# Final Validation Receipt

## Functional lanes

- Before cutover, the SDK plus pinned reference lane passed `54 / 54`; this is
  the deletion-before receipt for the former upstream execution path.
- After cutover, the full default suite passed `76`, skipped `20`, failed `0`.
- After cutover with SDK 6.0.2.650, the full suite passed `94`, skipped `2`,
  failed `0`. The two skips require an explicit
  project-scale performance corpus.
- Production exposes one local `arkts-diff` implementation and the independent
  `codelinter-diff` profile. Removed upstream/union/fix CLI options fail
  explicitly instead of falling back.

## Accuracy gates

- Static rule matrix covers every enabled/shadow detector and requires positive,
  near-negative, boundary, suppression, known-limit and oracle fields.
- Official checker fixtures assert exact compiler rules/messages and valid-file
  absence of errors.
- Ten locally implemented text/structure rules retain explicit positive,
  near-negative, boundary, suppression and known-limit fixtures. Regex rules
  remain shadow because comment/string false positives and multiline false
  negatives are still possible.
- Project integrity fails closed when a required permission catalog is absent;
  the production ArkTS seam propagates `unavailable/partial/success=false`.
- All three structured adapters reject JSON objects missing their required
  diagnostics schema instead of projecting an empty successful report.
- The pinned upstream checker is used only as a test oracle. Its hashes are in
  `docs/upstream-reference.json`; its canonical source URL/commit is unknown,
  so it is not production-eligible.
- Runtime-risk has a seven-rule matrix with executable positive and
  near-negative fact cases; V1/V2 findings remain shadow and runtime-required.

## Performance and packaging

Catima measured 31 findings across 12 files. Latest canonical cold, warm and
incremental local ArkTS runs measured `5.39s / 471708 KiB`, `4.11s / 473924
KiB`, and `4.34s / 475480 KiB`; the CodeLinter profile measured `7.31s /
481300 KiB`. All remain below the provisional `10s / 1 GiB` workflow budget;
run-to-run variance is larger than the small in-process discovery optimization.
The official SDK compiler owns most RSS. Incremental precheck now runs one
ArkTS checker over changed and impacted files; it creates no Git worktree and
does not compile a baseline checkout.

Production `node --check`, full test lanes, the canonical performance lane,
and `npm pack --dry-run --ignore-scripts --json` passed. Tests, runtime-risk,
and task records are excluded from the package; production backend modules are
included. Static precheck does not prove Hvigor build or device/runtime
behavior, and automatic source fixes are not implemented.