docs(gh62): round-5 cross-LLM review — scope expansion + 5 artefacts refined (refs #62)

Round-5 review (Codex GPT-5, DeepSeek, Gemini CLI, Opus 4.7) surfaced
WRONG-DATA in ThisJoinPointEmitter, ~11/17 demand-count divergences, missing
testVerdictMatchesWorstOfPipeline, optimistic verdicts on rows 19/23/43/70,
and the SILENT-GAP-permanent sub-bucket as MECE-breaking. Scope expanded
in-change to close 2 row families (JoinPoint Reflective API + Matcher
Correctness, per D8) so the matrix doesn't ship documenting wrong payloads
or optimistic always-match verdicts.

proposal.md: scope expansion block; ledger reduced to Fix-now/Follow-up/
EXPLICIT-NO-OP; capabilities gain reflective-API + matcher-correctness
invariants; impact lists advice-emitter + pointcut-engine touches; demand
table marked provisional pending DemandCounter; malformed-descriptor
narrative rewritten (.equals() strict honest).

spec.md: +2 Requirements (R4 Reflective API Behavioural Parity, R5 Pointcut
Matcher Correctness) with 6 Scenarios; INV-INS-89 path-β regex enforcement;
+INV-INS-94/95; INV-INS-91 gains ledger.snapshot.sha256 tripwire (D7);
INV-INS-93 extends DemandCounter to .aj + .mop; SILENT-GAP-permanent removed.

design.md: Context expanded to 4 failure classes; mapping table covers R4/R5
+ INV-INS-94/95; Non-Goals revised (no longer 'no production code');
+D7 SHA tripwire, +D8 in-change closure justification; OQ5 reclassifies
handler/declare-precedence/aspect-Foo/pointcut-decl as NOT-NEEDED β;
API surface lists testVerdictMatchesWorstOfPipeline + testReflectiveApi +
testMatcherCorrectness; includeClassNamePatterns synced with tasks.

tasks.md: §1.0 BLOCKER DemandCounter-first + counts regen; §1.3 buckets
rewritten; §3.2 +junit-pioneer GAV; §3.4 scans .aj+.mop with composition-
aware regex; §4.3/4.16/4.19 cenários adversariais (dynamic-vs-static target,
nested try-catch, 4-way binding); §4.5/4.14/4.19/4.20 rows COVERED in-change;
§4.15 @* family parser-crash rationale fix; §4.21 reclassifica 4 NOT-NEEDED β;
§4.22 +3 DEX edge cases (>256 regs, nested try-catch, sparse-switch);
§4.R impl reflective API (~600 LOC, 11 sub-tasks); §4.M impl matcher
correctness (~250 LOC, 8 sub-tasks); §6.1a testVerdictMatchesWorstOfPipeline;
§6.8a/b reflective+matcher invariant tests; §6.10a RVSEC_HOME sanity;
§6.S 5-APK smoke validation; §7.4a commit ledger.snapshot.sha256.

ledger.md: NEW — was missing (deepseek finding). 10 Fix-now entries (post
SILENT-GAP-permanent elimination); 15 Follow-up entries including 'AJC
parity diff' (gemini/deepseek/opus); 2 EXPLICIT-NO-OP Deferred entries.

Validation: openspec validate --strict PASS. Two parallel cross-LLM
verification agents confirmed 16/16 internal consistency checks and 18/18
file:line anchors against HEAD. The 'if() demand 37→8' divergence flagged
empirically is the exact failure mode §1.0 (DemandCounter regeneration)
is designed to surface.

Author: phtcosta