fix: address CodeRabbit review feedback

plugins/subminer-workflow/skills/subminer-change-verification/scripts/verify_subminer_change.sh

@@ -72,6 +72,14 @@ add_blocker() {
   BLOCKED=1
 }
 
+validate_artifact_dir() {
+  local candidate=$1
+  if [[ ! "$candidate" =~ ^[A-Za-z0-9._/@:+-]+$ ]]; then
+    echo "Invalid characters in --artifact-dir path" >&2
+    exit 2
+  fi
+}
+
 append_step_record() {
   printf '%s\t%s\t%s\t%s\t%s\t%s\t%s\t%s\n' \
     "$1" "$2" "$3" "$4" "$5" "$6" "$7" "$8" >>"$STEPS_TSV"
@@ -411,6 +419,7 @@ if [[ -z "${ARTIFACT_DIR:-}" ]]; then
   SESSION_ID=$(generate_session_id)
   ARTIFACT_DIR="$REPO_ROOT/.tmp/skill-verification/$SESSION_ID"
 else
+  validate_artifact_dir "$ARTIFACT_DIR"
   SESSION_ID=$(basename "$ARTIFACT_DIR")
 fi
 

src/core/services/immersion-tracker-service.test.ts

@@ -1982,6 +1982,7 @@ test('flushSingle reuses cached prepared statements', async () => {
         cardsMined?: number;
         lookupCount?: number;
         lookupHits?: number;
+        yomitanLookupCount?: number;
         pauseCount?: number;
         pauseMs?: number;
         seekForwardCount?: number;
@@ -2051,6 +2052,7 @@ test('flushSingle reuses cached prepared statements', async () => {
       cardsMined: 0,
       lookupCount: 0,
       lookupHits: 0,
+      yomitanLookupCount: 0,
       pauseCount: 0,
       pauseMs: 0,
       seekForwardCount: 0,

src/core/services/immersion-tracker/__tests__/query.test.ts

@@ -208,6 +208,104 @@ test('getAnimeEpisodes prefers the latest session media position when the latest
   }
 });
 
+test('getAnimeEpisodes includes unwatched episodes for the anime', () => {
+  const dbPath = makeDbPath();
+  const db = new Database(dbPath);
+
+  try {
+    ensureSchema(db);
+    const watchedVideoId = getOrCreateVideoRecord(db, 'local:/tmp/watched-episode.mkv', {
+      canonicalTitle: 'Watched Episode',
+      sourcePath: '/tmp/watched-episode.mkv',
+      sourceUrl: null,
+      sourceType: SOURCE_TYPE_LOCAL,
+    });
+    const unwatchedVideoId = getOrCreateVideoRecord(db, 'local:/tmp/unwatched-episode.mkv', {
+      canonicalTitle: 'Unwatched Episode',
+      sourcePath: '/tmp/unwatched-episode.mkv',
+      sourceUrl: null,
+      sourceType: SOURCE_TYPE_LOCAL,
+    });
+    const animeId = getOrCreateAnimeRecord(db, {
+      parsedTitle: 'Episode Coverage Anime',
+      canonicalTitle: 'Episode Coverage Anime',
+      anilistId: null,
+      titleRomaji: null,
+      titleEnglish: null,
+      titleNative: null,
+      metadataJson: null,
+    });
+    linkVideoToAnimeRecord(db, watchedVideoId, {
+      animeId,
+      parsedBasename: 'watched-episode.mkv',
+      parsedTitle: 'Episode Coverage Anime',
+      parsedSeason: 1,
+      parsedEpisode: 1,
+      parserSource: 'fallback',
+      parserConfidence: 1,
+      parseMetadataJson: '{"episode":1}',
+    });
+    linkVideoToAnimeRecord(db, unwatchedVideoId, {
+      animeId,
+      parsedBasename: 'unwatched-episode.mkv',
+      parsedTitle: 'Episode Coverage Anime',
+      parsedSeason: 1,
+      parsedEpisode: 2,
+      parserSource: 'fallback',
+      parserConfidence: 1,
+      parseMetadataJson: '{"episode":2}',
+    });
+
+    const watchedSessionId = startSessionRecord(db, watchedVideoId, 1_000_000).sessionId;
+    db.prepare(
+      `
+      UPDATE imm_sessions
+      SET
+        ended_at_ms = ?,
+        status = 2,
+        ended_media_ms = ?,
+        active_watched_ms = ?,
+        cards_mined = ?,
+        tokens_seen = ?,
+        yomitan_lookup_count = ?,
+        LAST_UPDATE_DATE = ?
+      WHERE session_id = ?
+      `,
+    ).run(1_005_000, 7_000, 3_000, 2, 20, 4, 1_005_000, watchedSessionId);
+
+    const episodes = getAnimeEpisodes(db, animeId);
+    assert.equal(episodes.length, 2);
+    assert.deepEqual(
+      episodes.map((episode) => ({
+        videoId: episode.videoId,
+        totalSessions: episode.totalSessions,
+        totalActiveMs: episode.totalActiveMs,
+        totalCards: episode.totalCards,
+        totalTokensSeen: episode.totalTokensSeen,
+      })),
+      [
+        {
+          videoId: watchedVideoId,
+          totalSessions: 1,
+          totalActiveMs: 3_000,
+          totalCards: 2,
+          totalTokensSeen: 20,
+        },
+        {
+          videoId: unwatchedVideoId,
+          totalSessions: 0,
+          totalActiveMs: 0,
+          totalCards: 0,
+          totalTokensSeen: 0,
+        },
+      ],
+    );
+  } finally {
+    db.close();
+    cleanupDbPath(dbPath);
+  }
+});
+
 test('getAnimeEpisodes falls back to the latest subtitle segment end when session progress checkpoints are missing', () => {
   const dbPath = makeDbPath();
   const db = new Database(dbPath);
@@ -586,6 +684,109 @@ test('getTrendsDashboard returns chart-ready aggregated series', () => {
   }
 });
 
