Merge origin/master — reconcile #1144 composition-aware brain_score with #3023 orphan-policy getHealth rewrite

Master's #3023 moved orphan_pages out of SQL into a TS filter over islanded rows (shared orphan-reporting policy). Kept that structure and threaded the #1144 narrative exclusion into the islanded query instead: code/calendar-index pages no longer enter the orphan numerator, matching the narrative_pages denominator. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
fix(openclaw): declare gbrain plugin manifest entry (takeover of #2551 ) (#3185 )
2026-08-16 18:02:30 +00:00 · 2026-07-22 11:12:32 -07:00 · 2026-07-22 01:32:11 -07:00 · 2026-07-22 00:23:10 -07:00 · 2026-07-21 20:22:27 -07:00 · 2026-07-21 19:52:18 -07:00
50 changed files with 1589 additions and 516 deletions
@@ -71,8 +71,8 @@ GBrain is designed to be installed and operated by an AI agent. The fastest path

 If you don't already have an AI agent platform running, start with one of these. Both are designed to read GBrain's install protocol and execute it:

- **[OpenClaw](https://github.com/openclawagents/openclaw)** — deploy [AlphaClaw on Render](https://render.com/deploy?repo=https://github.com/chrysb/alphaclaw) (one click, 8GB+ RAM)
- **[Hermes](https://github.com/openclawagents/hermes)** — deploy on [Railway](https://github.com/praveen-ks-2001/hermes-agent-template) (one click)
+- **[OpenClaw](https://github.com/openclaw/openclaw)** — deploy [AlphaClaw on Render](https://render.com/deploy?repo=https://github.com/chrysb/alphaclaw) (one click, 8GB+ RAM)
+- **[Hermes](https://github.com/NousResearch/hermes-agent)** — deploy on [Railway](https://github.com/praveen-ks-2001/hermes-agent-template) (one click)

 Then paste this into your agent:

@@ -131,7 +131,9 @@ into gbrain so other clients can scaffold it. Default behavior:
  `~/.gbrain/harvest-private-patterns.txt` plus built-in defaults
  (canonical private fork name, common email regex, Slack channel pattern). Any
  match → rollback (delete the harvested files) and exit non-zero.
- `openclaw.plugin.json` updated with the new slug, sorted.
+- `openclaw.plugin.json` updated with the new slug, sorted. Harvest must preserve
+  the top-level OpenClaw-native plugin fields (`id`, `configSchema`, `contracts`)
+  because OpenClaw validates those before it can install the package.
 - `--no-lint` bypasses the linter (after a manual editorial scrub).

 Use the `skillpack-harvest` skill (its companion editorial workflow)
@@ -233,13 +233,14 @@ keep it or `git checkout` to throw it away. Nothing is committed for you.

 **For a skill that ships with gbrain** (anything under the gbrain repo's own
 `skills/`): SkillOpt refuses to overwrite it by default and writes the winner to
-`skills/<name>/skillopt/best.md` instead, so an optimization pass can never
-silently mutate a skill other people depend on. Two ways to handle that:
+`skills/<name>/skillopt/proposed.md` instead (while keeping `best.md` as the
+optimizer's current-best pointer), so an optimization pass can never silently
+mutate a skill other people depend on. Two ways to handle that:

 ```bash
 # See the proposed improvement without touching SKILL.md (works for ANY skill):
 gbrain skillopt meeting-prep --split 1:1:1 --no-mutate
-# → writes skills/meeting-prep/skillopt/best.md (the proposed rewrite), prints its path. Copy what you want.
+# → writes skills/meeting-prep/skillopt/proposed.md, updates best.md, and prints the proposal path.

 # Actually rewrite a bundled skill (explicit opt-in + an independent held-out set):
 gbrain skillopt brain-ops --split 1:1:1 --allow-mutate-bundled \
@@ -1565,8 +1565,8 @@ GBrain is designed to be installed and operated by an AI agent. The fastest path

 If you don't already have an AI agent platform running, start with one of these. Both are designed to read GBrain's install protocol and execute it:

- **[OpenClaw](https://github.com/openclawagents/openclaw)** — deploy [AlphaClaw on Render](https://render.com/deploy?repo=https://github.com/chrysb/alphaclaw) (one click, 8GB+ RAM)
- **[Hermes](https://github.com/openclawagents/hermes)** — deploy on [Railway](https://github.com/praveen-ks-2001/hermes-agent-template) (one click)
+- **[OpenClaw](https://github.com/openclaw/openclaw)** — deploy [AlphaClaw on Render](https://render.com/deploy?repo=https://github.com/chrysb/alphaclaw) (one click, 8GB+ RAM)
+- **[Hermes](https://github.com/NousResearch/hermes-agent)** — deploy on [Railway](https://github.com/praveen-ks-2001/hermes-agent-template) (one click)

 Then paste this into your agent:

@@ -1,4 +1,5 @@
 {
+  "id": "gbrain-context-engine",
  "name": "gbrain",
  "version": "0.32.3.0",
  "description": "Personal knowledge brain with Postgres + pgvector hybrid search",
@@ -23,7 +23,7 @@ health_checks:
    label: "Auth provider"
    checks:
      - type: http
-        url: "$CLAWVISOR_URL/health"
+        url: "$CLAWVISOR_URL/ready"
        label: "ClawVisor"
      - type: env_exists
        name: GOOGLE_CLIENT_ID
@@ -135,7 +135,7 @@ Tell the user:

 Validate:
 ```bash
-curl -sf "$CLAWVISOR_URL/health" && echo "PASS: ClawVisor reachable" || echo "FAIL"
+curl -sf "$CLAWVISOR_URL/ready" && echo "PASS: ClawVisor reachable" || echo "FAIL"
 ```

 **STOP until ClawVisor validates.**
@@ -0,0 +1,308 @@
+---
+id: contacts-to-brain
+name: Contacts-to-Brain
+version: 0.1.0
+description: Google Contacts become canonical people/ pages, enriching brain entities with ground-truth name/email/phone/org data.
+category: sense
+requires: [credential-gateway]
+secrets:
+  - name: CLAWVISOR_URL
+    description: ClawVisor gateway URL (Option A — recommended, handles OAuth for you)
+    where: https://clawvisor.com — create an agent, activate Google Contacts service
+  - name: CLAWVISOR_AGENT_TOKEN
+    description: ClawVisor agent token (Option A)
+    where: https://clawvisor.com — agent settings, copy the agent token
+  - name: GOOGLE_CLIENT_ID
+    description: Google OAuth2 client ID (Option B — direct API access, you manage tokens)
+    where: https://console.cloud.google.com/apis/credentials — create OAuth 2.0 Client ID
+  - name: GOOGLE_CLIENT_SECRET
+    description: Google OAuth2 client secret (Option B)
+    where: https://console.cloud.google.com/apis/credentials — same page as client ID
+health_checks:
+  - type: any_of
+    label: "Auth provider"
+    checks:
+      - type: http
+        url: "$CLAWVISOR_URL/ready"
+        label: "ClawVisor"
+      - type: env_exists
+        name: GOOGLE_CLIENT_ID
+        label: "Google OAuth"
+setup_time: 15 min
+cost_estimate: "$0 (both options are free)"
+---
+
+# Contacts-to-Brain: Your Address Book Becomes Ground Truth
+
+Calendar attendees and email senders are the bulk of `people/<slug>` brain pages.
+Your Google Contacts is the ground-truth directory — canonical name, email, phone,
+organization — for those same entities. Syncing it closes the "who is this person"
+loop automatically and feeds enrichment.
+
+## IMPORTANT: Instructions for the Agent
+
+**You are the installer.** Follow these steps precisely.
+
+**Why this matters:** email-to-brain and calendar-to-brain create people pages
+from whatever name string the API returned — sometimes an email prefix, sometimes
+a nickname. Contacts carries the authoritative record. After this recipe runs,
+"j.smith@acme-example.com" and "Jon S" resolve to the same person page with the
+right display name, phone, and company.
+
+**The output is staging files, not direct writes:** the deterministic collector
+dumps contact records to `brain/contacts/.staging/`; YOU (the agent) merge them
+into `people/<slug>` pages using judgment — the Notability Gate in
+`skills/_brain-filing-rules.md` applies. Not every contact deserves a page.
+
+**Do not skip steps. Verify after each step.**
+
+## Architecture
+
+```
+Google Contacts (People API, paginated)
+  ↓ (ClawVisor credential gateway: list_contacts / get_contact / search_contacts)
+Contacts Sync Script (deterministic Node.js)
+  ↓ Outputs:
+  ├── brain/contacts/.raw/contacts-{date}.json      (raw API responses, provenance)
+  └── brain/contacts/.staging/{slug}.md             (one markdown record per contact)
+  ↓
+Agent reads staging files
+  ↓ Judgment calls (Notability Gate):
+  ├── Merge into existing people/<slug> pages (name/email/phone/org enrichment)
+  ├── Create new pages ONLY for notable contacts not already in brain
+  └── Skip the rest (staging is not the brain)
+```
+
+## Opinionated Defaults
+
+**Staging record format** (one file per contact, deterministic):
+```markdown
+# Alice Example
+
+- **Emails:** alice@acme-example.com, alice@gmail.com
+- **Phone:** +1 555 0100
+- **Organization:** Acme Example — VP Engineering
+- **Source:** Google Contacts (resourceName people/c123, synced 2026-07-21)
+```
+
+**Enrichment, not duplication:** if `people/alice-example.md` already exists,
+append missing fields to it with a `[Source: Google Contacts]` citation. Do NOT
+create a second page. Slug-match by normalized name, then by email against
+existing page content.
+
+**Notability Gate (from `skills/_brain-filing-rules.md`):** a contact with no
+brain presence gets a new page only if they appear elsewhere in the brain
+(calendar attendee, email correspondent) or the user confirms they matter.
+When in doubt, DON'T create — a junk page degrades search quality.
+
+## Prerequisites
+
+1. **GBrain installed and configured** (`gbrain doctor` passes)
+2. **Node.js 18+** (for the sync script)
+3. **Google Contacts access** via ONE of:
+   - **Option A: ClawVisor** (recommended, handles OAuth for you, no token management)
+   - **Option B: Google OAuth2 directly** (you manage tokens, no extra service needed)
+
+## Setup Flow
+
+### Step 1: Choose and Configure Contacts Access
+
+Ask the user: "How do you want to connect to Google Contacts?
+
+**Option A: ClawVisor (recommended)**
+ClawVisor handles OAuth, token refresh, and encryption. If you already use
+ClawVisor for email-to-brain or calendar-to-brain, this uses the same setup —
+just activate the Google Contacts service on your existing agent.
+
+**Option B: Google OAuth2 directly**
+Connect to the Google People API directly. No extra service needed, but you
+manage OAuth tokens yourself."
+
+#### Option A: ClawVisor Setup
+
+Tell the user:
+"I need your ClawVisor URL and agent token.
+1. Go to https://clawvisor.com
+2. Create an agent (or use existing)
+3. Activate the **Google Contacts** service
+4. Create a standing task with purpose: 'Full contacts access for people
+   enrichment: list contacts, read contact details, search contacts across
+   all connected Google accounts.'
+   IMPORTANT: Be EXPANSIVE in the task purpose. Narrow purposes block requests.
+5. Copy the gateway URL and agent token"
+
+Validate:
+```bash
+curl -sf "$CLAWVISOR_URL/ready" && echo "PASS: ClawVisor reachable" || echo "FAIL"
+```
+
+**STOP until ClawVisor validates.**
+
+#### Option B: Google OAuth2 Setup
+
+Same flow as `recipes/credential-gateway.md` Option B, with the contacts scope:
+
+1. https://console.cloud.google.com/apis/credentials — create an OAuth client ID
+   (Desktop app), consent screen scope: `https://www.googleapis.com/auth/contacts.readonly`
+2. Enable the People API: https://console.cloud.google.com/apis/library/people.googleapis.com
+3. Run the OAuth flow; store tokens in `~/.gbrain/google-tokens.json` (auto-refresh on expiry)
+
+Validate:
+```bash
+[ -n "$GOOGLE_CLIENT_ID" ] && [ -n "$GOOGLE_CLIENT_SECRET" ] \
+  && echo "PASS: Google OAuth credentials set" \
+  || echo "FAIL: Missing GOOGLE_CLIENT_ID or GOOGLE_CLIENT_SECRET"
+```
+
+**STOP until OAuth flow completes and tokens are stored.**
+
+### Step 2: Set Up the Contacts Sync Script
+
+```bash
+mkdir -p contacts-sync
+cd contacts-sync
+npm init -y
+```
+
+The sync script needs these capabilities:
+
+1. **Paginated retrieval** — `list_contacts` (People API `people.connections.list`)
+   returns pages of up to 100; follow `nextPageToken` until exhausted. Request
+   fields: names, emailAddresses, phoneNumbers, organizations, metadata.
+2. **Deterministic staging output** — one markdown file per contact at
+   `brain/contacts/.staging/{slug}.md`, slug from normalized display name
+   (fall back to email prefix). Same contact = same file on every run (idempotent).
+3. **Raw JSON preservation** — save raw API responses to
+   `brain/contacts/.raw/contacts-{date}.json` for provenance.
