fix(rerank): classify missing auth before fallback (#2059)

Missing ZEROENTROPY_API_KEY threw AIConfigError from auth resolution, which
rerank.ts recorded as reason 'unknown' — and doctor's reranker_health had no
unknown bucket, so it reported ok while every rerank silently failed open.

- gateway.rerank wraps AIConfigError from applyResolveAuth as
  RerankError(reason: 'auth') before any HTTP call.
- checkRerankerHealth warns on >=3 'unknown' failures in the 7-day window
  (covers historical pre-fix audit rows), with a ZEROENTROPY_API_KEY setup
  hint when the error summary points at a missing key.
- Tests: RerankError(auth) classification, applyReranker fail-open + audit
  reason, doctor warn on repeated unknowns.

Takeover of #2070.

Co-authored-by: maxpetrusenkoagent <maxpetrusenkoagent@users.noreply.github.com>
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

src/commands/doctor.ts

@@ -1551,6 +1551,24 @@ export async function checkRerankerHealth(engine: BrainEngine): Promise<Check> {
       };
     }
 
+    // Historical #2059 rows were logged as `unknown` before missing reranker
+    // auth was classified at the gateway. Surface repeated unknowns instead of
+    // reporting "ok" while every rerank fails open.
+    const unknownFails = failures.filter((f) => f.reason === 'unknown');
+    if (unknownFails.length >= 3) {
+      const setupHint = unknownFails.some((f) => {
+        const summary = String(f.error_summary ?? '');
+        return summary.includes('ZEROENTROPY_API_KEY') || summary.toLowerCase().includes('api key');
+      })
+        ? ' Fix: verify ZEROENTROPY_API_KEY and run `gbrain models doctor`.'
+        : '';
+      return {
+        name: 'reranker_health',
+        status: 'warn',
+        message: `${unknownFails.length} unknown reranker failure(s) in last 7 days.${setupHint}`,
+      };
+    }
+
     return {
       name: 'reranker_health',
       status: 'ok',

src/commands/extract.ts

@@ -1651,7 +1651,7 @@ async function extractTimelineFromDB(
  * make re-extraction idempotent). EVERY processed page is stamped, including
  * zero-link pages — they WERE processed.
  */
-export async function extractStaleFromDB(
+async function extractStaleFromDB(
   engine: BrainEngine,
   opts: {
     dryRun: boolean;

src/commands/jobs.ts

@@ -1479,31 +1479,7 @@ export async function registerBuiltinHandlers(
       embedSkipReason = 'auto_embed_disabled';
     }
 
-    // #2849: large-sync extract deferral follow-up. performSync skips inline
-    // link/timeline extraction when totalChanges > 100, leaving
-    // links_extracted_at unstamped. A standalone sync job (webhook push,
-    // sync trigger) has no autopilot extract phase behind it, so the pages
-    // would stay extraction-stale until a manual `gbrain extract --stale`.
-    // Queue a source-scoped stale sweep instead. Best-effort + idempotent:
-    // a duplicate sweep finds 0 stale pages and no-ops.
-    let extractJobId: number | null = null;
-    if (result.extractDeferred) {
-      try {
-        const { MinionQueue } = await import('../core/minions/queue.ts');
-        const queue = new MinionQueue(engine);
-        const followUp = await queue.add(
-          'extract',
-          { stale: true, ...(sourceId ? { sourceId } : {}) },
-          {
-            idempotency_key: `sync-extract-stale:${sourceId ?? 'default'}:${Math.floor(Date.now() / 30_000)}`,
-            maxWaiting: 1,
-          },
-        );
-        extractJobId = followUp.id;
-      } catch { /* best-effort: extract --stale sweeps it later */ }
-    }
-
-    return { ...result, embed_job_id: embedJobId, embed_skip_reason: embedSkipReason, extract_stale_job_id: extractJobId };
+    return { ...result, embed_job_id: embedJobId, embed_skip_reason: embedSkipReason };
   });
 
   registerBuiltinJob(worker, engine, 'embed', async (job) => {
@@ -1676,20 +1652,6 @@ export async function registerBuiltinHandlers(
   });
 
