test(supervisor): update runner fixture mock for #2750 lock path

tryAcquireDbLock now acquires via engine.executeRaw (signal-boundable) and releases via engine.executeRawDirect instead of the sql tagged template. The fixture's mock engine returned [] from executeRaw, so every spawned supervisor failed lock acquisition and exited LOCK_HELD (2), breaking all 8 supervisor integration tests. The mock now returns the lock row for gbrain_cycle_locks queries and stubs executeRawDirect. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
fix(cycle): enforce the extract-atoms drain deadline end to end (#2750 )
2026-08-15 01:12:20 +00:00 · 2026-07-22 10:53:10 -07:00 · 2026-07-21 14:41:34 -07:00
26 changed files with 656 additions and 578 deletions
@@ -170,14 +170,10 @@ export async function runImport(
  // v0.22.13 (PR #490 Q2): shared parseWorkers helper rejects bad input
  // (--workers 0, -3, "foo") with a loud error instead of silently falling
  // through to 1. Mirrors sync.ts's flag handling.
-  const { parseWorkers, autoConcurrency } = await import('../core/sync-concurrency.ts');
-  // #1207: undefined (no --workers flag) defers to autoConcurrency below —
-  // the shared sync/import policy (PGLite → 1, >100 files → 4) — instead of
-  // hardcoding serial. Large Postgres imports stop paying one embedding
-  // round-trip per file in sequence.
-  let workerCount: number | undefined;
+  const { parseWorkers } = await import('../core/sync-concurrency.ts');
+  let workerCount: number;
  try {
-    workerCount = parseWorkers(workersArg ?? undefined);
+    workerCount = parseWorkers(workersArg ?? undefined) ?? 1;
  } catch (e) {
    console.error(e instanceof Error ? e.message : String(e));
    process.exit(1);
@@ -256,9 +252,8 @@ export async function runImport(
  }
  const files = resumeFilter(allFiles, dir, completed);

-  // Determine actual worker count. Explicit --workers wins; otherwise the
-  // shared autoConcurrency policy decides from engine kind + file count.
-  const actualWorkers = autoConcurrency(engine, files.length, workerCount);
+  // Determine actual worker count
+  const actualWorkers = workerCount > 1 ? workerCount : 1;
  if (actualWorkers > 1) {
    console.log(`Using ${actualWorkers} parallel workers`);
  }
@@ -2059,6 +2059,8 @@ export async function registerBuiltinHandlers(
        sourceId,
        windowSeconds,
        brainDir: repoPath,
+        // #2750: worker cancel/timeout/lock-loss propagates into the drain.
+        abortSignal: job.signal,
      });
    } catch (e) {
      if (e instanceof LockUnavailableError) {
@@ -1513,21 +1513,12 @@ export async function embed(texts: string[], opts?: EmbedOpts): Promise<Float32A

  const embedding = recipe.touchpoints?.embedding;
  const maxBatchTokens = embedding?.max_batch_tokens;
-  const maxBatchCount = embedding?.max_batch_count;
  const charsPerToken = embedding?.chars_per_token ?? DEFAULT_CHARS_PER_TOKEN;

-  // Pre-split is gated on max_batch_tokens / max_batch_count. Recipes with
-  // neither (e.g. OpenAI) ride the fast path: one embedMany call, no
-  // recursion safety net.
-  const batches = (maxBatchTokens || maxBatchCount)
-    ? splitByTokenBudget(
-        truncated,
-        maxBatchTokens
-          ? Math.floor(maxBatchTokens * effectiveSafetyFactor(recipe))
-          : Number.MAX_SAFE_INTEGER,
-        charsPerToken,
-        maxBatchCount,
-      )
+  // Pre-split is gated on max_batch_tokens. Recipes without it (e.g. OpenAI)
+  // ride the fast path: one embedMany call, no recursion safety net.
+  const batches = maxBatchTokens
+    ? splitByTokenBudget(truncated, Math.floor(maxBatchTokens * effectiveSafetyFactor(recipe)), charsPerToken)
    : [truncated];

  const allEmbeddings: Float32Array[] = [];
@@ -1577,9 +1568,6 @@ export async function embed(texts: string[], opts?: EmbedOpts): Promise<Float32A
 *   responsible for applying any safety-factor shrink before passing in.
 * @param charsPerToken - Provider-specific character density. Defaults to
 *   `DEFAULT_CHARS_PER_TOKEN` (4) when omitted, matching OpenAI tiktoken.
- * @param maxBatchCount - #1199: optional cap on INPUTS per sub-batch, for
- *   providers that reject batches by count (DashScope: 10). When omitted,
- *   only the token budget governs.
 *
 * @internal exported for tests; not part of the public gateway API.
 */
@@ -1587,17 +1575,15 @@ export function splitByTokenBudget(
  texts: string[],
  budgetTokens: number,
  charsPerToken: number = DEFAULT_CHARS_PER_TOKEN,
-  maxBatchCount?: number,
 ): string[][] {
  const ratio = charsPerToken > 0 ? charsPerToken : DEFAULT_CHARS_PER_TOKEN;
-  const maxCount = maxBatchCount !== undefined && maxBatchCount > 0 ? maxBatchCount : Infinity;
  const batches: string[][] = [];
  let current: string[] = [];
  let currentTokens = 0;

  for (const text of texts) {
    const estTokens = Math.ceil(text.length / ratio);
-    if (current.length > 0 && (currentTokens + estTokens > budgetTokens || current.length >= maxCount)) {
+    if (current.length > 0 && currentTokens + estTokens > budgetTokens) {
      batches.push(current);
      current = [];
      currentTokens = 0;
@@ -1623,11 +1609,7 @@ export function isTokenLimitError(err: unknown): boolean {
    /token.*limit.*exceeded/i.test(msg) ||
    // OpenAI embeddings: "Invalid 'input': maximum request size is 300000 tokens per request."
    /maximum request size.*tokens/i.test(msg) ||
-    /max.*tokens.*per.*request/i.test(msg) ||
-    // DashScope: "batch size is invalid, it should not be larger than 10." (#1199)
-    // Count-cap error, but recursive halving shrinks count too, so the same
-    // safety net converges.
-    /batch size is invalid/i.test(msg)
+    /max.*tokens.*per.*request/i.test(msg)
  );
 }

@@ -31,10 +31,6 @@ export const dashscope: Recipe = {
      // path. Conservative declaration so the gateway pre-splits before
      // hitting whatever undocumented server-side limit exists.
      max_batch_tokens: 8192,
-      // #1199: DashScope hard-caps embeddings at 10 inputs per request
-      // ("batch size is invalid, it should not be larger than 10"). The
-      // token budget alone admits far more than 10 short chunks per batch.
-      max_batch_count: 10,
      // text-embedding-v3 mixes English + CJK heavily; the tokenizer is
      // closer to Voyage density than OpenAI tiktoken for CJK-dominant
      // content. Conservative chars_per_token=2 leaves headroom.
@@ -16,15 +16,6 @@ export const google: Recipe = {
      dims_options: [768, 1536, 3072],
      cost_per_1m_tokens_usd: 0.15,
      price_last_verified: '2026-04-20',
-      // #970: Gemini's documented limits are per-INPUT (2048 tokens,
-      // silently truncated beyond) and per-REQUEST count (batchEmbedContents
-      // caps at 100 inputs). There is no separate per-request token cap, so
-      // the token budget is derived: 100 inputs × 2048 tokens. The count cap
-      // binds first for typical chunk sizes. Do NOT copy the 2048 per-input
-      // limit into max_batch_tokens — that would over-split 50×.
-      max_batch_tokens: 204_800,
-      chars_per_token: 4,
-      max_batch_count: 100,
    },
    expansion: {
      models: ['gemini-2.0-flash', 'gemini-2.0-flash-lite'],
@@ -58,8 +58,5 @@ export function getRecipe(id: string): Recipe | undefined {
 }

 export function listRecipes(): Recipe[] {
-  // Read the map (not ALL) so there is one source of truth — getRecipe,
-  // model-resolver, and listRecipes all see the same registry, and tests
-  // can inject a synthetic recipe via RECIPES to exercise registry walks.
-  return [...RECIPES.values()];
+  return [...ALL];
 }
@@ -46,16 +46,6 @@ export interface EmbeddingTouchpoint {
   * Only consulted when `max_batch_tokens` is also set.
   */
  chars_per_token?: number;
-  /**
-   * #1199: maximum number of INPUTS per embedding request, for providers
-   * that hard-cap batch size by count rather than (or in addition to)
-   * tokens — DashScope text-embedding-v3 rejects batches > 10 with
-   * `InvalidParameter`, Gemini batchEmbedContents caps at 100 requests.
-   * When set, the gateway's pre-split flushes a sub-batch at this count
-   * even if the token budget still has room. Independent of
-   * `max_batch_tokens`; either alone triggers the pre-split.
-   */
-  max_batch_count?: number;
  /**
   * Budget-utilization ceiling in (0, 1]. The gateway pre-splits at
   * `safety_factor × max_batch_tokens` to leave headroom for tokenizer
@@ -23,6 +23,10 @@
 */

 import type { BrainEngine } from '../engine.ts';
+import { anySignal } from '../abort-check.ts';
+
+/** Fresh cleanup budget for the lock release after the window signal fires. */
+const LOCK_RELEASE_GRACE_MS = 5_000;

 export interface ExtractAtomsDrainDeps {
  /**
@@ -30,13 +34,13 @@ export interface ExtractAtomsDrainDeps {
   * via `withRefreshingLock`. MUST throw when the lock is held by another
   * process (e.g. `LockUnavailableError`) — the drain lets that propagate so
   * the caller can report `cycle_already_running` and exit, matching the
-   * routine cycle's skip contract.
+   * routine cycle's skip contract. The signal bounds lock acquisition too.
   */
-  withLock: <T>(work: () => Promise<T>) => Promise<T>;
-  /** Process one bounded batch (rediscovers eligibility). Returns counts. */
-  runBatch: () => Promise<{ extracted: number; skipped: number }>;
+  withLock: <T>(work: () => Promise<T>, signal: AbortSignal) => Promise<T>;
+  /** Process one batch. The signal fires at the drain wallclock deadline. */
+  runBatch: (signal: AbortSignal) => Promise<{ extracted: number; skipped: number }>;
  /** Count remaining eligible-but-unextracted pages, or null on query error. */
-  countRemaining: () => Promise<number | null>;
+  countRemaining: (signal: AbortSignal) => Promise<number | null>;
  /** Injectable clock. Production: Date.now. */
  now: () => number;
  /** Optional progress sink (one line per batch). */
@@ -48,6 +52,8 @@ export interface ExtractAtomsDrainOpts {
  windowMs: number;
  /** Hard cap on batches (belt-and-suspenders against a 0-progress loop). Default 1000. */
  maxBatches?: number;
+  /** External caller cancellation (worker timeout / shutdown). */
+  abortSignal?: AbortSignal;
 }

 export interface ExtractAtomsDrainResult {
@@ -68,35 +74,79 @@ export async function runExtractAtomsDrain(
  opts: ExtractAtomsDrainOpts,
 ): Promise<ExtractAtomsDrainResult> {
  const maxBatches = opts.maxBatches ?? 1000;
-  return deps.withLock(async () => {
-    const deadline = deps.now() + opts.windowMs;
+  const deadline = deps.now() + opts.windowMs;
+  // #2750: the window used to be checked only BETWEEN batches, so one slow
+  // batch (sequential LLM calls) or a hung lock/count/write overran it without
+  // bound (observed window=120s → 282.5s). A real-time deadline signal now
+  // cancels (Postgres) or abandons (PGLite, cooperative) whatever is in
+  // flight; the injected clock still drives loop-boundary checks so the pure
+  // loop stays unit-testable.
+  const signal = anySignal(
+    AbortSignal.timeout(Math.max(1, opts.windowMs)),
+    opts.abortSignal,
+  );
+  const result: ExtractAtomsDrainResult = await deps.withLock(async () => {
    let extracted = 0;
    let skipped = 0;
    let batches = 0;
    let stopped: ExtractAtomsDrainResult['stopped'] = 'window';