+test('getTrendsDashboard keeps local-midnight session buckets separate', () => {
+  const dbPath = makeDbPath();
+  const db = new Database(dbPath);
+
+  try {
+    ensureSchema(db);
+    const stmts = createTrackerPreparedStatements(db);
+    const videoId = getOrCreateVideoRecord(db, 'local:/tmp/local-midnight-trends.mkv', {
+      canonicalTitle: 'Local Midnight Trends',
+      sourcePath: '/tmp/local-midnight-trends.mkv',
+      sourceUrl: null,
+      sourceType: SOURCE_TYPE_LOCAL,
+    });
+    const animeId = getOrCreateAnimeRecord(db, {
+      parsedTitle: 'Local Midnight Trends',
+      canonicalTitle: 'Local Midnight Trends',
+      anilistId: null,
+      titleRomaji: null,
+      titleEnglish: null,
+      titleNative: null,
+      metadataJson: null,
+    });
+    linkVideoToAnimeRecord(db, videoId, {
+      animeId,
+      parsedBasename: 'local-midnight-trends.mkv',
+      parsedTitle: 'Local Midnight Trends',
+      parsedSeason: 1,
+      parsedEpisode: 1,
+      parserSource: 'test',
+      parserConfidence: 1,
+      parseMetadataJson: null,
+    });
+
+    const beforeMidnight = new Date(2026, 2, 1, 23, 30).getTime();
+    const afterMidnight = new Date(2026, 2, 2, 0, 30).getTime();
+    const firstSessionId = startSessionRecord(db, videoId, beforeMidnight).sessionId;
+    const secondSessionId = startSessionRecord(db, videoId, afterMidnight).sessionId;
+
+    for (const [sessionId, startedAtMs, tokensSeen, lookupCount] of [
+      [firstSessionId, beforeMidnight, 100, 4],
+      [secondSessionId, afterMidnight, 120, 6],
+    ] as const) {
+      stmts.telemetryInsertStmt.run(
+        sessionId,
+        startedAtMs + 60_000,
+        60_000,
+        60_000,
+        1,
+        tokensSeen,
+        0,
+        lookupCount,
+        lookupCount,
+        lookupCount,
+        0,
+        0,
+        0,
+        0,
+        startedAtMs + 60_000,
+        startedAtMs + 60_000,
+      );
+      db.prepare(
+        `
+        UPDATE imm_sessions
+        SET
+          ended_at_ms = ?,
+          status = 2,
+          total_watched_ms = ?,
+          active_watched_ms = ?,
+          lines_seen = ?,
+          tokens_seen = ?,
+          lookup_count = ?,
+          lookup_hits = ?,
+          yomitan_lookup_count = ?,
+          LAST_UPDATE_DATE = ?
+        WHERE session_id = ?
+        `,
+      ).run(
+        startedAtMs + 60_000,
+        60_000,
+        60_000,
+        1,
+        tokensSeen,
+        lookupCount,
+        lookupCount,
+        lookupCount,
+        startedAtMs + 60_000,
+        sessionId,
+      );
+    }
+
+    const dashboard = getTrendsDashboard(db, 'all', 'day');
+    assert.equal(dashboard.progress.lookups.length, 2);
+    assert.deepEqual(
+      dashboard.progress.lookups.map((point) => point.value),
+      [4, 10],
+    );
+    assert.equal(dashboard.ratios.lookupsPerHundred.length, 2);
+  } finally {
+    db.close();
+    cleanupDbPath(dbPath);
+  }
+});
+
 test('getQueryHints reads all-time totals from lifetime summary', () => {
   const dbPath = makeDbPath();
   const db = new Database(dbPath);
@@ -1024,6 +1225,36 @@ test('getMonthlyRollups returns all rows for the most recent rollup months', ()
   }
 });
 
+test('getMonthlyRollups derives rate metrics from stored monthly totals', () => {
+  const dbPath = makeDbPath();
+  const db = new Database(dbPath);
+  try {
+    ensureSchema(db);
+    const insertRollup = db.prepare(
+      `
+      INSERT INTO imm_monthly_rollups (
+        rollup_month, video_id, total_sessions, total_active_min, total_lines_seen,
+        total_tokens_seen, total_cards, CREATED_DATE, LAST_UPDATE_DATE
+      ) VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?)
+    `,
+    );
+    const nowMs = Date.now();
+    insertRollup.run(202602, 1, 2, 30, 20, 90, 15, nowMs, nowMs);
+    insertRollup.run(202602, 2, 1, 0, 10, 25, 5, nowMs, nowMs);
+
+    const rows = getMonthlyRollups(db, 1);
+    assert.equal(rows.length, 2);
+    assert.equal(rows[1]?.cardsPerHour, 30);
+    assert.equal(rows[1]?.tokensPerMin, 3);
+    assert.equal(rows[1]?.lookupHitRate ?? null, null);
+    assert.equal(rows[0]?.cardsPerHour ?? null, null);
+    assert.equal(rows[0]?.tokensPerMin ?? null, null);
+  } finally {
+    db.close();
+    cleanupDbPath(dbPath);
+  }
+});
+
 test('getAnimeDailyRollups returns all rows for the most recent rollup days', () => {
   const dbPath = makeDbPath();
   const db = new Database(dbPath);
@@ -2234,9 +2465,8 @@ test('cover art queries reuse a shared blob across duplicate anime art rows', ()
     const animeArt = getAnimeCoverArt(db, animeId);
     const library = getMediaLibrary(db);
 
