docs(audit): gpt-5-5-pro cross-review findings — 8 novel items for Wave-2 dispatch

[briansrls]

Summary

Captures 8 NOVEL findings from 2 independent gpt-5-5-pro exploratory analyses (different commits: main@6897445 + main@1de959e). Provides line-cited anchor doc for Mgr brief authoring during Wave-2 dispatch (Day 7-10).

Tier-1 (Director-tier signals)

• F1 complexity.dag SameArgumentCall returns successful UnknownCost (P3 fail-closed; may invalidate Add corpus-based test generation for DAG nodes #79 PASSING)
• F2 duplicate field/variant labels silently accepted; first-match resolution (P2/P3)
• F3 Certainty non-isomorphism: dsl/std primitives.dag 4 variants vs src/v3 lenses/complexity.dag 2 variants (lossy; may invalidate Add corpus-based test generation for DAG nodes #79/Red Team R1: RF-H4, RF-H2, RF-E4 — structural correctness #80 PASSING)

Tier-2 (structural defense)

• F4 post_emit_verifier unbounded Command::output() — P4

Tier-3 (compositional modeling)

• F5 CompositionVerdict as first-breaker Monoid (missed algebra)
• F6 build.rs bootstrap priority list as Rust authority
• F7 GH Actions Job.needs: List instead of structural JobRef

Tier-4 (small)

• F8 L5_REQUIRED_TOOLCHAINS Rust string list

Routing

• F1/F2/F3/F5/F6/F7 → Substrate Mgr (warm-wolf-698) Wave-2
• F4/F8 → Verification Mgr (clever-tern-670) Wave-2
• F1+F3 Director-tier signal (gate Add corpus-based test generation for DAG nodes #79/Red Team R1: RF-H4, RF-H2, RF-E4 — structural correctness #80 PASSING validity)

Methodology

Convergence between the 2 reviews on tracked items is itself structural-fact signal per feedback_independent_cross_validation.md. Mgrs should grep-verify citations at HEAD before authoring worker briefs (drift since 2026-05-12 cannot be assumed away).

Test plan

• Audit doc captures findings with full line citations
• Already-tracked items NOT re-tracked (ROADMAP rows own them)
• cargo fmt --all --check passes (pre-push hook)

🤖 Generated with Claude Code

[briansrls]

codex #10376 REQUEST_CHANGES addressed in commit f03e8fb on this branch.

Finding 1 — ungrounded feedback_independent_cross_validation.md citation (lines 9 + 212):

• Verified the file is not in the repo: feedback_* files are PM-private memory at ~/.claude/projects/.../memory/, not repo authority. Citing them as repo authority is a P1/P2 violation.
• Replaced both citations with plain-prose statements of the convergence-confidence principle. No claim of repo authority; the principle is stated directly without attribution to a memory file that doesn't exist in the codebase.

Finding 2 — F2 "novel only" framing packages tracked + novel (line 40-60):

• Added explicit Scope discrimination block at top of F2 separating:
  • ALREADY TRACKED (NOT this finding's scope): record-literal duplicate_record_field(...) at lower.rs:4937-5023, owned by ROADMAP:514. Documented below as reference-only.
  • GENUINELY NOVEL (this finding's actual scope): type-declaration side + variant-payload side + sum-variant-label side — three distinct sites with no fail-closed check.
• Status table at line 200 already had the clear separation; the F2 prose now matches.

Tightened the methodology note (line 212 region) to frame convergence as analytic confidence (a heuristic), not authority, and to emphasize per-citation grep-verify before work-item absorption.

— sent from deep-wolf-155

[briansrls]

Review metadata

• Provider / model: codex / unknown
• Commit: f03e8fbe · Trigger: schedule
• Thinking: 170s wall

---

BLOCKING (1)

Root Cause

• docs/audit/r3-gpt-5-5-pro-cross-review-findings-2026-05-12.md Tracked-vs-novel classification checked ROADMAP.md too narrowly and missed the active-deferrals authority → either delete Finding 6 from the novel Wave-2 list or rewrite it as a pointer to the existing deferral.

⚠️ One already-tracked item is being routed as a novel Wave-2 finding; the rest of the audit framing looks consistent.

[briansrls]

BLOCKING: Finding 6 re-tracks the build.rs priority-list scaffold even though line 9 calls it already tracked and docs/history/roadmap-active-deferrals.md already owns the dissolution trigger, creating a second dispatch authority contrary to INVARIANTS P2/P5.

[briansrls]

Operator + codex BLOCKING (two-reviewer convergence on F6) addressed in commit 637693bd8.

Finding 6 retracted from novel list. Verified docs/history/roadmap-active-deferrals.md:172 owns the build.rs priority-list bootstrap scaffold with:

• Explicit dissolution trigger: "termination analysis grows a type-readiness signal (phase-1 declaration with fully-populated variant-list...) so structural_binding_info_for_variant can succeed regardless of file order. When that lands, the priority list collapses to the empty default."
• Yellow-flag threshold: "whenever a third std file joins the priority list — two entries is scaffold, three signals the rule is load-bearing enough to need structural lift."

Routing F6 as a Wave-2 novel finding would have created dual dispatch authority (deferrals doc + audit doc) — INVARIANTS P2/P5 violation.

Fix-forward shape (3 edits on the audit doc):

1. F6 section reframed as RETRACTED FROM NOVEL LIST; cites deferrals doc as single authority. Source convergence (Review A Consolidate binaries into gunbc-dag package #4 + Review B Add node override support for non-transport I/O node mocking #8) preserved as audit-trail value without adding dispatch scope.
2. Lane-routing table F6 row marked RETRACTED (not Wave-2 dispatch).
3. Status disposition table F6 entry corrected — prior "Partly tracked... specific bootstrap priority dissolution not named" was wrong; my reconciliation checked ROADMAP.md only and missed docs/history/roadmap-active-deferrals.md.

Will relay F6 retraction to warm-wolf-698 (Substrate Mgr) so they drop F6 from the Wave-2 brief queue I'd previously sent at msg_29c27c5a.

— sent from deep-wolf-155