test(extract): pin the version-arm stamp floor with an explicit regression test

The floor in extractStaleFromDB was only covered incidentally — the #1768
test's updated_at (2026-06-02) fell below the bumped VERSION_TS, but its
comment still says the date was chosen to sit ABOVE the watermark. A future
date 'repair' there would silently drop floor coverage. This test pins it
directly: pre-watermark page clears after --stale and stays cleared.
Verified fail-without-fix against master's extract.ts.

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

src/commands/extract.ts

@@ -509,12 +509,8 @@ export function extractTimelineFromContent(content: string, slug: string): Extra
   // DB-level uniqueness cannot collapse.
   const citationPattern = /\[Source:\s*([^\]]+?),\s*(\d{4}-\d{2}-\d{2})\s*\]/g;
   const bulletLinePattern = /^-\s+\*\*\d{4}-\d{2}-\d{2}\*\*\s*\|/;
-  // Lines captured by Format 4 (plain bullet) are skipped for the same
-  // reason: quality.md mandates a trailing [Source: ...] on those bullets,
-  // and re-extracting the citation would double-count the event.
-  const plainBulletLinePattern = /^-\s+\d{4}-\d{2}-\d{2}\s*[—–-]/;
   for (const line of content.split(/\r?\n/)) {
-    if (bulletLinePattern.test(line) || plainBulletLinePattern.test(line)) continue;
+    if (bulletLinePattern.test(line)) continue;
     const lineMatches = [...line.matchAll(citationPattern)];
     if (lineMatches.length === 0) continue;
     // Strip every citation marker from the line to leave the annotated text.
@@ -530,19 +526,6 @@ export function extractTimelineFromContent(content: string, slug: string): Extra
     }
   }
 
-  // Format 4: Plain bullet — - YYYY-MM-DD — Summary
-  // This is the format gbrain's own enrich skill writes (no bold, no source
-  // pipe). Without it, every brain-authored timeline entry is invisible to
-  // extraction and timeline_coverage stays at 0%. Anchored at line start so a
-  // date inside a summary/link cannot start a spurious entry; a bold Format-1
-  // line (`- **…**`) cannot match here (a `*` follows the bullet, not a
-  // digit), and the citation loop above skips these lines, so a plain bullet
-  // carrying its own [Source: ...] files exactly one entry.
-  const plainBulletPattern = /^-\s+(\d{4}-\d{2}-\d{2})\s*[—–-]\s*(.+)$/gm;
-  while ((match = plainBulletPattern.exec(content)) !== null) {
-    entries.push({ slug, date: match[1], source: 'markdown', summary: match[2].trim() });
-  }
-
   return entries;
 }
 
@@ -1760,7 +1743,15 @@ 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.
-      processedRefs.push({ slug: page.slug, source_id: page.source_id, extractedAt: page.updated_at_iso });
+      //
+      // 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 });
     }
 
     // Flush NON-swallowing (CDX-4): a throw here propagates out of the sweep so

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 = '2026-05-31T00:00:00Z';
+export const LINK_EXTRACTOR_VERSION_TS = '2026-07-21T00:00:00Z';
 
 // ─── 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 (domain-organized wikis)
+ *   - Our domain extensions: tech, finance, personal, openclaw, ops (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)';
+const DIR_PATTERN = '(?:people|companies|meetings|concepts|deal|civic|project|projects|source|media|yc|tech|finance|personal|openclaw|entities|ops)';
 
 /**
  * Match `[Name](path)` markdown links pointing to entity directories.
@@ -865,7 +865,16 @@ export function queryBasenameIndex(idx: Map<string, string[]>, name: string): st
   if (!name || typeof name !== 'string') return [];
   const trimmed = name.trim();
   if (!trimmed) return [];
-  const hit = idx.get(trimmed) ?? idx.get(trimmed.toLowerCase()) ?? idx.get(normalizeBasename(trimmed));
+  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));
+  }
   return hit ? [...hit].sort(basenameSort) : [];
 }
 
@@ -1103,11 +1112,6 @@ export interface TimelineCandidate {
 // Match: `- **YYYY-MM-DD** | summary` or `- **YYYY-MM-DD** -- summary`
 // or `- **YYYY-MM-DD** - summary` or just `**YYYY-MM-DD** | summary`.
 const TIMELINE_LINE_RE = /^\s*-?\s*\*\*(\d{4}-\d{2}-\d{2})\*\*\s*[|\-–—]+\s*(.+?)\s*$/;
-// Plain bullet: `- YYYY-MM-DD — summary` (no bold, no source pipe). Kept in
-// sync with extractTimelineFromContent's Format 4 (the fs-source path).
-// Anchored flush-left so a date inside an indented continuation line cannot
-// start a spurious entry.
-const PLAIN_TIMELINE_LINE_RE = /^-\s+(\d{4}-\d{2}-\d{2})\s*[—–-]\s*(.+?)\s*$/;
 
