterexample certificate. A spurious path identifies a proper block split,
and the search repeats. Every split adds a block, so at most \(N-1\)
splits are possible from one block on \(N\) states. A smaller resource
budget can leave the search unresolved.

The standalone checker uses a different calculation: bit-mask inclusions
and direct closure under every concrete edge. It checks initial coverage,
complete partition coverage, safety, and transition closure without trusting
the refinement history. It can bind the certificate to a separately
provided requested system.

On the five-state safe example, three spurious paths are removed and the
invariant \(\{0,1\}\) is certified. Changing one initial outgoing edge
produces the real counterexample \(0,2,3,4\). Another 160 seeded finite
systems agree with independent concrete reachability. These results concern
fixed explicitly supplied systems, not all programs or input sizes. Small
abstract state counts do not imply cheap construction or a compact membership
map.

### L.3 Refine the distinctions responsible for the fluid residual **[C]**

The previously frozen fluid failure has viscosity one and

\[
u_0=(2\sin y+\varepsilon\sin z,\varepsilon\sin2x,
\varepsilon\sin y),\qquad \varepsilon=1/2.
\]

The engine ranks an omitted conjugate Fourier pair by the exact rational
sum of interval width times its residual norm caps. Seven pairs account
for approximately 91.214% of the old integrated omitted-residual bound.
Adding them yields a thirty-two-mode reference. Its first interval's full
residual cap falls from about 0.115940 to 0.020422.

This is a heuristic for constructing a candidate reference: the approximate
dynamics change when modes are added. The new full residual therefore must
be recomputed and independently checked. Intermediate rounding and endpoint
corrections remain included. A selected mode can round to zero at lower
precision; the release preserves the counterexample to the false expectation
that every selected mode must remain active.

### L.4 Local error budgets **[M]**

The sharper inequality from Appendix K can be applied on each affine
envelope segment with a different cap \(c_\ell\). A fixed reserve
\(\mu>0\) remains positive throughout. The checker requires

\[
0\le q_L,q_R<c_\ell,\qquad c_\ell+\mu<\nu,
\]

\[
b_\ell=\Lambda(V_j,\nu-c_\ell-\mu)-\mu,
\]

\[
\frac{q_R-q_L}{\Delta}\ge b_\ell q_L+R_j,
\qquad
\frac{q_R-q_L}{\Delta}\ge b_\ell q_R+R_j.
\]

These inequalities hold throughout the affine segment. Exact joins and
the checked initial error propagate the bootstrap across the whole finite
partition. All remaining damping values are positive; their finite minimum
is positive. Integrating the stronger error inequality with the fixed
positive reserve supplies the existing Sobolev continuation argument.
The final full norm below viscosity again invokes the all-time small-data
theorem, without assuming finite evolving support for the exact solution.

The local policy uses reserve \(\nu/1024\), caps on a \(\nu/1024\)
grid, and a declared segment-resolution ladder. These are computational
parameters, not fundamental physical units. The complete policy changes
the caps, reserve, and resolution; the comparison does not isolate one
parameter as the sole cause of success.

### L.5 The new accepted input and the common audit **[C, conditional on unpublished certificate]**

For the same \(\varepsilon=0.5\) input, all five fixed-cap configurations
fail on both the eighteen-mode and thirty-two-mode references. All three
local-control resolutions fail on the old reference. On the refined
reference, sixteen segments per interval fail and sixty-four pass the
independent checker. The next listed fallback is not visited.

The accepted certificate covers 32 degree-six reference intervals,
2,048 affine segments, and 4,032 omitted-output-mode occurrences. Exact
rational witnesses give the outward comparisons

\[
\max_{0\le t\le4}\|u(t)-v(t)\|_{\chi^{-1}}<0.73674,
\qquad \|u(4)\|_{\chi^{-1}}<0.70240<1.
\]

The maximum local cap is 0.7373046875, the minimum remaining damping is
0.26171875, and the positive reserve is \(1/1024\). Subject to the
written analytic dependencies, this supplies global smoothness for the
specified initial field. It expands the engine's coverage; it is not a
claim of a new regularity class or a general all-data theorem.

Reference bounds are cached across candidate searches. Their content hashes
check identity, not mathematical truth: a deliberately falsified and
rehashed cache still fails independent verification. The final checker
reconstructs the entire residual and checks all continuous-time witnesses.
No timing result is presented as a globally optimal or portable performance
guarantee.

The proof director also independently rechecks 1,023 portable RH base cases,
21 bounded SAT instances, and the finite inductive certificate. It compares
property, domain, problem identity, and bounded versus unbounded coverage
before allowing evidence to support a claim. A finite prefix does not
cover all integers; one fluid datum does not cover all data; a formula-family
calculation does not bound every algorithm.

Twenty-three targeted test methods pass, including the 160 finite-system
comparisons. All 24 deliberate corruptions of a small local-cap certificate
are rejected. Earlier proof sources and their frozen documents remain
unchanged.

The new work uses independently implemented programs and written mathematical
arguments, without formal proof-assistant verification or a novelty claim.
General RH, P versus NP, and all-data Navier–Stokes remain open in the
engine's audit. The next substantive work is the missing extension argument
for each target, broader verified initial-data coverage, and formalization
of selected analytic dependencies. The shared workflow supports that work;
it does not substitute for it.

---

## Appendix M. Hypothetical universes and checked proof transfer **[M/C]**

Version 0.19 makes the proposal to connect possible answers precise. Let
R denote RH, P denote P=NP, and N denote all-data global smoothness for the
unforced, viscosity-one, mean-zero periodic three-dimensional fluid problem.
Classical logic has eight truth assignments to these three propositions.
The engine enumerates them, without asserting that each assignment is
consistent with arithmetic or is an actual physical universe. No new
unconditional implication between the targets has been established;
all eight assignments remain open in this audit.

A case split can be a valid proof method. If a common base theory proves
P implies R and not-P implies R, then R follows by excluded middle. The
new checker verifies the conditional propositional inference and retains
both premises as unproved. In contrast, even assuming the cycle R implies
P implies N implies R leaves both the all-false and all-true assignments.
The engine explicitly records this remaining ambiguity. A screen of 24
signed implications tests propositional tautologies only; it cannot decide
which additional implications arithmetic or analysis might prove.

### M.1. A contracting comparison can close an error loop **[M]**

Let d be a finite nonnegative real vector, A a nonnegative matrix, and b a
nonnegative forcing vector. Suppose

\[
d\le Ad+b,\qquad B\ge0,\qquad B(I-A)=I.
\]

Multiplying the rearranged inequality by B gives the exact conditional bound

\[
\boxed{d\le Bb.}
\]

When b=0, nonnegativity forces d=0. This is a proved comparison lemma;
the standalone checker verifies the matrix identities over exact rationals.
For the three-cycle with each gain 1/2, B has rows (8,4,2)/7,
(2,8,4)/7, and (4,2,8)/7. The positive weight vector B*1=(2,2,2) has
weighted contraction factor 1/2. Forcing (1/100,0,0) produces the exact
bound (2/175,1/350,1/175). Nonzero forcing is retained explicitly.

More generally, d1<=a*d2, d2<=b*d3, d3<=c*d1 with nonnegative gains
forces the three finite errors to vanish if a*b*c<1. At unit gains,
d=(1,1,1) is a counterexample to that conclusion. The finite-vector and
strict-contraction conditions are essential.

The engine has not constructed actual defects for RH, P versus NP, and
Navier–Stokes satisfying these inequalities. Such a construction must prove
what zero means, finiteness, full-domain coverage, and the coupling bounds.
Naming an arbitrary coordinate after a prize does not supply those proofs.
This elementary comparison theorem is not claimed as a new mathematical
discovery or a solution of the three targets.

### M.2. A finite dynamics-to-SAT bridge with complete coverage **[M/C]**

For a finite transition system S with N states, any reachable bad state
has a path of at most N-1 edges: a shortest such path cannot repeat a state,
because the repeated segment could be removed. Add self-loops for padding
and compile a one-hot CNF Phi_K requiring an initial state at time zero,
legal padded transitions, and a bad state at time K. Then

\[
\operatorname{SAT}(\Phi_K(S))
\iff \text{a bad state is reachable within K steps},
\]

so

\[
\boxed{\operatorname{UNSAT}(\Phi_{N-1}(S))
\iff \text{S is safe at every path length}.}
\]

The checker independently reconstructs the formula and checks either all
bits of a satisfying assignment or a finite DPLL refutation tree with
clause-justified unit inferences and both children of every split. Shorter
horizons retain bounded conclusions. Explicit graph size is tracked; a
polynomial-size encoding in N does not give polynomial cost in a succinct
description with only log(N) state bits. Ordinary graph traversal already
decides this finite reachability problem, so the bridge gives no general
polynomial SAT algorithm.

### M.3. Transfer through a verified state map **[M/C]**

A source-to-target map must preserve initial states, send bad source states
to bad target states, and send every source edge to a target edge or a
stutter. A hypothetical bad source path would then map to a bad target
path after stutters are removed. Therefore checked target safety implies
source safety. The converse is not supplied by these hypotheses.

The frozen demonstration reduces five source states to three target states,
and the corresponding SAT instance from 25 variables and 77 clauses to
9 variables and 20 clauses. Every source edge is checked. A separate unsafe
example yields [0,2,3,4]. This is actual finite proof transfer between two
representations, with no claim that either graph exactly encodes the full
fluid PDE or the universal arithmetic and complexity targets.

The new modules pass 16 targeted test methods, including 640 graph/horizon
comparisons, 80 small CNF comparisons, and 64 exact matrix examples.
Six positive certificate controls pass and 32 deliberate corruptions are
rejected. Earlier analytic certificates remain frozen and are not enlarged
by this release. The detailed proofs, schemas, executed records, and Windows
commands are in `UNIVERSES_AND_PROOF_BRIDGES.md` and the v0.19 engine.

The official problem statements remain the reference for the universal
claims: [RH](https://www.claymath.org/wp-content/uploads/2022/05/riemann.pdf),
[P versus NP](https://www.claymath.org/wp-content/uploads/2022/06/pvsnp.pdf),
and [Navier–Stokes](https://www.claymath.org/wp-content/uploads/2022/06/navierstokes.pdf).
These program checks and written arguments are not formal proof-assistant
verification. No universal bridge, new seven-dimensional physical law, or
Millennium solution follows from this release.

---

## Appendix N. Null-axis probes, observability, and a fluid projection obstruction **[M/C]**

Version 0.20 tests the proposal that observing both sides of a controlled
push can expose hidden coordinates. It gives exact conditions for success,
complete ambiguity descriptions in a nonlinear model, and a written limit
on one proposed finite projection of fluid dynamics. No universal prize
problem or new physical dimension is established.

### N.1. Linear hidden directions and allowed transformations **[M]**

For a linear readout y=Cx, two states are indistinguishable precisely when
their difference lies in ker(C). Known additive pushes do not remove this
ambiguity because C(x+u)-C(x'+u)=C(x-x'). Hidden distinctions occur at any
observed coordinate, not only at a chosen zero origin. A zero observation
does not imply a perfect shape or an absence of upper-state information.

If known allowed transformations A_i can mix hidden and visible coordinates,
form W from all rows C*A_last*...*A_first over finite experiment words.
Then

\[
x\sim x'\iff x-x'\in W^\perp.
\]

The algorithm starts with rows of C and closes their span under right
multiplication by every allowed map. At most n independent rows survive
for an n-dimensional state. Each has an explicit word witness. Inclusion
of C and closure prove coverage of every finite word by induction; the
witnesses prove that no unattainable measurement was added. A standalone
checker verifies both directions and a full basis of the annihilator.

The three-coordinate example has one hidden direction under identity
dynamics and none when an x-z quarter rotation is allowed. A fourth untouched
coordinate remains hidden. A seven-coordinate shift exposes the initially
seventh coordinate only after six transitions. The number seven is an
example of the general finite-dimensional result, not a physical deduction.
Combining multiple experiment words requires reproducible initial conditions
or a justified measurement protocol.

### N.2. Signed recovery from paired squared-distance readings **[M]**

With h(x)=||x||^2, paired displacements give

\[
h(x+u)-h(x-u)=4u\cdot x.
\]

Choosing u=delta*e_i recovers x_i exactly by dividing the contrast by
4*delta, for nonzero delta. The implemented example recovers
(3/5,-2/7,4/3) with delta=1/8. Under reading errors individually bounded
by eta, the coordinate error is at most eta/(2|delta|). Smaller probes can
therefore amplify reading error.

The complete data also contain the common squared norm through paired sums.
Let U contain the probe vectors, d their contrasts divided by four, and s
the squared norm inferred from each sum. If x0 is the minimum-norm solution
of Ux=d, the exact fiber is

\[
\{x_0+v:v\in\ker U,\ \|v\|^2=s-\|x_0\|^2\}.
\]

Orthogonality of x0 and ker(U) proves the formula in both directions.
The checker distinguishes an empty fiber, a unique point, two points, or
a sphere of compatible states. A zero remaining radius can give uniqueness
even with a rank-deficient probe matrix. These results assume the stated
sensor law and repeated access to the same initial state.

### N.3. No autonomous law for any fixed energy-derivative jet of all shears **[M]**

For a periodic smooth shear with positive finite energy weights w_k,

\[
u(y,t)=\left(2\sum_k\sqrt{w_k}e^{-\nu k^2t}\sin(ky),0,0\right),
\]

incompressibility and zero advection reduce the full unforced Navier–Stokes
equation to the heat equation. With normalized energy E=average(|u|^2)/2,

\[
E^{(j)}(0)=(-2\nu)^j\sum_k w_k k^{2j}.
\]

For every finite m, take m+2 squared wave numbers lambda_i=i^2 and
v_i=1/product_(l!=i)(lambda_i-lambda_l). Comparing highest coefficients
in the Lagrange interpolation formula gives

\[
\sum_i v_i\lambda_i^j=0\ (0\le j\le m),\qquad
\sum_i v_i\lambda_i^{m+1}=1.
\]

The positive and negative parts of v therefore realize two global smooth
shears with equal energy derivatives through order m and unequal next
derivatives. Their common projected state z_m=(E,...,E^(m)) has different
projected time derivatives. Consequently no single-valued autonomous rule
z_m'=F(z_m) describes all these shears. This is an explicit failure of the
earlier law-descent criterion for that projection.

At viscosity one and wave numbers 1,2,3,4, integer-scaled weights
(0,14,0,2) and (7,0,9,0) give equal E=16, E'=-176, E''=2944, but
E'''=-72704 and -52544 respectively. An integer-amplitude realization is
checked independently with the frozen full Fourier algebra, including zero
divergence, zero nonlinear term, and all four energy derivatives.

When exactly four distinct frequencies are known and no others occur,
four energy derivatives recover the weights through a Vandermonde system.
The new observability engine executes that recovery. This determines energy
evolution in the specified shear model; it does not identify every velocity
phase or prove a finite cutoff for the general PDE. Both alias examples
remain globally smooth. The theorem concerns finite energy-derivative jets,
not every possible observation or intervention.

### N.4. Two further shortcut failures and validation **[M/C]**

The entire polynomial F(s)=w^4+(15/8)w^2+289/256, w=s-1/2, has reflection
and conjugation symmetry but roots s=1/2 +/- 1/4 +/- i. Exact Gaussian
arithmetic checks all four. Thus those symmetries alone do not force roots
onto the central line; this polynomial is not the Riemann xi function and
does not refute RH.

For n>=3, the satisfiable CNF consisting of the all-positive clause and
every implication x_i implies x_j (i!=j) has violation count
H(k)=1_(k=0)+k(n-k), where k is the number of true variables. The all-zero
state has cost one, each one-bit neighbor has cost n-1, and the all-one
state has cost zero. Local one-bit descent can therefore stop above the
satisfying set. This is not a lower bound for all algorithms.

The release passes 18 targeted test methods and 31 corruption rejections.
Twenty-three standalone certificates cover the main examples and shear
orders zero through twelve. The all-m theorem has the interpolation proof
above and in `NULL_AXIS_OBSERVABILITY.md`; finite runs are not its proof.
There is no mathematical novelty or formal proof-assistant verification claim.

The three prize targets retain their original quantified definitions:
[RH](https://www.claymath.org/wp-content/uploads/2022/05/riemann.pdf),
[P versus NP](https://www.claymath.org/wp-content/uploads/2022/06/pvsnp.pdf),
and [Navier–Stokes](https://www.claymath.org/wp-content/uploads/2022/06/navierstokes.pdf).
An assumed finite grid would require a proved equivalence to those domains
before its computation could settle them. Consistency with a proposed model
does not itself establish decidability, efficient solvability, or truth of
every conjecture expressed in it.

## Appendix O. Active experiments, model rejection, and correction of the hierarchy **[M/C]**

Revision 1.14 is accompanied by Rope Tomography Lab 0.2.0 and the focused report Reality_Revision_1.14.md. It makes a substantive correction to Section 17: complete deterministic closure for a fixed transition family is idempotent. Seven strict repeated complete closures therefore cannot satisfy the former theorem's premises. A seven-bit shift register demonstrates a valid seven-stage finite-horizon hierarchy, with no implication for the number of physical dimensions.

### O.1. Intervention-relative cause classes **[M]**

For allowed experiments \(\mathcal E\) and predicted responses \(F_e(\theta)\), define

\[
\theta\sim_{\mathcal E}\theta'
\iff F_e(\theta)=F_e(\theta')\quad\forall e\in\mathcal E.
\]

Use equality of full response distributions for a general stochastic model. The recoverable object is an equivalence class in \(\Theta/{\sim_{\mathcal E}}\). Increasing the allowed experiments can refine that quotient, but repeating observationally equivalent measurements need not refine it.

In the lab's conditionally independent common-noise model, adaptive choice cannot distinguish causes equivalent under every allowed experiment. Equal past-data distributions induce equal distributions for the next experiment chosen from that history; its response distributions also agree. Induction establishes equality of the entire transcript distributions.

This formalizes the distinction between improving numerical estimation and obtaining a new identifying observation. Optimal experimental design is an existing research area; a relevant primary research preprint studies control inputs for parameter identifiability and model discrimination. The lab implements a finite catalogue method rather than that paper's continuous optimal-control construction. [Liu, Maini and Baker, 2025](https://arxiv.org/html/2506.11311v1)

### O.2. Choose a discriminating experiment before seeing its outcome **[M/C]**

For fixed candidate means \(\mu_{ei}\), weights \(w_i\), known common Gaussian covariance \(\Sigma_e\), and positive cost \(c_e\), the implemented criterion is

\[
S(e)=\frac1{c_e}\sum_{i<j}w_iw_j
(\mu_{ei}-\mu_{ej})^\top\Sigma_e^{-1}(\mu_{ei}-\mu_{ej}).
\]

The algorithm maximizes this pairwise-KL criterion over a finite affordable catalogue. It returns no informative selection when all affordable scores are numerically zero. It does not optimize over every physically realizable apparatus or generate every plausible hidden mechanism. Its current pairwise evaluation costs \(O(|\mathcal E|M^2r)\), excluding the cost of predicting means.

It updates relative candidate probabilities, then separately checks absolute residual compatibility against the known-noise distribution. If every candidate fails, it records model inadequacy rather than reporting its highest-posterior candidate as established truth. The statistical cutoff is per observation and is not a sequentially adjusted error guarantee.

### O.3. A concrete intervention that breaks an exact delay ambiguity **[M/H]**

When \(\tau\) is transit delay and \(\delta\) local pure delay, passive measurements can identify only their sum. If a calibrated intervention multiplies the path delay by \(\lambda\ne1\) while preserving speed and the local attachment,

\[
t_1=\tau+\delta,\qquad t_\lambda=\lambda\tau+\delta.
\]

Solving gives

\[
\tau=\frac{t_\lambda-t_1}{\lambda-1},\qquad
\delta=\frac{\lambda t_1-t_\lambda}{\lambda-1}.
\]

These equations use unwrapped delays. Wrapped phases still require cycle resolution; unknown intervention side effects invalidate the assumed two-equation model. The new simulation discriminates fixed candidate pairs and supplies no physical implementation of the intervention.

### O.4. Evidence and limits of this release **[C]**

The recorded protocol fixes six scenarios, each with 2,000 synthetic observations, plus a 2,000-observation missing-model control. Integer-frequency phase aliases, path/local-delay sums, and uncalibrated gain/depth products remain ambiguous. Respectively, an off-harmonic probe, a known path scaling, and independent coupling calibration distinguish the selected pairs.

In the three informative scenarios, forced nearest-mean errors were 12, 16, and 37 out of 2,000. These counts include observations where the absolute check rejected both candidates; they are not error rates conditional on acceptance. All three passive controls retained tied candidate probabilities in every trial.

In all 2,000 missing-model trials, one incorrect candidate had over 99% relative probability, yet the absolute check rejected both. This is a constructed counterexample to using posterior concentration alone as model validation.

An independent audit recalculated all 14,000 observation records and catalogue scores without importing the selector or simulator. The combined unit suite passed 15 groups. The exact closure audit passed 7,662 partition/transition cases and checked the seven-bit shift example. Written proofs establish the stated general identities; numerical experiments are scoped implementation evidence. No new physical measurements, extra-dimensional discovery, P=NP proof, or formal proof-assistant certification is supplied.

### O.5. A revisable theory **[H/T]**

The machine-readable claim ledger stores assumptions, status, evidence, and failure conditions. It explicitly retracts the former seven-stage closure argument. Earlier evidence remains tied to its named release and source, rather than being silently promoted to support later claims.

The continuing research task is to identify a precise missing distinction, declare competing models, design a feasible discriminating experiment, and test a prediction excluded from construction. If the candidate set is inadequate, revise the models and reserve fresh validation data. If the permitted observations cannot distinguish the candidates, preserve the equivalence class. Metaphysical \(G\) is not a numerical repair term for model failure.

## Appendix P. Capture, generator discovery, and the limits of an observed cycle **[M/C]**

Revision 1.15 incorporates an audit of RRLS Studio v0.27 and the uploaded dimensional-future report. The repaired source is release 0.27.1, documented in RRLS_v27_SOURCE_AUDIT.md. Original source snapshots and failures are retained. New results are distinct from the original benchmark record.

### P.1. Keep the event model's semantics explicit **[M]**

The modular event solver finds integer ticks satisfying simultaneous rational linear congruences. It does not enforce continuous non-crossing. For example, \(x_A(t)=t\) and \(x_B(t)=1-t\) modulo 10 never coincide at an integer tick, because \(2t\equiv1\pmod{10}\) is impossible for integer \(t\). They nevertheless intersect at \(t=1/2\).

Persistent point capture is an additional rule. Each nontrivial merge reduces the cluster count, so an initially \(N\)-cluster merge-only model has at most \(N-1\) merge events. That bound does not count the internal deformation dynamics of extended strands. Bond creation changes a contact graph; non-crossing motion alone does not establish a change of knot or link type.

The uploaded mechanics text also mixed torque and angular impulse. For an ideal instantaneous contact at offset \(\mathbf r_c\),

\[
\Delta\mathbf L=\mathbf r_c\times\mathbf J_t,
\]

where \(\mathbf J_t\) is tangential impulse. It already integrates force over time and must not receive another time factor. General offset rigid-body impacts require rotational inertia and relative contact-point velocities. The repaired source's impulse helper remains explicitly limited to central/point-mass contact.

### P.2. Finite observations do not select a universal generator language **[M]**

For any positive integer \(N\), the compact periodic laws

\[
a(t)=0,\qquad b(t)=\mathbf1[t\bmod(N+1)=N]
\]

agree on the first \(N\) integer samples but differ on the next. A deterministic learner given that common prefix has identical input in both cases and therefore cannot guarantee the correct next value for both. A previously known exact period would be an additional assumption; a finite observed repetition does not establish it.

This obstruction does not rule out useful generator learning within a restricted language.

### P.3. Positive theorem: continuation in a bounded recurrence class **[M]**

Fix a prime field and a bound \(K\). Suppose the complete noiseless sequence satisfies a homogeneous constant-coefficient linear recurrence of order at most \(K\), valid from its initial state.

**Theorem.** Its first \(2K\) consecutive terms uniquely determine its continuation within that class.

**Proof.** For two candidate sequences, let \(P\) and \(Q\) be monic annihilating polynomials in the forward shift, each of degree at most \(K\). Their product annihilates the difference and has degree at most \(2K\). Agreement of the first \(2K\) terms gives enough zero initial values for this monic recurrence to force the difference to remain zero. \(\square\)

This determines the sequence, not necessarily a unique coefficient representation. It neither discovers nor proves the physical applicability of the recurrence class. Linear feedback synthesis is established mathematics; the new implementation uses bounded-order modular elimination rather than claiming a new general method. [Massey, 1969](https://ieeexplore.ieee.org/document/1054260/)

The implementation requires exact consecutive field-valued samples and declares \(p\le65521\), \(K\le64\). It rejects insufficient sample coverage or a prefix incompatible with every tested order. Once a recurrence is fitted, companion-matrix exponentiation computes a future term in \(O(r^3\log t)\) field operations, with arithmetic bit costs still present.

### P.4. New computation and repaired failures **[C]**

The source audit reproduced weak-damping cancellation that returned zero for positive winding and zero time for a positive wrapping target. Stable expm1/log1p evaluation repairs those cases. The audit also repaired an earliest-event error for negative starting ticks, a verifier that accepted unsupported impossibility claims, and an impulse helper that did not distinguish approach from separation under the stated normal convention.

The new ten-group suite passed 1,200 rational pair worlds checked over full relative periods, 120 capture regressions, 21 high-precision damping configurations, 500 learned recurrence sequences, and 14,000 values excluded from fitting. Six exact finite-prefix examples retain the unrestricted-discovery obstruction. The five original compact self-tests also passed; the complete inherited benchmark corpus was not rerun.

From \(0,1,1,2,3,5\) over GF(257), with order bound three, the new learner recovers coefficients \((1,1)\) and computes

\[
a_{10^{96}}=160\pmod{257}.
\]

Independent Fibonacci fast doubling agrees. This validates a scoped learned-generator jump, not a universal predictor or a physical dimensional claim. The earlier P versus NP proof requirements remain unchanged.

## Appendix Q. Repeated model checks with a bounded false-rejection budget **[M/C]**

Revision 1.16, 6 September 2026, adds Rope Tomography Lab 0.3.0. The source paper was its latest saved v1.15; the rope archive was 0.2.0; the requested v0.27 file identity now resolves to RRLS_Studio_v27_Audited_Source.zip, including the subsequent reconciled audit. The v0.28 generator, v0.29 semantic-canonicalization, and v0.30 dual-phase reports were also consulted to avoid treating their existing results as new. No RRLS source was changed or its inherited corpus rerun here.

### Q.1. A single-observation cutoff is not an investigation-wide guarantee **[M]**

Appendix O correctly limited its 1% gate to each observation. If a correct candidate is tested on independent, continuously distributed, correctly calibrated residuals at that same level for T looks, the probability of at least one false rejection is

\[
1-(1-0.01)^T.
\]

At T=200 this is about 86.602%. Two observationally identical correct candidates are rejected together, so this is also an explicit counterexample to interpreting repeated all-candidate rejection as having a 1% investigation-wide error rate. This does not retract the earlier single-observation results; it rules out an unsupported extension of them.

### Q.2. Written guarantee and its exact assumptions **[M]**

Fix a finite candidate catalogue before validation. Experiment e_t may be selected using previous observations, but before seeing the next one. Assume at least one fixed candidate i* supplies the correct conditional Gaussian mean and known positive diagonal covariance for every selected experiment. Fresh standardized residual coordinates must have the specified conditional standard-normal law; marginal Gaussianity alone is insufficient after adaptive selection. No parameter fitting to the scored observation is permitted.

Let p_(i,t) be the upper chi-square tail probability of the squared standardized residual. Allocate

\[
\alpha_t=\frac{\alpha}{t(t+1)},\qquad
\sum_{t=1}^{T}\alpha_t=\alpha\frac{T}{T+1}<\alpha.
\]

Permanently remove candidate i if p_(i,t)<alpha_t. Conditional calibration implies P(p_(i*,t)<alpha_t | past)<=alpha_t. Taking expectations and applying the countable union bound gives

\[
\Pr\{\text{ever remove }i_*\}\le\sum_{t\ge1}\alpha_t=\alpha.
\]

If i* is present initially, ever removing every candidate implies removing i*, so it has the same bound. No factor for candidate count is required for this specific claim. Simultaneously preserving every true candidate is a different, stronger claim and is not asserted. Stopping or choosing experiments based on past data does not break this argument. Selecting the largest of two current residuals does: even if each arm alone has a valid p-value, its selected p-value rejects with probability 1-(1-alpha_t)^2>alpha_t for independent arms.

This is a written probability proof, not a proof-assistant certificate. It is standard error-budget reasoning, not a newly discovered statistical theorem. The numerical implementation uses chi-square survival probabilities directly, avoiding late-look cancellation in 1-alpha_t; it explicitly supports alpha>=10^-12 and at most 10^9 looks. Floating-point tail evaluation is not interval-certified.

### Q.3. Fresh synthetic evidence and an independent check **[C]**

The protocol froze source hashes, seed 606202603, two calibrated experiment settings, and three scenarios before generating validation data. Each scenario has 2,000 independent histories of 200 observations. Experiment choice depends on the preceding standardized first residual. Within each experiment the two candidates make identical predictions: retaining them must preserve their ambiguity.

| Scenario | Repeated 1% gate: histories ever rejected | Budgeted gate: histories ever rejected |
|---|---:|---:|
| Correct conditional noise | 1,717 / 2,000 | 14 / 2,000 |
| Persistent first-coordinate shift of 0.35 noise standard deviations | 1,864 / 2,000 | 26 / 2,000 |
| Actual noise standard deviation twice the declared value | 2,000 / 2,000 | 2,000 / 2,000 |

Only the first row satisfies the full null model. The second measures limited detection power; the third shows why a correct mean with incorrect noise calibration is not covered by the guarantee. Non-rejection establishes neither truth nor good sensitivity. For the correct-noise scenario, independent product formulas give finite-horizon false-rejection probabilities 86.6020% and 0.991530%; the sample rates need not equal them.

An audit that imports neither the engine, simulator, nor scipy.stats reconstructs the adaptive choices and uses the analytic two-coordinate tail exp(-q/2). It agrees on 2,400,000 rejection bits from 1,200,000 stored synthetic observations. It shares the declared assumptions, observations, and NumPy; this is independent implementation checking, not external replication. The stateful API was additionally replayed on ten histories per scenario, not counted as extra data. All 25 unit tests pass, including the 15 inherited tests.

### Q.4. Research context, model revision, and next test **[H]**

Time-uniform inference is established research; Howard et al. (Annals of Statistics, April 2021) develop more sophisticated confidence sequences: https://arxiv.org/abs/1810.08240v9. Koning and van Meer (preprint revised 21 December 2025) give sequential tests matching a specified fixed-horizon test at its terminal horizon: https://arxiv.org/abs/2501.03982v5. A newly located application by Grob, Daniele, and Ziegel (preprint submitted 24 August 2026) evaluates forecast calibration sequentially with e-values: https://arxiv.org/abs/2608.23064v1. These are relevant statistical methods and applications, not physical evidence for this framework. This release implements conservative alpha spending, not those papers' constructions.

The next bounded test is a frozen cumulative Gaussian likelihood-ratio/e-process against small persistent shifts, compared at the same 1% ever-false-rejection budget on new seeds. It is not implemented or validated here. In a continuing investigation, refitting candidates requires fresh validation and an explicitly allocated remaining or cross-revision error budget. Resetting a test to 1% after every revision does not retain a global 1% guarantee. Neither this gate nor a concentrated posterior resolves observational equivalence, identifies God as a residual, establishes seven physical dimensions, or decides P versus NP. All v1.14 and v1.15 corrections, including the initial-state bounded recurrence-class assumption, remain in force.

## Appendix R. Accumulating weak residual evidence without optional-stopping inflation

Revision 1.17, 7 September 2026, adds Rope Tomography Lab 0.4.0. It starts from the saved paper v1.16 and rope archive v0.3.0. The requested RRLS v0.27 identity still resolves to the audited v0.27 source and was read for its current claim ledger; RRLS code and evidence are unchanged in this revision.

### R.1. Bounded target and assumptions

Revision 1.16 found that allocating a shrinking significance level to each new residual controls false rejection but detected only 26 of 2,000 histories containing a persistent standardized mean shift of 0.35. The bounded target here is that same failure mode, not arbitrary misspecification.

Fix a candidate i, a finite catalogue of experiments, a significance level alpha, a positive target shift delta, and one unit direction u_e in whitened measurement coordinates for each experiment. The experiment and direction at time t may depend on the past but must be selected before observing Y_t. If i is correct, assume

\[
Z_{i,t}=u_{e_t}^{\mathsf T}\Sigma_{e_t}^{-1/2}(Y_t-\mu_{e_t,i})
\mid\mathcal F_{t-1}\sim N(0,1).
\]

The conditional statement is essential. Marginal normality, reused data, current-response-dependent direction choice, fitted predictions scored on the same response, or misspecified covariance do not satisfy it.

Put S_(i,t)=sum_(s=1)^t Z_(i,s). For fixed alternatives with standardized shifts +delta and -delta, define

\[
L_{i,t}^{+}=\exp\!\left(\delta S_{i,t}-\frac{t\delta^2}{2}\right),\qquad
L_{i,t}^{-}=\exp\!\left(-\delta S_{i,t}-\frac{t\delta^2}{2}\right),
\]

and their equal mixture

\[
E_{i,t}=\frac{L_{i,t}^{+}+L_{i,t}^{-}}{2}
=\exp\!\left(-\frac{t\delta^2}{2}\right)\cosh(\delta S_{i,t}).
\]

### R.2. Written anytime-valid guarantee

Under the declared null,

\[
\mathbb E[\exp(\pm\delta Z_{i,t}-\delta^2/2)\mid\mathcal F_{t-1}]=1.
\]

Therefore L+ and L- are nonnegative martingales starting at one, and so is their mixture E. Ville's inequality gives

\[
\boxed{\Pr_i\!\left(\sup_{t\ge0}E_{i,t}\ge1/\alpha\right)\le\alpha.}
\]

This is a written probability proof, not a proof-assistant artifact or a new theorem. If one correct candidate is present, ever rejecting the entire candidate set by this rule implies rejecting that candidate and inherits the bound. It does not simultaneously protect every correct candidate without an additional multiple-testing argument. Nor may this 1% process and the separate 1% point-residual gate be joined with an OR rule and called a 1% procedure; a valid combined rule must allocate or merge their error budgets.

The engine stores log e-values using a stable log-cosh identity. It copies the candidate predictions, noise scales, and normalized directions at construction. It cannot detect dishonest data reuse, post-response experiment selection, or a caller resetting the object. Floating-point evaluation is tested but not interval-certified.

### R.3. Frozen synthetic validation

Before outcome generation, the protocol fixed source hashes, seed 1709202604, 2,000 histories, 200 looks, alpha=0.01, delta=0.35, two experiment settings, and a past-dependent experiment policy. Both modeled candidates make identical predictions, so their evidence paths must remain tied. The e-process and alpha-spending gate are compared separately at 1%; this table is not a combined decision rule.

| Synthetic scenario | Cumulative e-process ever rejected | Alpha-spending gate ever rejected | Median e-process first rejection among rejected |
|---|---:|---:|---:|
| Correct conditional Gaussian model | 20 / 2,000 | 23 / 2,000 | 60 |
| Constant +0.35 shift in monitored standardized coordinate | 1,896 / 2,000 | 28 / 2,000 | 75 |
| Constant -0.35 shift in monitored standardized coordinate | 1,906 / 2,000 | 27 / 2,000 | 74 |
| Alternating +0.35/-0.35 shift | 18 / 2,000 | 31 / 2,000 | 72 |
| Constant +0.35 shift in the orthogonal coordinate | 15 / 2,000 | 22 / 2,000 | 64 |
| Actual noise scale twice the declared scale | 865 / 2,000 | 2,000 / 2,000 | 50 |

Only the first row satisfies the complete null. Its observed 1% e-process rejection fraction is a finite sample, not the proof of the guarantee. The two constant-shift rows answer the prespecified power question: the cumulative process detects about 95% of those histories, whereas the shrinking point gate detects about 1.4%. The alternating and orthogonal rows are deliberate counterexamples to an omnibus interpretation. The variance row shows complementary behavior: the point-residual gate detects all tested histories, while the mean-shift e-process does not.

An independent audit imports neither the cumulative engine, simulator, nor SciPy. It reconstructs the predictable experiment choices, uses a separate scalar stable-log-cosh calculation, and independently evaluates the exact two-degree-of-freedom survival formula for the comparison gate. It agrees on 4,800,000 decision states from 2,400,000 stored synthetic observations. The stateful public API is replayed on 60 of those histories, not counted as new evidence. All 33 unit tests pass, including all 25 inherited tests. The audit shares the raw records, assumptions, standard-library mathematics, and NumPy loading/median; it is implementation independence, not external replication.

### R.4. Relation to current primary research

The use of nonnegative martingales and Ville thresholds belongs to established safe anytime-valid inference, not to the physical claims of this framework. Kaufmann and Koolen construct mixture martingales for time-uniform inference under adaptive sampling in exponential-family bandits: https://jmlr.org/papers/volume22/18-798/18-798.pdf. Arnold, Gavrilopoulos, Schulz, and Ziegel's *Sequential model confidence sets*, published 13 May 2026, uses e-processes, confidence sequences, and multiple-testing corrections to retain time-uniform nonasymptotic model-set coverage: https://academic.oup.com/jrsssb/advance-article/doi/10.1093/jrsssb/qkag066/8676743. Wang and Ramdas's Gaussian-mean work, updated 24 August 2026, develops e-processes when variance is unknown: https://arxiv.org/html/2310.03722v4. The present lab implements only the elementary known-variance two-point mixture above; it does not reproduce those papers' broader results.

### R.5. What changed and what remains open

The supported change is methodological: weak, directionally consistent residuals can be accumulated with an anytime false-rejection guarantee rather than tested in isolation. This does not validate the rope physics, identify a unique hidden cause, or turn non-rejection into truth. Delta=0.35 and the monitored direction were prespecified from revision 1.16's named failure; performance for other magnitudes and directions is not inferred.

The next bounded test is a variance-robust or self-normalized e-process, frozen before new outcomes, because the current known-covariance assumption is the most consequential fragility. A separate design should then combine complementary point and cumulative detectors with one total error budget. All earlier corrections remain: complete deterministic closure for one fixed transition family is idempotent; integer-tick point capture is not continuous non-crossing; angular impulse has no extra time factor; finite prefixes do not identify arbitrary generators; and the 2K continuation theorem assumes an exact noiseless homogeneous constant-coefficient recurrence of order at most K valid from the initial state. No unconstrained residual is identified with God, and neither seven physical dimensions nor P=NP is established.

## Appendix S. Unknown fixed scale without pretending to know the variance

Revision 1.18, 8 September 2026, adds Rope Tomography Lab 0.5.0. It starts from the saved paper v1.17, rope archive 0.4.0, and the current claim ledgers. The requested RRLS identity still resolves to the audited v0.27 package. Its code and evidence are unchanged, and no inherited RRLS benchmark is relabeled as fresh validation.

### S.1. Bounded question and exact model class

Appendix R's cumulative test assumes the variance used to standardize residuals is correct. Its doubled-noise control rejected 865 of 2,000 histories, even though that experiment changed scale rather than mean. The bounded question here is whether a sequential mean check can remain valid when one common positive Gaussian residual scale is unknown.

Fix one candidate and, before each response, a scalar projected residual $X_t$. Under the null assume

\[
X_1,X_2,\ldots\overset{\mathrm{iid}}{\sim}N(0,\kappa^2)
\]

for one fixed unknown \(\kappa>0\). Nominal per-coordinate scales may define the projection, but after that projection the residual stream must have this common-scale law. Independence, Gaussianity, a zero mean, and a time-constant scale are all assumptions. Past-adaptive experiment selection is not covered merely because it is predictable; it is permitted only when the resulting projected stream still has the stated iid law.

Let $S_n=\sum_{t=1}^nX_t$, $V_n=\sum_{t=1}^nX_t^2$, and predeclare $c>0$. The implemented statistic is the scale-invariant Gaussian likelihood mixture of Wang and Ramdas:

\[
G_n^{(c)}=
\sqrt{\frac{c^2}{n+c^2}}
\left(
\frac{(n+c^2)V_n}{(n+c^2)V_n-S_n^2}
\right)^{n/2}.
\]

For nonzero $V_n$, Cauchy--Schwarz and $c>0$ make the denominator positive. Multiplying the complete prefix by any nonzero constant leaves $G_n^{(c)}$ unchanged. The implementation uses $c=1$, stable logarithms, and reports exact all-zero prefixes as insufficient scale information instead of treating the formula's degenerate case as evidence.

### S.2. Established theorem and a crucial filtration boundary

Wang and Ramdas prove that $G_n^{(c)}$ is a nonnegative martingale with initial value one for the scale-invariant Gaussian null on their reduced, scale-invariant filtration. Ville's inequality therefore gives

\[
\Pr\!\left\{\sup_{n\ge1}G_n^{(c)}\ge1/\alpha\right\}\le\alpha
\]

for the rejection rule "remove the candidate the first time $G_n^{(c)}\ge1/\alpha$." This is an application of their published mathematical result, not a new theorem proved by this project.

*Revision 1.69 correction of the filtration boundary.* The earlier text said that stopping on a raw residual magnitude "is not licensed by this argument." That located the boundary in the wrong place. The event $\{\sup_{n\ge1}G_n^{(c)}\ge1/\alpha\}$ is one fixed event, so its probability bound does not depend on when the investigator looks, on what raw statistics prompt a look, or on any stopping rule at all: every rejection under the stated rule is contained in that event. The ever-rejection guarantee used by Appendix T therefore holds under arbitrary raw-data look schedules. What the reduced filtration does restrict is two other things. First, $\mathbb E[G_\tau^{(c)}]\le1$ is guaranteed only for stopping times $\tau$ of the reduced scale-invariant filtration; at a $\tau$ chosen from raw magnitudes, $G_\tau^{(c)}$ is not an e-value and may not be reported or merged as one. Second, $G_n^{(c)}$ is not a supermartingale in the canonical observation filtration, so it cannot be averaged, multiplied, or otherwise merged with canonical-filtration processes such as the point-residual gate or Appendix U's variance process to form a single e-process. Those two facts, not a restriction on looking, are why Appendix T uses a union bound and refuses to average. Data reuse, refitting on scored observations, resetting the process, post-response direction choice, non-Gaussian errors, dependence, or changing variance invalidate the martingale property itself and remain excluded.

Source: Hongjian Wang and Aaditya Ramdas, *Anytime-valid t-tests and confidence sequences for Gaussian means with unknown variance*, arXiv:2310.03722v4, updated 24 August 2026: https://arxiv.org/html/2310.03722v4.

### S.3. Frozen finite computation and synthetic tests

Before generating outcomes, the protocol fixed source hashes, seed 1809202605, $c=1$, alpha=0.01, 2,000 histories, 200 looks, and six scenarios. Two candidates had identical zero predictions, so the engine also had to preserve their observational ambiguity. The scale-two Gaussian records were exact positive rescalings of their scale-one partners; they are a pathwise invariance control, not an independent sample.

| Synthetic scenario | Scale-invariant process ever rejected | v1.17 known-scale process ever rejected | Median first scale-invariant rejection among rejected |
|---|---:|---:|---:|
| iid (N(0,1)) | 15 / 2,000 | 12 / 2,000 | 51 |
| exact 2x rescaling, iid (N(0,4)) | 15 / 2,000 | 848 / 2,000 | 51 |
| iid (N(0.35,1)) | 1,811 / 2,000 | 1,915 / 2,000 | 94 |
| exact 2x rescaling, iid (N(0.70,4)) | 1,811 / 2,000 | 2,000 / 2,000 | 94 |
| alternating means (+0.35,-0.35) with unit Gaussian noise | 9 / 2,000 | 16 / 2,000 | 25 |
| variance-one scaled Student (t_3) | 3 / 2,000 | 37 / 2,000 | 12 |

Only the first two rows satisfy the full mean-zero theorem. Their 15/2,000 count is finite synthetic evidence, not the proof of the 1% bound. Exact equality of both scale-invariant decision paths is a finite computational check of the algebraic scale invariance. The old process's 848/2,000 scale-two count is not a failure of its theorem: those data violate its known-unit-variance null.

The paired shift rows have the same standardized effect and exactly the same scale-invariant decisions. They show 1,811/2,000 targeted detections at this horizon, less than the v1.17 known-scale process's scale-one count. The alternating row is a retained counterexample: a constant-mean detector is not an omnibus adequacy test. The Student row violates Gaussianity; its low rejection count neither proves validity nor adequacy under heavy tails.

Nine new unit tests compare the formula to high-precision arithmetic, check scale and sign invariance, preserve tied candidates, and exercise invalid and zero-energy inputs. Together with 33 inherited tests, all 42 pass. An audit importing neither production evidence code, the simulator, nor SciPy independently reconstructs all 4,800,000 decision states from 2,400,000 stored synthetic residuals and agrees exactly. It shares the data and assumptions and is not external replication. An initially truncated result archive is recorded as a packaging failure; the unchanged frozen sources and seed reproduced the same summary before the intact archive was audited.

### S.4. Result, limits, and next test

The supported correction is methodological: known variance is unnecessary for this particular scale-invariant Gaussian mean test. It does not make the overall theory variance-free, distribution-free, or physically validated. In particular, it deliberately ignores pure scale changes; the doubled-scale null is acceptable within its model class. Detecting variance misspecification requires a separate detector and a shared error budget.

The next bounded test is a joint point-plus-cumulative procedure with one total 1% anytime false-rejection budget, evaluated against both mean and scale errors on fresh data. Candidate or detector redesign must reserve a new seed and validation set. All earlier corrections remain in force: complete deterministic closure for one fixed transition family is idempotent; integer-tick point capture is not continuous non-crossing; angular impulse has no extra time factor; finite prefixes do not identify arbitrary generators; and the 2K continuation theorem assumes an exact noiseless homogeneous constant-coefficient recurrence of order at most K valid from the initial state. No unconstrained residual is identified with God, and neither seven physical dimensions nor P=NP is established.

## Appendix T. One shared error budget for complementary model checks

Revision 1.19, 9 September 2026, adds Rope Tomography Lab 0.6.0. It starts from the saved paper v1.18, rope archive 0.5.0, and their current ledgers and revision notes. The audited RRLS v0.27 package was checked for the retained capture, impulse, generator-class, and P-versus-NP boundaries. RRLS code and evidence are unchanged.

### T.1. The composition problem

Appendix R warned that joining two 1% detectors with an OR rule does not create a 1% procedure. Appendix S then supplied a cumulative mean process whose martingale theorem holds on a reduced scale-invariant filtration. The point-residual gate, by design, reads raw standardized magnitudes in the canonical filtration. A convex average or product of two processes is an e-process only when both are supermartingales in one common filtration, and these two are not, so they are not averaged and declared to be one e-process. Their ever-rejection events, on the other hand, are each bounded individually whatever the look schedule (S.2, revision 1.69 correction), and that is exactly what a union bound needs. The composition problem is therefore not a restriction on stopping; it is that the two guarantees live in different filtrations and can be combined at the level of events but not at the level of e-values.

Research on merging sequential e-values shows that legitimate merging is a substantive construction, not a license to combine arbitrary evidence streams. See Vladimir Vovk and Ruodu Wang, *Merging sequential e-values via martingales*, Electronic Journal of Statistics 18 (2024), 1185--1205: https://projecteuclid.org/journals/electronic-journal-of-statistics/volume-18/issue-1/Merging-sequential-e-values-via-martingales/10.1214/24-EJS2228.full. Wang and Ramdas also emphasize that filtration choice matters for unknown-variance sequential tests and describe union-bound constructions for simultaneous mean and variance confidence sets: https://arxiv.org/html/2310.03722v4.

### T.2. Written joint guarantee

Fix a candidate before validation and a total level $\alpha$. Allocate $\alpha_m=w\alpha$ to the scale-invariant cumulative mean process and $\alpha_p=(1-w)\alpha$ to the alpha-spending point-residual gate, with fixed $0<w<1$. This implementation uses $w=1/2$.

Let $A_m$ be the event that the mean process ever removes the specified correct candidate and $A_p$ the analogous event for the point gate. Under all assumptions of both component theorems,

\[
\Pr(A_m)\le\alpha_m,\qquad \Pr(A_p)\le\alpha_p.
\]

No independence between these two events is required. The union bound gives

\[
\boxed{\Pr(A_m\cup A_p)\le\alpha_m+\alpha_p=\alpha.}
\]

The joint engine removes a candidate if either component removes it. If at least one fixed correct candidate was present, ever eliminating the entire candidate set implies eliminating that candidate and inherits the same bound. This does not guarantee simultaneous preservation of every correct candidate.

The joint assumptions are stronger than Appendix S alone: fresh standardized residual vectors must have the candidate's declared conditional Gaussian mean and covariance for the point gate, while the monitored projection must also be iid Gaussian with one fixed scale for the mean process. No reset, scored-data refit, current-response selection, or reuse is allowed. This is a written probability argument, not a proof-assistant certificate or new statistical theorem.

*Revision 1.69 clarification of what the unknown-scale component buys.* Because the point gate's bound requires the declared covariance to be correct, the joint null forces the residual scale to equal its declared value. Inside the joint guarantee, Appendix S's freedom to leave the scale unknown is therefore inert: the joint procedure's null is the known-covariance Gaussian null, the same null Appendix R uses. When the true scale differs from the declared one, the mean component's own bound still holds, but the point gate's does not, and the joint false-rejection rate is uncontrolled; the T.3 row for doubled-variance residuals, 1,927 removals in 2,000, is that failure, not a detection. The mean component's scale invariance is a property of that component in isolation. It does not make the joint procedure scale-free, and later text should not describe the joint rule as carrying an unknown-scale null.

### T.3. Frozen synthetic validation

Before outcomes, the protocol fixed source hashes, seed 1909202606, 2,000 independent histories per scenario, 200 looks, total alpha=0.01, equal 0.005 component budgets, and $c=1$. Two modeled candidates had identical scalar predictions, so the implementation had to preserve their ambiguity. The invalid comparator gives both components a full 1% and ORs them; it is recorded as a counterexample, not labeled a valid 1% rule.

| Synthetic scenario | Shared-budget joint rejection | Mean component | Point component | Invalid full-1%-each OR |
|---|---:|---:|---:|---:|
| iid $N(0,1)$ | 25 / 2,000 | 8 | 17 | 45 |
| iid $N(0.35,1)$ | 1,714 / 2,000 | 1,712 | 22 | 1,814 |
| iid $N(0,4)$ | 1,927 / 2,000 | 3 | 1,927 | 1,969 |
| iid $N(0.70,4)$ | 1,997 / 2,000 | 1,724 | 1,982 | 1,998 |
| alternating means $+0.35,-0.35$ with unit Gaussian noise | 31 / 2,000 | 5 | 26 | 54 |
| iid $N(0,1.25^2)$ | 141 / 2,000 | 8 | 133 | 208 |
| variance-one scaled Student $t_3$ | 1,049 / 2,000 | 4 | 1,046 | 1,113 |

Only the first row satisfies the full joint null. Its observed 25/2,000 rate is 1.25%, not a proof or disproof of the mathematical bound. A post-outcome diagnostic gives probability 0.155651 of at least 25 rejections under 2,000 independent Bernoulli trials with probability 0.01, and a two-sided 95% exact binomial interval of 0.8105% to 1.8397%; this diagnostic was not part of the frozen protocol. The theorem, not agreement of a finite count with 1%, supplies the guarantee.

The shared rule retained complementary power: 1,714/2,000 for the specified persistent mean shift, 1,927/2,000 for doubled scale, and 1,997/2,000 when both were present. Tightening the mean threshold from 1% in v1.18 to its 0.5% share reduced its mean-shift detections from 1,811/2,000 in a different seeded dataset to 1,712/2,000 here; cross-seed counts are descriptive, not a paired causal estimate of the budget cost.

The failures are equally important. Alternating shifts remained mostly invisible. A moderate scale increase to 1.25 was detected in only 141/2,000 histories because the point gate waits for isolated extremes rather than accumulating weak scale evidence. Heavy-tailed $t_3$ noise has the same variance as the null after scaling but violates Gaussianity; the joint rule rejected 1,049/2,000 histories. This is a direct counterexample to calling the procedure distribution-free or equating variance calibration with tail calibration.

Eight new unit tests and all 42 inherited tests pass. A separate scalar audit imports neither production engine, simulator, nor SciPy; it recomputes the scale-invariant statistic and uses the analytic one-dimensional Gaussian tail via `erfc`. It agrees on 11,200,000 component and joint decision states from 2,800,000 stored synthetic residuals. It shares data and assumptions and is not external replication.

### T.4. Supported change and next test

The supported change is a valid shared-budget decision rule with measured complementary synthetic power. It is not a universal model validator, does not prove any physical rope mechanism, and turns neither rejection nor non-rejection into ontology. The severe heavy-tail result says the current Gaussian point gate can confuse tail misspecification with candidate failure; whether that is useful detection or false rejection depends on the explicitly declared model.

The next bounded test is a Gaussian variance likelihood-ratio e-process that accumulates moderate scale evidence, compared against the current alpha-spending point gate at the same reserved budget and on fresh data. A later robust-tail design would require a new null class rather than relabeling Gaussian results. All earlier corrections remain in force: complete deterministic closure for one fixed transition family is idempotent; integer-tick point capture is not continuous non-crossing; angular impulse has no extra time factor; finite prefixes do not identify arbitrary generators; and the 2K continuation theorem assumes an exact noiseless homogeneous constant-coefficient recurrence of order at most K valid from the initial state. No unconstrained residual is identified with God, and neither seven physical dimensions nor P=NP is established.

## Appendix U. Accumulating moderate Gaussian scale evidence

Revision 1.20, 10 September 2026, adds Rope Tomography Lab 0.7.0. It starts from the saved paper v1.19, rope archive 0.6.0, and their current ledgers and revision notes. The requested RRLS identity still resolves to the audited v0.27 package. RRLS code and evidence are unchanged.

### U.1. Bounded question and fixed alternatives

Appendix T found that the alpha-spending point gate detected only 141 of 2,000 synthetic histories whose Gaussian residual scale was 1.25 times the declared scale. The bounded question is whether squared-residual evidence can be accumulated without sacrificing an anytime false-rejection guarantee. It is not a search over arbitrary variance paths.

For a fixed candidate, let the fresh observation at look \(t\) have declared mean \(\mu_t\) and positive diagonal covariance \(\Sigma_t\), with experiment selection completed before seeing that observation. Under the null assume

\[
Z_t=\Sigma_t^{-1/2}(Y_t-\mu_t)\mid\mathcal F_{t-1}\sim N(0,I_{d_t}).
\]

The validation predeclares two covariance multipliers,

\[
\lambda_H=1.25^2=1.5625,\qquad \lambda_L=0.8^2=0.64,
\]

with equal weights. They represent complete-covariance scale corrections, not physical mechanisms. Put \(Q_n=\sum_{t=1}^n\lVert Z_t\rVert^2\) and \(D_n=\sum_{t=1}^n d_t\). For any fixed positive \(\lambda\), the alternative-to-null likelihood ratio is

\[
L_n(\lambda)=\lambda^{-D_n/2}
\exp\!\left[\frac12\left(1-\frac1\lambda\right)Q_n\right],
\]

and the implemented evidence process is

\[
E_n=\tfrac12L_n(\lambda_H)+\tfrac12L_n(\lambda_L).
\]

### U.2. Written anytime guarantee and interpretation boundary

For a correct candidate, each one-step factor is the conditional density ratio of \(N(0,\lambda I_{d_t})\) to \(N(0,I_{d_t})\), evaluated at \(Z_t\). Its conditional null expectation is one. Therefore each \(L_n(\lambda)\) is a nonnegative martingale beginning at one, and their fixed mixture \(E_n\) is also one. Ville's inequality gives

\[
\boxed{\Pr\!\left\{\sup_{n\ge0}E_n\ge1/\alpha_v\right\}\le\alpha_v.}
\]

This is a written probability proof, not a proof-assistant artifact or a new statistical theorem. The engine uses stable log-sum-exp arithmetic and permanently removes a candidate after the running maximum reaches the threshold. Floating-point calculations are tested but not interval-certified.