+4. **Skip empty records** — contacts with no name AND no email are noise; drop them.
+
+### Step 3: Run the Full Sync
+
+```bash
+node contacts-sync.mjs
+```
+
+Verify:
+```bash
+ls brain/contacts/.staging/ | head -10
+```
+
+Should show one file per contact, e.g. `alice-example.md`, `charlie-example.md`.
+
+### Step 4: Enrich People Pages (Agent Judgment)
+
+This is YOUR job (the agent). For each staging record:
+
+1. **Check brain**: `gbrain search "contact name"` — do they have a
+   `people/<slug>` page? Also search by email address.
+2. **Existing page** → merge the ground-truth fields (canonical name, emails,
+   phone, organization) into the page, each with a
+   `[Source: Google Contacts]` citation. Fix a wrong/partial display name.
+3. **No page** → apply the Notability Gate: create a page only if the contact
+   already appears in the brain (calendar, email) or is clearly relevant.
+   Otherwise skip.
+4. **Back-link** per the Iron Law in `skills/_brain-filing-rules.md`: an
+   organization with a brain page gets a link from the person's page and back.
+
+After enrichment, import and embed:
+```bash
+gbrain sync --no-pull --no-embed && gbrain embed --stale
+```
+
+Verify:
+```bash
+gbrain search "alice-example" --limit 3
+```
+
+Should return the enriched people page with contact details.
+
+### Step 5: Set Up Weekly Sync
+
+Contacts change slowly; once a week is plenty:
+```bash
+# Cron: every Sunday at 9 AM
+0 9 * * 0 cd /path/to/contacts-sync && node contacts-sync.mjs
+```
+
+After each sync, re-run the Step 4 enrichment pass over CHANGED staging files
+only (compare mtime or diff against git), then:
+```bash
+gbrain sync --no-pull --no-embed && gbrain embed --stale
+```
+
+### Step 6: Log Setup Completion
+
+```bash
+mkdir -p ~/.gbrain/integrations/contacts-to-brain
+echo '{"ts":"'$(date -u +%Y-%m-%dT%H:%M:%SZ)'","event":"setup_complete","source_version":"0.1.0","status":"ok","details":{"contacts":"CONTACT_COUNT"}}' >> ~/.gbrain/integrations/contacts-to-brain/heartbeat.jsonl
+```
+
+Tell the user: "Contacts-to-brain is set up. [N] contacts staged; [M] people
+pages enriched with ground-truth contact data. Weekly sync keeps it current."
+
+## Implementation Guide
+
+### Pagination
+
+```
+list_all_contacts():
+  contacts = []
+  token = null
+  do:
+    page = list_contacts({ pageSize: 100, pageToken: token,
+      personFields: 'names,emailAddresses,phoneNumbers,organizations,metadata' })
+    contacts += page.connections
+    token = page.nextPageToken
+  while token
+  return contacts
+```
+
+### Slug Normalization
+
+```
+slugify(contact):
+  name = contact.names?[0]?.displayName
+  if not name:
+    name = contact.emailAddresses?[0]?.value.split('@')[0]
+  return name.toLowerCase().normalize('NFD')
+             .replace(/[^a-z0-9]+/g, '-').replace(/^-|-$/g, '')
+```
+
+Collision (two contacts, same slug): suffix with the email domain
+(`alice-example-acme-example`) rather than overwrite.
+
+### What the Agent Should Test After Setup
+
+1. **Idempotency:** run the sync twice. `git status` on `brain/contacts/.staging/`
+   shows no changes on the second run.
+2. **Pagination:** with 250+ contacts, verify the staging count matches the
+   Google Contacts count (not capped at 100).
+3. **Notability Gate:** verify a one-off contact with no brain presence did NOT
+   get a `people/` page.
+4. **Enrichment merge:** verify an existing people page gained contact fields
+   without losing its prior content, each with a `[Source: Google Contacts]`
+   citation.
+
+## Cost Estimate
+
+| Component | Monthly Cost |
+|-----------|-------------|
+| ClawVisor (free tier) | $0 |
+| Google People API | $0 (within free quota) |
+| **Total** | **$0** |
+
+## Troubleshooting
+
+**No contacts returned:**
+- Check ClawVisor has the Google Contacts service activated
+- Check the standing task purpose is expansive enough
+- Option B: verify the People API is enabled and the token carries
+  `contacts.readonly`
+
+**Duplicate people pages after enrichment:**
+- The agent matched by name but the brain page slug differs — search by email
+  address too before creating, then merge and delete the duplicate
+
+**Contacts with no name:**
+- The sync script falls back to the email prefix; records with neither name
+  nor email are dropped as noise
@@ -23,7 +23,7 @@ health_checks:
    label: "Auth provider"
    checks:
      - type: http
-        url: "$CLAWVISOR_URL/health"
+        url: "$CLAWVISOR_URL/ready"
        label: "ClawVisor"
      - type: env_exists
        name: GOOGLE_CLIENT_ID
@@ -75,7 +75,7 @@ Tell the user:
 3. Activate the services you need:
   - **Gmail** (for email-to-brain)
   - **Google Calendar** (for calendar-to-brain)
-   - **Google Contacts** (for enrichment)
+   - **Google Contacts** (for contacts-to-brain)
 4. Create a standing task with a broad purpose. CRITICAL: be EXPANSIVE.

   Good purpose: 'Full executive assistant access to Gmail, Calendar, and
@@ -88,7 +88,7 @@ Tell the user:

 Validate:
 ```bash
-curl -sf "$CLAWVISOR_URL/health" \
+curl -sf "$CLAWVISOR_URL/ready" \
  && echo "PASS: ClawVisor reachable" \
  || echo "FAIL: ClawVisor not reachable — check the URL"
 ```
@@ -171,7 +171,7 @@ can now access your Google services."

 ## How to Verify

-1. **ClawVisor:** `curl $CLAWVISOR_URL/health` returns OK.
+1. **ClawVisor:** `curl $CLAWVISOR_URL/ready` returns OK.
 2. **Google OAuth:** Tokens exist at `~/.gbrain/google-tokens.json`.
 3. **Gmail access:** Run the email collector — it should pull recent messages.
 4. **Calendar access:** Run the calendar sync — it should pull today's events.
@@ -23,7 +23,7 @@ health_checks:
    label: "Auth provider"
    checks:
      - type: http
-        url: "$CLAWVISOR_URL/health"
+        url: "$CLAWVISOR_URL/ready"
        label: "ClawVisor"
      - type: env_exists
        name: GOOGLE_CLIENT_ID
@@ -130,7 +130,7 @@ Tell the user:

 Validate:
 ```bash
-curl -sf "$CLAWVISOR_URL/health" && echo "PASS: ClawVisor reachable" || echo "FAIL"
+curl -sf "$CLAWVISOR_URL/ready" && echo "PASS: ClawVisor reachable" || echo "FAIL"
 ```

 **STOP until ClawVisor validates.**
@@ -328,7 +328,7 @@ threads you already replied to. Sent mail acts as a negative filter.
 ## Troubleshooting

 **No emails collected:**
- Check ClawVisor health: `curl $CLAWVISOR_URL/health`
+- Check ClawVisor health: `curl $CLAWVISOR_URL/ready`
 - Check standing task is active and has Gmail service enabled
 - Check task purpose is expansive enough (narrow purposes block requests)

@@ -266,4 +266,5 @@ editorial pass.
  (e.g. `src/commands/<slug>.ts` if the host SKILL.md declares it
  in frontmatter)
 - gbrain's `openclaw.plugin.json` — adds the slug to `skills:`
-  array, sorted alphabetically
+  array, sorted alphabetically, without removing OpenClaw-native plugin fields
+  like `id`, `configSchema`, or `contracts`
@@ -57,6 +57,8 @@ This mode guarantees:
 - `skills/manifest.json` lists every skill directory
 - `skills/RESOLVER.md` references every skill in the manifest
 - `openclaw.plugin.json` `skills[]` round-trips with both
+- `openclaw.plugin.json` keeps OpenClaw install-required native plugin fields
+  (`id`, object `configSchema`, and `contracts.contextEngines` when applicable)
 - No MECE violations (duplicate triggers across skills)

 ### Phases
@@ -72,7 +74,7 @@ This mode guarantees:
 ### Automation

 ```bash
-bun test test/skills-conformance.test.ts test/resolver.test.ts
+bun test test/skills-conformance.test.ts test/resolver.test.ts test/openclaw-plugin-manifest.test.ts
 ```

 The CI-gated check is the package.json `test` script.
@@ -54,7 +54,7 @@ export function bigintToStringReplacer(_key: string, value: unknown): unknown {
 }

 // CLI-only commands that bypass the operation layer
-export const CLI_ONLY = new Set(['init', 'reinit-pglite', 'upgrade', 'post-upgrade', 'check-update', 'integrations', 'publish', 'check-backlinks', 'lint', 'report', 'import', 'export', 'files', 'embed', 'serve', 'call', 'config', 'doctor', 'migrate', 'eval', 'sync', 'extract', 'extract-conversation-facts', 'enrich', 'features', 'autopilot', 'graph-query', 'jobs', 'agent', 'apply-migrations', 'skillpack-check', 'skillpack', 'resolvers', 'integrity', 'repair-jsonb', 'orphans', 'sources', 'mounts', 'dream', 'check-resolvable', 'routing-eval', 'skillify', 'smoke-test', 'providers', 'storage', 'repos', 'code-def', 'code-refs', 'reindex', 'reindex-code', 'reindex-frontmatter', 'code-callers', 'code-callees', 'reconcile-links', 'frontmatter', 'auth', 'friction', 'claw-test', 'book-mirror', 'takes', 'think', 'salience', 'anomalies', 'calibration', 'transcripts', 'models', 'remote', 'recall', 'forget', 'edges-backfill', 'facts', 'cache', 'ze-switch', 'founder', 'brainstorm', 'lsd', 'schema', 'capture', 'onboard', 'conversation-parser', 'status', 'connect', 'skillopt', 'quarantine', 'self-upgrade', 'advisor', 'watch', 'reindex-search-vector']);
+export const CLI_ONLY = new Set(['init', 'reinit-pglite', 'upgrade', 'post-upgrade', 'check-update', 'integrations', 'publish', 'check-backlinks', 'lint', 'report', 'import', 'export', 'files', 'embed', 'serve', 'call', 'config', 'doctor', 'migrate', 'eval', 'sync', 'extract', 'extract-conversation-facts', 'enrich', 'features', 'autopilot', 'graph-query', 'jobs', 'agent', 'apply-migrations', 'skillpack-check', 'skillpack', 'resolvers', 'integrity', 'repair-jsonb', 'orphans', 'maintain', 'sources', 'mounts', 'dream', 'check-resolvable', 'routing-eval', 'skillify', 'smoke-test', 'providers', 'storage', 'repos', 'code-def', 'code-refs', 'reindex', 'reindex-code', 'reindex-frontmatter', 'code-callers', 'code-callees', 'reconcile-links', 'frontmatter', 'auth', 'friction', 'claw-test', 'book-mirror', 'takes', 'think', 'salience', 'anomalies', 'calibration', 'transcripts', 'models', 'remote', 'recall', 'forget', 'edges-backfill', 'cache', 'ze-switch', 'founder', 'brainstorm', 'lsd', 'schema', 'capture', 'onboard', 'conversation-parser', 'status', 'connect', 'skillopt', 'quarantine', 'self-upgrade', 'advisor', 'watch', 'reindex-search-vector']);
 // CLI-only commands whose handlers print their own --help text. These are
 // excluded from the generic short-circuit so detailed per-command and
 // per-subcommand usage stays reachable.
@@ -78,6 +78,8 @@ const CLI_ONLY_SELF_HELP = new Set([
  'capture',
  // v0.42 self-upgrade ships its own usage (flags + the agent-skill story).
  'self-upgrade',
+  // maintain (#3015) prints its own usage block (modes + not-auto-applied list).
+  'maintain',
  // v0.43 (#2095): watch ships WATCH_HELP (flags + the stdin-turn protocol).
  'watch',
  // v0.37 fix wave (Lane D.4 + CDX2-12): sync's --no-embed flag was
@@ -991,8 +993,6 @@ const THIN_CLIENT_REFUSED_COMMANDS = new Set([
  // hint pointing at the routable MCP tools; per-subcommand splits are
  // a v0.31.x follow-up TODO.
  'takes', 'sources',
-  // #1867: fence-backfill edits local .md fences + stamps the local DB.
-  'facts',
  // v0.32 thin-client routing audit (Codex round 2 findings #2, #4):
  // - `pages` purge-deleted is admin+localOnly (operations.ts:856-864)
  // - `files` list / file_url MCP ops are localOnly (operations.ts:1769-1879)
@@ -1028,7 +1028,6 @@ const THIN_CLIENT_REFUSE_HINTS: Record<string, string> = {
  migrate: "migrate runs on the host's local engine. Run on the host machine.",
  'apply-migrations': 'schema migrations run on the host. SSH and run there.',
  'repair-jsonb': 'repair-jsonb operates on the local DB only.',
-  facts: 'facts fence-backfill edits local entity-page fences. Run on the host machine.',
  integrity: 'integrity scans local files. Run on the host machine.',
  serve: 'serve starts a server. Run on the host, not the thin client.',
  dream: 'dream runs the autopilot cycle on the host. `gbrain remote ping` queues one. (Native `gbrain dream` thin-client routing planned for v0.31.2.)',
@@ -1760,6 +1759,11 @@ async function handleCliOnly(command: string, args: string[]) {
        await runOrphans(engine, args);
        break;
      }
+      case 'maintain': {
+        const { runMaintain } = await import('./commands/maintain.ts');
+        await runMaintain(engine, args);
+        break;
+      }
      // v0.32.7 CJK wave — post-upgrade markdown re-chunk sweep.
