fix(embed): extend #1717 model labeling to the embed-backfill stale path

src/core/embed-stale.ts (used by the embed-backfill minion handler) built
its merged ChunkInput without a model field, so every chunk on a touched
page — re-embedded AND preserved — was relabeled to the engine default on
each backfill pass. Mirror the embed.ts semantics: stamp the resolved
gateway label on re-embedded chunks, carry the existing label on untouched
ones. Pinned by a PGLite test that fails without the fix.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

src/commands/embed.ts

@@ -1,5 +1,5 @@
 import type { BrainEngine } from '../core/engine.ts';
-import { embedBatch, currentEmbeddingSignature } from '../core/embedding.ts';
+import { embedBatch, currentEmbeddingSignature, resolveEmbeddingModelLabel } from '../core/embedding.ts';
 import type { ChunkInput } from '../core/types.ts';
 import { chunkText } from '../core/chunkers/recursive.ts';
 import { createProgress, type ProgressReporter } from '../core/progress.ts';
@@ -581,11 +581,16 @@ async function embedPage(
   for (let j = 0; j < toEmbed.length; j++) {
     embeddingMap.set(toEmbed[j].chunk_index, embeddings[j]);
   }
+  // #1717: label each (re)embedded chunk with the model that actually
+  // produced its vector. Preserved chunks (not re-embedded this pass) keep
+  // their existing model so a mixed-model page isn't relabeled wholesale.
+  const embedModelLabel = resolveEmbeddingModelLabel();
   const updated: ChunkInput[] = chunks.map(c => ({
     chunk_index: c.chunk_index,
     chunk_text: c.chunk_text,
     chunk_source: c.chunk_source,
     embedding: embeddingMap.get(c.chunk_index),
+    model: embeddingMap.has(c.chunk_index) && embedModelLabel ? embedModelLabel : c.model,
     token_count: c.token_count || Math.ceil(c.chunk_text.length / 4),
   }));
 
@@ -717,12 +722,16 @@ async function embedAll(
       for (let j = 0; j < toEmbed.length; j++) {
         embeddingMap.set(toEmbed[j].chunk_index, embeddings[j]);
       }
+      // #1717: stamp the resolved embedding model on (re)embedded chunks;
+      // preserve the existing model on chunks left untouched.
+      const embedModelLabel = resolveEmbeddingModelLabel();
       // Preserve ALL chunks, only update embeddings for stale ones
       const updated: ChunkInput[] = chunks.map(c => ({
         chunk_index: c.chunk_index,
         chunk_text: c.chunk_text,
         chunk_source: c.chunk_source,
         embedding: embeddingMap.get(c.chunk_index) ?? undefined,
+        model: embeddingMap.has(c.chunk_index) && embedModelLabel ? embedModelLabel : c.model,
         token_count: c.token_count || Math.ceil(c.chunk_text.length / 4),
       }));
       await observed(pacer, () => engine.upsertChunks(page.slug, updated, pageOpts));
@@ -1012,11 +1021,15 @@ async function embedAllStale(
           for (let j = 0; j < stale.length; j++) {
             staleIdxToEmbedding.set(stale[j].chunk_index, embeddings[j]);
           }
+          // #1717: label the re-embedded (stale) chunks with the resolved
+          // model; preserve the existing model on the non-stale chunks.
+          const embedModelLabel = resolveEmbeddingModelLabel();
           const merged: ChunkInput[] = existing.map(c => ({
             chunk_index: c.chunk_index,
             chunk_text: c.chunk_text,
             chunk_source: c.chunk_source,
             embedding: staleIdxToEmbedding.get(c.chunk_index) ?? undefined,
+            model: staleIdxToEmbedding.has(c.chunk_index) && embedModelLabel ? embedModelLabel : c.model,
             token_count: c.token_count || Math.ceil(c.chunk_text.length / 4),
           }));
           await observed(pacer, () => engine.upsertChunks(slug, merged, { sourceId: keySourceId }));