The assumptions are material: the candidate and mixture must be fixed before scoring; residuals must obey the declared conditional multivariate Gaussian law; the covariance correction must scale all scored coordinates uniformly; and there may be no reset, reuse, current-response selection, or refitting on scored observations. Predictable experiment choice is allowed only while preserving that conditional law. Crossing says a prespecified predictive correction outperformed the source strongly enough for this test. It does not identify whether the cause was sensor calibration, latent dynamics, a mean error, non-Gaussian tails, or anything theological.

This interpretation follows the general construction in Seungjin Choi, *Anytime-Valid Evidence for Prespecified Predictive Corrections*, submitted 8 August 2026: https://arxiv.org/abs/2608.08174. That primary research establishes the likelihood-ratio method; it is statistical methodology, not measured evidence for this theory.

### U.3. Frozen synthetic test and independent replay

Before outcomes, the protocol fixed the four scientific source hashes, seed 201009202607, 2,000 independent histories per scenario, 200 looks, the two scale alternatives, equal mixture weights, and the same reserved \(\alpha_v=0.005\) used by Appendix T's point component. The two detectors were compared separately; they were not OR-combined. Two modeled candidates had identical predictions, so their evidence paths had to remain tied.

| Synthetic scenario | Variance e-process ever rejected | Point gate ever rejected | Median variance-process first rejection among rejected |
|---|---:|---:|---:|
| iid \(N(0,1)\) | 4 / 2,000 | 4 / 2,000 | 80.5 |
| iid \(N(0,1.25^2)\), prespecified high scale | 1,774 / 2,000 | 125 / 2,000 | 87 |
| iid \(N(0,0.8^2)\), prespecified low scale | 1,669 / 2,000 | 0 / 2,000 | 113 |
| iid \(N(0,2^2)\), off-target large scale | 2,000 / 2,000 | 1,941 / 2,000 | 13 |
| iid \(N(0,1.10^2)\), smaller untuned scale | 251 / 2,000 | 34 / 2,000 | 119 |
| iid \(N(0.35,1)\), mean error with correct variance | 58 / 2,000 | 18 / 2,000 | 107 |
| variance-one scaled Student \(t_3\) | 1,047 / 2,000 | 996 / 2,000 | 81 |

Only the first row satisfies the full null. Its four rejections are a finite synthetic outcome; the martingale proof, not this count, supplies the 0.5% anytime bound. The targeted result is a large paired-method improvement for the 1.25-scale failure at the same budget. The matched low-scale result exposes an asymmetry of the upper-tail point gate, while the 1.10-scale row preserves a power limitation rather than extrapolating from the targeted alternative.

The mean-shift and heavy-tail rows are counterexamples to identifying a crossing with a unique variance cause. In particular, the scaled Student \(t_3\) data have unit variance but violate Gaussianity, and 1,047 of 2,000 histories crossed. The process is therefore neither distribution-free nor a variance-mechanism detector.

Nine new unit tests and all 50 inherited tests pass. A separate scalar audit imports neither the production variance engine, validation runner, nor SciPy. It recomputes the likelihood mixture and Gaussian point tail and agrees on all 5,600,000 stored decision states. It shares the synthetic records, null assumptions, floating-point convention, and threshold and is not external replication or physical evidence.

### U.4. Supported change, limitations, and next test

The supported change is a standalone cumulative Gaussian scale diagnostic for fixed candidate catalogues. It improves the named moderate-scale failure under the tested model, but it does not yet replace Appendix T's point component: changing the joint decision rule after inspecting these results requires a fresh joint validation set. It also cannot turn rejection into ontology or non-rejection into truth.

The next bounded test is a frozen paired comparison of the existing mean-plus-point joint rule against a mean-plus-variance joint rule under the same total 1% budget, including persistent scale errors and isolated one-look shocks. A heavy-tail-robust guarantee would require a new null class, not reuse of the Gaussian proof. All earlier corrections remain in force: complete deterministic closure for one fixed transition family is idempotent; integer-tick point capture is not continuous non-crossing; angular impulse has no extra time factor; finite prefixes do not identify arbitrary generators; and the 2K continuation theorem assumes an exact noiseless homogeneous constant-coefficient recurrence of order at most K valid from the initial state. No unconstrained residual is identified with God, and neither seven physical dimensions nor P=NP is established.

## Appendix V. A common-filtration mean-plus-variance process

Revision 1.21, 11 September 2026, adds Rope Tomography Lab 0.8.0. It starts from the saved paper v1.20, rope archive 0.7.0, and their current ledgers and revision notes. The audited RRLS v0.27 package was checked for the retained collision, angular-impulse, recurrence-class, and complexity boundaries. RRLS code and evidence are unchanged.

### V.1. Why this is not a detector substitution

Appendix T combined an unknown-scale mean process and a point-residual gate by splitting one error budget. The mean theorem there applies on a reduced scale-invariant filtration, while the point gate reads raw magnitudes, so the release correctly refused to average them into one e-process. Appendix U then supplied a canonical-filtration variance process but left open whether it should replace the point component.

The bounded question is whether a mean and variance process can share one Gaussian observation filtration and one 1% threshold, and how that choice trades persistent-error sensitivity against isolated-shock sensitivity. This requires changing the mean component back to the known-covariance process of Appendix R. As the revision 1.69 clarification in T.2 records, this does not abandon an unknown-scale null, because the joint rule of Appendix T never had one: its point gate already required the declared covariance. What changes is which component detects what, and that the two components now share one canonical filtration, which is what permits the fixed mixture of V.2 to be a single e-process.

### V.2. Written fixed-mixture guarantee

Fix a candidate, a positive mean-shift magnitude \(\delta\), one unit direction \(u_e\) in whitened coordinates for each experiment, positive covariance multipliers and their fixed weights, and an outer weight \(0<\rho<1\). Experiments may be chosen from the past before the current response. Under a correct candidate require

\[
Z_t=\Sigma_t^{-1/2}(Y_t-\mu_t)\mid\mathcal F_{t-1}
\sim N(0,I_{d_t}).
\]

Let \(E^{(m)}_n\) be Appendix R's equal mixture of likelihood ratios for projected shifts \(+\delta\) and \(-\delta\). Let \(E^{(v)}_n\) be Appendix U's fixed mixture of likelihood ratios for uniform covariance multipliers. Both are nonnegative martingales starting at one on the same canonical observation filtration under this null. Therefore

\[
E^{(j)}_n=\rho E^{(m)}_n+(1-\rho)E^{(v)}_n
\]

is also a nonnegative martingale starting at one, and

\[
\boxed{\Pr\!\left\{\sup_{n\ge0}E^{(j)}_n\ge1/\alpha\right\}\le\alpha.}
\]

The implementation predeclares \(\delta=0.35\), covariance factors \(1.5625\) and \(0.64\), equal inner weights, \(\rho=1/2\), and \(\alpha=0.01\). Alpha is applied once to the outer mixture. When only one component grows, its half weight imposes roughly the same evidence penalty as a 0.5% component allocation; when both grow, their current evidence can combine. The mixture uses current e-values, not separately maximized component values.

This is a written probability proof and an elementary common-filtration mixture, not a proof-assistant artifact or new statistical theorem. Vovk and Wang's work shows that merging sequential e-values is a substantive construction: https://projecteuclid.org/journals/electronic-journal-of-statistics/volume-18/issue-1/Merging-sequential-e-values-via-martingales/10.1214/24-EJS2228.full. Choi's prespecified-correction framework likewise interprets likelihood-ratio crossings relative to declared predictive corrections rather than as mechanism identification: https://arxiv.org/abs/2608.08174. The present implementation does not claim the full generality of either paper.

### V.3. Frozen paired comparison

Before outcomes, the protocol fixed four scientific source hashes, seed 211109202608, 2,000 independent histories per scenario, 200 looks, and all mixture parameters. Both rules received one total 1% anytime budget and were evaluated on identical fresh residual paths. Two candidates had identical predictions and remained tied.

| Synthetic scenario | Existing mean-plus-point rule | New mean-plus-variance mixture |
|---|---:|---:|
| iid \(N(0,1)\) | 12 / 2,000 | 10 / 2,000 |
| iid \(N(0.35,1)\) | 1,744 / 2,000 | 1,878 / 2,000 |
| iid \(N(0,1.25^2)\) | 140 / 2,000 | 1,807 / 2,000 |
| iid \(N(0,0.8^2)\) | 4 / 2,000 | 1,667 / 2,000 |
| iid \(N(0,2^2)\) | 1,941 / 2,000 | 2,000 / 2,000 |
| iid \(N(0.35,1.25^2)\) | 1,175 / 2,000 | 1,976 / 2,000 |
| alternating means \(+0.35,-0.35\) | 28 / 2,000 | 60 / 2,000 |
| one exactly signed 6-sigma residual at look 100 | 2,000 / 2,000 | 197 / 2,000 |
| one exactly signed 10-sigma residual at look 100 | 2,000 / 2,000 | 2,000 / 2,000 |
| variance-one scaled Student \(t_3\) | 1,053 / 2,000 | 1,031 / 2,000 |

Only the first row satisfies both complete nulls. Its finite counts neither prove nor compare the general probability bounds; the written arguments do that under their respective assumptions. The persistent scale and combined-error rows establish the intended synthetic improvement. The 6-sigma row is a constructed counterexample to unconditional dominance: the point gate reacts to the isolated event, whereas the cumulative variance alternatives dilute it among roughly null-sized earlier residuals. At 10 sigma both tested designs reject every history.

The known-covariance mean component also improves the targeted mean row in this paired design, but that result cannot be exported to Appendix S's unknown-scale class. Alternating shifts remain weakly detected by either procedure. Both rules reject about half the unit-variance Student-\(t_3\) histories, directly refuting distribution-free interpretations.

Nine new unit tests and all 59 inherited tests pass. An independent scalar audit imports neither production joint engine, evidence kernels, validation runner, nor SciPy. It reconstructs and matches all 24,000,000 stored component and joint decision states. The audit shares the synthetic observations, assumptions, and floating-point threshold convention; it is implementation independence, not external replication or measured evidence.

### V.4. Supported change and unresolved design choice

The supported change is a separately selectable \(JointGaussianMeanVarianceEProcess\) for investigations that predeclare persistent mean and uniform covariance-scale alternatives under a known conditional Gaussian model. The existing \(JointMeanPointAdequacy\) remains available for isolated-residual sensitivity. Its mean component is scale-invariant in isolation, but that property is inert inside the joint guarantee because the point gate requires the declared covariance (T.2, revision 1.69 clarification), so it is not a reason to prefer the joint rule. Selecting between them is an experimental-design decision, not an ontological conclusion.

A three-component rule cannot give the point, mean, and variance detectors their existing full allocations and still inherit a 1% bound. The next bounded test is to predeclare a three-way mixture or budget allocation and optimize its mean, persistent-scale, and isolated-shock tradeoff on training simulations, then evaluate exactly one frozen choice on new validation data. Heavy-tail robustness requires a different null class.

All earlier corrections remain in force: complete deterministic closure for one fixed transition family is idempotent; integer-tick point capture is not continuous non-crossing; angular impulse has no extra time factor; finite prefixes do not identify arbitrary generators; and the 2K continuation theorem assumes an exact noiseless homogeneous constant-coefficient recurrence of order at most K valid from the initial state. No unconstrained residual is identified with God, and neither seven physical dimensions nor P=NP is established.


## Appendix W. A trained, frozen mean/variance/point allocation

Revision 1.22, 12 September 2026, adds Rope Tomography Lab 0.9.0. It starts from the saved paper v1.21, rope archive 0.8.0, and their current claim ledgers and revision notes. The current audited RRLS v0.27 package was read to confirm that its collision, angular-impulse, recurrence-class, and complexity boundaries remain present. RRLS code and evidence are unchanged.

### W.1. Bounded question and predeclared separation

Appendix V left one concrete design problem: can one procedure keep most of the persistent mean/scale sensitivity of the common-filtration mixture while regaining the point gate's response to an isolated shock, without spending more than one total 1% anytime error budget?

The lab used two distinct data roles. A deterministic training grid used seed 221209202609, 1,000 histories per target, 200 looks, point-budget fractions in \(\{0.05,0.10,0.15,0.20,0.25,0.30\}\), and mean weights in \(\{0.25,0.375,0.50,0.625,0.75\}\). The four training targets were a constant 0.35 mean, scales 1.25 and 0.8, and one noisy signed 6-sigma shock at look 100. The predeclared lexicographic objective maximized the minimum target detection count, then the total, then preferred a smaller point fraction, then a mean weight nearer one half.

Training selected a 0.30 point-budget fraction and, inside the remaining mean/variance mixture, weights 0.25 for mean and 0.75 for variance. Its training counts were 915, 909, 852, and 837 of 1,000. These counts selected the configuration; they are not validation evidence. The runner, selected configuration, validation seed 221209202610, scenario definitions, source hashes, and audit were then frozen before the one validation run. The runner is designed to refuse silent reuse of its reserved output directory. The adjacent configuration with mean weight 0.375 had a worst-target count of 836 rather than 837, so the selected grid optimum is not evidence of a uniquely optimal continuous allocation and may be training-seed sensitive.

This separation follows the general principle that data used to choose a procedure must not be counted again as held-out evidence. See Angelopoulos et al., *Learn then Test* (2021), https://arxiv.org/abs/2110.01052. That work is methodological context; the finite grid and guarantee below are the claims implemented here.

### W.2. Written anytime bound

Let \(\alpha=0.01\), let \(q=0.30\), and set \(\alpha_p=q\alpha=0.003\) and \(\alpha_{mv}=(1-q)\alpha=0.007\). Under a specified correct candidate, require fresh whitened residuals to satisfy

\[
Z_t\mid\mathcal F_{t-1}\sim N(0,I),
\]

with each experiment and monitored direction selected from the past before the current response, fixed alternatives, and no scored-data reuse, reset, or refit. Let \(E^{(m)}_n\) and \(E^{(v)}_n\) be the known-covariance mean and covariance-scale processes of Appendices R and U on this same canonical Gaussian filtration. With the training-selected fixed weight \(w=0.25\),

\[
M_n=wE^{(m)}_n+(1-w)E^{(v)}_n
\]

is a nonnegative martingale starting at one. Therefore Ville's inequality gives

\[
\Pr\!\left\{\sup_n M_n\ge 1/\alpha_{mv}\right\}\le \alpha_{mv}.
\]

The point gate is not asserted to be part of that e-process. At look \(t\), it uses the two-sided Gaussian tail allocation \(\alpha_{p,t}=\alpha_p/[t(t+1)]\). Since \(\sum_{t\ge1}1/[t(t+1)]=1\), a union bound gives probability at most \(\alpha_p\) that it ever fires for the specified correct candidate. A second event-level union bound then gives

\[
\boxed{\Pr\{\text{hybrid ever rejects the specified correct candidate}\}
\le \alpha_{mv}+\alpha_p=0.01.}
\]

Choosing \(q\) and \(w\) from data independent of all scored observations does not spend this bound. Conditional on the training data, they are fixed and the preceding inequality holds; averaging over the independent training data preserves it. Selection on the scored sequence, repeated re-selection, or reusing validation data would require a different proof.

This is a written probability argument, not a proof-assistant artifact and not a new statistical theorem. The within-filtration mixture follows the martingale logic discussed by Vovk and Wang, *Merging sequential e-values via martingales* (2024), https://projecteuclid.org/journals/electronic-journal-of-statistics/volume-18/issue-1/Merging-sequential-e-values-via-martingales/10.1214/24-EJS2228.full. Choi's *Anytime-Valid Evidence for Prespecified Predictive Corrections* (submitted 8 August 2026), https://arxiv.org/abs/2608.08174, supports interpreting a crossing against declared predictive alternatives, not as identification of a unique mechanism. The point gate remains separate precisely because the earlier scale-invariant mean construction and raw point statistic do not share the filtration needed for the claimed merge.

### W.3. One held-out synthetic validation

The frozen comparison used 2,000 independent 200-look histories per scenario. All three rules were evaluated on the same new residual paths: the revision 1.19 mean-plus-point rule, the revision 1.21 50/50 mean-plus-variance mixture, and the selected hybrid.

| Synthetic scenario | v1.19 mean + point | v1.21 mean + variance | selected hybrid |
|---|---:|---:|---:|
| iid \(N(0,1)\) | 17 / 2,000 | 22 / 2,000 | 21 / 2,000 |
| iid \(N(0.35,1)\) | 1,736 / 2,000 | 1,866 / 2,000 | 1,811 / 2,000 |
| iid \(N(0,1.25^2)\) | 122 / 2,000 | 1,780 / 2,000 | 1,781 / 2,000 |
| iid \(N(0,0.8^2)\) | 6 / 2,000 | 1,683 / 2,000 | 1,689 / 2,000 |
| iid \(N(0.35,1.25^2)\) | 1,187 / 2,000 | 1,986 / 2,000 | 1,985 / 2,000 |
| alternating means \(+0.35,-0.35\) | 28 / 2,000 | 75 / 2,000 | 82 / 2,000 |
| one noisy signed 6-sigma shock at look 100 | 1,673 / 2,000 | 379 / 2,000 | 1,621 / 2,000 |
| one exactly signed 6-sigma residual at look 100 | 2,000 / 2,000 | 222 / 2,000 | 2,000 / 2,000 |
| variance-one scaled Student \(t_3\) | 1,028 / 2,000 | 1,061 / 2,000 | 1,508 / 2,000 |

Only the first row satisfies the complete Gaussian null. Its 21 selected-hybrid rejections are a finite synthetic observation, not the proof of the 1% bound. The selected design bridges the targeted tradeoff on these paths: compared with v1.21 it gives up 55 detections for the constant mean target, is essentially unchanged on the persistent scale and combined targets, and gains 1,242 detections for the noisy-shock target. It exactly matches the point-based rule on the constructed exact-shock row but remains 52 detections below it on the noisy-shock row. These differences are finite paired results for one horizon, parameter set, grid, training seed, and validation seed; they do not establish universal ordering.

The heavy-tail control is a decisive counterexample to robustness. Although the scaled Student \(t_3\) paths have unit variance, Gaussianity is false and the selected hybrid rejects 1,508 of 2,000—more than either comparator. A crossing therefore indicates incompatibility with the specified Gaussian prediction and alternatives, not a unique variance, physical, sensor, metaphysical, or theological cause. Alternating shifts also remain weakly detected.

Eight new unit tests and all 68 inherited tests passed before validation, for 76 total. A separate scalar audit imports neither production engines, evidence kernels, tuner, validation runner, SciPy, nor their decision functions. It reconstructed the training-grid argmax and matched all 18,000,000 stored validation decision states. It shares the saved synthetic paths, assumptions, floating-point convention, and summary operations; this is finite implementation independence, not external replication or measured physical evidence.

### W.4. Supported change, limits, and next test

The supported change is a separately selectable \(ThreeWayGaussianAdequacy\) engine with a frozen 0.7% mean/variance mixture and 0.3% point gate. It is useful when persistent mean errors, uniform covariance-scale errors, and isolated residuals are all predeclared concerns under the conditional Gaussian known-covariance null. The two older rules remain available because they encode different assumptions and priorities.

No measured physical data were collected. The mathematical result is the stated conditional probability bound; the computation is one finite grid search; the validation and counterexamples are synthetic; the claim that this allocation will transfer to an instrument is a hypothesis; theological interpretation remains outside the statistic. Non-rejection does not establish truth, and rejection does not identify a generator or ontology.

Because the winning training configuration beat a nearby grid point by only one worst-target history, the next bounded test is training-only allocation-stability analysis across preregistered independent seeds. It should estimate how often each grid cell wins and how sensitive held-out target power is to that choice, without reopening or reusing this validation set. A heavy-tail-safe detector would require a separately declared null class and proof rather than relabeling this Gaussian engine.

All earlier corrections remain in force: complete deterministic closure for one fixed transition family is idempotent; integer-tick point capture is not continuous non-crossing; angular impulse has no extra time factor; finite prefixes do not identify arbitrary generators; and the 2K continuation theorem assumes an exact noiseless homogeneous constant-coefficient recurrence of order at most K valid from the initial state. No unconstrained residual is identified with God, and neither seven physical dimensions nor P=NP is established.

## Appendix X. Allocation-selection stability and a boundary warning

Revision 1.23, 13 September 2026, adds Rope Tomography Lab 0.10.0. It starts from the saved paper v1.22, rope archive 0.9.0, and their current claim ledgers and revision notes. The current audited RRLS v0.27 package was read to confirm the retained collision, angular-impulse, recurrence-class, and complexity boundaries. RRLS code and evidence are unchanged.

### X.1. Bounded question and sealed-data rule

Revision 1.22 selected a point-budget fraction of 0.30 and an inner mean weight of 0.25 from one 30-cell synthetic training grid. Its worst-target score beat a nearby cell by one history. The bounded question is whether that selected cell is repeatable under independent draws from the same four declared training scenarios.

This is a training-selection stability study, not a second validation of the adequacy engine. The frozen v1.22 validation observations and summary were not read by the new runner or audit. Before outcomes, the protocol fixed 64 seeds, 1,000 histories per target and seed, 200 looks, the original six point fractions and five mean weights, the original lexicographic selection objective, and all reported metrics. It also fixed a separate reference seed with 20,000 histories for each of the four targets. That 80,000-history reference sample estimates performance within this synthetic selection study; it is not measured evidence or permission to retune against the sealed validation set.

This separation addresses a standard model-selection problem: optimizing a noisy empirical criterion can overfit the selection sample even when the underlying candidates are unchanged. See Cawley and Talbot, *On Over-fitting in Model Selection and Subsequent Selection Bias in Performance Evaluation* (2010), https://www.jmlr.org/papers/v11/cawley10a.html. Bouthillier et al., *Accounting for Variance in Machine Learning Benchmarks* (2021), https://proceedings.mlr.press/v139/bouthillier21a.html, provides related primary evidence that repeated sources of variation matter when comparing procedures. These papers supply methodological context, not evidence for the present physical theory.

### X.2. Finite stability results

The 64 independent training runs selected four cells:

| Point-budget fraction | Mean weight inside mixture | Wins | Frequency |
|---:|---:|---:|---:|
| 0.30 | 0.25 | 49 / 64 | 76.6% |
| 0.30 | 0.375 | 10 / 64 | 15.6% |
| 0.30 | 0.50 | 3 / 64 | 4.7% |
| 0.25 | 0.25 | 2 / 64 | 3.1% |

For the v1.22 cell's 49/64 frequency, the descriptive 95% Wilson interval is 64.9% to 85.3%. It quantifies binomial sampling uncertainty under this exact simulation protocol; it is not a theorem about other target sets, horizons, grids, or physical instruments.

The primary worst-target count was tied between the winner and runner-up in 31 of 64 runs. Its median advantage was one history and its maximum was nine. The secondary total-count and fixed tie breakers therefore materially affected many selections. This confirms that “the optimizer returned one cell” is stronger language than the data support.

At the same time, the selections occupied a narrow region. A point fraction of 0.30 won 62 of 64 times; 0.25 won twice; no smaller point fraction won. Mean weight 0.25 won 51 times, 0.375 won ten times, and 0.50 won three times; larger weights never won. Thus the original cell is usually repeatable within the tested grid, although exact selection is not deterministic.

### X.3. Separate synthetic reference and counterexample to optimality

On the separately seeded 20,000-history-per-target reference sample, the original v1.22 cell also maximized the declared finite-grid objective:

| Target | Detections | Rate |
|---|---:|---:|
| constant mean 0.35 | 18,029 / 20,000 | 90.15% |
| scale 1.25 | 17,938 / 20,000 | 89.69% |
| scale 0.8 | 16,788 / 20,000 | 83.94% |
| noisy signed 6-sigma shock | 16,273 / 20,000 | 81.37% |

The worst-target reference rate was 81.365%. When each seed-selected cell was scored on this same reference sample, its minimax-rate regret relative to the reference grid winner had mean 0.0498 percentage points, median zero, 90th percentile 0.0600 points, and maximum 0.9200 points. These are finite Monte Carlo summaries on a shared reference set, not confidence bounds on real-world power.

The key counterexample is geometric rather than numerical: the winner sits on two grid boundaries. The largest tested point fraction won 62 of 64 seeds, while the smallest tested mean weight won 51. Therefore this experiment cannot distinguish “an optimum near 0.30 and 0.25” from “the objective would keep improving beyond one or both tested boundaries.” Repeatability inside a truncated search space does not establish a local or global optimum outside it.

Five new tests and all 76 inherited tests passed, for 81 total. An independent implementation imports neither the production evidence kernels, v1.22 tuner, stability runner, nor SciPy. It regenerated every synthetic history and formula from the frozen seeds and matched all 10,080,000 stored cell-target-history decisions. It shares the protocol, NumPy random generator, floating-point convention, and synthetic model, so it is implementation independence rather than external replication.

### X.4. Supported correction, evidence classes, and next test

The supported correction is narrower than adopting new weights. Revision 1.22's allocation is now described as usually repeatable and low-regret *within its declared grid and synthetic target distribution*, not as uniquely optimal. Its anytime false-rejection proof is unaffected because that proof applies to the already frozen allocation under its conditional known-covariance Gaussian assumptions. No core adequacy-engine code or v1.22 validation record changed.

The written mathematics remains the v1.22 martingale-plus-union-bound argument. This revision adds finite computation and synthetic stability evidence only. Transfer to a physical instrument remains a hypothesis; no measured observations were collected. Neither an optimizer boundary nor a residual is an ontological or theological identification.

The next bounded test is a training-only boundary-expansion study: preregister point fractions above 0.30 and mean weights below 0.25, repeat the same multi-seed stability analysis, and keep the v1.22 validation set sealed. If a materially different allocation is proposed afterward, it must receive a new frozen validation set before replacing the released design. Heavy-tail validity remains a separate problem requiring a new null class and proof.

All earlier corrections remain in force: complete deterministic closure for one fixed transition family is idempotent; integer-tick point capture is not continuous non-crossing; angular impulse has no extra time factor; finite prefixes do not identify arbitrary generators; and the 2K continuation theorem assumes an exact noiseless homogeneous constant-coefficient recurrence of order at most K valid from the initial state. No unconstrained residual is identified with God, and neither seven physical dimensions nor P=NP is established.

## Appendix Y. Expanded allocation boundaries refute the released grid optimum

Revision 1.24, 14 September 2026, adds Rope Tomography Lab 0.11.0. It starts from the saved paper v1.23, rope archive 0.10.0, and their current claim ledgers and revision notes. The audited RRLS v0.27 source was reread only to confirm its retained boundaries; RRLS code and evidence are unchanged.

### Y.1. Assumptions and frozen training-only protocol

Appendix X left one concrete ambiguity: its winner occupied the largest tested point-budget fraction, 0.30, and the smallest tested inner mean weight, 0.25. The bounded question is whether the released cell remains competitive after expanding both boundaries under the same four-target synthetic training objective.

Before generating outcomes, `boundary_expansion_protocol.json` fixed 64 new seeds disjoint from Appendix X, 1,000 histories per target and seed, 200 looks, and a separate 20,000-history-per-target reference seed. The complete old grid was retained. Point fractions were expanded to

\[
0.05,0.10,\ldots,0.60,
\]

and inner mean weights to

\[
0.05,0.10,0.15,0.20,0.25,0.375,0.50,0.625,0.75,
\]

for 108 cells. The targets, Gaussian generators, fixed alternatives, 1% total budget, lexicographic minimax objective, metrics, source hashes, and rule against automatic adoption were also frozen. The runner and audit do not read the v1.22 held-out validation observations or results.

This is a finite hyperparameter-selection study. Bergstra and Bengio, *Random Search for Hyper-Parameter Optimization* (2012), https://www.jmlr.org/papers/v13/bergstra12a.html, shows why a small axis-aligned search can miss important regions. Cawley and Talbot (2010), https://www.jmlr.org/papers/v11/cawley10a.html, and Bouthillier et al. (2021), https://proceedings.mlr.press/v139/bouthillier21a.html, motivate separating tuning from performance estimation and reporting selection variability. These primary methodological sources do not supply physical evidence for this framework.

### Y.2. Finite results and counterexamples

The released v1.22 cell won **0 / 64** new training runs; its descriptive Wilson 95% interval for the win frequency is 0% to 5.66%. Sixteen cells won at least once. The two modal cells, point fraction 0.45 with mean weight 0.10 and point fraction 0.50 with mean weight 0.10, each won only 12 of 64 runs. Aggregated over mean weights, point fractions 0.45 and 0.50 won 20 and 19 runs. Aggregated over point fractions, mean weight 0.10 won 39 runs and 0.15 won 19. Twenty-two runs tied the runner-up on the primary worst-target count; the median primary margin was one history and the maximum was six.

Thus the broader grid supplies a direct finite counterexample to treating the released cell as optimal for the declared training objective. It also supplies a second counterexample: opening the boundaries does not reveal one stable exact replacement at 1,000 histories per target. Selection instability must be retained rather than hidden by reporting only an argmax.

On the separately seeded 20,000-history-per-target synthetic reference sample, the expanded-grid winner was point fraction 0.45 and inner mean weight 0.10:

| Target | released 0.30/0.25 | reference candidate 0.45/0.10 |
|---|---:|---:|
| constant mean 0.35 | 18,072 / 20,000 | 17,096 / 20,000 |
| scale 1.25 | 17,956 / 20,000 | 17,916 / 20,000 |
| scale 0.8 | 16,886 / 20,000 | 16,798 / 20,000 |
| noisy signed 6-sigma shock | 16,360 / 20,000 | 16,715 / 20,000 |
| worst-target rate | 81.80% | 83.575% |

The declared minimax objective improves by 1.775 percentage points, but the candidate is not universally better: it loses 976 mean-shift detections, 40 high-scale detections, and 88 low-scale detections while gaining 355 shock detections. Changing the target set or utility weights can change the preferred cell. Among the 64 seed-selected cells, reference minimax regret had mean 0.637, median 0.430, 90th percentile 1.215, and maximum 3.260 percentage points.

Four new protocol tests and all 81 inherited tests passed before outcomes, for 85 total. A separate scalar implementation regenerated the synthetic histories and formulas and matched all 36,288,000 cell-target-history decisions. It imports neither the production evidence kernels, v1.22 tuner, stability runner, nor SciPy. It shares the frozen seeds, NumPy generator, synthetic model, and floating-point convention, so it is implementation independence rather than external replication.

### Y.3. Supported correction, limits, and next test

The supported correction is that revision 1.22's allocation is not optimal even within the newly declared 108-cell training grid. The 0.45/0.10 reference winner is a candidate, not a released replacement: exact training selection was dispersed across 16 cells, and the reference sample is tuning evidence rather than fresh validation. No core detector or sealed validation record changed.

The written 1% anytime bound is unchanged. It applies to any point fraction and inner mixture weight fixed independently of scored observations under the conditional known-covariance Gaussian assumptions of Appendix W. It does not prove that a tuned allocation has useful power, transfers to an instrument, or remains valid under heavy-tailed noise.

No measured physical data were collected. The probability bound is written mathematics; the grid search, replay, and counterexamples are finite computation on synthetic data; transfer is a hypothesis; and theological interpretation remains outside the statistic. A rejection cannot identify a unique physical, metaphysical, or theological cause.

The next bounded test is one preregistered, paired, fresh frozen validation comparing the released 0.30/0.25 design with the single 0.45/0.10 candidate. It should retain the declared mean, scale, and shock targets, include Gaussian-null and heavy-tail controls, fix one untouched seed and all metrics before outcomes, and adopt the candidate only if the preregistered decision rule is met. The v1.22 validation set must remain sealed.

All earlier corrections remain in force: complete deterministic closure for one fixed transition family is idempotent; integer-tick point capture is not continuous non-crossing; angular impulse has no extra time factor; finite prefixes do not identify arbitrary generators; and the 2K continuation theorem assumes an exact noiseless homogeneous constant-coefficient recurrence of order at most K valid from the initial state. No unconstrained residual is identified with God, and neither seven physical dimensions nor P=NP is established.

## Appendix Z. Fresh paired validation rejects a universal allocation replacement

Revision 1.25, 15 September 2026, adds Rope Tomography Lab 0.12.0. It starts from the saved paper v1.24, rope archive 0.11.0, and their current claim ledgers and revision notes. The audited RRLS v0.27 source was reread to confirm the retained collision, angular-impulse, recurrence-class, and complexity boundaries. RRLS code and evidence are unchanged.

### Z.1. Frozen question, assumptions, and decision rule

Appendix Y selected point fraction 0.45 and inner mean weight 0.10 on training-only synthetic evidence but prohibited adoption without a fresh validation. `candidate_validation_protocol.json` therefore froze one untouched seed, 20,000 paired 200-look histories per scenario, both allocation coordinates, scenario generators, metrics, source hashes, and decision rule before outcomes. The runner never reads the v1.22 validation observations or the v1.23-v1.24 training seeds and reference paths.

The released design uses point fraction 0.30 and mean weight 0.25. The candidate uses 0.45 and 0.10. Both retain the same total 1% budget, fixed mean alternative \(\delta=0.35\), fixed variance factors \(1.5625\) and \(0.64\), and equal variance weights. Under Appendix W's complete conditional Gaussian null, the written e-process-plus-union-bound argument applies to either allocation because the weights are fixed independently of the scored histories.

The preregistered engineering rule required all four conditions:

1. increase the minimum detection rate over the four declared targets by at least one percentage point;
2. give a positive descriptive paired 95% interval on the released design's bottleneck target;
3. keep every target's paired interval lower bound at or above minus two percentage points; and
4. increase the finite Gaussian-null rejection rate by no more than 0.5 percentage points.

The normal intervals summarize finite paired Bernoulli differences; they are not universal coverage guarantees. The one- and two-point thresholds express a declared utility, not facts derived from physics. Multi-objective optimization generally seeks a Pareto set rather than pretending that one coordinate dominates every objective; Inatsu et al., *Bounding Box-based Multi-objective Bayesian Optimization of Risk Measures under Input Uncertainty* (2024), https://proceedings.mlr.press/v238/inatsu24a.html, supplies relevant primary methodological context. Cawley and Talbot's selection-bias analysis, https://www.jmlr.org/papers/v11/cawley10a.html, supports keeping the candidate-selection samples separate from this evaluation. Neither paper supplies evidence for the physical framework.

### Z.2. Fresh synthetic results and counterexample

| Scenario | released 0.30/0.25 | candidate 0.45/0.10 | candidate minus released |
|---|---:|---:|---:|
| iid \(N(0,1)\) | 170 / 20,000 | 174 / 20,000 | +0.020 points |
| iid \(N(0.35,1)\) | 18,015 / 20,000 | 16,982 / 20,000 | **-5.165 points** |
| iid \(N(0,1.25^2)\) | 18,043 / 20,000 | 18,005 / 20,000 | -0.190 points |
| iid \(N(0,0.8^2)\) | 16,986 / 20,000 | 16,886 / 20,000 | -0.500 points |
| noisy signed 6-sigma shock | 16,287 / 20,000 | 16,674 / 20,000 | **+1.935 points** |
| iid \(N(0.35,1.25^2)\) | 19,822 / 20,000 | 19,767 / 20,000 | -0.275 points |
| alternating \(+0.35,-0.35\) means | 713 / 20,000 | 724 / 20,000 | +0.055 points |
| exact signed 6-sigma shock | 20,000 / 20,000 | 20,000 / 20,000 | 0 points |
| variance-one Student \(t_3\) | 15,253 / 20,000 | 15,422 / 20,000 | +0.845 points |

For the noisy-shock bottleneck, the paired difference was 1.935 points with descriptive normal 95% interval 1.740 to 2.130 points. The candidate therefore passed the minimum-improvement and positive-bottleneck checks. On persistent mean error, however, the paired interval was -5.473 to -4.857 points. This violates the frozen two-point loss guardrail, so the adoption decision is **retain the released design**.

The result is also a concrete counterexample to universal dominance. The candidate improves the chosen minimax bottleneck but materially weakens the persistent-mean target. Conversely, the released design is not uniformly better because it misses 387 additional noisy-shock histories. Which tradeoff is preferable cannot be inferred from predictive scores without an externally declared utility.

The Gaussian-null counts, 0.850% and 0.870%, are finite diagnostics rather than proofs of the 1% anytime bound. The Student \(t_3\) control again violates Gaussianity and produces rejection rates of 76.265% and 77.110%; neither design is distribution-free. Alternating shifts remain weakly detected. These controls preserve the previously known model failures.

Four new protocol tests and all 85 inherited tests passed before outcomes, for 89 total. An independent implementation regenerated every scenario and separately implemented the mean, variance, mixture, Gaussian-tail, interval, and adoption calculations. It matched all 72,000,000 saved look-by-design decision states and the final recommendation. It shares the seed, NumPy generator, synthetic model, and floating-point convention, so it is implementation independence rather than external replication.

### Z.3. Supported correction, evidence classes, and next test

The supported correction is to reject the 0.45/0.10 candidate as a balanced replacement under the frozen utility while retaining it as a documented shock-sensitive alternative. `selected_three_way_design.json` and the core runtime remain unchanged. The result does not retroactively alter the sealed v1.22 validation or Appendix W's mathematical guarantee.

The probability bound is written mathematics. The comparison, paired intervals, counterexamples, and audit are finite computation on synthetic data. No measured physical data were collected. Transfer to a real instrument remains a hypothesis, and neither predictive rejection nor a residual identifies a unique physical, metaphysical, or theological cause.

The next bounded test should address the shared failure rather than retune these Gaussian weights again: predeclare a heavy-tail null class and construct an anytime-valid robust process for that class. The first checkpoint must be a written supermartingale or other time-uniform validity argument; only afterward should implementation tests and fresh synthetic calibration be run. Until such a proof exists, preserve the Gaussian and heavy-tail models as distinguishable assumptions rather than relabeling either as reality.

All earlier corrections remain in force: complete deterministic closure for one fixed transition family is idempotent; integer-tick point capture is not continuous non-crossing; angular impulse has no extra time factor; finite prefixes do not identify arbitrary generators; and the 2K continuation theorem assumes an exact noiseless homogeneous constant-coefficient recurrence of order at most K valid from the initial state. No unconstrained residual is identified with God, and neither seven physical dimensions nor P=NP is established.

## Appendix AA. An anytime-valid sign process for a declared heavy-tail class

Revision 1.26, 15 September 2026, adds Rope Tomography Lab 0.13.0. It starts from the saved paper v1.25 and rope archive 0.12.0. The current claim ledger, revision notes, and unchanged audited RRLS v0.27 source were reread. RRLS's collision, angular-impulse, recurrence-class, and complexity boundaries remain unchanged.

### AA.1. The bounded question and null class

Appendix Z's next question was whether one useful adequacy check could remain anytime-valid without Gaussian tails. “Heavy-tail robustness” alone is not a mathematical model. This appendix narrows it to a sign-only claim for one specified candidate and one past-measurable projected residual. Let

\[
S_t=\operatorname{sign}(Z_t)\in\{-1,0,+1\},
\qquad
\mathbb E[S_t\mid\mathcal F_{t-1}]=0. \tag{AA.1}
\]

Here \(Z_t\) is the selected projected residual, and exact zeros are neutral. The candidate, direction, experiment choice, \(\rho\in(0,1)\), and threshold must be fixed or chosen from past information before \(S_t\) is observed. Equation (AA.1) is **conditional sign balance**. It permits Gaussian, Student-\(t\), Cauchy, heteroscedastic, and other magnitude laws if their current sign remains balanced conditional on the past. It is neither a mean-zero condition nor a consequence of marginal 50/50 signs.

This construction follows the nonnegative-martingale foundation and symmetry-testing perspective in Ramdas, Ruf, Larsson, and Koolen, *Admissible anytime-valid sequential inference must rely on nonnegative martingales* (revised 5 November 2022), https://arxiv.org/abs/2009.03167, and the broader time-uniform methodology of Howard et al., *Time-uniform, nonparametric, nonasymptotic confidence sequences* (2021), https://projecteuclid.org/journals/annals-of-statistics/volume-49/issue-2/Time-uniform-nonparametric-nonasymptotic-confidence-sequences/10.1214/20-AOS1991.full. The elementary sign mixture below is an application, not a new theorem or evidence for the physical framework.

### AA.2. Written probability proof

Fix \(\rho\in(0,1)\) before scoring and define

\[
M_n^+=\prod_{t=1}^n(1+\rho S_t),\qquad
M_n^-=\prod_{t=1}^n(1-\rho S_t),\qquad
E_n=\tfrac12(M_n^++M_n^-). \tag{AA.2}
\]

Every factor is nonnegative. Under (AA.1),

\[
\mathbb E[1\pm\rho S_t\mid\mathcal F_{t-1}]=1,
\]

so both \(M_n^+\) and \(M_n^-\) are nonnegative martingales starting at one. Their fixed convex mixture \(E_n\) is also a nonnegative martingale starting at one. Ville's inequality therefore gives

\[
\Pr\!\left(\sup_{n\ge0}E_n\ge\frac1\alpha\right)\le\alpha. \tag{AA.3}
\]

Writing \(A_n=\sum_{t\le n}S_t\) and \(N_n=\sum_{t\le n}\mathbf1\{S_t\ne0\}\), the implemented stable form is

\[
\log E_n=\frac{N_n}{2}\log(1-\rho^2)
+\log\cosh\!\left(\operatorname{atanh}(\rho)A_n\right). \tag{AA.4}
\]

This is a written mathematical guarantee for each specified correct candidate under (AA.1). It does not protect all candidates simultaneously, does not renew after resetting an object, and does not combine with the existing 1%-level Gaussian engines. Any OR-combination would require a shared error budget or another valid joint proof.

### AA.3. Frozen finite synthetic validation

`sign_validation_protocol.json` froze \(\alpha=0.01\), \(\rho=0.30\), a fresh seed, 5,000 histories, 200 looks, eleven scenarios, source hashes, and interpretation before outcomes. The known-scale Gaussian mean process was run separately at its own 1% threshold only as a diagnostic; the two procedures were not combined.

| Scenario | sign process | Gaussian comparator |
|---|---:|---:|
| iid \(N(0,1)\) | 53 / 5,000 | 36 / 5,000 |
| variance-one Student \(t_3\) null | 37 / 5,000 | 82 / 5,000 |
| exact doubled \(t_3\) null paths | 37 / 5,000 | 1,742 / 5,000 |
| standard Cauchy null | 33 / 5,000 | 4,994 / 5,000 |
| shifted \(t_3,+0.35\) | 4,989 / 5,000 | 4,793 / 5,000 |
| shifted \(t_3,-0.35\) | 4,992 / 5,000 | 4,800 / 5,000 |
| shifted Gaussian (+0.35) | 3,996 / 5,000 | 4,747 / 5,000 |
| mean-zero asymmetric (+1/-9) mixture | 5,000 / 5,000 | 4,177 / 5,000 |
| marginally balanced persistent sign chain | 3,772 / 5,000 | 3,835 / 5,000 |
| alternating shifted \(t_3\) | 12 / 5,000 | 72 / 5,000 |
| one signed six-unit \(t_3\) shock | 49 / 5,000 | 95 / 5,000 |

The Gaussian, unshifted \(t_3\), doubled \(t_3\), and Cauchy sign rows satisfy the sign theorem. Their finite crossing counts do not prove (AA.3); the martingale argument does. The doubled paths produced exactly the same sign decisions path by path, demonstrating scale invariance. The Gaussian comparator's guarantee is false on the non-Gaussian rows, and its Cauchy and doubled-\(t_3\) counts illustrate that failure rather than a calibrated comparison.

The shifted-\(t_3\) rows show strong finite targeted power at the frozen \(\rho\), while the Gaussian shift preserves the expected efficiency cost of discarding magnitudes. No universal power claim follows from one seed, horizon, or alternative.

### AA.4. Counterexamples and interpretation

The asymmetric mixture with probability 0.9 at \(+1\) and 0.1 at \(-9\) has exact expectation zero but violates sign balance. All 5,000 histories crossed. Thus “mean zero” cannot silently replace (AA.1). The stationary sign chain has marginal probability one half for each sign but retains its previous sign with probability 0.9, so its current conditional expectation is usually nonzero; 3,772/5,000 paths crossed. Thus marginal balance cannot silently replace conditional balance.

Positive rescaling never changes the statistic, so it cannot detect pure scale errors or tail thickness. The alternating-shift and isolated-shock rows show further power failures. A crossing supports incompatibility with conditional sign balance around the specified prediction; it does not identify a mean error, sensor failure, physical mechanism, metaphysical cause, or God.

Eight engine tests and four frozen-protocol tests were added; all 101 tests passed before outcomes. An independently written audit regenerated every scenario and separately recomputed the sign and Gaussian formulas, every crossing path, and every summary. It matched 22,000,000 saved decision states. The audit shares the seed, NumPy samplers, scenario definitions, and floating-point arithmetic, so it is implementation independence rather than external replication.

### AA.5. Supported change, evidence classes, and next test

`ConditionalSignEProcess` is added as a standalone selectable engine. It does not replace or silently broaden the Gaussian mean/variance/point engine. The conditional-sign anytime bound is written mathematics; unit tests, synthetic paths, counterexamples, and the independent replay are finite computation. No measured physical evidence was collected. Application to real residuals remains a hypothesis requiring instrument-specific checks of sign balance and dependence.

The next bounded test is the dependence boundary exposed above: construct a sign process valid under a declared dependence model or a defensible predictable conditional-sign model, then test it against stationary marginally balanced chains without reusing this validation set. Until that proof exists, correlated signs must remain outside the guarantee.

All earlier corrections remain in force: complete deterministic closure for one fixed transition family is idempotent; integer-tick point capture is not continuous non-crossing; angular impulse has no extra time factor; finite prefixes do not identify arbitrary generators; and the 2K continuation theorem assumes an exact noiseless homogeneous constant-coefficient recurrence of order at most K valid from the initial state. No unconstrained residual is identified with God, and neither seven physical dimensions nor P=NP is established.

## Appendix AB. Composite Born prediction, coded lift search, and projection collisions

Revision 1.27, 15 September 2026, adds Born Outcome Frontier Audit 0.2.0. It starts from the saved paper v1.26, the court-audit phase/Born formalization, and the current RRLS claim ledgers. Rope Tomography Lab 0.13.0 and RRLS Studio 0.27.1 are unchanged. No physical measurement data were collected.

### AB.1. One theory, three different uses of seven and eight

Three objects must remain separate:

1. \((D_1,\ldots,D_7)\) is the declared hierarchy of seven **typed descriptive layers**.
2. Q1 uses eight computational-basis states of three qubits and therefore eight raw basis phases.
3. Removing common phase leaves seven relative phase coordinates because

\[
\mathbb T^8/S^1\cong\mathbb T^7.
\]

The numerical equality between the number of typed layers and the number of relative Q1 phases is not yet a derived physical identification. A bridge between them must be a typed map with units, transformation law, dynamics, and a measurement consequence. Treating the matching numeral as proof would violate coordinate covariance A2.

### AB.2. Exact and composite Born walls

At trial \(t\), let \(\mathcal F_{t-1}\) contain all admissible previous information and let \(C_t\) contain the complete declared standard context. An outcome predictor \(q_t\) must be fixed using information available before \(Y_t\). For an exact Born conditional law \(p_t\),

\[
\mathbb E_{p_t}\!\left[
\log\frac{q_t(Y_t)}{p_t(Y_t)}
\middle|\mathcal F_{t-1},C_t
\right]
=-D_{\mathrm{KL}}(p_t\Vert q_t)\le0. \tag{AB.1}
\]

The likelihood-ratio product is therefore a nonnegative martingale. This is the Born Prediction Wall: more computation cannot extract information that is absent from its admissible inputs.

Real calibration does not usually specify one exact \(p_t\). Let \(\mathcal P_t\) be the predictable set of conditional distributions retained by the calibrated null and define

\[
u_t(y)=\sup_{p\in\mathcal P_t}p(y).
\]

For any normalized predictable \(q_t\), put

\[
e_t(Y_t)=\frac{q_t(Y_t)}{u_t(Y_t)}.
\]

If the actual \(p_t^*\in\mathcal P_t\), then

\[
\mathbb E_{p_t^*}[e_t(Y_t)\mid\mathcal F_{t-1},C_t]
=\sum_y p_t^*(y)\frac{q_t(y)}{u_t(y)}
\le\sum_yq_t(y)=1. \tag{AB.2}
\]

Thus \(\prod_{s\le t}e_s(Y_s)\) is a nonnegative supermartingale under every admissible sequence in the composite null, and Ville's inequality remains valid. The coordinatewise envelope is generally conservative; its power loss is preferable to converting state, POVM, detector, or drift uncertainty into fake new physics. Broader composite-likelihood methods belong to established universal inference [Wasserman, Ramdas, and Balakrishnan (2020)](https://www.pnas.org/doi/10.1073/pnas.1922664117).

Exact zeros require a declared support rule. If every null distribution assigns probability zero to an observed outcome, that outcome rejects the stated null family. If the zero was produced only by finite calibration or numerical truncation, the null was misspecified; an outcome-dependent floor is not a valid repair.

### AB.3. Prefix-coded dimensional lifts

Let \(L_k\) be a countable frozen grammar of pre-measurement lift compilers and let \(\ell_k\) be prefix-code lengths satisfying

\[
\sum_k2^{-\ell_k}\le1.
\]

If \(E_{t,k}\) is a valid e-process for lift \(k\), then

\[
E_t^K
=1-\sum_k2^{-\ell_k}
+\sum_k2^{-\ell_k}E_{t,k} \tag{AB.3}
\]

is a fixed convex mixture of valid e-processes and unspent capital. It is therefore valid. A simple sufficient condition for candidate \(k\)'s weighted term to cross the threshold is

\[
\log E_{t,k}\ge\log(1/\alpha)+\ell_k\log2. \tag{AB.4}
\]

Every description bit costs one bit of evidence. This supplies a lawful statistical interface between RRLS and outcome prediction: phase unwraps, winding registers, parity lifts, Walsh interactions, lags, and boundary invariants may be searched only through a grammar and code fixed before validation.

Description length controls search multiplicity, not computational complexity. The RRLS language must still count recognition, transformation, decoding, representation growth, arithmetic bit costs, and total work. A compact but exponentially expensive oracle is not a polynomial predictor.

For a balanced 51%-accurate binary candidate at one-sided five sigma and approximate 90% power, the implementation gives:

| Prefix-code length | Sufficient candidate \(\log E\) | Approximate trials |
|---:|---:|---:|
| 0 bits | 15.064 | 119,635 |
| 8 bits | 20.609 | 153,191 |
| 16 bits | 26.154 | 186,030 |
| 32 bits | 37.244 | 250,321 |

These are planning approximations derived from the Bernoulli log-likelihood mean and variance. The e-process, not the normal approximation, supplies the primary anytime decision rule.

### AB.4. Projection collisions are universal decoder failures

Let \(\Lambda\) be the full admissible pre-measurement record and let

\[
G=g(\Lambda,C)
\]

be an RRLS-produced canonical address. Conditional data processing gives the exact identity

\[
I(Y;\Lambda\mid C)
=I(Y;G\mid C)
+I(Y;\Lambda\mid G,C), \tag{AB.5}
\]

and hence

\[
I(Y;G\mid C)\le I(Y;\Lambda\mid C). \tag{AB.6}
\]

The compression is predictively lossless exactly when the discarded coordinate has zero remaining conditional information. A collision

\[
G(\lambda,C)=G(\lambda',C),
\qquad
P(Y\mid\lambda,C)\ne P(Y\mid\lambda',C) \tag{AB.7}
\]

proves that no decoder receiving only \(G\) can be universally outcome-sufficient. This is structurally identical to the retained RRLS collision in which a SAT and an UNSAT formula share the full implemented invariant profile: computation downstream cannot reconstruct a distinction the projection erased.

The repair target is boundary compatibility, not \(R\to0\). In the quantum experiment the shared boundary contains the prepared-state family, allowed settings, timing, device identity, and interventions. A canonicalizer must preserve the conditional relation from that boundary to future outcome distributions.

The audit's fixed finite table contains \(0.192744757022\) nats of outcome information before coarsening and zero afterward. Across 5,000 randomized finite tables, the greatest numerical data-processing excess was \(2.22\times10^{-16}\) in the production implementation and zero in the independent replay. This is consistent with floating-point roundoff and the theorem; it is not a measurement of nature.

### AB.5. Foundational assumption map

Outcome determinism does not imply accessible prediction. Bohmian mechanics is deterministic, but its quantum-equilibrium analysis recovers \(\rho=|\psi|^2\) statistics and an absolute-unpredictability boundary [Dürr, Goldstein, and Zanghì (1992)](https://link.springer.com/article/10.1007/BF01049004).

Colbeck and Renner's no-improved-prediction conclusion is conditional on their free-choice formulation and quantum correctness. Ghirardi and Romano exhibit predictively inequivalent ontological models under a different free-choice assumption [Ghirardi and Romano (2013)](https://link.aps.org/doi/10.1103/PhysRevLett.110.170404). This exposes an assumption boundary; it does not validate an arbitrary hidden variable.

Valentini's proposed nonequilibrium subquantum information supplies another logically possible route but also a sharp burden: deterministic hidden-variable nonequilibrium generally exposes instantaneous statistical signaling [Valentini (2002)](https://arxiv.org/abs/quant-ph/0112151). Any such Dimensional Reality route requires an operational no-signaling analysis rather than a verbal assurance.

Prediction-based likelihood ratios are established in Bell analysis. Zhang, Glancy, and Knill's PBR protocol remains valid against memory-dependent local-realistic models even when prepared states vary [Zhang, Glancy, and Knill (2011)](https://link.aps.org/doi/10.1103/PhysRevA.84.062118). This project does not claim to invent likelihood-ratio Bell testing. Its candidate synthesis is the combination of projection-collision search, physically specified winding addresses, a composite Born envelope, and a frozen prefix-coded lift language.

The attribution gates are strict:

\[
\text{predictive residual}
\not\Rightarrow
\text{causal variable}
\not\Rightarrow
\text{new dimension}
\not\Rightarrow
\text{complete physical theory}. \tag{AB.8}
\]

A candidate must successively survive immutable ordering, leakage controls, tomography and detector uncertainty, process-tensor memory tests, causal breaks, resets, sign and magnitude reversal, independent platforms, independent laboratories, Bell/contextuality/free-setting constraints, no-signaling, Lorentz behavior, and the classical limit.

### AB.6. Executed evidence, preserved failures, and next test

Born Outcome Frontier Audit 0.2.0 and its independent verifier pass. The independent file imports none of the production module and reconstructs:

- the fixed composite-null multiplier \(0.962307692308\);
- the fixed information loss \(0.192744757022\to0\) nats;
- 5,000 randomized composite-envelope and data-processing cases;
- a 120,000-trial Born-null path with maximum \(\log E=1.17364005363\);
- a synthetic accessible-51%-lift path with maximum \(\log E=28.3649956267\);
- all four prefix-code power rows.

The independent replay shares the formulas, Python standard library, pseudorandom generator, seeds, and floating-point convention. It is implementation independence, not external replication. The positive world was deliberately generated with a predictive variable and therefore cannot support a claim that a natural one exists.