      // v0.36 Phase 3 wave — `gbrain reindex --multimodal` re-embeds content_chunks
      // into the unified Voyage multimodal-3 column.
@@ -1858,13 +1862,6 @@ async function handleCliOnly(command: string, args: string[]) {
        await runEdgesBackfill(engine, args);
        break;
      }
-      case 'facts': {
-        // #1867 — re-runnable fence-backfill for row_num-NULL legacy fact
-        // rows (idempotent v0_32_2 phase B, exposed as an operator command).
-        const { runFactsCommand } = await import('./commands/facts.ts');
-        await runFactsCommand(engine, args);
-        break;
-      }
      case 'whoknows': {
        // v0.33 (Issue #?): expertise + relationship-proximity routing.
        // MCP op `find_experts` (read-scoped) backs the same code path; CLI
@@ -2345,7 +2342,6 @@ TOOLS
  check-backlinks <check|fix> [dir]  Find/fix missing back-links across brain
  lint <dir|file> [--fix]            Catch LLM artifacts, placeholder dates, bad frontmatter
  orphans [--json] [--count]         Find pages with no inbound wikilinks
-  facts fence-backfill [--dry-run]   Fence legacy fact rows (row_num NULL) onto entity pages
  salience [--days N] [--kind P]     v0.29: pages ranked by emotional + activity salience
  anomalies [--since D] [--sigma N]  v0.29: cohort-based statistical anomalies (tag, type)
  transcripts recent [--days N]      v0.29: recent raw .txt transcripts (local-only)
@@ -581,7 +581,7 @@ async function embedPage(
  for (let j = 0; j < toEmbed.length; j++) {
    embeddingMap.set(toEmbed[j].chunk_index, embeddings[j]);
  }
-  const updated: ChunkInput[] = chunks.map(c => ({
+  const updated: ChunkInput[] = chunks.map(c => preserveCodeMetadata(c, {
    chunk_index: c.chunk_index,
    chunk_text: c.chunk_text,
    chunk_source: c.chunk_source,
@@ -605,6 +605,31 @@ async function embedPage(
  slog(`${slug}: embedded ${toEmbed.length} chunks`);
 }

+/**
+ * Carry code-chunk metadata (language, symbol_name, symbol_type, line range,
+ * parent scope, doc comment, qualified name) from a loaded Chunk back into a
+ * ChunkInput destined for upsertChunks.
+ *
+ * Issue #769: every re-embed used to strip these fields, and upsertChunks
+ * overwrites (does not COALESCE) the metadata columns from EXCLUDED, so
+ * each pass clobbered code-def's primary index to NULL. Pulling the
+ * preservation into one helper keeps the three re-embed call sites
+ * (embedPage, embedAll non-stale, embedAllStale) in lock-step.
+ */
+function preserveCodeMetadata(loaded: any, base: ChunkInput): ChunkInput {
+  return {
+    ...base,
+    language: loaded.language ?? undefined,
+    symbol_name: loaded.symbol_name ?? undefined,
+    symbol_type: loaded.symbol_type ?? undefined,
+    start_line: loaded.start_line ?? undefined,
+    end_line: loaded.end_line ?? undefined,
+    parent_symbol_path: loaded.parent_symbol_path ?? undefined,
+    doc_comment: loaded.doc_comment ?? undefined,
+    symbol_name_qualified: loaded.symbol_name_qualified ?? undefined,
+  };
+}
+
 async function embedAll(
  engine: BrainEngine,
  staleOnly: boolean,
@@ -717,8 +742,10 @@ async function embedAll(
      for (let j = 0; j < toEmbed.length; j++) {
        embeddingMap.set(toEmbed[j].chunk_index, embeddings[j]);
      }
-      // Preserve ALL chunks, only update embeddings for stale ones
-      const updated: ChunkInput[] = chunks.map(c => ({
+      // Preserve ALL chunks, only update embeddings for stale ones.
+      // preserveCodeMetadata threads code-chunk metadata (#769) so re-embed
+      // doesn't clobber language/symbol_name/symbol_type to NULL.
+      const updated: ChunkInput[] = chunks.map(c => preserveCodeMetadata(c, {
        chunk_index: c.chunk_index,
        chunk_text: c.chunk_text,
        chunk_source: c.chunk_source,
@@ -1012,7 +1039,10 @@ async function embedAllStale(
          for (let j = 0; j < stale.length; j++) {
            staleIdxToEmbedding.set(stale[j].chunk_index, embeddings[j]);
          }
-          const merged: ChunkInput[] = existing.map(c => ({
+          // preserveCodeMetadata threads code-chunk metadata (#769) so the
+          // autopilot --stale path doesn't clobber language/symbol_name/etc
+          // to NULL on every cycle.
+          const merged: ChunkInput[] = existing.map(c => preserveCodeMetadata(c, {
            chunk_index: c.chunk_index,
            chunk_text: c.chunk_text,
            chunk_source: c.chunk_source,
@@ -1651,7 +1651,7 @@ async function extractTimelineFromDB(
 * make re-extraction idempotent). EVERY processed page is stamped, including
 * zero-link pages — they WERE processed.
 */
-async function extractStaleFromDB(
+export async function extractStaleFromDB(
  engine: BrainEngine,
  opts: {
    dryRun: boolean;
@@ -1,48 +0,0 @@
-/**
- * gbrain facts — fact-store maintenance surface (#1867).
- *
- * `fence-backfill` re-runs the v0_32_2 fence-backfill phase on demand.
- * Remote `extract_facts` deposits that predate the fence-write backstop
- * (and any legacy DB-only insert) leave `row_num IS NULL` rows that the
- * cycle extract_facts guard refuses to reconcile past — previously the
- * only remedy was the one-shot v0_32_2 migration, which the ledger marks
- * complete and never re-runs. The phase is idempotent (only touches
- * `row_num IS NULL` rows), so exposing it as a command is safe to re-run
- * any time the backlog reappears.
- */
-import type { BrainEngine } from '../core/engine.ts';
-import { setCliExitVerdict } from '../core/cli-force-exit.ts';
-import { phaseBFenceFacts } from './migrations/v0_32_2.ts';
-
-function printHelp(): void {
-  process.stderr.write(
-    `Usage: gbrain facts fence-backfill [--dry-run]\n\n` +
-      `Fence-backfill: appends every legacy fact row (row_num IS NULL) to its\n` +
-      `entity page's \`## Facts\` fence and stamps row_num + source_markdown_slug\n` +
-      `back onto the DB row. Idempotent — re-runs only pick up rows still\n` +
-      `missing a fence assignment. Clears the backlog that makes the cycle's\n` +
-      `extract_facts phase skip fence→DB reconciliation.\n\n` +
-      `  --dry-run   report what would be fenced; no FS or DB writes\n`,
-  );
-}
-
-export async function runFactsCommand(engine: BrainEngine, args: string[]): Promise<void> {
-  const sub = args[0];
-  if (!sub || sub === '--help' || sub === '-h') {
-    printHelp();
-    return;
-  }
-  if (sub !== 'fence-backfill') {
-    process.stderr.write(`Unknown facts subcommand: ${sub}\n`);
-    printHelp();
-    setCliExitVerdict(1);
-    return;
-  }
-
-  const dryRun = args.includes('--dry-run');
-  const result = await phaseBFenceFacts(engine, { dryRun });
-  process.stderr.write(
-    `fence-backfill: ${result.status}${result.detail ? ` — ${result.detail}` : ''}\n`,
-  );
-  if (result.status === 'failed') setCliExitVerdict(1);
-}
@@ -334,6 +334,10 @@ export async function scanIntegrity(
    if (!page) continue;
    // Skip grandfathered pages (opted out of brain-integrity enforcement)
    if ((page.frontmatter as Record<string, unknown> | undefined)?.validate === false) continue;
+    // Skip code pages: indexed source files aren't prose. 'tweet' in an
+    // identifier or comment is not a bare-tweet citation gap, and auto-repair
+    // would inject wikilink brackets into source code.
+    if (page.type === 'code') continue;
    pagesScanned++;
    bareHits.push(...findBareTweetHits(page.compiled_truth, slug));
    externalHits.push(...findExternalLinks(page.compiled_truth, slug));
@@ -363,6 +367,9 @@ async function scanIntegrityBatch(
  // YAML) diverges from the sequential path's strict === false check. Intentional
  // — gbrain lint should reject stringly-typed validate at write time.
  const validateCondition = sql`AND (frontmatter->>'validate' IS NULL OR frontmatter->>'validate' != 'false')`;
+  // Mirror of the sequential path's `page.type === 'code'` skip: code pages
+  // (indexed source files) are never prose-integrity candidates.
+  const codeCondition = sql`AND type IS DISTINCT FROM 'code'`;

  // v0.32.8: scan ONE row per (source_id, slug) pair, not one per slug.
  // Pre-fix used DISTINCT ON (slug) which collapsed multi-source rows into
@@ -372,7 +379,7 @@ async function scanIntegrityBatch(
  const rows = await sql`
    SELECT slug, compiled_truth, frontmatter
    FROM pages
-    WHERE 1=1 ${typeCondition} ${validateCondition}
+    WHERE 1=1 ${typeCondition} ${validateCondition} ${codeCondition}
    ORDER BY source_id, slug
    LIMIT ${limit}
  `;
@@ -461,6 +468,9 @@ async function cmdAuto(args: string[]): Promise<void> {

      const page = await engine.getPage(slug, { sourceId: source_id });
      if (!page) continue;
+      // Never auto-repair code pages — injecting tweet citations into
+      // indexed source files corrupts them. Same gate as scanIntegrity.
+      if (page.type === 'code') continue;

      pagesProcessed++;
      progress.tick(1, slug);
@@ -0,0 +1,224 @@
+/**
+ * gbrain maintain — conservative self-healing maintenance.
+ *
+ * This command automates the safe parts of the operator runbook:
+ *   - stale link/timeline extraction
+ *   - stale per-source dream cycles when doctor reports cycle_freshness
+ *
+ * It deliberately does NOT mutate source files, apply schema-pack upgrades, or
+ * invent semantic hub links. Those need review or a separate command with an
+ * auditable proposal surface.
