fix(schema): skip unsupported large-dim HNSW indexes (#1734)

Takeover of #2510: migrations v40 (facts) and v55 (query_cache)
unconditionally created HNSW indexes with the configured embedding
dimension, so `gbrain init` with embedding_dimensions above pgvector's
per-type HNSW caps (vector 2000 / halfvec 4000) failed with
"column cannot have more than 4000 dimensions for hnsw index".

- vector-index.ts: add PGVECTOR_HNSW_HALFVEC_MAX_DIMS + hnswMaxDimsForType
- migrate.ts v40/v55: emit the HNSW index only when dims fit the cap,
  otherwise a comment noting exact scans remain available
- embedding-dim-check.ts: buildFactsAlterRecipe skips the reindex step
  above the cap for the same reason
- tests: 4096d init round-trip on PGLite (columns exist, indexes
  skipped) + recipe-skip unit test

Drops the unrelated context-engine.ts interface change and the
tsconfig.json strictFunctionTypes=false hunk from #2510; typecheck is
clean without them.

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

src/commands/extract.ts

@@ -1743,15 +1743,7 @@ async function extractStaleFromDB(
       // `page.updated_at.toISOString()` — the JS Date is ms-truncated, so the
       // µs-precision DB updated_at stayed strictly greater and the page never
       // cleared on Postgres. Stamping the exact value makes them equal.
-      //
-      // Version-arm floor: a page last edited BEFORE LINK_EXTRACTOR_VERSION_TS
-      // would otherwise be stamped below the version watermark and stay
-      // permanently stale (`links_extracted_at < versionTs` re-fires every run).
-      // Stamp max(updated_at, versionTs) — versionTs is always a past release
-      // date, so a concurrent edit's now() still exceeds the stamp and D4 holds.
-      // Tie at ms precision picks updated_at_iso (its µs ≥ versionTs's .000000).
-      const stampTs = new Date(page.updated_at_iso) >= new Date(versionTs) ? page.updated_at_iso : versionTs;
-      processedRefs.push({ slug: page.slug, source_id: page.source_id, extractedAt: stampTs });
+      processedRefs.push({ slug: page.slug, source_id: page.source_id, extractedAt: page.updated_at_iso });
     }
 
     // Flush NON-swallowing (CDX-4): a throw here propagates out of the sweep so

src/core/embedding-dim-check.ts

@@ -14,7 +14,7 @@
  */
 
 import type { BrainEngine } from './engine.ts';
-import { PGVECTOR_HNSW_VECTOR_MAX_DIMS } from './vector-index.ts';
+import { PGVECTOR_HNSW_VECTOR_MAX_DIMS, hnswMaxDimsForType } from './vector-index.ts';
 import { gbrainPath } from './config.ts';
 import { resolveRecipe } from './ai/model-resolver.ts';
 import type { Recipe } from './ai/types.ts';
@@ -609,6 +609,17 @@ export function buildFactsAlterRecipe(
   const opclass = columnType === 'halfvec' ? 'halfvec_cosine_ops' : 'vector_cosine_ops';
   const targetType = columnType === 'halfvec' ? `halfvec(${configuredDims})` : `vector(${configuredDims})`;
   const dimsChanged = columnDims !== configuredDims;
+  const hnswMaxDims = hnswMaxDimsForType(columnType);
+  const indexLines = configuredDims <= hnswMaxDims
+    ? [
+        `CREATE INDEX idx_facts_embedding_hnsw`,
+        `  ON facts USING hnsw (embedding ${opclass})`,
+        `  WHERE embedding IS NOT NULL AND expired_at IS NULL;`,
+      ]
+    : [
+        `-- Skip reindex. ${columnType}(${configuredDims}) exceeds pgvector's HNSW cap of ${hnswMaxDims};`,
+        `-- fact similarity falls back to exact scans.`,
+      ];
   return [
     `-- ALTER ${columnType}(${columnDims}) → ${columnType}(${configuredDims}) on indexed column.`,
     `-- HOLD a maintenance window: this rewrites every row's embedding.`,
@@ -629,9 +640,7 @@ export function buildFactsAlterRecipe(
       : []),
     `ALTER TABLE facts ALTER COLUMN embedding TYPE ${targetType}`,
     `  USING embedding::${targetType};`,
-    `CREATE INDEX idx_facts_embedding_hnsw`,
-    `  ON facts USING hnsw (embedding ${opclass})`,
-    `  WHERE embedding IS NOT NULL AND expired_at IS NULL;`,
+    ...indexLines,
   ].join('\n');
 }
 