-    while (deps.now() < deadline) {
+    while (deps.now() < deadline && !signal.aborted) {
      if (batches >= maxBatches) { stopped = 'max_batches'; break; }

-      const before = await deps.countRemaining();
+      let before: number | null;
+      try {
+        before = await deps.countRemaining(signal);
+      } catch (err) {
+        if (signal.aborted) break;
+        throw err;
+      }
      if (before === 0) { stopped = 'drained'; break; }

-      const r = await deps.runBatch();
+      // The backlog count consumed the same wallclock budget — re-check so a
+      // slow count can't hand the batch a window that already expired.
+      if (deps.now() >= deadline || signal.aborted) break;
+
+      let r: { extracted: number; skipped: number };
+      try {
+        r = await deps.runBatch(signal);
+      } catch (err) {
+        if (signal.aborted) break;
+        throw err;
+      }
      extracted += r.extracted;
      skipped += r.skipped;
      batches++;
      deps.onBatch?.({ batch: batches, extracted: r.extracted, remaining: before });

+      // A deadline abort inside the batch can surface as zero progress;
+      // window exhaustion wins over the generic no_progress label.
+      if (deps.now() >= deadline || signal.aborted) break;
+
      // Stop if a batch made zero forward progress — extraction is failing or
      // everything left is ineligible (e.g. all skipped). Prevents a hot loop
      // that spends budget without draining.
      if (r.extracted === 0 && r.skipped === 0) { stopped = 'no_progress'; break; }
    }

-    const remaining = await deps.countRemaining();
+    // After the window elapsed, don't spend more unbounded time on a final
+    // count — report remaining as unknown instead of overrunning further.
+    const windowElapsed = signal.aborted || deps.now() >= deadline;
+    let remaining: number | null = null;
+    if (!windowElapsed) {
+      try {
+        remaining = await deps.countRemaining(signal);
+      } catch (err) {
+        if (!signal.aborted) throw err;
+      }
+    }
    if (remaining === 0) stopped = 'drained';
    return { phase: 'extract_atoms', status: 'ok', extracted, skipped, remaining, batches, stopped };
-  });
+  }, signal);
+  // Internal window expiry is a normal partial result. An EXTERNAL abort
+  // (worker cancel/timeout/shutdown) must reject so Minion records the abort.
+  if (opts.abortSignal?.aborted) throw opts.abortSignal.reason;
+  return result;
 }

 // ─── Shared wiring helper (v0.42.x #1685 DECISION 5A) ──────────────────────
@@ -134,6 +184,8 @@ export interface DrainForSourceOpts {
  maxBatches?: number;
  /** Optional per-batch progress sink (stderr line in dream; job progress in the handler). */
  onBatch?: ExtractAtomsDrainDeps['onBatch'];
+  /** Worker cancellation / shutdown signal (Minion `job.signal`). */
+  abortSignal?: AbortSignal;
 }

 export async function runExtractAtomsDrainForSource(
@@ -149,12 +201,17 @@ export async function runExtractAtomsDrainForSource(

  return runExtractAtomsDrain(
    {
-      withLock: (work) => withRefreshingLock(engine, lockId, work, { ttlMinutes: 5 }),
-      runBatch: async () => {
+      withLock: (work, signal) => withRefreshingLock(engine, lockId, work, {
+        ttlMinutes: 5,
+        signal,
+        releaseTimeoutMs: LOCK_RELEASE_GRACE_MS,
+      }),
+      runBatch: async (signal) => {
        const r = await runPhaseExtractAtoms(engine, {
          sourceId: extractionSourceId,
          dryRun: false,
          brainDir: opts.brainDir,
+          abortSignal: signal,
        });
        const d = (r.details ?? {}) as Record<string, unknown>;
        return {
@@ -162,10 +219,14 @@ export async function runExtractAtomsDrainForSource(
          skipped: Number(d.duplicates_skipped ?? 0),
        };
      },
-      countRemaining: () => countExtractAtomsBacklog(engine, extractionSourceId),
+      countRemaining: (signal) => countExtractAtomsBacklog(engine, extractionSourceId, signal),
      now: Date.now,
      onBatch: opts.onBatch,
    },
-    { windowMs: opts.windowSeconds * 1000, maxBatches: opts.maxBatches },
+    {
+      windowMs: opts.windowSeconds * 1000,
+      maxBatches: opts.maxBatches,
+      abortSignal: opts.abortSignal,
+    },
  );
 }
@@ -58,6 +58,10 @@ import { createHash } from 'crypto';
 import { slugifySegment } from '../sync.ts';

 const DEFAULT_BUDGET_USD = 0.3;
+// #2750: fresh wallclock budget for the receipt/rollup bookkeeping writes when
+// the caller's deadline already fired — committed atoms must not lose their
+// cost/receipt trail, but the writes can't be unbounded either.
+const BOOKKEEPING_GRACE_MS = 5_000;

 // v0.42+ TODO: read atom_type enum from active pack manifest at runtime.
 const ATOM_TYPES = [
@@ -155,6 +159,13 @@ export interface ExtractAtomsOpts {
   * `heartbeat()` on the passed reporter.
   */
  progress?: ProgressReporter;
+  /**
+   * #2750: caller deadline/cancellation. Forwarded to every gateway call and
+   * DB query/write so the drain window bounds real lifetime, plus a
+   * cooperative between-item check (the PGLite path, where query abort only
+   * abandons the waiter).
+   */
+  abortSignal?: AbortSignal;
 }

 interface ExtractedAtom {
@@ -212,6 +223,7 @@ export async function discoverExtractablePages(
  engine: BrainEngine,
  sourceId: string,
  affectedSlugs?: string[],
+  abortSignal?: AbortSignal,
 ): Promise<DiscoveredPage[]> {
  const hasFilter = Array.isArray(affectedSlugs) && affectedSlugs.length > 0;
  const sql = `
@@ -251,13 +263,16 @@ export async function discoverExtractablePages(
      slug: string;
      compiled_truth: string;
      content_hash: string;
-    }>(sql, params);
+    }>(sql, params, { signal: abortSignal });
    return rows.map((r) => ({
      slug: r.slug,
      content: r.compiled_truth,
      contentHash: r.content_hash,
    }));
  } catch (err) {
+    // A deadline abort is not a fail-soft condition — propagate so the
+    // caller stops instead of proceeding with an empty page list.
+    if (abortSignal?.aborted) throw err;
    const msg = err instanceof Error ? err.message : String(err);
    console.error(`[extract_atoms] page-discovery query failed: ${msg}`);
    return []; // fail-soft: transcript path still proceeds
@@ -282,6 +297,7 @@ export async function discoverExtractablePages(
 export async function countExtractAtomsBacklog(
  engine: BrainEngine,
  sourceId?: string,
+  abortSignal?: AbortSignal,
 ): Promise<number | null> {
  try {
    // Two modes: scoped (the phase's per-source `remaining`) vs brain-wide
@@ -321,9 +337,10 @@ export async function countExtractAtomsBacklog(
    const params = scoped
      ? [sourceId, extractableTypes, MIN_PAGE_CHARS_FOR_EXTRACTION]
      : [extractableTypes, MIN_PAGE_CHARS_FOR_EXTRACTION];
-    const rows = await engine.executeRaw<{ cnt: string | number }>(sql, params);
+    const rows = await engine.executeRaw<{ cnt: string | number }>(sql, params, { signal: abortSignal });
    return Number(rows[0]?.cnt ?? 0);
  } catch (err) {
+    if (abortSignal?.aborted) throw err;
    const msg = err instanceof Error ? err.message : String(err);
    console.error(`[extract_atoms] backlog count failed: ${msg}`);
    return null;
@@ -350,6 +367,7 @@ export async function atomsExistingForHashes(
  engine: BrainEngine,
  sourceId: string,
  contentHash16s: string[],
+  abortSignal?: AbortSignal,
 ): Promise<Set<string>> {
  if (contentHash16s.length === 0) return new Set();
  try {
@@ -361,9 +379,11 @@ export async function atomsExistingForHashes(
          AND deleted_at IS NULL
          AND frontmatter->>'source_hash' = ANY($2::text[])`,
      [sourceId, contentHash16s],
+      { signal: abortSignal },
    );
    return new Set(rows.map(r => r.h));
  } catch (err) {
+    if (abortSignal?.aborted) throw err;
    const msg = err instanceof Error ? err.message : String(err);
    console.error(`[extract_atoms] batch idempotency check failed (assuming none extracted): ${msg}`);
    return new Set();
@@ -384,6 +404,7 @@ export async function runPhaseExtractAtoms(
 ): Promise<PhaseResult> {
  const sourceId = opts.sourceId ?? 'default';
  const chat = opts._chat ?? gatewayChat;
+  if (opts.abortSignal?.aborted) throw opts.abortSignal.reason;

  // 1a. Get transcripts (test seam OR production discovery).
  //     v0.41.2.1: config loader switched to loadConfigWithEngine() so the
@@ -425,7 +446,7 @@ export async function runPhaseExtractAtoms(
  if (opts._pages !== undefined) {
    pages = opts._pages;
  } else {
-    pages = await discoverExtractablePages(engine, sourceId, opts.affectedSlugs);
+    pages = await discoverExtractablePages(engine, sourceId, opts.affectedSlugs, opts.abortSignal);
  }

  // 2. Apply transcript-side source-hash idempotency in ONE batch query
@@ -437,7 +458,7 @@ export async function runPhaseExtractAtoms(
  // Surface a heartbeat before the batch query so even an instant
  // short-circuit shows a sign of life (closes Issue 2 silent-phase pain).
  opts.progress?.heartbeat(`checking existing atoms for ${allHashes16.length} transcripts`);
-  const existingHashes = await atomsExistingForHashes(engine, sourceId, allHashes16);
+  const existingHashes = await atomsExistingForHashes(engine, sourceId, allHashes16, opts.abortSignal);
  for (const t of transcripts) {
    if (existingHashes.has(t.contentHash.slice(0, 16))) {
      duplicatesSkipped++;
@@ -501,6 +522,7 @@ export async function runPhaseExtractAtoms(
  const failures: Array<{ source: string; error: string }> = [];
  let estimatedSpendUsd = 0;
  const budgetCap = DEFAULT_BUDGET_USD;
+  let deadlineAborted = false;