+ */
+
+import { existsSync } from 'fs';
+import type { BrainEngine } from '../core/engine.ts';
+import type { BrainHealth } from '../core/types.ts';
+import { buildChecks, computeDoctorReport, type DoctorReport, type Check } from './doctor.ts';
+import { extractStaleFromDB } from './extract.ts';
+import { runCycle, type CycleReport } from '../core/cycle.ts';
+
+type ActionStatus = 'ok' | 'would_apply' | 'applied' | 'blocked' | 'skipped';
+
+export interface MaintenanceAction {
+  name: string;
+  status: ActionStatus;
+  message: string;
+  details?: Record<string, unknown>;
+}
+
+export interface MaintainOptions {
+  json: boolean;
+  safe: boolean;
+  dryRun: boolean;
+  help: boolean;
+}
+
+export interface MaintainReport {
+  mode: 'dry-run' | 'safe';
+  before: {
+    health: BrainHealth;
+    doctor: DoctorReport;
+  };
+  actions: MaintenanceAction[];
+  after: {
+    health: BrainHealth;
+    doctor: DoctorReport;
+  };
+}
+
+export function parseMaintainArgs(args: string[]): MaintainOptions {
+  const safe = args.includes('--safe');
+  return {
+    json: args.includes('--json'),
+    safe,
+    dryRun: args.includes('--dry-run') || !safe,
+    help: args.includes('--help') || args.includes('-h'),
+  };
+}
+
+export function extractCycleFreshnessSourceIds(checks: Check[]): string[] {
+  const ids = new Set<string>();
+  for (const check of checks) {
+    if (check.name !== 'cycle_freshness' || check.status === 'ok') continue;
+    const re = /Source '([^']+)' last cycled/g;
+    for (const match of check.message.matchAll(re)) {
+      const id = match[1]?.trim();
+      if (id) ids.add(id);
+    }
+  }
+  return [...ids].sort();
+}
+
+async function buildDoctorReport(engine: BrainEngine): Promise<DoctorReport> {
+  const checks = await buildChecks(engine, ['--json', '--scope=brain']);
+  return computeDoctorReport(checks);
+}
+
+async function runStaleExtraction(
+  engine: BrainEngine,
+  beforeHealth: BrainHealth,
+  dryRun: boolean,
+): Promise<MaintenanceAction> {
+  if (beforeHealth.stale_pages <= 0) {
+    return { name: 'extract_stale', status: 'ok', message: 'No stale pages.' };
+  }
+
+  if (dryRun) {
+    return {
+      name: 'extract_stale',
+      status: 'would_apply',
+      message: `Would run DB-backed stale extraction for ${beforeHealth.stale_pages} page(s).`,
+      details: { stale_pages: beforeHealth.stale_pages },
+    };
+  }
+
+  const result = await extractStaleFromDB(engine, {
+    dryRun: false,
+    jsonMode: false,
+    includeFrontmatter: false,
+    catchUp: false,
+  });
+
+  return {
+    name: 'extract_stale',
+    status: 'applied',
+    message: `Processed ${result.pagesProcessed} stale page(s); ${result.staleRemaining} remain.`,
+    details: {
+      links_created: result.linksCreated,
+      timeline_created: result.timelineCreated,
+      pages_processed: result.pagesProcessed,
+      stale_remaining: result.staleRemaining,
+    },
+  };
+}
+
+async function runCycleFreshnessMaintenance(
+  engine: BrainEngine,
+  beforeDoctor: DoctorReport,
+  dryRun: boolean,
+): Promise<MaintenanceAction[]> {
+  const sourceIds = extractCycleFreshnessSourceIds(beforeDoctor.checks);
+  if (sourceIds.length === 0) {
+    return [{ name: 'cycle_freshness', status: 'ok', message: 'All sources cycled recently.' }];
+  }
+
+  if (dryRun) {
+    return sourceIds.map((sourceId) => ({
+      name: 'cycle_freshness',
+      status: 'would_apply',
+      message: `Would run source-scoped dream cycle for ${sourceId}.`,
+      details: { source_id: sourceId },
+    }));
+  }
+
+  const sources = await engine.listAllSources();
+  const actions: MaintenanceAction[] = [];
+
+  for (const sourceId of sourceIds) {
+    const source = sources.find((s) => s.id === sourceId);
+    const localPath = source?.local_path ?? null;
+    const brainDir = localPath && existsSync(localPath) ? localPath : null;
+    const report: CycleReport = await runCycle(engine, {
+      brainDir,
+      dryRun: false,
+      pull: false,
+      sourceId,
+    });
+    actions.push({
+      name: 'cycle_freshness',
+      status: report.status === 'failed' ? 'blocked' : 'applied',
+      message: `Ran source-scoped dream cycle for ${sourceId}: ${report.status}.`,
+      details: {
+        source_id: sourceId,
+        brain_dir: brainDir,
+        cycle_status: report.status,
+        phases: report.phases.map((p) => ({ phase: p.phase, status: p.status })),
+      },
+    });
+  }
+
+  return actions;
+}
+
+export async function runMaintain(engine: BrainEngine, args: string[]): Promise<MaintainReport | void> {
+  const opts = parseMaintainArgs(args);
+  if (opts.help) {
+    console.log(`Usage: gbrain maintain [--safe] [--dry-run] [--json]
+
+Conservative self-healing maintenance.
+
+Modes:
+  --dry-run   Preview safe actions without writes. Default when --safe is absent.
+  --safe      Apply safe actions: stale extraction and source cycle freshness.
+  --json      Emit a structured before/action/after report.
+
+Not auto-applied:
+  source-file frontmatter fixes, schema-pack upgrades, atom-pack changes,
+  semantic hub-link guesses, and destructive cleanup.
+`);
+    return;
+  }
+
+  const beforeHealth = await engine.getHealth();
+  const beforeDoctor = await buildDoctorReport(engine);
+  const actions: MaintenanceAction[] = [];
+
+  actions.push(await runStaleExtraction(engine, beforeHealth, opts.dryRun));
+  actions.push(...await runCycleFreshnessMaintenance(engine, beforeDoctor, opts.dryRun));
+
+  const afterHealth = await engine.getHealth();
+  const afterDoctor = await buildDoctorReport(engine);
+  const report: MaintainReport = {
+    mode: opts.dryRun ? 'dry-run' : 'safe',
+    before: { health: beforeHealth, doctor: beforeDoctor },
+    actions,
+    after: { health: afterHealth, doctor: afterDoctor },
+  };
+
+  if (opts.json) {
+    console.log(JSON.stringify(report, null, 2));
+  } else {
+    printMaintainReport(report);
+  }
+  return report;
+}
+
+function printMaintainReport(report: MaintainReport): void {
+  console.log(`GBrain maintain (${report.mode})`);
+  console.log(
+    `Before: brain_score=${Math.round(report.before.health.brain_score)}/100 ` +
+    `stale=${report.before.health.stale_pages} islands=${report.before.health.orphan_pages} ` +
+    `doctor=${report.before.doctor.status}`,
+  );
+  for (const action of report.actions) {
+    console.log(`  ${action.status}: ${action.name} — ${action.message}`);
+  }
+  console.log(
+    `After:  brain_score=${Math.round(report.after.health.brain_score)}/100 ` +
+    `stale=${report.after.health.stale_pages} islands=${report.after.health.orphan_pages} ` +
+    `doctor=${report.after.doctor.status}`,
+  );
+  if (report.mode === 'dry-run') {
+    console.log('Run `gbrain maintain --safe` to apply safe actions.');
+  }
+}
@@ -39,7 +39,6 @@ import type { BrainEngine } from '../../core/engine.ts';
 import { loadConfig, toEngineConfig } from '../../core/config.ts';
 import { createEngine } from '../../core/engine-factory.ts';
 import { upsertFactRow, parseFactsFence } from '../../core/facts-fence.ts';
-import { resolvePageFilePath } from '../../core/markdown.ts';

 let testEngineOverride: BrainEngine | null = null;
 export function __setTestEngineOverride(engine: BrainEngine | null): void {
@@ -149,16 +148,9 @@ function isLocalPathDirty(localPath: string): boolean {
  }
 }

-/**
- * Exported (not just via `__testing`) because `gbrain facts fence-backfill`
- * (#1867) re-runs this phase on demand: remote `extract_facts` deposits that
- * predate the fence-write backstop leave row_num-NULL rows the cycle guard
- * refuses to reconcile past. The phase is idempotent (only touches
- * `row_num IS NULL` rows), so re-running is always safe.
- */
-export async function phaseBFenceFacts(
+async function phaseBFenceFacts(
  engine: BrainEngine | null,
-  opts: Pick<OrchestratorOpts, 'dryRun'>,
+  opts: OrchestratorOpts,
 ): Promise<OrchestratorPhaseResult> {
  if (opts.dryRun) {
    // Dry-run: report what WOULD happen without touching FS or DB.
@@ -246,11 +238,7 @@ export async function phaseBFenceFacts(
    for (const [key, group] of groups) {
      const [sourceId, entitySlug] = key.split('\0');
      const localPath = localPathById.get(sourceId)!;
-      // resolvePageFilePath, NOT a bare join — non-default sources fence
-      // into `<local_path>/.sources/<id>/<slug>.md`, the same path the
-      // fence-write backstop and put_page write-through compute. A bare
-      // join here diverges fence and DB for non-default sources (#2044).
-      const filePath = resolvePageFilePath(localPath, entitySlug, sourceId);
+      const filePath = join(localPath, `${entitySlug}.md`);
      const tmpPath = `${filePath}.tmp`;

      try {
@@ -281,21 +269,6 @@ export async function phaseBFenceFacts(
        const existingFence = parseFactsFence(body);
        const existingKeySet = new Set(existingFence.facts.map(f => `${f.claim}\0${f.source ?? ''}`));

-        // Seed appended row_nums from MAX(fence max, DB max) — same #2044
-        // divergence guard as writeFactsToFence. When the fence at the
-        // resolved path is missing/behind but the DB already holds stamped
-        // rows for this slug (legacy wrong-path fence writes), fence-max+1
-        // collides with idx_facts_fence_key on the post-rename UPDATE,
-        // failing the page and leaving fence and DB disagreeing.
-        const dbMaxRows = await engine.executeRaw<{ max: number | string | null }>(
-          `SELECT MAX(row_num) AS max FROM facts
-            WHERE source_id = $1 AND source_markdown_slug = $2`,
-          [sourceId, entitySlug],
-        );
-        const dbMaxRowNum = Number(dbMaxRows[0]?.max ?? 0) || 0;
-        const fenceMaxRowNum = existingFence.facts.reduce((m, f) => Math.max(m, f.rowNum), 0);
-        let nextRowNum = Math.max(fenceMaxRowNum, dbMaxRowNum) + 1;
-
        const assignments: Array<{ id: string; row_num: number }> = [];
        for (const row of group) {
          const key = `${row.fact}\0${row.source ?? ''}`;
@@ -318,7 +291,6 @@ export async function phaseBFenceFacts(
                .toISOString().slice(0, 10)
            : undefined;
          const { body: updated, rowNum } = upsertFactRow(body, {
-            rowNum:     nextRowNum++,
            claim:      row.fact,
            kind:       row.kind,
            confidence: row.confidence,
@@ -409,7 +381,7 @@ async function phaseCVerify(
    for (const g of groups) {
      const localPath = localPathById.get(g.source_id);
      if (!localPath) continue;
-      const filePath = resolvePageFilePath(localPath, g.source_markdown_slug, g.source_id);
+      const filePath = join(localPath, `${g.source_markdown_slug}.md`);
      if (!existsSync(filePath)) {
        mismatches.push(`${g.source_markdown_slug} (file missing)`);
        continue;
@@ -15,6 +15,11 @@
 import type { BrainEngine } from '../core/engine.ts';
 import { createProgress, startHeartbeat } from '../core/progress.ts';
 import { getCliOptions, cliOptsToProgressOptions } from '../core/cli-options.ts';
+import {
+  shouldExcludeFromOrphanReporting,
+  loadOrphanPolicyOverrides,
+  type OrphanPolicyOverrides,
+} from '../core/orphan-policy.ts';

 // --- Types ---

@@ -32,65 +37,14 @@ export interface OrphanResult {
  excluded: number;
 }

-// --- Filter constants ---
-
-/** Slug suffixes that are always auto-generated root files */
-const AUTO_SUFFIX_PATTERNS = ['/_index', '/log'];
-
-/** Page slugs that are pseudo-pages by convention */
-const PSEUDO_SLUGS = new Set(['_atlas', '_index', '_stats', '_orphans', '_scratch', 'claude']);
-
-/** Slug segment that marks raw sources */
-const RAW_SEGMENT = '/raw/';
-
-/** Slug prefixes where no inbound links is expected */
-const DENY_PREFIXES = [
-  'output/',
-  'dashboards/',
-  'scripts/',
-  'templates/',
-  'openclaw/config/',
-];
-
-/** First slug segments where no inbound links is expected */
-const FIRST_SEGMENT_EXCLUSIONS = new Set([
-  'scratch',
-  'thoughts',
-  'catalog',
-  'entities',
-  'raw',
-  'atoms',
-  'skills',
-]);
-
 // --- Filter logic ---

 /**
 * Returns true if a slug should be excluded from orphan reporting by default.
 * These are pages where having no inbound links is expected / not a content problem.
 */
-export function shouldExclude(slug: string): boolean {
-  // Pseudo-pages (exact match)
-  if (PSEUDO_SLUGS.has(slug)) return true;
-
-  // Auto-generated suffix patterns
-  for (const suffix of AUTO_SUFFIX_PATTERNS) {
-    if (slug.endsWith(suffix)) return true;
-  }
-
-  // Raw source slugs
-  if (slug.includes(RAW_SEGMENT)) return true;
-
-  // Deny-prefix slugs
-  for (const prefix of DENY_PREFIXES) {
-    if (slug.startsWith(prefix)) return true;
-  }
-
-  // First-segment exclusions
-  const firstSegment = slug.split('/')[0];
-  if (FIRST_SEGMENT_EXCLUSIONS.has(firstSegment)) return true;
-
-  return false;
+export function shouldExclude(slug: string, overrides?: OrphanPolicyOverrides): boolean {
+  return shouldExcludeFromOrphanReporting(slug, overrides);
 }

 /**
@@ -156,6 +110,7 @@ export async function findOrphans(
  let allOrphans: { slug: string; title: string; domain: string | null }[];
  let total: number;
  let excludedAll: number;
+  const overrides = includePseudo ? undefined : await loadOrphanPolicyOverrides(engine);
  try {
    allOrphans = await engine.findOrphanPages(
      sourceIds ? { sourceIds } : sourceId ? { sourceId } : undefined,
@@ -184,7 +139,7 @@ export async function findOrphans(
    total = liveRows.length;
    excludedAll = includePseudo
      ? 0
-      : liveRows.reduce((n, r) => n + (shouldExclude(r.slug) ? 1 : 0), 0);
+      : liveRows.reduce((n, r) => n + (shouldExclude(r.slug, overrides) ? 1 : 0), 0);
  } finally {
    stopHb();
    progress.finish();
@@ -192,7 +147,7 @@ export async function findOrphans(

  const filtered = includePseudo
    ? allOrphans
-    : allOrphans.filter(row => !shouldExclude(row.slug));
+    : allOrphans.filter(row => !shouldExclude(row.slug, overrides));

  const orphans: OrphanPage[] = filtered.map(row => ({
    slug: row.slug,
@@ -54,6 +54,7 @@ import {
 import {
  generatePerChunkSynopsis,
  SYNOPSIS_PROMPT_VERSION,
+  SYNOPSIS_DOC_MAX_CHARS,
  type GeneratePerChunkSynopsisResult,
 } from './page-summary.ts';
 import {
@@ -103,8 +104,17 @@ function getEmbeddingModelTag(): string {
 export function computeCorpusGeneration(args: {
  crMode: CRMode;
  haikuModel: string;
+  /**
+   * Resolved `SYNOPSIS_DOC_MAX_CHARS` for per_chunk_synopsis runs. When
+   * present, folded into the hash so changes to
+   * `GBRAIN_SYNOPSIS_DOC_MAX_CHARS` invalidate the prior cache cleanly.