   worker.register('extract', async (job) => {
-    // #2849: stale-sweep mode — the sync handler's large-sync deferral
-    // follow-up. DB-source (reads page content from the DB, so it runs on
-    // checkout-less brains), source-scopable, idempotent. Same core as
-    // `gbrain extract --stale`.
-    if (job.data.stale === true) {
-      const { extractStaleFromDB } = await import('./extract.ts');
-      return await extractStaleFromDB(engine, {
-        dryRun: !!job.data.dryRun,
-        jsonMode: false,
-        includeFrontmatter: false,
-        sourceIdFilter: typeof job.data.sourceId === 'string' ? job.data.sourceId : undefined,
-        catchUp: false,
-      });
-    }
     const { runExtractCore } = await import('./extract.ts');
     const mode = (typeof job.data.mode === 'string' && ['links', 'timeline', 'all'].includes(job.data.mode))
       ? (job.data.mode as 'links' | 'timeline' | 'all')

src/commands/serve-http.ts

@@ -2146,13 +2146,8 @@ export async function runServeHttp(engine: BrainEngine, options: ServeHttpOption
   //     Other event types (ping, pull_request, etc.) return 202 'ignored'
   //     so GitHub doesn't retry.
   // D15.5: HMAC compare uses the shared safeHexEqual helper.
-  // D18: submits 'sync' job with extraction + auto_embed_backfill enabled and
-  //     priority -10 (above autopilot's 0). noExtract:false opts normal
-  //     incremental pushes into sync's inline link/timeline extraction (#2849
-  //     — the standalone sync handler defaults noExtract to TRUE, which left
-  //     webhook-imported pages permanently stale). Large (>100 file) pushes
-  //     defer inline extract; the sync handler queues an extract --stale
-  //     follow-up job for that branch.
+  // D18: submits 'sync' job with auto_embed_backfill=true and priority -10
+  //     (above autopilot's 0).
   // ---------------------------------------------------------------------------
   const githubWebhookLimiter = rateLimit({
     windowMs: 60_000,
@@ -2272,7 +2267,6 @@ export async function runServeHttp(engine: BrainEngine, options: ServeHttpOption
           'sync',
           {
             sourceId: source.id,
-            noExtract: false,
             auto_embed_backfill: true,
             embed_reason: 'webhook',
           },

src/commands/sync.ts

@@ -222,14 +222,6 @@ export interface SyncResult {
    * everything," the exact misdiagnosis in the #1794 recurrence report.
    */
   bankedFiles?: number;
-  /**
-   * #2849: true when extraction was REQUESTED (noExtract false) but this sync
-   * skipped inline link/timeline extraction because totalChanges > 100 (the
-   * #1794 large-sync deferral). links_extracted_at stays unstamped for the
-   * imported pages. The standalone `sync` job handler queues a source-scoped
-   * `extract --stale` follow-up when set; CLI runs print the manual hint.
-   */
-  extractDeferred?: boolean;
 }
 
 /**
@@ -1387,10 +1379,6 @@ See also:
     {
       sourceId: sourceIdArg,
       repoPath: source.local_path,
-      // #2849: opt in to inline extraction — the standalone sync handler
-      // defaults noExtract to TRUE (dedupe for doctor's [sync, extract]
-      // remediation plan), which would leave triggered syncs extraction-stale.
-      noExtract: false,
       auto_embed_backfill: true,
       embed_reason: 'sync_trigger',
     },
@@ -3299,16 +3287,11 @@ async function performSyncInner(engine: BrainEngine, opts: SyncOpts): Promise<Sy
   // the stale sweep scans the whole source, so banked-across-runs pages are
   // covered regardless.
   const extractOpts = opts.sourceId ? { sourceId: opts.sourceId } : undefined;
-  let extractDeferred = false;
   if (!opts.noExtract && totalChanges > 100 && pagesAffected.length > 0) {
-    // #2849: surface the deferral to callers. A standalone sync job (webhook
-    // push, sync trigger) has no autopilot extract phase behind it, so the
-    // job handler queues an `extract --stale` follow-up off this flag.
-    extractDeferred = true;
     slog(
       `  Large sync: deferring link/timeline extraction. ` +
       `Run 'gbrain extract --stale${opts.sourceId ? ` --source-id ${opts.sourceId}` : ''}' ` +
-      `(sync jobs queue this follow-up automatically).`,
+      `(or let the autopilot cycle's extract phase sweep it).`,
     );
   }
   if (!opts.noExtract && totalChanges <= 100 && pagesAffected.length > 0) {
@@ -3417,7 +3400,6 @@ async function performSyncInner(engine: BrainEngine, opts: SyncOpts): Promise<Sy
     chunksCreated,
     embedded,
     pagesAffected,
-    extractDeferred,
   };
 }
 