-    assert.equal(artOne?.coverBlob?.length, 4);
-    assert.equal(artTwo?.coverBlob?.length, 4);
-    assert.deepEqual(artOne?.coverBlob, artTwo?.coverBlob);
+    assert.deepEqual(artOne?.coverBlob, Buffer.from([1, 2, 3, 4]));
+    assert.deepEqual(artTwo?.coverBlob, Buffer.from([9, 9, 9, 9]));
     assert.equal(animeArt?.coverBlob?.length, 4);
     assert.deepEqual(
       library.map((row) => ({
@@ -2254,6 +2484,52 @@ test('cover art queries reuse a shared blob across duplicate anime art rows', ()
   }
 });
 
+test('upsertCoverArt prefers freshly fetched bytes over a reused shared hash', () => {
+  const dbPath = makeDbPath();
+  const db = new Database(dbPath);
+
+  try {
+    ensureSchema(db);
+    const originalVideoId = getOrCreateVideoRecord(db, 'local:/tmp/shared-cover-original.mkv', {
+      canonicalTitle: 'Shared Cover Original',
+      sourcePath: '/tmp/shared-cover-original.mkv',
+      sourceUrl: null,
+      sourceType: SOURCE_TYPE_LOCAL,
+    });
+    const refreshedVideoId = getOrCreateVideoRecord(db, 'local:/tmp/shared-cover-refresh.mkv', {
+      canonicalTitle: 'Shared Cover Refresh',
+      sourcePath: '/tmp/shared-cover-refresh.mkv',
+      sourceUrl: null,
+      sourceType: SOURCE_TYPE_LOCAL,
+    });
+
+    upsertCoverArt(db, originalVideoId, {
+      anilistId: 999,
+      coverUrl: 'https://images.test/shared-refresh.jpg',
+      coverBlob: Buffer.from([1, 2, 3, 4]),
+      titleRomaji: 'Shared Cover Refresh',
+      titleEnglish: 'Shared Cover Refresh',
+      episodesTotal: 12,
+    });
+    upsertCoverArt(db, refreshedVideoId, {
+      anilistId: 999,
+      coverUrl: 'https://images.test/shared-refresh.jpg',
+      coverBlob: Buffer.from([9, 8, 7, 6]),
+      titleRomaji: 'Shared Cover Refresh',
+      titleEnglish: 'Shared Cover Refresh',
+      episodesTotal: 12,
+    });
+
+    const originalArt = getCoverArt(db, originalVideoId);
+    const refreshedArt = getCoverArt(db, refreshedVideoId);
+    assert.deepEqual(originalArt?.coverBlob, Buffer.from([1, 2, 3, 4]));
+    assert.deepEqual(refreshedArt?.coverBlob, Buffer.from([9, 8, 7, 6]));
+  } finally {
+    db.close();
+    cleanupDbPath(dbPath);
+  }
+});
+
 test('anime/media detail and episode queries use ended-session metrics when telemetry rows are absent', () => {
   const dbPath = makeDbPath();
   const db = new Database(dbPath);
@@ -2836,13 +3112,13 @@ test('deleteSession rebuilds word and kanji aggregates from retained subtitle li
 
     assert.ok(sharedWordRow);
     assert.equal(sharedWordRow.frequency, 1);
-    assert.equal(sharedWordRow.first_seen, keptTs);
-    assert.equal(sharedWordRow.last_seen, keptTs);
+    assert.equal(sharedWordRow.first_seen, Math.floor(keptTs / 1000));
+    assert.equal(sharedWordRow.last_seen, Math.floor(keptTs / 1000));
     assert.equal(deletedOnlyWordRow ?? null, null);
     assert.ok(sharedKanjiRow);
     assert.equal(sharedKanjiRow.frequency, 1);
-    assert.equal(sharedKanjiRow.first_seen, keptTs);
-    assert.equal(sharedKanjiRow.last_seen, keptTs);
+    assert.equal(sharedKanjiRow.first_seen, Math.floor(keptTs / 1000));
+    assert.equal(sharedKanjiRow.last_seen, Math.floor(keptTs / 1000));
     assert.equal(deletedOnlyKanjiRow ?? null, null);
   } finally {
     db.close();

src/core/services/immersion-tracker/query-library.ts

@@ -156,7 +156,7 @@ export function getAnimeEpisodes(db: DatabaseSync, animeId: number): AnimeEpisod
       COALESCE(SUM(COALESCE(asm.yomitanLookupCount, s.yomitan_lookup_count, 0)), 0) AS totalYomitanLookupCount,
       MAX(s.started_at_ms) AS lastWatchedMs
     FROM imm_videos v
-    JOIN imm_sessions s ON s.video_id = v.video_id
+    LEFT JOIN imm_sessions s ON s.video_id = v.video_id
     LEFT JOIN active_session_metrics asm ON asm.sessionId = s.session_id
     WHERE v.anime_id = ?
     GROUP BY v.video_id

src/core/services/immersion-tracker/query-maintenance.ts

@@ -352,15 +352,16 @@ export function upsertCoverArt(
   const sharedCoverBlobHash = findSharedCoverBlobHash(db, videoId, art.anilistId, art.coverUrl);
   const fetchedAtMs = toDbMs(nowMs());
   const coverBlob = normalizeCoverBlobBytes(art.coverBlob);
-  let coverBlobHash = sharedCoverBlobHash ?? null;
-  if (!coverBlobHash && coverBlob && coverBlob.length > 0) {
-    coverBlobHash = createHash('sha256').update(coverBlob).digest('hex');
-  }
+  const computedCoverBlobHash =
+    coverBlob && coverBlob.length > 0
+      ? createHash('sha256').update(coverBlob).digest('hex')
+      : null;
+  let coverBlobHash = computedCoverBlobHash ?? sharedCoverBlobHash ?? null;
   if (!coverBlobHash && (!coverBlob || coverBlob.length === 0)) {
     coverBlobHash = existing?.coverBlobHash ?? null;
   }
 