+   * Omit for `crMode !== 'per_chunk_synopsis'` — title / none modes
+   * don't consult the cap and the field stays out of the hash for
+   * back-compat with pre-cap embeddings.
+   */
+  synopsisDocMaxChars?: number;
 }): string {
-  return createHash('sha256')
+  const h = createHash('sha256')
    .update(args.crMode)
    .update('|')
    .update(String(SYNOPSIS_PROMPT_VERSION))
@@ -113,9 +123,11 @@ export function computeCorpusGeneration(args: {
    .update('|')
    .update(String(TITLE_WRAPPER_VERSION))
    .update('|')
-    .update(getEmbeddingModelTag())
-    .digest('hex')
-    .slice(0, 16);
+    .update(getEmbeddingModelTag());
+  if (args.synopsisDocMaxChars !== undefined) {
+    h.update('|doc_cap=').update(String(args.synopsisDocMaxChars));
+  }
+  return h.digest('hex').slice(0, 16);
 }

 /**
@@ -253,7 +265,11 @@ export async function reembedPageWithContextualRetrieval(
      args.pageSlug,
      args.sourceId,
      resolution.mode,
-      computeCorpusGeneration({ crMode: resolution.mode, haikuModel: args.haikuModel ?? DEFAULT_HAIKU_MODEL }),
+      computeCorpusGeneration({
+        crMode: resolution.mode,
+        haikuModel: args.haikuModel ?? DEFAULT_HAIKU_MODEL,
+        synopsisDocMaxChars: resolution.mode === 'per_chunk_synopsis' ? SYNOPSIS_DOC_MAX_CHARS : undefined,
+      }),
    );
    return { kind: 'skipped', reason: 'no_chunks' };
  }
@@ -282,6 +298,7 @@ export async function reembedPageWithContextualRetrieval(
      const corpus_generation = computeCorpusGeneration({
        crMode: attemptMode,
        haikuModel,
+        synopsisDocMaxChars: attemptMode === 'per_chunk_synopsis' ? SYNOPSIS_DOC_MAX_CHARS : undefined,
      });

      // ── PHASE 2: single DB transaction ───────────────────────────
@@ -176,11 +176,9 @@ export async function runExtractFacts(
  if (legacyCount > 0) {
    result.guardTriggered = true;
    result.warnings.push(
-      `extract_facts: ${legacyCount} legacy fact rows pending fence backfill ` +
-      `(row_num IS NULL — v0.31 rows or remote extract_facts deposits that ` +
-      `predate the fence backstop). Run \`gbrain facts fence-backfill\` ` +
-      `(idempotent, re-runnable) before this phase can safely reconcile ` +
-      `fence → DB.`,
+      `extract_facts: ${legacyCount} legacy v0.31 fact rows pending fence backfill. ` +
+      `Run \`gbrain apply-migrations --yes\` to complete v0_32_2 before this phase ` +
+      `can safely reconcile fence → DB.`,
    );
    return result;
  }
@@ -1739,14 +1739,13 @@ export interface BrainEngine {
   * single-row supersede flow because fence reconciliation is the canonical
   * source-of-truth direction, not the consolidator path.
   *
-   * Insertion runs in a single transaction. A collision on the v51
-   * partial UNIQUE index `(source_id, source_markdown_slug, row_num)`
-   * skips ONLY that row (ON CONFLICT DO NOTHING, #2044) — the rest of
-   * the batch still commits, so a redundant deposit against an
-   * already-indexed fence row is idempotent instead of a hard failure.
+   * Insertion is atomic per call: all rows commit in a single transaction
+   * or none commit (the transaction rolls back on any constraint
+   * violation, e.g. the v51 partial UNIQUE index on
+   * `(source_id, source_markdown_slug, row_num)`).
   *
-   * Returns the inserted ids in input-order (colliding rows omitted) so
-   * callers can correlate fence-row → DB-id without a separate lookup.
+   * Returns the inserted ids in input-order so callers can correlate
+   * fence-row → DB-id without a separate lookup.
   */
  insertFacts(
    rows: Array<NewFact & { row_num: number; source_markdown_slug: string }>,
@@ -218,27 +218,11 @@ export async function writeFactsToFence(
      }

      // 2. Upsert each fact onto the fence in input order. row_num
-      //    monotonically increases, append-only, seeded from the MAX of
-      //    the fence and the DB index (#2044): when fence and DB have
-      //    diverged (e.g. legacy writes that stamped DB rows against a
-      //    fence at a path this code no longer reads), fence-max+1 can
-      //    collide with an existing DB row_num, tripping
-      //    idx_facts_fence_key and rolling back the whole batch.
-      const dbMaxRows = await engine.executeRaw<{ max: number | string | null }>(
-        `SELECT MAX(row_num) AS max FROM facts
-          WHERE source_id = $1 AND source_markdown_slug = $2`,
-        [target.sourceId, target.slug],
-      );
-      const dbMaxRowNum = Number(dbMaxRows[0]?.max ?? 0) || 0;
-      const fenceMaxRowNum = parseFactsFence(body).facts
-        .reduce((m, f) => Math.max(m, f.rowNum), 0);
-      let nextRowNum = Math.max(fenceMaxRowNum, dbMaxRowNum) + 1;
-
+      //    monotonically increases (max-existing + 1 per call, append-only).
      const assignedRowNums: number[] = [];
      for (const f of facts) {
        const validFromStr = (f.validFrom ?? new Date()).toISOString().slice(0, 10);
        const { body: updated, rowNum } = upsertFactRow(body, {
-          rowNum:      nextRowNum++,
          claim:       f.fact,
          kind:        (f.kind ?? 'fact') as 'fact' | 'event' | 'preference' | 'commitment' | 'belief',
          confidence:  f.confidence ?? 1.0,
@@ -733,6 +733,11 @@ export async function importFromContent(
      : computeCorpusGeneration({
          crMode: effectiveCRMode,
          haikuModel: 'anthropic:claude-haiku-4-5-20251001',
+          // Inline import-file path never uses per_chunk_synopsis (refuses
+          // upstream); pass undefined so the doc-cap field stays out of
+          // the hash here. Per_chunk_synopsis runs through the Minion
+          // backfill handler which threads SYNOPSIS_DOC_MAX_CHARS through
+          // the service layer.
        });

  // Transaction wraps all DB writes. Every per-page tx call carries the
@@ -489,7 +489,22 @@ export async function extractPageLinks(
      // text inside `[[...]]` before any `|`), NOT the display alias
      // (ref.name = match[2]). `[[struktura|the project]]` must resolve
      // `struktura`, not "the project". The display text is for context only.
-      const matches = await resolver.resolveBasenameMatches(ref.slug);
+      //
+      // The literal may be path-qualified (`[[notes/struktura]]`). The FS
+      // path (resolveSlugAll) strips the dirname before its basename lookup,
+      // but this path passed the raw literal to an index keyed by final
+      // segments only — so every slash-containing wikilink outside
+      // DIR_PATTERN silently resolved to nothing. Query by the final
+      // segment, then use the written path as a disambiguation filter
+      // (the analogue of the FS ancestor walk honoring the written path):
+      // a match must end with the literal, so `[[notes/struktura]]` can
+      // resolve to `vault/notes/struktura` but never to `wiki/struktura`.
+      const slashIdx = ref.slug.lastIndexOf('/');
+      const basename = slashIdx === -1 ? ref.slug : ref.slug.slice(slashIdx + 1);
+      let matches = await resolver.resolveBasenameMatches(basename);
+      if (slashIdx !== -1) {
+        matches = matches.filter(m => m === ref.slug || m.endsWith(`/${ref.slug}`));
+      }
      if (matches.length === 0) continue;
      const idx = content.indexOf(ref.slug);
      const context = idx >= 0 ? excerpt(content, idx, 240) : ref.name;
@@ -0,0 +1,116 @@
+/**
+ * Shared orphan-reporting exclusion policy.
+ *
+ * These are pages where "no inbound links" is expected and should not count
+ * against health. Keep this in core so the CLI orphan report and engine health
+ * dashboard cannot drift.
+ *
+ * Defaults are GBrain-wide conventions only. Brain-specific exclusions
+ * (private folder names, one-off fixture slugs) belong in the brain's own
+ * config, not here:
+ *
+ *   gbrain config set orphans.exclude_prefixes "my-private-folder/,archive/"
+ *   gbrain config set orphans.exclude_slugs "some-one-off-page"
+ */
+
+const AUTO_SUFFIX_PATTERNS = ['/_index', '/log'];
+
+const PSEUDO_SLUGS = new Set(['_atlas', '_index', '_stats', '_orphans', '_scratch', 'claude']);
+
+const RAW_SEGMENT = '/raw/';
+
+const DENY_PREFIXES = [
+  'output/',
+  'dashboards/',
+  'scripts/',
+  'templates/',
+  '_templates/',
+  'openclaw/config/',
+  'extracts/',
+];
+
+const FIRST_SEGMENT_EXCLUSIONS = new Set([
+  'scratch',
+  'thoughts',
+  'catalog',
+  'entities',
+  'raw',
+  'atoms',
+  'skills',
+  'dreaming',
+  'daily',
+]);
+
+const ROOT_DATE_SLUG = /^\d{4}-\d{2}-\d{2}(?:-.+)?$/;
+
+function isAgentWorkspaceConvention(slug: string): boolean {
+  if (!slug.startsWith('agents/')) return false;
+  if (slug.includes('/memory/dreaming/')) return true;
+  return /^agents\/[^/]+\/(?:agents|identity|soul|tools|user|heartbeat|dreams|dormant)$/.test(slug);
+}
+
+/** Per-brain additions to the convention defaults (from config). */
+export interface OrphanPolicyOverrides {
+  excludePrefixes?: string[];
+  excludeSlugs?: string[];
+}
+
+/** Config keys for per-brain orphan exclusions (comma-separated values). */
+export const ORPHAN_EXCLUDE_PREFIXES_KEY = 'orphans.exclude_prefixes';
+export const ORPHAN_EXCLUDE_SLUGS_KEY = 'orphans.exclude_slugs';
+
+function parseList(value: string | null): string[] {
+  if (!value) return [];
+  return value.split(',').map(s => s.trim()).filter(Boolean);
+}
+
+/**
+ * Load per-brain orphan exclusions from the brain config table. Callers with
+ * an engine in hand (getHealth, `gbrain orphans`) pass the result as the
+ * second argument to shouldExcludeFromOrphanReporting.
+ */
+export async function loadOrphanPolicyOverrides(
+  engine: { getConfig(key: string): Promise<string | null> },
+): Promise<OrphanPolicyOverrides> {
+  const [prefixes, slugs] = await Promise.all([
+    engine.getConfig(ORPHAN_EXCLUDE_PREFIXES_KEY),
+    engine.getConfig(ORPHAN_EXCLUDE_SLUGS_KEY),
+  ]);
+  return { excludePrefixes: parseList(prefixes), excludeSlugs: parseList(slugs) };
+}
+
+export function shouldExcludeFromOrphanReporting(
+  slug: string,
+  overrides?: OrphanPolicyOverrides,
+): boolean {
+  if (PSEUDO_SLUGS.has(slug)) return true;
+
+  for (const suffix of AUTO_SUFFIX_PATTERNS) {
+    if (slug.endsWith(suffix)) return true;
+  }
+
+  if (slug.includes(RAW_SEGMENT)) return true;
+  if (slug.includes('/daily/')) return true;
+
+  for (const prefix of DENY_PREFIXES) {
+    if (slug.startsWith(prefix)) return true;
+  }
+
+  const firstSegment = slug.split('/')[0];
+  if (FIRST_SEGMENT_EXCLUSIONS.has(firstSegment)) return true;
+
+  if (ROOT_DATE_SLUG.test(slug)) return true;
+
+  if (slug.startsWith('_brain-')) return true;
+
+  if (isAgentWorkspaceConvention(slug)) return true;
+
+  if (overrides) {
+    if (overrides.excludeSlugs?.includes(slug)) return true;
+    for (const prefix of overrides.excludePrefixes ?? []) {
+      if (slug.startsWith(prefix)) return true;
+    }
+  }
+
+  return false;
+}
@@ -44,6 +44,33 @@ const HAIKU_MAX_TOKENS = 200;
 /** Default model when caller doesn't override. Resolves through the gateway. */
 const DEFAULT_SYNOPSIS_MODEL = 'anthropic:claude-haiku-4-5-20251001';

+/**
+ * Hard cap on `documentText` length (chars) before send.
+ *
+ * 2026-05-25 fix wave: small local chat models (Gemma 4 E2B, Qwen3 4B) get
+ * dramatically slower on long contexts even with 131K-token windows declared.
+ * A 73K-char page synopsis on Gemma 4 E2B takes 60-120s, exceeding the
+ * worker's default 30s `lockDuration` and tripping `lock-lost` errors.