src/core/ai/gateway.ts

@@ -3656,7 +3656,15 @@ export async function rerank(input: RerankInput): Promise<RerankResult[]> {
   // whose request/response shape differs from ZE/llama.cpp (e.g. Voyage with
   // `top_k` / `data[]`) needs separate adapter hooks in a follow-up plan.
   const url = `${compat.baseURL.replace(/\/$/, '')}${tp.path ?? '/models/rerank'}`;
-  const auth = applyResolveAuth(recipe, cfg, 'reranker');
+  let auth: { apiKey?: string; headers?: Record<string, string> };
+  try {
+    auth = applyResolveAuth(recipe, cfg, 'reranker');
+  } catch (err) {
+    if (err instanceof AIConfigError) {
+      throw new RerankError(err.message, 'auth');
+    }
+    throw err;
+  }
   // applyResolveAuth returns { apiKey } for Bearer-style auth (SDK's native
   // path) or { headers } for custom-header providers (Azure). v0.37.6.0:
   // recipes can ALSO declare default_headers (attribution etc.) which flow

test/ai/rerank.test.ts

@@ -154,6 +154,28 @@ describe('gateway.rerank() — happy path', () => {
 describe('gateway.rerank() — error classification', () => {
   beforeEach(() => configureZE());
 
+  test('missing required reranker API key → RerankError(auth) before HTTP call', async () => {
+    configureGateway({
+      reranker_model: 'zeroentropyai:zerank-2',
+      env: {},
+    });
+    let called = false;
+    __setRerankTransportForTests(async () => {
+      called = true;
+      return mockResp({ results: [{ index: 0, relevance_score: 0.5 }] });
+    });
+
+    try {
+      await rerank({ query: 'q', documents: ['d'] });
+      throw new Error('should have thrown');
+    } catch (err) {
+      expect(err).toBeInstanceOf(RerankError);
+      expect((err as RerankError).reason).toBe('auth');
+      expect((err as Error).message).toContain('ZEROENTROPY_API_KEY');
+      expect(called).toBe(false);
+    }
+  });
+
   test('401 → auth', async () => {
     __setRerankTransportForTests(async () => new Response('Unauthorized', { status: 401 }));
     try {

test/doctor.test.ts

@@ -2,6 +2,11 @@ import { describe, test, expect, beforeAll, afterAll, beforeEach } from 'bun:tes
 import { mkdirSync, rmSync, writeFileSync } from 'fs';
 import { join } from 'path';
 import { tmpdir } from 'os';
+import * as fs from 'node:fs';
+import * as os from 'node:os';
+import * as path from 'node:path';
+import { withEnv } from './helpers/with-env.ts';
+import { logRerankFailure } from '../src/core/rerank-audit.ts';
 
 describe('doctor command', () => {
   test('doctor module exports runDoctor', async () => {
@@ -47,6 +52,34 @@ describe('doctor command', () => {
     expect(check.issues![0].action).toContain('trigger');
   });
 