-  if (coverBlobHash && coverBlob && coverBlob.length > 0 && !sharedCoverBlobHash) {
+  if (computedCoverBlobHash && coverBlob && coverBlob.length > 0) {
     db.prepare(
       `
         INSERT INTO imm_cover_art_blobs (blob_hash, cover_blob, CREATED_DATE, LAST_UPDATE_DATE)
@@ -368,7 +369,7 @@ export function upsertCoverArt(
         ON CONFLICT(blob_hash) DO UPDATE SET
           LAST_UPDATE_DATE = excluded.LAST_UPDATE_DATE
       `,
-    ).run(coverBlobHash, coverBlob, fetchedAtMs, fetchedAtMs);
+    ).run(computedCoverBlobHash, coverBlob, fetchedAtMs, fetchedAtMs);
   }
 
   db.prepare(

src/core/services/immersion-tracker/query-sessions.ts

@@ -204,7 +204,7 @@ export function getQueryHints(db: DatabaseSync): {
 
   const now = new Date();
   const todayLocal = Math.floor(
-    new Date(now.getFullYear(), now.getMonth(), now.getDate()).getTime() / 86_400_000,
+    (now.getTime() / 1000 - now.getTimezoneOffset() * 60) / 86_400,
   );
 
   const episodesToday =
@@ -333,9 +333,15 @@ export function getMonthlyRollups(db: DatabaseSync, limit = 24): ImmersionSessio
       total_lines_seen AS totalLinesSeen,
       total_tokens_seen AS totalTokensSeen,
       total_cards AS totalCards,
-      0 AS cardsPerHour,
-      0 AS tokensPerMin,
-      0 AS lookupHitRate
+      CASE
+        WHEN total_active_min > 0 THEN (total_cards * 60.0) / total_active_min
+        ELSE NULL
+      END AS cardsPerHour,
+      CASE
+        WHEN total_active_min > 0 THEN total_tokens_seen * 1.0 / total_active_min
+        ELSE NULL
+      END AS tokensPerMin,
+      NULL AS lookupHitRate
     FROM imm_monthly_rollups
     WHERE rollup_month IN (SELECT rollup_month FROM recent_months)
     ORDER BY rollup_month DESC, video_id DESC

src/core/services/immersion-tracker/query-shared.ts

@@ -197,7 +197,12 @@ function refreshWordAggregates(db: DatabaseSync, wordIds: number[]): void {
       deleteStmt.run(row.wordId);
       continue;
     }
-    updateStmt.run(row.frequency, row.firstSeen, row.lastSeen, row.wordId);
+    updateStmt.run(
+      row.frequency,
+      Math.floor(row.firstSeen / 1000),
+      Math.floor(row.lastSeen / 1000),
+      row.wordId,
+    );
   }
 }
 
@@ -241,7 +246,12 @@ function refreshKanjiAggregates(db: DatabaseSync, kanjiIds: number[]): void {
       deleteStmt.run(row.kanjiId);
       continue;
     }
-    updateStmt.run(row.frequency, row.firstSeen, row.lastSeen, row.kanjiId);
+    updateStmt.run(
+      row.frequency,
+      Math.floor(row.firstSeen / 1000),
+      Math.floor(row.lastSeen / 1000),
+      row.kanjiId,
+    );
   }
 }
 

src/core/services/immersion-tracker/query-trends.ts

@@ -112,6 +112,16 @@ function makeTrendLabel(value: number): string {
   });
 }
 
+function getLocalEpochDay(timestampMs: number): number {
+  const date = new Date(timestampMs);
+  return Math.floor((timestampMs - date.getTimezoneOffset() * 60_000) / 86_400_000);
+}
+
+function getLocalDateForEpochDay(epochDay: number): Date {
+  const utcDate = new Date(epochDay * 86_400_000);
+  return new Date(utcDate.getTime() + utcDate.getTimezoneOffset() * 60_000);
+}
+
 function getTrendSessionWordCount(session: Pick<TrendSessionMetricRow, 'tokensSeen'>): number {
   return session.tokensSeen;
 }
@@ -188,7 +198,7 @@ function buildWatchTimeByHour(sessions: TrendSessionMetricRow[]): TrendChartPoin
 }
 
 function dayLabel(epochDay: number): string {
-  return new Date(epochDay * 86_400_000).toLocaleDateString(undefined, {
+  return getLocalDateForEpochDay(epochDay).toLocaleDateString(undefined, {
     month: 'short',
     day: 'numeric',
   });
@@ -200,7 +210,7 @@ function buildSessionSeriesByDay(
 ): TrendChartPoint[] {
   const byDay = new Map<number, number>();
   for (const session of sessions) {
-    const epochDay = Math.floor(session.startedAtMs / 86_400_000);
+    const epochDay = getLocalEpochDay(session.startedAtMs);
     byDay.set(epochDay, (byDay.get(epochDay) ?? 0) + getValue(session));
   }
   return Array.from(byDay.entries())
@@ -213,7 +223,7 @@ function buildLookupsPerHundredWords(sessions: TrendSessionMetricRow[]): TrendCh
   const wordsByDay = new Map<number, number>();
 
   for (const session of sessions) {
-    const epochDay = Math.floor(session.startedAtMs / 86_400_000);
+    const epochDay = getLocalEpochDay(session.startedAtMs);
     lookupsByDay.set(epochDay, (lookupsByDay.get(epochDay) ?? 0) + session.yomitanLookupCount);
     wordsByDay.set(epochDay, (wordsByDay.get(epochDay) ?? 0) + getTrendSessionWordCount(session));
   }
@@ -237,7 +247,7 @@ function buildPerAnimeFromSessions(
 