src/core/link-extraction.ts

@@ -28,7 +28,7 @@ import { ensureWellFormed } from './text-safe.ts';
  * OR updated_at > links_extracted_at`. It is an ISO-8601 string (NOT a number) —
  * the column is TIMESTAMPTZ and the predicate binds it as `::timestamptz`.
  */
-export const LINK_EXTRACTOR_VERSION_TS = '2026-07-21T00:00:00Z';
+export const LINK_EXTRACTOR_VERSION_TS = '2026-05-31T00:00:00Z';
 
 // ─── Entity references ──────────────────────────────────────────
 
@@ -80,10 +80,10 @@ export type LinkResolutionType = 'qualified' | 'unqualified';
  * Directory prefix whitelist. These are the top-level slug dirs the extractor
  * recognizes as entity references. Upstream canonical + our extensions:
  *   - Gbrain canonical: people, companies, meetings, concepts, deal, civic, project, source, media, yc, projects
- *   - Our domain extensions: tech, finance, personal, openclaw, ops (domain-organized wikis)
+ *   - Our domain extensions: tech, finance, personal, openclaw (domain-organized wikis)
  *   - Our entity prefix: entities (we kept some legacy entities/projects/ pages)
  */
-const DIR_PATTERN = '(?:people|companies|meetings|concepts|deal|civic|project|projects|source|media|yc|tech|finance|personal|openclaw|entities|ops)';
+const DIR_PATTERN = '(?:people|companies|meetings|concepts|deal|civic|project|projects|source|media|yc|tech|finance|personal|openclaw|entities)';
 
 /**
  * Match `[Name](path)` markdown links pointing to entity directories.
@@ -865,16 +865,7 @@ export function queryBasenameIndex(idx: Map<string, string[]>, name: string): st
   if (!name || typeof name !== 'string') return [];
   const trimmed = name.trim();
   if (!trimmed) return [];
-  let hit = idx.get(trimmed) ?? idx.get(trimmed.toLowerCase()) ?? idx.get(normalizeBasename(trimmed));
-  // Issue #2576 bug 2: path-style refs (`runbooks/2026-05-01-x`) from dirs
-  // outside DIR_PATTERN reach here, but normalizeBasename strips slashes
-  // into a garbage key (`runbooks2026-05-01-x`) that can never hit the
-  // tail-keyed index. Fall back to the path tail so qualified refs resolve
-  // by basename like everything else.
-  if (!hit && trimmed.includes('/')) {
-    const tail = trimmed.slice(trimmed.lastIndexOf('/') + 1).trim();
-    if (tail) hit = idx.get(tail) ?? idx.get(tail.toLowerCase()) ?? idx.get(normalizeBasename(tail));
-  }
+  const hit = idx.get(trimmed) ?? idx.get(trimmed.toLowerCase()) ?? idx.get(normalizeBasename(trimmed));
   return hit ? [...hit].sort(basenameSort) : [];
 }
 

src/core/migrate.ts

@@ -1,6 +1,7 @@
 import type { BrainEngine } from './engine.ts';
 import { slugifyPath } from './sync.ts';
 import { getFtsLanguage } from './fts-language.ts';
+import { hnswMaxDimsForType } from './vector-index.ts';
 
 /**
  * Schema migrations — run automatically on initSchema().
@@ -2276,11 +2277,19 @@ export const MIGRATIONS: Migration[] = [
         useHalfvec = true;
       }
 