  // v0.41.19.0 (T3): throttled yield helper. Fires `opts.yieldDuringPhase`
  // every 30s. Cycle.ts threads `buildYieldDuringPhase(lock, outer)` so
@@ -526,6 +548,12 @@ export async function runPhaseExtractAtoms(
  }

  for (const item of work) {
+    // #2750: cooperative between-item abort. Works on every engine — this is
+    // the primary bound on PGLite, where query abort only abandons the waiter.
+    if (opts.abortSignal?.aborted) {
+      deadlineAborted = true;
+      break;
+    }
    await maybeYield();
    if (estimatedSpendUsd >= budgetCap) {
      if (item.kind === 'transcript') transcriptsSkipped++;
@@ -544,16 +572,22 @@ export async function runPhaseExtractAtoms(
          },
        ],
        maxTokens: 2000,
+        abortSignal: opts.abortSignal,
      });
+      // Rough cost estimate — Haiku at ~$0.80/M input + $4/M output.
+      // A completed gateway call is billable even if the deadline fires
+      // immediately afterward, so record usage BEFORE the abort check.
+      estimatedSpendUsd +=
+        (result.usage.input_tokens * 0.8 + result.usage.output_tokens * 4.0) / 1_000_000;
+      if (opts.abortSignal?.aborted) {
+        deadlineAborted = true;
+        break;
+      }
      // Post-await yield: closes the "long LLM call past TTL" hazard
      // codex flagged. The 30s throttle inside maybeYield bounds the
      // actual refresh rate so this is cheap when calls are fast.
      await maybeYield();

-      // Rough cost estimate — Haiku at ~$0.80/M input + $4/M output
-      estimatedSpendUsd +=
-        (result.usage.input_tokens * 0.8 + result.usage.output_tokens * 4.0) / 1_000_000;
-
      const atoms = parseAtomsResponse(result.text);
      if (atoms.length === 0) {
        if (item.kind === 'transcript') transcriptsProcessed++;
@@ -592,7 +626,7 @@ export async function runPhaseExtractAtoms(
              },
              timeline: '',
            },
-            { sourceId },
+            { sourceId, signal: opts.abortSignal },
          );
          totalAtomsExtracted++;
        }
@@ -605,6 +639,11 @@ export async function runPhaseExtractAtoms(
      // Reporter rate-limits to ~1 line/sec; safe to tick every iter.
      opts.progress?.tick(1, `${totalAtomsExtracted} atoms / ${duplicatesSkipped} skipped`);
    } catch (err) {
+      // A deadline abort is a partial result, not a per-item failure.
+      if (opts.abortSignal?.aborted) {
+        deadlineAborted = true;
+        break;
+      }
      failures.push({
        source: originLabel,
        error: err instanceof Error ? err.message : String(err),
@@ -615,6 +654,12 @@ export async function runPhaseExtractAtoms(
  // v0.42 Wave B2: write extract receipt + rollup row when the phase
  // actually extracted atoms. Both are best-effort per F-OUT-19 —
  // audit-trail / search-visibility surfaces don't block the phase result.
+  //
+  // #2750: bookkeeping runs on a FRESH short grace signal, never the caller's
+  // work deadline — the deadline may have already fired (partial run) and
+  // committed atoms must not lose their receipt/cost trail; but the writes
+  // stay bounded so the overrun is capped at the grace window.
+  const bookkeepingSignal = opts.dryRun ? undefined : AbortSignal.timeout(BOOKKEEPING_GRACE_MS);
  if (!opts.dryRun && totalAtomsExtracted > 0) {
    const runId = `atoms-${Date.now().toString(36)}-${sourceId.slice(0, 4)}`;
    try {
@@ -629,19 +674,24 @@ export async function runPhaseExtractAtoms(
        summary:
          `Extracted ${totalAtomsExtracted} atoms from ` +
          `${transcriptsProcessed} transcripts + ${pagesProcessed} pages.`,
-      });
+      }, { signal: bookkeepingSignal });
    } catch (err) {
      console.error(`[extract_atoms] receipt write failed: ${(err as Error).message}`);
    }
  }
  if (!opts.dryRun) {
-    await upsertExtractRollup(engine, {
-      kind: 'atoms',
-      source_id: sourceId,
-      cost_delta: estimatedSpendUsd,
-      round_completed_delta: failures.length === 0 ? 1 : 0,
-      halt_delta: failures.length > 0 ? 1 : 0,
-    });
+    try {
+      await upsertExtractRollup(engine, {
+        kind: 'atoms',
+        source_id: sourceId,
+        cost_delta: estimatedSpendUsd,
+        // A deadline-truncated run is not a completed round.
+        round_completed_delta: failures.length === 0 && !deadlineAborted ? 1 : 0,
+        halt_delta: failures.length > 0 ? 1 : 0,
+      }, { signal: bookkeepingSignal });
+    } catch (err) {
+      console.error(`[extract_atoms] rollup write failed: ${(err as Error).message}`);
+    }
  }

  return {
@@ -670,6 +720,7 @@ export async function runPhaseExtractAtoms(
      budget_usd: budgetCap,
      source_id: sourceId,
      dry_run: opts.dryRun ?? false,
+      deadline_aborted: deadlineAborted,
    },
  };
 }
@@ -26,7 +26,8 @@ import type { BrainEngine } from './engine.ts';

 export interface DbLockHandle {
  id: string;
-  release: () => Promise<void>;
+  /** Optional signal bounds the release DELETE (deadline-bound callers). */
+  release: (signal?: AbortSignal) => Promise<void>;
  refresh: () => Promise<void>;
 }

@@ -173,6 +174,7 @@ export async function tryAcquireDbLock(
  engine: BrainEngine,
  lockId: string,
  ttlMinutes: number = DEFAULT_TTL_MINUTES,
+  opts: { signal?: AbortSignal } = {},
 ): Promise<DbLockHandle | null> {
  const pid = process.pid;
  const host = hostname();
@@ -205,20 +207,26 @@ export async function tryAcquireDbLock(
    // `gbrain sync --break-lock --max-age <s>` uses last_refreshed_at (not
    // acquired_at) to identify wedged-but-alive holders without stealing
    // healthy long-running holders that are actively refreshing.
-    const rows: Array<{ id: string }> = await sql`
-      INSERT INTO gbrain_cycle_locks (id, holder_pid, holder_host, acquired_at, ttl_expires_at, last_refreshed_at)
-      VALUES (${lockId}, ${pid}, ${host}, NOW(), NOW() + ${ttl}::interval, NOW())
-      ON CONFLICT (id) DO UPDATE
-        SET holder_pid = ${pid},
-            holder_host = ${host},
-            acquired_at = NOW(),
-            ttl_expires_at = NOW() + ${ttl}::interval,
-            last_refreshed_at = NOW()
-        WHERE gbrain_cycle_locks.ttl_expires_at < NOW()
-          AND (gbrain_cycle_locks.last_refreshed_at IS NULL
-               OR gbrain_cycle_locks.last_refreshed_at < NOW() - ${stealGraceSeconds} * INTERVAL '1 second')
-      RETURNING id
-    `;
+    // #2750: routed through executeRaw so a deadline-bound caller's signal
+    // can cancel a hung acquire (pool exhaustion). Cancellation is
+    // transactional; in the rare ambiguous-commit case the row's TTL is the
+    // backstop (drain locks use a short 5-minute TTL).
+    const rows = await engine.executeRaw<{ id: string }>(
+      `INSERT INTO gbrain_cycle_locks (id, holder_pid, holder_host, acquired_at, ttl_expires_at, last_refreshed_at)
+       VALUES ($1, $2, $3, NOW(), NOW() + $4::interval, NOW())
+       ON CONFLICT (id) DO UPDATE
+         SET holder_pid = $2,
+             holder_host = $3,
+             acquired_at = NOW(),
+             ttl_expires_at = NOW() + $4::interval,
+             last_refreshed_at = NOW()
+         WHERE gbrain_cycle_locks.ttl_expires_at < NOW()
+           AND (gbrain_cycle_locks.last_refreshed_at IS NULL
+                OR gbrain_cycle_locks.last_refreshed_at < NOW() - $5 * INTERVAL '1 second')
+       RETURNING id`,
+      [lockId, pid, host, ttl, stealGraceSeconds],
+      { signal: opts.signal },
+    );
    if (rows.length === 0) return null;
    const deregister = registerCleanup(`db-lock:${lockId}`, async () => {
      await sql`
@@ -241,12 +249,17 @@ export async function tryAcquireDbLock(
          [ttl, lockId, pid],
        );
      },
-      release: async () => {
+      release: async (signal?: AbortSignal) => {
        deregister();
-        await sql`
-          DELETE FROM gbrain_cycle_locks
-          WHERE id = ${lockId} AND holder_pid = ${pid}
-        `;
+        // Direct session pool (same rationale as refresh, #1794) + optional
+        // signal so a deadline-bound caller's release can't hang forever on
+        // an exhausted pooler. TTL is the backstop if the DELETE is cancelled.
+        await engine.executeRawDirect(
+          `DELETE FROM gbrain_cycle_locks
+            WHERE id = $1 AND holder_pid = $2`,
+          [lockId, pid],
+          { signal },
+        );
      },
    };
  }
@@ -303,6 +316,11 @@ export async function tryAcquireDbLock(
  const first = await acquireOnce();
  if (first) return first;

+  // #2750: deadline-bound callers prefer an honest busy result over the
+  // best-effort same-host takeover below, whose inspect/delete/retry calls
+  // are not signal-bounded. The initial upsert already reclaims expired locks.
+  if (opts.signal) return null;
+
  // v0.42 (#1780 Gap 3): the lock is held and its TTL hasn't expired (the
  // upsert's ON CONFLICT ... WHERE ttl_expires_at < NOW() returned no row).
  // If the holder is on THIS host, provably dead, and past the grace window,
@@ -796,6 +814,10 @@ export interface WithRefreshingLockOpts {
  ttlMinutes?: number;
  /** Heartbeat-fail threshold in ms — abort if SELECT 1 takes longer. Default 30000. */
  heartbeatTimeoutMs?: number;
+  /** #2750: bound lock acquisition with the caller's deadline signal. */
+  signal?: AbortSignal;
+  /** Fresh cleanup budget for the release DELETE when `signal` is set. Default 5000. */
+  releaseTimeoutMs?: number;
 }

 /**
@@ -815,7 +837,7 @@ export async function withRefreshingLock<T>(
  // Refresh 6x per TTL window so a missed tick doesn't expire the lock.
  const refreshIntervalMs = Math.max(15000, (ttlMinutes * 60 * 1000) / 6);