+ *
+ * Truncate to a budget that fits a small model's effective throughput while
+ * preserving enough document context for the synopsis to be useful. Truncates
+ * the TAIL because the head (title, frontmatter, intro) carries the
+ * document-level anchor the synopsis needs.
+ *
+ * Override per workload via `GBRAIN_SYNOPSIS_DOC_MAX_CHARS`. Default 32768
+ * (~8K tokens at 4 chars/tok) keeps small-model synopsis under ~30s.
+ * Anthropic Haiku is unaffected at this cap; bump higher when running
+ * frontier models if you want richer document anchoring.
+ */
+export const SYNOPSIS_DOC_MAX_CHARS = (() => {
+  const env = process.env.GBRAIN_SYNOPSIS_DOC_MAX_CHARS;
+  if (env && /^\d+$/.test(env)) {
+    const n = parseInt(env, 10);
+    if (n >= 512 && n <= 1_048_576) return n;
+  }
+  return 32768;
+})();
+
 /**
 * Synopsis prompt version. Folded into corpus_generation so prompt edits
 * invalidate prior embeddings via the v0.40.3.0 query_cache.page_generations
@@ -188,11 +215,19 @@ function buildUserPrompt(
  documentText: string,
  chunkText: string,
 ): string {
+  // Tail-truncate `documentText` to `SYNOPSIS_DOC_MAX_CHARS` so small local
+  // chat models don't stall on >100KB pages. Head preserved (title block,
+  // frontmatter, intro paragraphs carry the document-level anchor).
+  let trimmedDoc = documentText;
+  if (documentText.length > SYNOPSIS_DOC_MAX_CHARS) {
+    trimmedDoc = documentText.slice(0, SYNOPSIS_DOC_MAX_CHARS) +
+      `\n\n[... ${documentText.length - SYNOPSIS_DOC_MAX_CHARS} chars truncated for synopsis budget ...]`;
+  }
  return [
    `<page_title>${pageTitle}</page_title>`,
    '',
    '<full_document>',
-    documentText,
+    trimmedDoc,
    '</full_document>',
    '',
    '<chunk>',
@@ -57,6 +57,8 @@ import { finalizeLastSeen } from './chronicle/last-seen.ts';
 import { computeAnomaliesFromBuckets } from './cycle/anomaly.ts';
 import { resolveBoostMap, resolveHardExcludes } from './search/source-boost.ts';
 import { buildSourceFactorCase, buildHardExcludeClause, buildVisibilityClause, buildRecencyComponentSql, buildBestPerPagePoolCte, buildOrFallbackWebsearchQuery } from './search/sql-ranking.ts';
+import { shouldExcludeFromOrphanReporting, loadOrphanPolicyOverrides } from './orphan-policy.ts';
+import { LINK_EXTRACTOR_VERSION_TS } from './link-extraction.ts';
 import {
  normalizeEngineColumn,
  buildVectorCastFragment,
@@ -2322,6 +2324,10 @@ export class PGLiteEngine implements BrainEngine {
    // v0.40.3.0 D24 NULL→non-NULL race fix mirrors postgres-engine.ts. Two writers
    // racing on the same chunk previously raced last-write-wins; the fix lets the
    // fresher `embedded_at` win in the text-unchanged branch.
+    //
+    // Code-chunk metadata columns follow the same chunk_text-gated CASE pattern as `embedding`
+    // (#769). Re-chunk trusts EXCLUDED outright; pure re-embed COALESCEs so a caller carrying
+    // only embedding-shaped fields doesn't clobber metadata to NULL.
    await this.db.query(
      `INSERT INTO content_chunks ${cols} VALUES ${rowParts.join(', ')}
       ON CONFLICT (page_id, chunk_index) DO UPDATE SET
@@ -2345,14 +2351,14 @@ export class PGLiteEngine implements BrainEngine {
                THEN EXCLUDED.embedded_at
           ELSE content_chunks.embedded_at
         END,
-         language = EXCLUDED.language,
-         symbol_name = EXCLUDED.symbol_name,
-         symbol_type = EXCLUDED.symbol_type,
-         start_line = EXCLUDED.start_line,
-         end_line = EXCLUDED.end_line,
-         parent_symbol_path = EXCLUDED.parent_symbol_path,
-         doc_comment = EXCLUDED.doc_comment,
-         symbol_name_qualified = EXCLUDED.symbol_name_qualified,
+         language = CASE WHEN EXCLUDED.chunk_text != content_chunks.chunk_text THEN EXCLUDED.language ELSE COALESCE(EXCLUDED.language, content_chunks.language) END,
+         symbol_name = CASE WHEN EXCLUDED.chunk_text != content_chunks.chunk_text THEN EXCLUDED.symbol_name ELSE COALESCE(EXCLUDED.symbol_name, content_chunks.symbol_name) END,
+         symbol_type = CASE WHEN EXCLUDED.chunk_text != content_chunks.chunk_text THEN EXCLUDED.symbol_type ELSE COALESCE(EXCLUDED.symbol_type, content_chunks.symbol_type) END,
+         start_line = CASE WHEN EXCLUDED.chunk_text != content_chunks.chunk_text THEN EXCLUDED.start_line ELSE COALESCE(EXCLUDED.start_line, content_chunks.start_line) END,
+         end_line = CASE WHEN EXCLUDED.chunk_text != content_chunks.chunk_text THEN EXCLUDED.end_line ELSE COALESCE(EXCLUDED.end_line, content_chunks.end_line) END,
+         parent_symbol_path = CASE WHEN EXCLUDED.chunk_text != content_chunks.chunk_text THEN EXCLUDED.parent_symbol_path ELSE COALESCE(EXCLUDED.parent_symbol_path, content_chunks.parent_symbol_path) END,
+         doc_comment = CASE WHEN EXCLUDED.chunk_text != content_chunks.chunk_text THEN EXCLUDED.doc_comment ELSE COALESCE(EXCLUDED.doc_comment, content_chunks.doc_comment) END,
+         symbol_name_qualified = CASE WHEN EXCLUDED.chunk_text != content_chunks.chunk_text THEN EXCLUDED.symbol_name_qualified ELSE COALESCE(EXCLUDED.symbol_name_qualified, content_chunks.symbol_name_qualified) END,
         modality = EXCLUDED.modality,
         embedding_image = COALESCE(EXCLUDED.embedding_image, content_chunks.embedding_image)`,
      params
@@ -4102,13 +4108,11 @@ export class PGLiteEngine implements BrainEngine {
  ): Promise<{ inserted: number; ids: number[] }> {
    if (rows.length === 0) return { inserted: 0, ids: [] };

-    // Single transaction; per-row INSERTs (not multi-row VALUES) keep the
-    // embedding-vs-no-embedding branching readable; batch sizes are small
-    // (5-30 rows per page in practice) so the loop overhead is negligible
-    // vs the embedding compute cost. #2044: ON CONFLICT DO NOTHING on the
-    // v51 partial UNIQUE index makes a residual fence/DB row_num collision
-    // skip that row instead of rolling back the whole batch (parity with
-    // postgres-engine.ts).
+    // Single transaction so the v51 partial UNIQUE index can roll back the
+    // whole batch on constraint violation. Per-row INSERTs (not multi-row
+    // VALUES) keep the embedding-vs-no-embedding branching readable; batch
+    // sizes are small (5-30 rows per page in practice) so the loop overhead
+    // is negligible vs the embedding compute cost.
    const ids = await this.db.transaction(async (tx) => {
      const out: number[] = [];
      for (const input of rows) {
@@ -4151,11 +4155,7 @@ export class PGLiteEngine implements BrainEngine {
                 $14, $15,
                 $16, $17, $18, $19,
                 $20
-               )
-               ON CONFLICT (source_id, source_markdown_slug, row_num)
-                 WHERE row_num IS NOT NULL
-                 DO NOTHING
-               RETURNING id`
+               ) RETURNING id`
            : `INSERT INTO facts (
                 source_id, entity_slug, fact, kind, visibility, notability, context,
                 valid_from, valid_until, source, source_session, confidence,
@@ -4169,16 +4169,12 @@ export class PGLiteEngine implements BrainEngine {
                 $15, $16,
                 $17, $18, $19, $20,
                 $21
-               )
-               ON CONFLICT (source_id, source_markdown_slug, row_num)
-                 WHERE row_num IS NOT NULL
-                 DO NOTHING
-               RETURNING id`,
+               ) RETURNING id`,
          embedStr === null
            ? [ctx.source_id, entitySlug, input.fact, kind, visibility, notability, context, validFrom, validUntil, input.source, sourceSession, confidence, embeddedAt, input.row_num, input.source_markdown_slug, claimMetric, claimValue, claimUnit, claimPeriod, eventType]
            : [ctx.source_id, entitySlug, input.fact, kind, visibility, notability, context, validFrom, validUntil, input.source, sourceSession, confidence, embedStr, embeddedAt, input.row_num, input.source_markdown_slug, claimMetric, claimValue, claimUnit, claimPeriod, eventType],
        );
-        if (ins.rows[0]) out.push(ins.rows[0].id);
+        out.push(ins.rows[0].id);
      }
      return out;
    });
@@ -5212,26 +5208,31 @@ export class PGLiteEngine implements BrainEngine {
    const { rows: [h] } = await this.db.query(`
      WITH entity_pages AS (
        SELECT id, slug FROM pages WHERE type IN ('person', 'company')
+      ),
+      narrative_pages AS (
+        -- Composition-aware score: code source files and calendar daily
+        -- files are orphans-by-design (no inbound wikilinks, no Timeline
+        -- fence). Excluding them from the orphan/link-density/timeline
+        -- denominators keeps a bulk code/calendar import from cratering
+        -- brain_score.
+        SELECT id FROM pages WHERE type IS NULL OR type NOT IN ('code', 'calendar-index')
      )
      SELECT
        (SELECT count(*) FROM pages) as page_count,
+        (SELECT count(*) FROM narrative_pages) as narrative_page_count,
        (SELECT count(*) FROM content_chunks WHERE embedded_at IS NOT NULL)::float /
          GREATEST((SELECT count(*) FROM content_chunks), 1)::float as embed_coverage,
-        (SELECT count(*) FROM pages p
-         WHERE p.updated_at < (SELECT MAX(te.created_at) FROM timeline_entries te WHERE te.page_id = p.id)
-        ) as stale_pages,
-        -- Bug 11 — orphan = islanded (no inbound AND no outbound).
-        -- See BrainHealth.orphan_pages docstring; docs updated to match this.
-        (SELECT count(*) FROM pages p
-         WHERE NOT EXISTS (SELECT 1 FROM links l WHERE l.to_page_id = p.id)
-           AND NOT EXISTS (SELECT 1 FROM links l WHERE l.from_page_id = p.id)
-        ) as orphan_pages,
+        0 as stale_pages,
+        -- Bug 11 — orphan = islanded (no inbound AND no outbound). The raw
+        -- list is filtered in TS using the shared orphan-reporting policy.
+        0 as orphan_pages,
        (SELECT count(*) FROM links l
         WHERE NOT EXISTS (SELECT 1 FROM pages p WHERE p.id = l.to_page_id)
        ) as dead_links,
        (SELECT count(*) FROM content_chunks WHERE embedded_at IS NULL) as missing_embeddings,
        (SELECT count(*) FROM links) as link_count,
-        (SELECT count(DISTINCT page_id) FROM timeline_entries) as pages_with_timeline,
+        (SELECT count(DISTINCT te.page_id) FROM timeline_entries te
+         WHERE te.page_id IN (SELECT id FROM narrative_pages)) as pages_with_timeline,
        (SELECT count(*) FROM entity_pages e
         WHERE EXISTS (SELECT 1 FROM links l WHERE l.to_page_id = e.id))::float /
          GREATEST((SELECT count(*) FROM entity_pages), 1)::float as link_coverage,
@@ -5250,17 +5251,34 @@ export class PGLiteEngine implements BrainEngine {
      LIMIT 5
    `);

+    const { rows: islandedRows } = await this.db.query(`
+      SELECT p.slug
+      FROM pages p
+      -- Narrative pages only (same type filter as the narrative_pages CTE):
+      -- code/calendar-index pages are orphans-by-design and must not count
+      -- against the noOrphans component (#1144).
+      WHERE (p.type IS NULL OR p.type NOT IN ('code', 'calendar-index'))
+        AND NOT EXISTS (SELECT 1 FROM links l WHERE l.to_page_id = p.id)
+        AND NOT EXISTS (SELECT 1 FROM links l WHERE l.from_page_id = p.id)
+    `);
+
    const r = h as Record<string, unknown>;
    const pageCount = Number(r.page_count);
+    // Composition-aware denominators (excludes code/calendar-index pages);
+    // a code-only brain has nothing narrative to penalize → full marks.