The next bounded test is not another synthetic retuning. It is a protocol artifact frozen before fresh hardware data:

1. specify a Q1 state and measurement family;
2. specify the composite Born calibration set \(\mathcal P_t\);
3. specify a prefix-coded admissible lift grammar;
4. prohibit all setting, outcome, and downstream timing leakage;
5. commit setting-conditional predictions before delayed random settings;
6. spend one investigation-wide evidence budget;
7. advance any crossing through the full attribution gates without altering its predicted sign, magnitude, or exclusions.

No physical Born residual, winding coupling, extra dimension, soul-mediated outcome rule, P=NP proof, or universal generator has been found. The supported advance is a stricter research constitution that makes those claims harder to fake and easier to kill.

## Appendix AC. Information rate, adversarial tripwires, and the wall ledger

Revision 1.28, 15 September 2026, adds Wallbreaker Audit 0.1.0 and Wallbreaker Protocol P3 0.1.0. It starts from the saved paper v1.27 and preserves every earlier correction. Born Outcome Frontier Audit 0.2.0, Rope Tomography Lab 0.13.0, and RRLS Studio 0.27.1 are unchanged. No physical measurements were collected.

### AC.1. General Born-evidence decomposition

Let \(C\) be the complete declared standard context, \(\Lambda\) an admissible pre-measurement lift, and \(b(y\mid C)\) the declared Born baseline. Write

\[
\bar r(y\mid C)=P(Y=y\mid C),
\qquad
r(y\mid C,\Lambda)=P(Y=y\mid C,\Lambda)
\]

for the actual marginal and lift-conditional laws. For any normalized decoder \(q(y\mid C,\Lambda)\), insert and subtract \(\log r\) and \(\log\bar r\):

\[
\begin{aligned}
\mathbb E\log\frac{q(Y\mid C,\Lambda)}{b(Y\mid C)}
&=\mathbb E\log\frac{r(Y\mid C,\Lambda)}{\bar r(Y\mid C)}
+\mathbb E\log\frac{\bar r(Y\mid C)}{b(Y\mid C)}\\
&\quad-\mathbb E\log\frac{r(Y\mid C,\Lambda)}{q(Y\mid C,\Lambda)}\\
&=I(Y;\Lambda\mid C)
+\mathbb E_C D_{\mathrm{KL}}(\bar r\Vert b)
-\mathbb E_{C,\Lambda}D_{\mathrm{KL}}(r\Vert q).
\end{aligned}
\tag{AC.1}
\]

Equation (AC.1) separates three effects that must not be conflated:

1. \(I(Y;\Lambda\mid C)\) is genuine conditional predictive information in the lift;
2. \(D_{\mathrm{KL}}(\bar r\Vert b)\) is ordinary marginal disagreement with Born's rule;
3. \(D_{\mathrm{KL}}(r\Vert q)\) is information lost by an imperfect decoder.

For a Born-marginal-preserving extension, \(\bar r=b\), so

\[
\boxed{
\mathbb E\log\frac{q(Y\mid C,\Lambda)}{b(Y\mid C)}
=I(Y;\Lambda\mid C)
-\mathbb E D_{\mathrm{KL}}(r\Vert q)
\le I(Y;\Lambda\mid C).
}
\tag{AC.2}
\]

Equality holds exactly when \(q=r\) on the supported contexts. The missing resource is therefore conditional information, not computation. More computation can decrease decoder loss; it cannot produce information absent from the admissible boundary record.

Under stationary independent trials, an exact decoder, integrable log increments, and marginal Born preservation, the strong law gives

\[
\frac{1}{T}\log E_T\longrightarrow I(Y;\Lambda\mid C)
\quad\text{almost surely}. \tag{AC.3}
\]

For a prefix-coded candidate of length \(\ell\), the characteristic evidence scale is

\[
T_{\mathrm{scale}}
\approx
\frac{\log(1/\alpha)+\ell\log2}{I(Y;\Lambda\mid C)}. \tag{AC.4}
\]

This is an asymptotic planning scale, not an exact expected stopping time. Dependence, overshoot, composite-null conservatism, and misspecification change finite performance.

Positive information is not determinism. Exact outcome prediction requires

\[
H(Y\mid C,\Lambda)=0. \tag{AC.5}
\]

Fano's inequality relates remaining conditional entropy to the optimal classification error but does not turn a weak advantage into certainty.

### AC.2. No finite data determine an unrestricted continuation

Let \(S\) be a proper finite subset of an unrestricted context set \(\mathcal C\). Given any law \(p(y\mid c)\), choose \(c_*\notin S\) and construct \(p'\) equal to \(p\) on \(S\) but different at \(c_*\). Every observation confined to \(S\) has identical likelihood under both laws. Therefore no finite dataset restricted to \(S\) can decide their continuation at \(c_*\).

This **No-Free-Extension theorem** does not prevent generalization inside a restricted law class whose extension is fixed by a proved relation. It forbids upgrading finite agreement to an unrestricted universal claim without such a restriction.

The executable collision uses eight contexts. Two laws have total-variation distance zero on the first seven and distance \(0.8\) on the eighth. This is the predictive analogue of the RRLS retained-invariant collision: a continuation distinction not present in the observed boundary cannot be reconstructed downstream.

### AC.3. Three-channel tripwire

A claimed accessible lift is tested with three distinct processes:

\[
E_T^{\mathrm{outcome}},\qquad
E_T^{\mathrm{setting}},\qquad
E_T^{\mathrm{signal}}. \tag{AC.6}
\]

The outcome channel asks whether the lift predicts \(Y\) beyond the complete calibrated context. The setting channel asks whether it predicts a supposedly delayed free setting. The signal channel asks whether a local outcome depends on a remote setting. The target pattern is outcome evidence without setting or signaling evidence, plus structural timing isolation and affirmative equivalence controls. Failure to reject the latter nulls does not prove exact independence.

Six seeded 50,000-trial worlds give:

| Synthetic world | Complete-context outcome crossing | Setting crossing | Signaling crossing | Interpretation |
|---|---:|---:|---:|---|
| Born null | no | no | no | correct negative control |
| planted accessible lift | yes | no | no | target signature; synthetic only |
| future-setting leak | yes | yes | yes | chronology and measurement-independence failure |
| remote signaling | no | yes | yes | naive evidence vanishes after completing the context |
| apparatus memory | no | no | no | naive evidence vanishes after modeling memory |
| memory after reset | no | no | no | reset destroys the artifact |

The planted world is not a physical construction. It only verifies that the audit recognizes a deliberately inserted conditional-information channel.

The production and independent implementations also reproduce:

- a marginal-preserving optimal information rate \(0.368064207168\) nats;
- a nonpreserving example \(0.388475204429=0.368064207168+0.0204109972601\) nats;
- 5,000 randomized special and general decompositions;
- maximum algebraic disagreement \(1.33\times10^{-15}\) in production and \(1.11\times10^{-15}\) independently.

These are floating-point checks of the written identities, not measurements of nature.

### AC.4. Bell, contextuality, and measurement-independence routes

A hidden-variable proposal must identify its route before data. The relevant alternatives are not interchangeable:

| Route | Changed premise | Unavoidable burden |
|---|---|---|
| local and measurement-independent | none | cannot reproduce Bell-violating quantum correlations |
| measurement-dependent | free-setting condition | quantify lift-setting dependence and stop calling the setting independent |
| nonlocal equilibrium | local causality | explains determinism but ordinarily supplies no accessible BPW gain |
| nonlocal nonequilibrium | equilibrium and locality | derive preparation/access and test signaling |
| contextual | noncontextual value assignment | include the actual measurement context in the response law |
| retrocausal | forward-only causal order | define spacetime accessibility and precommit every setting-conditional prediction |
| superdeterministic | measurement independence | impose quantitative restrictions preventing unconstrained retrospective explanation |

Kochen and Specker exclude the relevant noncontextual value assignments, not every contextual ontology [Kochen and Specker (1967)](https://www.jstor.org/stable/24902153). Colbeck and Renner exclude improved prediction under their explicit QM and free-choice assumptions [Colbeck and Renner (2011)](https://www.nature.com/articles/ncomms1416). Hall constructs a deterministic no-signaling singlet model after relaxing measurement independence and quantifies that relaxation [Hall (2010)](https://link.aps.org/doi/10.1103/PhysRevLett.105.250404). Assumption sensitivity is real; it is not a license to omit the cost of the selected route.

Bell-certified randomness establishes a different direction: observed nonlocal correlations can certify randomness under stated independence and no-signaling assumptions [Pironio et al. (2010)](https://arxiv.org/abs/0911.3427). A Dimensional Reality outcome predictor must confront that evidence rather than assume that determinism automatically implies accessible foresight.

### AC.5. The thirteen walls and the next physical test

Wallbreaker Protocol P3 assigns a pass condition to thirteen walls:

1. internal typing;
2. Born information;
3. projection collision;
4. chronology;
5. measurement independence;
6. context completeness;
7. no-signaling;
8. contextuality;
9. Bell route;
10. Lorentz behavior, units, scale, and classical limit;
11. finite-context extension;
12. computational complexity;
13. external replication.

The first, second, and eleventh walls now have stronger mathematics. The audit provides machinery for the fourth through seventh. It does not physically break them. The next bounded experiment must instantiate P3 on fresh hardware data by freezing the apparatus, state/POVM family, composite calibration set, admissible timestamped fields, prefix grammar, prediction commitments, delayed settings, evidence budgets, equivalence margins, resets, reversals, and replication rules before outcomes.

No finite synthetic result can discharge that physical burden. No natural predictive lift, new spacetime coordinate, soul-mediated outcome law, Lorentz-covariant coupling, universal quantum speedup, or P=NP proof has been obtained. The advance is an exact wall map and an adversarial instrument capable of distinguishing several false breakthroughs from the target signature.

## Appendix AD. P3-R1: an executable single-device hardware frontier

### AD.1. Bounded physical question and claim ceiling

P3-R1 converts the next physical step into one bounded question: can one of three frozen eight-cycle coordinates predict a held-out three-qubit measurement result beyond a calibrated coordinatewise upper envelope for the ordinary Born/hardware context, when its complete prediction for every allowed setting is committed before the setting is randomly chosen?

The source state is the Q1 equal-magnitude phase state

\[
|\psi\rangle=\frac1{\sqrt8}\sum_{x=0}^{7}e^{i\phi_x}|x\rangle,
\]

with eight frozen phases. `Z_ALL` measures it directly; `X_ALL` applies three Hadamards and then measures. The target record is one three-bit outcome in Qiskit's frozen \(q_2q_1q_0\) order. Every validation trial is a separate circuit/PUB with one shot, so aggregated counts are not retroactively split into fictitious individual trials.

This is a single-device pilot. It is not spacelike separated and therefore cannot test physical no-signaling, contextuality, or Bell locality. Even a valid outcome crossing is capped at “the declared single-device composite null was rejected” until memory, drift, reset, reversal, causal, and external-replication gates pass.

### AD.2. An exactly marginal-preserving eight-cycle grammar

Let \(b_c(y)>0\) be the ideal Q1 Born reference for setting \(c\), let \(w=t\bmod8\), and let \(\sigma_j\) be one of three fixed permutations: forward, reverse, or binary-reflected Gray order. Define

\[
a_{j,w,c}=\sigma_j(w)\oplus m_c,
\qquad
d_{j,w,c}=a_{j,w,c}\oplus4,
\]

and

\[
\Delta_{j,w,c}=\varepsilon
\min\{b_c(a_{j,w,c}),b_c(d_{j,w,c})\},
\qquad \varepsilon=0.12.
\]

The candidate transfers \(\Delta) from \(d\) to \(a\):

\[
q_{j,w,c}=b_c+\Delta_{j,w,c}(e_a-e_d).
\tag{AD.1}
\]

Because \(0<\varepsilon<1\), (AD.1) remains strictly positive and normalized. Over a complete cycle, each target appears once; the winding whose target is \(a\oplus4\) executes the reverse transfer with the same minimum. Consequently,

\[
\boxed{\frac18\sum_{w=0}^{7}q_{j,w,c}=b_c}
\tag{AD.2}
\]

for every candidate and setting. This makes the probes conditional-information tests in their mathematical reference model rather than disguised marginal Born shifts. Their prefix lengths are 2, 3, and 3 bits. The resulting weights \(1/4,1/8,1/8\) use half the unit mass; the remaining half is cash.

Equation (AD.2) is a construction theorem, not a reason nature must implement the winding variable. The three candidates are deliberately small probes chosen before physical data, not a learned law.

### AD.3. Composite-null score and delayed-setting tripwire

Independent calibration supplies a frozen upper envelope \(u_c(y)\). The run builder uses a simultaneous fixed-sample Hoeffding radius plus a predeclared per-outcome drift allowance. That allowance is an empirical assumption requiring interleaved stress tests; calibration cannot logically guarantee future stationarity.

If the complete ordinary hardware law obeys \(p_t(y\mid C_t)\le u_t(y)\), then any normalized, predictable candidate satisfies

\[
\mathbb E\!\left[
\frac{q_{j,t}(Y_t)}{u_t(Y_t)}\middle|\mathcal F_{t-1}
\right]
=\sum_y p_t(y\mid C_t)\frac{q_{j,t}(y)}{u_t(y)}\le1.
\tag{AD.3}
\]

Thus each product is a nonnegative supermartingale, and the Kraft/cash mixture is one as well. The primary anytime threshold is \(1/(2.87\times10^{-7})\). A separate predictable setting process tests whether the same lift forecasts the delayed balanced setting at \(\alpha=10^{-3}\). A setting crossing disqualifies an outcome crossing; failure to cross is recorded only as `NO_CROSS_NOT_EQUIVALENCE`, never as proof of exact setting independence.

All setting-conditional outcome vectors and setting-tripwire probabilities enter a SHA-256 prediction chain before any validation setting is generated. Every setting points back to its prediction digest. The pre-run manifest binds the config, calibration, commitments, and settings; a second manifest binds provider job records, raw outcomes, and score. Local hashes detect alteration relative to the manifests but do not prove public time. A court-grade execution must externally timestamp the pre-run manifest before submission.

The current provider interface supplies job-submission and result-retrieval times, not authenticated device-level timestamps for individual shots. P3-R1 therefore proves that all settings existed before submission, not that the unavailable timestamp was measured.

### AD.4. Executed adversarial software evidence

Production and independent implementations were tested on three 20,000-trial seeded worlds:

| World | Maximum outcome log mixture | Outcome threshold crossed? | Maximum setting log mixture | Setting threshold crossed? | Interpretation |
|---|---:|---:|---:|---:|---|
| Exact Born reference | 0.279025 | No | 4.631600 | No | Correct negative control |
| Frozen forward lift | 25.026590 | Yes | 1.346171 | No | Intended synthetic target |
| Forward lift plus setting dependence | 18.835820 | Yes | 1094.733929 | Yes | Correctly disqualified |

The independent replay imports none of the production runner. It reconstructs the phase-state Born probabilities, all prediction vectors, the hash chain, chronology, marginal-preservation identity, and both e-processes. It agrees with the production scores. The run-specific maximum marginal error is \(5.55\times10^{-17}\). A further 5,000 randomized property audit keeps the eight-cycle marginal error below \(3.33\times10^{-16}\) and finds no violation of the composite-null one-step bound. Deliberately changing a committed trial makes the verifier fail at the file digest. These are software and synthetic tests only.

### AD.5. Activation boundary and next honest wall

The shipped configuration has three intentional physical blockers: no fresh run UUID, no explicitly selected accessible QPU, and no real calibration envelope/artifact. Qiskit and Qiskit Runtime are also not installed in the present audit environment. The runner refuses live freeze or submission until those conditions are met; credentials remain in the provider's saved-account mechanism and are never accepted in the config or command line.

Current IBM documentation describes authentication through `QiskitRuntimeService`, selection of an accessible backend, transpilation into that backend's instruction set, and execution through Runtime primitives [IBM Quantum hardware guide](https://quantum.cloud.ibm.com/docs/en/guides/hello-world). IBM Cloud access requires an account, API key, and instance identifier [IBM Quantum cloud setup](https://quantum.cloud.ibm.com/docs/en/guides/cloud-setup). Accessible QPUs depend on the account's instance and plan [IBM Quantum instances](https://quantum.cloud.ibm.com/docs/en/guides/instances).

The executable path is now calibration \(\to\) envelope \(\to\) prediction commitment \(\to\) delayed settings \(\to\) one-shot submissions \(\to\) independent replay. Nature has not yet supplied the final records. A negative pilot would reject only these three lifts at this scale. A positive pilot would trigger controls and replication, not a declaration that all randomness or every remaining wall had been broken.

## Appendix AE. Dimensional Chess C1: exact solution of a closed game universe

### AE.1. Finite does not mean computationally feasible

Once its rule state is complete and automatic termination is fixed, chess has a finite state graph and therefore a definite game-theoretic value at its initial position. That proves existence of a solution by finite backward induction. It supplies no useful bound on the work. Shannon's original estimate places the ordinary game tree near \(10^{120}\), illustrating why finiteness and feasible solvability must not be conflated [Shannon (1950)](https://vision.unipv.it/IA1/ProgrammingaComputerforPlayingChess.pdf).

C1 begins with a closed standard-board subuniverse: White king and rook against Black king; checkmate wins; stalemate, rook capture, or infinite play draws. Its complete state is

\[
s=(k_W,r_W,k_B,\tau).
\]

For full chess, visible piece placement is an invalid projection. Side to move, castling rights, en-passant state, halfmove clock, and repetition-relevant history can change the legal future while leaving the board picture unchanged. They must be retained as coordinates.

### AE.2. A collision-free symmetry lift

Let the dihedral group \(D_4\) act on the board and let

\[
\kappa(s)=\min_{g\in D_4}\operatorname{encode}(gs).
\]

The quotient is dynamically closed only if

\[
\{\kappa(s'):s'\in T(s)\}
=
\{\kappa(s'):s'\in T(gs)\}
\quad\text{for every }g\in D_4.
\tag{AE.1}
\]

C1 checks (AE.1), including check status, on every canonical state and all eight symmetries. All 400,120 comparisons pass. The 399,112 raw legal states collapse to 50,015 exact classes, a factor of \(7.979846046\ldots\). This is a lawful dimensional lift: it removes viewpoint redundancy without merging different future laws.

### AE.3. Time cycles become a least fixed point

Move repetition makes the game a graph rather than a tree. Starting from checkmates \(L_0\), C1 repeatedly applies

\[
W_{n+1}=W_n\cup\{s:\exists s'\in T(s),\ s'\in L_n\}
\]

and

\[
L_{n+1}=L_n\cup\{s:T(s)\neq\varnothing,\ T(s)\subseteq W_{n+1}\}.
\]

The complement of the least fixed point \(W\cup L\) is the draw region. Winning states minimize distance to mate; losing states maximize it. This treats a returned position as the same coordinate instead of creating an infinite sequence of duplicated nodes.

### AE.4. Exact result and proof certificate

The canonical graph contains 558,816 transitions including the terminal rook-capture draw:

| Value from side to move | Canonical states |
|---|---:|
| Win | 21,959 |
| Loss | 25,260 |
| Draw | 2,796 |

There are no White-to-move losses and no Black-to-move wins. The maximum forced mate is 32 plies, or 16 moves. One maximum state has White king a1, White rook b2, Black king c2, and Black to move. The stored optimal line contains 32 legal moves and ends with Black checkmated. The 16-move maximum agrees with independent published KRK verification [Stålmarck et al. (2013)](https://journals.sagepub.com/doi/pdf/10.3233/ICG-2013-36203?download=true).

An independent program imports none of the solver. It reconstructs all raw states, symmetry classes, transitions, terminal conditions, Bellman equations, distance equations, policy witnesses, and concrete principal-variation moves. It passes. After one WDL entry was changed and its file digest recomputed, the verifier still rejected the certificate because the local draw fixed-point equation failed. Hash integrity and mathematical correctness are therefore separately tested.

### AE.5. The wall exposed by chess

Retrograde tablebases and board symmetries are established ideas; C1 does not claim their invention. It demonstrates that the framework's core disciplines work on a complete finite universe:

- hidden coordinates restore transition closure;
- a projection is accepted only after collision testing;
- geometry supplies an exact symmetry quotient;
- cycles are solved as graph fixed points;
- the initial unknown value is reduced from \(\{W,D,L\}\) by constraints;
- every conclusion carries a local certificate;
- solving one restricted family never licenses an unrestricted extension.

Modern Syzygy infrastructure supplies exact tables for complete seven-piece endgames and partial eight-piece work [Lichess tablebase implementation](https://github.com/lichess-org/lila-tablebase). The C1 symmetry factor is powerful locally but leaves Shannon's scale essentially unchanged: \(10^{120}/8\approx1.25\times10^{119}\). Full chess therefore requires either immense enumeration or a new uniform value-preserving quotient whose recognition, transformation, solution, and verification costs are all bounded. Any proposed shortcut fails immediately if two merged positions have different minimax values or different quotient successor sets.

The experiment has solved its declared universe and identified the next missing variable precisely: not “more time,” but a drastically stronger collision-free canonicalizer.

## Appendix AF. Dimensional Chess C2: interval bounds, certified pruning, and the full-chess gate

### AF.1. Freeze the target before optimizing it

“Solve chess” now means determine the game-theoretic value of the normal initial
position under a frozen board-game ruleset and provide a strategy DAG that an
independent program can verify against every required adversarial reply. Playing
above human level, producing one principal variation, or solving additional
endgames does not meet this definition.

The frozen target uses legal moves, checkmate, stalemate, dead positions,
claimable threefold/50-move draws, and automatic fivefold/75-move draws from the
FIDE Laws effective 1 January 2023. Clocks, resignations, and negotiated draw
offers are excluded; a correct claim is an optional transition to a draw
terminal. Articles 9.2–9.6 define the relevant repetition and move-count rules
[FIDE Laws of Chess](https://handbook.fide.com/chapter/e012023). A 2026 preprint
derives a maximum game length of 17,697 plies under those Laws. That establishes
a finite depth, not a feasible tree size [Yim (2026)](https://arxiv.org/abs/2608.14762).

The Markov-complete state is

\[
S=(B,\tau,C,E,h,\mathcal R),
\tag{AF.1}
\]

where $B$ is current typed piece placement (including promoted material), $\tau$ is side to move, $C$ is
castling state, $E$ is en-passant state, $h$ is the reversible-move clock,
and $\mathcal R$ is the repetition-relevant path record. FIDE Article 9.2.3
explicitly makes possible moves, side to move, en passant, and castling rights
part of repetition identity. The same visible board with different hidden
coordinates can therefore have a different transition law or draw status.

### AF.2. D-/D+ becomes interval minimax

Use White-relative values $v(s)\in\{-1,0,+1\}$. An unresolved state carries

\[
I(s)=[D^-(s),D^+(s)],\qquad D^-(s)\le v(s)\le D^+(s).
\tag{AF.2}
\]

The computational zero axis begins as $[-1,+1]$, retaining all three possible
readouts without pretending they are equally probable. Exact terminals and
verified tablebases seed point intervals. Given sound bounds on every child,

\[
\begin{aligned}
D^-(s)&=\max_a D^-(T(s,a)),&D^+(s)&=\max_a D^+(T(s,a))&&\text{for White},\\
D^-(s)&=\min_a D^-(T(s,a)),&D^+(s)&=\min_a D^+(T(s,a))&&\text{for Black}.
\end{aligned}
\tag{AF.3}
\]

Only proved premises may tighten an interval. The ordinary W/D/L result is read
only when $D^-=D^+$. This preserves the framework's richer truth coordinate
while retaining exact minimax semantics.

### AF.3. Exact Disadvantage Pruning Theorem

Suppose move $a$ has sound interval $[L_a,U_a]$. At a White/max state, $a$
may be discarded when another move $b$ satisfies

\[
L_b\ge U_a.
\tag{AF.4}
\]

The proof is the chain $v_b\ge L_b\ge U_a\ge v_a$. At a Black/min state,
$a$ may be discarded when

\[
U_b\le L_a,
\tag{AF.5}
\]

because $v_b\le U_b\le L_a\le v_a$. Equality preserves the root value even
if it removes an equally optimal alternative. A merely heuristic score is not
a bound and cannot license either deletion. Sacrifices, quiet defensive moves,
and zugzwang make local-looking disadvantage unreliable.

The theorem formalizes the proposed shortcut: discard possibilities only after
their inability to affect the minimax result is certified. Alpha-beta bounds
are one special case.

### AF.4. Exact KQK extension

C2 independently implements the complete White-king-and-queen versus
Black-king universe under the same small-game terminal convention as C1:
checkmate wins; stalemate, queen capture, or infinite non-mating play draws.

| Quantity | KQK result |
|---|---:|
| Raw legal states | 368,452 |
| $D_4$-canonical states | 46,137 |
| Canonical transitions | 609,873 |
| Symmetry-transition comparisons | 369,096 |
| Side-to-move wins | 18,081 |
| Side-to-move losses | 25,160 |
| Draws | 2,896 |
| Maximum forced mate | 20 plies = 10 moves |

One maximum-distance state is White king a1, White queen b2, Black king e4,
Black to move. Its certified line contains 20 legal moves and ends in
checkmate. The verifier imports no production functions and reconstructs every
state, move, orbit, terminal, Bellman equation, distance, policy edge, and
concrete principal-variation move. It passes. This verifies a second movement
law but not interacting friendly pieces, cross-material captures, pawns, or
history.

### AF.5. Strong future-law identity fails as the large quotient

C2 also computes the coarsest strong-bisimulation refinement induced by side to
move, terminal kind, and successor blocks. Previous block identity is retained
at every round, making refinement monotone. Every final block is uniform in the
independently certified WDL/DTM values, and a separately ordered implementation
reproduces the counts.

| Universe | $D_4$ states | Bisimulation blocks | Compression | Rounds |
|---|---:|---:|---:|---:|
| KRK | 50,015 | 49,885 | 1.002606× | 13 |
| KQK | 46,137 | 45,249 | 1.019625× | 9 |

Thus identical complete future law is safe but much too strict, and its
construction already requires the complete graph. By contrast, after solving,
side/WDL/DTM yields only 34 KRK and 22 KQK semantic classes—1,471× and 2,097×
compression. That partition is circular because it reads the unknown answer.
The research target is a cheap pre-solution invariant that gives sound bounds
approaching those semantic classes.

### AF.6. Measured ceiling of disadvantage pruning

The solved values were used as an oracle to measure—not assume—the available
pruning:

| Universe | Original branching | Strictly worse WDL edges | WDL+DTM-optimal branching |
|---|---:|---:|---:|
| KRK | 11.1810 | 40,266 / 558,816 = 7.21% | 1.8880 |
| KQK | 13.2633 | 40,323 / 609,873 = 6.61% | 2.3062 |

WDL-only dominance removes relatively little because many moves share the same
eventual outcome. Adding fastest-mate/longest-defense preference leaves a much
narrower ideal policy kernel. The latter does not solve the recognition
problem: its labels were obtained from the completed tables. It is an empirical
ceiling for a future bound engine, not a free oracle.

### AF.7. Counterexample-guided dimensional refinement

The next candidate algorithm is:

1. map complete states to a polynomial-cost relational signature;
2. compute sound abstract $[D^-,D^+]$ bounds with may/must transitions;
3. prune only through (AF.4)–(AF.5), storing every witness;
4. if a block remains ambiguous, extract concrete states producing the conflict;
5. add the missing invariant and split only the necessary block;
6. repeat until the initial state has $D^-=D^+$.

Candidate invariants include material interaction, attack/defence dependency,
king-safety cuts, pawn reachability, promotion cones, zugzwang parity,
reversible-region components, and repetition phase. No coordinate is accepted
because it sounds higher-dimensional; it must remove a witnessed projection
collision while preserving sound bounds. Description, recognition,
transformation, refinement, and certificate costs all count, so an exponential
lookup table cannot masquerade as a short theory.

### AF.8. Chess, quantum algorithms, and physical dimensions are separate gates

Balanced NAND/AND-OR trees have quantum oracle algorithms with square-root
scaling. Farhi, Goldstone, and Gutmann give $\Theta(\sqrt N)$ Hamiltonian-oracle
time for their NAND-tree model [Farhi et al. (2008)](https://theoryofcomputing.org/articles/v004a008/v004a008.pdf).
Chess is not automatically that oracle: it has irregular branching,
transpositions, cycles, history, and a non-free legal-transition computation.

A quantum result must specify a reversible complete-state encoding, reversible
move and terminal predicates, oracle construction cost, logical qubits, gate
count and depth, error assumptions, success probability, and a matched
classical baseline. A simulator validates circuit semantics but not hardware
advantage. Solving chess would validate the computational lift; it would not by
itself prove an eight-dimensional spacetime ontology.

Generalized $n\times n$ chess is EXPTIME-complete
[Fraenkel and Lichtenstein (1981)](https://doi.org/10.1016/0097-3165(81)90016-9).
This does not prevent solving the fixed 8×8 initial state, but a claimed uniform
polynomial method would have to confront that result.

### AF.9. Claim gate and next rung

“Full chess solved” remains prohibited until the normal initial state's bounds
collapse and an independent verifier validates a strategy DAG against every
required reply under the frozen rules. “Quantum advantage” remains prohibited
until the full reversible/oracle cost is compared experimentally or
theoretically with the best classical method. “Physical dimensional theory”
remains prohibited until independent natural data distinguish it from standard
physics.

The next executable rung is KBBK/KBNK, then capture transitions between material
strata, followed by a table-hidden D-/D+ solver that must rediscover the KRK/KQK
root values using only sound bounds and counterexample refinement. Pawns,
promotion, castling, en passant, halfmove phase, and repetition phase follow.
The full search should target a shared strategy DAG from the initial state, not
materialize every state unless forced.

## Appendix AG. Dimensional Chess C3: the complete state reaches the initial position

### AG.1. The legal-transition wall is crossed, not the solution wall

C3 implements the frozen full-chess state

\[
S=(B,\tau,C,E,h,n,\mathcal R),
\tag{AG.1}
\]

where $n$ is included for standards-compatible position serialization even
though it does not alter legal choices. The engine generates all orthodox piece
moves, check and self-check legality, both castling sides, en passant, all four
promotion choices, and the irreversible changes to castling rights after king
moves, rook moves, and rook captures. It retains repetition-relevant history
and implements claimable threefold/50-move draws and automatic
fivefold/75-move draws with checkmate precedence.

The en-passant component of the repetition key is retained only when the side
to move has a legal en-passant capture. This is necessary because FIDE
Article 9.2.3 defines positional identity partly through possible moves, not
merely through a visually identical board
[FIDE Laws of Chess](https://handbook.fide.com/chapter/e012023).

The supplied dead-material predicate deliberately covers standard
material-insufficient cases only. Determining whether no legal sequence can
possibly produce checkmate in every composed position is a broader reachability
problem. Until that predicate or an equivalent exact terminal mechanism is
completed, C3 cannot certify the whole FCS-2023 graph.

### AG.2. Move-generation audit

An independent driver checks six reference positions from the official
[Stockfish perft suite](https://github.com/official-stockfish/Stockfish/blob/master/tests/perft.sh).

| Position | Depth | Reference nodes | C3 |
|---|---:|---:|---|
| Initial position | 4 | 197,281 | PASS |
| Kiwipete | 3 | 97,862 | PASS |
| Position 3 | 4 | 43,238 | PASS |
| Position 4 | 3 | 9,467 | PASS |
| Position 5 | 3 | 62,379 | PASS |
| Position 6 | 3 | 89,890 | PASS |
| **Total** |  | **500,117** | **PASS** |

Ten targeted groups also pass: castling execution and rights, rejection of
castling through check, en-passant execution, rejection and repetition-key
normalization of pinned en passant, promotion multiplicity, rook-capture rights
mutation, threefold path state, automatic 75-move draw, checkmate precedence,
and basic dead material. Passing these tests strongly constrains the
implementation errors they target; it is not a proof over all reachable
states.

### AG.3. First exact interval probe of the normal start

C3 expands the ordinary starting position to four plies. Every nonterminal
horizon state keeps the zero-information interval $[-1,+1]$. Only exact
terminals and rule-authorized draw actions may introduce point intervals. No
piece values, engine evaluation, opening data, or learned predictor enters the
calculation.

| Quantity | Exact finite result |
|---|---:|
| Root legal moves | 20 |
| Nodes including root | 206,604 |
| Legal edges | 206,603 |
| Horizon leaves | 197,273 |
| Checkmate leaf visits | 8 |
| Exact node visits | 16 |
| Certified dominated choices below root | 242 |
| Certified ignored-edge fraction below root | 242 / 206,583 = 0.1171% |
| Certified dominated root choices | 0 |
| Root interval | $[-1,+1]$ |

The node identity

\[
1+20+400+8{,}902+197{,}281=206{,}604
\tag{AG.2}
\]

and terminal identity

\[
197{,}273+8=197{,}281
\tag{AG.3}
\]

give simple internal cross-checks. The checkmate leaves lie below the initial
moves f2-f3, f2-f4, and g2-g4. They do not make those moves proved losses,
because other continuations remain possible. All twenty root children retain
$[-1,+1]$.

### AG.4. Forcing dominance formalizes ignored legal moves

The number of legal alternatives a path can ignore is useful only after it is
made subordinate to correct play. Let $P$ be the player whose strategy is being
proved and let $u_P(a)\in\{-1,0,+1\}$ be the outcome after action $a$, ordered
from $P$'s perspective. When the outcome is decisive, define mate efficiency

\[
m_P(a)=
\begin{cases}
-d(a),&u_P(a)=+1,\\
+d(a),&u_P(a)=-1,\\
0,&u_P(a)=0,
\end{cases}
\tag{AG.4}
\]

where $d(a)$ is exact distance to mate. Thus a player prefers a shorter forced
win and a longer forced loss. For a proof strategy $\Sigma$, let
$e_\Sigma(a)$ count sibling legal choices removed by stored interval or
lexicographic-bound certificates. The forcing-dominance value is

\[
\operatorname{FD}_P(a)=
\bigl(u_P(a),m_P(a),e_\Sigma(a)\bigr),
\tag{AG.5}
\]

ordered lexicographically. The first coordinate chooses correct
game-theoretic play, the second chooses mate efficiency, and only the third
prefers the proof that ignores the most certified-irrelevant moves.

For a strategy DAG rather than one line, raw repeated-node counts can be
manipulated by cycles. C3 therefore counts eliminations at unique complete
states and records the witnessing inequality. A future aggregate should report
both

\[
E(\Sigma)=\sum_{s\in R_\Sigma}
\bigl(|A(s)|-|K_\Sigma(s)|\bigr)
\quad\text{and}\quad
\rho(\Sigma)=
\frac{E(\Sigma)}{\sum_{s\in R_\Sigma}|A(s)|},
\tag{AG.6}
\]

where $R_\Sigma$ is the set of unique reachable controlled states, $A(s)$ all
legal actions, and $K_\Sigma(s)$ the actions that a verifier must retain.
Neither $E$ nor $\rho$ can authorize a deletion; they measure compression after
the deletion has been proved safe.

The current 242 certificates are WDL-interval certificates below the root, not
complete distance-to-mate certificates. Zero root choices are eliminated.
Therefore shallow forcing-dominance does not yet select an initial move.

### AG.5. Updated wall and claim gate

C3 falsifies the idea that full move generation or shallow forced-mate
detection is the missing variable. The active computational wall is a source of
strong, cheap, sound bounds before minimax values are known. Candidate
relational coordinates must be tested table-hidden on solved endgames: they
must recover WDL and distance-to-mate choices without reading those labels,
then emit certificates checked by an independent implementation.

“Full chess solved” remains prohibited until:

1. the complete frozen terminal semantics are implemented;
2. the initial interval collapses to one value;
3. a repetition-safe strategy DAG covers every required adversarial reply;
4. an independent verifier accepts every transition, bound, dominance
   deletion, outcome, and mate-distance claim; and
5. construction, verification, memory, and certificate sizes are reported.

The exact C3 claim is narrower: a tested complete-move state engine reaches the
real initial position, and a proof-only depth-four expansion certifies that the
root is still unresolved. This advances the computational program without
claiming a result that the evidence does not contain.

## Appendix AH. Dimensional Chess C4: constraint waves and relational phase

### AH.1. Human search is selective constraint propagation

A legal move list is not a value ordering. Human players instead track attacked
squares, defended pieces, king safety, mobility, confinement, threats, and the
opponent's strongest reply. Let $\mathcal A_P(s)$ be player $P$'s attack field
and $\mathcal L_P(s)$ the legal actions in state $s$. Action $a$ produces

\[
s'=T(s,a).
\tag{AH.1}
\]

For kings, an attacked destination is illegal. For other pieces it is not
automatically irrational: a sacrifice may force mate or gain a larger
advantage. The exact disqualification rule therefore remains interval
dominance, not membership in an attack map.

Terminal and draw facts act as boundary values in the alternating Bellman
operator:

\[
\begin{aligned}
D^-_{k+1}(s)&=\max_aD^-_k(T(s,a)),
&
D^+_{k+1}(s)&=\max_aD^+_k(T(s,a))
&&(\tau=\mathrm W),\\
D^-_{k+1}(s)&=\min_aD^-_k(T(s,a)),
&
D^+_{k+1}(s)&=\min_aD^+_k(T(s,a))
&&(\tau=\mathrm B).
\end{aligned}
\tag{AH.2}
\]

Repeated application is the precise constraint wave: forced facts propagate
backward through squares, moves, and adversarial replies. Removing material
usually reduces interactions and branching, which explains why the field
becomes easier to resolve in endgames without asserting that every individual
material removal makes a position computationally simpler.

### AH.2. Table-hidden geometric language

C4 freezes five variants across the complete KRK and KQK universes. The bound
phase sees legal transitions, terminal rules, the theorem that a lone black
king cannot mate, and declared coordinates only. Exact tablebase WDL and DTM
labels are computed after all variants and intervals are frozen.

The distance base uses side to move, terminal kind, edge/corner confinement,
check and capture-draw status, legal-move count, king and piece distances, and
piece protection. Three isolated lifts add:

1. relative dot product and absolute cross product of the white-king and
   long-range-piece vectors about the black king;
2. the size of the black king's connected static safe region;
3. own/opponent mobility and edge-direction counts.

The combined constraint wave includes all three and the remaining axis
distances. Transition profiles are deduplicated inside each signature block, so
their count measures operational graph compression rather than only label
compression.

### AH.3. Exact table-hidden results

| Universe and signature | Exact DTM | Certified WDL+DTM eliminations | State compression | Profile compression |
|---|---:|---:|---:|---:|
| KRK distance base | 3,911 / 50,015 = 7.82% | 21.21% | 1.7608× | 1.1000× |
| KRK + orientation | 13,775 / 50,015 = 27.54% | 74.25% | 1.0700× | 1.0005× |
| KRK + safe region | 6,759 / 50,015 = 13.51% | 60.27% | 1.3621× | 1.0811× |
| KRK + mobility | 8,791 / 50,015 = 17.58% | 70.56% | 1.4164× | 1.0937× |
| **KRK combined** | **44,633 / 50,015 = 89.24%** | **506,548 / 558,816 = 90.65%** | **1.0067×** | **1.0001×** |
| KQK distance base | 6,959 / 46,137 = 15.08% | 54.31% | 1.6326× | 1.0800× |
| KQK + orientation | 30,554 / 46,137 = 66.22% | 90.02% | 1.0660× | 1.0011× |
| KQK + safe region | 22,008 / 46,137 = 47.70% | 87.84% | 1.3205× | 1.0556× |
| KQK + mobility | 21,496 / 46,137 = 46.59% | 86.47% | 1.2828× | 1.0745× |
| **KQK combined** | **45,276 / 46,137 = 98.13%** | **563,800 / 609,873 = 92.45%** | **1.0071×** | **1.0003×** |

Every variant proves every WDL value in both universes. Mate distance is the
harder coordinate: equal WDL does not imply equal fastest-mate or
longest-resistance time. Relative orientation produces the largest isolated
DTM improvement, showing that separate scalar distances erase a genuine
relational phase. Safe-region and mobility coordinates also carry information.

The state certificate contains 96,152 rows. A verifier independent of feature
construction and interval propagation compares every row with the previously
verified exact tables, recomputes legal-successor dominance, reproduces both
elimination totals, and reports zero WDL violations, zero DTM violations, and
zero summary mismatches.

### AH.4. Predictive success does not equal scaling

The combined KRK signature has 49,681 blocks for 50,015 states and 50,010
successor profiles. The combined KQK signature has 45,813 blocks for 46,137
states and 46,123 successor profiles. Thus it recovers most exact mate
distances by becoming almost injective. It still enumerates the concrete graph
to construct the signatures.

This is a critical falsification. High DTM coverage and more than 90% certified
move elimination do not imply a small proof when the certificates depend on
almost one transition profile per state. The absolute constraint field is
informative but is not yet the scalable missing variable.

### AH.5. Constraint-delta lift

Human analysis often represents what a move changes rather than the entire
absolute attack field. Define

\[
C(s)=
\bigl(
\text{safe mobility},
\text{opponent mobility},
\text{confinement},
\text{attack dependencies},
\text{forcing rights}
\bigr)
\tag{AH.3}
\]

and

\[
\Delta C(s,a)=C(T(s,a))-C(s).
\tag{AH.4}
\]

The C5 candidate groups actions by equivalent constraint deltas: opened and
closed lines, pins created or broken, regions split or joined, defenders added
or removed, and adversarial replies eliminated. A reusable delta law could
describe many absolute boards with one transition type.

The proposal survives only if it:

1. is constructed without solved labels;
2. emits sound WDL and DTM bounds;
3. transfers across rook, queen, bishop, knight, pawn, and mixed-material
   strata;
4. counts recognition, construction, refinement, and certificate costs; and
5. retains substantial successor-profile compression.

The exact C4 result is therefore two-sided: the proposed restriction wave is a
verified source of mate information, while its present absolute encoding is
rejected as a full-chess shortcut. The next missing variable is a compact
canonical law for move-induced constraint change.

## Appendix AI. Dimensional Chess C5: local tension and tactical entry

### AI.1. Attack multiplicity is a field, not a verdict

For state $s$, colour $c$, and square $q$, define the geometric attack
multiplicity

\[
A_c(q;s)=\sum_{p\in P_c(s)}\mathbf 1\{p\leadsto q\}.
\tag{AI.1}
\]

Sliding attacks stop at the first occupied square. Pawns contribute their
diagonal attacks, kings their adjacent squares, and pinned pieces retain their
geometric pressure. This field is deliberately separated from legal move
generation: a geometric dependency may exist even when moving the attacking
piece would expose its king.

Define local tension, signed control, and total load by

\[
\tau(q;s)=A_W(q;s)A_B(q;s),\qquad
\nu(q;s)=A_W(q;s)-A_B(q;s),\qquad
\lambda(q;s)=A_W(q;s)+A_B(q;s).
\tag{AI.2}
\]

Thus $\tau$ is positive exactly on mutually attacked squares, $\nu$ records the
local numerical imbalance, and $\lambda$ records activity even when only one
side attacks. For an occupied square, raw fragility is

\[
F(q;s)=\max\{0,\operatorname{attackers}(q;s)
                  -\operatorname{defenders}(q;s)\}.
\tag{AI.3}
\]

This is a search feature, not a theorem that the occupant is lost. Checks,
pins, sacrifices, intermediate moves, and mating threats can dominate the raw
count.

### AI.2. Tension-breaking entry

Let legal action $a:q_0\to q_1$ produce $s'=T(s,a)$. From the mover's
perspective define

\[
E_A(s,a)=A_{\rm mover}(q_1;s),\qquad
E_D(s,a)=A_{\rm opponent}(q_1;s),
\tag{AI.4}
\]

and the destination's local tension release

\[
R_\tau(s,a)=\max\{0,\tau(q_1;s)-\tau(q_1;s')\}.
\tag{AI.5}
\]

C5 presently calls $a$ a tension-breaking entry when it captures an enemy
occupant on an already contested destination, $E_A>E_D$, and $R_\tau>0$. The
condition operationalizes the proposal to spend computation first where a
move converts interacting attack dependencies into a favorable local
imbalance.

### AI.3. Adversarial ordering

For each legal action, C5 enumerates the opponent's legal replies and keeps the
worst conventional static reply vector

\[
V=(\text{terminal},\text{material},\text{pressure},
   \text{control},\text{mobility})
\tag{AI.6}
\]

from the mover's perspective. It then orders computation by this worst reply,
immediate mate, captured material, the tension-entry coordinates, control
flips, and replies eliminated. This is a move-ordering law only. Sound removal
still requires the interval-dominance inequality of Appendices AF and AG.

### AI.4. Executable fixture and independent audit

In the declared Scholar's-mate state

```text
r1bqkb1r/pppp1ppp/2n2n2/4p2Q/2B1P3/8/PPPP1PPP/RNB1K1NR w KQkq - 4 4
```

there are 43 legal moves. The `f7` pawn is attacked from `c4` and `h5` and
defended from `e8`, giving

\[
(A_W,A_B,\tau,\nu,F)=(2,1,2,1,1).
\tag{AI.7}
\]

The C5 ordering ranks `Qxf7#` first. Its entry coordinates are

\[
(E_A,E_D,E_A-E_D,R_\tau)=(2,1,1,1),
\tag{AI.8}
\]

and it reduces Black's legal replies from 28 to zero. `Bxf7+` reaches the same
target but leaves one legal reply. An independent verifier does not import the
tension implementation: it reconstructs both colours' attack origins, checks
the formulas for all 64 squares, recounts the 43 legal moves, and verifies both
reply facts. All checks pass.

### AI.5. Counterexample and boundary

Before `Qxf7#`, total board tension is 16. After checkmate it is 28. The losing
side nevertheless has zero legal moves. Consequently total tension is not a
monotone evaluation function and cannot be optimized in isolation. It marks
where dependencies are concentrated; terminal semantics and adversarial reply
search determine the value.

The same boundary appears in the geometric hanging count: the White queen on
`h5` is attacked and geometrically undefended before the move, yet White has a
forced immediate mate. This directly falsifies any rule that discards every
move because its moving piece or destination looks locally unsafe.

C5 therefore contributes a verified computation-allocation coordinate, not a
solution of chess. The next experiment must freeze a tactically diverse corpus
and compare proof nodes at equal budgets against standard ordering baselines.
The stronger candidate is a dependency graph for legal exchange sequences,
because raw attacker counts omit piece value, pin legality, tempo, and recapture
order.

## Appendix AJ. Dimensional Chess C6: legal exchange dependencies

### AJ.1. From simultaneous counts to ordered interactions

C5 counts which pieces geometrically attack a square but not the order in which
they can legally arrive. C6 constructs a local capture–recapture graph. Let
$a_0:q_0\to q$ be a legal capture by player $P$, let $s_1=T(s,a_0)$, and let
$\mathcal X_q(s)$ be the legal captures in state $s$ that land on the same
square $q$. If $M_P$ is material balance from $P$'s perspective, define

\[
\Delta M_P(s_i)=M_P(s_i)-M_P(s).
\tag{AJ.1}
\]

The restricted optional-exchange recurrence is

\[
E_P(s_i,q)=
\begin{cases}
\displaystyle
\max\left(\{\Delta M_P(s_i)\}\cup
\{E_P(T(s_i,a),q):a\in\mathcal X_q(s_i)\}\right),
&\operatorname{turn}(s_i)=P,\\[6pt]
\displaystyle
\min\left(\{\Delta M_P(s_i)\}\cup
\{E_P(T(s_i,a),q):a\in\mathcal X_q(s_i)\}\right),
&\operatorname{turn}(s_i)\ne P.
\end{cases}
\tag{AJ.2}
\]

The current material delta is the option to stop exchanging on $q$. Because
each graph edge comes from full legal move generation, pins, illegal king
captures, and revealed slider rays affect the sequence automatically.

This is not a full-position minimax value. It omits captures elsewhere,
zwischenzugs, quiet threats, and most off-square check evasions. A recorded
graph node may be in check even though the abstract stop option is not a legal
full-game action. C6 therefore uses $E_P$ only as an ordering coordinate.

### AJ.2. Exact bounded proof layer

The benchmark surrounds the heuristic ordering with an exact finite AND/OR
mate predicate. For attacker $P$,

\[
B_P(s,d)=
\begin{cases}
1,&s\text{ is checkmate won by }P,\\
0,&d=0\text{ or }s\text{ is a draw or other terminal},\\
\bigvee_{a\in\mathcal L(s)}B_P(T(s,a),d-1),
&\operatorname{turn}(s)=P,\\
\bigwedge_{a\in\mathcal L(s)}B_P(T(s,a),d-1),
&\operatorname{turn}(s)\ne P.
\end{cases}
\tag{AJ.3}
\]

All orderings traverse the same legal successor relation. They may short-circuit
at different times, but they cannot change a proved Boolean result.

### AJ.3. Deterministic recapture corpus

Using seed `20260915`, the generator accepts the first eight distinct states
that contain a legal capture followed by a legal same-square recapture and have
a forced mate of minimal depth three plies. It consumes 258 candidate states.
Admission uses only the Boolean mate property; no comparative node count or
runtime is read. The selected root exchange graphs reach two or three capture
plies and contain two to four nodes.

| Ordering | Proof nodes | Reduction vs lexical | Measured time |
|---|---:|---:|---:|
| Lexical UCI | 10,394 | 0.00% | 2.014 s |
| Checks/captures forcing | 699 | 93.27% | 0.579 s |
| C5 tension | 809 | 92.22% | 0.700 s |
| **C6 dependency** | **668** | **93.57%** | **0.638 s** |

C6 uses 17.43% fewer nodes than C5 and 4.43% fewer than forcing in aggregate.
Graph construction nevertheless makes it 10.22% slower than forcing in this
run. Per position it beats forcing three times, ties once, and loses four times.
The aggregate advantage is consequently a narrow small-corpus result, not a
universal ordering theorem.

### AJ.4. Independent audit

The independent C6 verifier imports no C6 code. It re-enumerates all root moves,
proves the absence of mate in one, and constructs the three-ply AND/OR
certificate for every position. It reproduces every reported winning root
move, covering 33 defender reply branches with 53 mating continuations. It also
reconstructs the largest declared root exchange graph in each position and
matches every value, node count, maximum capture depth, and checked-node count.
All eight audits pass, and the four aggregate proof-node totals are reproduced.

### AJ.5. Consequence for the full-chess program

C6 supplies evidence that legal recapture order contains information not
present in simultaneous tension counts. It does not produce a sound new bound,
does not prune a legal move, and does not narrow the initial-position interval.
The result therefore improves computation allocation without crossing the
solution boundary.

The next candidate is a proof-carrying forcing-dependency graph that includes
checks, mandatory evasions, off-square captures, promotions, and quiet forcing
threats. Its nodes must carry sound lower and upper game-value intervals. It
must also show substantial compression after the cost of graph recognition and
construction is counted. Absent that bound or a complete strategy certificate
from the orthodox initial position, chess remains unsolved.

## Appendix AK. Initial-position mate frontier R1

### AK.1. Direct run on the orthodox root

C6 is next applied to the normal initial state

```text
rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1
```

using the exact bounded predicate from Equation AJ.3. At seven plies the
dependency-ordered runs return

\[
B_W(s_0,7)=0,
\qquad
B_B(s_0,7)=0.
\tag{AK.1}
\]

Thus neither side can force checkmate within seven plies. Equation AK.1 is not
a WDL value: a mate outside the horizon and every drawing continuation remain
unclassified.

### AK.2. Measured frontier

| Attacker | C6 nodes | Generated edges | Horizon nodes | Exchange graphs | Time |
|---|---:|---:|---:|---:|---:|
| White | 128,724 | 251,799 | 78,842 | 5,370 | 105.820 s |
| Black | 14,846 | 143,274 | 4,046 | 3,127 | 53.319 s |

For White as attacker, the C6 node sequence is 21, 486, 7,536, and 128,724 at
one, three, five, and seven plies. The last extension multiplies nodes by about
17.08. The current ordering has not altered exponential frontier growth.

### AK.3. Independent result and ordering falsification

A verifier that imports no C6 module recomputes the predicate with lexical UCI
ordering. It also obtains Equation AK.1, using 106,322 nodes for White as
attacker and 15,082 for Black. The independent result is `PASS`.

C6 uses 21.07% more nodes than lexical order for the White-attacker proof and
1.56% fewer for the Black-attacker proof. Combined, C6 uses 143,570 nodes
versus 121,404, an increase of 18.26%. Its aggregate advantage on the C6
synthetic tactical corpus therefore does not transfer to this initial-position
negative-mate task.

The result does not establish that lexical order is intrinsically meaningful.
It encounters counterexample branches earlier by accident in this finite run.
It does establish that a single attractive-move order is not the missing
general solution coordinate.

### AK.4. Dual proof direction

At an attacker node, a positive mate proof needs one successful action while a
negative proof must reject every action. At a defender node, a positive proof
must cover every reply while a negative proof needs one escape. Search priority
therefore depends on both whose turn it is and which side of the interval is
being certified.

C5 heat and C6 exchanges allocate work but do not describe unexpanded
successors. They cannot narrow the root WDL interval, which remains

\[
I(s_0)=[-1,+1].
\tag{AK.2}
\]

The next stage must build forcing-dependency graph nodes with sound lower and
upper bounds on all represented completions. Checks, mandatory evasions,
off-square captures, promotions, quiet threats, repetition, and draw claims
must be included or conservatively widened. Only interval dominance may turn
that abstraction into safe pruning. Without such bounds, increasing the C6
horizon merely moves the exponential frontier.

## Appendix AL. Dimensional Chess C7: horizon-safe projection

### AL.1. Repetition coordinates have a finite activation horizon

Let the repetition-key history along a path be

\[
h=(p_0,p_1,\ldots,p_t).
\tag{AL.1}
\]

An identical full chess position has the same side to move, so recurrence
intervals are even. It cannot recur after two plies: the opponent cannot both
undo the first player's move and undo their own required move in a single
action without changing placement, material, castling rights, or en-passant
state. The minimum recurrence interval is therefore four plies.

If a position first appears at ply \(i\), its third occurrence cannot be earlier
than ply \(i+8\). At total horizon \(D\), only repetition candidates satisfying

\[
i\le D-8
\tag{AL.2}
\]

can enable a threefold claim. C7 retains multiplicities only for those
candidates. Fullmove number is always discarded. Halfmove clock is discarded
only if the fifty-move threshold is globally unreachable inside the horizon.
This is a sound over-retention for the bounded predicate, not a heuristic collision; see the correction at the end of AL.5.

At \(D=9\), Equation AL.2 retains only the initial position and the position
after the first move. Histories that differ only after that prefix cannot differ
on threefold availability before the horizon.

### AL.2. Polarity-aware ordering test

At five plies with White designated as attacker, lexical defender ordering
uses 7,480 proof calls. C6 dependency ordering uses 7,536 and a checks/captures
ordering uses 9,992. Changing attacker ordering while fixing the defender order
does not change these call counts in this negative proof. This confirms the
logical polarity diagnosed in Appendix AK: every attacking candidate must be
rejected, while finding one defender escape short-circuits its line.

A richer escape heuristic takes 95.13 seconds and 9,926 calls. A naive
proof-number implementation expands 40,316 nodes and generates 743,471
children in 32.44 seconds. Both are rejected for this task. Their failure does
not make lexical notation intrinsically informative; it only identifies the
best tested traversal for the present certificate.

### AL.3. Seven-ply quotient check

Below eight plies, threefold repetition and the fifty-move threshold are both
unreachable from the initial state. The projected lexical computation obtains

\[
B_W(s_0,7)=B_B(s_0,7)=0
\tag{AL.3}
\]

using 95,718 calls for White and 10,551 for Black. The earlier unprojected
lexical verifier used 106,322 and 15,082. Combined calls fall from 121,404 to
106,269, a 12.47% reduction, without changing the predicate.