   for (const session of sessions) {
     const animeTitle = resolveTrendAnimeTitle(session);
-    const epochDay = Math.floor(session.startedAtMs / 86_400_000);
+    const epochDay = getLocalEpochDay(session.startedAtMs);
     const dayMap = byAnime.get(animeTitle) ?? new Map();
     dayMap.set(epochDay, (dayMap.get(epochDay) ?? 0) + getValue(session));
     byAnime.set(animeTitle, dayMap);
@@ -258,7 +268,7 @@ function buildLookupsPerHundredPerAnime(sessions: TrendSessionMetricRow[]): Tren
 
   for (const session of sessions) {
     const animeTitle = resolveTrendAnimeTitle(session);
-    const epochDay = Math.floor(session.startedAtMs / 86_400_000);
+    const epochDay = getLocalEpochDay(session.startedAtMs);
 
     const lookupMap = lookups.get(animeTitle) ?? new Map();
     lookupMap.set(epochDay, (lookupMap.get(epochDay) ?? 0) + session.yomitanLookupCount);
@@ -462,7 +472,7 @@ function buildNewWordsPerDay(db: DatabaseSync, cutoffMs: number | null): TrendCh
   const whereClause = cutoffMs === null ? '' : 'AND first_seen >= ?';
   const prepared = db.prepare(`
     SELECT
-      CAST(first_seen / 86400 AS INTEGER) AS epochDay,
+      CAST(julianday(first_seen, 'unixepoch', 'localtime') - 2440587.5 AS INTEGER) AS epochDay,
       COUNT(*) AS wordCount
     FROM imm_words
     WHERE first_seen IS NOT NULL

src/core/services/immersion-tracker/storage-session.test.ts

@@ -1078,6 +1078,56 @@ test('executeQueuedWrite inserts event and telemetry rows', () => {
   }
 });
 
+test('executeQueuedWrite rejects partial telemetry writes instead of zero-filling', () => {
+  const dbPath = makeDbPath();
+  const db = new Database(dbPath);
+
+  try {
+    ensureSchema(db);
+    const stmts = createTrackerPreparedStatements(db);
+    const videoId = getOrCreateVideoRecord(db, 'local:/tmp/partial-telemetry.mkv', {
+      canonicalTitle: 'Partial Telemetry',
+      sourcePath: '/tmp/partial-telemetry.mkv',
+      sourceUrl: null,
+      sourceType: SOURCE_TYPE_LOCAL,
+    });
+    const { sessionId } = startSessionRecord(db, videoId, 5_000);
+
+    assert.throws(
+      () =>
+        executeQueuedWrite(
+          {
+            kind: 'telemetry',
+            sessionId,
+            sampleMs: 6_000,
+            totalWatchedMs: 1_000,
+            activeWatchedMs: 900,
+            linesSeen: 3,
+            cardsMined: 1,
+            lookupCount: 2,
+            lookupHits: 1,
+            yomitanLookupCount: 0,
+            pauseCount: 1,
+            pauseMs: 50,
+            seekForwardCount: 0,
+            seekBackwardCount: 0,
+            mediaBufferEvents: 0,
+          },
+          stmts,
+        ),
+      /Incomplete telemetry write/,
+    );
+
+    const telemetryCount = db
+      .prepare('SELECT COUNT(*) AS total FROM imm_session_telemetry WHERE session_id = ?')
+      .get(sessionId) as { total: number };
+    assert.equal(telemetryCount.total, 0);
+  } finally {
+    db.close();
+    cleanupDbPath(dbPath);
+  }
+});
+
 test('executeQueuedWrite inserts and upserts word and kanji rows', () => {
   const dbPath = makeDbPath();
   const db = new Database(dbPath);

src/core/services/immersion-tracker/storage.ts

@@ -1406,27 +1406,46 @@ function incrementKanjiAggregate(
 export function executeQueuedWrite(write: QueuedWrite, stmts: TrackerPreparedStatements): void {
   const currentMs = toDbMs(nowMs());
   if (write.kind === 'telemetry') {
+    if (
+      write.totalWatchedMs === undefined ||
+      write.activeWatchedMs === undefined ||
+      write.linesSeen === undefined ||
+      write.tokensSeen === undefined ||
+      write.cardsMined === undefined ||
+      write.lookupCount === undefined ||
+      write.lookupHits === undefined ||
+      write.yomitanLookupCount === undefined ||
+      write.pauseCount === undefined ||
+      write.pauseMs === undefined ||
+      write.seekForwardCount === undefined ||
+      write.seekBackwardCount === undefined ||
+      write.mediaBufferEvents === undefined
+    ) {
+      throw new Error('Incomplete telemetry write');
+    }
     const telemetrySampleMs = toDbMs(write.sampleMs ?? Number(currentMs));
     stmts.telemetryInsertStmt.run(
       write.sessionId,
       telemetrySampleMs,
-      write.totalWatchedMs ?? 0,
-      write.activeWatchedMs ?? 0,
-      write.linesSeen ?? 0,
-      write.tokensSeen ?? 0,
-      write.cardsMined ?? 0,
-      write.lookupCount ?? 0,
-      write.lookupHits ?? 0,
-      write.yomitanLookupCount ?? 0,
-      write.pauseCount ?? 0,
-      write.pauseMs ?? 0,
-      write.seekForwardCount ?? 0,
-      write.seekBackwardCount ?? 0,
-      write.mediaBufferEvents ?? 0,
+      write.totalWatchedMs,
+      write.activeWatchedMs,
+      write.linesSeen,
+      write.tokensSeen,
+      write.cardsMined,
+      write.lookupCount,
+      write.lookupHits,
+      write.yomitanLookupCount,
+      write.pauseCount,
+      write.pauseMs,
+      write.seekForwardCount,
+      write.seekBackwardCount,
+      write.mediaBufferEvents,
       currentMs,
       currentMs,
     );
-    stmts.sessionCheckpointStmt.run(write.lastMediaMs ?? null, currentMs, write.sessionId);
+    if (write.lastMediaMs !== undefined) {
+      stmts.sessionCheckpointStmt.run(write.lastMediaMs ?? null, currentMs, write.sessionId);
+    }
     return;
   }
   if (write.kind === 'word') {