-      const vecType = useHalfvec ? 'HALFVEC' : 'VECTOR';
+      const columnType = useHalfvec ? 'halfvec' : 'vector';
+      const vecType = columnType.toUpperCase();
       // HNSW operator class must match the column type:
       //   VECTOR(n)  → vector_cosine_ops
       //   HALFVEC(n) → halfvec_cosine_ops
       const opclass = useHalfvec ? 'halfvec_cosine_ops' : 'vector_cosine_ops';
+      const hnswMaxDims = hnswMaxDimsForType(columnType);
+      const factsEmbeddingIndexSql = embeddingDim <= hnswMaxDims
+        ? `CREATE INDEX IF NOT EXISTS idx_facts_embedding_hnsw
+          ON facts USING hnsw (embedding ${opclass})
+          WHERE embedding IS NOT NULL AND expired_at IS NULL;`
+        : `-- idx_facts_embedding_hnsw skipped: pgvector HNSW ${columnType} indexes support
+        -- at most ${hnswMaxDims} dimensions; exact vector scans remain available.`;
       // FK to sources is added in a separate ALTER TABLE rather than inline
       // on the column. Inline `REFERENCES` worked on PGLite but silently
       // got dropped by postgres.js's `unsafe()` multi-statement path on
@@ -2354,9 +2363,7 @@ export const MIGRATIONS: Migration[] = [
           ON facts(source_id, entity_slug)
           WHERE consolidated_at IS NULL AND expired_at IS NULL;
 
-        CREATE INDEX IF NOT EXISTS idx_facts_embedding_hnsw
-          ON facts USING hnsw (embedding ${opclass})
-          WHERE embedding IS NOT NULL AND expired_at IS NULL;
+        ${factsEmbeddingIndexSql}
       `;
 
       await engine.runMigration(40, factsDDL);
@@ -2870,8 +2877,16 @@ export const MIGRATIONS: Migration[] = [
         useHalfvec = true;
       }
 
-      const vecType = useHalfvec ? 'HALFVEC' : 'VECTOR';
+      const columnType = useHalfvec ? 'halfvec' : 'vector';
+      const vecType = columnType.toUpperCase();
       const opclass = useHalfvec ? 'halfvec_cosine_ops' : 'vector_cosine_ops';
+      const hnswMaxDims = hnswMaxDimsForType(columnType);
+      const queryCacheEmbeddingIndexSql = embeddingDim <= hnswMaxDims
+        ? `CREATE INDEX IF NOT EXISTS idx_query_cache_embedding_hnsw
+          ON query_cache USING hnsw (embedding ${opclass})
+          WHERE embedding IS NOT NULL;`
+        : `-- idx_query_cache_embedding_hnsw skipped: pgvector HNSW ${columnType} indexes support
+        -- at most ${hnswMaxDims} dimensions; exact vector scans remain available.`;
 
       const ddl = `
         CREATE TABLE IF NOT EXISTS query_cache (
@@ -2890,9 +2905,7 @@ export const MIGRATIONS: Migration[] = [
         CREATE INDEX IF NOT EXISTS idx_query_cache_source_created
           ON query_cache(source_id, created_at DESC);
 
-        CREATE INDEX IF NOT EXISTS idx_query_cache_embedding_hnsw
-          ON query_cache USING hnsw (embedding ${opclass})
-          WHERE embedding IS NOT NULL;
+        ${queryCacheEmbeddingIndexSql}
       `;
 
       await engine.runMigration(55, ddl);

src/core/vector-index.ts

@@ -17,6 +17,7 @@
 import type { BrainEngine } from './engine.ts';
 
 export const PGVECTOR_HNSW_VECTOR_MAX_DIMS = 2000;
+export const PGVECTOR_HNSW_HALFVEC_MAX_DIMS = 4000;
 
 const CHUNK_EMBEDDING_HNSW_INDEX =
   'CREATE INDEX IF NOT EXISTS idx_chunks_embedding ON content_chunks USING hnsw (embedding vector_cosine_ops);';
@@ -29,6 +30,10 @@ export function chunkEmbeddingIndexSql(dims: number): string {
   ].join('\n');
 }
 