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/embed-stale.ts

@@ -20,6 +20,7 @@
 import type { BrainEngine } from './engine.ts';
 import type { ChunkInput } from './types.ts';
 import { embedBatchWithBackoff } from '../commands/embed.ts';
+import { resolveEmbeddingModelLabel } from './embedding.ts';
 import { type DbPacer, createNoopPacer, observed } from './db-pacer.ts';
 import { AbortError } from './abort-check.ts';
 
@@ -200,11 +201,17 @@ export async function embedStaleForSource(
         for (let j = 0; j < stale.length; j++) {
           staleIdxToEmbedding.set(stale[j].chunk_index, embeddings[j]);
         }
+        // #1717: label re-embedded chunks with the model that produced the
+        // vector; preserved chunks keep their existing model. Without this,
+        // upsertChunks falls back to DEFAULT_EMBEDDING_MODEL for every chunk
+        // (the same mislabel the embed.ts paths fixed).
+        const embedModelLabel = resolveEmbeddingModelLabel();
         const merged: ChunkInput[] = existing.map((c) => ({
           chunk_index: c.chunk_index,
           chunk_text: c.chunk_text,
           chunk_source: c.chunk_source,
           embedding: staleIdxToEmbedding.get(c.chunk_index) ?? undefined,
+          model: staleIdxToEmbedding.has(c.chunk_index) && embedModelLabel ? embedModelLabel : c.model,
           token_count: c.token_count || Math.ceil(c.chunk_text.length / 4),
           // Carry through per-chunk metadata. upsertChunks writes these as
           // EXCLUDED.<col> (not COALESCE), so omitting them here resets image

src/core/embedding.ts

@@ -113,6 +113,21 @@ export async function embedBatch(
   return results;
 }
 
+/**
+ * Resolve the embedding model label (`provider:model`) to stamp onto
+ * `content_chunks.model`, so each chunk records the model that actually
+ * produced its vector instead of the engine's hardcoded default (#1717).
+ * Returns undefined if the gateway is unconfigured; callers then fall back
+ * to the chunk's existing model rather than mislabeling it.
+ */
+export function resolveEmbeddingModelLabel(): string | undefined {
+  try {
+    return gatewayGetModel();
+  } catch {
+    return undefined;
+  }
+}
+
 /** Currently-configured embedding model (short form without provider prefix). */
 export function getEmbeddingModelName(): string {
   return gatewayGetModel().split(':').slice(1).join(':') || 'text-embedding-3-large';

src/core/import-file.ts

@@ -8,7 +8,7 @@ import { chunkText } from './chunkers/recursive.ts';
 import { chunkCodeText, chunkCodeTextFull, detectCodeLanguage, CHUNKER_VERSION } from './chunkers/code.ts';
 import { findChunkForOffset } from './chunkers/edge-extractor.ts';
 import { extractCodeRefs, imageOfCandidates } from './link-extraction.ts';
-import { embedBatch, embedMultimodal, currentEmbeddingSignature } from './embedding.ts';
+import { embedBatch, embedMultimodal, currentEmbeddingSignature, resolveEmbeddingModelLabel } from './embedding.ts';
 import { slugifyPath, slugifyCodePath, isCodeFilePath } from './sync.ts';
 import type { ChunkInput, PageInput, PageType } from './types.ts';
 import { computeEffectiveDate } from './effective-date.ts';
@@ -716,8 +716,12 @@ export async function importFromContent(
       ? chunks.map((c) => wrapChunkForEmbedding(c.chunk_text, prefix, c.chunk_source))
       : chunks.map((c) => c.chunk_text);
     const embeddings = await embedBatch(wrappedTexts);
+    // #1717: label each chunk with the model that actually produced its
+    // vector, not the engine's hardcoded default.
+    const embedModelLabel = resolveEmbeddingModelLabel();
     for (let i = 0; i < chunks.length; i++) {
       chunks[i].embedding = embeddings[i];
+      if (embedModelLabel) chunks[i].model = embedModelLabel;
       // token_count tracks the wrapped string length so cost reporting
       // reflects what we actually sent to the embedder.
       chunks[i].token_count = Math.ceil(wrappedTexts[i].length / 4);
@@ -1141,7 +1145,10 @@ export async function importCodeFile(
     const matched = existingByKey.get(key);
     if (matched && matched.embedding) {
       // Reuse the existing embedding verbatim. No API call, no cost.
+      // #1717: carry the existing model label along with the reused vector
+      // so the upsert doesn't relabel it with the engine default.
       chunks[i]!.embedding = matched.embedding as Float32Array;
+      chunks[i]!.model = matched.model ?? undefined;
       chunks[i]!.token_count = matched.token_count ?? undefined;
     } else {
       needsEmbedIndexes.push(i);
@@ -1153,9 +1160,12 @@ export async function importCodeFile(
     try {
       const textsToEmbed = needsEmbedIndexes.map((i) => chunks[i]!.chunk_text);
       const embeddings = await embedBatch(textsToEmbed);
+      // #1717: stamp the model that produced these vectors.
+      const embedModelLabel = resolveEmbeddingModelLabel();
       for (let j = 0; j < needsEmbedIndexes.length; j++) {
         const i = needsEmbedIndexes[j]!;
         chunks[i]!.embedding = embeddings[j]!;
+        if (embedModelLabel) chunks[i]!.model = embedModelLabel;
         chunks[i]!.token_count = Math.ceil(chunks[i]!.chunk_text.length / 4);
       }
     } catch (e: unknown) {