src/core/services/immersion-tracker/time.test.ts

@@ -0,0 +1,7 @@
+import assert from 'node:assert/strict';
+import test from 'node:test';
+import { nowMs } from './time.js';
+
+test('nowMs returns wall-clock epoch milliseconds', () => {
+  assert.ok(nowMs() > 1_600_000_000_000);
+});

src/core/services/immersion-tracker/time.ts

@@ -1,10 +1,8 @@
-const SQLITE_SAFE_EPOCH_BASE_MS = 2_000_000_000;
-
 export function nowMs(): number {
   const perf = globalThis.performance;
-  if (perf) {
-    return SQLITE_SAFE_EPOCH_BASE_MS + Math.floor(perf.now());
+  if (perf && Number.isFinite(perf.timeOrigin)) {
+    return Math.floor(perf.timeOrigin + perf.now());
   }
 
-  return SQLITE_SAFE_EPOCH_BASE_MS;
+  return Date.now();
 }

src/main/character-dictionary-runtime/zip.test.ts

@@ -0,0 +1,98 @@
+import assert from 'node:assert/strict';
+import fs from 'node:fs';
+import os from 'node:os';
+import path from 'node:path';
+import test from 'node:test';
+import { buildDictionaryZip } from './zip';
+import type { CharacterDictionaryTermEntry } from './types';
+
+function makeTempDir(): string {
+  return fs.mkdtempSync(path.join(os.tmpdir(), 'subminer-character-zip-'));
+}
+
+function cleanupDir(dirPath: string): void {
+  fs.rmSync(dirPath, { recursive: true, force: true });
+}
+
+function readStoredZipEntries(zipPath: string): Map<string, Buffer> {
+  const archive = fs.readFileSync(zipPath);
+  const entries = new Map<string, Buffer>();
+  let cursor = 0;
+
+  while (cursor + 4 <= archive.length) {
+    const signature = archive.readUInt32LE(cursor);
+    if (signature === 0x02014b50 || signature === 0x06054b50) {
+      break;
+    }
+    assert.equal(signature, 0x04034b50, `unexpected local file header at offset ${cursor}`);
+
+    const compressedSize = archive.readUInt32LE(cursor + 18);
+    const fileNameLength = archive.readUInt16LE(cursor + 26);
+    const extraLength = archive.readUInt16LE(cursor + 28);
+    const fileNameStart = cursor + 30;
+    const dataStart = fileNameStart + fileNameLength + extraLength;
+    const fileName = archive.subarray(fileNameStart, fileNameStart + fileNameLength).toString(
+      'utf8',
+    );
+    const data = archive.subarray(dataStart, dataStart + compressedSize);
+    entries.set(fileName, Buffer.from(data));
+    cursor = dataStart + compressedSize;
+  }
+
+  return entries;
+}
+
+test('buildDictionaryZip writes a valid stored zip without fs.writeFileSync', () => {
+  const tempDir = makeTempDir();
+  const outputPath = path.join(tempDir, 'dictionary.zip');
+  const termEntries: CharacterDictionaryTermEntry[] = [
+    ['アルファ', 'あるふぁ', '', '', 0, ['Alpha entry'], 0, 'name'],
+  ];
+  const originalBufferConcat = Buffer.concat;
+
+  try {
+    Buffer.concat = ((...args: Parameters<typeof Buffer.concat>) => {
+      throw new Error(`buildDictionaryZip should not Buffer.concat the full archive (${args[0].length} chunks)`);
+    }) as typeof Buffer.concat;
+
+    const result = buildDictionaryZip(
+      outputPath,
+      'Dictionary Title',
+      'Dictionary Description',
+      '2026-03-27',
+      termEntries,
+      [{ path: 'images/alpha.bin', dataBase64: Buffer.from([1, 2, 3]).toString('base64') }],
+    );
+
+    assert.equal(result.zipPath, outputPath);
+    assert.equal(result.entryCount, 1);
+
+    const entries = readStoredZipEntries(outputPath);
+    assert.deepEqual([...entries.keys()].sort(), [
+      'images/alpha.bin',
+      'index.json',
+      'tag_bank_1.json',
+      'term_bank_1.json',
+    ]);
+
+    const indexJson = JSON.parse(entries.get('index.json')!.toString('utf8')) as {
+      title: string;
+      description: string;
+      revision: string;
+      format: number;
+    };
+    assert.equal(indexJson.title, 'Dictionary Title');
+    assert.equal(indexJson.description, 'Dictionary Description');
+    assert.equal(indexJson.revision, '2026-03-27');
+    assert.equal(indexJson.format, 3);
+
+    const termBank = JSON.parse(entries.get('term_bank_1.json')!.toString('utf8')) as
+      CharacterDictionaryTermEntry[];
+    assert.equal(termBank.length, 1);
+    assert.equal(termBank[0]?.[0], 'アルファ');
+    assert.deepEqual(entries.get('images/alpha.bin'), Buffer.from([1, 2, 3]));
+  } finally {
+    Buffer.concat = originalBufferConcat;
+    cleanupDir(tempDir);
+  }
+});

src/main/character-dictionary-runtime/zip.ts

@@ -5,8 +5,8 @@ import type { CharacterDictionarySnapshotImage, CharacterDictionaryTermEntry } f
 
 type ZipEntry = {
   name: string;
-  data: Buffer;
   crc32: number;
+  size: number;
   localHeaderOffset: number;
 };
 
