# P3-R1 Hardware Package Audit Report

**Release:** 1.0.0  
**Date:** 15 September 2026  
**Verdict:** Software release passes; physical experiment not run

## Scope

This report audits the P3-R1 single-device Born-outcome pilot package. It distinguishes four claims:

1. the probability construction is mathematically valid;
2. the two implementations agree on stored data;
3. the software fails closed under missing or altered provenance;
4. a physical QPU supplied evidence.

The first three pass within the tests below. The fourth is **not run**.

## Release checks

| Check | Result | Evidence |
|---|---|---|
| Python syntax | PASS | Production runner, independent verifier, and predecessor audits byte-compile |
| JSON syntax | PASS | Config and trial schema parse as JSON |
| Template structure | PASS | No structural errors; candidate grammar digest reproduces |
| Live readiness of shipped template | BLOCKED AS DESIGNED | Fresh UUID, explicit accessible backend, calibration envelope/artifact, and calibrated status are absent |
| Offline calibration-to-envelope reconstruction | PASS | Deterministic Hoeffding-plus-drift envelope rebuilds and freezes |
| Calibration binding | PASS | Config, copied calibration, and pre-run manifest digests agree |
| Prediction chronology | PASS | Every complete setting-conditional vector precedes every delayed setting in the audit ledgers |
| Eight-cycle marginal theorem | PASS | Run-specific maximum numerical error \(5.55\times10^{-17}\) |
| Randomized mathematical audit | PASS | 5,000 cases; maximum marginal error \(3.33\times10^{-16}\); no composite-null multiplier violation |
| Independent replay | PASS | No production function imported; scores agree |
| Commitment tamper | DETECTED | Modified commitment fails at commitment-ledger digest |
| Outcome tamper | DETECTED | Modified outcome fails at result digest |
| Score tamper | DETECTED | Modified score fails at result digest |
| Missing IBM dependencies | REFUSED AS DESIGNED | Hardware mode stops before access when Qiskit packages are absent |
| Real IBM job | NOT RUN | No credentials, accessible backend, calibration record, or QPU outcome was supplied |

The environment did not contain `jsonschema` or `ruff`; JSON parsing and Python byte-compilation passed, while those optional meta-validation/lint passes were not claimed.

## Frozen adversarial worlds

Each world uses 20,000 trials. Production and independent replay agree.

| World and seed | Max outcome log mixture | Outcome gate | Max setting log mixture | Setting gate | Correct classification |
|---|---:|---|---:|---|---|
| Born reference, 73021 | 0.279024506801 | NO CROSS | 4.631599926069 | NO CROSS, not equivalence | Yes |
| Planted forward lift, 73022 | 25.026590321580 | CROSS | 1.346171293282 | NO CROSS, not equivalence | Yes; synthetic target only |
| Lift plus setting dependence, 73023 | 18.835819634315 | CROSS | 1094.733928533300 | CROSS | Yes; disqualified |

The outcome threshold is \(\log(1/2.87\times10^{-7})\); the setting-leak threshold is \(\log(1/10^{-3})\). The synthetic verdict remains `SYNTHETIC_ONLY_NO_PHYSICAL_CLAIM` even when a planted process crosses.

## Independent mathematical checks

The verifier reconstructs, rather than trusts:

- the exact `Z_ALL` and `X_ALL` Born probabilities of the eight-phase state;
- the forward, reverse, and Gray winding maps;
- every probability transfer for every candidate, trial, and setting;
- candidate normalization, positivity, and eight-cycle marginal preservation;
- Kraft weights and retained cash mass;
- the prediction hash chain and config links;
- the rule that all predictions precede all delayed settings;
- setting/outcome joins, backend identity, and submission/retrieval order;
- the outcome and setting e-processes;
- the calibration-derived envelope for a physical record;
- agreement with the sealed production score.

The randomized audit additionally draws arbitrary positive eight-outcome baselines, transfer fractions, permutations, setting masks, composite-null laws, conservative upper envelopes, and normalized predictors. It found no violation in 5,000 trials. Finite randomized testing supplements the written algebra; it does not replace proof.

## Security and provenance boundary

The configuration stores no API key or token. Hardware authentication is delegated to the provider's saved-account mechanism. The local manifests detect changes relative to the manifest copies, but a controller of the host could rewrite every local artifact together. A court-grade run must publish or externally timestamp the pre-run manifest before QPU submission.

The provider path records job submission and result retrieval. It does not invent an authenticated device timestamp for each shot. One-shot PUB identity and the fact that every setting precedes submission are the available chronology evidence.

## Physical blockers and claim ceiling

The shipped configuration intentionally contains no live run identity, backend, or calibration. The current workspace also lacks Qiskit and Qiskit Runtime. No hardware connector or account was available, so no external mutation or QPU submission occurred.

P3-R1 is not spacelike separated. Its signal channel is therefore `NOT_TESTED_SINGLE_DEVICE`. It cannot adjudicate no-signaling, Bell locality, relativistic propagation, external replication, or universal predictability. A 256-trial pilot is also not a discovery-power calculation.

If a future physical outcome crossing survives the setting-leak tripwire, it licenses only:

> The declared single-device composite null was rejected, pending drift, memory, reset, reversal, causal, and independent-replication controls.

## Release digests

| SHA-256 | Artifact |
|---|---|
| `5e2c114908661482ded49526efc821e31988785f5bcd7e0cefd0d7644921e82b` | `p3_hardware_runner.py` |
| `e1b5bb72f70ca9c576e78b07a206682566291fb8dce884ef6bc1f3e870bf0d74` | `verify_p3_hardware_package.py` |
| `9f03fbb499c96d62b29eec34256de2d04a36dcd0bd50c277e07c1b853259aae5` | `p3_run_config.json` |
| `ceb54ab14f8e5eb17871280d7a620394fbdc45433a0ef6fb6dff7b84e9ff819e` | `p3_trial_schema.json` |
| `c94337709c96840dea2b8e401adf53c081bfa3b51470cebb915553d77895b261` | `P3_Hardware_Runbook_R1.md` |
| `ef2aa36ff943be9eeaf7ba55827f8a675d255a01f0c0f7ccd72151db26921e6b` | `Wallbreaker_Protocol_P3.md` |
| `9c556a0548e7929b3732aa76acefa864993f5b26b2a3ab35184ec1deaf8d6058` | `Dimensional_Reality_Court_Audit.md` |
| `e5d0675b7cc61ef08d658e1cf20a9d4f8cb2d4f696b544e6429b491c511df8f0` | `Theory_of_Reality.md` v1.29 |

These digests identify the audited local release. Library versions may carry storage metadata outside file contents; the content digest remains the comparison key.