test/embed-stale.serial.test.ts

@@ -15,6 +15,7 @@ import { describe, test, expect, beforeAll, afterAll, beforeEach } from 'bun:tes
 import { PGLiteEngine } from '../src/core/pglite-engine.ts';
 import { resetPgliteState } from './helpers/reset-pglite.ts';
 import { embedStaleForSource } from '../src/core/embed-stale.ts';
+import { configureGateway, resetGateway } from '../src/core/ai/gateway.ts';
 import type { ChunkInput } from '../src/core/types.ts';
 
 let engine: PGLiteEngine;
@@ -276,4 +277,49 @@ describe('embedStaleForSource', () => {
     // The stale text row actually got its embedding.
     expect(txtRow.embedded_at).not.toBeNull();
   });
+
+  // #1717: the backfill path must label re-embedded chunks with the model
+  // that produced the vector, and preserve the existing label on chunks it
+  // did not touch (before the fix, both were reset to the engine default).
+  test('labels re-embedded chunks with the gateway model, preserves untouched labels (#1717)', async () => {
+    configureGateway({
+      embedding_model: 'openai:text-embedding-3-large',
+      env: { OPENAI_API_KEY: 'sk-test-embed-stale-1717' },
+    });
+    try {
+      await engine.putPage('notes/model-label', {
+        type: 'note',
+        title: 'model-label',
+        compiled_truth: '# model-label\n\nseeded',
+      });
+      await engine.upsertChunks('notes/model-label', [
+        {
+          chunk_index: 0,
+          chunk_text: 'already embedded elsewhere',
+          chunk_source: 'compiled_truth',
+          embedding: new Float32Array(1536).fill(0.01),
+          model: 'voyage:voyage-3',
+          token_count: 4,
+        },
+        {
+          chunk_index: 1,
+          chunk_text: 'stale chunk needing embed',
+          chunk_source: 'compiled_truth',
+          token_count: 5,
+          embedding: undefined, // stale
+        },
+      ]);
+
+      const result = await embedStaleForSource(engine, 'default', { embedFn: fakeEmbedFn });
+      expect(result.embedded).toBe(1);
+
+      const after = await engine.getChunks('notes/model-label');
+      const preserved = after.find((c) => c.chunk_index === 0)!;
+      const reembedded = after.find((c) => c.chunk_index === 1)!;
+      expect(reembedded.model).toBe('openai:text-embedding-3-large');
+      expect(preserved.model).toBe('voyage:voyage-3');
+    } finally {
+      resetGateway();
+    }
+  });
 });

test/embed.serial.test.ts

@@ -37,6 +37,8 @@ mock.module('../src/core/embedding.ts', () => ({
   // setPageEmbeddingSignature / invalidateStaleSignatureEmbeddings resolve to
   // null via the Proxy default, so the signature value is inert here.
   currentEmbeddingSignature: () => 'test:model:1536',
+  // #1717: embed paths stamp this label on (re)embedded chunks.
+  resolveEmbeddingModelLabel: () => 'openai:text-embedding-3-large',
 }));
 
