__tests__/tefArchiveStage1Mirror.test.ts (1)

181-197: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick win

Restore the prototype spy in a teardown hook, not only inline.

vi.spyOn(IDBDatabase.prototype, 'transaction') mutates shared global state. If an awaited assertion or waitFor* rejects before the inline mockRestore() (lines 703, 738), the patched prototype leaks into every subsequent test in the file, turning one failure into a cascade. Same concern for vi.stubGlobal('indexedDB', undefined) at line 775.

♻️ Suggested teardown

   beforeEach(() => {
     localStorage.clear();
     vi.resetModules();
     vi.stubGlobal('indexedDB', new IDBFactory());
   });
+
+  afterEach(() => {
+    vi.restoreAllMocks();
+    vi.unstubAllGlobals();
+  });

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@__tests__/tefArchiveStage1Mirror.test.ts` around lines 181 - 197, Restore the
shared `IDBDatabase.prototype.transaction` spy created by
`failReadWriteTransactionsForStore` in an `afterEach` teardown so cleanup runs
even when awaited assertions fail; also restore `indexedDB` after tests that
call `vi.stubGlobal('indexedDB', undefined)`. Remove reliance on the inline
`mockRestore()` calls while preserving the existing failure-injection behavior.

services/tefArchiveService.ts (2)

503-546: 📐 Maintainability & Code Quality | 🔵 Trivial | 💤 Low value

writeFailedVerificationMetadata drops prior diagnostic counts.

On failure, the written metadata only carries verificationError/lastReconciledAt (from previous); mismatchCounts, repairCounts, and integrity counts are all omitted rather than carried forward from the last known-good previous record. Not a correctness bug, but a reader inspecting a 'failed' record loses visibility into the last observed mismatch/integrity state, which is useful for troubleshooting recurring failures.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@services/tefArchiveService.ts` around lines 503 - 546, The metadata
constructed in writeFailedVerificationMetadata must preserve prior diagnostic
counts from previous. Carry forward mismatchCounts, repairCounts, and integrity
counts from the existing previous metadata when creating the failed record,
while retaining the current verificationError, timestamps, and record counts.

---

362-425: 🚀 Performance & Scalability | 🔵 Trivial | ⚡ Quick win

Redundant full-table reads of SAVED_ADS_STORE on every topic-archive mirror pass.

inspectRecordIntegrity → findArchivesWithMissingAds performs a full readAllFromStore(SAVED_ADS_STORE) with no memoization. Within a single mirrorLatestSource attempt it's invoked up to 3 times for the topic-archive config: once via createShadowComparison (Line 570-574), once for postRepairIntegrity (Line 589), and once for integrityAfterMetadata (Line 609). Since mirrorLatestSource now runs on every topic-archive save/delete/shadow-verify (not just background verification), this means every single archive mutation triggers up to 3 full ads-table scans, and each retry attempt (up to MAX_MIRROR_STABILITY_ATTEMPTS) multiplies that further. This will scale poorly as the saved-ads collection grows.

Consider fetching the ads collection once per attempt and threading it through the three integrity checks instead of re-reading per call.

♻️ Sketch: memoize per-attempt ads read

-async function findArchivesWithMissingAds(archives: TefTopicArchive[]): Promise<string[]> {
-  const ads = await readAllFromStore<TefSavedAd>(SAVED_ADS_STORE);
-  const savedAdIds = new Set(ads.map((ad) => ad.id));
-  return archives
-    .filter((archive) => !savedAdIds.has(archive.adId))
-    .map((archive) => archive.id);
-}
+function findArchivesWithMissingAds(archives: TefTopicArchive[], savedAdIds: Set<string>): string[] {
+  return archives
+    .filter((archive) => !savedAdIds.has(archive.adId))
+    .map((archive) => archive.id);
+}

Then in mirrorLatestSource, fetch savedAdIds once per attempt (when config.inspectRelationships applies) and pass it down through createShadowComparison/inspectRecordIntegrity for all three call sites within that attempt.

Also applies to: 567-624

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@services/tefArchiveService.ts` around lines 362 - 425, Memoize the saved-ad
lookup once per mirror attempt instead of rereading SAVED_ADS_STORE in
findArchivesWithMissingAds. Update mirrorLatestSource to fetch and pass the
saved-ad IDs through createShadowComparison and every inspectRecordIntegrity
call, and adjust findArchivesWithMissingAds/relationship inspection to reuse
that per-attempt data across all three integrity checks and retries.

types.ts (1)

164-184: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚖️ Poor tradeoff

Consider a discriminated union for stronger verification-state typing.

All Stage 2 fields (lastVerifiedAt, mismatchCounts, repairCounts, integrity counts) are optional regardless of verificationStatus, so nothing at the type level prevents a consumer from reading a 'verified' record with missing counts (or a 'failed' record with stale counts left over). Every current write site (writeVerifiedMirrorMetadata/writeFailedVerificationMetadata) already sets these consistently, so this is not an active bug, but a discriminated union keyed on verificationStatus would make that invariant compiler-enforced for future writers/readers (e.g. App.tsx).

♻️ Sketch of a discriminated union

interface DurableDataMigrationMetadataBase {
  name: DurableDataMigrationName;
  version: 1;
  state: 'mirroring';
  lastReconciledAt: number;
  sourceRecordCount: number;
  destinationRecordCount: number;
}

export type DurableDataMigrationMetadata =
  | (DurableDataMigrationMetadataBase & {
      verificationStatus: 'verified';
      lastVerifiedAt: number;
      mismatchCounts: DurableDataMismatchCounts;
      repairCounts: DurableDataRepairCounts;
      preRepairIntegrityCounts: DurableDataIntegrityCounts;
      postRepairIntegrityCounts: DurableDataIntegrityCounts;
      relationshipInvalidRecordCount: number;
      legacyShapeRecordCount: number;
    })
  | (DurableDataMigrationMetadataBase & {
      verificationStatus: 'failed';
      lastVerifiedAt?: number;
      verificationError: string;
    });

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@types.ts` around lines 164 - 184, Replace DurableDataMigrationMetadata with a
discriminated union keyed by verificationStatus, using a shared base for the
common migration fields. Require all verification, mismatch, repair, integrity,
and flat post-repair count fields on the 'verified' variant, while the 'failed'
variant must require verificationError and may only retain lastVerifiedAt as
optional. Update dependent type references as needed while preserving the
existing writeVerifiedMirrorMetadata and writeFailedVerificationMetadata
contracts.