CHECKS.txt

Recorded local simulation evidence. This readable view displays the supplied source file; it does not verify authorship or production experience.

Worked rollout reference: verification record — 7 October 2026

EXECUTION
Six configurations were executed with preserved configs, reports, stdout,
stderr, commands, UTC timestamps, exit codes and SHA-256 values.
Order of checks: candidate_readiness / capacity_safety / rollback_rehearsal.
  baseline:          fail / fail / fail; exit 1
  readiness-only:    pass / fail / fail; exit 1
  final:             pass / pass / pass; exit 0
  wrong-port:        fail / pass / pass; exit 1
  rollback-disabled: pass / pass / fail; exit 1
  wrong-target:      pass / pass / fail; exit 1
The five failing configurations are expected negative evidence.

Eight lab behavior tests and five reproduction-runner tests passed; their
captured outputs and execution records are in evidence/. A separate Codex
agent reproduced all six reports and exit codes. This is an AI consistency
check, not independent human review or verified professional competence.

INTEGRITY AND FORMAT CHECKS
Six source hashes, 39 recorded evidence hashes and six config/report pairs
matched the actual files. The runner-test sidecar's two source hashes and two
log hashes matched; its original-manifest hash remained unchanged.
Both documents contain the same 40 material facts and 23 working source links.
Their lengths, repetition and organization differ; see STUDY.txt for limits.
The actual app schema accepts the sample: four requirements, three artifacts,
three claims, zero reviews and zero application observations. All five reviewer
checklist prompts are labelled not completed. Demo-selection flags disclose
Codex authorship and do not assert human confirmation.

The new showcase HTTP route test and the existing static-path allowlist test
passed together. Directory listing, path traversal and database paths remain
unavailable. The user's local service workspace was not replaced or imported.
JavaScript syntax, local HTML references and ZIP integrity are checked during
packaging. The package excludes SQLite databases and Python caches.

BROWSER VERIFICATION
Verified on the local server in the Codex in-app browser:
- Initial unselected chooser; either document can be read first and switched.
- Both documents have identical reading width and typography.
- Both rendered documents contain 23 shared links and disclosed authorship.
- All six run rows display the expected check results and exit codes.
- At a 390px viewport, chooser, both documents and evidence view have no page
  overflow; the wider run table scrolls inside its container.
- No warning/error console entries observed in this tested session.
Desktop and mobile views were visually inspected. This is not a comprehensive
accessibility audit. A raw report link did not confirm a browser disk download;
report serving is covered by HTTP tests and files are supplied in the ZIP.

SCOPE
Codex executed a deterministic Python simulation after reading the model,
tests and reference solution. No real cluster, traffic, production recovery,
unaided exercise by the profile owner, human pilot or hiring benefit is demonstrated.
Local hashes identify bytes and are editable, not tamper-resistant signatures.
See PROVENANCE.txt, evidence/manifest.json and STUDY.txt.