+    const narrativePageCount = Number(r.narrative_page_count);
    const embedCoverage = Number(r.embed_coverage);
-    const orphanPages = Number(r.orphan_pages);
+    const stalePages = await this.countStalePagesForExtraction({ versionTs: LINK_EXTRACTOR_VERSION_TS });
+    const orphanOverrides = await loadOrphanPolicyOverrides(this);
+    const orphanPages = (islandedRows as { slug: string }[])
+      .filter(row => !shouldExcludeFromOrphanReporting(row.slug, orphanOverrides)).length;
    const deadLinks = Number(r.dead_links);
    const linkCount = Number(r.link_count);
    const pagesWithTimeline = Number(r.pages_with_timeline);

-    const linkDensity = pageCount > 0 ? Math.min(linkCount / pageCount, 1) : 0;
-    const timelineCoverageDensity = pageCount > 0 ? Math.min(pagesWithTimeline / pageCount, 1) : 0;
-    const noOrphans = pageCount > 0 ? 1 - (orphanPages / pageCount) : 1;
+    const linkDensity = narrativePageCount > 0 ? Math.min(linkCount / narrativePageCount, 1) : 1;
+    const timelineCoverageDensity = narrativePageCount > 0 ? Math.min(pagesWithTimeline / narrativePageCount, 1) : 1;
+    const noOrphans = narrativePageCount > 0 ? 1 - (orphanPages / narrativePageCount) : 1;
    const noDeadLinks = pageCount > 0 ? 1 - Math.min(deadLinks / pageCount, 1) : 1;
    // Bug 11 — per-component points. Sum equals brainScore by construction
    // so `doctor` can render a breakdown that adds up to the total.
@@ -5281,7 +5299,7 @@ export class PGLiteEngine implements BrainEngine {
    return {
      page_count: pageCount,
      embed_coverage: embedCoverage,
-      stale_pages: Number(r.stale_pages),
+      stale_pages: stalePages,
      orphan_pages: orphanPages,
      missing_embeddings: Number(r.missing_embeddings),
      brain_score: brainScore,
@@ -67,6 +67,8 @@ import { resolveBoostMap, resolveHardExcludes } from './search/source-boost.ts';
 import { buildSourceFactorCase, buildHardExcludeClause, buildVisibilityClause, buildRecencyComponentSql, buildBestPerPagePoolCte, buildOrFallbackWebsearchQuery } from './search/sql-ranking.ts';
 import { DEFAULT_EMBEDDING_MODEL, DEFAULT_EMBEDDING_DIMENSIONS } from './ai/defaults.ts';
 import { DELETE_BATCH_SIZE } from './engine-constants.ts';
+import { shouldExcludeFromOrphanReporting, loadOrphanPolicyOverrides } from './orphan-policy.ts';
+import { LINK_EXTRACTOR_VERSION_TS } from './link-extraction.ts';

 function escapeSqlStringLiteral(value: string): string {
  return value.replace(/'/g, "''");
@@ -2473,6 +2475,13 @@ export class PostgresEngine implements BrainEngine {
    //   - new is fresher (embedded_at > existing.embedded_at) → take new
    //   - otherwise → keep existing (slower writer with stale embedding loses)
    // Mirrored in pglite-engine.ts; pinned by test/e2e/concurrent-embed-race.test.ts.
+    //
+    // Code-chunk metadata columns (language / symbol_name / symbol_type / line range /
+    // parent_symbol_path / doc_comment / symbol_name_qualified) follow the SAME chunk_text-gated
+    // CASE pattern as `embedding` (#769). Re-chunk (chunk_text changed) trusts EXCLUDED outright;
+    // pure re-embed (chunk_text unchanged) COALESCEs so a caller that only carries embedding
+    // doesn't clobber metadata to NULL. Without this, every embed --stale pass nuked code-def's
+    // primary index for thousands of chunks at once.
    await sql.unsafe(
      `INSERT INTO content_chunks ${cols} VALUES ${rows.join(', ')}
       ON CONFLICT (page_id, chunk_index) DO UPDATE SET
@@ -2496,14 +2505,14 @@ export class PostgresEngine implements BrainEngine {
                THEN EXCLUDED.embedded_at
           ELSE content_chunks.embedded_at
         END,
-         language = EXCLUDED.language,
-         symbol_name = EXCLUDED.symbol_name,
-         symbol_type = EXCLUDED.symbol_type,
-         start_line = EXCLUDED.start_line,
-         end_line = EXCLUDED.end_line,
-         parent_symbol_path = EXCLUDED.parent_symbol_path,
-         doc_comment = EXCLUDED.doc_comment,
-         symbol_name_qualified = EXCLUDED.symbol_name_qualified,
+         language = CASE WHEN EXCLUDED.chunk_text != content_chunks.chunk_text THEN EXCLUDED.language ELSE COALESCE(EXCLUDED.language, content_chunks.language) END,
+         symbol_name = CASE WHEN EXCLUDED.chunk_text != content_chunks.chunk_text THEN EXCLUDED.symbol_name ELSE COALESCE(EXCLUDED.symbol_name, content_chunks.symbol_name) END,
+         symbol_type = CASE WHEN EXCLUDED.chunk_text != content_chunks.chunk_text THEN EXCLUDED.symbol_type ELSE COALESCE(EXCLUDED.symbol_type, content_chunks.symbol_type) END,
+         start_line = CASE WHEN EXCLUDED.chunk_text != content_chunks.chunk_text THEN EXCLUDED.start_line ELSE COALESCE(EXCLUDED.start_line, content_chunks.start_line) END,
+         end_line = CASE WHEN EXCLUDED.chunk_text != content_chunks.chunk_text THEN EXCLUDED.end_line ELSE COALESCE(EXCLUDED.end_line, content_chunks.end_line) END,
+         parent_symbol_path = CASE WHEN EXCLUDED.chunk_text != content_chunks.chunk_text THEN EXCLUDED.parent_symbol_path ELSE COALESCE(EXCLUDED.parent_symbol_path, content_chunks.parent_symbol_path) END,
+         doc_comment = CASE WHEN EXCLUDED.chunk_text != content_chunks.chunk_text THEN EXCLUDED.doc_comment ELSE COALESCE(EXCLUDED.doc_comment, content_chunks.doc_comment) END,
+         symbol_name_qualified = CASE WHEN EXCLUDED.chunk_text != content_chunks.chunk_text THEN EXCLUDED.symbol_name_qualified ELSE COALESCE(EXCLUDED.symbol_name_qualified, content_chunks.symbol_name_qualified) END,
         modality = EXCLUDED.modality,
         embedding_image = COALESCE(EXCLUDED.embedding_image, content_chunks.embedding_image)`,
      params as Parameters<typeof sql.unsafe>[1],
@@ -4302,12 +4311,10 @@ export class PostgresEngine implements BrainEngine {
    // ONCE per process so the cast matches the actual column type
    // (halfvec vs vector). The probe is cached after first call.
    const castSuffix = await this.resolveFactsEmbeddingCast();
-    // Single transaction; per-row INSERTs (not multi-row VALUES) keep the
-    // embedding-vs-no-embedding branching readable; batch sizes are small
-    // (5-30 rows per page in practice). #2044: ON CONFLICT DO NOTHING on
-    // the v51 partial UNIQUE index makes a residual fence/DB row_num
-    // collision skip that row instead of rolling back the whole batch —
-    // the fence stays system-of-record and reconciliation catches up.
+    // Single transaction so the v51 partial UNIQUE index can roll back
+    // the whole batch on constraint violation. Per-row INSERTs (not
+    // multi-row VALUES) keep the embedding-vs-no-embedding branching
+    // readable; batch sizes are small (5-30 rows per page in practice).
    // No supersede flow in this path — fence reconciliation is the
    // canonical source-of-truth direction, not the consolidator path.
    const ids = await sql.begin(async (tx) => {
@@ -4348,13 +4355,9 @@ export class PostgresEngine implements BrainEngine {
            ${input.row_num}, ${input.source_markdown_slug},
            ${claimMetric}, ${claimValue}, ${claimUnit}, ${claimPeriod},
            ${eventType}
-          )
-          ON CONFLICT (source_id, source_markdown_slug, row_num)
-            WHERE row_num IS NOT NULL
-            DO NOTHING
-          RETURNING id
+          ) RETURNING id
        `;
-        if (ins[0]) out.push(Number(ins[0].id));
+        out.push(Number(ins[0].id));
      }
      return out;
    });
@@ -5319,32 +5322,35 @@ export class PostgresEngine implements BrainEngine {
  async getHealth(): Promise<BrainHealth> {
    const sql = this.sql;
    // Bug 11 doc-drift fix — orphan_pages means "islanded" (no inbound AND
-    // no outbound links), aligning both engines with the user-facing
-    // definition. The type comment previously said "no inbound" but the
-    // SQL required both — docs now match code so users can trust the
-    // number. A hub page that links out to many but has no back-references
-    // is working as intended, not an orphan.
+    // no outbound links). The raw islanded list is filtered through the same
+    // policy as `gbrain orphans` so convention pages do not count against
+    // dashboard health.
    const [h] = await sql`
      WITH entity_pages AS (
        SELECT id, slug FROM pages WHERE type IN ('person', 'company')
+      ),
+      narrative_pages AS (
+        -- Composition-aware score: code source files and calendar daily
+        -- files are orphans-by-design (no inbound wikilinks, no Timeline
+        -- fence). Excluding them from the orphan/link-density/timeline
+        -- denominators keeps a bulk code/calendar import from cratering
+        -- brain_score.
+        SELECT id FROM pages WHERE type IS NULL OR type NOT IN ('code', 'calendar-index')
      )
      SELECT
        (SELECT count(*) FROM pages) as page_count,
+        (SELECT count(*) FROM narrative_pages) as narrative_page_count,
        (SELECT count(*) FROM content_chunks WHERE embedded_at IS NOT NULL)::float /
          GREATEST((SELECT count(*) FROM content_chunks), 1)::float as embed_coverage,
-        (SELECT count(*) FROM pages p
-         WHERE p.updated_at < (SELECT MAX(te.created_at) FROM timeline_entries te WHERE te.page_id = p.id)
-        ) as stale_pages,
-        (SELECT count(*) FROM pages p
-         WHERE NOT EXISTS (SELECT 1 FROM links l WHERE l.to_page_id = p.id)
-           AND NOT EXISTS (SELECT 1 FROM links l WHERE l.from_page_id = p.id)
-        ) as orphan_pages,
+        0 as stale_pages,
+        0 as orphan_pages,
        (SELECT count(*) FROM links l
         WHERE NOT EXISTS (SELECT 1 FROM pages p WHERE p.id = l.to_page_id)
        ) as dead_links,
        (SELECT count(*) FROM content_chunks WHERE embedded_at IS NULL) as missing_embeddings,
        (SELECT count(*) FROM links) as link_count,
-        (SELECT count(DISTINCT page_id) FROM timeline_entries) as pages_with_timeline,
+        (SELECT count(DISTINCT te.page_id) FROM timeline_entries te
+         WHERE te.page_id IN (SELECT id FROM narrative_pages)) as pages_with_timeline,
        (SELECT count(*) FROM entity_pages e
         WHERE EXISTS (SELECT 1 FROM links l WHERE l.to_page_id = e.id))::float /
          GREATEST((SELECT count(*) FROM entity_pages), 1)::float as link_coverage,
@@ -5362,17 +5368,33 @@ export class PostgresEngine implements BrainEngine {
      LIMIT 5
    `;

+    const islandedRows = await sql<{ slug: string }[]>`
+      SELECT p.slug
+      FROM pages p
+      -- Narrative pages only (same type filter as the narrative_pages CTE):
+      -- code/calendar-index pages are orphans-by-design and must not count
+      -- against the noOrphans component (#1144).
+      WHERE (p.type IS NULL OR p.type NOT IN ('code', 'calendar-index'))
+        AND NOT EXISTS (SELECT 1 FROM links l WHERE l.to_page_id = p.id)
+        AND NOT EXISTS (SELECT 1 FROM links l WHERE l.from_page_id = p.id)
+    `;
+
    const pageCount = Number(h.page_count);
+    // Composition-aware denominators (excludes code/calendar-index pages);
+    // a code-only brain has nothing narrative to penalize → full marks.
+    const narrativePageCount = Number(h.narrative_page_count);
    const embedCoverage = Number(h.embed_coverage);
-    const orphanPages = Number(h.orphan_pages);
+    const stalePages = await this.countStalePagesForExtraction({ versionTs: LINK_EXTRACTOR_VERSION_TS });
+    const orphanOverrides = await loadOrphanPolicyOverrides(this);
+    const orphanPages = islandedRows.filter(row => !shouldExcludeFromOrphanReporting(row.slug, orphanOverrides)).length;
    const deadLinks = Number(h.dead_links);
    const linkCount = Number(h.link_count);
    const pagesWithTimeline = Number(h.pages_with_timeline);

    // brain_score: 0-100 weighted average
-    const linkDensity = pageCount > 0 ? Math.min(linkCount / pageCount, 1) : 0;
-    const timelineCoverageWhole = pageCount > 0 ? Math.min(pagesWithTimeline / pageCount, 1) : 0;
-    const noOrphans = pageCount > 0 ? 1 - (orphanPages / pageCount) : 1;
+    const linkDensity = narrativePageCount > 0 ? Math.min(linkCount / narrativePageCount, 1) : 1;
+    const timelineCoverageWhole = narrativePageCount > 0 ? Math.min(pagesWithTimeline / narrativePageCount, 1) : 1;
+    const noOrphans = narrativePageCount > 0 ? 1 - (orphanPages / narrativePageCount) : 1;
    const noDeadLinks = pageCount > 0 ? 1 - Math.min(deadLinks / pageCount, 1) : 1;
    // Per-component points. Sum equals brainScore by construction.