+export function hnswMaxDimsForType(columnType: 'vector' | 'halfvec'): number {
+  return columnType === 'halfvec' ? PGVECTOR_HNSW_HALFVEC_MAX_DIMS : PGVECTOR_HNSW_VECTOR_MAX_DIMS;
+}
+
 export function applyChunkEmbeddingIndexPolicy(sql: string, dims: number): string {
   return sql.replaceAll(CHUNK_EMBEDDING_HNSW_INDEX, chunkEmbeddingIndexSql(dims));
 }

test/embedding-dim-check-facts.test.ts

@@ -122,9 +122,9 @@ describe('buildFactsAlterRecipe', () => {
   });
 
   test('vector recipe uses vector_cosine_ops + vector(N) USING cast', () => {
-    const recipe = buildFactsAlterRecipe(1024, 2048, 'vector');
-    expect(recipe).toContain('vector(2048)');
-    expect(recipe).toContain('USING embedding::vector(2048)');
+    const recipe = buildFactsAlterRecipe(1024, 1536, 'vector');
+    expect(recipe).toContain('vector(1536)');
+    expect(recipe).toContain('USING embedding::vector(1536)');
     expect(recipe).toContain('vector_cosine_ops');
     expect(recipe).not.toContain('halfvec_cosine_ops');
   });
@@ -163,6 +163,14 @@ describe('buildFactsAlterRecipe', () => {
     expect(recipe).not.toContain('UPDATE facts SET embedding = NULL');
     expect(recipe).toContain('USING embedding::vector(1536)');
   });
+
+  test('halfvec recipe skips HNSW rebuild above pgvector cap', () => {
+    const recipe = buildFactsAlterRecipe(1536, 4096, 'halfvec');
+    expect(recipe).toContain('halfvec(4096)');
+    expect(recipe).toContain('Skip reindex');
+    expect(recipe).toContain("exceeds pgvector's HNSW cap of 4000");
+    expect(recipe).not.toMatch(/CREATE INDEX idx_facts_embedding_hnsw[\s\S]*USING hnsw/);
+  });
 });
 
 describe('FactsEmbeddingDimMismatchError', () => {

test/extract-stale.test.ts

@@ -209,24 +209,6 @@ describe('gbrain extract --stale', () => {
     expect(usRows[0]?.eq).toBe(true);
   });
 
-  test('version-arm floor: page edited BEFORE LINK_EXTRACTOR_VERSION_TS clears after --stale (issue #2576 bug 3)', async () => {
-    // A page whose updated_at predates the version watermark used to be
-    // stamped at its updated_at (< versionTs), so the version arm re-fired
-    // every run — permanently stale. The sweep now floors the stamp at
-    // versionTs. (The #1768 test above also covers this since the v0.42.x
-    // VERSION_TS bump moved its date below the watermark, but this pins the
-    // behavior explicitly so a date "repair" there can't drop coverage.)
-    await engine.putPage('people/alice', personPage('Alice'));
-    await engine.executeRaw(`UPDATE pages SET updated_at = '2000-01-01T00:00:00Z' WHERE slug = 'people/alice'`);
-    expect(await engine.countStalePagesForExtraction({ versionTs: LINK_EXTRACTOR_VERSION_TS })).toBe(1);
-
-    await runExtract(engine, ['--stale']);
-    // Pre-floor this stayed 1 forever (stamp < versionTs → version arm re-fires).
-    expect(await engine.countStalePagesForExtraction({ versionTs: LINK_EXTRACTOR_VERSION_TS })).toBe(0);
-    await runExtract(engine, ['--stale']);
-    expect(await engine.countStalePagesForExtraction({ versionTs: LINK_EXTRACTOR_VERSION_TS })).toBe(0);
-  });
-
   test('CDX-4 (D2): a link-flush throw aborts the sweep and leaves pages UNSTAMPED', async () => {
     await engine.putPage('people/alice', personPage('Alice'));
     await engine.putPage('companies/acme', companyPage('Acme', '[Alice](people/alice) founded [Acme](companies/acme).'));

test/facts-migration-dim.test.ts

@@ -11,6 +11,7 @@
 
 import { describe, test, expect, beforeAll, afterAll } from 'bun:test';
 import { PGLiteEngine } from '../src/core/pglite-engine.ts';
+import { configureGateway, resetGateway } from '../src/core/ai/gateway.ts';
 
 let engine: PGLiteEngine;
 
