[DealAppSeo]

Red-team of the merged bricks (#1513 Lane A, #1514 Evidence Pack)

Adversarial pass on the two inert bricks, hunting a false-pass — a brick reporting ok/verified/correct when the underlying reality is not. One real no-silent-failure gap found and fixed here; two others flagged (one Sean-MAJOR, one theoretical).

Fixed here (LOW, inert)

Silent skip of the existence check. When rows cite trinity_tasks:<id> but the caller omits taskEvidence, the increment-2 existence/done join never runs, yet the pack still returns ok:true — a skipped check reading as a pass (r10 / THE_ONE §10). Live: all 656 rows are task citations [V 2026-11-04], so it is 100% applicable.

Fix: a non-gating signal task_existence_check_skipped (detail {citing_rows, cited_tasks}) emitted whenever rows cite tasks but no taskEvidence was supplied. ok is unchanged — the opt-in is intentional; this is the tripwire for when a task-bearing caller forgets it, and it is distinct from the cited_task_exists invariant (check RAN and FAILED).

• Pure reducer, zero callers in any scoring path → cannot change live state (r23: ship on convergence).
• 17/17 jest green (+3 adversarial/signal tests; the net-negative test updated to assert the honest signal set).

Flagged, NOT changed

• Lane A deliverable binding (Sean MAJOR). lane-a-dryrun.ts builds the receipt's artifact_hash from the hash of the verdict string {id,verdict:TRUE}, not from deliverable_sha256. The deliverable hash is surfaced but never bound into the signed receipt or the would-write row — an eventual +1 would credit contractId+verdict with no cryptographic tie to the artifact. Binding it changes the signed public receipt = Sean's gate.
• Opaque evidence_id (theoretical). A non-task, non-hash, non-empty evidence_id passes increment-1 grounding and is exempt from increment-2. 0 such rows in prod today; recommend a shape signal once the immune evidence_kind starts writing, not a gating tightening now.

Grok co-signed the fix and both flags (D-054 concur). D-147 to be logged.

🤖 Generated with Claude Code

[strix-security]

Strix Security Review

✅ No open security findings.

PR summary: Adds a non-gating `task_existence_check_skipped` signal to the inert Evidence Pack reducer so a pack built from task-citing rows without caller-supplied `taskEvidence` reports that the existence/done join never ran, plus tests for the new signal. `ok` and the existing invariants are unchanged.

Review summary

This PR only adds a non-gating diagnostic to a pure, caller-driven reducer: rows that cite trinity_tasks:<id> while no taskEvidence was supplied now produce a task_existence_check_skipped signal (with citing_rows/cited_tasks detail), leaving ok and all invariants untouched. I read the full reducer and the accompanying test changes. The new code introduces no attacker-reachable sink — it writes fixed-key numbers into an in-memory signal list, emits on every skipped-join case without a way to suppress it, and cannot be bypassed with an empty-but-present taskEvidence object (the join still runs and cited_task_exists trips). Search confirms the reducer has no callers, so no downstream gate on signals or ok changes behaviour. The item the author flagged (Lane A not binding deliverable_sha256 into the signed receipt) is in lane-a-dryrun.ts, which is not part of this diff, so it is outside this review's scope. No security issue introduced by the changed lines.

_{Review 1 on this PR · e4ed9a6 · requested Nov 5 00:43 UTC when the PR was opened}

---

Reviewed by Strix
_{Re-run review · Configure security review settings}

[strix-security]

This PR only adds a non-gating diagnostic to a pure, caller-driven reducer: rows that cite trinity_tasks:<id> while no taskEvidence was supplied now produce a task_existence_check_skipped signal (with citing_rows/cited_tasks detail), leaving ok and all invariants untouched. I read the full reducer and the accompanying test changes. The new code introduces no attacker-reachable sink — it writes fixed-key numbers into an in-memory signal list, emits on every skipped-join case without a way to suppress it, and cannot be bypassed with an empty-but-present taskEvidence object (the join still runs and cited_task_exists trips). Search confirms the reducer has no callers, so no downstream gate on signals or ok changes behaviour. The item the author flagged (Lane A not binding deliverable_sha256 into the signed receipt) is in lane-a-dryrun.ts, which is not part of this diff, so it is outside this review's scope. No security issue introduced by the changed lines.

---

Reviewed by Strix
_{Configure security review settings}