+  test('reranker_health warns on repeated unknown rerank failures', async () => {
+    const { checkRerankerHealth } = await import('../src/commands/doctor.ts');
+    const tmpDir = fs.mkdtempSync(path.join(os.tmpdir(), 'gbrain-rerank-doctor-'));
+    try {
+      await withEnv({ GBRAIN_AUDIT_DIR: tmpDir }, async () => {
+        for (let i = 0; i < 3; i++) {
+          logRerankFailure({
+            model: 'zeroentropyai:zerank-2',
+            reason: 'unknown',
+            query_hash: `unknown${i}`,
+            doc_count: 30,
+            error_summary: 'ZeroEntropy reranker requires ZEROENTROPY_API_KEY.',
+          });
+        }
+        const check = await checkRerankerHealth({
+          async getConfig(key: string): Promise<string | null> {
+            return key === 'search.reranker.enabled' ? 'true' : null;
+          },
+        } as any);
+        expect(check.status).toBe('warn');
+        expect(check.message).toContain('unknown');
+        expect(check.message).toContain('ZEROENTROPY_API_KEY');
+      });
+    } finally {
+      fs.rmSync(tmpDir, { recursive: true, force: true });
+    }
+  });
+
   test('runDoctor accepts null engine for filesystem-only mode', async () => {
     const { runDoctor } = await import('../src/commands/doctor.ts');
     // runDoctor should accept null engine — it runs filesystem checks only.

test/search/rerank.test.ts

@@ -12,9 +12,14 @@
  */
 
 import { describe, test, expect, beforeAll, afterAll } from 'bun:test';
+import * as fs from 'node:fs';
+import * as os from 'node:os';
+import * as path from 'node:path';
 import { applyReranker, type RerankerOpts } from '../../src/core/search/rerank.ts';
 import { RerankError, type RerankResult } from '../../src/core/ai/gateway.ts';
+import { readRecentRerankFailures } from '../../src/core/rerank-audit.ts';
 import type { SearchResult } from '../../src/core/types.ts';
+import { withEnv } from '../helpers/with-env.ts';
 
 function makeResult(slug: string, score: number, chunk: string): SearchResult {
   return {
@@ -160,6 +165,36 @@ describe('applyReranker — fail-open on every RerankError reason', () => {
     expect(out).toEqual(results);
   });
 
+  test('missing gateway reranker API key fail-opens and audits auth', async () => {
+    const { configureGateway } = await import('../../src/core/ai/gateway.ts');
+    const tmpDir = fs.mkdtempSync(path.join(os.tmpdir(), 'gbrain-rerank-search-'));
+    try {
+      await withEnv({ GBRAIN_AUDIT_DIR: tmpDir }, async () => {
+        configureGateway({
+          reranker_model: 'zeroentropyai:zerank-2',
+          env: {},
+        });
+
+        const results = [makeResult('a', 1.0, 'doc a')];
+        const out = await applyReranker('q', results, {
+          enabled: true,
+          topNIn: 1,
+          topNOut: null,
+          model: 'zeroentropyai:zerank-2',
+        });
+
+        expect(out).toEqual(results);
+        const failures = readRecentRerankFailures(1);
+        expect(failures).toHaveLength(1);
+        expect(failures[0]!.reason).toBe('auth');
+        expect(failures[0]!.error_summary).toContain('ZEROENTROPY_API_KEY');
+      });
+    } finally {
+      fs.rmSync(tmpDir, { recursive: true, force: true });
+      configureGateway({ env: { ZEROENTROPY_API_KEY: 'test-key' } });
+    }
+  });
+
   test('fail-open on non-RerankError throw too', async () => {
     const results = [makeResult('a', 1.0, 'a')];
     const opts: RerankerOpts = {

test/sources-webhook.test.ts

@@ -16,7 +16,6 @@
  */
 import { describe, test, expect } from 'bun:test';
 import { createHmac } from 'node:crypto';
-import { readFileSync } from 'node:fs';
 import { safeHexEqual } from '../src/core/timing-safe.ts';
 