@@ -67,97 +67,78 @@ function crc32(data: Buffer): number {
   return (crc ^ 0xffffffff) >>> 0;
 }
 
-function createStoredZip(files: Array<{ name: string; data: Buffer }>): Buffer {
-  const chunks: Buffer[] = [];
-  const entries: ZipEntry[] = [];
-  let offset = 0;
+function createLocalFileHeader(fileName: Buffer, fileCrc32: number, fileSize: number): Buffer {
+  const local = Buffer.alloc(30 + fileName.length);
+  let cursor = 0;
+  writeUint32LE(local, 0x04034b50, cursor);
+  cursor += 4;
+  local.writeUInt16LE(20, cursor);
+  cursor += 2;
+  local.writeUInt16LE(0, cursor);
+  cursor += 2;
+  local.writeUInt16LE(0, cursor);
+  cursor += 2;
+  local.writeUInt16LE(0, cursor);
+  cursor += 2;
+  local.writeUInt16LE(0, cursor);
+  cursor += 2;
+  writeUint32LE(local, fileCrc32, cursor);
+  cursor += 4;
+  writeUint32LE(local, fileSize, cursor);
+  cursor += 4;
+  writeUint32LE(local, fileSize, cursor);
+  cursor += 4;
+  local.writeUInt16LE(fileName.length, cursor);
+  cursor += 2;
+  local.writeUInt16LE(0, cursor);
+  cursor += 2;
+  fileName.copy(local, cursor);
+  return local;
+}
 
-  for (const file of files) {
-    const fileName = Buffer.from(file.name, 'utf8');
-    const fileData = file.data;
-    const fileCrc32 = crc32(fileData);
-    const local = Buffer.alloc(30 + fileName.length);
-    let cursor = 0;
-    writeUint32LE(local, 0x04034b50, cursor);
-    cursor += 4;
-    local.writeUInt16LE(20, cursor);
-    cursor += 2;
-    local.writeUInt16LE(0, cursor);
-    cursor += 2;
-    local.writeUInt16LE(0, cursor);
-    cursor += 2;
-    local.writeUInt16LE(0, cursor);
-    cursor += 2;
-    local.writeUInt16LE(0, cursor);
-    cursor += 2;
-    writeUint32LE(local, fileCrc32, cursor);
-    cursor += 4;
-    writeUint32LE(local, fileData.length, cursor);
-    cursor += 4;
-    writeUint32LE(local, fileData.length, cursor);
-    cursor += 4;
-    local.writeUInt16LE(fileName.length, cursor);
-    cursor += 2;
-    local.writeUInt16LE(0, cursor);
-    cursor += 2;
-    fileName.copy(local, cursor);
+function createCentralDirectoryHeader(entry: ZipEntry): Buffer {
+  const fileName = Buffer.from(entry.name, 'utf8');
+  const central = Buffer.alloc(46 + fileName.length);
+  let cursor = 0;
+  writeUint32LE(central, 0x02014b50, cursor);
+  cursor += 4;
+  central.writeUInt16LE(20, cursor);
+  cursor += 2;
+  central.writeUInt16LE(20, cursor);
+  cursor += 2;
+  central.writeUInt16LE(0, cursor);
+  cursor += 2;
+  central.writeUInt16LE(0, cursor);
+  cursor += 2;
+  central.writeUInt16LE(0, cursor);
+  cursor += 2;
+  central.writeUInt16LE(0, cursor);
+  cursor += 2;
+  writeUint32LE(central, entry.crc32, cursor);
+  cursor += 4;
+  writeUint32LE(central, entry.size, cursor);
+  cursor += 4;
+  writeUint32LE(central, entry.size, cursor);
+  cursor += 4;
+  central.writeUInt16LE(fileName.length, cursor);
+  cursor += 2;
+  central.writeUInt16LE(0, cursor);
+  cursor += 2;
+  central.writeUInt16LE(0, cursor);
+  cursor += 2;
+  central.writeUInt16LE(0, cursor);
+  cursor += 2;
+  central.writeUInt16LE(0, cursor);
+  cursor += 2;
+  writeUint32LE(central, 0, cursor);
+  cursor += 4;
+  writeUint32LE(central, entry.localHeaderOffset, cursor);
+  cursor += 4;
+  fileName.copy(central, cursor);
+  return central;
+}
 