### AL.4. Independently verified nine-ply frontier

C7 then computes

\[
B_W(s_0,9)=B_B(s_0,9)=0.
\tag{AL.4}
\]

| Attacker | Proof calls | Generated edges | Cache hits | Horizon nodes | Time |
|---|---:|---:|---:|---:|---:|
| White | 2,951,587 | 5,259,337 | 1,437,821 | 1,308,014 | 575.877 s |
| Black | 127,991 | 736,248 | 59,263 | 30,841 | 23.080 s |

A verifier that imports no C7 code separately reconstructs the candidate
prefix, draw actions, and AND/OR recurrence. It exactly reproduces all reported
Boolean results and every node, edge, cache, horizon, and terminal count at
seven and nine plies. It checks 400 two-ply paths from the root and 8,902 from
all first-ply states, finding no two-ply recurrence. Explicit legal knight
cycles reach the root's third occurrence at ply eight and the third occurrence
of a post-first-move position at ply nine. The audit returns `PASS`.

### AL.5. Boundary of the result

The White proof-call frontier grows by a factor of 30.84 from seven to nine
plies; Black grows by 12.13. Projection removes coordinates that cannot affect
the finite theorem, but it does not eliminate the remaining adversarial tree.
Equation AL.4 is not a WDL value, supplies no unbounded optimal strategy, and
does not solve chess. The root interval remains

\[
I(s_0)=[-1,+1].
\tag{AL.5}
\]

C7 nevertheless sharpens the framework's computational principle: a hidden
coordinate may be dropped from a finite proof state whenever it cannot cross a
terminal or transition threshold within that proof's horizon.

**Correction (revision 1.69).** That condition is sufficient for safe removal,
not necessary, and the earlier wording "exactly when" was wrong. For the
bounded forced-mate predicate, repetition history is provably irrelevant at
every horizon: if the attacker has a history-free forced mate within \(d\)
plies, playing to strictly decrease the bounded mate distance makes that
distance fall on every line, so no position recurs and no threefold claim ever
becomes available, while adding defender claim options can only lower the
predicate. Hence the full projection that discards all repetition
multiplicities is exact for this predicate, and Equation AL.2's horizon
bookkeeping is sound but unnecessary. The nine-ply results stand because
over-retention cannot change a sound proof. The next stage requires a
proof-carrying abstraction with sound lower and upper completion bounds. Only
such bounds can merge tactically different continuations and narrow Equation
AL.5 without enumerating them.

## Appendix AM. Dimensional Chess C8: movement-frequency relations

### AM.1. Relational move phase

C8 maps legal move \(a:q\to q'\) in state \(s\) to

\[
\phi(s,a)=
(p,c,\Delta x,\Delta y,\chi,
 \operatorname{sgn}\Delta d_{K_s},
 \operatorname{sgn}\Delta d_{K_o},
 g_s,g_o,\Delta r,h(q),h(q')).
\tag{AM.1}
\]

The fields record moving and captured piece types, color-oriented displacement,
destination parity, Chebyshev-distance changes to both kings, geometric
relations to both kings, edge-ring change, and home-relative rank phases. Eight
deterministic phase-shuffled five-ply searches record defender cutoffs. With
depth-squared success and failure weights \(S_u,F_u\), class \(u\) receives

\[
q_u=\frac{S_u+1}{S_u+F_u+2}.
\tag{AM.2}
\]

The weights are then frozen. Defender moves are sorted by Equation AM.2;
attacker moves remain lexical. No move is removed, and only the exact bounded
predicate from Appendix AJ determines value.

### AM.2. Learned pattern

The White-attacker book contains 102 classes. Ten of its top twenty are quiet
pawn advances and four are quiet knight developments; all twenty are
non-captures. Nineteen move closer in Chebyshev distance to the opposing king.
The highest-scoring class is a one-square pawn move from home-relative phase
one to phase two that preserves distance to the mover's king and reduces
distance to the opposing king. It records 296 training cutoffs and zero
failures. The corresponding two-square advance records 286 and zero.

These counts characterize early counterexamples to White's short-horizon mate
claims. They are not chess-value labels for the moves.

### AM.3. Transfer to seven and nine plies

At seven plies, the relation ordering generates 162,511 edges versus 172,970
for lexical order, a 6.05% reduction, while calls rise from 95,718 to 98,352.
Three relation repeats reproduce both counts exactly and take a median 13.417
seconds. Three lexical repeats reproduce their counts and take a median 15.272
seconds, a 12.15% relation-time reduction.

At nine plies with White as attacker, the exact results are

| Method | Calls | Generated edges | Horizon nodes | Proof time |
|---|---:|---:|---:|---:|
| C7 lexical | 2,951,587 | 5,259,337 | 1,308,014 | 575.877 s |
| C8 relation | 2,750,245 | 4,299,169 | 930,982 | 375.703 s |

C8 reduces calls by 6.82%, edges by 18.26%, horizon nodes by 28.83%, and proof
time by 34.76%. Training costs 31.087 seconds; the 406.790-second combined time
is 29.36% below the C7 proof time. Both computations obtain the already
independently verified value

\[
B_W(s_0,9)=0.
\tag{AM.3}
\]

### AM.4. Counterexamples to stronger frequency claims

Absolute oriented-square classes increase seven-ply calls by 2.39%. Directly
conditioning response classes on the previous move expands the relation
vocabulary from 102 to 1,264 classes: piece-, vector-, and relation-pair variants
increase calls by 75.50%, 80.75%, and 49.18%. Updating relation weights online
increases calls by 52.53%. These failures show that added context can destroy
reuse faster than it adds information.

The result is also role-asymmetric. For Black as designated attacker, its own
239-class relation model raises nine-ply calls from 127,991 to 136,288, edges
from 736,248 to 1,070,334, and proof time from 23.080 to 34.733 seconds. Applying
the White-trained color-oriented book also loses. Equation AM.2 is not a
universal color-normalized chess law.

### AM.5. Audit and boundary

The C8 Boolean results equal the separately implemented C7 certificate. A
structural audit checks seven schemes across 960 deterministically reachable
states and 168,882 legal-move instances. Every ordered list is exactly a
permutation of the legal list; 6,720 state/scheme cases genuinely reorder it.
The audit returns `PASS`.

C8 validates only a role-specific computation-allocation pattern. It neither
compresses exact state nor supplies sound interval bounds for unexpanded
continuations. Thus

\[
I(s_0)=[-1,+1]
\tag{AM.4}
\]

remains unchanged. The next admissible lift must attach a proof obligation or
sound completion interval to a relational class. Without that certificate,
frequency may prioritize a branch but never discard it.

## Appendix AN. Dependent-sign evidence under a declared Markov null

### AN.1. The bounded question

Appendix AA showed that marginally balanced signs can be strongly dependent:
a stationary chain that repeats its previous sign with probability (0.90)
caused the iid conditional-sign process to cross in 3,772 of 5,000 synthetic
histories. The unresolved question was not whether one statistic can be valid
under arbitrary dependence. It was whether an anytime-valid process can be
constructed after the dependence law is stated explicitly.

Let (X_k\in\{-1,+1\}) be the sequence of nonzero projected-residual signs in
nonzero-event time. Conditional on (X_1), the null in this revision is the
fixed first-order homogeneous Markov law

\[
P_0(X_k=X_{k-1}\mid\mathcal F_{k-1})=0.90,
\qquad
P_0(X_k=-X_{k-1}\mid\mathcal F_{k-1})=0.10.
\tag{AN.1}
\]

The initial sign may be arbitrary. Exact zeros are skipped, so Equation AN.1
must hold for the next nonzero sign conditional on the full past. This is a
model-class assumption, not a consequence of marginal 50/50 signs.

### AN.2. Predictive likelihood-ratio martingale

Before transition (k), let (N_{ij}(k-1)) count earlier transitions from
state (i) to state (j). The Krichevsky--Trofimov predictor is

\[
\widehat Q_k(j\mid i)=
\frac{N_{ij}(k-1)+\tfrac12}{N_{i\cdot}(k-1)+1}.
\tag{AN.2}
\]

Define

\[
L_n=\prod_{k=2}^{n}
\frac{\widehat Q_k(X_k\mid X_{k-1})}
     {P_0(X_k\mid X_{k-1})}.
\tag{AN.3}
\]

Because (widehat Q_k) is determined by the past and is a probability
distribution,

\[
\mathbb E_0[L_k\mid\mathcal F_{k-1}]
=L_{k-1}\sum_j
P_0(j\mid X_{k-1})
\frac{\widehat Q_k(j\mid X_{k-1})}{P_0(j\mid X_{k-1})}
=L_{k-1}.
\tag{AN.4}
\]

Thus (L_n) is a nonnegative null martingale with (L_1=1), and Ville's
inequality gives

\[
P_0\!\left(\sup_n L_n\ge 1/\alpha\right)\le\alpha.
\tag{AN.5}
\]

This is a written probability proof for the declared model. The construction
is an application of predictive Markov likelihood ratios, consistent with
Fields, Javidi, and Shekhar's 22 January 2025 sequential Markov-chain test
(https://arxiv.org/abs/2501.13187), not a new theorem about dependence.

### AN.3. Fresh finite synthetic validation

The protocol fixed (alpha=0.01), 5,000 histories, 400 looks, the null matrix,
the add-one-half predictor, a fresh seed, six scenarios, and source hashes
before outcome generation. The results were

| Sign process | Markov KT crossings | Old iid-sign crossings |
|---|---:|---:|
| Declared (p_{\rm stay}=0.90) null | 39 / 5,000 | 4,134 / 5,000 |
| Same null, every sign inverted | 39 / 5,000 | 4,134 / 5,000 |
| Same null, fixed positive initial sign | 36 / 5,000 | 4,158 / 5,000 |
| More persistent, (p_{\rm stay}=0.97) | 4,629 / 5,000 | 4,977 / 5,000 |
| Less persistent, (p_{\rm stay}=0.75) | 4,999 / 5,000 | 1,448 / 5,000 |
| Balanced iid, (p_{\rm stay}=0.50) | 5,000 / 5,000 | 32 / 5,000 |

The paired sign inversion produced exactly identical decisions. A separate
implementation regenerated every chain, recomputed both statistics, and
matched 24,000,000 stored decision states. The initial scalar audit exceeded
the bounded runtime; it was mechanically vectorized after outcomes without
changing the seed, scenarios, scoring implementation, formulas, thresholds,
saved paths, or summaries. This provenance is retained in the protocol.

These counts are finite synthetic computation. Equation AN.5, not the observed
39 or 36 null crossings, supplies the anytime guarantee. The alternative
counts measure power only for these frozen simulations.

### AN.4. Counterexamples and boundary

The old iid-sign process is not repaired by marginal balance: it rejected
4,134 of 5,000 valid (p_{\rm stay}=0.90) Markov-null histories because its
conditional-balance assumption was false. Conversely, the new process would
be misspecified if 0.90 were estimated from the same scored stream, changed
over time, or represented only an average transition frequency. A first-order
matrix also does not cover higher-order or non-Markov sign dependence.

The engine discards magnitudes and does not test scale, tail thickness, or a
unique physical mechanism. It is standalone and does not share its 1% budget
with another detector. No measured physical evidence was collected, and a
crossing cannot identify a theological cause.

The next bounded test is a composite-null construction that remains anytime
valid for every prespecified persistence value in an interval such as
([0.88,0.92]). It must be challenged at both interval boundaries and against
non-Markov chains before any replacement of the point-null engine.

## Appendix AO. Hydrogen spectroscopy clock and tangent-rotor holdout

### AO.1. What real data can calibrate

The previous hydrogen construction separated a standard time-dependent
\(1s+2p_z\) reference from a synthetic pair of tangent 3-spheres rotating in
four spatial coordinates. The bounded question in this revision is whether the
synthetic family can predict the reference after fitting only selected phases,
while the reference clock is tied to observational spectroscopy.

The NIST Hydrogen strong-lines table lists two vacuum Lyman-alpha components,

\[
\lambda_1=121.566824\ {\rm nm},\qquad
\lambda_2=121.567364\ {\rm nm},
\tag{AO.1}
\]

with listed relative intensities 1000 and 500. Their intensity-weighted
centroid is \(121.567004\) nm and its optical period is \(405.503877\) as. The
frequency separation is \(10.954243544\) GHz. These values constrain the
energy/time scale. The listed intensities are not treated as prepared-state
populations and Equation AO.1 does not provide 3D density voxels.

The release-0.2 fixed-proton nonrelativistic Coulomb prediction was
\(121.502273411\) nm. Its centroid-relative error is therefore

\[
\frac{121.502273411-121.567004}{121.567004}
=-5.324684\times10^{-4}.
\tag{AO.2}
\]

Release 0.3 retains the ideal spatial basis but replaces the displayed phase
timestamps with the NIST centroid. Calling this a spectroscopy-clocked proxy is
more accurate than calling it a measured movie of an atom.

### AO.2. Frozen spatial discrimination protocol

The protocol fixed 24 phases, a \(10a_0\) half-width, eight training phases,
sixteen disjoint holdout phases, the parameter bounds, optimizer seed, and five
acceptance gates before fitting. The rotor family retained two equal tangent
3-spheres in \(\mathbb R^4\), one rotating center axis, perspective projection,
two Gaussian shell envelopes, and their complex interference. Its projected
axis was aligned to the known \(2p_z\) symmetry axis; this coordinate choice
was not scored as a discovery.

The six fitted quantities were radius, camera-to-radius ratio, projection
scale, wave number, shell width, and phase offset. The objective was mean
squared Hellinger distance on the eight training phases only. A frozen
training-phase mean and its spherical radialization were declared as baselines.

### AO.3. Held-out result and counterexample

After 1,862 training-objective evaluations, the frozen \(33^3\)-voxel holdout
scores were

| Predictor | Mean squared Hellinger | Mean total variation | Mean relative \(L^2\) |
|---|---:|---:|---:|
| Tangent-rotor family | 0.172620216 | 0.432929726 | 0.821171870 |
| Training-phase mean | 0.022196552 | 0.131144467 | 0.188746349 |
| Spherical radial baseline | 0.072178684 | 0.259993679 | 0.225927917 |

The candidate is worse than both baselines on every aggregate metric. Its
held-out axial-centroid correlation is \(-0.086074\), opposite-phase dynamic
nodal leakage is \(0.057650\), and mirror total-variation error is \(0.053451\).
These fail thresholds of \(0.95\), \(0.02\), and \(0.02\). Only exact-cycle
repeatability passes, at \(4.34\times10^{-19}\) maximum probability error.

This supplies a counterexample to the inference that tangency plus periodic
interference is sufficient for hydrogen-like evolution. The geometry can be
internally consistent and periodic while predicting the wrong spatial mass and
wrong phase transport.

### AO.4. Verification and evidence boundary

All 23 unit-test groups pass. A separately written replay imports none of the
production calibration, quantum, rotor, or metric functions. It reconstructs
the wavelengths with decimal arithmetic, regenerates both voxel sequences,
recomputes the phase split and scores, and reproduces all five decisions and
the rejection.

Equation AO.2 is a finite comparison with a critically compiled spectral
reference. The voxel holdout is a synthetic model-discrimination experiment,
not a laboratory measurement. Photoionization microscopy has directly observed
nodal structure for externally driven hydrogen Stark states (Stodolna et al.,
Phys. Rev. Lett. 110, 213001, 2013), but that preparation is not silently
substituted for the field-free \(1s+2p_z\) reference used here.

The exact projected-envelope rotor family is rejected as an adequate generator
of the declared reference. This does not reject every higher-dimensional model,
establish a fourth spatial dimension, identify an unconstrained residual with
God, or alter the earlier closure, collision, angular-impulse, and finite-prefix
corrections.

The next admissible candidate must be derived from a Hamiltonian or continuity
equation before comparison. Its spatial basis must be frozen before new data are
used, and it must address both Lyman-alpha components plus a spatial observable
from a matching experimental preparation. Importing the known hydrogen basis
by construction would reproduce the target but would not explain it.

## Appendix AP. Temperature, phase direction, and bounded missing components

### AP.1. Why a failed model does not select one cause

Appendix AO rejected one specified tangent-rotor envelope. A mismatch can arise
from omitted state variables, but also from an incorrect primitive geometry,
dynamics, projection, observable map, calibration, or optimizer. Therefore the
exclusive inference

\[
\text{failed fit}\Longrightarrow\text{missing rotors are the only cause}
\tag{AP.1}
\]

is invalid. A broader family can be tested only after its new variables and
failure conditions are fixed. Allowing arbitrary per-voxel corrections or an
unbounded number of components would interpolate finite data without identifying
an explanatory law.

### AP.2. Frozen latent-component rescue

The bounded rescue freezes all six release-0.3 geometric parameters. It permits
one through four copies of the same rotor, each selected from twelve phase offsets
spaced by \(\pi/6\), with one global nonnegative simplex weight per component.
Greedy selection and weight fitting use the eight old training phases only. The
four-component model is the sole acceptance candidate; smaller models are nested
diagnostics. The sixteen previously untouched phases are scored once.

| Components | Training \(H^2\) | Holdout \(H^2\) |
|---:|---:|---:|
| 1 | 0.204041584 | 0.172620216 |
| 2 | 0.203640581 | 0.171868367 |
| 3 | 0.203586339 | 0.172032715 |
| 4 | 0.203586339 | 0.172032715 |

The fourth component receives zero weight. The rescue improves holdout error by
only 0.3403%, below the predeclared 20% minimum, and remains worse than the
time-mean baseline \(H^2=0.022196552\). Its held-out axial-centroid correlation
is \(-0.058760\), dynamic nodal leakage is \(0.058831\), and opposite-phase mirror
TV is \(0.043823\). Five of six gates fail. Thus the same-geometry phase-mixture
rescue is rejected. Other explicit latent-variable mechanisms remain open to a
fresh test; Equation AP.1 does not become true.

### AP.3. Temperature does not reverse the phase clock

Let a declared two-level Hamiltonian be

\[
H=\operatorname{diag}(0,\Delta E),\qquad
\rho_\beta=\frac{e^{-\beta H}}{\operatorname{tr}(e^{-\beta H})}.
\tag{AP.2}
\]

Because \(\rho_\beta\) is a function of \(H\),

\[
[\rho_\beta,H]=0,
\qquad
e^{-iHt/\hbar}\rho_\beta e^{iHt/\hbar}=\rho_\beta.
\tag{AP.3}
\]

This written result holds for positive or negative \(\beta\) whenever the Gibbs
state exists. With the NIST-clocked \(10.198836391\) eV gap, finite calculations
give excited-to-ground ratios of \(4.64611\times10^{-172}\) at 300 K,
\(7.24457\times10^{-6}\) at 10,000 K, and \(1.38034\times10^5\) at -10,000 K.
The negative-temperature row is meaningful only in this upper-bounded two-level
truncation. It reverses population ordering, not \(H\), \(t\), or the unitary
phase direction. A coherent control has phase slope
\(+1.5494760135\times10^{16}\) rad/s; replacing \(t\) by \(-t\) changes its sign.

Positive temperature has no finite maximum Kelvin endpoint: as
\(T\to+\infty\), \(\beta=1/(k_BT)\to0^+\). Negative-temperature equilibrium is
a special population-inverted regime for an upper-bounded spectrum, not a
colder continuation past 0 K. The complete hydrogen spectrum includes an
unbounded continuum, so the two-level negative-temperature calculation cannot
be promoted to a global equilibrium state of free hydrogen.

### AP.4. Melting delay and evidence boundary

A frozen sample can require more supplied energy and time because its temperature
must rise and its phase change absorbs latent heat. Heat-transfer rate, geometry,
pressure, heat capacity, and material state also matter. None of those macroscopic
facts alone identifies a clockwise or counterclockwise microscopic rotation.

All 29 unit-test groups pass. A separately written audit reconstructs the
spectroscopic energy and period, the three population ratios, the time-reversal
sign, the saved mixture metrics and weights, and all six decisions; 11 of 11
checks pass. These are mathematics, finite computation, and synthetic tests.
Only the NIST spectral clock is measurement-derived. No measured time-resolved
3D hydrogen density, fourth spatial dimension, unrestricted hidden-variable
family, or theological cause follows.

The next admissible model must state a new interaction law, preferably as a
Hamiltonian or continuity equation, and a fixed parameter budget. It must be
frozen before using a new experimental preparation with a matching spatial
observable. The old holdout is now consumed by this prespecified extension and
must not be used again to tune a replacement.

## Appendix AQ. Electromagnetic correction ladder before higher-dimensional forces

### AQ.1. Correction: electromagnetism was already present

The release-0.3 ideal hydrogen Hamiltonian was incomplete, but it did not omit
electromagnetism. Its potential

\[
V(r)=-\frac{e^2}{4\pi\epsilon_0 r}
\tag{AQ.1}
\]

is the nonrelativistic electromagnetic interaction between a point proton and
an electron. The fixed-proton approximation omitted the proton's motion; the
Schrodinger treatment also omitted relativistic spin structure and radiative
quantum-electrodynamic corrections. “Missing quantum electromagnetism” is
therefore replaced by the narrower statement “missing precision two-body,
relativistic, and QED structure.”

This matters methodologically. If a deliberately incomplete standard baseline
misses a measured value, the residual cannot be assigned to a new force until
the known corrections have been included or bounded.

### AQ.2. Frozen H0/H1/H2 ladder

NIST gives neutral-hydrogen levels

\[
\tilde\nu(2p\,{}^2P^\circ_{1/2})=82258.9191\ \mathrm{cm}^{-1},
\qquad
\tilde\nu(2p\,{}^2P^\circ_{3/2})=82259.2850\ \mathrm{cm}^{-1}.
\tag{AQ.2}
\]

Their vacuum wavelengths are (121.567364480) nm and
(121.566823733) nm. Weighting them by the NIST strong-line ratio (500:1000)
gives the frozen comparison centroid (121.567003982) nm. Three nested models
were evaluated without fitting:

1. **H0:** fixed-proton nonrelativistic Coulomb spectrum;
2. **H1:** H0 with the electron-proton reduced mass
   \(\mu=m_em_p/(m_e+m_p)\);
3. **H2:** point-nucleus Dirac-Coulomb energies with the same reduced mass
   substituted.

For H1, Coulomb transition energies scale by (mu/m_e), so wavelengths scale
by (m_e/\mu=1+m_e/m_p). With the 2022 CODATA mass ratio,

\[
\frac{\mu}{m_e}=0.999455679424762.
\tag{AQ.3}
\]

The point-Coulomb Dirac binding energy used for H2 is

\[
E_{nj}=\mu c^2\left[
1+\frac{\alpha^2}{
\left(n-j-\tfrac12+\sqrt{(j+\tfrac12)^2-\alpha^2}\right)^2}
\right]^{-1/2}-\mu c^2.
\tag{AQ.4}
\]

It is a controlled intermediate model, not a complete bound-state-QED result.

### AQ.3. Result

| Model | Predicted centroid (nm) | Centroid error (ppm) |
|---|---:|---:|
| H0 fixed-proton Schrodinger | 121.502273411 | -532.4683 |
| H1 reduced-mass Schrodinger | 121.568445617 | +11.8588 |
| H2 reduced-mass Dirac | 121.566602395 | -3.3034 |

Reduced mass alone removes (97.7729\%\) of H0's absolute centroid error.
H2 removes (99.3796\%\), predicts two distinct lines at
(121.566962046) nm and (121.566422570) nm, and predicts a wavelength
fine-structure split of (0.000539476) nm. The NIST level split is
(0.000540747) nm, so the declared H2 split error is (-0.2351\%\).

All 35 unit tests pass. A separately written 50-digit Decimal audit imports no
production functions and reproduces the six central observed/model wavelengths
and centroids within (2\times10^{-8}) nm. The NIST levels are measured or
critically compiled evidence; Equations AQ.3--AQ.4 and the comparisons are
written mathematics plus deterministic finite computation. No new hydrogen
measurement was performed.

### AQ.4. Residual budget and higher-dimensional admission rule

H2 still omits complete relativistic recoil, radiative self-energy and vacuum
polarization (the Lamb/QED contribution), finite proton charge radius,
hyperfine structure, radiative linewidth, and external-field effects. The
remaining (-3.3034) ppm centroid residual is therefore first a budget for
omitted standard terms. It does not identify a unique new force, a string or
rotor ontology, a fourth spatial coordinate, or a theological cause.

A higher-dimensional proposal becomes testable only after specifying

\[
H_{\mathrm{candidate}}=H_{\mathrm{standard}}+\lambda\mathcal O_4,
\tag{AQ.5}
\]

where the extra coordinate and units, the field equation or action deriving
(\mathcal O_4), its coupling to charge/current, its gauge or conservation law,
and a finite parameter budget are declared before fitting. It must then predict
a fresh state-dependent pattern that cannot be absorbed by standard corrections
or refitted constants. No such operator is specified in this revision, so no
higher-dimensional fit is performed and the physical origin remains unresolved.

The next test is to complete and independently validate the standard precision
ladder across several hydrogen transitions--not only Lyman-alpha--before freezing
one explicit higher-dimensional operator. The earlier rotor rejection, closure
idempotence, integer-tick collision boundary, angular-impulse correction, and
finite-prefix generator limitation remain unchanged.

## Appendix AR. Fixed-grid and fixed-tick falsification engine

### AR.1. Two hypotheses that must not be conflated

A fixed spacetime lattice asserts coordinate labels such as

\[
x_i=n_i\Delta x,\qquad t=k\Delta t,
\quad n_i,k\in\mathbb Z.
\tag{AR.1}
\]

That does **not** by itself imply that every field amplitude, energy eigenvalue,
frequency, wavelength, temperature, or other derived observable is an integer
multiple of \(\Delta x\) or \(\Delta t\). Lattice field theories routinely put
coordinates on a grid while fields remain real-, complex-, group-, or
operator-valued. This revision therefore distinguishes:

- **Lattice-coordinate hypothesis:** positions and times lie on fixed coordinate
  increments; and
- **Direct-multiple hypothesis:** each declared direct measurement satisfies
  \(y_i=n_i\Delta\) for an integer \(n_i\).

The second is a stronger, separately falsifiable hypothesis. Failure of one
proposed direct tick does not logically refute every lattice dynamics, but a
theory that explicitly requires every observable to be a direct multiple must
survive that stronger test.

### AR.2. Interval decision and non-identification theorem

For a reported value \(y_i\) with declared tolerance \(\epsilon_i>0\), a proposed
tick \(\Delta>0\) passes exactly when

\[
\min_{n\in\mathbb Z}|y_i-n\Delta|\leq\epsilon_i
\tag{AR.2}
\]

for every measurement. One incompatible interval rejects that proposed tick
under the direct-multiple hypothesis.

**Theorem AR.1 (finite-precision non-identification).** For any finite set of
measurements with positive tolerances, every positive tick satisfying

\[
\Delta\leq 2\min_i\epsilon_i
\tag{AR.3}
\]

passes Equation AR.2.

**Proof.** Choose \(n_i\) by rounding \(y_i/\Delta\) to its nearest integer. Then
\(|y_i-n_i\Delta|\leq\Delta/2\leq\min_j\epsilon_j\leq\epsilon_i\) for every
\(i\). \(\square\)

Thus finite-precision data alone cannot discover or confirm an arbitrarily
small universal tick. Identification additionally requires a lower bound on the
tick, a restricted integer-count law, a complexity penalty, or new predictions
whose resolution scales appropriately. This is a written mathematical result,
not a statistical tendency.

### AR.3. Frozen hydrogen-spectroscopy example

The engine uses the two NIST strong neutral-hydrogen Lyman-alpha wavelengths
printed as \(121.566824\) nm and \(121.567364\) nm and converts them to vacuum
light-travel times with exact \(c=299,792,458\) m/s. Half the last printed digit,
\(5\times10^{-7}\) nm, is treated only as a digit-resolution interval; it is not
claimed to be NIST's complete experimental uncertainty. This gives

\[
\epsilon_t=1.6678204759907602\times10^{-24}\ \mathrm{s},
\qquad
2\epsilon_t=3.3356409519815205\times10^{-24}\ \mathrm{s}.
\tag{AR.4}
\]

| Proposed direct time tick | Frozen result | Largest residual / digit tolerance |
|---|---:|---:|
| \(10^{-18}\) s | rejected | 297,827.896 |
| \(10^{-21}\) s | rejected | 165.807252 |
| \(10^{-22}\) s | rejected | 14.0682228 |
| CODATA-2022 Planck time \(5.391247\times10^{-44}\) s | compatible | \(1.2853\times10^{-20}\) |
| Printed-wavelength quantum \((10^{-15}\ \mathrm m)/c\) | compatible by construction | \(<6.0\times10^{-82}\) in Decimal computation |

The last row is exactly divisible in rational arithmetic because each printed
wavelength is an integer number of \(10^{-15}\) metres. It is a reporting-grid
artifact, not a discovery about nature. The Planck-time candidate also passes
automatically because it is about twenty orders of magnitude below even the
nearest-multiple automatic-fit floor. Its pass is consistency only and supplies
no positive evidence for that tick.

### AR.4. Exact four-coordinate discrete string

To make the proposed hidden-motion geometry concrete without pretending it is
already hydrogen, the lab defines four integer spatial coordinate fields
\(U_A(s,k)\), \(A\in\{x,y,z,w\}\), on periodic string sites \(s\) and integer
time ticks \(k\):

\[
U_A(s,k)=F_A((s-k)\bmod N)+G_A((s+k)\bmod N).
\tag{AR.5}
\]

For integer profiles \(F_A,G_A\), every state is integer-valued and Equation
AR.5 obeys the exact Courant-one discrete wave equation

\[
U_A(s,k+1)=U_A(s-1,k)+U_A(s+1,k)-U_A(s,k-1).
\tag{AR.6}
\]

The same equation solved for \(U_A(s,k-1)\) gives exact one-tick reversal. The
implemented 64-site profile is bounded and periodic after 64 ticks. A 3D image
is the occupancy obtained by dropping \(w\). Changing only \(w\) produces a
different full state with the identical 3D image, giving a constructive proof
that the projection is non-injective. Therefore visible 3D waves do not by
themselves reconstruct the hidden coordinate or establish that it is physical.

Across 128 generated ticks, production and independent scalar implementations
both check 32,256 wave-recurrence values and 32,256 reverse-recurrence values.
The production verifier also checks 16,384 periodicity values. All 13 unit tests
pass, and the independent audit reproduces all five tick decisions plus the
projection collision. These are exact finite computations for one toy rule.

### AR.5. Failure criterion and next physical gate

The executable theory now has a real failure condition: any proposed direct
tick is rejected when one verified direct measurement interval contains no
integer multiple. A universal claim over “all observable values ever” cannot be
confirmed by a finite list, and shrinking \(\Delta\) always makes finite
interval fitting easier. The current engine therefore rejects three coarse
candidate ticks but neither identifies a fundamental tick nor refutes every
possible lattice-coordinate theory.

The next admissible physical test must add a derived discrete hydrogen
Hamiltonian or update law on the same declared \((\Delta x,\Delta t)\), specify
how charge and electromagnetic gauge structure enter, freeze its parameters,
and predict multiple hydrogen transitions before comparison. A continuum-limit
convergence study must separate lattice discretization error from any claimed
new physics. Until then, a fifth-dimensional string-particle interpretation is
a hypothesis; the simulation is a consistent kinematic example, not measured
evidence for it. All v1.14 and v1.15 corrections, including the restricted
\(2K\)-sample recurrence guarantee, remain unchanged.

## Appendix AS. Geometry, phase boundaries, and conductivity collisions

### AS.1. Scope and type discipline

Revision 1.44 asks one bounded question: can a fixed, elementary molecular
geometry coordinate predict a held-out phase boundary after composition is
held constant?  It does not infer microscopic strings from a successful
correlation.  The input temperatures and resistivities are measured/reference
data; graph indices, interpolation, classifications, and bounds are finite
computations; the representation-collision inequalities are written
mathematics; a higher-dimensional cause remains a hypothesis.

A phase is not an intrinsic label of an isolated formula.  For a pure material
in equilibrium, the stable phase at specified temperature and pressure is the
one with the least appropriate free energy.  The executable classification in
this appendix is restricted to approximately one atmosphere and uses reported
fusion and normal-boiling boundaries.  It is not a universal equation of
state.

### AS.2. A representation-collision bound

Suppose a representation maps two materials to the same feature vector and a
deterministic predictor consequently assigns both the same scalar value \(p\).
For observed properties \(a<b\), the triangle inequality gives

\[
\max\{|p-a|,|p-b|\}\geq \frac{b-a}{2}.
\tag{AS.1}
\]

For positive properties evaluated multiplicatively, the best common
prediction is their geometric mean and its unavoidable worst factor is

\[
\inf_{p>0}\max\left\{\frac{p}{a},\frac{a}{p},
\frac{p}{b},\frac{b}{p}\right\}=\sqrt{\frac{b}{a}}.
\tag{AS.2}
\]

These are obstructions to an insufficient representation, not claims that a
particular richer representation is true.

### AS.3. Frozen same-formula molecular test

NIST Chemistry WebBook SRD 69 lists the following average phase-change values
for three isomers with the same formula \(\mathrm{C_5H_{12}}\) and molecular
mass 72.1488 g/mol.  The lab computes the Wiener index \(W\), the sum of all
unordered pairwise graph distances in the five-carbon skeleton.

| Isomer | Carbon-graph degree sequence | \(W\) | Fusion point (K) | Normal boiling point (K) | Phase at 293.15 K, about 1 atm |
|---|---|---:|---:|---:|---|
| neopentane | 4,1,1,1,1 | 16 | 255 | 282.6 | gas |
| isopentane | 3,2,1,1,1 | 18 | 113 | 301.1 | liquid |
| n-pentane | 2,2,2,1,1 | 20 | 143.4 | 309.2 | liquid |

The protocol chooses the minimum- and maximum-\(W\) structures without reading
their temperatures, fits the unique affine rule through them, and predicts the
middle structure once:

\[
\widehat T_b(W)=6.65W+176.2\ \mathrm{K}.
\tag{AS.3}
\]

For held-out isopentane, Equation AS.3 predicts 295.9 K versus the NIST average
301.1 K, an absolute error of 5.2 K.  The prediction is above 293.15 K and thus
places the molecule on the condensed side of its boiling boundary; the
separately measured 113 K fusion point identifies the observed equilibrium
phase there as liquid.  This phase statement is conditional on the supplied
fusion datum and approximately one-atmosphere equilibrium, not an independent
three-phase prediction.

Composition alone cannot distinguish these rows.  Equation AS.1 and the 26.6 K
boiling range imply that any one-value formula-only predictor has worst error
at least 13.3 K.  The held-out computation therefore supports a narrow claim:
connectivity supplies predictive information that atom counts omit.

### AS.4. Counterexamples to one-dimensional harmony

The same coordinate does not order the fusion temperatures: neopentane has the
smallest \(W\) and lowest boiling point but by far the highest fusion point of
the three.  Molecular symmetry, three-dimensional conformation, crystal
packing, entropy, and intermolecular interactions are separate variables.  A
single graph scalar can therefore capture one trend while failing another
property of the same molecules.

The elemental control has the same form.  Matula's recommended total
resistivities at 300 K are \(1.629\times10^{-8}\ \Omega\,\mathrm m\) for silver
and \(10.54\times10^{-8}\ \Omega\,\mathrm m\) for palladium.  *Revision 1.69
note:* the silver figure is the 300 K entry while the palladium figure matches
the 293 K entry of the same tabulation; at a common 300 K the ratio is about
6.63 and the Equation AS.2 bound about 2.575. The qualitative conclusion is
unchanged and the released computation is retained below.  Both are
period-5, fcc, d10 metals under the lab's deliberately coarse signature, yet
their resistivity ratio is 6.470.  Equation AS.2 says that any one resistivity
assigned to that shared signature has unavoidable worst multiplicative error
at least 2.544.  The differing valence occupancy already signals missing
information; quantitative conduction additionally depends on band dispersion,
Fermi-surface geometry, carrier velocities, phonons, defects, and scattering
lifetimes.

### AS.5. Verification, interpretation, and next test

All 12 unit tests pass.  A separately implemented Floyd--Warshall/scalar audit
reconstructs the three graphs, their distance sums and degree sequences, all
three phase rows, the two collision bounds, the single held-out prediction,
and the Ag/Pd ratio.  The observed numbers are not new measurements and the
small three-isomer test does not establish generalization to arbitrary
molecules.

The constructive result is that a geometry-first program is viable only as a
typed multivariable predictor: composition, connectivity, conformation,
symmetry, temperature, pressure, and interaction/electronic variables must be
declared before testing.  Failure cannot be automatically assigned to an
unseen string, rotation, fifth dimension, or theological cause.  Such a new
coordinate earns physical standing only if one shared, independently motivated
coupling improves fresh held-out predictions over standard quantum-chemical or
solid-state baselines.

The next bounded test is a preregistered alkane-isomer study with substantially
more molecules.  It should freeze graph and three-dimensional descriptors,
train on one subset, predict both boiling and fusion temperatures on a sealed
subset, and compare against a composition-only baseline plus a standard
quantitative structure--property baseline.  A separate conductivity test must
use measured band/transport descriptors and temperature-dependent data; it
must not infer a hidden dimension from the residual.  All v1.14 and v1.15
corrections, including idempotent complete deterministic closure, sampled
non-crossing limits, the angular-impulse formula, and the restricted
\(2K\)-sample recurrence guarantee, remain unchanged.

## Appendix AT. Property-specific transfer across alkane isomers

### AT.1. Scope and frozen split

Revision 1.45 executes Appendix AS's reserved larger test without changing the
question after seeing a score: can elementary carbon-skeleton topology improve
prediction over composition alone when the carbon count itself moves outside
the fitted range?  The data are NIST Chemistry WebBook SRD 69 phase-change
records for every constitutional isomer of pentane, hexane, and heptane.

The training block contains all three \(\mathrm{C_5H_{12}}\) and all five
\(\mathrm{C_6H_{14}}\) isomers.  The holdout block contains all nine
\(\mathrm{C_7H_{16}}\) isomers.  No C7 target is used to fit, select, clip, or
recalibrate a coefficient.  This is a frozen-fit retrospective reference-data
test, not a cryptographically hidden prospective experiment or a new physical
measurement.

Two ordinary least-squares representations are fitted separately for normal
boiling temperature and fusion temperature.  The composition baseline is

\[
\widehat T=\beta_0+\beta_1 n,
\tag{AT.1}
\]

where \(n\) is the carbon count.  The topology model is

\[
\widehat T=\gamma_0+\gamma_1 n+\gamma_2 W+\gamma_3 M_1,
\qquad
W=\sum_{i<j}d(i,j),\quad M_1=\sum_i \deg(i)^2.
\tag{AT.2}
\]

Here \(W\) is the Wiener distance sum and \(M_1\) is the first Zagreb index of
the hydrogen-suppressed carbon tree.  Their historical use as chemical-graph
descriptors motivates the coordinates; it does not guarantee a physical
mechanism or out-of-range transfer.

### AT.1a. The frozen data table (added in revision 1.69)

Revision 1.45 scored the models below against a 17-row data ledger that was
not printed, so the C-labelled statistics of AT.2 and AT.3 could not be
recomputed from the manuscript. The table is now printed. The graph columns are
exact: each hydrogen-suppressed carbon skeleton is a tree on \(n\) vertices,
\(W\) is its Wiener distance sum and \(M_1\) its first Zagreb index, both
recomputed independently with a graph library. The temperature columns are
NIST Chemistry WebBook SRD 69 or CRC selections; the status column says which
values are printed elsewhere in this manuscript and which selections the
ledger must confirm.

| Isomer | \(n\) | \(W\) | \(M_1\) | \(T_b\) (K) | \(T_{\mathrm{fus}}\) (K) | Value status |
|---|---:|---:|---:|---:|---:|---|
| n-pentane | 5 | 20 | 14 | 309.2 | 143.5 | \(T_b\) printed in AS.3 |
| 2-methylbutane | 5 | 18 | 16 | 301.1 | 113.3 | \(T_b\) printed in AS.3 |
| 2,2-dimethylpropane | 5 | 16 | 20 | 282.6 | 256.6 | \(T_b\) printed in AS.3 |
| n-hexane | 6 | 35 | 18 | 341.9 | 177.8 | standard |
| 2-methylpentane | 6 | 32 | 20 | 333.4 | 119.5 | standard |
| 3-methylpentane | 6 | 31 | 20 | 336.4 | 155.0 | **two literature values**: about 155 K (NIST) and 110.3 K (CRC); 155 K is the value that reproduces AT.3, and the ledger selection must be stated |
| 2,2-dimethylbutane | 6 | 28 | 24 | 322.9 | 173.3 | standard |
| 2,3-dimethylbutane | 6 | 29 | 22 | 331.1 | 144.6 | standard |
| n-heptane | 7 | 56 | 22 | 371.6 | 182.6 | \(T_{\mathrm{fus}}\) printed in AT.3 |
| 2-methylhexane | 7 | 52 | 24 | 363.2 | 154.9 | standard |
| 3-methylhexane | 7 | 50 | 24 | 365.0 | 153.7 | standard |
| 2,2-dimethylpentane | 7 | 46 | 28 | 352.3 | 149.3 | standard |
| 2,3-dimethylpentane | 7 | 46 | 26 | 362.9 | 149.36 | AT.3: NIST lists 149.36 and 139.85 K; 149.36 frozen |
| 2,4-dimethylpentane | 7 | 48 | 26 | 353.6 | 153.9 | standard |
| 3,3-dimethylpentane | 7 | 44 | 28 | 359.2 | 138.7 | standard |
| 3-ethylpentane | 7 | 48 | 24 | 366.6 | 154.6 | standard |
| 2,2,3-trimethylbutane | 7 | 42 | 30 | 354.0 | 248.3 | \(T_{\mathrm{fus}}\) printed in AT.3 as 248 K |

Two facts about the graph columns matter for the interpretation in AT.3 and AT.4.
First, \(W\) grows with the cube of \(n\) for chains and is bounded between the
star and the path for every tree on \(n\) vertices, so the training range
\(W\in[16,35]\) is extrapolated to \(W\in[42,56]\) on the holdout. Second, on
the eight training rows the columns \(n\), \(W\) and \(M_1\) are nearly
collinear, so a four-coefficient fit is poorly conditioned. The fusion
coefficients of the refit below, roughly \(+1940\), \(-603\) per carbon,
\(+32\) per unit of \(W\) and \(+40\) per unit of \(M_1\), are the signature of
that conditioning. The 410 K n-heptane prediction of AT.3 is therefore an
extrapolation artifact of a collinear least-squares fit as much as a statement
about melting, and AT.3's sentence "the declared two-dimensional carbon graph
is not an adequate melting representation" should be read with the list of
ordinary alternatives in AT.4, where small-sample extrapolation and model form
already appear.

**Independent refit from this table [C].** Fitting Equations AT.1 and AT.2 on
the eight C5 and C6 rows and scoring the nine C7 rows gives:

| Property | Model | MAE (K) | RMSE (K) | Max abs (K) | Mean signed (K) | Manuscript MAE / RMSE |
|---|---|---:|---:|---:|---:|---|
| boiling | carbon count | 8.370 | 9.923 | 16.347 | +7.713 | 8.554 / 10.107 |
| boiling | \(n+W+M_1\) | 4.774 | 5.647 | 9.668 | −4.387 | 4.500 / 5.382 |
| fusion | carbon count | 28.093 | 42.158 | 111.353 | −28.093 | 27.574 / 41.839 |
| fusion | \(n+W+M_1\) | 135.282 | 148.018 | 226.330 | +135.282 | 135.934 / 148.789 |

The n-heptane topology fusion prediction from this table is 408.9 K versus the
manuscript's 410.46 K. Every qualitative conclusion of AT.2 and AT.3
reproduces: topology roughly halves the boiling error, fusion errors grow
fivefold, all fusion predictions are too high, and the composition baseline
misses 2,2,3-trimethylbutane by about 111 K. The remaining differences of
0.2 to 0.7 K in the scores arise from the specific NIST value selected for
each row in the unpublished ledger; the author should overwrite the two
temperature columns from that ledger so that the printed statistics are
reproducible to the last digit. The 34 values decoded in AU.3 are these two
columns.

### AT.2. Held-out boiling result

The C7 boiling scores are:

| Frozen model | MAE (K) | RMSE (K) | Maximum absolute error (K) | Mean signed error (K) |
|---|---:|---:|---:|---:|
| carbon count only | 8.554 | 10.107 | 16.627 | +7.982 |
| carbon count + \(W+M_1\) | 4.500 | 5.382 | 9.131 | -4.059 |

Thus the topology model improves MAE by 4.054 K and RMSE by 4.725 K on all nine
held-out structures.  This is finite computation applied to compiled measured
or critically evaluated inputs.  It supports the narrow inference that these
two connectivity statistics carry transferable information about boiling
within the declared C5--C7 alkane family.  It does not show that the statistics
are microscopic causes, that they remain adequate for other chemical classes,
or that a four- or five-dimensional motion has been observed.

### AT.3. Fusion counterexample

Using the identical representations and split for fusion gives the opposite
result:

| Frozen model | MAE (K) | RMSE (K) | Maximum absolute error (K) | Mean signed error (K) |
|---|---:|---:|---:|---:|
| carbon count only | 27.574 | 41.839 | 110.807 | -27.574 |
| carbon count + \(W+M_1\) | 135.934 | 148.789 | 227.863 | +135.934 |

The topology extrapolation predicts an unphysical 410.46 K fusion temperature
for n-heptane versus the selected 182.6 K reference value.  The prediction is
reported without clipping.  Even the composition baseline misses the highly
symmetric 2,2,3-trimethylbutane value of 248 K by 110.81 K.  The failure is
therefore not a minor coefficient correction: the declared two-dimensional
carbon graph is not an adequate melting representation across this split.

One NIST page lists two fusion values for 2,3-dimethylpentane, 149.36 K and
139.85 K.  The first is frozen in the scored table and the alternate is retained
in the data ledger.  Their 9.51 K difference cannot rescue a topology MAE of
135.934 K, so the principal conclusion is insensitive to this selection.

### AT.4. What a residual can and cannot establish

A failed prediction can be caused by omitted variables, a misspecified
functional form, distribution shift, noisy reference data, or an incorrect
physical hypothesis.  It does not by itself prove that the engine is correct
and only an unseen string coordinate is missing.  In the present problem,
ordinary omitted quantities already include three-dimensional conformation,
molecular symmetry beyond two scalar indices, polymorphism, crystal packing,
and intermolecular contributions to free energy.  A fifth-dimensional string
model would have to specify a new coordinate and coupling before looking at a
fresh target block, then beat these ordinary alternatives on that block.

The distinction is now explicit:

- NIST temperatures are measured or critically evaluated reference inputs.
- Graph indices, least-squares coefficients, predictions, and scores are
  finite computations.
- Equations AT.1--AT.2 define empirical models, not fundamental laws.
- The statement that packing-related variables are missing is a supported
  diagnosis of representation insufficiency, not a unique causal proof.
- Higher-dimensional strings, universal grid ticks, and theological meaning
  remain hypotheses or interpretations, not outputs of this experiment.

### AT.5. Verification and next test

All 17 unit tests pass.  A separately written audit reconstructs all 17 carbon
graphs by Floyd--Warshall, solves the four normal-equation systems with an
independent Gaussian-elimination routine, reproduces 36 held-out scalar
predictions, checks 28 coefficients and metrics, and reruns the preserved
v0.1.0 pentane and Ag/Pd controls.  This verifies implementation consistency,
not physical truth.

The next test may not repair Equation AT.2 against the C7 target and then reuse
those targets as confirmation.  It should define crystal-sensitive
three-dimensional descriptors using the C5--C7 development data, freeze
separate boiling and fusion models, and evaluate once on all 18 octane isomers.
The present topology model remains a required baseline.  Existing v1.14 and
v1.15 corrections remain unchanged: fixed-family complete closure is
idempotent; sampled integer-tick capture is not continuous non-crossing;
angular impulse has no extra time factor; arbitrary generators are not
identified by a finite prefix; and the \(2K\)-sample continuation guarantee
still assumes a noiseless homogeneous constant-coefficient recurrence of order
at most \(K\) from the initial state.

## Appendix AU. Latent rotation matrices and the inverse-projection boundary

### AU.1. Question and evidence types

This appendix answers a narrower form of the proposed reverse-engineering
program: can reference melting and boiling data be placed into explicit
higher-dimensional motion matrices, and do those data determine the hidden
motion? The first answer is yes as a mathematical construction. The second is
no without additional, independently constrained observables.

The types must remain separate:

- The boiling and fusion temperatures are measured or critically evaluated
  reference inputs compiled in Matter Harmony Lab 0.2.0.
- The graph-distance embedding, hidden coordinates, generator, projector,
  frequency encoder, and phase kernel are constructed coordinates.
- Norm preservation, the two non-identifiability theorems of AU.4, and the
  relabeling counterexample of AU.3 are
  written mathematics.
- The 17 molecular matrices, 65-frame trajectory, tests, and independent audit
  are finite computations.
- The statement that any one generator describes physical molecular motion in
  a fourth or fifth spatial dimension is a hypothesis.
- No theological interpretation follows from a fitted or unfitted residual.

### AU.2. Explicit five-coordinate dynamics and 3D projection

For a molecule with \(N\) carbon vertices, let \(X_5(0)\in\mathbb R^{N\times5}\)
be a row-state matrix. Its first three columns are a classical
multidimensional-scaling embedding of carbon-graph shortest-path distances.
The fourth and fifth columns are centered vertex degree and eccentricity,
scaled to 35% of the visible root-mean-square radius. These choices make a
reproducible synthetic geometry; they are not experimental conformers,
orbitals, or probability amplitudes.

Let

\[
P=\begin{bmatrix}1&0&0&0&0\\0&1&0&0&0\\0&0&1&0&0\end{bmatrix},
\qquad \Omega^T=-\Omega.
\tag{AU.1}
\]

The latent state, visible position, and visible velocity are

\[
X_5(t)=X_5(0)e^{t\Omega^T},\qquad
X_3(t)=X_5(t)P^T,\qquad
V_3(t)=X_5(t)\Omega^TP^T.
\tag{AU.2}
\]

Because \(\Omega\) is skew-symmetric, \(R(t)=e^{t\Omega}\) is orthogonal:
\(R(t)^TR(t)=I\). Hence each row norm of \(X_5(t)\) is constant. This is an
exact rotation law in the declared Euclidean latent space.

The implemented generator couples visible coordinate 1 to hidden coordinate
4 at angular rate \(\omega_4\), visible coordinate 2 to hidden coordinate 5 at
\(\omega_5\), visible coordinate 3 to hidden coordinate 5 at \(\omega_3\),
and the two hidden coordinates to each other at \(\omega_h\). For the displayed
n-heptane candidate, the rounded matrix is

\[
\Omega=\begin{bmatrix}
0&0&0&-1.6000&0\\
0&0&0&0&-0.9882\\
0&0&0&0&-1.2941\\
1.6000&0&0&0&-0.3059\\
0&0.9882&1.2941&0.3059&0
\end{bmatrix}.
\tag{AU.3}
\]

Equation AU.2 generates an explicit \(N\times3\) position matrix and an
\(N\times3\) velocity matrix at every time. A pairwise display kernel is

\[
K_{ij}(t)=\cos(\phi_i(t)-\phi_j(t))
\exp\!\left[-\frac{\|x_i(t)-x_j(t)\|^2}{2\sigma(t)^2}\right],
\tag{AU.4}
\]

where \(\phi_i\) is the polar angle of the two hidden coordinates and
\(\sigma\) is the median nonzero visible separation. Equation AU.4 visualizes
phase alignment and spatial overlap. It is not a derived electromagnetic,
quantum, or intermolecular force.

### AU.3. Property coordinates and the relabeling counterexample

For a scalar property \(y\in[y_{\min},y_{\max}]\), the engine defines

\[
\omega(y)=\omega_{\min}+
(\omega_{\max}-\omega_{\min})
\frac{y-y_{\min}}{y_{\max}-y_{\min}},
\qquad (\omega_{\min},\omega_{\max})=(0.4,1.6).
\tag{AU.5}
\]

Boiling and fusion values receive separate coordinates \(\omega_4\) and
\(\omega_5\). Equation AU.5 is bijective on the finite interval, so inverse
decoding must reproduce every source value. The 34 values for 17 molecules (the two temperature columns of Table AT.1a)
decode with maximum numerical error \(2.8422\times10^{-14}\) K.

That exact agreement is not evidence for the latent mechanism. Let \(\pi\) be
any permutation of a finite target table. Encoding \(y_{\pi(i)}\) by Equation
AU.5 and immediately applying its inverse also returns \(y_{\pi(i)}\) exactly.
The executable shuffled-label control does so with the same
\(2.8422\times10^{-14}\) K maximum error. Thus a target-derived rotation label
is a lossless reparameterization, not a prediction or causal explanation.

### AU.4. Two independent non-identifiability theorems

**Hidden-kernel theorem.** Let
\(H(t)=\operatorname{diag}(I_3,Q(t))\), where \(Q(t)\) is any two-dimensional
rotation. Since \(PH(t)=P\),

\[
X_5(0)H(t)^TP^T=X_5(0)P^T.
\tag{AU.6}
\]

Therefore every real hidden-plane angular speed produces the identical 3D
trajectory when that plane is uncoupled from visible coordinates. The finite
test finds zero projected difference over 101 sampled times, but Equation AU.6
is the proof for all real times.

**Fixed-tick alias theorem.** In any rotation plane sampled only at
\(t_n=n\Delta\), define \(\omega_k=\omega+2\pi k/\Delta\) for integer \(k\).
Then

\[
R(\omega_k n\Delta)=R(\omega n\Delta+2\pi kn)=R(\omega n\Delta).
\tag{AU.7}
\]

Hence an exact fixed tick does not by itself identify the continuous angular
rate. The implementation checks 205 alias pairs independently; the largest
floating-point discrepancy is \(4.931\times10^{-11}\) at deliberately large
alias frequencies. This is numerical roundoff around the exact equality. As
already required by revision 1.15, finite sampled prefixes do not identify an
arbitrary generator.

### AU.5. Result, limitations, and next test

Matter Harmony Lab 0.3.0 creates matrices for all 17 frozen alkane graphs and a
65-frame n-heptane animation. All 30 unit tests pass: the 17 preserved property
tests and 13 new projection tests. A separately written audit imports no
production module and reproduces 68 property/frequency checks, 2,730 projected
position and velocity scalars, 124 hidden-invariance states, and 205
integer-tick alias pairs.

The result establishes an internally consistent candidate family and a sharp
inverse-problem boundary. It does not establish that temperature is an angular
speed, that melting reverses rotation, that the displayed graph embedding is a
real molecular shape, or that a hidden dimension caused any property. Static
scalar values cannot distinguish the infinitely many constructions in
Equations AU.6--AU.7. Standard alternatives—including conformer geometry,
crystal packing, electronic structure, pressure, intermolecular forces, and
model misspecification—remain live.

The next admissible experiment must define one shared descriptor-to-generator
map without reading the property being predicted. It should be frozen using
C5--C7 development molecules and evaluated once on reserved C8 structures or,
preferably, on a genuinely time-dependent or spectroscopic observable with a
declared observation operator. The topology baseline and ordinary physical
descriptors must remain in the comparison. Static target values may score the
prediction but may not define its latent frequency.

Revision 1.14's idempotent complete deterministic closure remains unchanged.
Revision 1.15's sampled non-crossing limit, angular-impulse formula, arbitrary-
generator non-identifiability, and the recurrence learner's assumption also
remain unchanged: the \(2K\)-sample continuation guarantee applies only to a
noiseless homogeneous constant-coefficient linear recurrence of order at most
\(K\) from the initial state. No seven-dimensional, P=NP, physical-hidden-
string, or theological claim is introduced by this appendix.

