test(rich): consolidate #669 permanent regression ownership

[jnhu76]

TEST_ONLY = YES

PRODUCTION_CODE_CHANGED = NO — 8 *.test.* files, +159/−231. No production behavior change; every surviving invariant keeps a deterministic owner-layer regression.

RECENT_PRS_AUDITED = #687, #690, #691, #692, #693, #694, #696, #700, #702

What changed

Change	Rationale (owner stays intact)
DELETE QuestionRenderer string-in-rich-slot test	RichTextAnswerInput R1 owns the fail-closed mechanism; the wrapper forwards value verbatim (QuestionRenderer.tsx:69-77)
DELETE saveAnswerReceiptsD2 UQ probe	constraint existence owned by the 0044 migration test (raw SQL, /exam_answer_save_receipts_pk/); duplicate-through-adapter behavior is the premise and outcome of the C1 serialization case
MOVE hostile-LaTeX live-DOM audit → MathRenderer.evidence	actual component seam; selector widened to the file's 16-tag active-content set (Reviewer C suggestion applied)
TABLE-DRIVE D5.1 publish-trust rejects	{corrupt, unsupported_version} × {question, option} slots; every cell now asserts ValidationError class AND message (old R3/R5 asserted message only — stronger)
PARAMETERIZE D-F01 wire rejections	3 families × per-row zero-durable-write witness (was: shared witness)
PARAMETERIZE authoring cross-writer seams	2 writer shapes (prompt/option) × representative family; family enumeration owned by contracts leaf tests over the same schema object
FOLD preflight-fired assertion into grammar-bomb test	semantics-preserving merge

Explicitly KEPT despite apparent duplication (load-bearing)

• C15 typed input-rule E2E — the real-event layer. The reality probe enters at view.someProp("handleTextInput"), downstream of the entire browser event chain (keydown → SyncPendingSelectionOnKeydown's domObserver.flush — the fix(rich-ui): sync pending DOM selection before keydown input (#701) #702 fix has NO jsdom witness — → DOMObserver dispatch). Reviewer B proposed deletion; Reviewer C adjudicated the keep: "the deletion would have been a load-bearing loss."
• Replay-receipt layering: engine semantics / DB constraints / API concurrency prove distinct mechanisms (fix(rich): phase-D2 compact append-only replay receipts #691).
• Formula chain: policy matrix / dialog arming / confirm-time clamp / open-context / reality / MathLive E2E — four distinct mechanisms.
• Canonical closure seeded property + fixed examples (different falsification power), T−1/T+1 boundary pair, D4-G frozen-mode counterexample, counterexample+inverse-control pairs.

TESTS_DELETED = 4 tests (2 standalone + 1 describe of 2, migrated not deleted)
TESTS_CONSOLIDATED = 10 tests → 5 table rows (D5.1 4→2, D-F01 wire 2→3 rows, cross-writer 2→2 rows, preflight 2→1)
TESTS_RETAINED_DESPITE_APPARENT_DUPLICATION = 8 groups (see ledger §KEEP)

LOAD_BEARING_INVARIANTS = 40 invariant groups adjudicated in the ownership ledger; all retain ≥1 deterministic owner-layer regression; historical P0/P1 defects #676 / D-F01 / #701 CE-1 each retain clearly identifiable named regressions.

E2E_BEFORE = 13 · E2E_AFTER = 13 (NET_NEW_E2E = 0)
TEST_LOC_BEFORE = 14683 · TEST_LOC_AFTER = 14610 (NET_TEST_LOC = −73)
RUNTIME_BEFORE / AFTER = focused suites marginally reduced (engine 1.96s→1.90s, contracts 1.10s→1.0s); no noisy performance claims.

Evidence (all first-hand on this branch)

• contracts 447 ✓ · engine 760 ✓ · domain 92 ✓ · db 715 ✓ · api 2807 ✓ (full suite; one unrelated load-timing flake passed isolated and did not recur) · web 2346 ✓ · rich E2E 2/2 shards ✓ before AND after · pnpm verify PASS

REVIEWER_B = independent test-architecture audit (KEEP/DELETE/CONSOLIDATE/UNCERTAIN per group; verdict MIXED — suite already substantially owner-layer-correct)
REVIEWER_C = independent cleanup-diff review; verdict NO_LOAD_BEARING_LOSS; 2×P3 + 2×INFO (all bounded, none blocking; one P3 addressed by the selector widening above)

Ownership ledger: /tmp/exam-669-phase-g-test-ledger.md (session artifact)

Closes nothing — parent #669 adjudication continues after merge per the Phase-G protocol. Human merge gate applies.

[coderabbitai]

Important

• 🔍 Trigger review

This repository does not receive automatic reviews because it has fewer than 10 stars.

⚙️ Run configuration

• Configuration used: defaults
• Review profile: CHILL
• Plan: Advanced
• Run ID: c7eb2a34-a3d8-4ce6-8b55-e9a1aa21667c

• Autopilot · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[jnhu76]

Final focused review for #669 Phase G test consolidation @ 128f8ac5.

VERDICT: MERGE / READY_FOR_HUMAN_MERGE = YES.

Independent checks:

• PR is test-only: 8 *.test.* files; no production code changed.
• Deleted QuestionRenderer corrupt-string case is redundant: RichTextAnswerInput.readTrust.test.tsx still directly owns R1 (unexplained string in Rich slot -> fail-closed integrity state).
• Deleted D2 API UQ probe is redundant at that layer: migration suite 0044-answer-save-receipts.test.ts still directly proves exam_answer_save_receipts_pk exists and rejects a duplicate replay key; engine/API receipt semantics remain separately owned.
• D5.1 and D-F01 changes consolidate/table-drive existing semantic states without collapsing distinct authority classes.
• Hostile-LaTeX live-DOM evidence moved to the actual MathRenderer seam and strengthened with the wider active-content selector audit.
• No E2E removed; browser-only C15/#702-style event evidence remains intact.
• Exact-head CI run 37279387388: all 10 GitHub Actions jobs successful; CodeRabbit commit status success.

No new blocker found. After human merge, perform the planned §38 checkpoint (verify master contains 128f8ac5, run the targeted Rich suites / pnpm verify as planned), then #669 may be closed if no production drift is present.