-  const handle = await tryAcquireDbLock(engine, lockId, ttlMinutes);
+  const handle = await tryAcquireDbLock(engine, lockId, ttlMinutes, { signal: opts.signal });
  if (!handle) throw new LockUnavailableError(lockId);

  let healthOk = true;
@@ -854,7 +876,13 @@ export async function withRefreshingLock<T>(
    return await work();
  } finally {
    clearInterval(interval);
-    try { await handle.release(); } catch { /* idempotent */ }
+    // #2750: when the caller is deadline-bound, its work signal may already
+    // have fired — release on a FRESH short grace signal so cleanup neither
+    // inherits the spent deadline nor hangs unbounded. TTL is the backstop.
+    const releaseSignal = opts.signal
+      ? AbortSignal.timeout(opts.releaseTimeoutMs ?? 5_000)
+      : undefined;
+    try { await handle.release(releaseSignal); } catch { /* idempotent; TTL backstop */ }
    if (!healthOk) {
      // Surface that the heartbeat detected backend trouble — caller can
      // log to the connection-events audit if desired.
@@ -79,34 +79,15 @@ export interface EmbedBatchOptions {
   * and amplify rate-limit pressure.
   */
  maxRetries?: number;
-  /**
-   * #1818: bounded parallelism across BATCH_SIZE sub-batches. Defaults to
-   * `GBRAIN_EMBED_BATCH_CONCURRENCY` env, else 4. Results are
-   * index-addressed so output order always matches input order. Set 1 to
-   * force the pre-v0.42 serial dispatch.
-   */
-  concurrency?: number;
 }

 /**
 * Embed a batch of texts via the gateway. Sub-batches of 100 so upstream
 * progress callbacks fire incrementally on large imports. The gateway owns
 * adaptive batch splitting and per-recipe token-budget logic; this paginator
- * owns progress-callback granularity and (#1818) bounded parallel dispatch
- * of the sub-batches — the embed-stale.ts worker-pool pattern, scoped down.
+ * is purely about progress-callback granularity.
 */
 const BATCH_SIZE = 100;
-const DEFAULT_EMBED_BATCH_CONCURRENCY = 4;
-
-function resolveEmbedBatchConcurrency(options: EmbedBatchOptions): number {
-  if (options.concurrency !== undefined) {
-    return Math.max(1, Math.floor(options.concurrency));
-  }
-  const env = Number(process.env.GBRAIN_EMBED_BATCH_CONCURRENCY);
-  if (Number.isFinite(env) && env >= 1) return Math.floor(env);
-  return DEFAULT_EMBED_BATCH_CONCURRENCY;
-}
-
 export async function embedBatch(
  texts: string[],
  options: EmbedBatchOptions = {},
@@ -122,44 +103,13 @@ export async function embedBatch(
  if (texts.length <= BATCH_SIZE && !options.onBatchComplete) {
    return gatewayEmbed(texts, gwOpts);
  }
-  // #1818: dispatch sub-batches through a bounded worker pool instead of a
-  // serial loop. Results are written into a preallocated index-addressed
-  // array so output order matches input order regardless of completion
-  // order; onBatchComplete reports a monotonic completed-embedding count.
-  const slices: Array<{ start: number; texts: string[] }> = [];
+  const results: Float32Array[] = [];
  for (let i = 0; i < texts.length; i += BATCH_SIZE) {
-    slices.push({ start: i, texts: texts.slice(i, i + BATCH_SIZE) });
+    const slice = texts.slice(i, i + BATCH_SIZE);
+    const out = await gatewayEmbed(slice, gwOpts);
+    results.push(...out);
+    options.onBatchComplete?.(results.length, texts.length);
  }
-  const results = new Array<Float32Array>(texts.length);
-  let next = 0;
-  let done = 0;
-  const numWorkers = Math.min(resolveEmbedBatchConcurrency(options), slices.length);
-  // Once any sub-batch fails, `failed` stops the surviving workers from
-  // dispatching FURTHER slices — the whole call is rejecting anyway, so
-  // continuing would burn real provider spend in the background and fire
-  // onBatchComplete after the caller already saw the failure (worst with
-  // embedBatchWithBackoff, whose 429 backoff assumes nothing is in flight).
-  // In-flight sibling calls still run to completion (bounded by numWorkers-1).
-  let failed = false;
-  const worker = async (): Promise<void> => {
-    while (!failed && next < slices.length) {
-      // NOTE: no local aborted-check here — an aborted signal makes the next
-      // gatewayEmbed call throw (SDK-side), which rejects the pool. Returning
-      // silently instead would resolve with holes in `results`.
-      const slice = slices[next++];
-      let out: Float32Array[];
-      try {
-        out = await gatewayEmbed(slice.texts, gwOpts);
-      } catch (err) {
-        failed = true;
-        throw err;
-      }
-      for (let j = 0; j < out.length; j++) results[slice.start + j] = out[j];
-      done += out.length;
-      if (!failed) options.onBatchComplete?.(done, texts.length);
-    }
-  };
-  await Promise.all(Array.from({ length: numWorkers }, () => worker()));
  return results;
 }

@@ -696,8 +696,16 @@ export interface BrainEngine {
   * is included in the INSERT column list so ON CONFLICT (source_id, slug)
   * DO UPDATE actually targets the intended row instead of fabricating a
   * duplicate at (default, slug). Multi-source brains MUST pass sourceId.
+   *
+   * `opts.signal` (#2750): optional cancellation for deadline-bound writers.
+   * Postgres cancels the in-flight statement; PGLite pre-checks only (query
+   * cancellation is not possible in-process — cooperative abort between calls).
   */
-  putPage(slug: string, page: PageInput, opts?: { sourceId?: string }): Promise<Page>;
+  putPage(
+    slug: string,
+    page: PageInput,
+    opts?: { sourceId?: string; signal?: AbortSignal },
+  ): Promise<Page>;
  /**
   * v0.41.13 (#1309) — identity-based dedup pre-check for the import pipeline.
   *
@@ -187,6 +187,7 @@ function buildReceiptFrontmatter(input: ExtractReceiptInput): Record<string, unk
 export async function writeReceipt(
  engine: BrainEngine,
  input: ExtractReceiptInput,
+  opts?: { signal?: AbortSignal },
 ): Promise<{ slug: string; page: Page }> {
  const slug = receiptSlug(input);
  const title = `${input.kind} — ${input.round} — ${input.source_id}`;
@@ -201,7 +202,7 @@ export async function writeReceipt(
      compiled_truth,
      frontmatter,
    },
-    { sourceId: input.source_id },
+    { sourceId: input.source_id, signal: opts?.signal },
  );

  return { slug, page };
@@ -70,6 +70,7 @@ function today(): string {
 export async function upsertExtractRollup(
  engine: BrainEngine,
  input: RollupUpsertInput,
+  opts?: { signal?: AbortSignal },
 ): Promise<{ ok: boolean; error?: string }> {
  const day = input.day ?? today();
  const cost = input.cost_delta ?? 0;
@@ -96,9 +97,12 @@ export async function upsertExtractRollup(
         rollup_write_failures  = extract_rollup_7d.rollup_write_failures  + EXCLUDED.rollup_write_failures,
         updated_at             = now()`,
      [input.kind, input.source_id, day, cost, halts, evalFails, evalPasses, completed, failures],
+      { signal: opts?.signal },
    );
    return { ok: true };
  } catch (err) {
+    // Signal-bounded callers get the abort surfaced, not a swallowed `ok:false`.
+    if (opts?.signal?.aborted) throw err;
    const msg = (err as Error).message || String(err);
    // Don't spam: log once per process per (kind, day) error class.
    rollupErrorLogOnce(input.kind, day, msg);
@@ -1003,7 +1003,15 @@ export class PGLiteEngine implements BrainEngine {
    return { slug: r.slug, id: Number(r.id) };
  }

-  async putPage(slug: string, page: PageInput, opts?: { sourceId?: string }): Promise<Page> {
+  async putPage(
+    slug: string,
+    page: PageInput,
+    opts?: { sourceId?: string; signal?: AbortSignal },
+  ): Promise<Page> {
+    // #2750: PGLite is in-process WASM — no query cancellation. Pre-check so
+    // an already-fired deadline skips the write; abort is cooperative
+    // between calls (same posture as executeRaw's documented gap).
+    if (opts?.signal?.aborted) throw new DOMException('aborted', 'AbortError');
    slug = validateSlug(slug);
    const hash = page.content_hash || contentHash(page);
    const frontmatter = page.frontmatter || {};
@@ -72,6 +72,32 @@ function escapeSqlStringLiteral(value: string): string {
  return value.replace(/'/g, "''");
 }

+/**
+ * #2750: race a promise against an AbortSignal, detaching the listener once
+ * settled (long-lived drain signals are reused across many calls, so a bare
+ * Promise.race would leak one listener per call). The abandoned promise keeps
+ * running; used only for pool-acquisition waits where that is harmless.
+ */
+function waitForSignal<T>(work: Promise<T>, signal?: AbortSignal): Promise<T> {
+  if (!signal) return work;
+  if (signal.aborted) return Promise.reject(new DOMException('aborted', 'AbortError'));
+  return new Promise<T>((resolve, reject) => {
+    let settled = false;
+    const finish = (fn: () => void) => {
+      if (settled) return;
+      settled = true;
+      signal.removeEventListener('abort', onAbort);
+      fn();
+    };
+    const onAbort = () => finish(() => reject(new DOMException('aborted', 'AbortError')));
+    signal.addEventListener('abort', onAbort, { once: true });
+    work.then(
+      (value) => finish(() => resolve(value)),
+      (err) => finish(() => reject(err)),
+    );
+  });
+}
+
 export function getPostgresSchema(
  dims: number = DEFAULT_EMBEDDING_DIMENSIONS,
  model: string = DEFAULT_EMBEDDING_MODEL,
@@ -1061,7 +1087,12 @@ export class PostgresEngine implements BrainEngine {
    });
  }

-  async putPage(slug: string, page: PageInput, opts?: { sourceId?: string }): Promise<Page> {
+  async putPage(
+    slug: string,
+    page: PageInput,
+    opts?: { sourceId?: string; signal?: AbortSignal },
+  ): Promise<Page> {
+    if (opts?.signal?.aborted) throw new DOMException('aborted', 'AbortError');
    slug = validateSlug(slug);
    const sql = this.sql;
    const hash = page.content_hash || contentHash(page);
@@ -1096,7 +1127,7 @@ export class PostgresEngine implements BrainEngine {
    const sourceUri = page.source_uri ?? null;
    const ingestedVia = page.ingested_via ?? null;
    const ingestedAt = (sourceKind || sourceUri || ingestedVia) ? new Date() : null;
-    const rows = await sql`
+    const pending = sql`
      INSERT INTO pages (source_id, slug, type, page_kind, title, compiled_truth, timeline, frontmatter, content_hash, updated_at, effective_date, effective_date_source, import_filename, chunker_version, source_path, source_kind, source_uri, ingested_via, ingested_at)
      VALUES (${sourceId}, ${slug}, ${page.type}, ${pageKind}, ${page.title}, ${page.compiled_truth}, ${page.timeline || ''}, ${sql.json(frontmatter as Parameters<typeof sql.json>[0])}, ${hash}, now(), ${effectiveDate}, ${effectiveDateSource}, ${importFilename}, COALESCE(${chunkerVersion}::smallint, ${MARKDOWN_CHUNKER_VERSION}), ${sourcePath}, ${sourceKind}, ${sourceUri}, ${ingestedVia}, ${ingestedAt})
      ON CONFLICT (source_id, slug) DO UPDATE SET
@@ -1119,6 +1150,22 @@ export class PostgresEngine implements BrainEngine {
        ingested_at           = COALESCE(EXCLUDED.ingested_at,           pages.ingested_at)
      RETURNING id, source_id, slug, type, title, compiled_truth, timeline, frontmatter, content_hash, created_at, updated_at, effective_date, effective_date_source, import_filename, source_kind, source_uri, ingested_via, ingested_at
    `;
+    // #2750: cancel the in-flight statement when the caller's deadline fires,
+    // same .cancel() wiring as runUnsafe (postgres.js pending queries).
+    if (opts?.signal) {
+      const signal = opts.signal;
+      const onAbort = () => {
+        try { (pending as unknown as { cancel?: () => void }).cancel?.(); } catch { /* best-effort */ }
+      };
+      signal.addEventListener('abort', onAbort, { once: true });
+      try {
+        const rows = await pending;
+        return rowToPage(rows[0]);
+      } finally {
+        signal.removeEventListener('abort', onAbort);
+      }
+    }
+    const rows = await pending;
    return rowToPage(rows[0]);
  }