-    chunks.push(local, fileData);
-    entries.push({
-      name: file.name,
-      data: fileData,
-      crc32: fileCrc32,
-      localHeaderOffset: offset,
-    });
-    offset += local.length + fileData.length;
-  }
-
-  const centralStart = offset;
-  const centralChunks: Buffer[] = [];
-  for (const entry of entries) {
-    const fileName = Buffer.from(entry.name, 'utf8');
-    const central = Buffer.alloc(46 + fileName.length);
-    let cursor = 0;
-    writeUint32LE(central, 0x02014b50, cursor);
-    cursor += 4;
-    central.writeUInt16LE(20, cursor);
-    cursor += 2;
-    central.writeUInt16LE(20, cursor);
-    cursor += 2;
-    central.writeUInt16LE(0, cursor);
-    cursor += 2;
-    central.writeUInt16LE(0, cursor);
-    cursor += 2;
-    central.writeUInt16LE(0, cursor);
-    cursor += 2;
-    central.writeUInt16LE(0, cursor);
-    cursor += 2;
-    writeUint32LE(central, entry.crc32, cursor);
-    cursor += 4;
-    writeUint32LE(central, entry.data.length, cursor);
-    cursor += 4;
-    writeUint32LE(central, entry.data.length, cursor);
-    cursor += 4;
-    central.writeUInt16LE(fileName.length, cursor);
-    cursor += 2;
-    central.writeUInt16LE(0, cursor);
-    cursor += 2;
-    central.writeUInt16LE(0, cursor);
-    cursor += 2;
-    central.writeUInt16LE(0, cursor);
-    cursor += 2;
-    central.writeUInt16LE(0, cursor);
-    cursor += 2;
-    writeUint32LE(central, 0, cursor);
-    cursor += 4;
-    writeUint32LE(central, entry.localHeaderOffset, cursor);
-    cursor += 4;
-    fileName.copy(central, cursor);
-    centralChunks.push(central);
-    offset += central.length;
-  }
-
-  const centralSize = offset - centralStart;
+function createEndOfCentralDirectory(entriesLength: number, centralSize: number, centralStart: number): Buffer {
   const end = Buffer.alloc(22);
   let cursor = 0;
   writeUint32LE(end, 0x06054b50, cursor);
@@ -166,17 +147,63 @@ function createStoredZip(files: Array<{ name: string; data: Buffer }>): Buffer {
   cursor += 2;
   end.writeUInt16LE(0, cursor);
   cursor += 2;
-  end.writeUInt16LE(entries.length, cursor);
+  end.writeUInt16LE(entriesLength, cursor);
   cursor += 2;
-  end.writeUInt16LE(entries.length, cursor);
+  end.writeUInt16LE(entriesLength, cursor);
   cursor += 2;
   writeUint32LE(end, centralSize, cursor);
   cursor += 4;
   writeUint32LE(end, centralStart, cursor);
   cursor += 4;
   end.writeUInt16LE(0, cursor);
+  return end;
+}
 
-  return Buffer.concat([...chunks, ...centralChunks, end]);
+function writeBuffer(fd: number, buffer: Buffer): void {
+  let written = 0;
+  while (written < buffer.length) {
+    written += fs.writeSync(fd, buffer, written, buffer.length - written);
+  }
+}
+
+function writeStoredZip(outputPath: string, files: Iterable<{ name: string; data: Buffer }>): void {
+  const entries: ZipEntry[] = [];
+  let offset = 0;
+  const fd = fs.openSync(outputPath, 'w');
+
+  try {
+    for (const file of files) {
+      const fileName = Buffer.from(file.name, 'utf8');
+      const fileSize = file.data.length;
+      const fileCrc32 = crc32(file.data);
+      const localHeader = createLocalFileHeader(fileName, fileCrc32, fileSize);
+      writeBuffer(fd, localHeader);
+      writeBuffer(fd, file.data);
+      entries.push({
+        name: file.name,
+        crc32: fileCrc32,
+        size: fileSize,
+        localHeaderOffset: offset,
+      });
+      offset += localHeader.length + fileSize;
+    }
+
+    const centralStart = offset;
+    for (const entry of entries) {
+      const centralHeader = createCentralDirectoryHeader(entry);
+      writeBuffer(fd, centralHeader);
+      offset += centralHeader.length;
+    }
+
+    const centralSize = offset - centralStart;
+    writeBuffer(fd, createEndOfCentralDirectory(entries.length, centralSize, centralStart));
+  } catch (error) {
+    fs.closeSync(fd);
+    fs.rmSync(outputPath, { force: true });
+    throw error;
+  }
+
+  fs.closeSync(fd);
 }
 
 export function buildDictionaryZip(
@@ -187,36 +214,37 @@ export function buildDictionaryZip(
   termEntries: CharacterDictionaryTermEntry[],
   images: CharacterDictionarySnapshotImage[],
 ): { zipPath: string; entryCount: number } {
-  const zipFiles: Array<{ name: string; data: Buffer }> = [
-    {
+  ensureDir(path.dirname(outputPath));
+
+  function* zipFiles(): Iterable<{ name: string; data: Buffer }> {
+    yield {
       name: 'index.json',
       data: Buffer.from(
         JSON.stringify(createIndex(dictionaryTitle, description, revision), null, 2),
         'utf8',
       ),
-    },
-    {
+    };
+    yield {
       name: 'tag_bank_1.json',
       data: Buffer.from(JSON.stringify(createTagBank()), 'utf8'),
-    },
-  ];
+    };
 
-  for (const image of images) {
-    zipFiles.push({
-      name: image.path,
-      data: Buffer.from(image.dataBase64, 'base64'),
-    });
+    for (const image of images) {
+      yield {
+        name: image.path,
+        data: Buffer.from(image.dataBase64, 'base64'),
+      };
+    }
+
+    const entriesPerBank = 10_000;
+    for (let i = 0; i < termEntries.length; i += entriesPerBank) {
+      yield {
+        name: `term_bank_${Math.floor(i / entriesPerBank) + 1}.json`,
+        data: Buffer.from(JSON.stringify(termEntries.slice(i, i + entriesPerBank)), 'utf8'),
+      };
+    }
   }
 
-  const entriesPerBank = 10_000;
-  for (let i = 0; i < termEntries.length; i += entriesPerBank) {
-    zipFiles.push({
-      name: `term_bank_${Math.floor(i / entriesPerBank) + 1}.json`,
-      data: Buffer.from(JSON.stringify(termEntries.slice(i, i + entriesPerBank)), 'utf8'),
-    });
-  }
-
-  ensureDir(path.dirname(outputPath));
-  fs.writeFileSync(outputPath, createStoredZip(zipFiles));
+  writeStoredZip(outputPath, zipFiles());
   return { zipPath: outputPath, entryCount: termEntries.length };
 }