# D30-F01-A03 validation evidence

Date: 25 October 2026. Status: **video-ready private draft**. No remote publication, manifest edit, deployment, recorded narration or encoded video is claimed.

## Executable acceptance

- Four Python suites passed:84 independently specified fixture cases, input immutability, whole-group receipt boundaries, checkpoint roundtrip/alias isolation and key-order determinism.
- Strict TypeScript compilation and matching fixture assertions passed.
- Complete JSON parity:84 cases, comparing exact balances, typed codes, nulls, complete coverage, ordered traces, checkpoint fields and full returned registry payloads.
- Canonical Ada 100000→97500→97600→97350; earlier E = 20October result 97600. The correction pair is absent between row recording clocks and appears together at receipt equality.
- Exact huge cash 9007199254740993 and maximum 32-leg aggregate 15999999999999999999999999999984 match independent arithmetic.
- High/low lone-surrogate correction reasons reject in both runtimes; valid 240-scalar emoji reason accepts; 241 rejects. Padded historical adjustment evidence remains unchanged while its trimmed content is checked.
- Structural full-package and playground validators: final command results are preserved in tests/evidence/acceptance-results.json.

Reproduce from this topic directory: `python tests/run_acceptance.py`. The runner captures argument arrays, raw stdout/stderr and exit codes, compiles the TS source, rebuilds the embedded lab kernel and runs the two skill validators. Ordinary test/parity use does not require the skill. Browser QA is separately reproducible through `node tests/visual-qa.mjs` after compilation.

## Standalone browser and rendered visual QA

Playwright Chromium passed at 390,768 and 1440px:

- Nine evidence scenarios per width, each stepping through all nine meaningful transitions:27 scenario runs.
- All 84 fixtures compared as **complete JSON outputs** against the Node reference in each browser width:252 comparisons.
- Initial canonical preview 97350, visible/validated/head 5 and four complete group audit cards.
- No partial correction in audit; before receipt 97500, at equality 97600. Earlier E excludes fee with selected journal IDs empty and raw delta 0.
- Begin replay, Back, Step, Play/Pause, changed speed, scenario/E recomputation, canonical reset and ArrowLeft/Right navigation.
- Controls≥44×44px; no document horizontal overflow in any scenario/stage.
- Reduced-motion Play advances one observation without a timer; no delayed auto-advance; Step and Reset remain deterministic.
- Zero browser warning/error messages and zero page exceptions.

Both SVGs rendered with title/description metadata. Mermaid parsed and rendered. The author visually inspected narrow/wide lab screenshots, both SVGs and the complete flow figure: labels legible, no clipping, exact financial values, light surfaces and responsive single-column mobile layout. Evidence files: tests/evidence/visual-qa.json, lab-390.png, lab-768.png, lab-1440.png, balance-waterfall.png, atomic-receipt.png and fold-flow.png.

The HTML embeds the compiled self-contained TypeScript kernel and nine copied scenario fixtures. It makes no network requests. Timing/stage indexes use ordinary Number; all monetary arithmetic remains BigInt/string. The build script is tests/build_lab.py.

## Visitor integration boundary

Standalone package acceptance passed. The coordinating author owns the isolated website preview: actual production build, descriptive H1, supported examples downloads, inline images/Mermaid, standalone and iframe HTTP routes, responsive containment and console. That visitor receipt is pending at this handoff and is not inferred from local-file screenshots.

Article evidence links use examples rather than research, because the visitor reader exports examples. Related topics use exact approved article routes.

## Trust and limits

The head, catalog and checkpoint registry are trusted host inputs. Prefix cache recomputation establishes consistency relative to supplied evidence; it does not authenticate the source or prove tamper resistance. The source adapter must produce one consistent committed snapshot and an actual post-COMMIT observer receipt. recorded_at is not physical COMMIT proof.

The 4096-group/128-character aggregate representation guards are portable resource bounds, not production ledger limits or a benchmark. Current closed-account status/open periods are not rerun as historical posting policy. Pending reservations, available funds, recognition, collectability and legal/compliance conclusions are outside scope.