 const GITHUB_SECRET = 'super-secret-webhook-key';
@@ -124,25 +123,3 @@ describe('Branch ref construction (D5)', () => {
     expect(pushedRef === `refs/heads/${trackedBranch}`).toBe(false);
   });
 });
-
-describe('Webhook sync job extraction contract (#2849)', () => {
-  test('opts into extraction before the pushed commit is consumed', () => {
-    const serveSource = readFileSync(
-      new URL('../src/commands/serve-http.ts', import.meta.url),
-      'utf8',
-    );
-    const routeStart = serveSource.indexOf("'/webhooks/github'");
-    const queueStart = serveSource.indexOf('const job = await queue.add(', routeStart);
-    const responseStart = serveSource.indexOf('res.status(202)', queueStart);
-    expect(routeStart).toBeGreaterThanOrEqual(0);
-    expect(queueStart).toBeGreaterThan(routeStart);
-    expect(responseStart).toBeGreaterThan(queueStart);
-
-    const routeSource = serveSource.slice(queueStart, responseStart);
-    const payload = routeSource.match(
-      /queue\.add\(\s*'sync',\s*\{([\s\S]*?)\}\s*,\s*\{/,
-    );
-    expect(payload).not.toBeNull();
-    expect(payload?.[1]).toMatch(/\bnoExtract:\s*false\b/);
-  });
-});