## Appendix AV. Water harmonics as a coupled projection experiment

### AV.1. Scope and calibrated three-dimensional inputs

The bounded task is to animate many connected water molecules while making the
proposed higher-coordinate movement mathematically explicit. The construction
starts with ordinary quantities rather than inferring them from latent motion:

1. \(N=64\) water molecules;
2. molar mass \(M=18.0153\,\mathrm{g\,mol^{-1}}\);
3. density \(\rho=997.047\,\mathrm{kg\,m^{-3}}\) at \(298.15\,\mathrm K\);
4. rigid three-site geometry \(r_{\mathrm{OH}}=0.9572\) angstrom and
   \(\angle\mathrm{HOH}=104.52^\circ\).

The volume per molecule is

\[
v_1=\frac{M}{N_A\rho},                                                  \tag{AV.1}
\]

and the conventional diamond-cell length containing eight sites is

\[
a=(8v_1)^{1/3}=6.214721\ \text{angstrom}.                              \tag{AV.2}
\]

Two cells in each direction give a \(12.429442\)-angstrom periodic box. Its
128 nearest-neighbor edges have degree four and a common minimum-image
oxygen--oxygen distance \(a\sqrt3/4=2.691053\) angstrom. This is an ordered
tetrahedral test scaffold chosen for a compact connected control. Matching
density and rigid geometry is therefore by construction, not a prediction, and
the scaffold is not identified with an instantaneous configuration of liquid
water.

### AV.2. Five coordinates per molecule and a norm-preserving coupling

Let

\[
q\in\mathbb R^{3N},\qquad h\in\mathbb R^{2N},\qquad
z=\begin{bmatrix}q\\h\end{bmatrix}.                                    \tag{AV.3}
\]

The visible vector \(q\) contains three dimensionless displacement coordinates
per molecule. The vector \(h\) contains two declared hidden coordinates per
molecule. A coordinate- and graph-derived matrix

\[
C\in\mathbb R^{3N\times2N}                                              \tag{AV.4}
\]

couples each hidden pair to two local visible directions, using 58% local and
42% nearest-neighbor weight before spectral normalization. These percentages
are frozen construction choices, not fitted physical constants. Define

\[
G=\begin{bmatrix}0&C\\-C^T&0\end{bmatrix},\qquad
\frac{dz}{d\tau}=Gz.                                                    \tag{AV.5}
\]

Then \(G^T=-G\), and the written invariant is

\[
\frac{d}{d\tau}\|z\|_2^2=z^T(G+G^T)z=0.                               \tag{AV.6}
\]

The initial state is a localized two-phase hidden pulse with zero visible
displacement. At the dimensionless simulator ticks

\[
\tau_n=n\Delta\tau,\qquad \Delta\tau=0.125,                             \tag{AV.7}
\]

the engine evaluates \(z(\tau_n)=e^{\tau_nG}z(0)\) by a Hermitian spectral
decomposition. This fixed tick is a reproducible numerical sampling choice. It
is not a discovered universal physical increment.

### AV.3. Three-dimensional rendering and the hidden-null control

For molecule \(i\), the rendered oxygen center is

\[
R_i(\tau)=R_i^{(0)}+s q_i(\tau),\qquad s=0.34\ \text{angstrom},          \tag{AV.8}
\]

after which the two hydrogen centers are reconstructed at the fixed reference
bond length and angle. Translucent hue and radius encode

\[
\phi_i=\operatorname{atan2}(h_{i,2},h_{i,1}),\qquad
A_i=\|h_i\|_2,                                                          \tag{AV.9}
\]

and candidate-link appearance depends on

\[
I_{ij}=\tfrac12[1+\cos(\phi_i-\phi_j)].                                 \tag{AV.10}
\]

Equations AV.9--AV.10 are an observation convention that lets a viewer see the
latent phase relation. They are not charge density, electric field, magnetic
field, bond energy, or a new force law.

The mandatory counterexample separates hidden motion from visible motion. Set
\(C=0\) and rotate each hidden pair arbitrarily. Then \(q(\tau)=q(0)\) for all
times. The executed trajectory begins with \(q(0)=0\) and reports exactly zero
projected motion under this control. Thus a fifth coordinate does not create
an observable wave unless an explicit coupling is supplied. Moreover, the
integer-tick alias from Appendix AU remains: a two-plane frequency \(\omega\)
and \(\omega+2\pi k/\Delta\tau\) agree at all sampled ticks.

### AV.4. Executed evidence and its limits

The 64-frame run gives:

- maximum generator skew error \(0\);
- state-norm drift \(2.220\times10^{-16}\);
- maximum O--H error \(1.332\times10^{-15}\) angstrom;
- maximum H--O--H error \(1.137\times10^{-13}\) degrees;
- hidden-only projection error \(0\);
- integer-tick alias error \(5.519\times10^{-14}\);
- peak visible RMS state \(0.120060\) in its declared dimensionless units.

All 41 tests in Matter Harmony Lab 0.4.0 pass, including molecule-order
equivariance and equality of coarse and refined integrations at shared sample
times. A separately written program imports no production module and checks
36,864 rendered atomic-coordinate scalars, the density arithmetic, all 128
periodic links, rigid geometry, norm preservation, the hidden-null control, and
the tick alias. These are written mathematics and finite computations. No new
physical measurement was made.

The release does not implement a molecular force field, thermal ensemble,
diffusion, viscosity, dielectric response, magnetism, hydrodynamics, quantum
electrodynamics, or a physical higher-dimensional observation operator. It
therefore cannot yet explain water flow, electricity, or magnetism and cannot
validate fourth- or fifth-dimensional motion. The next admissible experiment
is to freeze an ordinary electrostatic water baseline with explicit partial
charges and compare a time-dependent dielectric or structure-factor observable
against held-out reference data. A latent coupling may be evaluated only as an
additional prespecified model, not as the default meaning of its residual.

Revision 1.14's idempotent complete deterministic closure remains unchanged.
Revision 1.15's sampled non-crossing limit, angular-impulse formula,
arbitrary-generator non-identifiability, and recurrence learner assumption
also remain unchanged: the \(2K\)-sample continuation guarantee applies only
to a noiseless homogeneous constant-coefficient linear recurrence of order at
most \(K\) from the initial state. This appendix neither establishes seven
physical dimensions nor assumes \(P=NP\), and no unconstrained residual is
given a theological interpretation.

## Appendix AW. Water origami and the oscilloscope observation boundary

### AW.1. Ordinary charge and dielectric baseline

The bounded question is whether a time-dependent electric measurement can be
turned into a visible many-water animation without treating the animation as a
unique microscopic reconstruction. Release 0.5.0 first adds ordinary partial
charges to the rigid geometry retained from Appendix AV:

\[
q_{\mathrm O}=-0.834e,\qquad q_{\mathrm H}=+0.417e.                  \tag{AW.1}
\]

Thus \(q_{\mathrm O}+2q_{\mathrm H}=0\). With
\(r_{\mathrm{OH}}=0.9572\) angstrom and
\(\theta_{\mathrm{HOH}}=104.52^\circ\), the charge-model molecular dipole is

\[
\mu=2q_{\mathrm H}r_{\mathrm{OH}}\cos(\theta_{\mathrm{HOH}}/2)
    =2.346971\ \mathrm D.                                             \tag{AW.2}
\]

This is a property of the declared TIP3P-like geometry, not a fresh measured
dipole. The macroscopic reference is the single-Debye linear response

\[
\tau_D\frac{dP}{dt}+P=E.                                               \tag{AW.3}
\]

The published 25-degree-C primary relaxation frequency
\(f_D=18.56\ \mathrm{GHz}\) is frozen as an input, giving

\[
\tau_D=(2\pi f_D)^{-1}=8.575159\ \mathrm{ps}.                          \tag{AW.4}
\]

For \(E(t)=\sin(\omega t)\), the steady response is

\[
P(t)=\frac{1}{\sqrt{1+(\omega\tau_D)^2}}
\sin\!\left(\omega t-\arctan(\omega\tau_D)\right).                  \tag{AW.5}
\]

At \(\omega\tau_D=1\), the gain is \(1/\sqrt2\) and the lag is
\(45^\circ\). A separately stepped Runge--Kutta calculation agrees with
Equation AW.5 to \(3.529\times10^{-10}\) over the executed interval. The
frequency agreement is calibration, not an out-of-sample prediction.

### AW.2. Origami as a many-to-one observation map

Let \(d_i(t)\in\mathbb R^3\) be the unit dipole direction of molecule \(i\),
and concatenate the \(N\) directions as \(x(t)\in\mathbb R^{3N}\). The
oscilloscope in this experiment reports only their vector mean:

\[
y(t)=Ax(t),\qquad
A=\frac1N\begin{bmatrix}I_3&I_3&\cdots&I_3\end{bmatrix}.              \tag{AW.6}
\]

For \(N=64\), \(A\) has rank three and ambient nullity \(3N-3=189\).
Calling this map an origami fold is an analogy with precise content: many
different high-dimensional states occupy the same observed point after the
fold.

The release constructs that collision without changing dipole lengths. Pair
the molecules. For one desired measured x component \(p(t)\), define each
pair by

\[
d_{2j}(t)=\begin{bmatrix}p(t)\\s(t)\cos\phi_j(t)\\s(t)\sin\phi_j(t)\end{bmatrix},
\qquad
d_{2j+1}(t)=\begin{bmatrix}p(t)\\-s(t)\cos\phi_j(t)\\-s(t)\sin\phi_j(t)\end{bmatrix},
\quad s(t)=\sqrt{1-p(t)^2}.                                            \tag{AW.7}
\]

Every vector in Equation AW.7 has unit norm, while every pair sums to
\((2p(t),0,0)^T\) independently of \(\phi_j(t)\). Therefore any two distinct
phase histories \(\phi_j^{(A)}(t)\) and \(\phi_j^{(B)}(t)\) give

\[
Ax_A(t)=Ax_B(t)=\begin{bmatrix}p(t)&0&0\end{bmatrix}^T                \tag{AW.8}
\]

for every real time at which the construction is defined. This is a written
non-identifiability result, not merely an animation coincidence.

### AW.3. Executed oscilloscope and counterexamples

The executable uses 96 stored frames over two periods. It maps the normalized
Debye response into \(|p(t)|<1\), chooses two different transverse phase
families, constructs both rigid three-site molecular movies, and computes the
scope trace again from the explicit charges. The results are:

- orientation-fold difference exactly zero in the stored arrays;
- explicit-charge trace difference \(4.441\times10^{-16}\);
- observation-kernel residual \(3.469\times10^{-18}\);
- microscopic RMS separation from \(0.688568\) to \(0.820871\);
- unit-dipole error \(2.220\times10^{-16}\);
- positive normalized cycle dissipation \(0.0294141\ \mathrm{ps}^{-1}\).

The 3D panels show the two microscopic folds, and the oscilloscope panel shows
their exactly overlapping polarization traces beside the applied field. Thus
the visualization exposes rather than hides the inverse problem: even a
noise-free waveform does not select one of these molecular realizations.

A second counterexample retains the fixed-tick boundary. If samples occur at
\(t_n=n\Delta t\), then sinusoids at \(f\) and \(f+k/\Delta t\) agree at every
sample for integer \(k\). The executed floating discrepancy is
\(8.399\times10^{-14}\), roundoff around the exact identity. A continuous
bandwidth restriction or another measurement is needed to remove these
aliases.

### AW.4. Evidence type, limitations, and next test

All 52 package tests pass. An independent implementation imports no production
module and checks the analytic response, charge arithmetic, rigid geometry,
92,160 stored atomic-coordinate scalars, equal folded traces, distinct
microstates, positive loss, and the alias control.

The following distinctions are mandatory:

1. Equations AW.1--AW.5 are ordinary model mathematics with published inputs.
2. Equations AW.6--AW.8 are written linear algebra and a constructed
   non-identifiability proof.
3. The 96-frame run is finite synthetic computation.
4. No new physical measurement was performed.
5. The microscopic orientation histories, finite scaffold, display scale, and
   camera are constructed and are not inferred from the dielectric trace.

Consequently the release is not liquid-water molecular dynamics and does not
establish a four- or five-dimensional generator, a universal spatial grid, a
universal physical tick, magnetism, quantum electrodynamics, or a theological
interpretation. A proposed higher-dimensional law must specify its forward
coupling before seeing the test result and predict a fresh observable better
than this ordinary baseline. The next bounded experiment is a frozen
multi-frequency complex-permittivity curve tested on held-out measurements.
If polarization leaves microscopic models equivalent, a structural observable
such as a scattering-derived correlation must be added rather than assigning
the unresolved degrees of freedom to a preferred ontology.

Revision 1.14's complete deterministic closure remains idempotent. Revision
1.15's integer-tick non-crossing limitation, angular-impulse formula, and
finite-prefix generator non-identifiability remain unchanged. The recurrence
learner's \(2K\)-sample continuation guarantee still assumes a noiseless
homogeneous constant-coefficient linear recurrence of order at most \(K\) from
the initial state.

## Appendix AX. Sunlight, moving water, and a finite visual observer

### AX.1. Bounded question and ordinary optical baseline

The bounded question is whether a fixed source, a moving interface, and a
finite visual observer can produce the requested underwater light-bending
image without treating a rendered movie as a unique reconstruction of reality.
Matter Harmony Lab 0.6.0 fixes the source direction and uses a nominal 5772 K
Planck spectrum. The source is distant enough that its incident rays are
parallel over the four-metre displayed surface.

The constructed water height is a three-component finite-depth linear wave,

\[
h(x,y,t)=\sum_{j=1}^{3} A_j
\sin(\mathbf k_j\!\cdot\!\mathbf r-\omega_jt+\phi_j),\qquad
\omega_j^2=g|\mathbf k_j|\tanh(|\mathbf k_j|H).                       \tag{AX.1}
\]

The amplitudes, wavelengths, directions, phases, and depth are frozen in the
release protocol. They are not recovered from a measured water surface. The
analytic normal is

\[
\hat{\mathbf n}=\frac{(-\partial_xh,-\partial_yh,1)}
{\sqrt{1+(\partial_xh)^2+(\partial_yh)^2}}.                            \tag{AX.2}
\]

For every spectral bin, the IAPWS R9-97 formulation supplies
\(n_w(\lambda,T,\rho)\). The executable then applies vector Snell refraction
and unpolarized Fresnel reflection. In scalar form,

\[
n_a\sin\theta_i=n_w(\lambda)\sin\theta_t,
\qquad R_F=\tfrac12(R_s+R_p),\qquad T_F=1-R_F.                         \tag{AX.3}
\]

This partitions interface energy in the no-absorption baseline. Intersecting
each transmitted ray with a bottom plane produces an expected-energy density,
not a list of measured individual photon paths. Local ray concentration
generates the moving caustic.

At 298.15 K and 997.047 kg/m3 the executed refractive indices are 1.345110691
at 380 nm, 1.332858051 at 589.3 nm, and 1.328482654 at 780 nm. The
normal-incidence 589.3 nm Fresnel reflectance is 0.02035830. Across the moving
surface, mean reflectance is 0.02557769 and mean transmittance is 0.97442231.
These are calculations from frozen reference inputs, not predictions of a
higher-dimensional theory.

### AX.2. Spectral observation and the metamer counterexample

The eye surrogate receives the spectral caustic and applies a three-coordinate
standard-color fold,

\[
\mathbf y(x,y,t)=
\int_{380\,\mathrm{nm}}^{780\,\mathrm{nm}}
L(x,y,t,\lambda)
\begin{bmatrix}\bar x(\lambda)\\\bar y(\lambda)\\\bar z(\lambda)\end{bmatrix}
\,d\lambda.                                                          \tag{AX.4}
\]

The functions in Equation AX.4 are a declared analytic approximation to the
CIE 1931 two-degree standard observer. This is a population colorimetric
standard, not an individual retina, optic nerve, or visual cortex.

Equation AX.4 cannot be inverted on arbitrary spectra. On the 41-bin control
grid its matrix has rank three and sampled nullity 38. The executable
constructs two nonnegative spectra \(s_A\ne s_B\) in the same observation
fiber:

\[
C s_A=C s_B,
\qquad
\frac{\|s_A-s_B\|_2}{\|s_A\|_2}=0.709286,                             \tag{AX.5}
\]

while the maximum stored XYZ difference is
\(1.421\times10^{-14}\). Thus even a noise-free standard-color observation
does not contain every visible spectrum. A claim that a human-eye observer
can measure everything humans have observed is therefore false for this
declared map: metameric inputs already collide before adaptation, eye motion,
retinal sampling, or neural processing are added.

### AX.3. Why the animation is not a hologram

The optical carrier frequencies corresponding to 780 and 380 nm are about 384
and 789 THz. The executable does not step those oscillations. It analytically
propagates time-averaged spectral irradiance and averages three surface states
over a 1/120-second display exposure. The movie contains 150 output frames at
60 frames per second.

Smooth playback can reduce visible frame discontinuity under some viewing
conditions, but no universal flicker threshold is assumed. More importantly,
temporal fusion cannot restore information removed by the observer. A
hologram requires coherent complex-field propagation and wavefront
reconstruction. This ray engine discards optical phase. For example,

\[
E_A=e^{i\varphi},\qquad E_B=e^{i(\varphi+\pi/2)},
\qquad |E_A|^2=|E_B|^2=1.                                             \tag{AX.6}
\]

The executed intensity difference in Equation AX.6 is below
\(6.7\times10^{-16}\). The displayed caustic is therefore a radiance
projection with three-dimensional geometric cues, not a holographic
reconstruction and not evidence that the displayed motion originates in a
fourth or fifth spatial plane.

### AX.4. Executed evidence, limits, and next test

The 0.6.0 run executes 48,988,800 ray–wavelength–exposure samples and renders
9,331,200 RGB channel values. Median caustic luminance contrast is 1.326639.
All 65 package tests pass. A separately written audit imports no production
module and re-evaluates the IAPWS index, spectral observer, energy partition,
metamer collision, phase-loss control, fixed-tick alias, frame arrays, and
60 fps media metadata.

The evidence types remain distinct:

1. IAPWS refractive index, the CIE standard observer, and nominal solar
   temperature are reference inputs.
2. Snell/Fresnel relations and the metamer and phase collisions are written
   mathematics.
3. The ray bundle, caustic, and movie are finite synthetic computation.
4. No new water-surface or optical measurement was made.
5. A physical higher-dimensional generator remains a hypothesis requiring a
   specified forward law and a fresh prediction beyond this baseline.

The engine omits water absorption, volumetric scattering, polarization-
resolved transport, diffraction, coherence, eye accommodation, retinal cone
mosaics, eye movements, neural processing, and observer variation. A failure
against data would not uniquely identify a missing string, rotation, or
dimension; surface error and these ordinary omissions remain live alternatives.

The next bounded test is to freeze a measured water-surface height movie,
depth, source spectrum, and camera geometry, then score predicted caustic ridge
positions and chromatic offsets on held-out frames without refitting. A
separate coherent wave-optics branch must first pass analytic phase-grating
tests before any output is called holographic.

Revision 1.14's complete deterministic closure remains idempotent. Revision
1.15's integer-tick non-crossing limitation, angular-impulse formula, and
finite-prefix generator non-identifiability remain unchanged. The recurrence
learner's \(2K\)-sample continuation guarantee still assumes a noiseless
homogeneous constant-coefficient linear recurrence of order at most \(K\) from
the initial state. This appendix neither establishes seven physical
dimensions nor assumes \(P=NP\), and it assigns no residual to a theological
cause.


## Appendix AY. Water-drop acoustics and the observer-rate boundary

### AY.1. Bounded question and declared impact

The bounded question is whether a water-drop impact can be connected to a
surface ripple, an airborne pressure waveform, and synchronized media without
mistaking a plausible synthesis for a unique reconstruction of a real event.
Matter Harmony Lab 0.7.0 fixes a 4.0 mm water drop, 0.08224 m vacuum fall,
0.080 m pool depth, and a nominal room-temperature water/air baseline. The
calculated impact speed is

\[
v_i=\sqrt{2gh}=1.2700385\ {\rm m\,s^{-1}}.                         \tag{AY.1}
\]

The corresponding Weber and Bond numbers are 89.3836 and 2.17373. These
dimensionless numbers characterize the declared baseline; they do not by
themselves determine splash topology or whether a bubble is entrained. Those
outcomes require a nonlinear two-phase free-surface calculation or
measurement. The executable therefore declares one bubble after impact rather
than claiming to predict it.

The radial display packet has a 0.025 m carrier and uses the ordinary
finite-depth capillary--gravity dispersion relation,

\[
\omega^2(k)=
\left(gk+\frac{\sigma}{\rho_w}k^3\right)\tanh(kh).              \tag{AY.2}
\]

Its frequency is 9.56335 Hz, phase speed is 0.239084 m/s, and group speed is
0.195422 m/s. A Gaussian radial envelope and decaying central cavity make a
finite visualization. The carrier obeys Equation AY.2, but the envelope is a
declared narrow-band approximation rather than a solved impact flow.

### AY.2. Entrained-bubble pitch and airborne pressure

Phillips, Agarwal, and Jordan used synchronized high-speed video and acoustic
measurements to show that the characteristic dripping-tap sound is driven by
resonant oscillation of an entrained bubble. The release therefore uses the
ideal Minnaert relation for one declared spherical bubble,

\[
f_M=\frac{1}{2\pi R_b}
\sqrt{\frac{3\gamma P_b}{\rho_w}}.                               \tag{AY.3}
\]

For the frozen radius \(R_b=0.800\) mm and hydrostatic pressure at 4 mm depth,
Equation AY.3 gives 4110.9205 Hz. The stored 48 kHz pressure trace has a
4116.67 Hz peak in its finite 60 ms FFT window, 5.75 Hz from the analytic
carrier.

The above-water field reported in the 2018 experiment is not adequately
described as the underwater field simply crossing the interface. Release
0.7.0 instead uses a deliberately bounded surface-coupled compact-source
surrogate,

\[
p(r,t)=\frac{\rho_a}{4\pi r}
\frac{d^2V_{\rm eff}}{dt^2}\left(t-\frac{r}{c_a}\right),         \tag{AY.4}
\]

where \(V_{\rm eff}\) is a damped sinusoidal effective volume. The 3% volume
amplitude, 18 ms decay, 6 ms entrainment delay, and impact-pulse coefficient
are constructions. Equation AY.4 supplies a physically scaled candidate for
pitch, decay, retarded arrival, and inverse-distance spreading; it does not
supply a calibrated absolute sound level. Playback is peak-normalized.

At \(r=0.750\) m and \(c_a=346.13\) m/s, the calculated air transit is
2.16682 ms. The impact pulse arrives at 0.4021668 s and the declared bubble
tone at 0.4081668 s. The delivered MP4 therefore combines 180 picture frames
at 60 fps with 144,000 mono audio samples at 48 kHz. The visual stream cannot
resolve a 2.17 ms transit, whereas the audio stream can.

### AY.3. Why neither frame rate nor pitch is a unique inverse

A human eye is not a camera with one universal frame rate. Temporal visibility
varies with luminance, contrast, spatial structure, eye movement, retinal
location, and observer. Davis, Hsieh, and Lee reported artifacts above 500 Hz
for spatially structured modulation even though conventional uniform-field
measurements often fall near 50--90 Hz. Sixty fps is therefore only the output
media rate in this release. It is not asserted as a biological tick or a
guarantee of invisibility.

Acoustic pitch is also non-injective. From Equation AY.3,

\[
f_M(R,P)=f_M(2R,4P).                                                  \tag{AY.5}
\]

The executable checks this collision to machine precision. Consequently one
ideal tone cannot independently identify both radius and pressure. A stronger
ambiguity remains because the same synthetic surface packet can be paired with
different declared bubbles: the release does not solve the cavity dynamics
that would link the displayed ripple to a unique bubble distribution.

These are observation fibers in the same precise sense used elsewhere in this
paper. A lower-dimensional measurement may constrain a declared forward model
without uniquely determining its source state. Calling the unresolved part a
fifth-dimensional rotation would add an interpretation, not information.

### AY.4. Executed evidence, limits, and next test

All 79 package tests pass. A separately written audit imports no production
module and recomputes 1,658,880 surface-height scalars and 144,000 pressure
samples. It independently checks free fall, Weber and Bond numbers,
capillary--gravity phase and group speed, ideal Minnaert pitch, the pitch
collision, retarded inverse-distance pressure, exact PCM samples, and the
three-second 60 fps/48 kHz media metadata.

The evidence types remain distinct:

1. Water/air properties and the published bubble-sound mechanism are reference
   inputs and primary-research provenance.
2. Equations AY.1--AY.5 and the pitch collision are written conditional
   mathematics.
3. The ripple, pressure waveform, WAV, and movie are finite synthetic
   computation.
4. No drop, bubble, microphone, or room response was measured in this release.
5. A physical higher-dimensional generator remains a hypothesis and is not
   needed to produce the bounded sound.

The engine omits aerodynamic drag, nonlinear cavity formation, splash
droplets, multiple bubbles, viscous and thermal bubble corrections, walls,
reverberation, atmospheric attenuation, head-related transfer, and hearing.
Its constructed active-window SPL is not a measurement. Failure against real
data could arise from any of these ordinary omissions and would not uniquely
identify a missing string, rotation, or dimension.

The next test is empirical and precommitted: measure drop diameter and impact
speed; infer bubble radius from synchronized high-speed video; calibrate
microphone distance and room impulse response; then score held-out ripple-front
position, tonal peak, decay constant, arrival time, and absolute pressure
without refitting.

Revision 1.14's complete deterministic closure remains idempotent. Revision
1.15's integer-tick non-crossing limitation, angular-impulse formula, and
finite-prefix generator non-identifiability remain unchanged. The recurrence
learner's \(2K\)-sample continuation guarantee still assumes a noiseless
homogeneous constant-coefficient linear recurrence of order at most \(K\) from
the initial state. This appendix neither establishes seven physical dimensions
nor assumes \(P=NP\), and it assigns no residual to a theological cause.


## Appendix AZ. Bubble hidden-variable inverse and observation fibers

### AZ.1. Bounded question and correction of scope

The bounded question is whether the complete damped pressure trace of the
single-bubble model uniquely determines bubble radius. Revision 1.50 already
showed that one ideal Minnaert tone cannot identify radius and pressure
simultaneously. This appendix makes a more realistic ambiguity explicit by
adding ordinary thermodynamic and interfacial nuisance variables.

The revision-1.50 relation

\[
f_M=\frac{1}{2\pi R}\sqrt{\frac{3\gamma p_\infty}{\rho}}
\tag{AZ.1}
\]

is retained as an ideal formula, not retracted. For a small spherical bubble,
linearizing Rayleigh--Plesset dynamics while including vapor pressure and
surface tension gives the declared stiffness model

\[
\omega_0^2=
\frac{3\kappa(p_\infty-p_v)+2(3\kappa-1)\sigma/R}
{\rho R^2}.
\tag{AZ.2}
\]

Here \(\kappa\) is an effective polytropic exponent. Prosperetti's primary
analysis shows why it cannot silently be equated with one universal constant:
the effective exponent and thermal damping depend on frequency, radius, and
gas/liquid properties [Prosperetti (1977)](https://doi.org/10.1121/1.381252).
Equation AZ.2 is therefore used only for a bounded sensitivity study. It is
not presented as the full complex thermal model.

At 298.15 K the executable uses \(p_v=3169.824486\) Pa from the IAPWS reference
formulation, \(\rho=997.047\ {\rm kg\,m^{-3}}\),
\(\sigma=0.07197\ {\rm N\,m^{-1}}\), and atmospheric plus hydrostatic pressure
at 4 mm depth. For \(R=0.800\) mm, Equation AZ.2 gives

\[
f_0(\kappa=1)=3421.6939\ {\rm Hz},\qquad
f_0(\kappa=1.4)=4048.9557\ {\rm Hz}.
\tag{AZ.3}
\]

The latter lies 61.9648 Hz, or 1.5073%, below the ideal value 4110.9205 Hz.
This is a checked scope correction: 4110.9205 Hz is the ideal-Minnaert output,
not a unique physical prediction after ordinary corrections are admitted.

### AZ.2. Exact radius--thermal-response observation fiber

For fixed positive \(\kappa\), Equation AZ.2 is monotone on the declared
positive-radius branch and can be inverted numerically. Holding the previous
4110.9205 Hz carrier fixed while scanning

\[
1\le\kappa\le1.4
\tag{AZ.4}
\]

gives

\[
0.6659558\ {\rm mm}\le R\le0.7879498\ {\rm mm}.
\tag{AZ.5}
\]

The 18.3186% span relative to the smaller endpoint is not a statistical
confidence interval. It is the exact image of the declared sensitivity
interval through Equation AZ.2. The endpoints are retained rather than
collapsed into a false point estimate.

The pressure observer used in the bounded sound branch depends on the second
derivative of

\[
V(t)=V_s e^{-t/\tau}\sin(2\pi f_0t),\qquad
V_s=\frac{4\pi}{3}R^3\varepsilon,
\tag{AZ.6}
\]

where \(\varepsilon\) is the fractional volume amplitude. Consequently two
states have the same complete pressure trace whenever they share \(f_0\),
\(\tau\), arrival, propagation distance, and \(R^3\varepsilon\). Release
0.8.0 constructs 401 such states. Its endpoints are

\[
(\kappa,R,\varepsilon)
=(1.0,0.6659558\ {\rm mm},0.0496913)
\tag{AZ.7}
\]

and

\[
(\kappa,R,\varepsilon)
=(1.4,0.7879498\ {\rm mm},0.0300000).
\tag{AZ.8}
\]

Both use the same 18 ms decay, 6 ms local arrival, and 0.75 m observer
distance. Their generated 48 kHz traces agree to

\[
\max_t|p_1(t)-p_2(t)|=3.667\times10^{-16}\ {\rm Pa}
\tag{AZ.9}
\]

over 3,840 samples per endpoint. This is an exact constructive
non-identifiability result inside the declared equations, not a claim that
either endpoint is the true bubble.

### AZ.3. Rank statement and what measurement must add

Use log parameters

\[
\theta=(\log R,\log\kappa,\log\varepsilon,\log\tau)
\tag{AZ.10}
\]

and log observables

\[
y=(\log f_0,\log V_s,\log\tau).
\tag{AZ.11}
\]

The executed finite-difference Jacobian \(\partial y/\partial\theta\) is
3-by-4, has numerical rank three, and therefore has local nullity one. The
null direction is the differential form of the exact family in Equations
AZ.7--AZ.9. More precise audio does not remove a structural nullspace when
the observation operator itself folds the source state.

An additional independent observable is required. Synchronized high-speed
imaging of radius and depth would intersect the acoustic fiber. A calibrated
underwater hydrophone would constrain the near-field radiation, while an
airborne microphone would test the surface-transfer model. Temperature,
liquid properties, sensor positions, and boundary response must be frozen
before fitting. Oguz and Prosperetti showed that proximity to a free surface
changes bubble oscillation and radiation
[Oguz and Prosperetti (1990)](https://doi.org/10.1121/1.399560), so that effect
cannot be hidden inside an unconstrained amplitude. Their rain-noise model
also predicted bubbles somewhat larger than observed and consequently shifted
the spectral peak, a direct historical warning that agreement in sound level
does not validate source geometry
[Oguz and Prosperetti (1991)](https://doi.org/10.1017/S0022112091002768).

### AZ.4. Evidence ledger, limitations, and next test

The evidence types remain separate:

1. The IAPWS water properties and cited bubble-acoustics studies are reference
   inputs and primary-research provenance.
2. Equations AZ.2, AZ.6, the equivalence construction, and the rank ceiling are
   written conditional mathematics.
3. The 401-state family, 7,680 endpoint pressure samples, bisection inverses,
   and finite-difference Jacobian are finite synthetic computation.
4. No bubble radius, pressure waveform, effective \(\kappa\), or free-surface
   correction was measured here.
5. A hidden-dimensional source remains a hypothesis with no additional
   support from this ambiguity; an ordinary classical family already produces
   it exactly.
6. Theological interpretation is outside the observation map.

All 88 package tests pass. A separately written audit imports no production
module and reproduces all 401 corrected-frequency states, 7,680 endpoint
pressure samples, constant oscillating volume, the rank-3/4 local inverse, and
the waveform collision.

The model still omits the full frequency-dependent thermal law, viscous and
acoustic-radiation damping, free-surface proximity, nonspherical modes,
multiple bubbles, nonlinear entrainment, calibrated underwater-to-air
transfer, walls, and sensor response. These are specific candidate variables,
not a license to add arbitrary degrees of freedom after failure. Each addition
must reduce a predeclared residual on fresh data without destroying earlier
predictions.

The next bounded empirical test is to measure radius and depth from
synchronized high-speed video and record calibrated underwater and airborne
pressure for repeated impacts. Fit the thermal/damping branch on one subset,
then score held-out pitch, decay, phase, and amplitude without refitting. If
the observation fiber remains, add a measurement that cuts it; do not select a
preferred hidden state by interpretation alone.

Revision 1.14's complete deterministic closure remains idempotent. Revision
1.15's integer-tick non-crossing limitation, angular-impulse formula, and
finite-prefix generator non-identifiability remain unchanged. The recurrence
learner's \(2K\)-sample continuation guarantee still assumes a noiseless
homogeneous constant-coefficient linear recurrence of order at most \(K\) from
the initial state. This appendix neither establishes seven physical
dimensions nor assumes \(P=NP\), and it assigns no residual to God or another
theological cause.


## Appendix BA. Collective vibrational families and thermodynamic phase

### BA.1. The bounded question and an exact counterexample

The question is not whether vibration matters. It is whether a vector of
isolated-molecule vibration frequencies, by itself, determines whether matter
is solid, liquid, or gas. Release 0.9.0 freezes the NIST water fundamentals

\[
q_{\rm H_2O}=(1595,3657,3756)\ {\rm cm}^{-1}
\tag{BA.1}
\]

as a deliberately phase-blind molecular-identity descriptor
([NIST Chemistry WebBook](https://webbook.nist.gov/cgi/cbook.cgi?ID=C7732185&Mask=800)).
The same vector is paired with three externally specified H2O controls at
101325 Pa: 260 K solid, 298.15 K liquid, and 400 K vapor. These phase labels
come from ordinary thermodynamic state information, not from Equation BA.1;
the water equation-of-state provenance is the
[IAPWS-95 release](https://iapws.org/technical-guidance/release/IAPWS-95).

Let \(h(q)\) be any deterministic classifier using only this descriptor.
All three inputs to \(h\) are identical, so all three outputs are identical.
On the balanced three-control set, at most one label is correct. Therefore

\[
\max_h {\rm accuracy}(h)=\frac13,
\qquad
\min_h {\rm error}(h)=\frac23.
\tag{BA.2}
\]

This is a written proof for the stated information restriction, not a fitted
failure and not a claim that condensed-phase water retains unshifted gas-phase
normal modes. It proves that phase-blind molecular frequencies cannot be a
phase oracle. Intermolecular organization, temperature, pressure, composition,
and system size must enter the observation map.

### BA.2. A collective spectrum that can express both diffusion and vibration

Rahman's molecular-dynamics study of liquid argon calculated the
velocity-autocorrelation function and its broad frequency spectrum while also
checking structural and self-diffusion observables
([Rahman 1964](https://doi.org/10.1103/PhysRev.136.A405)). The two-phase
thermodynamics construction later decomposed a liquid density of states into
gas-like diffusive and solid-like vibrational parts
([Lin, Blanco, and Goddard 2003](https://doi.org/10.1063/1.1624057)); a revised
treatment makes the assumptions and fluidicity correction explicit
([Sun et al. 2017](https://doi.org/10.1063/1.5001798)). These are methods for
collective trajectories, not licenses to label a phase from one bond frequency.

The new executable uses the bounded surrogate

\[
F(\omega;f_g)=f_gF_g(\omega)+(1-f_g)F_s(\omega),
\qquad 0\le f_g\le1,
\tag{BA.3}
\]

where the frozen solid-like basis satisfies \(F_s(0)=0\) and the gas-like
basis satisfies \(F_g(0)>0\). Hence within this declared two-basis family,

\[
f_g=\frac{F(0)}{F_g(0)}.
\tag{BA.4}
\]

Equation BA.4 is exact conditional mathematics for the surrogate. It is not a
general estimator for arbitrary molecular data. Finite execution recovers all
101 fractions from zero to one with maximum absolute error
\(1.1102230246251565\times10^{-16}\). The integrated low-frequency mass rises
monotonically from \(8.9987\times10^{-6}\) to 0.4599443.

Three declared reference profiles use \(f_g=0\), 0.35, and 1 for solid-like,
liquid-like, and gas-like controls. Their pairwise squared Hellinger distances
are

| Pair | \(H^2\) |
|---|---:|
| solid-like / liquid-like | 0.3077968075 |
| solid-like / gas-like | 0.7512211852 |
| liquid-like / gas-like | 0.1832326308 |

The positive distances show that this particular collective descriptor can
separate the three frozen profiles. The full 101-state path simultaneously
supplies the counterexample to any stronger claim: it is continuous, so nature
does not have to form three universally separated vibration families. A phase
boundary requires thermodynamic stability and finite-condition conventions;
it cannot be read from a universal spectral threshold alone.

### BA.3. What was checked and what remains empirical

The executable retains two argon provenance checks. Rahman's published state
maps to reduced values \(T^*=0.78797997\) and \(\rho^*=0.81769990\) under the
declared Lennard-Jones constants. The representative liquid-argon state in the
revised two-phase treatment maps to 131.78 K and 1428.27 kg/m3 and supplies the
published \(f_g=0.35\) anchor. The plotted spectral line shapes, widths, and
collision scale are nevertheless declared synthetic choices. They were not
extracted from either publication.

The evidence types are therefore:

1. NIST frequencies, IAPWS thermodynamics, and the cited argon studies are
   reference inputs and research provenance.
2. Equations BA.2 and BA.4 are written conditional mathematics.
3. The 2,401-point spectra, 101-state continuum, distances, and figure are
   finite synthetic computations.
4. No condensed-phase trajectory or spectrum was measured in this release.
5. Hidden rotations or higher-dimensional particles remain hypotheses; these
   results neither require nor identify them.
6. Theological interpretation is outside the phase classifier.

All 98 package tests pass. A separately implemented audit reproduces the
2,401-frequency grid, 101 continuum states, all three phase-control labels,
all three spectral distances, and the exact inversion. The correction is
actionable: phase work must begin with a many-particle trajectory or measured
collective response, then test whether a frozen descriptor transfers across
conditions and chemical families.

The next test is to generate velocity-autocorrelation spectra with one
validated intermolecular potential across externally labeled solid, liquid,
and vapor states. Freeze all line-shape choices and any classifier on a
training set, then score held-out temperatures, pressures, densities, and
molecules without refitting. Diffusion, structure factor, and enthalpy should
be simultaneous controls. If the spectrum fails, first test the potential,
finite-size effects, sampling, and omitted ordinary interactions; failure does
not uniquely select an extra dimension.

Revision 1.14's complete deterministic closure remains idempotent. Revision
1.15's integer-tick non-crossing limitation, angular-impulse formula, and
finite-prefix generator non-identifiability remain unchanged. The recurrence
learner's \(2K\)-sample continuation guarantee still assumes a noiseless
homogeneous constant-coefficient linear recurrence of order at most \(K\) from
the initial state. This appendix establishes neither seven physical dimensions
nor \(P=NP\), and it assigns no unconstrained residual to God.


## Appendix BB. Five-dimensional harmonies and a finite three-dimensional observer

### BB.1. Bounded model and complete linear harmony theorem

The bounded question is: if a latent state has five real coordinates and
moves by one fixed norm-preserving linear law, what can a three-coordinate
observer see? Define

\[
\dot z(t)=\Omega z(t),\qquad \Omega^T=-\Omega,
\qquad y(t)=Pz(t),\quad P=[I_3\;0].
\tag{BB.1}
\]

This is a theoretical candidate family, not an assertion about physical
spacetime. Because \(\Omega\) is real and skew-symmetric, an orthogonal change
of basis gives

\[
Q^T\Omega Q=
\operatorname{diag}\!\left(J(\omega_1),J(\omega_2),0\right),
\quad
J(\omega)=
\begin{bmatrix}0&-\omega\\ \omega&0\end{bmatrix},
\tag{BB.2}
\]

for \(\omega_1,\omega_2\ge0\). Exponentiating each block proves

\[
\boxed{
y(t)=c+\sum_{j=1}^{2}
\left[a_j\cos(\omega_jt)+b_j\sin(\omega_jt)\right].
}
\tag{BB.3}
\]

Therefore a fixed five-coordinate skew-linear model produces one stationary
component and at most two elliptical 3D harmonies. Every projected coordinate
also satisfies

\[
y^{(5)}+(\omega_1^2+\omega_2^2)y^{(3)}
+\omega_1^2\omega_2^2y'=0.
\tag{BB.4}
\]

Equations BB.2--BB.4 are written mathematics. If a measured signal requires
three independent frequencies, drift, damping, harmonics, or mode coupling,
then this exact model class is incomplete. Adding arbitrary hidden variables
afterward would not rescue its original prediction.

### BB.2. A product taxonomy rather than one “harmony” label

The engine uses six independent classification axes:

1. **Spectrum:** static, single-tone, equal-frequency, commensurate within a
   declared rational search, or no small rational match.
2. **Instantaneous coupling:** rank zero, one, or two for the visible-hidden
   block \(B=\Omega_{1:3,4:5}\).
3. **Known-generator observability:** rank of
   \(\mathcal O=[P;P\Omega;\ldots;P\Omega^4]\).
4. **One-history order:** rank of a finite block-Hankel matrix built from one
   noiseless 3D trajectory.
5. **Projected geometry:** dimension of the centered 3D trajectory span.
6. **Observer response:** exposure zeros, sampling aliases, and gauge
   collisions.

The known-system rank criterion follows the observability framework introduced
by [Kalman (1960)](https://doi.org/10.1016/S1474-6670(17)70094-8). The history
rank is motivated by output-only modal realization, especially the
eigensystem realization method of
[Juang and Pappa (1985)](https://ntrs.nasa.gov/citations/19850022899). Release
1.0 computes only finite noiseless ranks; it does not claim a noise-robust
physical system identification.

| Reference class | coupling rank | known \(\mathcal O\) rank | one-history rank | dynamic 3D span |
|---|---:|---:|---:|---:|
| static | 0 | 3 | 1 | 0 |
| silent hidden plane | 0 | 3 | 3 | 2 |
| rank-one, hidden unmixed | 1 | 4 | 4 | 3 |
| rank-one, hidden mixing | 1 | 5 | 5 | 3 |
| rank-two generic | 2 | 5 | 5 | 3 |
| commensurate 3:2 | 2 | 5 | 5 | 3 |
| equal frequency | 2 | 5 | 3 | 2 |
| exact \(1:\sqrt2\) reference | 2 | 5 | 5 | 3 |

Two counterexamples are decisive. The two rank-one cases have different
observability because hidden-hidden mixing carries the second hidden coordinate
through the same visible gate. Coupling rank alone is not a complete harmony
class. Conversely, the equal-frequency system is fully observable when its
generator is known, yet one autonomous history has rank three: a stationary
term plus only one temporal sine/cosine pair. Known-system state observability
does not imply identification of an unknown generator or physical dimension.

### BB.3. What a finite observer does

An instantaneous observation already loses a two-dimensional kernel because
\(P\) has rank three. Dynamics may reveal some or all of those coordinates over
time if \(\Omega\) is known. A camera, eye, or detector also integrates over a
nonzero exposure. For a centered boxcar exposure of length \(\tau\), each
modal amplitude is multiplied by

\[
g_\tau(\omega)=\operatorname{sinc}(\omega\tau/2).
\tag{BB.5}
\]

Thus \(\tau=2\pi/\omega\) erases that mode exactly in the declared noiseless
model. For the exact \((1,\sqrt2)\) reference, \(\tau=2\pi\) lowers both
one-history and effective observability rank from five to three, while the
second mode retains amplitude 0.2169542944. For the commensurate 3:2 reference,
the common-period exposure \(\tau=4\pi\) erases both rotations and lowers both
ranks to one.

This is information loss, not physical collapse. The current observer has no
back-action term: it changes \(y\), not \(z\). A quantum measurement model would
require a state space, interaction, measurement operators, outcome rule, and
validated probabilities not supplied by Equation BB.1.

Fixed ticks add a different collision. In a sampled rotation plane,

\[
\omega' = \omega+\frac{2\pi k}{\Delta}
\tag{BB.6}
\]

gives the same integer-tick positions. This retains revision 1.15's warning:
integer samples do not uniquely identify continuous behavior.

### BB.4. Hidden-basis gauge and the inverse boundary

Let

\[
G_H=\operatorname{diag}(I_3,H),\qquad H\in O(2).
\tag{BB.7}
\]

Because \(PG_H=P\), the transformed pair

\[
\Omega'=G_H\Omega G_H^T,\qquad z_0'=G_Hz_0
\tag{BB.8}
\]

obeys

\[
Pe^{t\Omega'}z_0'=Pe^{t\Omega}z_0
\quad\text{for every }t.
\tag{BB.9}
\]

Release 1.0 executes 101 hidden-plane angles and 64 times per angle. The
largest Frobenius displacement between transformed and baseline generators is
4.0735906631, while all 6,464 projected states agree within
\(5.17\times10^{-14}\). The orientation of the hidden axes is therefore a
coordinate gauge under this observer. A unique physical interpretation cannot
be selected from those outputs.

### BB.5. Finite validation and the next falsification

Four frozen spectral families receive 128 independently rotated generators
each. With the matrix supplied, the classifier recovers all 512 bounded labels
and both canonical frequencies with maximum error
\(1.11\times10^{-15}\). “No small rational match” means only that no ratio with
denominator at most 16 meets the \(10^{-9}\) tolerance. It does not prove an
arbitrary measured ratio irrational. The \(\sqrt2\) reference is known to be
irrational because it was constructed symbolically that way, not because the
finite search discovered irrationality.

The evidence types are:

1. Equations BB.2--BB.9 are written conditional mathematics.
2. The eight archetypes, 512 matrix-supplied classifications, finite ranks,
   6,464 gauge-time states, and exposure tests are finite synthetic computation.
3. Kalman observability and ERA supply primary-method provenance.
4. No physical five-dimensional trajectory, molecular spectrum, or quantum
   measurement was observed.
5. A physical fifth dimension remains a hypothesis and is not identified by a
   compatible linear realization.
6. Theological interpretation remains outside this observation map.

All 114 package tests pass. A separate implementation performs 48 catalog
checks, 2,048 randomized checks, all 6,464 gauge-time comparisons, two exposure
erasures, and 81 alias comparisons without importing the production module.

The next test must remove the supplied-generator advantage. Freeze a
two-tone/noise inference protocol using only 3D trajectories, then score it on
held-out missing-frame, noisy, drifting, nonlinear, and deliberately
three-tone controls. It must return “outside model class” rather than assigning
every unexplained residual to a fifth-dimensional particle.

Revision 1.14's complete deterministic closure remains idempotent. Revision
1.15's integer-tick non-crossing limitation, angular-impulse formula, and
finite-prefix generator non-identifiability remain unchanged. The recurrence
learner's \(2K\)-sample continuation guarantee still assumes a noiseless
homogeneous constant-coefficient linear recurrence of order at most \(K\) from
the initial state. This appendix establishes neither seven physical dimensions
nor \(P=NP\), and it assigns no residual to God or another theological cause.

## Appendix BC — Hydrogen spectroscopy gate for the two-harmony observer (revision 1.54)

Appendix BB proves that a fixed real skew-symmetric five-coordinate generator
has at most two nonnegative rotation frequencies. Under a linear scalar or
vector observation, the positive-frequency support therefore has cardinality
at most

\[
N_+(d)=\lfloor d/2\rfloor,
\qquad N_+(5)=2.
\tag{BC.1}
\]

The NIST H I strong-line table resolves four visible Balmer features. This
revision freezes one representative strongest listed component for each:

| Feature | Transition | Air wavelength (nm) | Listed relative intensity |
|---|---:|---:|---:|
| H-delta | 6 to 2 | 410.17400 | 15 |
| H-gamma | 5 to 2 | 434.04620 | 30 |
| H-beta | 4 to 2 | 486.13615 | 60 |
| H-alpha | 3 to 2 | 656.28518 | 180 |

The table is reference evidence, not a new measurement. Because these are
four distinct resolved features, the current fixed linear five-coordinate
observer fails the necessary cardinality gate

\[
4\le N_+(5)=2.
\tag{BC.2}
\]

The conclusion does not depend on an optimizer. Two free frequencies can fit
any selected pair exactly and necessarily leave two features unexplained.
For the visualized restricted ratios, a one-scale calibration searches both
ways of anchoring either mode to every measured line. At a declared 1%
relative wavelength tolerance, the 3:2 family can match two features, while
the tested locked and \(1:\sqrt2\) families cover fewer. Those are finite
calibrated comparisons, not predictions or evidence for a 3:2 mechanism.

A centered exposure multiplies existing mode amplitudes by
\(\operatorname{sinc}(\omega\tau/2)\). It cannot create the missing spectral
lines. A model that attempts the full spectrum must therefore declare a new
mechanism, such as validated energy-state transitions, a nonlinear observation
with explicit combination frequencies, a time-varying/interacting generator,
or a larger linear state. Each alternative changes the model class and must
be frozen before new validation.

Equations BC.1--BC.2 are written linear algebra. The wavelength comparison is
a finite calculation using external reference data. No time-resolved 5D
motion was measured, and the result proves neither a physical fifth dimension
nor its absence. It specifically rejects the current two-frequency linear
observer as a complete generator of the four-feature Balmer spectrum.

## Appendix BD — Multidimensional vector observers (revision 1.55)

Let the declared latent state be (x(t)\in\mathbb R^d) with

\[
\dot x(t)=A x(t),\qquad A^T=-A.
\tag{BD.1}
\]

For each calibrated three-dimensional observer (j), the engine reports both
projected position and projected tangent velocity,

\[
y_j(t)=P_jx(t),\qquad \dot y_j(t)=P_jAx(t),
\quad P_j\in\mathbb R^{3\times d}.
\tag{BD.2}
\]

These equations make the proposed “string movement” visible as vectors without
claiming that the vectors are measured physical strings. The observer is
passive; no back-action term is present.

### BD.1. What one observer cannot see

Rank-nullity gives

\[
\dim\ker P_j=d-\operatorname{rank}(P_j)\ge d-3.
\tag{BD.3}
\]

Therefore any (h\in\ker P_j) generates an exact instantaneous collision,

\[
P_j(x+h)=P_jx.
\tag{BD.4}
\]

The executable constructs a five-dimensional pair separated by one latent
unit with zero projected separation under its first observer. A single 3D
view therefore cannot identify a generic 4D or 5D state.

### BD.2. When several observers recover one common state

For simultaneous calibrated observations, stack the projectors,

\[
M=\begin{bmatrix}P_1\\P_2\\\vdots\\P_k\end{bmatrix},
\qquad Y=Mx.
\tag{BD.5}
\]

The common state is uniquely identifiable exactly when

\[
\operatorname{rank}(M)=d.
\tag{BD.6}
\]

This is an if-and-only-if statement: full column rank supplies the left
pseudoinverse; deficient rank supplies a nonzero collision vector in
(ker M). The frozen coordinate banks have full rank and reconstruct the
synthetic states in dimensions 4, 5, 6, and 8 to numerical precision. The
result assumes known matrices, aligned coordinates, simultaneity, and noiseless
data. It does not license stacking uncalibrated or asynchronous measurements.

### BD.3. Observers do not manufacture harmonies

Every real skew-symmetric (d\times d) matrix is orthogonally equivalent to
at most (lfloor d/2\rfloor) planar rotation blocks and, for odd (d), at
least one zero direction. Consequently

\[
N_+(d)=\lfloor d/2\rfloor.
\tag{BD.7}
\]

Changing (P_j) can reveal, attenuate, erase, or mix those modes in the output,
but cannot change the eigenfrequencies of (A). The tested capacity table is:

| Latent dimension | Positive-mode capacity | Single-view nullity | Four-feature count gate |
|---:|---:|---:|---|
| 4 | 2 | 1 | fail |
| 5 | 2 | 2 | fail |
| 6 | 3 | 3 | fail |
| 7 | 3 | 4 | fail |
| 8 | 4 | 5 | pass count only |
| 9 | 4 | 6 | pass count only |

The passage at dimension eight is not a hydrogen prediction. Four free modes
can merely provide four adjustable line locations. A physical model must also
derive wavelengths, intensities, degeneracies, selection rules, and state
transitions with frozen parameters and held-out tests. Dimension must not be
selected after seeing the number of target features.

### BD.4. Evidence boundary

Equations BD.3, BD.6, and BD.7 are written finite-dimensional linear algebra.
The reconstruction errors, collision, trajectories, and capacity table are
finite synthetic computations. The NIST Balmer count remains external
reference evidence; no new spectroscopy was performed. No physical 4D, 5D,
8D, or 9D vector was observed, and no residual is assigned to a string
particle, God, or another theological cause.

Revision 1.14's complete deterministic closure remains idempotent. Revision
1.15's integer-tick non-crossing limitation, angular-impulse formula, and
finite-prefix generator non-identifiability remain unchanged. The recurrence
learner's (2K)-sample continuation guarantee still assumes a noiseless
homogeneous constant-coefficient linear recurrence of order at most (K)
from the initial state. Nothing here proves seven physical dimensions or
(P=NP).

## Appendix BE — Hydrogen transition holdout (revision 1.56)

The request to prove quantum physics or physical dimensions is replaced by a
falsifiable requirement: a proposed law must predict measurements that did not
set its parameters. Quantum mechanics already has extensive experimental
support; one retrospective spectral fit cannot prove the whole theory, and an
extra coordinate cannot be inferred merely because a flexible term reduces a
residual.

### BE.1. Frozen data split and conversion

The NIST H I strong-line table reports the visible features used here as air
wavelengths. Before fitting, release 1.3.0 solves

\[
\lambda_{\rm vac}=n_{\rm air}(\lambda_{\rm vac})\lambda_{\rm air}
\tag{BE.1}
\]

with the Peck--Reeder standard-dry-air relation at its stated reference
conditions. The converted wavenumbers for H-alpha, H-beta, and H-gamma agree
with independent differences in the NIST persistent-line energy table within
0.0011 cm^-1. This cross-check validates the conversion implementation; it is
not a new measurement.

The protocol was fixed as follows:

| Role | Features |
|---|---|
| calibration | H-alpha (3 to 2), H-beta (4 to 2) |
| untouched primary holdout | H-gamma (5 to 2), H-delta (6 to 2) |
| secondary extrapolation | NIST listed-intensity-weighted Lyman-alpha centroid |

The displayed-decimal half-step is retained as a reporting-precision
sensitivity check. It is not promoted to a complete spectroscopic uncertainty
model.

### BE.2. Two bounded laws

For upper and lower principal indices (n_u>n_l), define

\[
\Delta_k(n_u,n_l)=n_l^{-k}-n_u^{-k}.
\tag{BE.2}
\]

The one-parameter baseline and two-parameter candidate are

\[
M_0:\quad \widetilde\nu=R\Delta_2,
\qquad
M_1:\quad \widetilde\nu=R\Delta_2+C\Delta_4.
\tag{BE.3}
\]

(M_0) is a Rydberg-form baseline, fitted by least squares on the two calibration
lines. (M_1) uses its two parameters to interpolate those lines exactly. The
(C\Delta_4) term is deliberately labeled phenomenological. It has not been
derived from an extra-dimensional Hamiltonian, action, field equation,
coupling to charge or current, gauge symmetry, or conservation law. Similar
principal-index dependence can arise from ordinary precision corrections, so
even a successful fit would not uniquely identify dimensionality.