 // Import AFTER mocking.
@@ -803,3 +805,34 @@ describe('embedAllStale --source threading (D7)', () => {
     expect((firstCallOpts as { sourceId?: string }).sourceId).toBe('media-corpus');
   });
 });
+
+// #1717: content_chunks.model must record the model that actually produced
+// each vector, not the gateway/engine default.
+describe('content_chunks.model labeling (#1717)', () => {
+  test('stamps the resolved embedding model on re-embedded chunks, preserves it on untouched chunks', async () => {
+    let upserted: any[] | undefined;
+    // Chunk 0 is stale (no embedded_at) → gets re-embedded this pass.
+    // Chunk 1 is already embedded with a DIFFERENT model → must be preserved,
+    // not relabeled to the current model.
+    const chunks = [
+      { chunk_index: 0, chunk_text: 'a', chunk_source: 'compiled_truth', embedded_at: null, model: 'zeroentropyai:zembed-1', token_count: 1 },
+      { chunk_index: 1, chunk_text: 'b', chunk_source: 'compiled_truth', embedded_at: '2026-01-01', embedding: new Float32Array(1536), model: 'voyage:voyage-3', token_count: 1 },
+    ];
+    const engine = mockEngine({
+      getPage: async () => ({ slug: 'notes/x', compiled_truth: 'a', timeline: '', source_id: 'default' }),
+      getChunks: async () => chunks,
+      upsertChunks: async (_slug: string, c: any[]) => { upserted = c; },
+      setPageEmbeddingSignature: async () => null,
+    });
+
+    await runEmbedCore(engine, { slugs: ['notes/x'] });
+
+    expect(upserted).toBeDefined();
+    const byIdx = Object.fromEntries(upserted!.map(c => [c.chunk_index, c]));
+    // Re-embedded chunk carries the model that produced its vector (was
+    // mislabeled with the default before the fix).
+    expect(byIdx[0].model).toBe('openai:text-embedding-3-large');
+    // Untouched chunk keeps its original model — no wholesale relabel.
+    expect(byIdx[1].model).toBe('voyage:voyage-3');
+  });
+});

test/import-signature-stamp.serial.test.ts

@@ -73,4 +73,21 @@ describe('importFromContent embedding_signature stamping (F1)', () => {
     await importFromContent(engine, 'concepts/unstamped', '# Unstamped\n\nbody content.', { noEmbed: true });
     expect(await signatureOf('concepts/unstamped')).toBeNull();
   });
+
+  // #1717: content_chunks.model must record the model that produced the
+  // vector (the configured gateway model), not the engine's hardcoded
+  // default. The gateway here is configured to openai:text-embedding-3-large,
+  // which differs from DEFAULT_EMBEDDING_MODEL — so this fails without the
+  // import-path model stamping.
+  test('inline embed labels content_chunks.model with the configured model (#1717)', async () => {
+    await importFromContent(engine, 'concepts/labeled', '# Labeled\n\nsome body content to chunk and embed.', {});
+    const rows = await engine.executeRaw<{ model: string }>(
+      `SELECT cc.model FROM content_chunks cc
+        JOIN pages p ON p.id = cc.page_id
+       WHERE p.slug = $1 AND p.source_id = 'default'`,
+      ['concepts/labeled'],
+    );
+    expect(rows.length).toBeGreaterThan(0);
+    for (const r of rows) expect(r.model).toBe('openai:text-embedding-3-large');
+  });
 });

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);
   });