@@ -5807,11 +5854,16 @@ export class PostgresEngine implements BrainEngine {
    params?: unknown[],
    opts?: { signal?: AbortSignal },
  ): Promise<T[]> {
+    // #2750: an already-fired signal short-circuits BEFORE any pool routing,
+    // and the direct-pool acquisition itself is signal-bounded — under pooler
+    // exhaustion `ddl()` can stall indefinitely, which used to make even a
+    // "bounded" lock release hang past its caller's deadline.
+    if (opts?.signal?.aborted) throw new DOMException('aborted', 'AbortError');
    // Inside an open transaction, _sql is the reserved tx connection (set via
    // defineProperty in transaction()); never reroute off it.
    const inTransaction = this._sql !== null && this.connectionManager?.peekReadPool() !== this._sql;
    const conn = (!inTransaction && this.connectionManager?.isDualPoolActive())
-      ? await this.connectionManager.ddl()
+      ? await waitForSignal(this.connectionManager.ddl(), opts?.signal)
      : this.sql;
    return this.runUnsafe<T>(conn, sql, params, opts);
  }
@@ -39,8 +39,6 @@ import {
  __getShrinkStateForTests,
 } from '../../src/core/ai/gateway.ts';
 import { AIConfigError, AITransientError } from '../../src/core/ai/errors.ts';
-import { RECIPES } from '../../src/core/ai/recipes/index.ts';
-import type { Recipe } from '../../src/core/ai/types.ts';

 // The last test in this file leaves the gateway configured with a remote
 // provider + fake key and a REAL embed transport. Without a final reset,
@@ -95,14 +93,6 @@ function configureGoogle(): void {
  });
 }

-function configureDashscope(): void {
-  configureGateway({
-    embedding_model: 'dashscope:text-embedding-v3',
-    embedding_dimensions: 1024,
-    env: { DASHSCOPE_API_KEY: 'sk-fake' },
-  });
-}
-
 // --------- 1. Pure helpers ---------

 describe('splitByTokenBudget (pure helper)', () => {
@@ -159,27 +149,6 @@ describe('splitByTokenBudget (pure helper)', () => {
    expect(splitByTokenBudget(texts, 96_000, 0)).toEqual(splitByTokenBudget(texts, 96_000, 4));
    expect(splitByTokenBudget(texts, 96_000, -1)).toEqual(splitByTokenBudget(texts, 96_000, 4));
  });
-
-  // #1199: count cap for providers that reject batches by input count.
-  test('max_batch_count flushes even when token budget has room', () => {
-    const texts = Array.from({ length: 25 }, (_, i) => `t${i}`);
-    const result = splitByTokenBudget(texts, 1_000_000, 4, 10);
-    expect(result.map(b => b.length)).toEqual([10, 10, 5]);
-    expect(result.flat()).toEqual(texts);
-  });
-
-  test('token budget still governs alongside max_batch_count', () => {
-    const texts = ['a'.repeat(50_000), 'b'.repeat(50_000), 'c'.repeat(50_000)];
-    const result = splitByTokenBudget(texts, 96_000, 1, 10);
-    expect(result).toHaveLength(3);
-  });
-
-  test('undefined / zero / negative max_batch_count is ignored', () => {
-    const texts = Array.from({ length: 25 }, () => 'x');
-    expect(splitByTokenBudget(texts, 1_000_000, 4, undefined)).toHaveLength(1);
-    expect(splitByTokenBudget(texts, 1_000_000, 4, 0)).toHaveLength(1);
-    expect(splitByTokenBudget(texts, 1_000_000, 4, -5)).toHaveLength(1);
-  });
 });

 describe('isTokenLimitError (pure helper)', () => {
@@ -210,12 +179,6 @@ describe('isTokenLimitError (pure helper)', () => {
    expect(isTokenLimitError(new Error('Exceeded 300000 max tokens per request'))).toBe(true);
  });

-  test('matches DashScope batch-count error (#1199)', () => {
-    expect(isTokenLimitError(new Error(
-      'InvalidParameter: batch size is invalid, it should not be larger than 10.',
-    ))).toBe(true);
-  });
-
  test('does not match unrelated errors', () => {
    expect(isTokenLimitError(new Error('Connection refused'))).toBe(false);
    expect(isTokenLimitError(new Error('Invalid API key'))).toBe(false);
@@ -424,92 +387,26 @@ describe('shrink-on-miss adaptive cache', () => {
  });
 });

-// --------- 8. Pre-split count cap through public embed() (#1199 / #970) ---------
-
-describe('embed() pre-split honors max_batch_count', () => {
-  beforeEach(() => resetGateway());
-  afterEach(() => __setEmbedTransportForTests(null));
-
-  test('dashscope never dispatches more than 10 inputs per call (#1199)', async () => {
-    configureDashscope();
-    const stub = mock(async ({ values }: { values: string[] }) => fakeEmbeddings(values, 1024));
-    __setEmbedTransportForTests(stub as any);
-
-    // 25 short texts fit trivially in the 8192-token budget; without the
-    // count cap they'd ship as ONE batch and DashScope would reject it.
-    const texts = Array.from({ length: 25 }, (_, i) => `short-${i}`);
-    const result = await embed(texts);
-
-    expect(result).toHaveLength(25);
-    const callLengths = stub.mock.calls.map(([arg]) => (arg as { values: string[] }).values.length);
-    expect(Math.max(...callLengths)).toBeLessThanOrEqual(10);
-    expect(callLengths.reduce((a, b) => a + b, 0)).toBe(25);
-    // Order preserved across sub-batches.
-    expect((stub.mock.calls[0][0] as { values: string[] }).values[0]).toBe('short-0');
-  });
-
-  test('google pre-splits at 100 inputs per batchEmbedContents call (#970)', async () => {
-    configureGoogle();
-    const stub = mock(async ({ values }: { values: string[] }) => fakeEmbeddings(values, 768));
-    __setEmbedTransportForTests(stub as any);
-
-    const texts = Array.from({ length: 250 }, (_, i) => `g${i}`);
-    const result = await embed(texts);
-
-    expect(result).toHaveLength(250);
-    const callLengths = stub.mock.calls.map(([arg]) => (arg as { values: string[] }).values.length);
-    expect(callLengths).toEqual([100, 100, 50]);
-  });
-});
-
 // --------- 7. Startup warning (D9-B) ---------

 describe('startup warning for recipes missing max_batch_tokens', () => {
  beforeEach(() => resetGateway());

-  // #970 closed google's missing cap, so no registered recipe is capless
-  // anymore. Inject a synthetic capless recipe to keep the warning path
-  // covered for the NEXT recipe that forgets the field.
-  const caplessRecipe: Recipe = {
-    id: 'capless-test',
-    name: 'Capless Test Provider',
-    tier: 'openai-compat',
-    implementation: 'openai-compatible',
-    base_url_default: 'https://example.invalid/v1',
-    auth_env: { required: [] },
-    touchpoints: {
-      embedding: { models: ['capless-embed-1'], default_dims: 768 },
-    },
-  };
-
-  function configureCapless(): void {
-    configureGateway({
-      embedding_model: 'capless-test:capless-embed-1',
-      embedding_dimensions: 768,
-      env: {},
-    });
-  }
-
  test('configured missing-cap recipe warns once; unrelated recipes stay quiet', () => {
    const warnings: string[] = [];
    const original = console.warn;
    console.warn = (msg: string) => warnings.push(String(msg));
-    RECIPES.set(caplessRecipe.id, caplessRecipe);
    try {
      configureOpenAI();
      expect(warnings.length).toBe(0);
-      // #970 regression: google now declares max_batch_tokens → quiet.
      configureGoogle();
-      expect(warnings.length).toBe(0);
-      configureCapless();
      const firstCallCount = warnings.length;
      // Reconfigure: the warning should NOT re-fire for the same recipes
      // within one process (we already told the operator).
-      configureCapless();
+      configureGoogle();
      expect(warnings.length).toBe(firstCallCount);
    } finally {
      console.warn = original;
-      RECIPES.delete(caplessRecipe.id);
    }

    // The warning text should match the documented contract.
@@ -518,12 +415,11 @@ describe('startup warning for recipes missing max_batch_tokens', () => {
    );
    expect(contractMatch.length).toBe(1);

-    // Voyage + google declare max_batch_tokens → suppressed. OpenAI is the
-    // canonical fast-path recipe → also suppressed by id. All must be
-    // absent from the warnings; only the synthetic capless recipe fires.
+    // Voyage declares max_batch_tokens → suppressed. OpenAI is the
+    // canonical fast-path recipe → also suppressed by id. Both must be
+    // absent from the warnings.
    expect(warnings.find(w => w.includes('"voyage"'))).toBeUndefined();
    expect(warnings.find(w => w.includes('"openai"'))).toBeUndefined();
-    expect(warnings.find(w => w.includes('"google"'))).toBeUndefined();
-    expect(warnings.find(w => w.includes('"capless-test"'))).toBeDefined();
+    expect(warnings.find(w => w.includes('"google"'))).toBeDefined();
  });
 });
@@ -52,7 +52,16 @@ describe('v0.32 #779: no_batch_cap suppresses the missing-max_batch_tokens warni
    }
  });