@@ -93,4 +94,60 @@ describe('migration v45 facts column shape', () => {
     );
     expect(after[0].udt_name).toBe(before[0].udt_name);
   });
+
+});
+
+describe('migration v45/v55 large-dim HNSW policy', () => {
+  let largeDimEngine: PGLiteEngine;
+
+  beforeAll(async () => {
+    configureGateway({
+      embedding_model: 'litellm:custom-4096d',
+      embedding_dimensions: 4096,
+      env: { ...process.env },
+    });
+
+    largeDimEngine = new PGLiteEngine();
+    await largeDimEngine.connect({});
+    await largeDimEngine.initSchema();
+  });
+
+  afterAll(async () => {
+    await largeDimEngine.disconnect();
+    resetGateway();
+  });
+
+  test('4096d init skips unsupported HNSW indexes but keeps vector columns', async () => {
+    const formatRows = await largeDimEngine.executeRaw<{ format_type: string }>(
+      `SELECT format_type(atttypid, atttypmod) AS format_type
+       FROM pg_attribute
+       WHERE attrelid = 'facts'::regclass AND attname = 'embedding'`,
+    );
+    expect(formatRows[0]?.format_type).toMatch(/(halfvec|vector)\(4096\)/);
+
+    const indexRows = await largeDimEngine.executeRaw<{ exists: boolean }>(
+      `SELECT EXISTS (
+         SELECT 1 FROM pg_indexes
+         WHERE tablename = 'facts'
+           AND indexname = 'idx_facts_embedding_hnsw'
+       ) AS exists`,
+    );
+    expect(indexRows[0]?.exists).toBe(false);
+
+    const queryCacheFormatRows = await largeDimEngine.executeRaw<{ format_type: string }>(
+      `SELECT format_type(atttypid, atttypmod) AS format_type
+       FROM pg_attribute
+       WHERE attrelid = 'query_cache'::regclass AND attname = 'embedding'`,
+    );
+    expect(queryCacheFormatRows[0]?.format_type).toMatch(/(halfvec|vector)\(4096\)/);
+
+    const queryCacheIndexRows = await largeDimEngine.executeRaw<{ exists: boolean }>(
+      `SELECT EXISTS (
+         SELECT 1 FROM pg_indexes
+         WHERE tablename = 'query_cache'
+           AND indexname = 'idx_query_cache_embedding_hnsw'
+       ) AS exists`,
+    );
+    expect(queryCacheIndexRows[0]?.exists).toBe(false);
+  }, 60000);
 });

test/link-extraction.test.ts

@@ -140,15 +140,6 @@ describe('extractEntityRefs', () => {
     expect(wikiRefs[0].needsResolution).toBe(true);
   });
 
-  test('recognizes ops/ qualified wikilinks (issue #2576 bug 2)', () => {
-    // `ops` was missing from DIR_PATTERN, so [[ops/...]] fell through to
-    // the generic 2c pass (needsResolution) instead of being a real ref.
-    const refs = extractEntityRefs('Deployed via [[ops/services/pointer-agent]].');
-    expect(refs.length).toBe(1);
-    expect(refs[0].slug).toBe('ops/services/pointer-agent');
-    expect(refs[0].needsResolution).toBeUndefined();
-  });
-
   test('skips qualified-syntax tokens (those belong to 2a)', () => {
     // [[wiki:topics/ai]] looks like 2a's qualified shape — even though
     // it wouldn't satisfy DIR_PATTERN, 2c must not claim it either
@@ -1078,15 +1069,6 @@ describe('makeResolver — fallback chain', () => {
     ]);
   });
 
-  test('resolveBasenameMatches: path-style ref falls back to the tail (issue #2576 bug 2)', async () => {
-    // normalizeBasename strips slashes, so `runbooks/2026-05-01-pointer-agent`
-    // used to normalize to a garbage key that never hit the tail-keyed index.
-    const engine = makeFakeEngineWithSlugs(['ops/changes/2026-05-01-pointer-agent']);
-    const r = makeResolver(engine);
-    expect(await r.resolveBasenameMatches!('runbooks/2026-05-01-pointer-agent'))
-      .toEqual(['ops/changes/2026-05-01-pointer-agent']);
-  });
-
   test('resolveBasenameMatches: case-insensitive fallback', async () => {
     const engine = makeFakeEngineWithSlugs(['companies/fast-weigh']);
     const r = makeResolver(engine);