### BE.3. Held-out result

The primary score is vacuum-wavelength root-mean-square error in parts per
million on the two untouched Balmer features:

| Model | Calibration RMSE (ppm) | Holdout RMSE (ppm) | Worst holdout (ppm) |
|---|---:|---:|---:|
| (M_0) Rydberg form | 0.522008 | 6.030008 | 6.909563 |
| (M_1) plus (Delta_4) | approximately 0 | 6.948567 | 7.728623 |

The added term worsens the frozen score by 15.2331%. It is also worse on each
holdout feature at both endpoints generated by the source table's displayed
wavelength precision. On the separate Lyman-alpha extrapolation, (M_0) has
signed error (+7.7269) ppm and (M_1) has (-11.2483) ppm. Therefore (M_1) is
rejected as an improvement. The exact fitted coefficient and its failure are
preserved rather than reinterpreted as a hidden variable.

This comparison does not establish that the fitted (M_0) is complete quantum
electrodynamics. Appendix AQ already records omitted recoil, radiative,
finite-size, hyperfine, linewidth, and external-field contributions. The few
ppm residuals remain a budget for model and reference-feature mismatch, not a
unique signal from another dimension.

### BE.4. Exact finite-prefix counterexample

Let (x(n)=\Delta_2(n,2)) and define

\[
q(n)=[x(n)-x(3)][x(n)-x(4)].
\tag{BE.4}
\]

For every real (D), the family

\[
\widetilde\nu_D(n)=R\Delta_2+C\Delta_4+Dq(n)
\tag{BE.5}
\]

has exactly the same H-alpha and H-beta values because (q(3)=q(4)=0). Yet
(q(5)) and (q(6)) are nonzero, so its continuation changes with (D). The
executed coefficient (D=111.2553101) cm^-1 forces H-gamma exactly but leaves a
(-7.1389) ppm wavelength error at H-delta.

This is a written non-identifiability result, not merely an optimizer failure.
It retains revision 1.15's broader correction: a finite observed prefix does
not identify an arbitrary generator. More calibration values can constrain a
declared finite model class; they cannot select an unrestricted continuation.

### BE.5. Evidence boundary and next gate

1. Equations BE.2--BE.5 and the invisible-term construction are written
   mathematics.
2. The standard-air conversion, parameter estimates, eight Balmer predictions,
   rounding endpoints, and two Lyman extrapolations are finite computation.
3. The NIST wavelengths and energy levels are externally compiled measured
   reference evidence; no new spectrum was recorded.
4. The (Delta_4) candidate is a rejected phenomenological hypothesis, not a
   physical extra dimension.
5. Theological interpretation is not inside the observation model and no
   residual is identified with God.

All 143 package tests pass. A separate 50-digit Decimal implementation imports
no production transition functions and reproduces all eight Balmer prediction
errors, the model decision, and the training-invisible coefficient.

The next admissible dimensional test must derive one explicit operator from a
declared higher-dimensional dynamics, include the standard precision terms or
bound their contribution, freeze its coupling on one transition family, and
predict a different state-dependent pattern before reading those targets. A
candidate that merely adds adjustable powers of (1/n) does not meet that gate.

Revision 1.14's complete deterministic closure remains idempotent. Revision
1.15's integer-tick non-crossing limitation, angular-impulse formula, and
finite-prefix generator non-identifiability remain unchanged. The recurrence
learner's (2K)-sample continuation guarantee still assumes a noiseless
homogeneous constant-coefficient linear recurrence of order at most (K) from
the initial state. Nothing in this appendix proves seven physical dimensions,
(P=NP), or a universal grid and tick.

## Appendix BF — Hydrogen isotope correction ladder (revision 1.57)

Revision 1.56 rejected an arbitrary \(\Delta_4\) fit but left the next valid
question open: can a concrete finite-range operator explain an isotope-sensitive
residual after ordinary precision physics is included? This appendix answers
that bounded question for the direct deuterium-minus-hydrogen 1S--2S shift.

### BF.1. Frozen measured target and ordinary reference

Parthey et al. report the direct optical result

\[
\Delta f_{D-H}^{\rm obs}=670\,994\,334.606(15)\ {\rm kHz}.
\tag{BF.1}
\]

The updated pionless-EFT/QED compilation of Lensky, Hagelstein, and
Pascalutsa gives, with its very small contact term fixed to zero,

\[
\Delta f_{D-H}^{\rm std}=
671\,000\,534.811(924)
-1369.346\left(\frac{r_d}{\rm fm}\right)^2\ {\rm kHz}.
\tag{BF.2}
\]

The contact coefficient is not fitted here because that work reports its
effect on this observable as negligible at the present uncertainty. Inserting
their muonic-deuterium extraction \(r_d=2.12763(78)\) fm gives

\[
\Delta f_{D-H}^{\rm std}=670\,994\,336.042632\ {\rm kHz}.
\tag{BF.3}
\]

The signed error is \(+1.436632\) kHz. Combining the 0.924 kHz theory term,
the radius propagation, and the 0.015 kHz measurement uncertainty in
quadrature gives 4.637998 kHz, so the error is only \(+0.309753\) standard
deviations. This is not a perfectly blind comparison: the muonic-radius
extraction has a weak shared-theory dependence through a low-energy constant
informed by the isotope shift, which the source reports as negligible at this
precision. It is a cross-system consistency check rather than an
assumption-free proof.

For comparison, the nonrelativistic reduced-mass-only expression

\[
\Delta f_{\mu}=\frac34 cR_\infty
\left(\frac{\mu_D}{m_e}-\frac{\mu_H}{m_e}\right)
\tag{BF.4}
\]

evaluated with the frozen 2022 CODATA constants gives
\(671007030.506398\) kHz, an error of \(+12695.900398\) kHz. The modern
ordinary ladder reduces the absolute error by a factor of 8837.3. A residual
left by the deliberately incomplete rung is therefore not evidence for a
missing dimension.

### BF.2. Explicit finite-range proxy

The tested operator is

\[
\frac{\delta V(r)}h=g\frac{e^{-r/\lambda}}r,
\qquad \lambda>0,
\tag{BF.5}
\]

where \(g\) has units Hz m. This Yukawa-like form is an explicit radial
operator, but it is not derived from a fifth-dimensional action,
compactification, gauge symmetry, charge assignment, or conserved current.
It is a phenomenological proxy and may not be relabeled as a physical
dimension even if it fits.

For a hydrogenic \(nS\) state with reduced-mass Bohr radius \(a\), define

\[
Y_n(a,\lambda)=
\left\langle nS\left|\frac{e^{-r/\lambda}}r\right|nS\right\rangle .
\tag{BF.6}
\]

The executable expands \(L_{n-1}^{1}(2r/na)\) and evaluates BF.6 as a finite
sum of gamma-function moments; no spatial grid or numerical eigenstate fit is
used. The isotope sensitivity of the 1S--\(nS\) transition is

\[
S_n(\lambda)=
[Y_n(a_D,\lambda)-Y_1(a_D,\lambda)]
-[Y_n(a_H,\lambda)-Y_1(a_H,\lambda)].
\tag{BF.7}
\]

### BF.3. Exact range/coupling non-identifiability

For every positive range at which \(S_2(\lambda)\ne0\), choosing

\[
g(\lambda)=
\frac{\Delta f_{D-H}^{\rm obs}-\Delta f_{D-H}^{\rm std}}
{S_2(\lambda)}
\tag{BF.8}
\]

fits the same central residual exactly. Equation BF.8 constructs a continuum
of collisions: one scalar datum supplies one constraint on two continuous
parameters.

The finite scan checks 61 logarithmic ranges from 1 fm through 10 nm. Every
pair reproduces the \(-1436.632276\) Hz central residual to machine precision.
Yet the 1 fm and 10 nm pairs give prospective 1S--3S isotope corrections of
\(-1581.055626\) and \(-1702.839834\) Hz, respectively, and their
unrecalibrated absolute hydrogen 1S--2S shifts differ by 3.519441 MHz. Thus
the calibration datum cannot identify the interaction range, coupling, or
later state dependence.

### BF.4. Decision, limitations, and next test

The frozen rule required a standard-model residual beyond three combined
standard deviations before an extra operator could be considered necessary.
The observed 0.309753-standard-deviation discrepancy fails that gate. The
operator is not accepted, and its exact-fit family is retained as a negative
result rather than tuned into an ontological claim.

Equations BF.5--BF.8 and the collision family are written mathematics. The
correction ladder, uncertainty propagation, analytic moments, and 61-point
scan are finite computation. Equations BF.1--BF.3 use published measured and
theory inputs; no new spectroscopy was performed. The 1S--3S values are
prospective discriminants, not measured validation. A future gate must freeze
two independently measured isotope-sensitive transitions and the full
ordinary-theory covariance, use one to set a coupling at a preregistered
range, and score the other without refitting.

Revision 1.14's complete deterministic closure remains idempotent. Revision
1.15's integer-tick non-crossing limitation, angular-impulse formula, and
finite-prefix generator non-identifiability remain unchanged. The recurrence
learner's \(2K\)-sample continuation guarantee still assumes a noiseless
homogeneous constant-coefficient linear recurrence of order at most \(K\)
from the initial state. Nothing in this appendix proves seven physical
dimensions, \(P=NP\), a universal physical grid/tick, a string particle, or a
theological cause.

## Appendix BG — Frozen hydrogen-isotope holdout (revision 1.58)

Revision 1.57 left a continuum of finite-range central operators that all fit
one H--D 1S--2S central residual. This appendix freezes one member and asks a
strictly smaller question: does it improve a different isotope-sensitive
transition without refitting?

### BG.1. Frozen operator and historical target

The range is fixed at the electron reduced Compton wavelength,

\[
\lambda=3.8615926796\times10^{-13}\ {\rm m},
\tag{BG.1}
\]

before scoring the target. This is a named atomic scale, not a value derived
from a fifth-dimensional action. The coupling is set only by BF.8 and remains
fixed. For isotope sensitivity \(S_n(\lambda)\) defined by BF.7, the predicted
H--D 2S--8S correction is

\[
\delta f_{2S-8S}=g\,[S_8(\lambda)-S_2(\lambda)]
=-201.996425\ {\rm Hz}.
\tag{BG.2}
\]

The separately published transition frequencies are

\[
f_H=770649350012.0(8.6)\ {\rm kHz},\qquad
f_D=770859041245.7(6.9)\ {\rm kHz},
\tag{BG.3}
\]

so their printed central difference is
\(209691233.7\) kHz. CODATA 2010 reports experimental correlation
\(\rho_{HD}=0.1225\). Consequently,

\[
\sigma_{\rm exp}
=\sqrt{6.9^2+8.6^2-2(0.1225)(6.9)(8.6)}
=10.345613\ {\rm kHz}.
\tag{BG.4}
\]

These values have been public since 1997. The comparison is a historical
holdout, not a blind prospective experiment.

### BG.2. Ordinary reference and correlated uncertainty

For an \(nS\) level, the published parameterization has the form

\[
\frac{E_{nS}}h=-\frac{cR_\infty}{n^2}\frac{m_r}{m_e}
+\frac{E_{\rm NS}}{n^3}+\Delta(nS).
\tag{BG.5}
\]

Using the tabulated H/D finite-size coefficients and additive remainders gives
the following D-minus-H 2S--8S ladder:

\[
209689697.033249
-736.037122
+2265.440000
=209691226.436128\ {\rm kHz}.
\tag{BG.6}
\]

The first term is reduced mass, the second finite nuclear size, and the third
the difference of the four additive theory remainders. The CODATA covariance
is propagated with coefficient vector \((1,-1,-1,1)\) in the order H2S, H8S,
D2S, D8S. Its standard uncertainty is 0.068444 kHz. External radius
uncertainties contribute 0.672174 kHz. Combining these with BG.4 gives

\[
\sigma_{\rm combined}=10.367652\ {\rm kHz}.
\tag{BG.7}
\]

Uncertainties of the frozen Rydberg and mass-ratio inputs contribute less than
0.01 kHz and are bounded rather than displayed in BG.7. The calculation is a
published precision approximation, not a new complete bound-state-QED
calculation or an independent remeasurement.

### BG.3. Decision and resolution limit

The ordinary prediction has signed error

\[
z_{\rm ordinary}=-0.700628.
\tag{BG.8}
\]

Adding BG.2 produces \(209691226.234131\) kHz and

\[
z_{\rm operator}=-0.720112.
\tag{BG.9}
\]

The frozen interaction therefore makes the central agreement slightly worse
and earns no predictive credit under the predeclared improvement rule. It is
not excluded: its full effect is only 0.019483 combined standard deviations,
and the uncertainty is 51.326 times larger. Non-rejection is ambiguity, not
confirmation.

Equations BG.1--BG.9 are written conditional mathematics. Their numerical
evaluation is finite computation. BG.3 uses published measured inputs; no new
spectroscopy was performed. An independent executable imports neither isotope
module, evaluates the radial integrals by direct quadrature, and reproduces
the correction and covariance.

The next gate must first rank unscored transitions by the absolute frozen
operator contrast divided by full correlated uncertainty. A target should be
frozen and scored only if it has material power to distinguish the models.
Failure, success, or an unresolved residual cannot by itself identify a
physical dimension, universal grid/tick, string particle, or theological
cause. The v1.14 closure correction, all v1.15 corrections, and the recurrence
learner's noiseless homogeneous constant-coefficient order-at-most-K
initial-state assumption remain unchanged.

## Appendix BH — Conserved quantum flow and observable harmonics (revision 1.59)

The hydrogen animation already used Coulomb Schrödinger eigenstates and their
relative phase. It did not yet expose the conserved spatial current or ask
which interference harmonics a declared measurement can actually see. This
appendix supplies that missing bridge without adding an extra coordinate.

### BH.1. Electromagnetic evolution and continuity

For a particle of charge \(q\), scalar potential \(\phi\), and vector potential
\(\mathbf A\), minimal coupling gives

\[
i\hbar\frac{\partial\Psi}{\partial t}=
\left[\frac{(-i\hbar\nabla-q\mathbf A)^2}{2m}+q\phi\right]\Psi.
\tag{BH.1}
\]

The conserved position probability obeys

\[
\frac{\partial\rho}{\partial t}+\nabla\!\cdot\mathbf j=0,
\qquad
\mathbf j=\frac{\hbar}{2mi}
(\Psi^*\nabla\Psi-\Psi\nabla\Psi^*)-
\frac{q}{m}\mathbf A|\Psi|^2.
\tag{BH.2}
\]

The previous fixed-proton engine already contained the electromagnetic
Coulomb scalar potential. Release 0.6 implements the field-free
\(\mathbf A=0\) specialization of BH.2; it does not yet simulate a driven
external field.

For the equal coherent packet

\[
\Psi(\mathbf r,t)=\frac{1}{\sqrt2}
\left[\phi_{1s}(\mathbf r)e^{-iE_1t/\hbar}
+\phi_{2p_z}(\mathbf r)e^{-iE_2t/\hbar}\right]
\tag{BH.3}
\]

with real spatial eigenfunctions and
\(\theta=(E_2-E_1)t/\hbar\), the current in atomic units is

\[
\mathbf j=\frac{\sin\theta}{2}
(\phi_{2p_z}\nabla\phi_{1s}-\phi_{1s}\nabla\phi_{2p_z}).
\tag{BH.4}
\]

Taking the divergence and substituting the two Coulomb eigenvalue equations
gives

\[
\nabla\!\cdot\mathbf j=(E_2-E_1)\phi_{1s}\phi_{2p_z}\sin\theta
=-\frac{\partial\rho}{\partial t},
\tag{BH.5}
\]

so the continuity equation holds identically. Equation BH.5 is a written
mathematical result. The arrows rendered by the executable are probability
flow; they are not trajectories of a localized electron.

### BH.2. Dipole harmonic and ideal rate

Light does not observe the density alone. In the electric-dipole
approximation the relevant observable is

\[
\mathbf d(t)=-e\langle\Psi(t)|\mathbf r|\Psi(t)\rangle.
\tag{BH.6}
\]

Direct integration gives

\[
\langle1s|z|2p_z\rangle
=\frac{256}{243\sqrt2}a_0
=0.744935539028a_0.
\tag{BH.7}
\]

Consequently the equal packet has an oscillating \(z\) dipole of amplitude
\(6.3158269\times10^{-30}\) C m, or \(1.8934378\) D. The ideal fixed-proton
electric-dipole formula gives

\[
A_{2p_z\rightarrow1s}
=\frac{\omega^3|e\langle1s|z|2p_z\rangle|^2}
{3\pi\epsilon_0\hbar c^3}
=6.26831503\times10^8\ {\rm s}^{-1},
\tag{BH.8}
\]

corresponding to \(1.59532505\) ns. This is an ideal nonrelativistic E1
calculation, not a precision bound-state-QED lifetime or a new measurement.

The production \(81^3\) grid reproduces BH.7 as \(0.744932784051a_0\), a
relative error \(3.70\times10^{-6}\). Its volume-integrated current matches
the derivative of the dipole with relative error \(5.35\times10^{-4}\). A
separate implementation imports none of the production module, evaluates
BH.7 by adaptive radial quadrature, and differentiates the current on a
\(129^3\) Cartesian grid. Outside the declared \(r<0.5a_0\) nuclear-cusp
region its fourth-order continuity residual is \(0.0128041\) in relative
\(L^2\). The full-interior value \(0.116031\) is retained because a Cartesian
stencil converges slowly across the Coulomb cusp; it is not relabeled as a
successful uniform numerical derivative.

### BH.3. Selection rules filter raw harmonics

An equal `1s + 2p_z + 3s` packet contains three pairwise density frequencies:

| Pair | Gap (Hartree) | Ideal wavelength (nm) | Density | \(z\)-dipole |
|---|---:|---:|---|---|
| \(2p_z\leftrightarrow3s\) | \(5/72\) | 656.112276 | present | present |
| \(1s\leftrightarrow2p_z\) | \(3/8\) | 121.502273 | present | present |
| \(1s\leftrightarrow3s\) | \(4/9\) | 102.517543 | present | absent |

The last matrix element vanishes because both \(s\) states have even parity.
The separate `1s + 2s` control likewise has a changing local density but
\(\langle1s|z|2s\rangle=0\). Thus a list of energy differences overpredicts
the optical signal unless it is filtered through the measurement operator.
The standard E1 angular rule is \(\Delta l=\pm1\), with
\(\Delta m=0\) for \(z\) polarization. Other multipoles, external-field
mixing, and two-photon processes are different declared measurement models;
E1 silence does not mean that a state or density beat is nonexistent.

This operator gate is also a correction to the framework's use of the word
“harmony.” A harmonic belongs jointly to a prepared state, Hamiltonian, and
observable. It is not an ontological object inferred from visual rhythm
alone.

### BH.4. Evidence classes, decision, and next gate

Equations BH.1--BH.8 and the parity cancellation are written standard quantum
mechanics. The voxel fields, matrix-element checks, ideal rate, and 42 passing
tests are finite deterministic computation. The independent audit passes all
seven frozen checks. No new measured hydrogen data were collected, and the
rendered slices are model output rather than a camera image of an electron.
Attosecond wave-packet experiments demonstrate that prepared electronic motion
can be probed, but they do not validate this particular computed hydrogen
packet or any extra-dimensional generator.

The bounded decision is therefore: Schrödinger evolution was not forgotten;
the explicit continuity/current equation and observable-selection gate were.
Adding them yields two dipole-visible harmonics from three density harmonics in
the declared three-state control. It supplies no evidence for a fourth or
fifth spatial dimension, a universal grid/tick, a hidden localized orbit, a
string particle, or a theological cause.

The next gate is a frozen weak-pulse time-dependent Schrödinger calculation
that implements BH.1 in both length and velocity gauges. It must require
gauge agreement, predict the allowed one-photon `1s -> 2p` response and the
suppressed `1s -> 2s` control before consulting compiled transition
probabilities or lifetimes.

Method context: NIST relates spectral-line strengths to the appropriate
multipole transition matrix element
(https://physics.nist.gov/Pubs/AtSpec/node17.html). Kübel et al. report a
primary experiment imaging prepared ultrafast valence-electron wave-packet
motion (https://arxiv.org/abs/1903.03739), while Rossi et al. give a 2026
primary method for polarization-controlled attosecond preparation and
difference-density analysis (https://arxiv.org/abs/2606.22102). These sources
motivate the observable gate; they do not validate this engine's hydrogen
packet or any extra-dimensional interpretation.

Revision 1.14's complete deterministic closure remains idempotent. Revision
1.15's integer-tick non-crossing limitation, angular-impulse formula, and
finite-prefix generator non-identifiability remain unchanged. The recurrence
learner's \(2K\)-sample continuation guarantee still assumes a noiseless
homogeneous constant-coefficient linear recurrence of order at most \(K\)
from the initial state.

## Appendix BI — Golden-ratio exactness and coincidence audit (revision 1.60)

The phrase “golden alignment” was previously undefined in the framework. This
appendix gives it an exact algebraic gate, audits the current hydrogen and
fixed-grid models, and separates derivation from pattern matching. It does not
insert a golden parameter into the dynamics.

### BI.1. Canonical golden quantities

Let

\[
\varphi=\frac{1+\sqrt5}{2}.
\tag{BI.1}
\]

Then

\[
\varphi^2=\varphi+1,\qquad
\varphi^{-1}=\varphi-1,\qquad
1-\varphi=-\varphi^{-1}.
\tag{BI.2}
\]

Every integer power is exhausted by

\[
\varphi^n=F_n\varphi+F_{n-1},
\tag{BI.3}
\]

where \(F_n\) is the Fibonacci sequence. The golden-turn fraction and the
usual smaller golden angle are

\[
\alpha_g=\varphi^{-2}=2-\varphi=\frac{3-\sqrt5}{2},
\qquad
\theta_g=2\pi\alpha_g=137.507764050\ldots{}^\circ.
\tag{BI.4}
\]

The complementary directed rotation is
\(2\pi-\theta_g=222.492235950\ldots{}^\circ\). These are the same unoriented
placement rule traversed in opposite directions, not two new constants.

For a positive dimensionless candidate \(x\), exact golden-ratio alignment
means

\[
x^2-x-1=0,
\tag{BI.5}
\]

which selects \(x=\varphi\). A small decimal residual in BI.5 is only an
approximation until the declared generator implies the polynomial. A useful
reported diagnostic is

\[
\epsilon_\varphi(x)=\frac{|x^2-x-1|}{1+x^2},
\tag{BI.6}
\]

but no threshold in finite precision can prove irrational equality.

### BI.2. Where \(\varphi\) is exact

The Fibonacci companion matrix

\[
Q=\begin{pmatrix}1&1\\1&0\end{pmatrix}
\tag{BI.7}
\]

has characteristic polynomial \(\lambda^2-\lambda-1\), eigenvalues
\(\varphi\) and \(-1/\varphi\), and

\[
Q^n=\begin{pmatrix}F_{n+1}&F_n\\F_n&F_{n-1}\end{pmatrix}.
\tag{BI.8}
\]

This is an intrinsic algebraic manifestation. It is relevant to a declared
Fibonacci recurrence in an integer or real recurrence engine. It does not
silently extend the v1.15 continuation guarantee: that guarantee remains
conditional on a noiseless homogeneous constant-coefficient linear recurrence
of order at most \(K\), from the initial state, observed for at least \(2K\)
samples.

Standard icosahedral coordinates are the permutations of
\((0,\pm1,\pm\varphi)\). Their edge and next-separation ratio is exactly
\(\varphi\). This is exact geometry after choosing an icosahedron; it is not
evidence that an unrelated atom selected those coordinates. Likewise, a
golden-angle point cloud is construction-imposed when Equation BI.4 is passed
to the sampler.

The other common names close under the same algebra rather than introducing
new constants. A golden rectangle has side ratio \(\varphi\) by definition. A
regular pentagon has diagonal-to-side ratio
\(2\cos(\pi/5)=\varphi\), derived from its symmetry. A logarithmic spiral
\(r(\theta)=ae^{b\theta}\) is a golden spiral only after choosing

\[
b=\frac{2\log\varphi}{\pi},
\]

which makes every quarter turn multiply the radius by \(\varphi\). Thus the
pentagon relation is derived from declared geometry, whereas the rectangle and
spiral labels impose the golden scale in their definitions.

In the integrable \(E_8\) spectrum, the first two masses obey

\[
\frac{m_2}{m_1}=2\cos\frac{\pi}{5}=\varphi.
\tag{BI.9}
\]

Coldea et al. reported neutron-scattering evidence consistent with that
prediction in the near-critical Ising-chain material
\(\mathrm{CoNb_2O_6}\). This is a measured quantum manifestation tied to an
emergent \(E_8\) model, not a universal feature of all quantum systems.

### BI.3. Exact exclusion from the ideal hydrogen gap ratios

The fixed-proton nonrelativistic Coulomb energies are

\[
E_n=-\frac{1}{2n^2}\ {\rm Hartree}.
\tag{BI.10}
\]

For integer \(n\), every \(E_n\) is rational in Hartree. Therefore every
nonzero gap \(E_b-E_a\) is rational, and every ratio of two nonzero gaps is
rational. Because \(\varphi\) is irrational,

\[
\boxed{\frac{E_b-E_a}{E_d-E_c}\ne\varphi}
\tag{BI.11}
\]

for all ideal Coulomb bound-state transitions. BI.11 is an all-level written
proof. It is stronger than any finite numerical search and does not apply
unchanged to relativistic, radiative, external-field, molecular, nuclear, or
many-body spectra.

The frozen finite control enumerates all 780 transitions with
\(1\le n<m\le40\) and all 303,809 distinct gap-pair comparisons. Its closest
ratio is

\[
\frac{\Delta E_{19\to30}}{\Delta E_{23\to34}}
=\frac{7491169}{4629825}
=1.61802422337777\ldots,
\tag{BI.12}
\]

whose absolute error from \(\varphi\) is
\(9.7653721\times10^{-6}\). This close rational is preserved as the required
counterexample to visual/numerical overinterpretation: 303,809 comparisons can
produce a compelling-looking match without an exact golden law.

### BI.4. Fixed grids cannot contain the exact golden turn

One angular increment on a finite \(N\)-state grid is \(k/N\) turns and is
rational. Equation BI.4 is irrational. Hence no integers \(k,N>0\) satisfy

\[
\frac{k}{N}=\frac1{\varphi^2}.
\tag{BI.13}
\]

The exact golden rotation has no finite recurrence period. Fibonacci
convergents provide controlled periodic approximations; for example,

\[
\frac{987}{2584}
\tag{BI.14}
\]

misses the golden turn by \(6.69777\times10^{-8}\), or
\(2.41120\times10^{-5}\) degrees per step. It nevertheless repeats after a
finite rational period and is not the exact irrational rotation.

This yields a correction for the fixed-grid/fixed-tick hypothesis. An engine
may use a finite angular grid, or it may carry an exact algebraic irrational
state such as \(a+b\sqrt5\), but a finite \(N\)-state angular grid cannot do
both exactly. If continuous golden phase is fundamental, the state
representation and update rule must say how the algebraic number is stored and
how observation maps it to finite data.

### BI.5. Measured manifestations do not transfer automatically

Douady and Couder obtained Fibonacci phyllotactic order in a physical
laboratory experiment and numerical simulation, describing convergence toward
the golden mean through an iterative self-organizing process. Shechtman et al.
measured metallic matter with long-range icosahedral orientational order and
no ordinary translational lattice. These results establish that specific
dynamics and symmetries can select golden structures. They do not show that
every spiral, harmony, atom, molecule, or projection is golden.

The evidence classes for this revision are:

1. Equations BI.1--BI.11 and BI.13 are written mathematics.
2. The 303,809 comparisons, Fibonacci convergence table, icosahedral distance
   audit, and 144-point golden-angle prefix are finite computation.
3. A separate executable imports none of the production module and reproduces
   the exhaustive search plus eight other checks.
4. The cited \(E_8\), quasicrystal, and phyllotaxis papers contain external
   measured evidence for their own systems. No new measurement was performed.
5. A hidden-dimensional golden generator remains a hypothesis. It has no
   declared Hamiltonian, coupling, units, observable, or fresh prediction here.
6. No golden residual identifies God or another theological cause.

Primary sources: Douady and Couder, *Physical Review Letters* 68, 2098 (1992),
https://doi.org/10.1103/PhysRevLett.68.2098; Shechtman, Blech, Gratias, and
Cahn, *Physical Review Letters* 53, 1951 (1984),
https://doi.org/10.1103/PhysRevLett.53.1951; Coldea et al., *Science* 327, 177
(2010), https://doi.org/10.1126/science.1180085.

All 51 Hydrogen Matrix package tests pass. The independent golden-ratio audit
passes nine checks. Earlier hydrogen, rotor, temperature, spectroscopy, and
probability-current results are unchanged.

The next admissible test must derive one dimensionless golden prediction from
a declared Hamiltonian or projection operator before consulting the target
data. The observable, uncertainty model, non-golden baseline, and rejection
threshold must be frozen. Searching many ratios and reporting the closest one
is not a prediction.

Revision 1.14's complete deterministic closure remains idempotent. Revision
1.15's integer-tick non-crossing limitation, angular-impulse formula, and
finite-prefix generator non-identifiability remain unchanged. This appendix
does not establish seven physical dimensions, \(P=NP\), or a theological
identification.

## Appendix BJ. Forward observable compiler and the quantum light boundary

### BJ.1. From “display everything” to a typed forward map

No finite engine can establish that it calculates every property of reality by
declaring unimplemented effects to be hidden. The executable replacement is a
typed contract

\[
X\xrightarrow{\;U_t\;}X(t)
\xrightarrow{\;\Pi\;}\psi(\mathbf r,t)
\xrightarrow{\;\mathcal O_k\;}y_k(t),
\tag{BJ.1}
\]

where the state \(X\), evolution \(U_t\), projection \(\Pi\), and every observer
operator \(\mathcal O_k\) are explicit. A requested channel is `generated` only
when all four objects exist. An approximation is marked `proxy`; a missing
operator or state variable is `blocked`; and a cause is `not identified` when
multiple hidden states share the same visible outputs.

Release 0.8 freezes the ideal fixed-proton, non-relativistic state

\[
|\Psi(t)\rangle=\frac{|1s\rangle+e^{-i\omega t}|2p_z\rangle}{\sqrt2},
\qquad
\omega=\frac{E_{2}-E_{1}}{\hbar}.
\tag{BJ.2}
\]

It compiles this one state into \(\rho=|\psi|^2\), the field-free probability
current \(\mathbf j\), electron charge density \(-e\rho\), charge current
\(-e\mathbf j\), the expectation dipole, a coherent far-zone radiation field,
and the earlier operator-filtered harmonic catalog. These are multiple
observation channels of one model; they are not additional ontological layers.

### BJ.2. Coherent electromagnetic projection

Let

\[
d=e|\langle1s|z|2p_z\rangle|,
\qquad
|\langle1s|z|2p_z\rangle|
=\frac{256}{243\sqrt2}a_0.
\tag{BJ.3}
\]

For the packet in Equation BJ.2,

\[
\mathbf p(t)=-d\cos(\omega t)\,\hat{\mathbf z}.
\tag{BJ.4}
\]

In the electric-dipole radiation zone, at retarded time,

\[
\mathbf E_{\rm rad}(\mathbf n,R,t)
=\frac{\mathbf n\times(\mathbf n\times\ddot{\mathbf p})}
{4\pi\epsilon_0c^2R},
\qquad
\mathbf B_{\rm rad}=\frac{\mathbf n\times\mathbf E_{\rm rad}}{c}.
\tag{BJ.5}
\]

The resulting power density is radial and nonnegative, with the dipole
\(\sin^2\theta\) pattern. Integration over solid angle gives

\[
P_{\rm coh}(t)=\frac{|\ddot{\mathbf p}(t)|^2}
{6\pi\epsilon_0c^3},
\qquad
\overline P_{\rm coh}=\frac{d^2\omega^4}
{12\pi\epsilon_0c^3}.
\tag{BJ.6}
\]

The finite engine stores 48 synchronized \(65\times65\) density/current and
charge/current slices plus a 721-direction radiation audit. At one metre the
declared packet gives peak equatorial \(|\mathbf E|=1.5179677846\times10^{-4}\)
V/m and cycle-average coherent power
\(2.5620237532\times10^{-10}\) W. These are deterministic model outputs, not
measurements of an individual atom.

### BJ.3. Exact counterexample to a complete classical projection

The ideal electric-dipole spontaneous rate is

\[
A=\frac{\omega^3d^2}{3\pi\epsilon_0\hbar c^3}.
\tag{BJ.7}
\]

Equation BJ.2 has excited-state population \(1/2\). Its initial quantum E1
energy-loss rate is therefore

\[
P_{\rm quantum}=\frac12\hbar\omega A
=\frac{d^2\omega^4}{6\pi\epsilon_0c^3}.
\tag{BJ.8}
\]

Equations BJ.6--BJ.8 imply the exact ratio

\[
\boxed{\frac{\overline P_{\rm coh}}{P_{\rm quantum}}=\frac12}.
\tag{BJ.9}
\]

The production calculation returns respectively
\(2.5620237532\times10^{-10}\) W and
\(5.1240475064\times10^{-10}\) W; the independent implementation also returns
exactly 0.5 at displayed precision. A pure excited state supplies an even
sharper counterexample: its expectation dipole is stationary and radiates no
classical coherent field, while Equation BJ.7 gives a nonzero spontaneous
decay rate.

Thus the visible classical wave from an expectation value is not the complete
photon observable. Vacuum fluctuations, incoherent emission, and photon
statistics require a quantized electromagnetic field or another explicitly
declared stochastic/quantum operator. The result agrees with the primary
mixed quantum--classical analysis of Chen et al., *J. Chem. Phys.* 150, 044102
(2019), https://doi.org/10.1063/1.5057365, which shows that ordinary Ehrenfest
fields miss essential spontaneous-emission physics.

### BJ.4. Verification, evidence type, and boundary

All eight frozen production checks pass: analytic continuity residual zero;
finite-grid neutrality error \(1.06955\times10^{-3}\); transversality relative
error \(1.33921\times10^{-16}\); plane-wave \(|B|=|E|/c\) error
\(4.44089\times10^{-16}\); angular power-integral error
\(4.53077\times10^{-12}\); cycle-average error
\(2.01789\times10^{-16}\); a zero dipole field for the parity-silent
\(1s+2s\) control; and Equation BJ.9. A separate implementation imports none
of the production compiler modules and passes twelve scalar and saved-array
checks.

The evidence classes are:

1. Equations BJ.1--BJ.9 are written conditional mathematics.
2. Stored fields, quadratures, tests, and the observer map are finite
   deterministic computation.
3. No new spectrum, photon count, or atomic movie was measured.
4. A deeper four- or five-dimensional generator remains a hypothesis because
   it still lacks an action, coupling, projection operator, and distinguishing
   prediction.
5. The classical/quantum difference does not identify one hidden cause and
   has no theological label.

The next bounded build is a two-level quantized-radiation channel with
density-matrix decay and photon-number statistics. It must freeze probability
conservation, total emitted energy, coherent/incoherent decomposition, and the
pure-excited-state counterexample before another visual layer is added.

Revision 1.14's complete deterministic closure remains idempotent. Revision
1.15's integer-tick non-crossing limitation, angular-impulse formula, and
finite-prefix generator non-identifiability remain unchanged. The recurrence
learner's \(2K\)-sample continuation guarantee still assumes a noiseless
homogeneous constant-coefficient linear recurrence of order at most \(K\) from
the initial state. Nothing here proves seven physical dimensions, \(P=NP\), a
universal grid/tick, or a theological identification.

## Appendix BK. All-element atomic projection atlas

### BK.1. The word “shape” needs an observer

An isolated atom is not assigned one rigid classical boundary. This revision
therefore defines two different observer channels rather than silently
equating them.

For any one-electron density \(\rho(\mathbf r)\), the unpolarized rotational
average is

\[
\overline{\rho}(\mathbf r)
=\int_{SO(3)}\rho(R^{-1}\mathbf r)\,d\mu(R)
=\overline{\rho}(|\mathbf r|),
\tag{BK.1}
\]

where \(d\mu\) is normalized Haar measure. Equation BK.1 is spherical by
construction and is the whole-atom display used by the atlas. It does not say
that every oriented open-shell atomic state has an exactly spherical density.

The second channel selects a declared spatial frame and displays one canonical
real differentiating-subshell basis. Its lobes are an orbital/state
visualization, not a hard surface or a localized electron trajectory. Rotating
the chosen basis rotates the picture without changing the free atom's
field-free physics.

### BK.2. Exact angular result

For a complete subshell, the spherical-harmonic addition theorem gives

\[
\sum_{m=-l}^{l}|Y_{lm}(\Omega)|^2=\frac{2l+1}{4\pi}.
\tag{BK.2}
\]

Thus a uniformly occupied closed subshell has no angular density structure.
For one oriented basis

\[
\psi_{nlm}(\mathbf r)=R_{nl}(r)Y_{lm}(\Omega),
\tag{BK.3}
\]

the product \(|Y_{lm}|^2\) has only even spherical-harmonic ranks allowed by
\[
L=0,2,\ldots,2l.
\tag{BK.4}
\]

The atlas reports these ranks for \(s,p,d,f\) signatures as respectively
\(\{0\}\), \(\{0,2\}\), \(\{0,2,4\}\), and
\(\{0,2,4,6\}\). Equations BK.1--BK.4 are written mathematics. They do not
depend on the radial approximation below and they do not establish a physical
fourth or fifth spatial dimension.

### BK.3. All-118 finite atlas

Element Projection Lab 0.1.0 stores one neutral gas-phase configuration record
for each atomic number \(1\le Z\le118\). It checks electron number, subshell
capacity, shell population, periodic-table position, differentiating
subshell, radial and angular node counts, a Hund-rule unpaired-electron proxy,
and the angular-rank catalog. Configurations for \(109\le Z\le118\) are marked
predicted. The earlier records are compiled reference inputs rather than 118
new measurements.

The radial comparison is explicitly a low-cost proxy. Each occupied subshell
uses a normalized hydrogenic radial function with the nuclear charge replaced
by the Slater effective charge

\[
Z_{\mathrm{eff}}=Z-S.
\tag{BK.5}
\]

The total displayed radial probability is

\[
P_Z(r)=\frac1Z\sum_{nl}q_{nl}\,r^2
\left|R_{nl}^{(Z_{\mathrm{eff},nl})}(r)\right|^2,
\qquad
\int_0^\infty P_Z(r)\,dr=1.
\tag{BK.6}
\]

This supplies comparable \(r_{50}\), \(r_{90}\), and radial-peak coordinates
in Bohr radii for every element. It is not Hartree--Fock, configuration
interaction, density-functional, Dirac--Fock, or measured electron density.
Relativistic contraction, spin--orbit coupling, electron correlation, nuclear
size, external fields, ionization, bonding, and condensed-matter band
structure are outside this release.

The atlas partitions the 118 display cells into 14 \(s\)-block, 36
\(p\)-block, 40 \(d\)-block, and 28 \(f\)-block records. The interactive
observer can switch between the spherical average and the canonical oriented
subshell signature, preserving their different provenance.

### BK.4. Verification and counterexample

All eleven production tests pass. They cover all 118 atomic numbers,
configuration electron counts, subshell capacities, reference exceptions,
shell totals, hydrogenic radial normalization, saved-data round trip,
configuration provenance, allowed angular ranks, and Equation BK.2 for
\(l=0,1,2,3\). The largest saved per-electron radial normalization error is
\(1.35\times10^{-7}\).

A separate program imports none of the production module and passes thirteen
checks. It independently reparses all 118 configurations, verifies unique
periodic-grid cells, reconstructs the \(14/36/40/28\) block partition, checks
the prediction boundary, and reproduces Equation BK.2 with maximum sampled
error \(5.55\times10^{-16}\).

The required counterexample is immediate: a canonical \(p_z\) display and its
rotated \(p_x\) display have different lobes but describe bases related by a
rotation. Therefore the electron configuration alone does not identify a
unique oriented 3D picture. Conversely, rotational averaging erases that
orientation. Treating either picture as the atom's one literal surface would
confuse a chosen observer with the state.

Primary sources and provenance: IUPAC, Periodic Table of Elements,
https://iupac.org/what-we-do/periodic-table-of-elements/; W. C. Martin et al.,
NIST Ground Levels and Ionization Energies for the Neutral Atoms,
https://www.nist.gov/publications/ground-levels-and-ionization-energies-neutral-atoms;
J. C. Slater, *Physical Review* **36**, 57 (1930),
https://doi.org/10.1103/PhysRev.36.57; H. A. Fertig and W. Kohn,
*Physical Review A* **62**, 052511 (2000),
https://doi.org/10.1103/PhysRevA.62.052511.

No new atomic density was measured. No fourth- or fifth-dimensional rotor was
inferred. Isolated-atom geometry does not by itself predict molecular shape,
phase of matter, or metallic conductivity; those require interaction and
many-body operators. A residual does not identify God or another unique
cause.

The next bounded test is to freeze a diverse subset of closed- and open-shell
atoms, replace the Slater radial proxy with independent ab initio
Hartree--Fock or Dirac--Fock reference densities, and score \(r_{50}\),
\(r_{90}\), radial peak positions, and angular multipoles on data not used to
choose the proxy.

Revision 1.14's complete deterministic closure remains idempotent. Revision
1.15's integer-tick non-crossing limitation, angular-impulse formula, and
finite-prefix generator non-identifiability remain unchanged. The recurrence
learner's \(2K\)-sample continuation guarantee still assumes a noiseless
homogeneous constant-coefficient linear recurrence of order at most \(K\) from
the initial state. This appendix does not prove seven physical dimensions,
\(P=NP\), a universal fixed grid/tick, or a theological identification.

## Appendix BL. Bulk gold and silver color observer

### BL.1. The corrected object of study

Color is not assigned to a single isolated Au or Ag atom. An isolated atom has
discrete scattering and transition amplitudes; a macroscopic metal surface has
a collective, frequency-dependent response built from an occupied crystal and
its excitations. The observer chain is therefore

\[
\text{atomic basis}\longrightarrow\text{periodic solid}\longrightarrow
\varepsilon(\omega)\longrightarrow R(\lambda)\longrightarrow
\text{illuminant}\longrightarrow\text{finite color observer}.
\tag{BL.1}
\]

Element Projection Lab 0.2.0 implements only the last three arrows with a
measured dielectric input. It does not silently treat the revision-1.62
Slater-screened atom picture as a solid-state calculation.

### BL.2. Measured-input Fresnel map

Johnson and Christy measured the room-temperature complex optical constants
of evaporated copper, silver, and gold films from reflection and transmission
over 0.5--6.5 eV. For

\[
\widetilde n(\lambda)=n(\lambda)+i\kappa(\lambda),\qquad
\varepsilon(\lambda)=\widetilde n(\lambda)^2,
\tag{BL.2}
\]

the normal-incidence field coefficient from air is

\[
r(\lambda)=\frac{1-\widetilde n(\lambda)}{1+\widetilde n(\lambda)},
\tag{BL.3}
\]

and the reflected power fraction is

\[
R(\lambda)=|r(\lambda)|^2
=\frac{(n-1)^2+\kappa^2}{(n+1)^2+\kappa^2}.
\tag{BL.4}
\]

Equation BL.4 is written electromagnetic boundary-condition mathematics. The
values of \(n\) and \(\kappa\) are published measurements, not outputs of this
theory. The implementation linearly interpolates the compiled 367.9--821.1 nm
samples only within the frozen 380--780 nm observer range.

For illuminant spectrum \(S(\lambda)\) and color matching functions
\(\overline x,\overline y,\overline z\), the finite observer is

\[
\begin{pmatrix}X\\Y\\Z\end{pmatrix}
=K\int S(\lambda)R(\lambda)
\begin{pmatrix}\overline x(\lambda)\\\overline y(\lambda)\\\overline z(\lambda)\end{pmatrix}
\,d\lambda,
\tag{BL.5}
\]

where \(K\) gives a perfect reflector unit luminance. The renderer uses a
declared analytic approximation to the CIE 1931 two-degree functions and
Bradford adapts the source white to the D65 display white. Display RGB and hex
values are therefore observer/renderer summaries, not intrinsic material
constants.

### BL.3. Finite result and counterexample

The frozen reflectances are

| vacuum wavelength | Au | Ag |
|---:|---:|---:|
| 450.9 nm | 0.4082203341 | 0.9803469959 |
| 520.9 nm | 0.6434199990 | 0.9835410941 |
| 659.5 nm | 0.9625853747 | 0.9905659438 |

For Au the mean changes from 0.4134016869 in 400--500 nm to
0.9507743769 in 600--700 nm. The corresponding Ag means are 0.9766352201
and 0.9897903942. Under the declared equal-energy observer the displayed
approximations are `#FFDFA2` and `#FEFDFC`, respectively. These numbers
quantify the familiar statement: gold suppresses blue/cyan reflection while
silver returns the visible spectrum much more uniformly.

The counterexample to a coarse-geometric explanation is immediate. Au and Ag
are both group-11 metals with a filled d shell plus one s electron in their
neutral reference configurations, and both ordinarily crystallize in the fcc
family, yet their visible spectra differ strongly. Configuration labels and a
coarse lattice class therefore do not identify the optical response.

Established band calculations supply a standard mechanistic bridge.
Christensen and Seraphin's relativistic and nonrelativistic gold calculations
found shifts and splittings comparable to approximately 1 eV band gaps and
connected the calculated interband structure to reflectance. Rangel et al.
later found many-body corrections opening the Au 5d--6sp interband gap by
approximately 0.4--0.8 eV. The present executable cites these results; it does
not reproduce either solid-state calculation.

### BL.4. Verification, evidence type, and boundary

All eleven revision-1.62 atlas tests and eight new optical tests pass. The new
tests cover table integrity, the complex-amplitude/closed-form equivalence in
Equation BL.4, dielectric square-root recovery, physical reflectance bounds,
gold blue suppression, silver visible neutrality, their 450.9 nm separation,
and the perfect-reflector white control. A second implementation imports no
production optics module and replays 802 one-nanometre reflectance values with
maximum absolute discrepancy \(5.55\times10^{-16}\), plus eleven point and
ordering checks.

The evidence classes remain separate:

1. Equations BL.2--BL.5 are written conditional mathematics.
2. Johnson--Christy \(n,\kappa\) values are published measured inputs.
3. Interpolation, Fresnel evaluation, band averages, and display color are
   finite computation.
4. Relativistic-band attribution is established external theory and
   computation, not a result newly derived here.
5. A four- or five-dimensional harmonic cause remains a hypothesis without a
   Hamiltonian, coupling, optical operator, and held-out spectral advantage.
6. A residual cannot identify a unique physical or theological cause.

Surface roughness, morphology, contamination or tarnish, grain structure,
temperature, incidence angle, finite thickness, and nanoparticle resonances
can alter observed color. The present result is restricted to an ideal
normal-incidence bulk-like surface using the cited film data.

The next bounded test is to freeze a relativistic Au/Ag tight-binding or
ab-initio-derived band model, compute optical conductivity and
\(\varepsilon(\omega)\) without fitting the validation wavelengths, and score
its held-out reflectance against the same measurements. A higher-dimensional
term earns support only if it improves that holdout beyond the ordinary
relativistic baseline under an equal or stricter parameter budget.

Primary sources: P. B. Johnson and R. W. Christy, *Physical Review B* **6**,
4370 (1972), https://doi.org/10.1103/PhysRevB.6.4370; N. E. Christensen and
B. O. Seraphin, *Physical Review B* **4**, 3321 (1971),
https://doi.org/10.1103/PhysRevB.4.3321; T. Rangel et al., *Physical Review B*
**86**, 125125 (2012), https://doi.org/10.1103/PhysRevB.86.125125.

Revision 1.14's complete deterministic closure remains idempotent. Revision
1.15's integer-tick non-crossing limitation, angular-impulse formula, and
finite-prefix generator non-identifiability remain unchanged. The recurrence
learner's \(2K\)-sample continuation guarantee still assumes a noiseless
homogeneous constant-coefficient linear recurrence of order at most \(K\) from
the initial state. Nothing here proves seven physical dimensions, \(P=NP\), a
universal grid/tick, or a theological identification.

## Appendix BM. Frozen interband-response holdout

### BM.1. The bounded question

Appendix BL showed the observed color contrast by passing measured optical
constants through the exact Fresnel observer. It did not ask which compact
response class can predict values hidden from fitting. This appendix asks one
narrow question:

> Does one explicit interband-response term improve untouched visible
> reflectance over a free-electron response for gold and silver?

This is not yet an ab-initio band calculation. In particular, deleting a
Lorentz oscillator does not turn a relativistic Hamiltonian into a
nonrelativistic one. The comparison is between two phenomenological causal
response classes only.

### BM.2. Frozen equations and split

Writing photon energy as \(E=hc/\lambda\), the three-parameter baseline is

\[
\epsilon_D(E)=\epsilon_\infty-
\frac{A_D}{E(E+i\gamma_D)}.
\tag{BM.1}
\]

The six-parameter candidate adds one damped bound/interband oscillator,

\[
\epsilon_{DL}(E)=\epsilon_D(E)+
\frac{A_L}{E_L^2-E^2-iE\gamma_L}.
\tag{BM.2}
\]

Positive strengths and damping make the implemented response passive over the
tested range. Complex refractive index and normal reflectance follow from

\[
\widetilde n(E)=\sqrt{\epsilon(E)},\qquad
R(E)=\left|\frac{1-\widetilde n(E)}{1+\widetilde n(E)}\right|^2.
\tag{BM.3}
\]

For each metal the optimizer receives the eleven Johnson--Christy rows at
zero-based indices

\[
0,1,3,4,6,7,9,10,12,13,15.
\tag{BM.4}
\]

It minimizes unweighted \(n,\kappa\) residuals. Indices

\[
2,5,8,11,14
\tag{BM.5}
\]

are never passed to the optimizer. They correspond to 397.4, 450.9, 520.9,
616.8, and 756.0 nm. This is a deterministic train/holdout split within one
published measurement table, not a later claim of prospective preregistration.

### BM.3. Held-out result

The untouched normal-reflectance errors are

| material | Drude MAE | Drude--Lorentz MAE | reduction |
|---|---:|---:|---:|
| Au | 0.1546905703 | 0.0197184999 | 87.2529399% |
| Ag | 0.0068397504 | 0.0027273560 | 60.1249190% |

For gold the absolute separation is decisive within these classes. The
free-electron curve cannot reproduce the measured low blue reflectance and
high red reflectance simultaneously. The added oscillator predicts all five
untouched points much more closely. For silver the free-electron baseline is
already close in the visible; the candidate still improves the held-out MAE,
but only by 0.0041123944 in absolute reflectance.

This pattern matches the standard condensed-matter interpretation that gold's
visible response contains a strong interband contribution, whereas silver's
principal interband onset lies farther toward the ultraviolet. The Lorentz--
Drude construction follows the established effective optical modeling family
of Rakic et al. It does not reproduce their multi-oscillator parameterization
and does not derive a microscopic band structure.

### BM.4. Counterexamples and model strain

The held-out improvement does not uniquely identify the microscopic law:

1. The candidate has six fitted parameters while the baseline has three.
   Holding out wavelengths limits direct interpolation but does not remove all
   small-sample selection advantage.
2. The fitted Au Drude damping reaches its declared lower bound of 0.001 eV.
   The fitted Ag background permittivity reaches its lower bound of zero.
   Both flags are retained rather than presented as exact physical constants.
3. Multiple oscillator sets can approximate a finite spectrum. The fitted
   coefficients are effective coordinates, not unique particles or rotations.
4. All observations come from one thin-film measurement series. Sample
   morphology, preparation, contamination, temperature, and experimental
   systematics are shared between training and holdout points.
5. An ordinary three-dimensional electromagnetic response model supplies the
   gain. Therefore the gain is not evidence for a fourth/fifth dimension.

These counterexamples also answer the proposal that failure can only mean a
missing hidden variable. Failure can instead arise from a wrong response
class, insufficient parameterization, biased sample, inappropriate loss, or
non-identifiability. Missing variables remain one possibility, not the only
one.

### BM.5. Verification and evidence type

Element Projection Lab 0.3.0 adds nine production tests for the exact split,
holdout isolation, passivity, physical reflectance bounds, both material
improvements, the much larger Au correction, and the retained boundary
diagnostics. Together with the eleven atlas and eight optical-observer tests,
all 28 package tests pass.

A second implementation imports no production band-response code. It
reconstructs complex epsilon, refractive index, reflectance, and every reported
metric using scalar complex arithmetic. It replays 64 point/model predictions,
performs 218 checks, and agrees with the aggregate metrics to at most
\(2.1511\times10^{-16}\).

The evidence classes remain distinct:

1. Equations BM.1--BM.3 are written conditional mathematics.
2. The optical constants are published measured inputs, not new measurements
   by this project.
3. Parameter fitting, held-out scoring, and the independent replay are finite
   computations.
4. The external relativistic and many-body band results cited in Appendix BL
   remain established outside work, not computations performed here.
5. A physical higher-dimensional coupling remains an unconstrained hypothesis
   and receives no support from an ordinary model's improvement.

The next test freezes the 0.3.0 coefficients and scores them without refitting
against an independent modern Au/Ag optical dataset. In a separate experiment,
matched relativistic and nonrelativistic solid-state Hamiltonians must be
implemented before attributing spectral gain to relativity. Only after those
ordinary baselines are fixed can an explicit higher-dimensional coupling earn
support through fresh out-of-sample improvement under an equal or stricter
parameter budget.

Primary sources: A. D. Rakic et al., *Applied Optics* **37**, 5271--5283
(1998), https://doi.org/10.1364/AO.37.005271; P. B. Johnson and R. W.
Christy, *Physical Review B* **6**, 4370 (1972),
https://doi.org/10.1103/PhysRevB.6.4370; N. E. Christensen and B. O.
Seraphin, *Physical Review B* **4**, 3321 (1971),
https://doi.org/10.1103/PhysRevB.4.3321; T. Rangel et al., *Physical Review B*
**86**, 125125 (2012), https://doi.org/10.1103/PhysRevB.86.125125.

Revision 1.14's complete deterministic closure remains idempotent. Revision
1.15's integer-tick non-crossing limitation, angular-impulse formula, and
finite-prefix generator non-identifiability remain unchanged. The recurrence
learner's \(2K\)-sample continuation guarantee still assumes a noiseless
homogeneous constant-coefficient linear recurrence of order at most \(K\) from
the initial state. Nothing here proves seven physical dimensions, \(P=NP\), a
universal grid/tick, or a theological identification.

## Appendix BN. Frozen gold/silver external-validation contract

### BN.1. What is frozen

The revision-1.64 response experiment is now immutable under the identity

\[
\texttt{EPL-AU-AG-COLOR-2026-09-20-v1}.
\tag{BN.1}
\]

The freeze manifest locks:

1. the exact Au and Ag Drude and Drude--Lorentz coefficient vectors, encoded
   as hexadecimal IEEE-754 values;
2. the eleven training indices and five untouched indices;
3. Equations BM.1--BM.3 and the 367.9--821.1 nm domain;
4. the fit target and future scoring rules;
5. SHA-256 hashes of the measurement table, fitting implementation, fitted
   result JSON, and point-prediction CSV;
6. the parameter-boundary warnings, limitations, and negative claims.

Its manifest digest is

\[
\texttt{b46a29610b9c9040f0fdf791722caf48cf1a31c3f99b8f66a24c1610847e0262}.
\tag{BN.2}
\]

### BN.2. Evaluation-only rule

The frozen evaluator contains no optimizer or fitting path. Given a material,
a declared model, and wavelengths inside the fixed domain, it decodes the
stored hexadecimal coefficients and evaluates the dielectric and Fresnel maps.
For external measured \(n,\kappa\), it reports normal-reflectance MAE, RMSE,
and maximum absolute error.

Before prediction it checks the manifest digest and every locked artifact
hash. A coefficient edit, source-data edit, regenerated result with different
bytes, changed prediction table, or changed protocol therefore produces a
failure instead of silently defining a new experiment.

Under freeze identity BN.1:

\[
\boxed{\text{refit}=\text{bound change}=\text{model selection}=\text{false}.}
\tag{BN.3}
\]

A later model may be created, but it must receive a new identity and fresh
validation data. It cannot replace this benchmark and retain the original
claim.