-  test('configureGateway does NOT warn for google now that it declares batch caps (#970)', () => {
+  test('configureGateway warns for google only when google embedding is configured', () => {
+    warnSpy.mockClear();
+    resetGateway();
+    configureGateway({ env: {} });
+    let messages = warnSpy.mock.calls.map(c => String(c[0] ?? ''));
+    expect(
+      messages.some(m => m.includes('"google"') && m.includes('without max_batch_tokens')),
+      'google should not warn while OpenAI default is configured',
+    ).toBe(false);
+
    warnSpy.mockClear();
    resetGateway();
    configureGateway({
@@ -60,20 +69,11 @@ describe('v0.32 #779: no_batch_cap suppresses the missing-max_batch_tokens warni
      embedding_dimensions: 768,
      env: { GOOGLE_GENERATIVE_AI_API_KEY: 'fake' },
    });
-    const messages = warnSpy.mock.calls.map(c => String(c[0] ?? ''));
+    messages = warnSpy.mock.calls.map(c => String(c[0] ?? ''));
    expect(
      messages.some(m => m.includes('"google"') && m.includes('without max_batch_tokens')),
-      'google declares max_batch_tokens/max_batch_count since #970 — no warning',
-    ).toBe(false);
-  });
-
-  test('google recipe declares its derived batch caps (#970)', () => {
-    const e = getRecipe('google')!.touchpoints.embedding!;
-    // Count cap is the REAL Gemini limit (batchEmbedContents: 100 inputs);
-    // the token budget is derived (100 × 2048 per-input tokens), NOT the
-    // 2048 per-input limit — copying that verbatim would over-split 50×.
-    expect(e.max_batch_count).toBe(100);
-    expect(e.max_batch_tokens).toBe(204_800);
+      'google should warn when configured because it has fixed-cap models',
+    ).toBe(true);
  });

  test('every recipe with empty models[] declares user_provided_models OR has openai-fast-path', () => {
@@ -55,11 +55,6 @@ describe('recipe: dashscope', () => {
    expect(r.touchpoints.embedding!.chars_per_token).toBeGreaterThan(0);
  });

-  test('declares max_batch_count: 10 — DashScope rejects larger batches (#1199)', () => {
-    const r = getRecipe('dashscope')!;
-    expect(r.touchpoints.embedding!.max_batch_count).toBe(10);
-  });
-
  test('dimsProviderOptions threads dimensions for text-embedding-v3 (Matryoshka)', async () => {
    // Codex finding #1: DashScope text-embedding-v3 is Matryoshka 64-1024.
    // Without `dimensions` on the wire, user-selected non-default dims are
@@ -17,6 +17,7 @@ import { runPhaseExtractAtoms, parseAtomsResponse } from '../../src/core/cycle/e
 import { runPhaseSynthesizeConcepts } from '../../src/core/cycle/synthesize-concepts.ts';
 import { resetPgliteState } from '../helpers/reset-pglite.ts';
 import type { ChatResult, ChatOpts } from '../../src/core/ai/gateway.ts';
+import type { BrainEngine } from '../../src/core/engine.ts';

 let engine: PGLiteEngine;

@@ -177,6 +178,180 @@ describe('v0.41 T5: runPhaseExtractAtoms via stubbed chat', () => {
    expect((result.details?.failures as unknown[]).length).toBe(1);
  });

+  // ── #2750: caller deadline bounds the phase ────────────────────────────
+
+  test('caller deadline aborts a hung chat before processing the next item', async () => {
+    let calls = 0;
+    const chat = async (opts: ChatOpts) => {
+      calls++;
+      return await new Promise<never>((_resolve, reject) => {
+        const signal = opts.abortSignal;
+        if (!signal) return reject(new Error('missing abort signal'));
+        if (signal.aborted) return reject(signal.reason);
+        signal.addEventListener('abort', () => reject(signal.reason), { once: true });
+      });
+    };
+    const started = Date.now();
+    const result = await runPhaseExtractAtoms(engine, {
+      _transcripts: [
+        { filePath: '/hung.txt', content: 'a', contentHash: 'hung-a' },
+        { filePath: '/never.txt', content: 'b', contentHash: 'hung-b' },
+      ],
+      _pages: [],
+      _chat: chat as typeof import('../../src/core/ai/gateway.ts').chat,
+      abortSignal: AbortSignal.timeout(25),
+    });
+    expect(Date.now() - started).toBeLessThan(2_000);
+    expect(calls).toBe(1);
+    expect(result.status).toBe('ok');
+    expect(result.details?.deadline_aborted).toBe(true);
+    expect(result.details?.atoms_extracted).toBe(0);
+    expect(result.details?.failures).toEqual([]);
+  });
+
+  test('billable chat usage is counted when the deadline fires as the response resolves', async () => {
+    const controller = new AbortController();
+    const chat = async (opts: ChatOpts): Promise<ChatResult> => {
+      controller.abort(new DOMException('deadline', 'TimeoutError'));
+      return stubChat(`[{"title":"late","atom_type":"insight","body":"b"}]`, {
+        input_tokens: 1_000,
+        output_tokens: 500,
+      })(opts);
+    };
+    const result = await runPhaseExtractAtoms(engine, {
+      _transcripts: [{ filePath: '/late.txt', content: 'a', contentHash: 'late' }],
+      _pages: [],
+      _chat: chat,
+      abortSignal: controller.signal,
+    });
+    expect(result.details?.deadline_aborted).toBe(true);
+    expect(Number(result.details?.estimated_spend_usd)).toBeGreaterThan(0);
+    expect(result.details?.atoms_extracted).toBe(0);
+  });
+
+  test('deadline after partial progress still writes receipt and incomplete rollup', async () => {
+    const controller = new AbortController();
+    let calls = 0;
+    let notifySecondChat!: () => void;
+    const secondChatStarted = new Promise<void>((resolve) => { notifySecondChat = resolve; });
+    const chat = async (opts: ChatOpts): Promise<ChatResult> => {
+      calls++;
+      if (calls === 1) {
+        return stubChat(`[{"title":"committed","atom_type":"insight","body":"b"}]`)(opts);
+      }
+      notifySecondChat();
+      return await new Promise<never>((_resolve, reject) => {
+        const signal = opts.abortSignal;
+        if (!signal) return reject(new Error('missing abort signal'));
+        signal.addEventListener('abort', () => reject(signal.reason), { once: true });
+      });
+    };
+
+    const pending = runPhaseExtractAtoms(engine, {
+      _transcripts: [
+        { filePath: '/committed.txt', content: 'a', contentHash: 'committed-a' },
+        { filePath: '/hung.txt', content: 'b', contentHash: 'hung-b' },
+      ],
+      _pages: [],
+      _chat: chat,
+      abortSignal: controller.signal,
+    });
+    await secondChatStarted;
+    controller.abort(new DOMException('deadline', 'TimeoutError'));
+    const result = await pending;
+
+    const atoms = await engine.executeRaw<{ n: number }>(
+      `SELECT COUNT(*)::int AS n FROM pages WHERE type = 'atom'`,
+    );
+    const receipts = await engine.executeRaw<{ n: number }>(
+      `SELECT COUNT(*)::int AS n FROM pages WHERE type = 'extract_receipt'`,
+    );
+    const rollups = await engine.executeRaw<{
+      cost_usd: string | number;
+      round_completed_count: string | number;
+    }>(
+      `SELECT cost_usd, round_completed_count
+         FROM extract_rollup_7d
+        WHERE kind = 'atoms' AND source_id = 'default'`,
+    );
+    expect(result.details?.deadline_aborted).toBe(true);
+    expect(atoms[0].n).toBe(1);
+    expect(receipts[0].n).toBe(1);
+    expect(Number(rollups[0].cost_usd)).toBeGreaterThan(0);
+    expect(Number(rollups[0].round_completed_count)).toBe(0);
+  });
+
+  test('bookkeeping runs on a fresh grace signal, not the fired work deadline', async () => {
+    const controller = new AbortController();
+    let putCalls = 0;
+    let receiptSignal: AbortSignal | undefined;
+    let rollupSignal: AbortSignal | undefined;
+    const signalAwareEngine = {
+      executeRaw: async (sql: string, _params?: unknown[], opts?: { signal?: AbortSignal }) => {
+        if (sql.includes('INSERT INTO extract_rollup_7d')) rollupSignal = opts?.signal;
+        return [];
+      },
+      putPage: async (_slug: string, _page: unknown, opts?: { signal?: AbortSignal }) => {
+        putCalls++;
+        if (putCalls === 1) {
+          // Atom write in flight; the work deadline fires before bookkeeping.
+          controller.abort(new DOMException('work deadline', 'TimeoutError'));
+        } else {
+          receiptSignal = opts?.signal;
+        }
+        return {};
+      },
+    } as unknown as BrainEngine;
+
+    const result = await runPhaseExtractAtoms(signalAwareEngine, {
+      _transcripts: [{ filePath: '/one.txt', content: 'a', contentHash: 'one' }],
+      _pages: [],
+      _chat: stubChat(`[{"title":"one","atom_type":"insight","body":"b"}]`),
+      abortSignal: controller.signal,
+    });
+
+    expect(result.details?.atoms_extracted).toBe(1);
+    expect(putCalls).toBe(2); // atom write + receipt write
+    expect(receiptSignal).toBeDefined();
+    expect(receiptSignal).not.toBe(controller.signal);
+    expect(receiptSignal?.aborted).toBe(false);
+    expect(rollupSignal).toBe(receiptSignal);
+  });
+
+  test('caller deadline cancels a hung atom write and stops the phase', async () => {
+    const controller = new AbortController();
+    let notifyWriteStarted!: () => void;
+    const writeStarted = new Promise<void>((resolve) => { notifyWriteStarted = resolve; });
+    let writeCalls = 0;
+    const signalAwareEngine = {
+      executeRaw: async () => [],
+      putPage: async (_slug: string, _page: unknown, opts?: { signal?: AbortSignal }) => {
+        writeCalls++;
+        notifyWriteStarted();
+        return await new Promise<never>((_resolve, reject) => {
+          const signal = opts?.signal;
+          if (!signal) return reject(new Error('missing abort signal'));
+          if (signal.aborted) return reject(signal.reason);
+          signal.addEventListener('abort', () => reject(signal.reason), { once: true });
+        });
+      },
+    } as unknown as BrainEngine;
+
+    const pending = runPhaseExtractAtoms(signalAwareEngine, {
+      _transcripts: [{ filePath: '/hung-write.txt', content: 'a', contentHash: 'hung-write' }],
+      _pages: [],
+      _chat: stubChat(`[{"title":"hung write","atom_type":"insight","body":"b"}]`),
+      abortSignal: controller.signal,
+    });
+    await writeStarted;
+    controller.abort(new DOMException('deadline', 'TimeoutError'));
+    const result = await pending;
+
+    expect(writeCalls).toBe(1);
+    expect(result.details?.deadline_aborted).toBe(true);
+    expect(result.details?.atoms_extracted).toBe(0);
+  });
+
  // v0.41.2.1 regression case (D9 #14 wording): with _pages:[] and same
  // _transcripts, all PRE-EXISTING PhaseResult.details fields match
  // pre-fix values byte-for-byte. The new fields (pages_processed,
@@ -1,161 +0,0 @@
-/**
- * #1818: embedBatch dispatches its 100-input sub-batches through a bounded
- * worker pool (the embed-stale.ts concurrency pattern) instead of a serial
- * `for` loop. This file pins:
- *
- *   - output order matches input order regardless of completion order
- *     (index-addressed results)
- *   - parallelism actually happens (max in-flight > 1) and stays bounded
- *     (max in-flight <= configured concurrency)
- *   - concurrency: 1 restores the serial pre-#1818 dispatch
- *   - GBRAIN_EMBED_BATCH_CONCURRENCY env is honored when the option is unset
- *   - onBatchComplete reports a monotonic completed count ending at total
- *
- * Transport is stubbed via the gateway's __setEmbedTransportForTests seam
- * (same pattern as test/ai/adaptive-embed-batch.test.ts). OpenAI recipe =
- * fast path (no pre-split), so each embedBatch sub-batch is exactly one
- * transport call.
- */
-
-import { afterAll, afterEach, beforeEach, describe, expect, test } from 'bun:test';
-import {
-  configureGateway,
-  resetGateway,
-  __setEmbedTransportForTests,
-} from '../src/core/ai/gateway.ts';
-import { embedBatch } from '../src/core/embedding.ts';
-import { withEnv } from './helpers/with-env.ts';
-
-const DIMS = 1536;
-
-function configureOpenAI(): void {
-  configureGateway({
-    embedding_model: 'openai:text-embedding-3-large',
-    embedding_dimensions: DIMS,
-    env: { OPENAI_API_KEY: 'sk-fake' },
-  });
-}
-
-/**
- * Install a transport whose returned embedding encodes the GLOBAL input
- * index in dim 0 (texts are `t<N>`), so order can be asserted end-to-end.
- * Tracks the max number of concurrently in-flight transport calls.
- */
-function installTrackingTransport(delayMs = 5): { maxInFlight: () => number } {
-  let inFlight = 0;
-  let maxInFlight = 0;
-  __setEmbedTransportForTests((async ({ values }: { values: string[] }) => {
-    inFlight++;
-    maxInFlight = Math.max(maxInFlight, inFlight);
-    await new Promise(r => setTimeout(r, delayMs));
-    inFlight--;
-    return {
-      embeddings: values.map(v => {
-        const idx = Number(v.slice(1));
-        return Array.from({ length: DIMS }, (_, j) => (j === 0 ? idx : 0.1));
-      }),
-    };
-  }) as any);
-  return { maxInFlight: () => maxInFlight };
-}
-
-const texts = Array.from({ length: 250 }, (_, i) => `t${i}`);
-
-afterAll(() => resetGateway());
-
-describe('embedBatch bounded parallelism (#1818)', () => {
-  beforeEach(() => {
-    resetGateway();
-    configureOpenAI();
-  });
-  afterEach(() => {
-    __setEmbedTransportForTests(null);
-  });
-
-  test('default pool dispatches sub-batches in parallel, order preserved', async () => {
-    const tracker = installTrackingTransport();
-    const result = await embedBatch(texts, { onBatchComplete: () => {} });
-    expect(result).toHaveLength(250);
-    for (let i = 0; i < 250; i++) {
-      expect(result[i][0]).toBe(i);
-    }
-    // 250 texts → 3 sub-batches; default concurrency 4 → all 3 in flight.
-    expect(tracker.maxInFlight()).toBeGreaterThan(1);
-    expect(tracker.maxInFlight()).toBeLessThanOrEqual(4);
-  });
-
-  test('concurrency: 1 keeps the serial dispatch', async () => {
-    const tracker = installTrackingTransport();
-    const result = await embedBatch(texts, { concurrency: 1, onBatchComplete: () => {} });
-    expect(result).toHaveLength(250);
-    expect(tracker.maxInFlight()).toBe(1);
-  });
-
-  test('GBRAIN_EMBED_BATCH_CONCURRENCY env bounds the pool when option unset', async () => {
-    const tracker = installTrackingTransport();
-    await withEnv({ GBRAIN_EMBED_BATCH_CONCURRENCY: '2' }, async () => {
-      await embedBatch(texts, { onBatchComplete: () => {} });
-    });
-    expect(tracker.maxInFlight()).toBeGreaterThan(1);
-    expect(tracker.maxInFlight()).toBeLessThanOrEqual(2);
-  });
-
-  test('onBatchComplete reports a monotonic count ending at total', async () => {
-    installTrackingTransport();
-    const seen: number[] = [];
-    await embedBatch(texts, {
-      onBatchComplete: (done, total) => {
-        expect(total).toBe(250);
-        seen.push(done);
-      },
-    });
-    expect(seen).toHaveLength(3); // 100 + 100 + 50 sub-batches
-    for (let i = 1; i < seen.length; i++) {
-      expect(seen[i]).toBeGreaterThan(seen[i - 1]);
-    }
-    expect(seen[seen.length - 1]).toBe(250);
-  });
-
-  test('a failing sub-batch rejects the whole call', async () => {
-    let call = 0;
-    __setEmbedTransportForTests((async ({ values }: { values: string[] }) => {
-      call++;
-      if (call === 2) throw new Error('boom');
-      await new Promise(r => setTimeout(r, 2));
-      return { embeddings: values.map(() => Array.from({ length: DIMS }, () => 0.1)) };
-    }) as any);
-    await expect(embedBatch(texts, { onBatchComplete: () => {} })).rejects.toThrow();
-  });
-
-  test('after a failure, surviving workers stop dispatching new slices', async () => {
-    // 1000 texts → 10 slices, concurrency 2. First call fails immediately;
-    // without the `failed` flag the second worker would keep draining all
-    // 10 slices in the background AFTER embedBatch already rejected —
-    // burning provider spend and firing onBatchComplete post-rejection.
-    let calls = 0;
-    const completions: number[] = [];
-    __setEmbedTransportForTests((async ({ values }: { values: string[] }) => {
-      calls++;
-      if (calls === 1) throw new Error('boom');
-      await new Promise(r => setTimeout(r, 5));
-      return { embeddings: values.map(() => Array.from({ length: DIMS }, () => 0.1)) };
-    }) as any);
-    const many = Array.from({ length: 1000 }, (_, i) => `t${i}`);
-    await expect(
-      embedBatch(many, { concurrency: 2, onBatchComplete: d => completions.push(d) }),
-    ).rejects.toThrow('boom');
-    const callsAtRejection = calls;
-    await new Promise(r => setTimeout(r, 50)); // would-be background drain window
-    expect(calls).toBe(callsAtRejection); // no new dispatch after rejection
-    expect(calls).toBeLessThanOrEqual(2); // only the in-flight sibling ran
-    expect(completions).toHaveLength(0); // no progress reported after failure
-  });
-
-  test('single small batch without callback stays on the one-call fast path', async () => {
-    const tracker = installTrackingTransport(1);
-    const result = await embedBatch(['t0', 't1', 't2']);
-    expect(result).toHaveLength(3);
-    expect(result[1][0]).toBe(1);
-    expect(tracker.maxInFlight()).toBe(1);
-  });
-});
@@ -43,26 +43,135 @@ describe('runExtractAtomsDrain (issue #1678)', () => {
    expect(batches).toBe(3);
  });

