task(#936): A3 Epic A verification

Verified, with what was actually run rather than a claim that it was.

**The gate fails on the real tree.** Added a temporary lib/ file carrying an
undeclared storage-shaped key, ran the inventory test, and it failed naming
both the key and the file:

    Actual: ['an_undeclared_global_key']
    undeclared candidate keys:
      an_undeclared_global_key  (lib\services\_gate_probe_tmp.dart)

Probe removed, test green again, tree clean. The three synthetic-source
tests inside the suite prove the same thing in CI; this proves it against
the real scanner over the real lib/, which is the version that matters.

**The gate passes for the declared set.** 27 surfaces classified: 14
perRadio, 3 knownHole pinned to #854/#855/#856, 10 globalOnPurpose, each
with a mandatory reason. Plus 21 justified non-keys.

**A2's repros characterize both bleeds.** Green rather than red, flagged in
A2's commit and again here: a red test cannot merge and Epic A lands before
Epic B. They go red the instant the bleed stops, which is the property Epic
B needs.

**Suite:** 1,408 pass, analyze clean, dart format clean.

What Epic A delivers that #853 did not have: its acceptance bar said every
store has "a verified scoping verdict", and verifying that meant a human
re-reading an issue body. It is now a test. On its first two runs it caught
two surfaces the hand audit had missed (channel_cyr2lat_, contact_cyr2lat_),
a third error sentinel, and one declaration bug of my own.

Citadel: meshcore-open-6ms
Refs #936, #853

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
feat/936-scoping-harness
diffwireauto 1 month ago
parent 483e61fb6a
commit fb1d688d5b

Diff Content Not Available

Powered by TurnKey Linux.