# MapleJockey independent implementation review

Reviewed 12 November 2026. Scope: `core.js`, `app.js`, `template.html`, `build.mjs`, generated internal and illustrative fixtures, and the isolated restoration result. This is a bounded source review and executable negative-input review, not a penetration test or external assurance.

## Results

- 17 independent core checks pass in `review_core.mjs`.
- Claim IDs, required scope/date/reason/observation, evidence references and service references are validated. Duplicate IDs and broken links are rejected.
- Observed, Supplier-declared, Document-reviewed, Inferred and Unknown remain provenance labels. Conflicting and Stale remain separate evidence states.
- A Supported finding is downgraded for Unknown, Conflicting or Stale claims, including claims linked through its requirements.
- Impossible calendar dates are rejected.
- JSON round trips retain records and links. Imported data retains its declared provenance but carries a distinct unauthenticated-origin warning in JSON and the Markdown report.
- Claims CSV preserves assessment ID, assessed-as-of date and unauthenticated import-origin context separately from declared claim provenance.
- Edits do not mutate their prior input and retain previous record values in change history.
- Displayed imported values are HTML-escaped. CSV formula prefixes are neutralised. The self-contained prototype's Content Security Policy prohibits network connections.
- Inventory control-dimension values are explicitly labelled unvalidated notes and accompanied by service claim links.
- The internal case and illustrative status fixture remain distinct. Three material findings are linked to evidence and requirements. Source inspection is not described as proof of provider residency, support access, key control or legal arrangements.
- GC framework applicability and customer-managed versus customer-controlled keys are expressly distinguished.

## Issues found and corrected during review

1. Supported findings originally ignored Unknown/Stale/Conflicting claims linked through requirements. The derivation now includes those claims.
2. Date validation originally accepted an impossible date such as 2026-02-31. Calendar consistency is now checked.
3. Import-origin trust warnings originally existed only in UI memory. They now persist in assessment metadata and appear in JSON/Markdown exports.
4. Editable service-dimension notes originally resembled evidence-backed conclusions. They now carry an explicit note label and linked claims.
5. The case selector now identifies imported data rather than retaining the previous case label.
6. Claims CSV now retains assessment and import-authenticity context. The added executable assertion passes.

## Boundaries and remaining checks

No open material issue remains in the reviewed source and test paths. Browser UI behaviour, downloaded file handling and the restored workload's DOM/interaction checks are performed by the owning agent, not this review. The owning agent reports verification of claim filters/details, action edit/save/load/reload, fixture separation and rendering, window selection and export in the restored Internal reporting application. Browser reimport remains explicitly unverified after a file-chooser stall; core import tests pass. The sanitised restoration uses synthetic report data and does not establish production recovery, hosted identity, DNS, vendor backup or key-service continuity.

Local browser storage and files are not an encrypted evidence vault or a multi-user access-control system. Source authenticity, imported history and source hashes are not cryptographically attested by this prototype. Its scope is a founder-led local demonstrator.