-  it('stops at the wallclock window with remaining > 0', async () => {
-    // SYNC stepping clock: now() #1 sets deadline (0+100=100); the while-check
-    // then sees 50, 50 (two batches), then 999999 → past deadline → stop.
-    const times = [0, 50, 50, 999_999];
-    let ti = 0;
-    const now = () => times[Math.min(ti++, times.length - 1)];
+  it('stops at the wallclock window; remaining is unknown (no post-window count)', async () => {
+    // Each batch consumes 60ms of the 100ms window: two batches fit, the
+    // third boundary check sees 120 ≥ 100 and stops. #2750: after the window
+    // elapses the final countRemaining is SKIPPED (it would overrun the
+    // window), so remaining reports null.
+    let now = 0;
    const result = await runExtractAtomsDrain(
      {
        withLock: passThroughLock,
        countRemaining: async () => 5, // never drains
-        runBatch: async () => ({ extracted: 1, skipped: 0 }),
-        now,
+        runBatch: async () => {
+          now += 60;
+          return { extracted: 1, skipped: 0 };
+        },
+        now: () => now,
      },
      { windowMs: 100 },
    );
    expect(result.stopped).toBe('window');
-    expect(result.remaining).toBe(5);
+    expect(result.remaining).toBeNull();
    expect(result.batches).toBe(2);
  });

