# Element Projection Lab 0.10.0 — complete HEPData contour audit

## Result

The unchanged CW/LD benchmark `k=500 GeV`, `M5=6000 GeV` lies on the excluded
side of all four published observed contours. Each is a 95% CL result within
its own ATLAS channel and threshold scenario. This report does not calculate
a combined confidence level or claim independent experimental replication.

| Table | Channel | Mass thresholds | Interpolated M5 boundary (TeV) | Boundary minus 6 TeV (TeV) |
|---|---|---|---:|---:|
| 1 | dielectron | yes | 7.063338 | 1.063338 |
| 7 | diphoton | yes | 9.815289 | 3.815289 |
| 13 | dielectron | no | 8.735511 | 2.735511 |
| 19 | diphoton | no | 10.896422 | 4.896422 |

The shortest margin is already much larger than the rounding of the source
vertices. The displayed decimal places are interpolation bookkeeping, not
experimental error bars. This is an extension of the Table 1 negative result,
not new positive evidence for a dimension.

## Independent verification

The archive has 26 members: one directory and 25 YAML files, comprising
submission metadata and 24 contour tables with **1,978 coordinate pairs**.
All 24 curves have a unique intersection at `k=500 GeV`.

An independent exact-Fraction parser and polyline implementation reproduces
all 24 boundaries. The output comparison passes **349 checks**, including
original brackets, source identities, row counts, classifications and observed
margins. The maximum boundary discrepancy is `1.8189894035458565e-12 GeV`.
The independent implementation also passes nine synthetic geometry controls.

All **120 unit tests pass**, including 37 new archive, schema, interpolation,
ambiguity, domain and provenance tests and all 83 earlier tests. An initial
run made while the test file was still being completed is retained in
`outputs/development_pre_final_tests_0.10.txt`; its lone failure compared Python
tuples directly with JSON arrays. The finalized test checks the JSON wire
representation, and the complete rerun passes. No source data or physical
parameter was changed to obtain this pass.

Original observed brackets, ordered as `(M5,k)` in GeV:

| Table | First vertex | Second vertex | Exact linear boundary (GeV) |
|---|---|---|---|
| 1 | (6973.0, 489.1) | (7082.4, 502.3) | 4661803/660 |
| 7 | (9775.4, 488.9) | (10000.0, 551.4) | 30672778/3125 |
| 13 | (8378.4, 463.8) | (8758.2, 502.3) | 16815858/1925 |
| 19 | (10902.1, 488.9) | (10862.3, 566.7) | 4238708/389 |

There are eleven horizontal segments elsewhere in the published curves.
The engine preserves source ordering and refuses ambiguous slices; it does not
assume every contour is globally a single-valued function.

## Expected-limit diagnostics

All numbers below are interpolated M5 limits in TeV at the same k. The labels
are exactly the source's expected-limit labels, not sorted or renamed by size.
These twenty curves are **not** additional observed exclusions.

| Channel / thresholds | Expected median | +1 sigma | -1 sigma | +2 sigma | -2 sigma |
|---|---:|---:|---:|---:|---:|
| dielectron / yes | 7.366416 | 6.396706 | 7.669757 | 5.658974 | 7.912023 |
| diphoton / yes | 9.033179 | 7.984627 | 9.333616 | 7.267511 | 10.196563 |
| dielectron / no | 7.960965 | 6.996987 | 8.864655 | 6.239672 | 9.879461 |
| diphoton / no | 9.951347 | 8.753814 | 11.005492 | 7.785506 | 12.224840 |

## What the archive does and does not contain

The upload's SHA-256 is
`cdfa32efdd4415054aa55554b6e318132d2cde115ef7b19059094efe2f86407e`.
Its original bytes are preserved under `sources/`; the full per-file inventory
and hashes are in `outputs/atlas_contour_archive_results.json`.

The archive contains 103,527 uncompressed file bytes. Metadata references eight
image filenames and one external ATLAS auxiliary webpage; none of those images
is embedded. There are no event spectra, background-bin vectors, acceptance
grids, response matrices, covariance, classifier weights, or likelihood
workspaces in this archive. This is an inventory of this upload, **not** a
claim that these objects cannot exist elsewhere.

## Scientific interpretation and next test

This rules out the chosen benchmark under the specified published CW/LD
analyses. It does not invalidate the conditional tower arithmetic, exclude all
extra-dimensional models, or identify the cause of any discrepancy. The CW/LD
benchmark is a borrowed test model, not a derivation of the project's fixed
grid/tick or string-harmonic hypothesis. No coupling from those hypotheses to
ATLAS observables has been established by this lookup.

The finite program has now completed the promised four-scenario check. The
next bounded test is an explicitly exploratory map of the published parameter
domain, preserving folds, source coverage and the non-perturbative boundary.
Regions labelled not excluded must not be called confirmed. No replacement
benchmark is selected in this release, and any new physical prediction must
be frozen before genuinely fresh validation data are used.

Run `bash run_hepdata.sh` for the source check, production lookup, independent
audit, comparison, and targeted tests. `bash run_atlas.sh` runs the full lab.
This release retains the prior Table 1 protocol/results and packages the
previous Theory 1.70 as a history snapshot.

Sources: [HEPData record](https://doi.org/10.17182/hepdata.140955.v1),
[ATLAS paper, v2 dated 30 November 2023](https://arxiv.org/html/2305.10894v2),
and the user-provided original YAML archive. No new measurement is made here.