test/sync-large-deferral-extract.serial.test.ts

@@ -1,130 +0,0 @@
-/**
- * #2849 — large-sync extract deferral queues an `extract --stale` follow-up.
- *
- * performSync's incremental path skips inline link/timeline extraction when
- * totalChanges > 100 (the #1794 large-sync deferral), leaving
- * links_extracted_at unstamped. Pre-fix, a standalone sync job (webhook push,
- * `gbrain sync trigger`) had NOTHING behind it to sweep those pages — the
- * autopilot cycle's extract phase only walks that cycle's changedSlugs — so a
- * large webhook push left extraction permanently stale until a manual
- * `gbrain extract --stale`.
- *
- * Pins:
- *   (a) performSync surfaces `extractDeferred: true` on the >100 branch and
- *       leaves the pages unstamped/unlinked.
- *   (b) the `sync` job handler queues an `extract` job with
- *       { stale: true, sourceId? } when extractDeferred is set.
- *   (c) the `extract` handler's stale mode actually sweeps: links created +
- *       watermark stamped (end-to-end recovery, no manual step).
- *
- * Marked .serial.test.ts — spawns git subprocesses + shares one PGLite engine.
- */
-
-import { describe, test, expect, beforeAll, afterAll } from 'bun:test';
-import { mkdtempSync, writeFileSync, rmSync, mkdirSync } from 'fs';
-import { execSync } from 'child_process';
-import { tmpdir } from 'os';
-import { join } from 'path';
-import { PGLiteEngine } from '../src/core/pglite-engine.ts';
-import { MinionWorker } from '../src/core/minions/worker.ts';
-import { MinionQueue } from '../src/core/minions/queue.ts';
-import { registerBuiltinHandlers } from '../src/commands/jobs.ts';
-
-let engine: PGLiteEngine;
-let worker: MinionWorker;
-let repoPath: string;
-
-function git(cmd: string): void { execSync(cmd, { cwd: repoPath, stdio: 'pipe' }); }
-
-describe('#2849 — large sync defers extract and queues a stale sweep', () => {
-  beforeAll(async () => {
-    engine = new PGLiteEngine();
-    await engine.connect({});
-    await engine.initSchema();
-    worker = new MinionWorker(engine, { queue: 'test' });
-    await registerBuiltinHandlers(worker, engine, { quiet: true });
-
-    repoPath = mkdtempSync(join(tmpdir(), 'gbrain-large-defer-'));
-    git('git init');
-    git('git config user.email "t@t.com"');
-    git('git config user.name "T"');
-    mkdirSync(join(repoPath, 'people'), { recursive: true });
-    mkdirSync(join(repoPath, 'notes'), { recursive: true });
-    writeFileSync(join(repoPath, 'people/alice.md'), [
-      '---', 'type: person', 'title: Alice', '---', '', 'Alice is a founder.',
-    ].join('\n'));
-    git('git add -A && git commit -m "initial"');
-
-    // Seed: full first sync imports the anchor page + sets last_commit.
-    const { performSync } = await import('../src/commands/sync.ts');
-    await performSync(engine, { repoPath, full: true, noPull: true, noEmbed: true });
-
-    // Second commit: 101 new pages → incremental totalChanges > 100.
-    for (let i = 0; i < 101; i++) {
-      writeFileSync(join(repoPath, `notes/n${i}.md`), [
-        '---', 'type: note', `title: Note ${i}`, '---', '',
-        `[Alice](people/alice) appears in note ${i}.`,
-      ].join('\n'));
-    }
-    git('git add -A && git commit -m "add 101 pages"');
-  }, 120_000);
-
-  afterAll(async () => {
-    if (repoPath) rmSync(repoPath, { recursive: true, force: true });
-    if (engine) await engine.disconnect();
-  }, 60_000);
-
-  test('sync handler defers inline extract and queues extract{stale} follow-up; stale sweep recovers', async () => {
-    const syncHandler = (worker as unknown as { handlers: Map<string, (job: unknown) => Promise<unknown>> })
-      .handlers.get('sync');
-    expect(syncHandler).toBeDefined();
-
-    // Same payload shape the webhook submits (minus embed backfill noise).
-    const result = await syncHandler!({
-      data: { repoPath, noExtract: false, noPull: true, auto_embed_backfill: false },
-      signal: { aborted: false },
-      updateProgress: async () => {},
-    }) as { status: string; extractDeferred?: boolean; extract_stale_job_id?: number | null };
-
-    expect(result.status).toBe('synced');
-    // (a) inline extract was deferred, pages left stale.
-    expect(result.extractDeferred).toBe(true);
-    const staleBefore = await engine.countStalePagesForExtraction();
-    expect(staleBefore).toBeGreaterThan(100);
-    expect(await engine.getLinks('notes/n0')).toHaveLength(0);
-
-    // (b) a follow-up extract job with stale:true was queued.
-    expect(result.extract_stale_job_id).toBeGreaterThan(0);
-    const queue = new MinionQueue(engine);
-    const extractJobs = await queue.getJobs({ name: 'extract', limit: 5 });
-    expect(extractJobs.length).toBe(1);
-    expect((extractJobs[0].data as { stale: boolean }).stale).toBe(true);
-
-    // (c) running the extract handler's stale mode recovers: links + stamps.
-    const extractHandler = (worker as unknown as { handlers: Map<string, (job: unknown) => Promise<unknown>> })
-      .handlers.get('extract');
-    await extractHandler!({
-      data: extractJobs[0].data,
-      signal: { aborted: false },
-      updateProgress: async () => {},
-    });
-    const links = await engine.getLinks('notes/n0');
-    expect(links.some(l => l.to_slug === 'people/alice')).toBe(true);
-    const rows = await engine.executeRaw<{ links_extracted_at: string | null }>(
-      `SELECT links_extracted_at FROM pages WHERE slug = 'notes/n0'`,
-    );
-    expect(rows[0]?.links_extracted_at).not.toBeNull();
-  }, 180_000);
-
-  test('sub-threshold sync does NOT set extractDeferred (no spurious follow-up)', async () => {
-    // One more small commit → inline extract path, no deferral.
-    writeFileSync(join(repoPath, 'notes/small.md'), [
-      '---', 'type: note', 'title: Small', '---', '', 'No big deal.',
-    ].join('\n'));
-    git('git add -A && git commit -m "one small page"');
-    const { performSync } = await import('../src/commands/sync.ts');
-    const result = await performSync(engine, { repoPath, noPull: true, noEmbed: true });
-    expect(result.status).toBe('synced');
-    expect(result.extractDeferred).toBeFalsy();
-  }, 60_000);
-});

test/sync-trigger-cli.test.ts

@@ -100,10 +100,6 @@ describe('runSyncTrigger', () => {
     const job = jobs[0];
     expect(job.priority).toBe(-10);
     expect((job.data as { sourceId: string }).sourceId).toBe('default');
-    // #2849: opt in to inline extraction — the standalone sync handler
-    // defaults noExtract to TRUE, which would leave triggered syncs
-    // extraction-stale.
-    expect((job.data as { noExtract: boolean }).noExtract).toBe(false);
     expect((job.data as { auto_embed_backfill: boolean }).auto_embed_backfill).toBe(true);
   });