+  it('passes one drain-level deadline signal into count and batch', async () => {
+    const seen: AbortSignal[] = [];
+    const controller = new AbortController();
+    let now = 0;
+    const result = await runExtractAtomsDrain(
+      {
+        withLock: passThroughLock,
+        countRemaining: async (signal) => {
+          seen.push(signal);
+          return 5;
+        },
+        runBatch: async (signal) => {
+          seen.push(signal);
+          now = 100;
+          return { extracted: 1, skipped: 0 };
+        },
+        now: () => now,
+      },
+      { windowMs: 100, abortSignal: controller.signal },
+    );
+    expect(result.stopped).toBe('window');
+    expect(result.batches).toBe(1);
+    expect(seen.length).toBe(2);
+    expect(seen[0]).toBe(seen[1]);
+    // Combined (timeout + external) signal, not the raw external one.
+    expect(seen[0]).not.toBe(controller.signal);
+  });
+
+  it('aborts a hung backlog count at the window deadline and releases the lock', async () => {
+    let released = false;
+    const result = await runExtractAtomsDrain(
+      {
+        withLock: async (work) => {
+          try { return await work(); }
+          finally { released = true; }
+        },
+        // Hangs until the drain's real-time deadline signal fires (10ms).
+        countRemaining: (signal) => new Promise((_resolve, reject) => {
+          signal.addEventListener('abort', () => reject(signal.reason), { once: true });
+        }),
+        runBatch: async () => ({ extracted: 0, skipped: 0 }),
+        now: () => 0, // injected clock never advances — the SIGNAL must save us
+      },
+      { windowMs: 10 },
+    );
+    expect(result.stopped).toBe('window');
+    expect(result.remaining).toBeNull();
+    expect(released).toBe(true);
+  });
+
+  it('rethrows external cancellation after releasing the lock', async () => {
+    const controller = new AbortController();
+    let released = false;
+    const pending = runExtractAtomsDrain(
+      {
+        withLock: async (work) => {
+          try { return await work(); }
+          finally { released = true; }
+        },
+        countRemaining: (signal) => new Promise((_resolve, reject) => {
+          signal.addEventListener('abort', () => reject(signal.reason), { once: true });
+        }),
+        runBatch: async () => ({ extracted: 0, skipped: 0 }),
+        now: () => 0,
+      },
+      { windowMs: 1_000_000, abortSignal: controller.signal },
+    );
+    controller.abort(new DOMException('worker timeout', 'AbortError'));
+    await expect(pending).rejects.toThrow('worker timeout');
+    expect(released).toBe(true);
+  });
+
+  it('classifies a deadline-exhausted zero-progress batch as window, not no_progress', async () => {
+    let now = 0;
+    const result = await runExtractAtomsDrain(
+      {
+        withLock: passThroughLock,
+        countRemaining: async () => 5,
+        runBatch: async () => {
+          now = 100; // batch consumed the whole window and returned nothing
+          return { extracted: 0, skipped: 0 };
+        },
+        now: () => now,
+      },
+      { windowMs: 100 },
+    );
+    expect(result.stopped).toBe('window');
+    expect(result.batches).toBe(1);
+  });
+
+  it('bounds a hung lock acquisition with the drain deadline signal', async () => {
+    const started = Date.now();
+    await expect(runExtractAtomsDrain(
+      {
+        withLock: (_work, signal) => new Promise((_resolve, reject) => {
+          signal.addEventListener('abort', () => reject(signal.reason), { once: true });
+        }),
+        countRemaining: async () => 1,
+        runBatch: async () => ({ extracted: 0, skipped: 0 }),
+        now: Date.now,
+      },
+      { windowMs: 10 },
+    )).rejects.toThrow();
+    expect(Date.now() - started).toBeLessThan(1_000);
+  });
+
  it('stops on a zero-progress batch (no hot loop)', async () => {
    let batches = 0;
    const result = await runExtractAtomsDrain(
@@ -133,4 +242,17 @@ describe('shared wiring helper holds the cycle lock (5A)', () => {
    expect(src).toContain('cycleLockIdFor(opts.sourceId)');
    expect(src).toContain('withRefreshingLock(engine, lockId');
  });
+
+  // #2750: the deadline signal must reach the phase, the backlog count, AND
+  // the lock wrapper — and the transcript path (brainDir) must stay wired
+  // exactly as the routine callers expect (PR #2752 takeover reverted its
+  // unsanctioned transcript-suppression scope change).
+  it('threads the drain deadline signal through phase, count, and lock', () => {
+    const jobsSrc = readFileSync(join(import.meta.dir, '../src/commands/jobs.ts'), 'utf8');
+    expect(src).toContain('abortSignal: signal');
+    expect(src).toContain('countExtractAtomsBacklog(engine, extractionSourceId, signal)');
+    expect(src).toContain('brainDir: opts.brainDir');
+    expect(src).not.toContain('_transcripts');
+    expect(jobsSrc).toContain('abortSignal: job.signal');
+  });
 });
@@ -24,15 +24,18 @@ import type { BrainEngine } from '../../src/core/engine.ts';
 // Mock engine: healthCheck() calls engine.executeRaw; return empty rows so
 // the query path exercises without needing Postgres.
 //
-// #1849: start() now acquires the queue-scoped DB singleton lock via
-// tryAcquireDbLock, which uses the postgres `sql` tagged-template escape hatch.
-// The stub returns a single row from every call so acquire succeeds (length 1
-// → acquired) and refresh/release are no-ops. Each spawned runner is a fresh
-// process, so there's no cross-test lock state to clean up.
+// #1849: start() acquires the queue-scoped DB singleton lock via
+// tryAcquireDbLock. #2750 routed the acquire upsert through engine.executeRaw
+// (signal-boundable) and release through engine.executeRawDirect, so the
+// stub returns a single row from the lock upsert (length 1 → acquired) and
+// empty rows everywhere else. Each spawned runner is a fresh process, so
+// there's no cross-test lock state to clean up.
 const sqlStub = (..._args: unknown[]) => Promise.resolve([{ id: 'supervisor-lock' }]);
 const mockEngine: Partial<BrainEngine> = {
  kind: 'postgres' as const,
-  executeRaw: async () => [],
+  executeRaw: async (query: string) =>
+    query.includes('gbrain_cycle_locks') ? [{ id: 'supervisor-lock' }] : [],
+  executeRawDirect: async () => [],
  sql: sqlStub,
 } as unknown as BrainEngine;

@@ -19,7 +19,7 @@
 * overwrites this preload.
 */
 import { configureGateway, getEmbeddingDimensions } from '../../src/core/ai/gateway.ts';
-import { afterEach, beforeEach } from 'bun:test';
+import { beforeEach } from 'bun:test';

 const LEGACY_CONFIG = {
  embedding_model: 'openai:text-embedding-3-large',
@@ -52,7 +52,7 @@ applyLegacy();
 //   2. file-local beforeAll → may overwrite to ZE/1280
 // Since beforeAll runs once per file BEFORE the first beforeEach,
 // file-local beforeAll wins for that file's tests. ✓
-function applyLegacyIfEmpty() {
+beforeEach(() => {
  try {
    // Only re-apply if the gateway was reset (or never configured).
    // Tests that explicitly configured a different model in their
@@ -62,28 +62,4 @@ function applyLegacyIfEmpty() {
  } catch {
    applyLegacy();
  }
-}
-
-beforeEach(applyLegacyIfEmpty);
-
-// PR #3130 shard-order fix: beforeEach alone leaves ONE window open — a file
-// whose LAST afterEach calls resetGateway() poisons the NEXT file's
-// beforeAll, which runs BEFORE any beforeEach fires. A beforeAll there that
-// does engine.initSchema() then sizes the embedding column from the gateway
-// DEFAULTS (zembed-1/1280d) instead of the pinned legacy 1536, and every
-// 1536-d Float32Array fixture in that file dies with
-// "expected 1280 dimensions, not 1536". Which file pair collides is a
-// function of shard composition, so adding/removing ANY test file can
-// surface it (that is exactly how it bit shard 9).
-//
-// Preload hooks are registered before any file-local hooks, and bun runs
-// after-hooks inside-out (file-local afterEach first, then this one), so
-// this repairs the empty slot immediately after the poisoning reset —
-// before the next file's beforeAll can observe it.
-//
-// Known remaining window: a file whose afterAll() resets the gateway (no
-// hook runs between its afterAll and the next file's beforeAll). Files
-// that reset in afterAll and can precede a schema-creating file should
-// re-apply their own config, or the victim file should configureGateway()
-// explicitly in its beforeAll.
-afterEach(applyLegacyIfEmpty);
+});
@@ -1,69 +0,0 @@
-/**
- * #1207: `gbrain import` without `--workers` used to hardcode workerCount=1,
- * so a large Postgres import paid one serial embedding round-trip per file.
- * runImport now routes the default through the shared autoConcurrency policy
- * (PGLite → 1, >100 files on Postgres → DEFAULT_PARALLEL_WORKERS), while an
- * explicit `--workers N` still wins.
- *
- * The engine here is a minimal postgres-kind stub with no database_url in
- * config — runImport's parallel branch then falls back to serial processing
- * (its PR #490 guard) but the WORKER-COUNT DECISION (the thing #1207 fixes)
- * is still observable via the "Using N parallel workers" log line. Per-file
- * imports fail against the stub engine and are swallowed by runImport's
- * per-file catch; that's fine — this test pins the policy, not the import.
- */
-
-import { afterEach, beforeEach, describe, expect, test } from 'bun:test';
-import { mkdtempSync, writeFileSync, mkdirSync, rmSync, realpathSync } from 'fs';
-import { tmpdir } from 'os';
-import { join } from 'path';
-import { withEnv } from './helpers/with-env.ts';
-import { runImport } from '../src/commands/import.ts';
-
-const fakePostgresEngine = {
-  kind: 'postgres',
-  executeRaw: async () => [],
-  logIngest: async () => {},
-  setConfig: async () => {},
-  getConfig: async () => null,
-} as any;
-
-let workspace: string;
-let brainDir: string;
-let logs: string[];
-const realLog = console.log;
-
-beforeEach(() => {
-  workspace = mkdtempSync(join(tmpdir(), 'gbrain-import-workers-home-'));
-  mkdirSync(join(workspace, '.gbrain'), { recursive: true });
-  brainDir = realpathSync(mkdtempSync(join(tmpdir(), 'gbrain-import-workers-brain-')));
-  // 101 files: one past AUTO_CONCURRENCY_FILE_THRESHOLD (100).
-  for (let i = 0; i < 101; i++) {
-    writeFileSync(join(brainDir, `page-${i}.md`), `# Page ${i}\n\nbody ${i}\n`);
-  }
-  logs = [];
-  console.log = (msg?: unknown) => logs.push(String(msg));
-});
-
-afterEach(() => {
-  console.log = realLog;
-  rmSync(workspace, { recursive: true, force: true });
-  rmSync(brainDir, { recursive: true, force: true });
-});
-
-describe('import default worker count (#1207)', () => {
-  test('no --workers flag → autoConcurrency picks 4 for >100 files on Postgres', async () => {
-    await withEnv({ GBRAIN_HOME: join(workspace, '.gbrain'), GBRAIN_SOURCE: undefined }, async () => {
-      await runImport(fakePostgresEngine, [brainDir, '--no-embed'], { sourceId: 'default' });
-    });
-    expect(logs.some(l => l.includes('Using 4 parallel workers'))).toBe(true);
-  });
-
-  test('explicit --workers 2 still wins over the auto policy', async () => {
-    await withEnv({ GBRAIN_HOME: join(workspace, '.gbrain'), GBRAIN_SOURCE: undefined }, async () => {
-      await runImport(fakePostgresEngine, [brainDir, '--no-embed', '--workers', '2'], { sourceId: 'default' });
-    });
-    expect(logs.some(l => l.includes('Using 2 parallel workers'))).toBe(true);
-    expect(logs.some(l => l.includes('Using 4 parallel workers'))).toBe(false);
-  });
-});
@@ -98,14 +98,39 @@ describe('PostgresEngine.executeRawDirect — routing decision (PR #1816)', () =
  });

  test('already-aborted signal short-circuits with AbortError before routing the query', async () => {
-    const readConn = fakeSql('read');
+    let unsafeCalls = 0;
+    let ddlCalls = 0;
+    const readConn: FakeSql = { unsafe: async () => { unsafeCalls++; return []; } };
    const directConn = fakeSql('direct');
    const engine = makeEngine({ dualPoolActive: true, readConn, directConn });
+    const e = engine as unknown as { connectionManager: { ddl: () => Promise<FakeSql> } };
+    e.connectionManager.ddl = async () => { ddlCalls++; return directConn; };

    const ac = new AbortController();
    ac.abort();
    await expect(
      engine.executeRawDirect('UPDATE minion_jobs SET x=1', [], { signal: ac.signal }),
    ).rejects.toThrow(/abort/i);
+    // #2750: short-circuits BEFORE pool routing — no ddl(), no unsafe().
+    expect(ddlCalls).toBe(0);
+    expect(unsafeCalls).toBe(0);
+  });
+
+  test('#2750: signal bounds a stalled direct-pool acquisition before unsafe starts', async () => {
+    let unsafeCalls = 0;
+    const readConn: FakeSql = { unsafe: async () => { unsafeCalls++; return []; } };
+    const directConn = fakeSql('direct');
+    const engine = makeEngine({ dualPoolActive: true, readConn, directConn });
+    const e = engine as unknown as { connectionManager: { ddl: () => Promise<FakeSql> } };
+    e.connectionManager.ddl = () => new Promise<FakeSql>(() => {}); // pooler exhausted: never resolves
+
+    const started = Date.now();
+    await expect(engine.executeRawDirect(
+      'DELETE FROM gbrain_cycle_locks',
+      [],
+      { signal: AbortSignal.timeout(10) },
+    )).rejects.toThrow(/abort/i);
+    expect(Date.now() - started).toBeLessThan(1_000);
+    expect(unsafeCalls).toBe(0);
  });
 });