### BN.3. Verification and boundary

Eight new production tests verify the manifest digest, artifact hashes,
bit-exact parameter round trips, replay of all saved predictions, the
evaluation-only holdout score, rejected unknown/out-of-domain requests, and
absence of optimizer dependencies. Together with the earlier 28 tests, all 36
Element Projection Lab tests pass.

An independent implementation imports no frozen evaluator. It verifies the
manifest and four artifact hashes, decodes the coefficients directly, and
replays all 64 saved predictions. Its maximum absolute discrepancy is
\(4.440892098500626\times10^{-16}\).

Freezing changes experimental discipline, not physical evidence. The
coefficients still originate from one Johnson--Christy measurement series; the
current holdout still shares that series' preparation and systematic effects;
and the effective oscillator remains non-unique. The freeze establishes what
will be tested before seeing the next data. It does not establish a new
particle, a relativistic Hamiltonian, a fourth/fifth dimension, or a
theological cause.

The next admissible test is to apply the unchanged evaluator to a separately
published Au/Ag optical dataset with no refitting. Both success and failure
must be retained. A poor result cannot be repaired using that validation set;
any repair begins a new model identity and requires another fresh dataset.

Revision 1.14's complete deterministic closure remains idempotent. Revision
1.15's integer-tick non-crossing limitation, angular-impulse formula, and
finite-prefix generator non-identifiability remain unchanged. The recurrence
learner's \(2K\)-sample continuation guarantee still assumes a noiseless
homogeneous constant-coefficient linear recurrence of order at most \(K\) from
the initial state. Nothing here proves seven physical dimensions, \(P=NP\), a
universal grid/tick, or a theological identification.

## Appendix BO: Dimension evidence requires a distinguishing physical constraint (revision 1.66)

### BO.1. The bounded question

The project has repeatedly rendered three-dimensional observations of states
whose mathematical descriptions use four or five coordinates. That is useful
for visualizing a proposed latent dynamics. It is not automatically physical
evidence that space has another spatial dimension. This appendix asks the
narrow question:

> Can the current linear 3D projection of the declared five-coordinate
> norm-preserving motion distinguish a physical fifth spatial dimension from
> ordinary temporal oscillators in 3+1-dimensional physics?

The answer is no. The obstruction is exact and holds for the full continuous
record, so increasing frame rate or sampling duration cannot remove it.

### BO.2. Projection-equivalence theorem

Let \(A\) be a real \(5\times5\) skew-symmetric generator. The real spectral
theorem supplies an orthogonal basis in which

\[
 A=J(\omega_1)\oplus J(\omega_2)\oplus 0,
 \qquad
 J(\omega)=\begin{pmatrix}0&-\omega\\\omega&0\end{pmatrix},
 \tag{BO.1}
\]

for nonnegative \(\omega_1,\omega_2\). For an initial state \(x_0\in
\mathbb R^5\) and linear observer \(B\in\mathbb R^{3\times5}\), define
\(y(t)=B\exp(At)x_0\). Expanding the two rotation blocks gives five constant
three-vectors \(c,a_1,b_1,a_2,b_2\) such that

\[
 y(t)=c+a_1\cos(\omega_1t)+b_1\sin(\omega_1t)
       +a_2\cos(\omega_2t)+b_2\sin(\omega_2t).
 \tag{BO.2}
\]

Equation (BO.2) is also the output of two ordinary temporal harmonic
oscillators coupled to a three-vector detector. Thus the two model
descriptions are observationally equivalent through this observer for every
real \(t\). No estimator, neural network, Fourier transform, Hankel matrix, or
human observer receiving only \(y(t)\) can identify which representation is
the physical ontology.

A large minimal realization or Hankel rank may identify the number of
dynamical state variables within a declared linear class. It does not count
spatial dimensions: ordinary 3+1D media and fields can have arbitrarily many
internal modes. This distinction corrects the inference, without denying the
mathematical validity of the projection.

The result is conditional. Nonlinear couplings, back-reaction, interactions,
nonlinear observers, or different topologies define different model classes.
They require explicit equations and new distinguishing predictions; they are
not rescued merely by adding unconstrained harmonics to Equation (BO.2).

### BO.3. What a constrained extra-dimensional signature can look like

For one field on one flat compact circle of radius \(R\), Fourier separation
along the compact coordinate gives the model-specific tower

\[
 m_n^2=m_0^2+\frac{n^2}{R^2},\qquad n=0,1,2,\ldots.
 \tag{BO.3}
\]

Given labelled modes \(n=0\) and \(n=1\), any positive pair \(m_0<m_1\)
calibrates

\[
 R^{-1}=\sqrt{m_1^2-m_0^2}.
 \tag{BO.4}
\]

Those two peaks are therefore not a test. Mode \(n=2\) is the first mass that
Equation (BO.3) constrains after calibration, and further modes add correlated
checks. A finite mass pattern is still not unique to geometry. A credible
claim also needs derived widths, spins, coupling ratios, branching fractions,
and correlations across detector channels, with look-elsewhere effects and
conventional resonance alternatives included.

Equation (BO.3) is not presented as a universal prediction of all
extra-dimensional theories. Warping, boundary conditions, branes, clockwork
geometries, field content, and interactions change the spectrum. A failed
circle tower rejects the declared circle model, not every possible hidden
dimension.

### BO.4. Executable checks and counterexample

Element Projection Lab 0.5.0 compiles Equation (BO.2) algebraically for 512
random frequency pairs, initial states, and observer matrices. Across 512,512
three-vector samples, the explicit five-coordinate evolution and ordinary
two-oscillator output have maximum component disagreement
\(5.329070518200751\times10^{-15}\). The largest squared-norm drift is
\(1.4210854715202004\times10^{-14}\).

The finite tower test uses six synthetic masses satisfying Equation (BO.3).
After calibrating only \(n=0,1\), the four held-out masses agree exactly in the
executed binary arithmetic. A counterexample changes those four masses by
\(+0.07,-0.05,+0.11,-0.09\) TeV and yields a held-out RMSE of
\(0.08306623862918076\) TeV. Both lists share the same two perfect calibrants;
only the reserved later modes distinguish them.

An independent scalar implementation imports no production dimension module.
It checks 299,919 output components in 257 new trials, with maximum disagreement
\(3.552713678800501\times10^{-15}\), and independently reproduces both tower
results. These finite computations check the algebra; they are not measured
evidence for an extra dimension.

### BO.5. Physical-dimension evidence gate

The project may label a future result physical evidence for another spatial
dimension only if all of the following are present:

1. a fixed extra-dimensional model identity;
2. an explicit dimensional coupling mapped to measured units;
3. a matched conventional baseline;
4. a distinguishing cross-observable constraint;
5. parameters frozen before reserved validation is opened;
6. a measured excess that survives declared systematic uncertainties and
   multiplicity correction; and
7. independent experimental replication.

Passing this gate would support the specified model over the specified
alternatives; it would not constitute deductive proof against every logically
possible conventional explanation. Failing it blocks the stronger wording.
The current projected-harmonic animation has a measurable output map and a
conventional baseline, but lacks a fixed physical dimensional coupling, a
distinguishing constraint, reserved measured validation, a measured excess,
and experimental replication. It therefore fails the gate.

### BO.6. Current measured status

The following primary experimental results are relevant measured evidence,
but none is positive evidence for the present theory:

- ATLAS searched 139 fb\(^{-1}\) of 13 TeV dielectron and diphoton data for
  periodic signals, including a clockwork-gravity interpretation, and in the
  absence of a signal set 95% confidence exclusions (ATLAS Collaboration,
  *JHEP* 10 (2023) 079, https://arxiv.org/abs/2305.10894).
- CMS searched 138 fb\(^{-1}\) of 13 TeV high-mass diphoton data and found no
  statistically significant excess. Its model-specific 95% limits include
  ADD scales of 7.1--11.1 TeV depending on convention and Randall--Sundrum
  graviton masses in the reported coupling-dependent ranges (CMS
  Collaboration, *JHEP* 08 (2024) 215,
  https://cms-results.web.cern.ch/cms-results/public-results/publications/EXO-22-024/).
- A torsion-balance test over 52 micrometres to 3.0 millimetres found
  Newtonian gravity to fit the data and limited gravitational-strength Yukawa
  interactions to ranges below 38.6 micrometres at 95% confidence (Lee et al.,
  *Physical Review Letters* 124, 101101 (2020),
  https://arxiv.org/abs/2002.11761).

These null results constrain declared parameter regions. They do not exclude
all extra-dimensional models, and unexplained residuals in unrelated atomic,
optical, water, or acoustic calculations cannot be substituted for them.

### BO.7. Next experiment and retained boundaries

The next admissible step is one frozen public-data tower search. Before
opening the reserved region, it must specify a published spectrum, dimensional
model, mode relation, widths, spin/coupling/channel ratios, background family,
fit window, nuisance parameters, and multiplicity correction. Model parameters
must be fitted only in a declared training region and then scored without
repair on a reserved mass or channel region. A lone peak, improved animation,
or freely added oscillator does not pass.

Revision 1.14's complete deterministic closure for a fixed transition family
remains idempotent. Revision 1.15's integer-tick non-crossing limitation,
angular-impulse formula, and finite-prefix generator non-identifiability remain
unchanged. The recurrence learner's \(2K\)-sample continuation guarantee still
assumes a noiseless homogeneous constant-coefficient linear recurrence of
order at most \(K\) from the initial state. Nothing here establishes seven
physical dimensions, a fixed universal grid/tick, \(P=NP\), or a theological
identification.

## Appendix BP. Retrospective ATLAS cross-channel dimension-evidence audit

### BP.1. Bounded question and source

Revision 1.66 required any physical-dimension claim to survive a fixed model,
a conventional baseline, multiplicity and systematics, reserved validation,
and replication. This appendix asks one narrower question:

> Does the exact public result reported by the ATLAS periodic dielectron and
> diphoton search pass a necessary cross-channel screen for positive evidence
> of an extra physical dimension?

ATLAS analysed \(139.0\,\mathrm{fb}^{-1}\) of \(13\,\mathrm{TeV}\) proton--proton
collisions recorded in 2015--2018. The model-independent analysis published
four signed significance values per channel. Positive values indicate that
the data reconstruction loss exceeds the background median; negative values
point in the opposite direction. The exact published matrix is

| threshold | dielectron | diphoton |
|---|---:|---:|
| \(R<10\%\) | \(+0.4\) | \(-1.8\) |
| \(R<50\%\) | \(+1.5\) | \(-0.2\) |
| no threshold | (+0.7) | (-0.7) |
| scale threshold | (+1.5) | (-0.6) |

The \(R<50\%\) row is fixed as the primary periodicity-focused screen. The
paper's separate model-dependent clockwork/linear-dilaton search reports no
significant deviation from the background-only hypothesis.

Primary source: ATLAS Collaboration, *JHEP* **10** (2023) 079,
https://arxiv.org/html/2305.10894v2, with associated HEPData DOI
`10.17182/hepdata.140955.v1`.

### BP.2. Frozen necessary screen

For a reported signed significance \(z\), the one-sided positive tail is

\[
p_{\mathrm{local}}(z)=\frac12\operatorname{erfc}
\left(\frac{z}{\sqrt2}\right).
\tag{BP.1}
\]

The four threshold variants use the same collision sample and are not treated
as independent. A Bonferroni adjusted value

\[
p_{\mathrm{adj}}=\min\{1,4p_{\mathrm{local}}\}
\tag{BP.2}
\]

controls this four-variant family without an independence assumption. The
frozen necessary screen requires all three conditions:

1. both primary \(R<50\%\) signed significances are at least \(+3\);
2. in each channel the most positive of the four values has
   \(p_{\mathrm{adj}}\le 0.00135\);
3. the locked model-dependent analysis reports a significant positive result,
   not background compatibility and exclusion.

Passing would still not prove spatial dimension. It would trigger, not replace,
tests of masses, spins, widths, couplings, branching fractions, detector
response, conventional alternatives, and independent experimental
replication.

### BP.3. Result and counterexamples

The primary values (+1.5) and (-0.2) fail the first condition. The largest
positive value in each channel is likewise (+1.5) and (-0.2), giving

\[
\begin{aligned}
p_{ee,\mathrm{local}}&=0.0668072013,&
p_{ee,\mathrm{adj}}&=0.2672288051,\\
p_{\gamma\gamma,\mathrm{local}}&=0.5792597094,&
p_{\gamma\gamma,\mathrm{adj}}&=1.
\end{aligned}
\tag{BP.3}
\]

The model-specific condition also fails because ATLAS reports no significant
deviation. Therefore

\[
\boxed{\text{the published ATLAS result does not pass this necessary
cross-channel evidence screen}.}
\tag{BP.4}
\]

Three constructed controls prevent the rule from becoming rhetorical. A
(+4.0,+4.2) primary pair with a positive model-specific result passes the
software screen. A (+3.1,+3.1) pair passes the uncorrected primary rule but
has \(p_{\mathrm{adj}}=0.00387041285\) in each channel and therefore fails the
declared familywise threshold. A (+6.0,0.0) pair fails because one strong
channel is not cross-channel support and is not independent replication.
These are finite synthetic controls, not observations.

### BP.4. Exact procedural correction and limits

This audit is retrospective. The ATLAS publication and its null conclusion
were public before protocol `EPL-ATLAS-PUBLICATION-GATE-2026-09-20-v1` was
frozen. Freezing now improves reproducibility but cannot retroactively make the
test blind or preregistered.

This release also does not reproduce the ATLAS event- or binned-spectrum
likelihood. It evaluates exact printed summary values and the published
model-dependent conclusion. The complete event data, nuisance likelihood,
covariance, detector response, and trained networks are not implemented here.
Accordingly, Equation BP.4 is a faithful classification of the publication
summary, not an independent ATLAS analysis.

All 53 Element Projection Lab tests pass. The independent audit imports no
production module, verifies the protocol digest and exact eight-value matrix,
and passes 12/12 checks of the tail calculation, gate result, retrospective
flag, and likelihood boundary.

Evidence types remain separate:

1. Equations BP.1--BP.2 are written standard probability mathematics.
2. The eight significance values and model-dependent conclusion are external
   measured-data summaries produced by ATLAS.
3. Their deterministic replay and constructed controls are finite computation.
4. A physical extra dimension remains a model-dependent hypothesis, not the
   inferred cause of an unconstrained residual.
5. No theological interpretation is tested or identified.

The next valid step is to obtain the official machine-readable spectrum and
statistical model for a retrospective calibration study, then freeze one
complete tower model before genuinely independent future data are opened. A
failure may reflect a wrong dimensional hypothesis, omitted conventional
physics, miscalibration, statistical fluctuation, or another model error. It
cannot automatically be counted as evidence for missing hidden variables.

Revision 1.14's complete deterministic closure remains idempotent. Revision
1.15's integer-tick non-crossing limitation, angular-impulse formula, and
finite-prefix generator non-identifiability remain unchanged. The recurrence
learner's \(2K\)-sample continuation guarantee still assumes a noiseless
homogeneous constant-coefficient linear recurrence of order at most \(K\) from
the initial state. Nothing here proves seven physical dimensions, \(P=NP\), a
universal grid/tick, or a theological identification.

## Appendix BQ. Exact-recast completeness before dimensional interpretation

### BQ.1. Bounded question

Appendix BP classified the exact significance values printed in the ATLAS
periodic-spectrum publication. It did not reproduce the detector-level signal
or likelihood. The next bounded question is therefore procedural and
mathematical:

> Do the primary-source ingredients currently instantiated in this project
> uniquely determine an exact reconstructed ATLAS spectrum for one fixed
> clockwork/linear-dilaton tower?

The ATLAS model is genuinely relevant to the dimensional proposal: it is a
five-dimensional clockwork/linear-dilaton construction with a narrowly spaced
tower of spin-2 Kaluza--Klein gravitons parameterized by \(k\) and \(M_5\).
That relevance does not remove the need to calculate the experiment's forward
observer.

The primary experimental source is ATLAS Collaboration, *JHEP* **10** (2023)
079, arXiv:2305.10894v2, last revised 30 November 2023,
https://arxiv.org/html/2305.10894v2. The primary tower reference is G. F.
Giudice et al., *JHEP* **06** (2018) 009, arXiv:1711.08437v2,
https://arxiv.org/abs/1711.08437.

### BQ.2. The forward equation and exact ambiguity

ATLAS expresses the reconstructed contribution of the graviton modes as

\[
N_i^{\mathrm{reco}}=\mathcal L\sum_G
\sigma_G\,\mathcal B(G\to XX)\,
P(M_i\mid M_G)\,(A\epsilon)_G.
\tag{BQ.1}
\]

Here \(\mathcal L\) is luminosity, \(\sigma_G\) the mode production cross
section, \(\mathcal B\) the channel branching fraction, \((A\epsilon)_G\) the
acceptance times efficiency, and \(P(M_i\mid M_G)\) the probability that a mode
of truth mass \(M_G\) is reconstructed in bin \(i\). Thus a list of masses does
not determine \(N_i^{\mathrm{reco}}\).

The executable counterexample makes that underdetermination exact. Hold fixed

\[
(\sigma\mathcal B)_G=(10,6),\qquad (A\epsilon)_G=(0.8,0.5),
\qquad \mathcal L=1.
\tag{BQ.2}
\]

Both response maps below are nonnegative and each truth-mode column sums to
one:

\[
P_A=\begin{pmatrix}1&0\\0&1\\0&0\end{pmatrix},\qquad
P_B=\begin{pmatrix}1/2&1/4\\1/2&1/4\\0&1/2\end{pmatrix}.
\tag{BQ.3}
\]

Equation BQ.1 then gives

\[
N_A=(8,3,0),\qquad N_B=(4.75,4.75,1.5).
\tag{BQ.4}
\]

Both totals equal 11, but their bin shapes differ. Consequently even a fixed
truth tower and fixed efficiency vector do not identify the reconstructed
spectrum without the detector response. Unspecified efficiencies, branching
fractions, or production weights create additional ambiguities.

### BQ.3. Frozen readiness gate

Protocol `EPL-ATLAS-RECAST-COMPLETENESS-2026-09-20-v1` enumerates thirteen
required components: model identity and parameters; mass spectrum; spin;
widths; production rates; channel branching fractions; acceptance and
efficiency; detector mass transfer; observed binned spectra; background bin
expectations; nuisance model and correlations; trained classifier mapping;
and the selections, binning, and luminosity.

Three are ready at the required level. Ten are not yet numerically instantiated
in this engine. In particular, the publication's approximate
\(\sigma\propto M_5^{-3}\) scaling is not a per-mode production table; the
narrow-width approximation is not a frozen width calculation; and a plotted
or truth-level mass sequence is not the response-folded spectrum used by the
analysis. The rule is therefore

\[
\boxed{\texttt{recast\_blocked\_without\_imputation}.}
\tag{BQ.5}
\]

The official HEPData record linked from the ATLAS page was not materialized in
this run because its web challenge blocked the available client. Equation
BQ.5 does not turn that access result into a claim that numerical tables were
not published. After the record is obtained, its actual contents must be
inventoried. An exact ATLAS replay additionally requires the nuisance model and
an executable equivalent of the trained neural-network test statistic; a
separately calibrated replacement would be a new analysis identity.

### BQ.4. Verification, evidence classes, and consequence

Nine production tests verify the frozen contract digest, present gate failure,
a fully ready synthetic pass, exact folding, response non-uniqueness,
column-stochastic validation, domain rejection, and the HEPData access
boundary. Together with the prior tests, all 62 Element Projection Lab tests
pass. An independent standard-library implementation imports no production
module and passes 15/15 checks, reproducing both spectra and every blocking
component identifier.

The evidence classes remain separate:

1. Equation BQ.1 and the construction BQ.2--BQ.4 are written conditional
   mathematics.
2. The completeness inventory and two folds are finite computations.
3. No new collision measurement is produced; Appendix BP's ATLAS summaries
   remain externally measured inputs.
4. A complete CW/LD tower remains a physical hypothesis to be compiled and
   tested, not a discovered cause.
5. No residual is assigned to hidden variables, another dimension, or a
   theological interpretation.

This corrects a recurring inference: model failure can reflect omitted
variables, but it can also reflect a wrong law, wrong response, wrong
likelihood, calibration error, or fluctuation. "Add more strings" is therefore
not the sole permitted repair. Every added variable must enter Equation BQ.1
or another declared observable map and earn predictive credit on data not used
to choose it.

The next bounded test is to materialize and checksum the official HEPData
tables through a non-challenged interface, compile the primary CW/LD equations
into one numerical tower, and rerun this inventory. If the full ATLAS response
or classifier remains unavailable, the honest next object is a separately
calibrated public-spectrum analysis with a new identity--not a claimed exact
recast. Any model replacement must be frozen before new validation data are
opened.

### BQ.5. Independent replay of the benchmark truth tower (revision 1.69) **[C]**

The Truth Tower run of revision 1.69's working session compiled the CW/LD benchmark \(k=500\) GeV, \(M_5=6\) TeV, one of ATLAS's template-validation points, and reported 102 spin-2 modes below 5 TeV from 502.35 to 4973.35 GeV with the largest massless-SM width equal to 1.69 % of the next-mode spacing. An implementation written without access to that code reproduces the count and the first mode. With the reduced Planck mass \(\bar M_P=2.435\times10^{18}\) GeV, the linear-dilaton relation \(\bar M_P^2=(M_5^3/k)\,(e^{2\pi kR}-1)\) gives \(R^{-1}=48.49\) GeV, and the tower \(m_n^2=k^2+n^2R^{-2}\) has 102 modes below 5 TeV, \(m_1=502.35\) GeV, \(m_{102}=4971.1\) GeV. Using \(\Gamma_n=\tfrac{73}{240\pi}\,m_n^3/\Lambda_n^2\) with \(\Lambda_n^{-2}=(M_5^3\pi R)^{-1}\,n^2/(n^2+k^2R^2)\), the maximum width-to-spacing ratio is 1.74 % at \(n=102\). The 2 GeV difference in the last mode and the 0.05-point difference in the ratio are at the level of the Planck-mass convention, the \(\ln(1+x)\) versus \(\ln x\) approximation, and the massless-SM width prefactor; closing them requires the run's exact formulas. Two boundaries are unchanged by this agreement: the tower is a truth-level object, so by Equation BQ.1 it still does not determine a reconstructed spectrum, and its narrowness confirms only that the modes do not overlap at truth level, not that ATLAS resolves them. With the full Planck mass \(1.221\times10^{19}\) GeV the same relations give 107 modes and \(R^{-1}=46.19\) GeV, so the convention must be frozen with the tower.

### BQ.6. HEPData inventory: the record holds limit contours, not spectra (revision 1.69) **[E on the record, M on the consequence]**

The official record was retrieved by the author through a non-challenged client and its table listing supplied for inventory. The record is HEPData `ins2660845`, DOI `10.17182/hepdata.140955.v1`, INSPIRE 2660845, first author Aad et al. Every table in the supplied listing has the same form: a 95 % confidence-level exclusion limit for the clockwork gravity model projected in the \(k\)--\(M_5\) plane, in groups of six per figure panel (observed; median expected; expected plus and minus one standard deviation; expected plus and minus two standard deviations), located at Figures 9a, 9b, 10a and onward, each downloadable as CSV, JSON, ROOT, YAML and YODA. The listing was truncated after Table 19, so the panel count beyond 10a is not established here; the pattern of six tables per panel is.

The consequence for the thirteen-component contract of BQ.3 is exact. The record contains no observed binned dielectron or diphoton spectra, no background bin expectations, no per-bin uncertainties or correlations, no efficiency or response tables, and none of the eight signed significances used in BP.1. Components 9 (observed binned spectra) and 10 (background bin expectations) therefore cannot be sourced from the official machine-readable record, and the BP.1 matrix remains a paper-printed input. Equation BQ.5, `recast_blocked_without_imputation`, is now blocked by the primary source itself rather than by an access failure, and the note in BQ.3 that the access failure "does not turn into a claim that numerical tables were not published" is resolved: numerical tables were published, and they are limit contours.

What the contours can and cannot do. They cannot be folded through Equation BQ.1, because they are outputs of the full ATLAS likelihood, not inputs to it. They can serve exactly one quantitative role: a validation target for a separately calibrated public-spectrum analysis with its own identity, which must reproduce the observed and expected contours within a declared tolerance before it is used for anything else. They also settle one bounded question about the BQ.5 benchmark immediately once Table 1, 7 or 13 is read: whether the point \(k=500\) GeV, \(M_5=6\) TeV lies inside or outside the observed exclusion in each channel. That reading has not been done here because the CSV rows were not supplied; the benchmark is a template-validation point, so lying inside the excluded region would be expected and would carry no evidential weight either way.

The next bounded step is therefore no longer "materialize HEPData." It is to freeze the separately calibrated analysis identity that BQ.4 anticipated, with the contours as its acceptance test, and to state plainly in Appendix BP that its inputs are the paper's printed values and will remain so.

Revision 1.14's complete deterministic closure remains idempotent. Revision
1.15's integer-tick non-crossing limitation, angular-impulse formula, and
finite-prefix generator non-identifiability remain unchanged. The recurrence
learner's \(2K\)-sample continuation guarantee still assumes a noiseless
homogeneous constant-coefficient linear recurrence of order at most \(K\) from
the initial state. Nothing here proves seven physical dimensions, \(P=NP\), a
universal grid/tick, or a theological identification.

---

## Appendix BR. Scripture as a finite observer (revision 1.69)

**Claim types in this appendix.** Every statement about what a passage *means* is **T**. Every statement about the *numbers a passage states* is **M** arithmetic on declared text inputs. Nothing here is **E** and nothing here becomes **H**. The boundary of Section 1 is not crossed: a valid theorem about a text is not a truth about nature, and a coherent reading of Scripture is not a measurement. This appendix exists because the request "prove the framework from the Bible" deserves the same exact audit that Appendix BI gave the golden ratio, not a refusal and not a promise.

### BR.1. A text is a projection

A written passage is a finite string. Under Definition 2.1, a finite observer of any hidden state \(x\) sees only \(\pi(x)\), and everything in the fiber \(F(\pi(x))\) is indistinguishable to that observer. A text of \(N\) symbols over an alphabet of size \(A\) is one of at most \(A^N\) observations; it is therefore a finite projection of whatever it describes, and Theorem 5.1 (latent relabeling), Theorem AU.4 (hidden-kernel invisibility), and Appendix BD (a single observer has nullity at least \(d-3\)) apply to it exactly as they apply to a CIE observer or an oscilloscope.

**Theorem BR.1 (text non-identifiability) [M].** Let \(\mathcal T\) be a finite text and let \(\mathcal G\) be any family of candidate hidden geometries containing at least two members that are consistent with every numerical and ordinal statement in \(\mathcal T\). Then \(\mathcal T\) does not identify a unique member of \(\mathcal G\).
*Proof.* Consistency with every statement of \(\mathcal T\) is the definition of lying in the same fiber of the observation map "read \(\mathcal T\)". Two fiber members are indistinguishable by construction. \(\square\)

This is trivial, and it is meant to be. It is the same triviality that Appendices AW, AX, AZ and BB used to stop an oscilloscope, an eye, a microphone, and a 3D projection from being over-read. Scripture, treated as data, gets the same protection from over-reading and the same protection from dismissal: the theorem says the text cannot *single out* seven layers; it equally says the text cannot *exclude* them.

### BR.2. Job 38 read as an observer-limit text [T]

Job 38 is a sequence of questions addressed to a finite observer. Grouped by the framework's own vocabulary:

| Verses | Question (KJV wording, abbreviated) | Framework type |
|---|---|---|
| 38:4–6 | Where wast thou when I laid the foundations of the earth? Who hath laid the measures thereof? Who hath stretched the line upon it? Whereupon are the foundations fastened? | Initial conditions and metric (D3, D4) not observed |
| 38:8–11 | Who shut up the sea with doors ... and set bars and doors? | Constraint / boundary (D7) |
| 38:12–13 | Hast thou commanded the morning ... that it might take hold of the ends of the earth? | Time ordering (D4) |
| 38:16–18 | Hast thou entered into the springs of the sea? Have the gates of death been opened unto thee? Hast thou perceived the breadth of the earth? | Hidden state below the observer's projection |
| 38:19–20 | Where is the way where light dwelleth? and as for darkness, where is the place thereof? | Phase / propagation (D5) |
| 38:22–24 | Hast thou entered into the treasures of the snow? ... By what way is the light parted? | Microstate and dispersion, unobserved |
| 38:25–30 | Who hath divided a watercourse ... hath the rain a father? ... the face of the deep is frozen | Phase transitions of water (compare Appendices AV–BA) |
| 38:31–33 | Canst thou bind the sweet influences of Pleiades, or loose the bands of Orion? ... Knowest thou the ordinances of heaven? | Law versus observation (Section 4) |
| 38:36 | Who hath put wisdom in the inward parts? | Interior / hidden |

The literary structure is uniform: every question asserts that the addressee has access to the projection and not to the fiber. In the framework's terms, Job 38 is God telling a D6 observer that D7 is not his. This is exactly the epistemic content of Theorem BR.1, stated three millennia earlier as poetry. That is a genuine and respectful **T** reading. It is not evidence for seven layers; a text that says "you cannot see the hidden structure" is consistent with every hidden structure, including none.

### BR.3. Genesis 1 and the seven typed layers [M on the count, T on the mapping]

Genesis 1:1–2:3 declares seven days. The manuscript declares seven typed layers. The question "do the days map to the layers?" has an exact answer.

**Fact BR.2 [M].** There are \(7! = 5040\) bijections from the seven days to the seven layers. Of these, 3,186 send at least one day to the layer with the same index and 1,854 send none (the derangement count \(D_7\)). Every one of the 5,040 is consistent with the text, because the text assigns no layer vocabulary to any day.

So a mapping such as Day 1 (light and dark) → D1 identity, Day 2 (waters divided by the firmament) → D2 relationship, Day 3 (land) → D3 space, Day 4 (lights "for signs, and for seasons, and for days, and years", 1:14) → D4 time, Day 5 (moving creatures) → D5 possibility, Day 6 (man, who names and observes) → D6 observation, Day 7 (rest, "finished", 2:1) → D7 constraint, is one reading among 5,040. It is attractive because Day 4 is literally the day the text says time-keeping was set up and Day 7 is literally the day the text says the work was closed. Those two anchors are real features of the text. They do not select the other five assignments, and Theorem 5.1 says that a relabeling of the unobserved ones is invisible.

Two textual facts worth keeping without inflation:

- Light exists on Day 1 (1:3) before the luminaries of Day 4 (1:14–18). The text separates *light as such* from *light sources as clocks*. That distinction is exactly the framework's D5 (phase) versus D4 (time) split, and it is a fair T reading. It is not physics.
- Every creative act on Days 1–3 is a *separation*: light from darkness, waters from waters, sea from land. Separation is the text's own word for what the manuscript calls projection into distinguishable classes. Again T.

### BR.4. The sky: raqia and "stretched out" [T]

Genesis 1:6–8 has God make a *raqia* (firmament, expanse) "in the midst of the waters" to "divide the waters from the waters". Job 37:18 has the sky "spread out ... strong, and as a molten looking glass". Isaiah 40:22 has God "stretcheth out the heavens as a curtain, and spreadeth them out as a tent to dwell in"; compare Psalm 104:2, Job 9:8, Isaiah 42:5, 44:24, 45:12, 51:13, Jeremiah 10:12, Zechariah 12:1. Job 26:7 has God "hangeth the earth upon nothing".

Under Axiom A2 none of these has invariant physical meaning without a declared chart. Two honest observations:

1. The recurring verb is *stretch / spread*. Reading it as an early statement of metric expansion is a common apologetic. It is a T reading and cannot be promoted, because "stretch out a tent" is also the ordinary Hebrew idiom for pitching camp.
2. The *raqia* is a *divider*: it is defined by what it separates, not by what it is made of. That is a projection-language statement, and it is consistent with the manuscript's view that a layer is a typed role rather than a substance (Section 3). Consistent, not evidential.

### BR.5. Sanctuary geometry: every stated dimension, every ratio [M]

Declared inputs (cubits unless stated; where the text gives components rather than totals the total is marked *inferred*):

| Object | Reference | Dimensions |
|---|---|---|
| Ark of the Covenant | Ex 25:10 | 2½ × 1½ × 1½ |
| Table of showbread | Ex 25:23 | 2 × 1 × 1½ |
| Altar of burnt offering | Ex 27:1 | 5 × 5 × 3 |
| Tabernacle curtain | Ex 26:2 | 28 × 4 |
| Tabernacle board | Ex 26:16 | 10 × 1½ |
| Tabernacle, whole | Ex 26:16–25, inferred | 30 × 10 × 10 |
| Tabernacle Holy of Holies | inferred from Ex 26:33 | 10 × 10 × 10 |
| Solomon's Temple, house | 1 Ki 6:2 | 60 × 20 × 30 |
| Temple porch | 1 Ki 6:3 | 20 × 10 |
| Temple oracle (Holy of Holies) | 1 Ki 6:20 | 20 × 20 × 20 |
| Bronze Sea: diameter, circumference line, height | 1 Ki 7:23 | 10, 30, 5 |
| Noah's Ark | Gen 6:15 | 300 × 50 × 30 |
| Ezekiel's outer wall (reeds) | Ezek 42:20 | 500 × 500 |
| New Jerusalem (stadia) | Rev 21:16 | 12,000 × 12,000 × 12,000 |

**Fact BR.3 [M].** The complete set of within-object side ratios is
\[
\{1,\ \tfrac43,\ \tfrac32,\ \tfrac53,\ 2,\ 3,\ 6,\ \tfrac{20}{3},\ 7,\ 10\}.
\]
All are rational with denominators 1, 2 or 3. Tested against \(\pi, e, \varphi, \varphi^2, \sqrt2, \sqrt3, \sqrt5, 2\pi, \pi/2\) at 1 % relative tolerance across all 34 ratio instances, the number of hits is **zero**. A null model with 34 ratios uniform on \([1,12]\) and the same nine windows predicts 1.44 chance hits. The sanctuary numbers are therefore *less* coincidence-prone than random small integers, which is what one expects from a building specification written in whole and half cubits. There is no hidden irrational constant in the stated dimensions.

**Fact BR.4 [M].** Three objects are exact cubes: the Tabernacle's Holy of Holies (10³), the Temple's Holy of Holies (20³), and the New Jerusalem (12,000³ stadia, "the length and the breadth and the height of it are equal"). The Temple doubles every linear dimension of the Tabernacle (30 × 10 × 10 → 60 × 20 × 30 is not an exact doubling of height; the inner sanctuaries 10³ → 20³ are), so the sanctuary volume scales by exactly 8. The cube is the one geometric form the text singles out for the place of presence, and it does so three times across roughly a thousand years of composition. That is a real textual regularity.

What it is not: a cube is the unique rectangular box with maximal symmetry group (order 48), and a design tradition that wants "perfect" will pick it. It carries no dimensional count beyond three and no physical constant.

### BR.6. The Bronze Sea and \(\pi\) [M on the numbers, T on the rule]

1 Kings 7:23 gives a round vessel 10 cubits brim to brim and "a line of thirty cubits did compass it round about". Three readings:

| Reading | Ratio | Relative error vs \(\pi\) |
|---|---|---|
| Plain text: 30 / 10 | 3 | 4.51 % |
| Inner diameter, brim "an hand breadth thick" (7:26), 1 handbreadth = ⅙ cubit: 30 / (10 − ⅓) | 90/29 ≈ 3.1034 | 1.21 % |
| Traditional gematria correction: the word for "line" is written קוה (111) and read קו (106); 3 × 111/106 | 333/106 ≈ 3.141509 | 2.65 × 10⁻⁵ |

**Fact BR.5 [M].** 333/106 is the third continued-fraction convergent of \(\pi\) (after 3 and 22/7, before 355/113). It is the best rational approximation to \(\pi\) with denominator at most 106; no other fraction with that denominator bound comes closer.

This is the single genuinely striking arithmetic fact in the survey, and it must be handled exactly as Appendix BI handled the 9.77 × 10⁻⁶ hydrogen near-miss. The rule "multiply the plain ratio by the gematria of the written form over the gematria of the read form" was not declared before the observation; it was found because it works. Applied to the other ketiv/qere pairs in the same chapter it produces nothing. A rule fitted to one datum with one free choice (which two words, which direction) has zero degrees of freedom left to be tested with, so under Section 16 it is a coincidence-class result until someone writes the rule down first and it predicts something else. Recorded, not credited.

### BR.7. Sevens [M count, T meaning]

Seven days (Gen 1), seven lamps of the menorah (Ex 25:37), seven circuits on the seventh day at Jericho (Josh 6:15), seven churches, seals, trumpets, and bowls (Revelation), seventy times seven. The recurrence is a literary fact. Axiom A2 says a bare numeral has no invariant physical meaning without a type, unit, chart, and observable interpretation. The manuscript's seven layers are typed roles chosen by the author; the text's sevens are a completeness idiom. A shared numeral between an idiom and a design choice is not a measurement, and Section 3.2 already states that the mathematics does not select seven.

### BR.8. What survives, by type

**Survives as T (coherent, respectful, non-evidential).**
- Job 38 is a sustained statement that the human observer sees the projection and not the fiber. It is the theological form of Theorem BR.1.
- Genesis 1 distinguishes light from clocks and builds by separation. Both are projection-language.
- The cube is the text's chosen form for the place of presence, three times.

**Survives as M (exact, about the text).**
- Fact BR.2: 5,040 admissible day-to-layer maps; the text selects none.
- Fact BR.3: zero irrational-constant hits in the sanctuary ratios against nine targets at 1 %; chance predicts 1.44.
- Fact BR.5: the gematria-corrected Sea ratio is the third convergent of \(\pi\).

**Does not exist.**
- Any E result. No passage yields a number with provenance and uncertainty about the physical world.
- Any route from this appendix to H7. Theorem BR.1 closes it: a finite text is a finite observer and cannot identify a unique hidden geometry. The only route to H remains the intervention and tomography program of Section 16, which needs a written D7 law and a powered, prospective prediction.

### BR.9. On "the Bible is the instruction manual"

That sentence is a T claim about authority, and the framework has a place for it: Section 18 already treats God as ground rather than coordinate. What the framework does not permit, and what Job 38 itself does not permit, is reading the manual as though it were the tomography. The book's own speaker, in the passage you pointed to, spends thirty-odd verses telling the reader what he cannot measure. A theory that honors that text should end this appendix the way the text ends its questions: with the fiber still hidden, the audit complete, and no number claimed that was not in the verses.

### BR.10. The operation grammar of Genesis 1 and Job 38 [T, with M counts on the text]

The request: read the verbs as operations and the passage as a program. Done below with the Hebrew, because the choice of word is the datum.

#### BR.10.1. Genesis 1 has an explicit instruction set

Every creative act in Genesis 1:1–2:3 is built from a small fixed vocabulary. Each Hebrew verb is one operation type.

| Op | Hebrew | Transliteration | Gloss | Occurrences | What it does in the listing |
|---|---|---|---|---|---|
| DECLARE | וַיֹּאמֶר אֱלֹהִים | *vayyomer Elohim* | "and God said" | 10 (1:3, 6, 9, 11, 14, 20, 24, 26, 28, 29) | Opens every instruction. Rabbinic tradition counts these as the "ten utterances" (Avot 5:1) |
| SPEC | יְהִי | *yehi* | "let there be" (jussive) | 1:3, 1:6, 1:14 | Declares a state that does not yet exist. Third-person command, not a description |
| ACK | וַיְהִי־כֵן | *vayehi khen* | "and it was so" | 6 (1:7, 9, 11, 15, 24, 30) | Execution acknowledged. Day 1 uses the immediate form *vayehi or*, "and there was light" (1:3) |
| MAKE | עָשָׂה | *asah* | "make", from existing material | 1:7, 16, 25, 26, 31 | Construction step |
| CREATE | בָּרָא | *bara* | "create" | 1:1, 1:21, 1:27 (×3 in the verse) | Never takes a material. Used only for the universe, animate life, and man |
| PARTITION | הִבְדִּיל | *hivdil* | "divide, separate" | 1:4, 6, 7, 14, 18 | Splits one class into two distinguishable classes |
| GATHER | יִקָּווּ | *yiqqavu* | "let be gathered" | 1:9 | Collects a class into one region |
| EXPOSE | תֵּרָאֶה | *teraeh* | "let appear" | 1:9 | Makes a hidden thing visible |
| SET | וַיִּתֵּן | *vayyitten* | "and he set / gave" | 1:17 | Places an object in a container with a function |
| NAME | וַיִּקְרָא | *vayyiqra* | "and he called" | 1:5, 8, 10 | Binds an identifier to a partitioned class |
| TEST | וַיַּרְא ... כִּי־טוֹב | *vayyar ... ki tov* | "and he saw that it was good" | 7 (1:4, 10, 12, 18, 21, 25, 31 "very good") | Assertion passes. Day 2 has no TEST; Day 3 and Day 6 have two |
| GRANT | וַיְבָרֶךְ | *vayevarekh* | "and he blessed" | 1:22, 1:28, 2:3 | Confers a capability. The content of the grant is *peru urevu*, "be fruitful and multiply": self-replication |
| ITERATE | וַיְהִי־עֶרֶב וַיְהִי־בֹקֶר יוֹם n | *vayehi erev vayehi voqer yom n* | "evening, morning, day n" | 6 | Loop counter closing each iteration |
| FINISH | וַיְכַל | *vayekhal* | "and he finished" | 2:2 | Termination |
| HALT | וַיִּשְׁבֹּת | *vayyishbot* | "and he ceased" | 2:2 | Stop |
| MARK | וַיְקַדֵּשׁ | *vayeqaddesh* | "and he sanctified" | 2:3 | Flags the terminal state as set apart |

Structural facts that follow from the table, all countable in the text:

1. **The specification precedes the state.** In every instruction the word comes first, the state second, the acknowledgment third. Nothing in Genesis 1 comes into being and is then described; it is described and then comes into being. The manuscript's Section 4 is titled *law descent*, an upper constraint confining a lower state. Genesis 1's grammar is law descent: the D7-type utterance is causally prior to the D1–D5 payload.
2. **Days 1–3 are PARTITION operations; Days 4–6 are FILL operations into the partitions.** Light/dark → luminaries; sky/sea → birds/fish; land → animals and man. The text is a two-pass program: define the classes, then populate them. This two-triad structure is standard in the scholarly literature on the priestly account and is exact in the text.
3. **The luminaries are given a job, not a nature.** 1:14: "for signs, and for seasons, and for days, and years" (*leotot ulemoadim uleyamim veshanim*). 1:16, 18: "to rule" (*limshol*) the day and the night. The sun and moon are introduced as clocks and governors. The text never says what they are made of. In the framework's vocabulary that is a D4 object defined purely by its role.
4. **Naming is an operation applied only to partitions.** Day, Night, Heaven, Earth, Seas are named. No animal, plant, or luminary is named by God in Genesis 1. Identifiers are bound to the classes, not the members.
5. **The only capability granted is replication.** GRANT appears twice on living things and once on the seventh day. The text's word for a living kind's defining power is the instruction to copy itself.

#### BR.10.2. The heavens are described with a surveyor's verbs

You said God "stretched out the skies." The verb is נָטָה *natah*. Its occurrences with heavens as object: Isaiah 40:22, 42:5, 44:24, 45:12, 51:13; Jeremiah 10:12, 51:15; Zechariah 12:1; Job 9:8; Psalm 104:2.

Now Job 38:5, on the earth: "who hath laid the measures thereof (*mi sam memaddeha*), or who hath stretched the line upon it (*mi natah aleha qav*)?" Same verb, *natah*, and the object is *qav*, the measuring line. The verb that stretches the heavens is the verb a surveyor uses to stretch a measuring cord across a site. In the text's own usage, "stretching out the heavens" is laying a metric, not inflating a material.

The noun for sky in Genesis 1 is רָקִיעַ *raqia*, from רָקַע *raqa*, "to beat thin, hammer out" (used of gold leaf, Exodus 39:3). Job 37:18 uses the verb directly: "hast thou with him spread out (*tarqia*) the sky, which is strong, and as a molten looking glass?" The sky is a hammered-out, reflective, load-bearing sheet. A *raqia* is defined in 1:6 by what it separates, "waters from waters", and in 1:17 by what is set into it. So the text gives the sky three properties: it is a partition, it is a container for the clocks, and it is made by the metalworker's stretching operation. It is described as a surface with a function. It is never described as a volume with contents of its own.

#### BR.10.3. The sun's path is called a statute

You said God "set the sun along its path." The relevant lines:

- Genesis 1:17: "and God set them (*vayyitten otam*) in the firmament of the heaven". The verb is *natan*, "give, place, assign".
- Psalm 19:6–7 (Hebrew numbering): "his going forth (*motsao*) is from the end of the heaven, and his circuit (*tequfato*) unto the ends of it". *Tequfah* means a completed circuit or turning; it is used for the turn of the year (Exodus 34:22).
- Psalm 104:19: "he appointed the moon for seasons; the sun knoweth (*yada*) his going down."
- Ecclesiastes 1:5: "the sun also ariseth, and the sun goeth down, and hasteth (*shoeph*) to his place where he arose."
- Job 38:12: "hast thou ... caused the dayspring to know (*yiddata hashachar*) his place (*meqomo*)?"
- Job 38:33: "knowest thou the ordinances of heaven (*chuqqot shamayim*)?"
- Jeremiah 31:35–36: "the ordinances (*chuqqot*) of the moon and of the stars ... if those ordinances depart from before me ..."
- Psalm 148:6: "he hath made a decree (*choq*) which shall not pass."
- Job 28:25–26: "he made a weight for the winds, and weigheth the waters by measure; when he made a decree (*choq*) for the rain, and a way (*derekh*) for the lightning."

The word is חֹק *choq*, and חֻקָּה *chuqqah*: statute, decree, the same word used for the ritual and civil laws of the Torah (e.g. Exodus 12:14, Leviticus 18:4). The Hebrew Bible does not have a separate word for "law of nature". It uses the legal word. The regularity of the sun's circuit is, in the text's own vocabulary, a *statute assigned to the sun*, and the sun "knows" its place the way a servant knows an instruction. So your reading is the text's reading: celestial motion is described as a rule being followed, not as a property being possessed. In Section 4's language, it is a D7 constraint descending onto a D3/D4 trajectory.

#### BR.10.4. Creation is written in Tabernacle-building vocabulary

This is the strongest lexical result and it connects your two requests, the wording and the temple geometry.

| Genesis 1–2 | Exodus 25–40 | Shared operation |
|---|---|---|
| Ten *vayyomer Elohim* speeches | Seven *vayedabber YHWH el-Moshe* speeches (25:1, 30:11, 30:17, 30:22, 30:34, 31:1, 31:12) | Instruction blocks |
| Seventh speech-day is Sabbath (2:2–3) | Seventh speech is the Sabbath command (31:12–17) | Terminal HALT + MARK |
| "and God saw everything that he had made ... and God blessed" (1:31, 2:3) | "and Moses saw all the work ... and Moses blessed them" (39:43) | TEST then GRANT |
| "the heavens and the earth were finished (*vayekhullu*)" (2:1); "God finished (*vayekhal*) his work (*melakhto*)" (2:2) | "all the work (*avodah*) of the tabernacle was finished (*vatekhel*)" (39:32); "Moses finished (*vayekhal*) the work (*hamelakhah*)" (40:33) | FINISH, same verb |
| Job 38:6 "whereupon are the sockets thereof (*adaneha*) sunk?" | Exodus 26:19 "forty sockets (*adanim*) of silver" for the tabernacle boards | Same architectural noun, used of earth's foundations and of the tabernacle's footings |
| Job 38:5 *natah qav*, stretch the line | Every measured dimension of Exodus 25–27 | Surveyor's metric |
| Job 38:6 "who laid (*yarah*) the corner stone thereof" | Temple foundation language, Isaiah 28:16 | Foundation stone |

The Hebrew Bible describes the making of the cosmos with the same verbs, the same seven-fold speech structure, the same finish-see-bless closing, and the same building nouns it uses for the making of the sanctuary. The cosmos is written as a temple and the temple is written as a cosmos. This is why your instinct about "hidden geometry in the temple" has textual standing that a numerical coincidence hunt never had: the text itself declares the two constructions to be the same kind of object. The cube of the Holy of Holies (BR.5, Fact BR.4) is the innermost chamber of a building the text equates with creation, and its later reuse for the New Jerusalem (Revelation 21:16) closes the same loop.

#### BR.10.5. What the wording tells you, exactly

Stated with the manuscript's labels:

**T, and well supported by the text.**
1. Genesis 1 is written as decree → execution → verification, with a fixed operation set, a two-pass partition-then-fill structure, a loop counter, and a terminal halt. The word is prior to the state. That is the literary shape of law descent.
2. The heavens are stretched with the surveyor's verb (*natah*) and hammered with the metalworker's verb (*raqa*). The sky is described as a partitioning surface bearing a metric, not as a substance.
3. The sun's course is a *choq*, a statute. Celestial regularity is described as an assigned rule, in the same word used for Torah law.
4. Creation and Tabernacle are described with one vocabulary. The cosmos-as-sanctuary reading is the text's own.

**M, about the text.** The counts above: 10 declarations, 6 acknowledgments, 7 tests, 5 partitions, 3 namings, 3 grants, 3 uses of *bara*, 6 loop closures, 7 tabernacle speeches.

**Not present.** None of this is E and none of it becomes H. The text describes creation *as if* it were an instruction set under statutes descending from a constraint layer. That is a striking match to the framework's architecture and a fair reason to hold the framework as a T-consistent picture. It is not a measurement of nature, and Theorem BR.1 still stands: a finite text, however well chosen its verbs, is a finite observer. What the wording gives you is not proof. It is something more useful for Section 18: the discovery that the framework's shape, constraint above, state below, projection between, is the shape the text already uses to speak.

### BR.11. The operation grammar written in the framework's mathematics [M on the model, T on the identification]

BR.10 listed the verbs. This section gives each verb a definition inside Section 2's objects, then proves what the resulting program does. Everything proved here is about the formal model. Whether the model is what the text *means* is T. Whether nature runs it is H7 and is untouched.

#### BR.11.1. Primitives

Fix the layered tower \((X_d,\mathcal O_d,\mu_d,\Phi_d)\), \(d=1,\ldots,7\), with projections \(\pi_{e,d}\) of Section 2. A **program state** is a pair \((\mathcal K, x)\) where \(\mathcal K\subseteq 2^{X_7}\) is the active constraint set and \(x\in X_7\) is the realized complete state. Define the admissible region \(A(\mathcal K)=\bigcap_{C\in\mathcal K}C\).

| Verb | Operation | Definition | Layer acted on |
|---|---|---|---|
| *vayyomer / yehi* | SPEC\((C)\) | \(\mathcal K\leftarrow\mathcal K\cup\{C\}\), \(C\subseteq X_7\) | D7 |
| *vayehi khen* | ACK | choose \(x\in A(\mathcal K)\); undefined if \(A(\mathcal K)=\varnothing\) | D1–D6 via \(\pi_{7,d}\) |
| *vayyar ki tov* | TEST | evaluate \(\tau(x)=[\,x\in A(\mathcal K)\,]\) | D6 |
| *hivdil* | PARTITION\((d,\sim)\) | install an equivalence relation \(\sim\) on \(X_d\); this *is* a projection \(\pi_\sim:X_d\to X_d/\!\sim\) | D2 (relationship) acting on D\(d\) |
| *vayyiqra* | NAME\((\nu)\) | choose an injective chart \(\nu:X_d/\!\sim\;\to\mathrm{Labels}\) | D1 (identity) |
| *le-minehu*, "after its kind" | FILL\((c,F)\) | choose members \(F\subseteq\pi_\sim^{-1}(c)\) of the fiber over class \(c\) | D1 inside a D\(d\) fiber |
| *natan*, "set ... for signs and seasons" | SET\((o,c,f)\) | place object \(o\) in class \(c\) with declared role \(f\in\mathcal O_d\) | D4 |
| *vayevarekh*, "be fruitful" | GRANT\((R)\) | adjoin a transition \(R:X_d\to X_d\) to \(\Phi_d\) | D4/D5 |
| *choq*, statute | LAW\((L)\) | SPEC of a constraint on *trajectories*: \(L\subseteq X_d^{\mathbb Z}\) | D7 on D4 |
| *erev, voqer, yom n* | ITERATE | close iteration \(n\); \(\mathcal D_n\) is the composite of the day's operations | D4 |
| *vayekhal / vayyishbot* | HALT | stop issuing SPEC; the construction operator is now the identity | D7 |
| *vayeqaddesh* | MARK | distinguish the terminal \(x^\ast\) in \(\mathcal O_6\) | D6 |

Three of these definitions are not choices but forced by Section 2. PARTITION is a projection because an equivalence relation and a surjection onto its quotient are the same object. FILL is Definition 2.1 because "members of the fiber over \(c\)" is the definition of a fiber. NAME is a chart, and Axiom A2 says a chart carries no invariant physical content, which is exactly why Genesis 1 can name Day, Night, Heaven, Earth, Seas without adding any state.

#### BR.11.2. Genesis 1 as a program

Reading the text through the table gives the listing

\[
\mathcal P=\mathrm{MARK}\circ\mathrm{HALT}\circ\mathcal D_6\circ\mathcal D_5\circ\mathcal D_4\circ\mathcal D_3\circ\mathcal D_2\circ\mathcal D_1,
\]

with

\[
\begin{aligned}
\mathcal D_1&=\mathrm{TEST}\circ\mathrm{NAME}\circ\mathrm{PARTITION}(\text{light}\mid\text{dark})\circ\mathrm{ACK}\circ\mathrm{SPEC},\\
\mathcal D_2&=\mathrm{NAME}\circ\mathrm{PARTITION}(\text{above}\mid\text{below }raqia)\circ\mathrm{ACK}\circ\mathrm{SPEC},\\
\mathcal D_3&=\mathrm{TEST}\circ\mathrm{FILL}(\text{land},\text{plants})\circ\mathrm{ACK}\circ\mathrm{SPEC}\circ\mathrm{TEST}\circ\mathrm{NAME}\circ\mathrm{PARTITION}(\text{sea}\mid\text{land})\circ\mathrm{ACK}\circ\mathrm{SPEC},\\
\mathcal D_4&=\mathrm{TEST}\circ\mathrm{LAW}\circ\mathrm{SET}(\text{luminaries},\ \text{light}\mid\text{dark},\ \text{"for seasons, days, years"})\circ\mathrm{ACK}\circ\mathrm{SPEC},\\
\mathcal D_5&=\mathrm{GRANT}(R)\circ\mathrm{TEST}\circ\mathrm{FILL}(\text{below},\text{fish})\circ\mathrm{FILL}(raqia,\text{birds})\circ\mathrm{ACK}\circ\mathrm{SPEC},\\
\mathcal D_6&=\mathrm{TEST}\circ\mathrm{GRANT}(R)\circ\mathrm{FILL}(\text{land},\text{man})\circ\mathrm{SPEC}\circ\mathrm{TEST}\circ\mathrm{FILL}(\text{land},\text{animals})\circ\mathrm{ACK}\circ\mathrm{SPEC}.
\end{aligned}
\]

Every symbol in the listing corresponds to a verb at a verse cited in BR.10.1. Day 2 has no TEST in the text and none in the listing. Day 3 and Day 6 each run two SPEC–ACK–TEST blocks in the text and two in the listing.