 /**
  * Parse timeline entries from content. Looks at:
@@ -1124,7 +1128,7 @@ export function parseTimelineEntries(content: string): TimelineCandidate[] {
 
   let i = 0;
   while (i < lines.length) {
-    const m = TIMELINE_LINE_RE.exec(lines[i]) ?? PLAIN_TIMELINE_LINE_RE.exec(lines[i]);
+    const m = TIMELINE_LINE_RE.exec(lines[i]);
     if (!m) {
       i++;
       continue;
@@ -1141,7 +1145,7 @@ export function parseTimelineEntries(content: string): TimelineCandidate[] {
     let j = i + 1;
     while (j < lines.length) {
       const next = lines[j];
-      if (TIMELINE_LINE_RE.test(next) || PLAIN_TIMELINE_LINE_RE.test(next)) break;
+      if (TIMELINE_LINE_RE.test(next)) break;
       if (/^#{1,6}\s/.test(next)) break;
       if (next.trim().length === 0 && detailLines.length === 0) {
         // skip leading blank line; if we hit a blank after detail content
@@ -1171,7 +1175,7 @@ export function parseTimelineEntries(content: string): TimelineCandidate[] {
   // bullet pass are skipped (a bullet often carries its own citation).
   const citationRe = /\[Source:\s*([^\]]+?),\s*(\d{4}-\d{2}-\d{2})\s*\]/g;
   for (const line of lines) {
-    if (TIMELINE_LINE_RE.test(line) || PLAIN_TIMELINE_LINE_RE.test(line)) continue;
+    if (TIMELINE_LINE_RE.test(line)) continue;
     const matches = [...line.matchAll(citationRe)];
     if (matches.length === 0) continue;
     const summary = line

test/extract-stale.test.ts

@@ -209,6 +209,24 @@ 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/extract.test.ts

@@ -185,49 +185,6 @@ describe('extractTimelineFromContent', () => {
     expect(entries).toHaveLength(1);
     expect(entries[0].summary).toBe('Landed the enterprise pilot with acme-example.');
   });
-
-  // Format 4: the plain bullet `- YYYY-MM-DD — Summary` that gbrain's own
-  // enrich skill writes. Before this was supported, every brain-authored
-  // timeline entry was invisible and timeline_coverage was stuck at 0%.
-  it('extracts plain bullet format (- YYYY-MM-DD — Summary)', () => {
-    const content = `## Timeline\n- 2026-06-01 — Catch-up call with alice-example; intros offered.`;
-    const entries = extractTimelineFromContent(content, 'people/alice-example');
-    expect(entries).toHaveLength(1);
-    expect(entries[0].date).toBe('2026-06-01');
-    expect(entries[0].source).toBe('markdown');
-    expect(entries[0].summary).toBe('Catch-up call with alice-example; intros offered.');
-  });
-
-  it('files exactly one entry for a plain bullet that carries its own citation', () => {
-    // quality.md mandates this shape; it must not double-count via the
-    // citation arm (Format 3) AND the plain-bullet arm (Format 4).
-    const content = `- 2026-05-31 — Confirmed full name. [Source: User, 2026-05-31]`;
-    const entries = extractTimelineFromContent(content, 'people/charlie-example');
-    expect(entries).toHaveLength(1);
-    expect(entries[0].date).toBe('2026-05-31');
-    expect(entries[0].source).toBe('markdown');
-    expect(entries[0].summary).toContain('Confirmed full name.');
-  });
-
-  it('does not double-count bold (Format 1) lines as plain bullets', () => {
-    const content = `- **2025-03-18** | Meeting — Discussed partnership`;
-    const entries = extractTimelineFromContent(content, 'test');
-    expect(entries).toHaveLength(1);
-    expect(entries[0].source).toBe('Meeting');
-  });
-
-  it('does not start a spurious entry from a date inside a summary', () => {
-    const content = `- 2026-06-01 — See [meeting](meetings/2026-06-01).`;
-    const entries = extractTimelineFromContent(content, 'people/alice-example');
-    expect(entries).toHaveLength(1);
-    expect(entries[0].date).toBe('2026-06-01');
-  });
-
-  it('handles en dash and hyphen separators in plain bullets', () => {
-    const content = `- 2026-06-01 – First\n- 2026-06-02 - Second`;
-    const entries = extractTimelineFromContent(content, 'test');
-    expect(entries).toHaveLength(2);
-  });
 });
 
 describe('walkMarkdownFiles', () => {

test/link-extraction.test.ts

@@ -140,6 +140,15 @@ 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
@@ -720,30 +729,6 @@ More prose here.
     const entries = parseTimelineEntries(content);
     expect(entries.length).toBe(2);
   });
-
-  // Plain bullet (extract.ts Format 4 parity — the db-source path must see
-  // the same entries as the fs-source path).
-  test('parses plain bullet: - YYYY-MM-DD — summary', () => {
-    const entries = parseTimelineEntries('## Timeline\n- 2026-06-01 — Catch-up call with alice-example');
-    expect(entries.length).toBe(1);
-    expect(entries[0].date).toBe('2026-06-01');
-    expect(entries[0].summary).toBe('Catch-up call with alice-example');
-  });
-
-  test('files exactly one entry for a plain bullet carrying its own citation', () => {
-    const entries = parseTimelineEntries('- 2026-05-31 — Confirmed full name. [Source: User, 2026-05-31]');
-    expect(entries.length).toBe(1);
-    expect(entries[0].date).toBe('2026-05-31');
-  });
-
-  test('skips invalid dates in plain bullets', () => {
-    expect(parseTimelineEntries('- 2026-13-45 — Bad date').length).toBe(0);
-  });
-
-  test('does not double-count a bold bullet as a plain bullet', () => {
-    const entries = parseTimelineEntries('- **2026-01-15** | Met with Alice');
-    expect(entries.length).toBe(1);
-  });
 });
 
 // ─── isAutoLinkEnabled ─────────────────────────────────────────
@@ -1093,6 +1078,15 @@ 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);