# Hawking Level 7 Results Intake

Intake version 0.1.0, prepared 7 November 2026.

Status is preparation for a pending evidence gate. This package implements the fixed
four-arm, 4,096-shots-per-setting protocol published on 7 November 2026. The supplied
count tables are explicitly synthetic software demonstrations.

Python 3.10 or later and the standard library are sufficient.

## Run the Demonstrations

```sh
python analyze_run.py synthetic-positive.json > positive-report.json
python analyze_run.py synthetic-dephased.json > dephased-report.json
python -m unittest test_analyze_run.py -v
```

The first fixture has ideal primary and SWAP correlations. The second has a
dephased primary and an ideal SWAP. Both reports keep Level 7 pending. A successful
program exit means that the input could be analysed. Scientific interpretation is
recorded separately in the report.

## Prepare a Real Run Record

Copy `run-template.json` into a dedicated run folder and complete its fields. The
template contains null measurement entries and requires real count tables before
analysis. Keep the raw records and supporting files together in that folder.

The protocol identifier is `h50-fixed-4096-v1`. The required arms are `exchange`,
`swap`, `qmm_output` and `global_inverse`. Each needs XX, YY and ZZ histograms with
exactly 4,096 raw integer counts, totalling 49,152 measurement shots before
calibration or diagnostics. Outcome keys are two-bit strings 00, 01, 10 and 11.
Zero-count outcomes may be omitted. Each character represents the declared
reference or output readout bit. Parity gives the same correlator in either bit
order. Preserve the actual register mapping in the layout record.

Record the provider, backend and job identifiers. Set each attestation only after
checking its basis in the protocol and execution records. A JSON boolean is a
declaration whose physical validity requires assessment.

## Evidence Files

Supply at least one file for each of these roles in the `evidence` list.

| Role | Required Content |
| --- | --- |
| protocol | Dated fixed analysis plan, decoder, output mapping, controls and exclusions |
| native_circuits | Executed circuits, gate parameters and a documented logical-equivalence check |
| layout | Logical-to-physical qubit and classical-bit mappings |
| calibration | Measurements and derivations supporting the bias bounds and their uncertainty |
| execution | Job identifiers, timestamps, shot ordering and raw export provenance |
| exclusions | All exclusions and diagnostics, including an explicit empty record where applicable |

Each record has `role`, `path` and `sha256` fields. Paths must be relative to the
run folder and resolve inside it. Generate a digest locally, for example with
`sha256sum evidence/calibration.json`, then copy the exact value into the record.
The program checks file identity. Reviewers assess the content, chronology,
native-circuit equivalence and provenance.

## Calibration and Confidence

Enter non-negative bounds for initial-reference preparation error `b_prep`, the
effective observable errors `epsilon_XX`, `epsilon_YY`, `epsilon_ZZ` and drift
`b_drift`. Observable bounds are in operator norm under the protocol's model.

The declared witness-bias budget is

```text
b = b_prep + (epsilon_XX + epsilon_YY + epsilon_ZZ)/4 + b_drift <= 0.02
```

Record `bound_failure_probability` for the joint validity of the calibration
bounds. Zero asserts a guaranteed bound and needs appropriate justification.
Point estimates alone do not supply the required error bounds.

The primary raw-shot test uses alpha 0.01, statistical margin approximately
0.0419162165 and the fixed decision threshold Fhat > 0.562. The report also gives
the union bound alpha plus the declared calibration-bound failure probability.
Overall confidence must account for both contributions. An overall 99% statement
requires a justified total error allocation, with any revision fixed prospectively.
The individual conditional intervals do not imply simultaneous confidence across
all arms. QMM-output and global-inverse arms remain comparisons.

## Readiness Checklist

1. Freeze the protocol and its scope before outcome-dependent decisions.
2. Bind the circuits to a selected device and document native equivalence, layout,
   timing, resource requirements and any authorised execution cost.
3. Establish calibration, preparation and drift bounds with uncertainty accounted for.
4. Execute the fixed protocol and preserve raw shots, controls and exclusions.
5. Collect the evidence files, hashes, attestations and completed run JSON.
6. Run `python analyze_run.py path/to/run.json > report.json` in a clean environment.
7. Reconstruct the report independently and assess the scientific result against
   the pending stage gate.

## Interpretation

Malformed histograms, missing settings, duplicate JSON keys, non-integer counts
and wrong shot totals are rejected. The program reports incomplete provenance,
unverified evidence, missing attestations and excessive calibration bias as
quality issues. A failed SWAP control blocks a positive primary interpretation.
A missed primary threshold is reported as recovery uncertified by the fixed rule.
Rejecting a candidate needs an appropriate uncertainty or equivalence analysis.

Even a complete record with a positive conditional witness leaves the milestone
at `pending_gate_assessment`. Inspecting raw-count structure and file hashes does
not establish physical calibration, independence of shots or authenticity. The
program assigns no development level. Synthetic fixtures and temporary test
records exercise software branches and provide no device evidence.

The analysis runs locally and makes no network calls. Provider connection,
calibrated execution and the gravitational interpretation remain separate tasks.