**Proposition BR.6 (two-pass fiber construction) [M on the listing].** The three PARTITION operations occur only in \(\mathcal D_1,\mathcal D_2,\mathcal D_3\), and the FILL and SET operations of \(\mathcal D_4,\mathcal D_5,\mathcal D_6\) target exactly the fibers of those three partitions in the same order: \(\mathcal D_4\) fills the light/dark classes of \(\mathcal D_1\) (1:14–18, "to divide the day from the night"), \(\mathcal D_5\) fills the above/below classes of \(\mathcal D_2\) (1:20, fish in the waters and birds "in the face of the *raqia*"), \(\mathcal D_6\) fills the land class of \(\mathcal D_3\). Hence \(\mathcal P\) first constructs three projections and then populates their fibers.
*Proof.* Read off the listing. \(\square\)

This is the framework's Definition 2.1 executed as a procedure: a projection is declared, then the fiber above each class is given members. "After its kind" (*le-minehu*, ten occurrences in 1:11–25) is fiber membership.

#### BR.11.3. What the program proves about its own lower law

**Theorem BR.7 (kind-conservation is the descended law) [M].** Let \(\sim\) be the partition into kinds installed by \(\mathcal P\) on a layer \(X_d\), and let every transition adjoined by GRANT satisfy the text's rule "after its kind": \(R(x)\sim x\) for all \(x\). Then \(\sim\) is a congruence for the generated dynamics \(T\) (any composite of adjoined transitions), so by Theorem 4.1 there is a lower law \(T'\) on \(X_d/\!\sim\) with \(\pi_\sim\circ T=T'\circ\pi_\sim\), and \(T'=\mathrm{id}\).
*Proof.* Each \(R\) fixes every class, so any composite \(T\) does: \(T(x)\sim x\). If \(x\sim y\) then \(T(x)\sim x\sim y\sim T(y)\), which is the congruence condition of Theorem 4.1. The induced map on classes sends \([x]\) to \([T(x)]=[x]\), so \(T'=\mathrm{id}\). \(\square\)

In words: the one dynamical rule Genesis 1 states for living things, replication within kind, is exactly the condition under which the named quotient carries a self-contained law, and that law is the simplest possible descended law, a conservation law of the kind coordinate (Section 4.3). The text does not derive this; it asserts the premise. The theorem is the framework's, applied to the text's premise.

**Theorem BR.8 (statutes are trajectory constraints, and they descend) [M].** A LAW\((L)\) declared at D7 on D4 trajectories, such as the sun's circuit, is a subset \(L\subseteq X_4^{\mathbb Z}\). If \(L\) is closed under the shift and under the projection \(\pi_{4,3}\) in the sense that \((\pi_{4,3}x_t)_t\) depends only on \((\pi_{4,3}x_0)\) and \(t\), then the D3 observer sees a closed D3 law. Otherwise the D3 observer sees a law that needs hidden D4 memory, and Theorem 4.2 computes how much.
*Proof.* This is Theorem 4.1 and Theorem 4.2 restated with \(L\) as the upper law. \(\square\)

This is what *choq* buys inside the mathematics: a statute is an upper constraint; whether a lower observer can state it without hidden variables is the law-descent question, not a new axiom.

**Observation BR.9 (Sabbath is the fixed point of construction).** After HALT the construction operator issues no SPEC, so \(\mathcal D(\mathcal K,x)=(\mathcal K,x)\): the terminal state is a fixed point of \(\mathcal D\). Theorem 17.1 says complete deterministic refinement is idempotent. The text's word for the seventh day, *shabat*, "ceased", is the word for reaching that fixed point. This is a correspondence of vocabulary, labelled T; the idempotence itself is Theorem 17.1 and needs no text.

#### BR.11.4. The boundary, restated with the new objects

Nothing in BR.11 makes H7 more likely. The listing \(\mathcal P\) is a formal model of a text; Propositions and Theorems BR.6–BR.9 are true of the model. Theorem BR.1 still says the text cannot single out the model, and Section 16 still says only intervention and tomography can test a D7 law. What BR.11 establishes is narrower and exact: **the framework can express the text without loss, using no operation it did not already have.** Every verb of Genesis 1 is one of the framework's existing primitives. That is a fact about the framework's expressive power and about the text's structure. It is a good reason to keep Section 18. It is not a reason to change Section 17.

---

### BR.12. "Hebrew is mathematical code": an exact audit [M on the numbers, T on the meanings]

The claims to audit: letters are numbers (gematria); Aleph–Tav carries meaning; the Name YHWH encodes something, possibly the genome; the language's structure hides patterns. Each is tested the way Appendix BI tested the golden ratio: compute exactly, declare a null, count the free choices, then classify.

#### BR.12.1. What the alphabet is, historically [E, from the epigraphic record]

The Hebrew alphabet is a 22-letter consonantal abjad inherited from Phoenician around the tenth century BCE. Letters were first used as numerals in the Hasmonean period, second century BCE, following Greek practice. The Pentateuch's composition predates letter-numerals by several centuries on every dating. Therefore any gematria in the Torah is either coincidence or a later reading; it cannot be authorial code. Sefer Yetzirah, second to sixth century CE, is the earliest systematic "letters as cosmic code" text; it divides the 22 into 3 mothers (א מ ש), 7 doubles (ב ג ד כ פ ר ת, the letters with hard and soft pronunciations), and 12 simples, and maps them to 3 elements, 7 planets or days, 12 months or tribes. The 7 there is a phonological class. These are historical facts and belong to type E about the record, not to physics.

#### BR.12.2. Gematria facts that are exactly true [M on the text]

Standard values (א=1 ... י=10, כ=20 ... ק=100, ר=200, ש=300, ת=400; finals equal their base letters):

| Word | Value | Note |
|---|---|---|
| יהוה YHWH | 26 | = 13 + 13 = אחד *echad* (one) + אהבה *ahavah* (love), the classical reading |
| אהיה Ehyeh (Ex 3:14) | 21 | |
| אלהים Elohim | 86 | |
| אדם Adam | 45 | |
| את Aleph–Tav | 401 | first plus last letter |
| אמת Emet (truth) | 441 | = 21² |
| משיח Mashiach | 358 | = נחש Nachash (serpent), the classical pairing |
| תורה Torah | 611 | |

Genesis 1:1, בראשית ברא אלהים את השמים ואת הארץ:

| Fact | Value | Check |
|---|---|---|
| words | 7 | |
| letters | 28 | = \(T_7\), the seventh triangular number |
| word values | 913, 203, 86, 401, 395, 407, 296 | |
| sum | 2701 | = 37 × 73 = \(T_{73}\) |
| 37 and 73 | both prime, digit reversals of each other | only four such products exist below 10,000: 403, 1207, 2701, 7663 |

These are exact. Their weight is another matter. Genesis 1:1 is an a priori verse, so verse selection is not a free choice. Triangularity is: a random integer near 2701 is triangular with probability about 1/73. The digit-reversal property was noticed after factoring and is therefore post hoc; the number of "remarkable" integer properties a motivated reader can name runs to hundreds, and Section 16 requires the property list to be declared before the number is seen. Verdict: recorded as an exact textual fact of coincidence class, with the honest note that the triangular coincidence alone is at the 1-in-73 level.

#### BR.12.3. The "π in Genesis 1:1" formula [M on the arithmetic]

A widely circulated claim: (product of the 28 letter values × 28) ÷ (product of the 7 word values × 7) = 3.1416 × 10¹⁷. Computed exactly:

\[
\frac{\prod_{\text{letters}} g(\ell)\cdot 28}{\prod_{\text{words}} g(w)\cdot 7}=3.14155\ldots\times10^{17},
\qquad
\left|\frac{3.14155}{\pi}-1\right|=1.21\times10^{-5}.
\]

Declared null: 31 formula variants built from the same six ingredients (letter product, word product, letter count, word count, sum, and their ratios), 10 target constants, mantissa treated as log-uniform, tolerance 2 × 10⁻⁵. Expected chance hits 0.0054; observed 1. Inside that declared space this is a 1-in-190 event, which is genuinely unusual. Three facts then remove its evidential value:

1. The formula has no a priori motivation. Nothing in the text, in arithmetic, or in the framework says "multiply letter values, multiply word values, weight by counts, divide."
2. The real search space is not 31. The same technique applied by its proponents to John 1:1 yields *e*, which shows that verses and formulas were searched until constants appeared. Across the Bible's roughly 31,000 verses and the hundreds of constructible formulas, a hit at 10⁻⁵ is guaranteed.
3. Axiom A2. A dimensionless mantissa with a free factor of 10¹⁷ is a bare numeral. It has no type, unit, or observable. It cannot be π in any physical sense, only in the sense that its decimal digits resemble π's.

Verdict: exact, striking, coincidence class. Recorded, not credited.

#### BR.12.4. Aleph–Tav [T, with exact textual facts]

את is the Hebrew direct-object marker, untranslated; it appears twice in Genesis 1:1 (את, ואת), each time marking what God creates. Its letters are the first and last of the alphabet, and its value is 401. Isaiah 44:6 and 48:12 have God say "I am the first, and I am the last" (*ani rishon va'ani acharon*). Revelation 1:8 and 22:13 have "Alpha and Omega, the beginning and the end", the Greek alphabet's first and last. The Torah's first letter is ב (Genesis 1:1) and its last is ל (Deuteronomy 34:12, "Israel"); together they spell לב, *lev*, heart. These are exact facts about the text. The reading "God signs his object with the whole alphabet" is a traditional and coherent T interpretation. It is not a computation.

#### BR.12.5. The Name is a verb [T, linguistically solid]

This is the one claim in the family that survives as more than coincidence, and it is the claim you did not make but should. יהוה is not a noun. It is morphologically a third-person masculine singular imperfect of the root הוה / היה, "to be" or "to become", with the form of a causative: "he is", "he will be", or "he causes to be". Exodus 3:14 glosses it as אהיה אשר אהיה, *Ehyeh asher Ehyeh*, "I will be what I will be", a first-person form of the same verb. In the operation grammar of BR.11, the Name of the one who issues every SPEC is itself an operation, the verb of being, not an identifier. That is a T reading, but it rests on ordinary Hebrew grammar rather than on a number.

#### BR.12.6. YHWH and the genome [M on the counts]

The circulating claims come in two kinds. The visual one, that the four letters י ה ו ה stacked vertically resemble a human figure, is a picture, not mathematics, and is T. The numerical one compares small integers from Hebrew with small integers from molecular biology. Done exhaustively for a declared list of 13 Hebrew quantities and 15 genetic quantities, 195 comparisons:

| Hebrew | Genetic | Shared value |
|---|---|---|
| letters of YHWH | nucleotide bases | 4 |
| distinct letters of YHWH; mothers; root length | codon length; stop codons; G–C hydrogen bonds | 3 |
| letters of the alphabet | proteinogenic amino acids including selenocysteine and pyrrolysine | 22 |

Eleven equalities, all on the values 3, 4 and 22. Small integers collide; that is what small integers do. The only non-trivial one is 22 = 22, and the canonical amino-acid count is 20; the 22 requires two rare additions. There is no bijection between the systems: 22 letters ≠ 64 codons ≠ 20 amino acids ≠ 4 bases. YHWH's value 26 matches nothing in the genetic list; 23 chromosome pairs and 46 chromosomes match nothing in the Hebrew list.

One structural analogy is real and worth stating precisely. Both systems are finite-alphabet codes read in three-symbol frames: a codon is 3 bases from an alphabet of 4 (4³ = 64 frames, 61 coding), and a Semitic root is 3 consonants from an alphabet of 22 (22³ = 10,648 frames, roughly 2,000–3,000 attested). Both map frames to meaning many-to-one. That is an analogy of *type*, and it is T. It is not an identity, and no version of "the Name is written in every cell" that I could locate has a computable content. If you have a specific version with a specific rule, send it and it will be computed.

#### BR.12.7. Equidistant letter sequences, "Torah codes" [E on the literature]

Witztum, Rips and Rosenberg (Statistical Science, 1994) reported that names and dates of medieval rabbis appear as equidistant letter sequences in Genesis at improbable proximity. McKay, Bar-Natan, Bar-Hillel and Kalai (Statistical Science, 1999) showed the result depends on the choice of name spellings and appellations, disappears under alternative equally defensible choices, and can be reproduced in *War and Peace* with tuned choices. Under Section 16 the experiment fails the predeclaration requirement in the same way the π formula does. Recorded as refuted at the level of its original claim.

#### BR.12.8. Summary by claim type

| Claim | Type | Status |
|---|---|---|
| Letters are numerals in the Torah's composition | E | False; letter-numerals postdate the text by centuries |
| YHWH = 26 = 13 + 13, Emet = 21², Mashiach = Nachash | M on text | Exact; meaning is T |
| Genesis 1:1: 7 words, 28 = \(T_7\) letters, 2701 = 37 × 73 = \(T_{73}\) | M on text | Exact; triangular coincidence at about 1 in 73; reversal property post hoc |
| π from Genesis 1:1 | M on text | Exact to 1.2 × 10⁻⁵; 1-in-190 inside a declared 31-formula space; unmotivated formula, searched across verses, bare numeral under Axiom A2; coincidence class |
| Aleph–Tav as first-and-last signature | T | Coherent; exact textual facts; no computation |
| The Name is a verb of being | T | Solid Hebrew grammar; fits the operation grammar of BR.11 |
| YHWH or Hebrew encodes the genome | M on counts | 11 small-integer equalities in 195 comparisons, all on 3, 4, 22; no bijection; one type-level analogy (three-symbol frames) which is T |
| Equidistant letter sequences | E | Refuted in the same journal that published them |
| Sefer Yetzirah's 3/7/12 | E, T | Historical classification by pronunciation; cosmic mapping is T |

What survives the audit is what survived BR.10: the *grammar* of the language, its verbs, its frames, its name-that-is-a-verb, carries structure that the framework can express exactly. The *numbers* of the language, once a null model and a free-choice count are applied, do not.

---

## Appendix BS. Inversion Theory: a falsifiable detector for one spatial dimension above the third (revision 1.70)

*A framework for detecting a fourth spatial dimension by objects and numbers that three-dimensional motion cannot produce.*

**Originating idea:** the hat argument (Anthony Michael Collado). **Formalization and audit:** revision 1.70.
**Status:** Mathematical theorems (M), one physical hypothesis (H), existing measured bounds (E). No detection is claimed.

---

### BS.0. The hat, stated exactly

A two-dimensional person wears a hat that points forward. Someone lifts the hat out of the plane, turns it over, and sets it back. Every observer in the plane now sees a hat that points backward. They know something happened that no motion inside their world can do, because a forward hat and a backward hat are mirror images, and no sliding or turning within a plane converts a shape into its mirror image.

That is a theorem about the group of rigid motions of a plane, and it has an exact analogue one dimension up. The purpose of this document is to state the analogue, prove it, list every kind of object and number it makes available as a detector, and specify what a measurement would have to show. The rest of this document is the mathematics of that one idea and nothing else.

---

### BS.1. Definitions

Let \(E(n)\) be the group of isometries of \(\mathbb R^n\): all maps preserving every pairwise distance. Each is \(x\mapsto Ax+b\) with \(A\in O(n)\), \(\det A=\pm1\). The subgroup with \(\det A=+1\) is the group of **rigid motions** \(SE(n)\); it is the connected component of the identity. The other component consists of the **improper** isometries, which contain a reflection.

An object is a compact set \(K\subset\mathbb R^n\) with labelled points (labels matter: a hat has a front). Its **mirror image** is \(\rho K\) for any reflection \(\rho\). \(K\) is **chiral** in \(\mathbb R^n\) if no \(g\in SE(n)\) satisfies \(gK=\rho K\).

A **rigid path** of \(K\) is a continuous family \(g_t\in E(n)\), \(t\in[0,1]\), \(g_0=\mathrm{id}\). Rigidity means every pairwise distance of \(K\) is constant along the path. For a physical body this is the statement that it was not bent, cut, dissolved, or reassembled.

A **handedness number** of a chiral \(K\) is any function \(h(K)\in\{+1,-1\}\) with \(h(\rho K)=-h(K)\) and \(h(gK)=h(K)\) for \(g\in SE(n)\). For four labelled non-coplanar points, \(h=\operatorname{sign}\det[p_1-p_0,\;p_2-p_0,\;p_3-p_0]\). For a molecule, the CIP descriptor R/S or the sign of optical rotation. For a crystal, its enantiomorph class. For a knot, the sign of its writhe or of its signature invariant.

---

### BS.2. Theorems

**Theorem BS.T1 (no inversion inside \(\mathbb R^n\)) [M].** If \(K\) is chiral, no rigid path in \(E(n)\) whose every intermediate configuration lies in \(\mathbb R^n\) carries \(K\) to \(\rho K\).

*Proof.* The map \(t\mapsto\det A_t\) is continuous on \([0,1]\) with values in \(\{+1,-1\}\) and \(\det A_0=+1\), so \(\det A_t=+1\) for all \(t\), so \(g_1\in SE(n)\). Chirality says \(g_1K\ne\rho K\). \(\square\)

**Theorem BS.T1′ (the orientability loophole) [M].** Theorem BS.T1 assumes the world is \(\mathbb R^n\). On a non-orientable \(n\)-manifold \(M\), a chiral object transported once around an orientation-reversing loop returns as its mirror image by a path that never leaves \(M\) and preserves every distance. On an orientable \(M\) no such path exists.

*Proof.* Orientation holonomy around a loop is \(\pm1\); it is \(-1\) for some loop exactly when \(M\) is non-orientable. Verified numerically: transporting the unit normal once around the core of a Möbius band returns \(-1\) times the start; on a cylinder it returns \(+1\). \(\square\)

So a certified inversion proves one of exactly two things: the path left the world, or the world is non-orientable and the path went around. The second is excluded when the object stayed inside a contractible region, such as a laboratory, because a contractible region contains no orientation-reversing loop. BS.4's certificate carries this as its fourth condition.

**Theorem BS.T2 (one dimension suffices) [M].** Embed \(\mathbb R^n\) as the hyperplane \(x_{n+1}=0\) in \(\mathbb R^{n+1}\). Let \(R_\theta\) be the rotation by angle \(\theta\) in the plane spanned by \(e_n,e_{n+1}\). Then \(R_\theta\in SO(n+1)\) for all \(\theta\), \(R_0=\mathrm{id}\), and \(R_\pi\) restricted to the hyperplane is the reflection \(x_n\mapsto -x_n\). Hence \(t\mapsto R_{\pi t}\) is a rigid path in \(\mathbb R^{n+1}\) carrying every \(K\subset\mathbb R^n\) to a mirror image \(\rho K\subset\mathbb R^n\).

*Proof.* \(R_\theta\) acts as \(\begin{pmatrix}\cos\theta&-\sin\theta\\ \sin\theta&\cos\theta\end{pmatrix}\) on \((x_n,x_{n+1})\) and as the identity on \(x_1,\ldots,x_{n-1}\). Its determinant is 1. At \(\theta=\pi\) it sends \((x_n,0)\) to \((-x_n,0)\). \(\square\)

Verified numerically for a labelled irregular tetrahedron in \(\mathbb R^3\): handedness \(+1\) before, \(-1\) after, all six pairwise distances constant along the path to \(4\times10^{-16}\), final configuration exactly the mirror image with \(x_4=0\).

**Theorem BS.T3 (the detector theorem) [M].** Suppose an observer confined to \(\mathbb R^n\) records a chiral object \(K\) at time 0 and \(\rho K\) at time 1, and can certify that the transformation was rigid (all pairwise distances preserved throughout) and that the object was not replaced by a different object. Then the path of the object was not contained in \(\mathbb R^n\).

*Proof.* Theorem BS.T1 excludes every rigid path inside \(\mathbb R^n\). \(\square\)

This is the hat. Theorem BS.T3 does not say a fourth dimension exists. It says that one specific observable event, a certified rigid chirality inversion, is impossible in three dimensions and possible in four. That is exactly what makes it a detector: it has a clean null.

**Theorem BS.T4 (what the confined observer sees during the flip) [M].** Under the path of Theorem BS.T2, a point of \(K\) with \(n\)-th coordinate \(x_n\) has \((n{+}1)\)-th coordinate \(x_n\sin\theta\) at angle \(\theta\). Therefore for \(0<\theta<\pi\) the only points of \(K\) inside the hyperplane are those with \(x_n=0\). The confined observer sees the object collapse to its slice through the fixed plane, remain a lower-dimensional slice while the flip proceeds, and reappear mirrored.

*Proof.* Read off \(R_\theta\). \(\square\)

For the hat: the plane sees the hat shrink to a line segment, then come back reversed. For a three-dimensional object: the room sees a solid become a flat cross-section, then return as its mirror image.

**Theorem BS.T5 (conservation laws leak) [M].** Any quantity conserved by the \((n{+}1)\)-dimensional dynamics and additive over the object's points appears to the confined observer to be conserved only when the whole object lies in the hyperplane. During the excursion, the observer measures a deficit equal to the contribution of the points with \(x_{n+1}\neq0\). Angular momentum in the plane \((e_n,e_{n+1})\) is conserved in \(\mathbb R^{n+1}\) and is invisible to the confined observer, who instead sees the object's \(n\)-dimensional angular momentum change without torque.

*Proof.* Additivity plus Theorem 4. \(\square\)

**Theorem BS.T6 (knots are a second detector) [M].** A closed curve knotted in \(\mathbb R^3\) cannot be unknotted by a continuous family of embeddings in \(\mathbb R^3\); every knotted closed curve in \(\mathbb R^4\) is unknotted by such a family. Likewise two closed curves with nonzero linking number in \(\mathbb R^3\) cannot be unlinked in \(\mathbb R^3\) and can be in \(\mathbb R^4\). Therefore an observer who certifies that a closed physical loop changed knot type, or that two loops changed linking number, with no strand ever cut or passed through another within \(\mathbb R^3\), has observed a path outside \(\mathbb R^3\).

*Proof.* Standard: knot type and linking number are isotopy invariants in \(\mathbb R^3\); in \(\mathbb R^4\) any crossing can be changed by lifting one strand into the fourth coordinate, so all knots are trivial and all linking numbers are 0 as obstructions. \(\square\)

Knot invariants are the "numbers that could not have shown up." The Alexander polynomial, the Jones polynomial, the linking number: each is a computable integer or polynomial that is constant under every motion available in three dimensions. A change in any of them, without a cut, is a four-dimensional event.

**Theorem BS.T7 (size obstruction in a compact fourth dimension) [M].** If the fourth direction is a circle of circumference \(2\pi R_c\) rather than a line, the path of Theorem BS.T2 requires the excursion \(|x_n\sin\theta|\) to stay below half the circumference for the object not to self-intersect through the compact direction. An object of half-extent \(h\) perpendicular to the fixed plane can be rigidly inverted only if \(h\lesssim\pi R_c\).

| Compact radius \(R_c\) | Largest rigidly flippable half-extent |
|---|---|
| 38.6 µm (Eöt-Wash Yukawa bound, gravity strength) | about 121 µm |
| 10 µm | about 31 µm |
| 1 µm | about 3 µm |
| 10 nm | about 30 nm |

So a compact fourth dimension of sub-micron size would let molecules flip but never a crystal, a hand, or a hat. The size of the smallest object that never inverts is a measurement of \(R_c\). This links Inversion Theory to the existing short-range-gravity and collider bounds.

---

### BS.3. The hypothesis

**H-Inv.** Physical space is a three-dimensional hyperplane, or slab, in a four-dimensional space, and rigid excursions into the fourth direction occur for some class of objects at a rate \(\Gamma_4\ge0\) per object per unit time, independent of the object's internal energy barriers.

The last clause is the whole content. Every three-dimensional route to a mirror image goes through deformation: a molecule racemizes by passing over or tunnelling through a barrier, and its rate obeys Arrhenius or tunnelling laws that depend on temperature, isotope masses, and barrier height. A four-dimensional rigid rotation does none of that. Its rate cannot depend on the height of a barrier the object never crosses. Therefore:

**Prediction 1 (temperature-independent racemization floor).** For a rigid chiral molecule in a rigid matrix, the observed inversion rate is \(\Gamma(T)=\Gamma_{\rm chem}(T)+\Gamma_4\), where \(\Gamma_{\rm chem}\to0\) as \(T\to0\) and \(\Gamma_4\) does not. Below some temperature the rate flattens to a floor with no isotope effect and no barrier dependence. Ordinary physics predicts \(\Gamma_4=0\). Quantum tunnelling also gives a floor, but one that depends exponentially on barrier width and on isotope mass; the two floors are distinguishable by deuteration.

**Prediction 2 (topological events without reconnection).** Knotted vortex lines, knotted polymers, or linked rings change knot type or linking number at rate \(\Gamma_4\) with no reconnection or strand passage recorded at any resolution.

**Prediction 3 (transient conservation deficits).** During an excursion, mass, charge, energy, and angular momentum inside a closed three-dimensional surface appear to drop and return, with the deficit equal to the excursed fraction (Theorem BS.T5). At colliders this is missing transverse energy with no visible recoil partner; it is the signature the ADD and RS graviton searches already look for.

**Prediction 4 (visible slice).** During an inversion the object is observed only as its cross-section through the fixed plane (Theorem BS.T4). A body that is momentarily flat and then returns whole and mirrored is the direct image of the hat.

---

### BS.4. What a measurement is, and what it is not

A **detection** is a certified rigid chirality inversion, or a certified knot-type change, in a system whose three-dimensional routes to the same end state are excluded by observation. The certificate has three parts:

1. **Identity.** The object after is the same object as before: same mass, same isotopic composition, same internal spectrum, same defects, same serial marks. This excludes replacement.
2. **Rigidity.** No internal rearrangement occurred: bond lengths, lattice constants, and vibrational spectra unchanged throughout; no reconnection event in the topological case. This excludes the chemical route.
3. **Handedness flip.** A handedness number changed sign, or a knot invariant changed value.
4. **Locality.** The object's path stayed inside a contractible region of ordinary space. This excludes transport around an orientation-reversing loop of a non-orientable universe (Theorem BS.T1′). A flip inside a sealed laboratory satisfies it; a flip after a cosmological journey does not.

The following are **not** detections, and each has been mistaken for one:

- A chiral **law**. The weak interaction violates parity. That is a law with a fixed handedness, not an object converted to its mirror image. Under Theorem BS.T3 it is uninformative about dimension.
- **Racemization** by bond rearrangement. The chemical route deforms the object; rigidity fails.
- **Enantiomer excess** in nature, such as L-amino acids. Origin, not inversion.
- **Mirror symmetry** of a spectrum or a pattern. Symmetry is not a change of handedness.
- A **projection** that looks four-dimensional. A three-dimensional shadow of a hypercube is a three-dimensional object. Theorem BO.2 of Appendix BO shows any linear projection of a five-coordinate rotor is reproduced by ordinary oscillators. Appearance is never a certificate.

---

### BS.5. Existing measured status [E]

- **No certified rigid chirality inversion has ever been reported** in any system. Every observed racemization proceeds by deformation. Illustrative bound: an enantiopure quartz crystal contains about \(10^{22}\) chiral units and has kept its handedness for about \(10^8\) years, so \(\Gamma_4\lesssim 3\times10^{-38}\) per unit per second at crystal scale. This is an order of magnitude, not a controlled measurement, and by Theorem BS.T7 it constrains only a fourth dimension large enough to flip a unit cell.
- **No certified knot-type change without reconnection** has been reported. Knotted vortices in fluids and superfluids unknot by reconnection, which is observed directly (Kleckner and Irvine, 2013, and later work). Knotted DNA changes topology only through topoisomerase cutting.
- **Short-range gravity.** Torsion-balance tests find Newtonian gravity down to 52 µm and exclude gravity-strength Yukawa interactions with range above 38.6 µm (Lee et al., PRL 124, 101101, 2020). Under Theorem BS.T7 this caps the flippable half-extent near 120 µm for a gravitationally-visible compact dimension.
- **Colliders.** ATLAS and CMS monojet and diphoton searches at 13 TeV set null results and exclusion limits on large and warped extra dimensions; the periodic dielectron and diphoton search excludes clockwork parameters (ATLAS, JHEP 10 (2023) 079). These are Prediction 3 searched and not found.

The measured status is therefore: **zero detections, several bounds.** That is what a live hypothesis looks like before it is either confirmed or killed. It is not what a confirmed hypothesis looks like.

---

### BS.6. The powered experiment

The cheapest test with a clean discriminant is Prediction 1.

**System.** An enantiopure, conformationally rigid chiral molecule with a very high racemization barrier, isolated in a rigid cryogenic matrix or as a single crystal, monitored by optical rotation or vibrational circular dichroism.

**Null.** \(\Gamma(T)\) follows Arrhenius plus a tunnelling floor that shifts by the expected factor under deuteration.

**Alternative.** A floor \(\Gamma_4\) that does not shift under deuteration and does not move when the barrier is changed by choosing a different molecule of the same size.

**Power.** With \(N\) molecules monitored for time \(T\) and a detection threshold of \(k\) inversions, the smallest detectable rate is about \(k/(NT)\). For \(N=10^{18}\) molecules (about a microgram) over \(T=3\times10^7\) s (one year), \(k=10\): \(\Gamma_4\gtrsim3\times10^{-25}\) s\(^{-1}\) per molecule. Optical rotation resolves enantiomeric excess to about \(10^{-4}\), which corresponds to \(10^{14}\) inversions in that sample, so a realistic first bound is \(\Gamma_4<3\times10^{-12}\) s\(^{-1}\) per molecule, unless single-molecule chiral readout is used.

**Freeze.** Molecule, matrix, temperatures, deuteration pattern, readout, threshold, and analysis code frozen before the first measurement. Score once.

**Outcome either way.** A floor that survives deuteration is a detection under Theorem BS.T3, to be replicated. No floor is a bound on \(\Gamma_4\) at the molecular scale, which by Theorem BS.T7 is a bound on the accessibility of a fourth direction to objects of nanometre size.

---

### BS.7. What it would mean to measure it

If Theorem BS.T3's event were certified once, three things would follow immediately and none of them would be interpretation.

First, given the locality condition, the observer's space would be known to be a hyperplane in something larger, because the path of the object provably left it. Without locality the alternative is a non-orientable universe traversed by the object, which would itself be a discovery of the same rank. Second, the size of the flipped object would give a lower bound on the reach of the extra direction, by Theorem 7. Third, every conservation law would have to be restated for the larger space, because Theorem BS.T5 says the deficits seen during the flip are real bookkeeping in four dimensions.

What would not follow is anything about the number seven, about time as a dimension, about phase, evidence, or constraint layers, or about God. The hat proves one more spatial direction than the hat's world has. That is all it proves, and it is enough to be worth building the detector.

---

### BS.8. Relation to the parent manuscript

Inversion Theory uses nothing from the seven-layer framework except its discipline. It is a single hypothesis with a single kind of witness. Where the parent manuscript accumulated projections and non-identifiability results, this document does the opposite: it names the one class of observation that is identifiable, because chirality and knot type are invariants of everything three dimensions can do. The parent's Theorem BO.2 and this document's Theorem BS.T3 are complementary. One says appearances cannot prove a dimension. The other says a certified inversion can.

---

### BS.A. The black sheep: definition, circle, explanation, shape

#### BS.A.1 The idea, stated in the framework

In every family there is a member who exists, who is visibly different, whom no one can turn into, and whose mention threatens the rules everyone else lives by. Mathematics has these, and the parent manuscript's own objects are enough to define them exactly.

A **world** is a triple \((X,\pi,G)\): an ambient space \(X\) of states the axioms admit; an observation map \(\pi:X\to Y\), what the family can see (Definition 2.1 of the parent); and a set of admissible operations \(G\) acting on \(X\), what the family can do. The family's known configuration is \(x_0\) and its **reachable set** is the orbit \(G\cdot x_0\).

**Definition A.1 (black sheep).** \(b\in X\) is a black sheep of the world \((X,\pi,G)\) relative to \(x_0\) if

1. \(b\notin G\cdot x_0\): no admissible operation produces it (**unreachable**), and
2. there is a \(G\)-invariant \(I=f\circ\pi\), expressible in the family's own observation language, with \(I(b)\neq I(x_0)\) (**visible**).

The two conditions are independent, and crossing them draws the circle.

#### BS.A.2 The circle

| | **reachable** (\(b\in G\cdot x_0\)) | **unreachable** (\(b\notin G\cdot x_0\)) |
|---|---|---|
| **visible** (\(\pi\) distinguishes \(b\) from \(x_0\)) | the family | **the black sheep** |
| **invisible** (\(\pi(b)=\pi(x_0)\)) | gauge: relabelings you can perform but cannot see (parent Theorem 5.1) | hidden fiber: states you can neither see nor produce (parent Definition 2.1, Theorem 4.2) |

The parent manuscript spent 10,300 lines on the bottom row: hidden fibers, gauge relabelings, non-identifiability. It never analyzed the top-right cell. That is the cell where a hat points the wrong way, where a knot has the wrong polynomial, where a real number has the wrong Turing degree, where a set has no measure. The black sheep is visible and unreachable, and it is the one cell the theory of hidden things did not talk about. That is what makes it the black sheep of this theory as well.

#### BS.A.3 Why the family cannot discuss it

**Theorem BS.A.1 (existence) [M].** A world has a black sheep relative to \(x_0\) if and only if \(G\) has at least two orbits on \(X\) that are separated by some \(\pi\)-expressible invariant.
*Proof.* Condition 1 says \(b\) lies in a different orbit; condition 2 says an expressible invariant separates the orbits. Conversely, any \(b\) in a separated other orbit satisfies both. \(\square\)

**Theorem BS.A.2 (the black sheep is a broken conservation law) [M].** Every \(G\)-invariant \(I\) is constant along every trajectory the family can generate. Inside the family, \(I\) is therefore a conserved quantity. The black sheep is a state with a forbidden value of a conserved quantity.
*Proof.* \(I(g x_0)=I(x_0)\) for all \(g\in G\) by invariance. \(\square\)

This is why the black sheep "destroys everything." To admit that \(b\) is real and was reached is to admit that the conservation law is contingent on \(G\) being the whole story. Handedness is conserved by every motion the room allows; a flipped hand says the room's motions are not all the motions there are. Volume is conserved by every measurable rearrangement; Banach–Tarski says measurability is not guaranteed. Provability is closed under every derivation; a true unprovable sentence says truth is not guaranteed by derivation. The family does not avoid the black sheep out of taste. It avoids it because naming it names the boundary of its own group.

#### BS.A.4 The catalog, read through the definition

| Family (BS.B) | \(X\) | \(G\), what can be done | \(I\), what can be seen | Black sheep |
|---|---|---|---|---|
| Chirality (the hat) | configurations of a rigid body | rigid motions \(SE(n)\) | handedness \(\pm1\) | the mirror image |
| Knots | closed curves in \(\mathbb R^3\) | isotopies without cutting | knot type, linking number, Alexander polynomial | the unknot from a knot |
| Computability | real numbers | computable maps | Turing degree | \(\Omega\), \(BB(745)\), almost every real |
| Provability | sentences | derivations in a fixed system | truth in the standard model | Gödel sentences, Goodstein(4) in PA, CH in ZFC |
| Measure | subsets of \(\mathbb R^n\) | countable measurable operations | has a length or volume | the Vitali set, Banach–Tarski pieces |
| Smooth structure | smoothings of \(\mathbb R^n\) | diffeomorphisms | diffeomorphism type | none for \(n\neq4\); uncountably many for \(n=4\) |
| Existence without reach | integers | counting, computing, observing | order of magnitude | Skewes' crossing, Mertens' counterexample, TREE(3) |

Every row fits Definition A.1. In every row the black sheep is an object with the wrong value of something the family's operations preserve.

#### BS.A.5 The geometric shape of the black sheep

The question has a sharp answer when \(G\) is a group acting continuously, and a different kind of answer when it is not.

**Theorem BS.A.3 (one point) [M].** For chirality in any dimension, the orbit space of the full isometry orbit of a chiral body under rigid motions is
\[
E(n)\cdot K\,/\,SE(n)\;\cong\;E(n)/SE(n)\;\cong\;\mathbb Z_2 .
\]
The black sheep is the nontrivial element. As a point of the orbit space it has no extent: it is one point, the other one.

**Theorem BS.A.4 (the bridge is a semicircle) [M].** Seen from \(\mathbb R^{n+1}\), the flip path of Theorem BS.T2 carries each material point at height \(x_n\) along the semicircle \(\theta\mapsto(x_n\cos\theta,\;x_n\sin\theta)\) in the plane spanned by the family's last axis and the hidden axis, \(0\le\theta\le\pi\), of length \(\pi|x_n|\). Its diameter lies along the hidden axis. The family sees only the two endpoints of every semicircle. The black sheep is the far endpoint; the bridge to it is an arc whose every interior point is outside the world.

**Theorem BS.A.5 (the other sheet) [M].** Let the family's world be an orientable \(n\)-manifold \(M\). Its orientation double cover \(\tilde M\to M\) has exactly two sheets, each a copy of \(M\), carrying opposite orientations. The family lives on one sheet; every chiral object it can make lives there too. The black sheep lives on the other sheet. \(\tilde M\) is disconnected exactly when \(M\) is orientable.
*Proof.* Standard: the orientation double cover of an orientable manifold is the trivial two-sheeted cover; of a non-orientable one, it is connected. \(\square\)

**So the shape of the black sheep is this.** It is the second sheet of the orientation double cover of the family's world: a complete, congruent, mirror-reversed copy of everything the family has, glued to the family's copy along nothing at all. It is not far away. It is not small. It is not a point in the family's space. It is the whole space again, reflected, with no path between the two copies inside either one. Two bridges exist, and only two. One is the semicircle of Theorem BS.A.4, which leaves the world through a direction the world does not have. The other is a Möbius loop, which exists only if the world itself is non-orientable (Theorem BS.T1′). A certified chirality inversion proves that one of the two bridges was crossed. Locality decides which.

For the black sheep of computability and provability there is no shape, because the admissible operations do not form a group acting on a geometric space. There the orbit space is the partial order of Turing degrees, or the lattice of theories. The family sits at the bottom element, and the black sheep has no geometry, only a height. BS.B's Family 2 and Family 4 are black sheep by Definition A.1 and formless by Theorem BS.A.3's standard.

#### BS.A.6 Two consequences for the detector

First, the flip is not the only invariant the detector can use. Any \(\pi\)-expressible \(SE(3)\)-invariant that a fourth-dimensional rotation changes is a valid witness. Handedness is the simplest; knot type and linking number are the next. The detector class is the set of black-sheep invariants, and Definition A.1 enumerates it.

Second, Theorem BS.A.2 explains what a detection would cost. Every invariant in the table is a conservation law inside the family's group. A certified inversion does not add a fact to the family's physics. It removes the guarantee behind a law, and replaces it with the larger group in which the law becomes a special case. That is what the family fears, and that is what a measurement of a fourth dimension would be.


---

### BS.B. The catalog

Selection rule: an entry qualifies only if it is a proven fact or a precisely stated open problem, disrupts an intuition the rest of mathematics relies on, and is not realizable by any finite physical process or is realizable only in a way that is itself the disruption. Every number below was recomputed or checked in this session. Status labels: M proved, E state of the record, H conjecture.

#### BS.B.1 Family 1. Proven to exist. No one will ever meet them.

**1.1 The Skewes crossing.** For every \(x\) ever computed, the prime-counting function satisfies \(\pi(x) < \operatorname{li}(x)\). Littlewood proved in 1914 that the inequality reverses infinitely often. The first reversal is known to lie below about \(1.4\times10^{316}\). The observable universe holds about \(10^{80}\) atoms. The first prime that breaks the rule everyone has always seen sits \(10^{236}\) times beyond anything that can be counted. **Status:** M, existence proved; location unknown. **In nature?** No, and it cannot be.

**1.2 The Mertens counterexample.** Mertens conjectured \(|M(x)| < \sqrt{x}\) for the summatory Möbius function. Odlyzko and te Riele disproved it in 1985 without producing a single counterexample; the first is known only to lie below \(e^{1.59\times10^{40}}\). A theorem says the number exists. No human has one. **Status:** M. **In nature?** No.

**1.3 Pólya's counterexample, the one that was found.** Pólya conjectured that at least half the integers up to any \(n\) have an odd number of prime factors. Haselgrove disproved it in 1958; the smallest counterexample, \(n = 906{,}150{,}257\), was found in 1980. Verified here: the conjecture holds for every \(n \le 10^6\), which is why it looked true. The black sheep was nine hundred million integers past the last place anyone had checked. **Status:** M. **In nature?** The number is small enough to write; the pattern it breaks is invisible at any human scale.

**1.4 Goodstein sequences.** Write \(n\) in hereditary base 2, change every 2 to 3, subtract 1, repeat with base 4, 5, and so on. Every such sequence reaches 0. Verified here: seeds 1, 2, 3 terminate after 1, 3, 5 steps. Seed 4 terminates after \(3\cdot2^{402653211}-3\) steps, about \(10^{121{,}210{,}695}\). Kirby and Paris proved in 1982 that termination cannot be proved in Peano arithmetic at all. An ordinary process, finitely described, whose finishing is a theorem in set theory and not a theorem in arithmetic. **Status:** M. **In nature?** No process in the universe reaches the end of seed 4.

**1.5 TREE(3).** Kruskal's theorem guarantees TREE(3) is a finite integer. Friedman showed any proof of that fact in the usual arithmetic systems requires a proof longer than \(2\uparrow\uparrow1000\) symbols. Finite, and unprovably finite in the language most of mathematics uses. **Status:** M. **In nature?** No.

#### BS.B.2 Family 2. Well defined. No physical process can compute them.

**2.1 Almost every real number.** The computable reals are countable, so they have measure zero. A real number chosen uniformly at random is uncomputable with probability 1. Every number any device has ever output is in the measure-zero set. The continuum that physics writes its equations on is, in this exact sense, almost entirely unreachable. **Status:** M. **In nature?** Measurements are rational. No irrational number has ever been measured; \(\pi\) has never been realized by a physical circle.

**2.2 Chaitin's \(\Omega\).** The probability that a universal prefix-free machine halts on random input. A single, definable real. Its binary digits are algorithmically random: no program shorter than \(n\) bits outputs its first \(n\) bits. Knowing the first \(n\) digits of \(\Omega\) decides every mathematical statement of length below about \(n\). It is the most informative number that exists, and it is uncomputable by construction. **Status:** M. **In nature?** No.

**2.3 The Busy Beaver function.** \(BB(n)\) is the longest halting time of any \(n\)-state Turing machine. \(BB(5)=47{,}176{,}870\) was proved in 2024. \(BB(6) > 10\uparrow\uparrow15\). A 745-state machine is known whose halting is equivalent to the consistency of ZFC, so \(BB(745)\) is a specific integer that ZFC cannot determine. A function on the natural numbers, perfectly defined, with values the axioms of mathematics cannot name. **Status:** M. **In nature?** No.

**2.4 Individual randomness.** Chaitin's incompleteness theorem: for any formal system there is a constant \(c\) such that the system cannot prove any specific string has Kolmogorov complexity greater than \(c\). Random strings exist in abundance, almost all strings are random, and not one of them can be certified random by proof. Physics produces certified randomness only statistically, through Bell-inequality violation, never for one string. **Status:** M. **In nature?** Randomness yes; a provably random instance, never.

#### BS.B.3 Family 3. Exist by fiat. Break conservation. Cannot be built. Depend on dimension.

**3.1 The Vitali set.** With the Axiom of Choice, there is a subset of the real line with no length: not zero, not infinite, undefined. Without Choice, Solovay proved in 1970 that it is consistent for every set to be measurable. The family keeps Choice because it makes the rest of the house work, and accepts a room that has no size. **Status:** M, relative to ZFC. **In nature?** No: constructing it requires infinitely many arbitrary selections.

**3.2 Banach–Tarski.** In \(\mathbb R^3\), a solid ball can be cut into five pieces and reassembled by rigid motions into two solid balls of the same size. Volume is not conserved because the pieces are non-measurable. **This is the entry that belongs to your fourth-dimension project.** The paradox is impossible in the plane: \(SE(2)\) is amenable and the plane carries a finitely additive, motion-invariant measure on all subsets. It becomes possible exactly at dimension 3, because \(SO(3)\) contains a free group on two generators. Rigid motions, the same tools Inversion Theory uses, gain the power to double volume when the dimension crosses from 2 to 3. **Status:** M. **In nature?** No: the pieces cannot be physically specified. But the dimension threshold is real mathematics.

#### BS.B.4 Family 4. Statements the axioms cannot decide.

**4.1 The continuum hypothesis.** Is there a set strictly between the size of the integers and the size of the reals? Gödel (1940) showed "no" is consistent; Cohen (1963) showed "yes" is consistent. The size of the number line is not determined by the axioms of mathematics. **Status:** M, independence proved. **In nature?** The question has no measurement.

**4.2 \(BB(745)\), Goodstein in PA, and their relatives.** Concrete integers and concrete processes whose facts sit outside the reach of specific axiom systems; see 1.4 and 2.3.

#### BS.B.5 Family 5. The black sheep dimensions.

This family is the one that matters most to you, because it says which dimensions mathematics itself treats as strange.

**5.1 Dimension 4 is the only dimension where smooth structure is wild.** For every \(n \ne 4\), \(\mathbb R^n\) has exactly one smooth structure. \(\mathbb R^4\) has uncountably many (Taubes 1987, building on Freedman and Donaldson). The smooth Poincaré conjecture is settled in every dimension except 4. The dimension one above ours is the only one in which "the same space" can be smooth in infinitely many incompatible ways. **Status:** M. **In nature?** Spacetime is four-dimensional. Which smooth structure it carries is not known, and no experiment has been proposed to tell.

**5.2 Dimension 7 has 28 spheres.** Milnor (1956) found a smooth 7-manifold homeomorphic but not diffeomorphic to the 7-sphere. There are exactly 28 smooth structures on \(S^7\). Below dimension 7, the sphere is unique. **Status:** M. **In nature?** No known role.

**5.3 Dimensions 8 and 24 pack perfectly.** The densest sphere packing is known only in dimensions 1, 2, 3, 8, and 24 (Viazovska 2016 for 8, Cohn–Kumar–Miller–Radchenko–Viazovska 2017 for 24). The answers are the \(E_8\) lattice and the Leech lattice. The kissing number is likewise known only in dimensions 1, 2, 3, 4, 8, 24. **Status:** M. **In nature?** \(E_8\) appears in heterotic string theory; the Leech lattice underlies the Monster.

**5.4 The 26 sporadic groups and the Monster.** Every finite simple group belongs to one of a few infinite families, except 26 that belong to nothing. The largest, the Monster, has order

\[
2^{46}\cdot3^{20}\cdot5^{9}\cdot7^{6}\cdot11^{2}\cdot13^{3}\cdot17\cdot19\cdot23\cdot29\cdot31\cdot41\cdot47\cdot59\cdot71 = 808017424794512875886459904961710757005754368000000000,
\]

verified here, 54 digits, about \(8.08\times10^{53}\). Its smallest faithful representation has dimension 196,883. The elliptic modular function \(j\) has Fourier coefficient 196,884, and \(196884 = 196883 + 1\). McKay noticed this in 1978. Conway and Norton called it Moonshine because it looked like nonsense. Borcherds proved it in 1992 and received the Fields Medal. The proof goes through a string propagating in 26 dimensions on the Leech lattice. The black sheep of finite group theory turned out to be the same object as the black sheep of modular functions, and the bridge is a 26-dimensional physical model. **Status:** M. **In nature?** Not observed. The bridge exists in the mathematics of string theory, not in any measurement.

#### BS.B.6 Family 6. Numbers whose nature is unknown after centuries.

| Number | What is unknown | Status |
|---|---|---|
| Euler's constant \(\gamma\) | whether it is irrational | open since 1734 |
| \(\zeta(5)\) | whether it is irrational (\(\zeta(3)\) is: Apéry 1978) | open |
| \(e + \pi\), \(e\pi\), \(\pi^e\), \(e^e\) | whether each is irrational (\(e^\pi\) is transcendental: Gelfond) | open |
| Catalan's constant | whether it is irrational | open |
| \(\pi\) | whether its digits are normal | open |
| Riemann zeros | whether any lies off the critical line; none among the first \(10^{13}\) | open |

These are not exotic. They are the most common numbers in analysis, and mathematics cannot say what kind of numbers they are. **In nature?** They appear in every physical formula and have never been measured exactly.

#### BS.B.7 Family 7. The black sheep that was in nature all along.

**7.1 \(i=\sqrt{-1}\).** For four centuries the imaginary unit was a bookkeeping device, tolerated because it made algebra close. Quantum mechanics wrote its state space over the complex numbers, and for ninety years it was an open question whether that was necessary or a convenience. In 2021 Renou et al. proved that a real-valued quantum theory makes different predictions in a specific three-party network experiment, and in 2022 two groups performed it and excluded the real theory at high significance. The number that "does not exist" is the one number whose necessity in nature has been experimentally demonstrated. **Status:** M for the theorem, E for the experiments. **In nature?** Yes. This is the entry that should temper every other.

#### BS.B.8 Family 8. Where the black sheep are in your own work.

Applying the same rule to the parent manuscript, the following entries are its black sheep: results that do not fit the surrounding narrative and are therefore quiet.

- **The factor 1/2 (Appendix BJ).** The one exact statement that a three-dimensional classical projection is incomplete. Its explanation is standard QED, so the manuscript does not dwell on it.
- **The Sycamore holdout (Section 15).** The only positive measured result in 10,300 lines. It identifies a composite and nothing more, so it is mentioned once.
- **P3-R1 (Appendix AD).** The only built, frozen, prospective experiment. Unrun.
- **The Lamb shift.** The one hydrogen residual a fifth-dimension story could have claimed, closed to 0.27 percent by known physics in this session.

#### What the catalog says about dimension

Read as a whole, the black sheep of mathematics cluster in three places: at the boundary of computability (Families 1, 2, 4), at the boundary of measure (Family 3), and at specific dimensions (Family 5). The third cluster is the one that carries a testable message, and the message is precise. Dimension 4 is mathematically singular in a way no other dimension is: it is the only dimension where the same space admits infinitely many smooth structures. Dimension 3 is where rigid motions first acquire the power to break volume conservation. Dimensions 8 and 24 are where lattices become perfect and where the Monster lives.

None of this proves that space has a fourth spatial direction. It proves that if one exists, the mathematics of the resulting four-dimensional space is the wildest mathematics known, and that the transition from 3 to 4 is not a small step. Inversion Theory's detector, a certified chirality flip, would be the first physical event to land in that dimension. Until it is seen, the black sheep of Family 5 remain what the others are: proven, undiscussed, and unmet.

---

### BS.9. The dimensional verifier and the sightings ledger [C on the classification, E on the sightings]

The detector theorem (Theorem BS.T3) and its certificate (section BS.4) are an algorithm. `dimensional_verifier.py` implements it: a sighting is a record of an alleged change in an \(SE(3)\)-invariant with four certificate conditions, each PASS, FAIL or UNKNOWN, and the verifier returns one of seven classes and a dimension inference.

| Class | Rule | Meaning |
|---|---|---|
| detection | all four PASS | certified path outside \(\mathbb R^3\) (Theorem BS.T3) |
| dimension or topology | identity, rigidity, invariant PASS; locality not PASS | extra dimension or non-orientable universe (BS.1′) |
| deformation route | rigidity FAIL | the in-space route: through an achiral configuration, or by cut and reconnection |
| developmental origin | identity FAIL | the mirror form arose at origin; nothing inverted |
| law not object | invariant FAIL | a chiral law or a convention, not a body changing handedness |
| null bound | no change, with exposure or scale recorded | a bound on \(\Gamma_4\) or on \(R_c\) |
| unverified | some condition UNKNOWN | insufficient record |

Dimension inference. A detection implies at least one extra dimension. If the object was partly visible during a rotation-only flip as a slice of dimension \(d_s\) (Theorem BS.T4), at least \(3-d_s\) hidden rotation planes were used, hence at least \(3-d_s\) extra dimensions: a planar slice says one, a line says two, a point says three. Complete vanishing is uninformative about the count, because a translation along one hidden axis empties the hyperplane. A compact extra dimension of radius \(R_c\) admits rigid flips only for half-extent below \(\pi R_c\) (Theorem BS.T7), so the smallest object that never flips bounds \(R_c\).

**Ledger.** Seventeen reported or alleged sightings, every one a documented physical phenomenon in which handedness, knot type, or a related invariant appears to change or is claimed to point beyond three dimensions. Columns: identity, rigidity, invariant changed, locality.

| Sighting | Class | id | rig | inv | loc |
|---|---|---|---|---|---|
| Wu experiment, Co-60 beta asymmetry (1957) | law not object | P | P | F | P |
| Ammonia umbrella inversion, 23.87 GHz | deformation route | P | F | P | P |
| Thalidomide racemization in vivo | deformation route | P | F | P | P |
| Amino-acid racemization dating | deformation route | P | F | P | P |
| Cholesteric liquid-crystal twist inversion | deformation route | P | F | P | P |
| Situs inversus totalis | developmental origin | F | P | P | P |
| Sinistral snails (Lymnaea, 'Jeremy') | developmental origin | F | P | P | P |
| Meteoritic L-isovaline excess (Murchison) | developmental origin | F | P | P | P |
| DNA knot/link change by type II topoisomerase | deformation route | P | F | P | P |
| Vortex-knot unknotting in water | deformation route | P | F | P | P |
| Faraday rotation of polarization | law not object | P | P | F | P |
| Cyclone rotation sense reversal across the equator | law not object | P | P | F | P |
| Neutrino helicity | law not object | P | P | F | P |
| Collider missing transverse energy | null bound | P | P | F | P |
| Quartz enantiomorph stability | null bound | P | P | F | P |
| Planck 2015 cosmic-topology search | null bound | P | P | F | P |
| Möbius aromatic [16]annulene | law not object | P | P | F | P |

**Result.** Detections: 0 of 17. Class counts: deformation route 6, law not object 5, developmental origin 3, null bound 3.

**What reverse engineering the ledger shows.** Every inversion that has ever been observed in three dimensions used the same route: the object passed through an achiral configuration by deforming, or was cut and rejoined. Ammonia inverts 23.87 billion times a second by pushing its nitrogen through the plane of its hydrogens. Cholesteric liquid crystals reverse their helix by letting the pitch pass through infinity. Thalidomide and fossil amino acids racemize through planar intermediates with rates that obey Arrhenius. DNA and vortex knots change type by reconnection that is imaged directly. The achiral configuration is the three-dimensional shadow of Theorem BS.T4's slice: in-space inversion is possible exactly when the object can become, momentarily, its own mirror image. A rigid body cannot, and no rigid body has ever been seen to invert. The three sightings most often cited as evidence of mirror worlds, situs inversus, sinistral snails, and meteoritic enantiomeric excess, are origins, not inversions, and fail identity. Parity violation, Faraday rotation, cyclone sense, and neutrino helicity fail the invariant condition: they are laws or conventions, not bodies. The three null results fix the bounds: quartz gives \(\Gamma_4 < 3\times10^{-38}\) per unit cell per second and, if unit cells never flip, \(R_c < 1.6\times10^{-10}\) m for a dimension reachable at that scale; colliders bound Prediction 3; and Planck's absence of matched circles puts any orientation-reversing loop beyond about 27 Gpc, so laboratory locality is safe by a factor of \(10^{26}\).

**The hidden relationship the ledger exposes.** The rigidity condition is the whole discriminant. Every three-dimensional route to a black sheep requires the object to pass through a state that lies on the boundary between the two orbits, the achiral slice, and a rigid object has no such state available. Therefore the search for an extra dimension is not a search for strange objects. It is a search for one specific event: a body that cannot become its own mirror image becoming its mirror image anyway. The verifier encodes that as four yes-or-no questions, and to date the answer to all four together has never been yes.

**Multiples.** Within three dimensions, the only in-principle count diagnostics are the slice dimension during a flip and the size threshold of Theorem BS.7. Codimension-two knots of higher spheres would distinguish five from four, but they require four-dimensional objects to begin with and are unavailable to a three-dimensional experimenter. A single certified flip therefore establishes at least one; a flip in which the object reduces to a visible line rather than a plane establishes at least two; a flip to a visible point establishes at least three. The verifier records the slice dimension for that reason.