    //
@@ -5392,7 +5414,7 @@ export class PostgresEngine implements BrainEngine {
    return {
      page_count: pageCount,
      embed_coverage: embedCoverage,
-      stale_pages: Number(h.stale_pages),
+      stale_pages: stalePages,
      orphan_pages: orphanPages,
      missing_embeddings: Number(h.missing_embeddings),
      brain_score: brainScore,
@@ -93,7 +93,13 @@ import { resolveLrSchedule } from './lr-schedule.ts';
 import { preflight, formatPreflightReport } from './preflight.ts';
 import { isRejected, loadRejectedBuffer, makeRejectedEntry, saveRejectedBuffer } from './rejected-buffer.ts';
 import { runReflect, runOneShotRewrite, describeJudges } from './reflect.ts';
-import { acceptCandidate, bestPath, revertAllPending, skillPath, writeProposed } from './version-store.ts';
+import {
+  acceptCandidate,
+  proposedPath as proposedFilePath,
+  revertAllPending,
+  skillPath,
+  writeProposed,
+} from './version-store.ts';
 import { runValidationGate, scoreSkillOnTasks } from './validate-gate.ts';
 import { ROLLOUT_SUCCESS_THRESHOLD } from './types.ts';
 import type { SkillOptOpts, EditOp, RunReceipt, BenchmarkTask } from './types.ts';
@@ -702,9 +708,9 @@ async function runOptimizationLoop(
  // to the catch's assignment values only (it can't prove the async callback ran).
  const finalOutcome = outcome as 'accepted' | 'no_improvement' | 'aborted' | 'errored';
  if (!mutateDecision.mutate && finalOutcome === 'accepted') {
-    // best.md was written by writeProposed() in the accept branch (no-mutate
-    // path); it doubles as proposed.md for human review. SKILL.md untouched.
-    proposedPath = bestPath(skillsDir, skillName);
+    // writeProposed() emitted both the best pointer and the stable review
+    // artifact in the accept branch. SKILL.md remains untouched.
+    proposedPath = proposedFilePath(skillsDir, skillName);
  } else if (mutateDecision.mutate) {
    mutatedSkillFile = finalOutcome === 'accepted';
  }
@@ -23,6 +23,7 @@
 *
 *   history.json
 *   best.md
+ *   proposed.md
 *   versions/
 *     v0001_e1_s1.md
 *     v0002_e1_s2.md
@@ -52,6 +53,10 @@ export function bestPath(skillsDir: string, skillName: string): string {
  return path.join(skilloptDir(skillsDir, skillName), 'best.md');
 }

+export function proposedPath(skillsDir: string, skillName: string): string {
+  return path.join(skilloptDir(skillsDir, skillName), 'proposed.md');
+}
+
 export function skillPath(skillsDir: string, skillName: string): string {
  return path.join(skillsDir, skillName, 'SKILL.md');
 }
@@ -171,17 +176,18 @@ export function acceptCandidate(input: AcceptInput): AcceptResult {
 }

 /**
- * Write the candidate to `best.md` (which doubles as `proposed.md`) WITHOUT
- * touching SKILL.md or the history ledger. Used by the `--no-mutate` /
- * bundled-without-allow paths: the optimizer found a better candidate but the
- * caller opted out of in-place mutation, so we surface it for human review.
- * Returns the path written. Atomic (.tmp + rename).
+ * Write the candidate to both `best.md` and `proposed.md` WITHOUT touching
+ * SKILL.md or the history ledger. `best.md` remains the optimizer's current
+ * best pointer; `proposed.md` is the stable human-review artifact promised by
+ * `--no-mutate`. Returns the proposal path. Each write is atomic (.tmp + rename).
 */
 export function writeProposed(skillsDir: string, skillName: string, candidateText: string): string {
-  const p = bestPath(skillsDir, skillName);
-  fs.mkdirSync(path.dirname(p), { recursive: true });
-  atomicWrite(p, candidateText);
-  return p;
+  const best = bestPath(skillsDir, skillName);
+  const proposed = proposedPath(skillsDir, skillName);
+  fs.mkdirSync(path.dirname(best), { recursive: true });
+  atomicWrite(best, candidateText);
+  atomicWrite(proposed, candidateText);
+  return proposed;
 }

 /**
@@ -63,25 +63,26 @@ interface PluginCtx {
  [key: string]: unknown;
 }

+export function register(api: PluginApi) {
+  api.registerContextEngine(ENGINE_ID, (ctx: PluginCtx) => {
+    const hostResolver =
+      typeof ctx.resolveEntities === 'function'
+        ? ctx.resolveEntities
+        : typeof ctx.brainQuery === 'function'
+          ? ctx.brainQuery
+          : undefined;
+    return createGBrainContextEngine({
+      workspaceDir: ctx.workspaceDir,
+      resolveEntities: hostResolver,
+    });
+  });
+}
+
 const entry: PluginEntry = {
  id: 'gbrain-context-engine',
  name: 'GBrain Context Engine',
  description: 'Deterministic temporal/spatial context injection on every turn',
-
-  register(api: PluginApi) {
-    api.registerContextEngine(ENGINE_ID, (ctx: PluginCtx) => {
-      const hostResolver =
-        typeof ctx.resolveEntities === 'function'
-          ? ctx.resolveEntities
-          : typeof ctx.brainQuery === 'function'
-            ? ctx.brainQuery
-            : undefined;
-      return createGBrainContextEngine({
-        workspaceDir: ctx.workspaceDir,
-        resolveEntities: hostResolver,
-      });
-    });
-  },
+  register,
 };

 export default entry;
@@ -119,6 +119,40 @@ describe('Bug 11 — orphan_pages is "no inbound links"', () => {
  });
 });

+describe('#1144 — brain_score is composition-aware', () => {
+  test('code and calendar-index pages do not count as orphans or dilute density metrics', async () => {
+    // Two linked narrative pages + a flood of orphan-by-design pages.
+    await engine.putPage('people/alice', { type: 'person', title: 'Alice', compiled_truth: 'a', frontmatter: {} });
+    await engine.putPage('people/bob', { type: 'person', title: 'Bob', compiled_truth: 'b', frontmatter: {} });
+    const aId = (await (engine as any).db.query(`SELECT id FROM pages WHERE slug='people/alice'`)).rows[0].id;
+    const bId = (await (engine as any).db.query(`SELECT id FROM pages WHERE slug='people/bob'`)).rows[0].id;
+    await (engine as any).db.query(
+      `INSERT INTO links (from_page_id, to_page_id, link_type) VALUES ($1, $2, 'mentions')`,
+      [aId, bId],
+    );
+    for (let i = 0; i < 10; i++) {
+      await engine.putPage(`code/src/f${i}.py`, { type: 'code', title: `f${i}.py`, compiled_truth: 'def f(): pass', frontmatter: {} });
+    }
+    await engine.putPage('daily/calendar/2026-01-01', { type: 'calendar-index', title: '2026-01-01', compiled_truth: 'events', frontmatter: {} });
+
+    const h = await engine.getHealth();
+    // 11 unlinked code/calendar pages exist, but both narrative pages are
+    // linked → 0 orphans, full no-orphans marks.
+    expect(h.orphan_pages).toBe(0);
+    expect(h.no_orphans_score).toBe(15);
+    // Link density: 1 link / 2 narrative pages, not 1 / 13 total pages.
+    expect(h.link_density_score).toBe(Math.round(0.5 * 25));
+  });
+
+  test('a brain with ONLY non-narrative pages gets full composition marks', async () => {
+    await engine.putPage('code/src/only.py', { type: 'code', title: 'only.py', compiled_truth: 'x = 1', frontmatter: {} });
+    const h = await engine.getHealth();
+    expect(h.no_orphans_score).toBe(15);
+    expect(h.link_density_score).toBe(25);
+    expect(h.timeline_coverage_score).toBe(15);
+  });
+});
+
 describe('Bug 11 — doctor renders brain_score breakdown', () => {
  test('doctor source contains brain_score breakdown rendering', async () => {
    const source = await Bun.file(new URL('../src/commands/doctor.ts', import.meta.url)).text();
@@ -763,3 +763,48 @@ describeBoth('Engine parity — federated sourceIds[] secondary reads (#2200)',
    }
  });
 });
+
+// #1144 — brain_score is composition-aware: code / calendar-index pages are
+// orphans-by-design and must not feed the orphan/link-density/timeline
+// denominators. Both engines must agree.
+async function seedComposition(eng: BrainEngine) {
+  await eng.putPage('people/comp-alice', { type: 'person', title: 'Alice', compiled_truth: 'a', timeline: '' });
+  await eng.putPage('people/comp-bob', { type: 'person', title: 'Bob', compiled_truth: 'b', timeline: '' });
+  await eng.addLink('people/comp-alice', 'people/comp-bob', 'knows', 'mentions', 'markdown');
+  for (let i = 0; i < 5; i++) {
+    await eng.putPage(`code/src/comp-f${i}.py`, { type: 'code', title: `f${i}.py`, compiled_truth: 'def f(): pass', timeline: '' });
+  }
+  await eng.putPage('daily/calendar/2026-01-01', { type: 'calendar-index', title: '2026-01-01', compiled_truth: 'events', timeline: '' });
+}
+
+describeBoth('Engine parity — getHealth composition-aware brain_score (#1144)', () => {
+  let pgEngine: BrainEngine;
+  let pgliteEngine: PGLiteEngine;
+
+  beforeAll(async () => {
+    pgEngine = await setupDB();
+    await seedComposition(pgEngine);
+    pgliteEngine = new PGLiteEngine();
+    await pgliteEngine.connect({});
+    await pgliteEngine.initSchema();
+    await seedComposition(pgliteEngine);
+  }, 90_000);
+
+  afterAll(async () => {
+    await pgliteEngine.disconnect();
+    await teardownDB();
+  }, 30_000);
+
+  test('orphan/link-density/timeline components identical and exclude code + calendar-index pages', async () => {
+    const pg = await pgEngine.getHealth();
+    const pglite = await pgliteEngine.getHealth();
+    for (const k of ['orphan_pages', 'no_orphans_score', 'link_density_score', 'timeline_coverage_score', 'brain_score'] as const) {
+      expect(pg[k]).toBe(pglite[k]);
+    }
+    // 6 non-narrative pages exist but both narrative pages are linked → 0 orphans.
+    expect(pg.orphan_pages).toBe(0);
+    expect(pg.no_orphans_score).toBe(15);
+    // 1 link / 2 narrative pages, not / 8 total pages.
+    expect(pg.link_density_score).toBe(Math.round(0.5 * 25));
+  });
+});
@@ -172,6 +172,59 @@ describe('issue #972 — DB-source (gbrain extract links --source db)', () => {
    expect(strk!.link_type).toBe('wikilink_basename');
  });

+  test('flag ON → path-qualified wikilink outside DIR_PATTERN resolves via DB path', async () => {
+    // `[[notes/struktura]]` — `notes` is not in DIR_PATTERN, so the ref
+    // reaches the generic pass with its dirname intact. Regression: the DB
+    // path queried the basename index with the raw literal (which is keyed
+    // by final segments only), so path-qualified wikilinks outside
+    // DIR_PATTERN silently produced zero edges while the FS path resolved
+    // the identical content.
+    await engine.putPage('notes/struktura', {
+      type: 'concept' as any, title: 'Struktura Notes',
+      compiled_truth: '', timeline: '',
+    });
+    await engine.putPage('concepts/knowledge-graph', {
+      type: 'concept', title: 'Knowledge Graph',
+      compiled_truth: 'Background in [[notes/struktura]].', timeline: '',
+    });
+    await engine.setConfig('link_resolution.global_basename', 'true');
+
+    await runExtract(engine, ['links', '--source', 'db']);
+
+    const outLinks = await engine.getLinks('concepts/knowledge-graph');
+    const strk = outLinks.find(l => l.to_slug === 'notes/struktura');
+    expect(strk).toBeDefined();
+    expect(strk!.link_type).toBe('wikilink_basename');
+    expect(strk!.link_source).toBe('wikilink-resolved');
+  });
+
+  test('path-qualified wikilink never attaches to a basename-only sibling', async () => {
+    // Both notes/struktura and wiki/struktura exist. The author wrote
+    // `[[notes/struktura]]` — the written path must exclude wiki/struktura
+    // (a bare `[[struktura]]` would legitimately match both).
+    await engine.putPage('notes/struktura', {
+      type: 'concept' as any, title: 'Struktura Notes',
+      compiled_truth: '', timeline: '',
+    });
+    await engine.putPage('wiki/struktura', {
+      type: 'concept' as any, title: 'Struktura Wiki',
+      compiled_truth: '', timeline: '',
+    });
+    await engine.putPage('concepts/x', {
+      type: 'concept', title: 'X',
+      compiled_truth: 'See [[notes/struktura]].', timeline: '',
+    });
+    await engine.setConfig('link_resolution.global_basename', 'true');
+
+    await runExtract(engine, ['links', '--source', 'db']);
+
+    const outLinks = await engine.getLinks('concepts/x');
+    const basenameLinks = outLinks
+      .filter(l => l.link_type === 'wikilink_basename')
+      .map(l => l.to_slug);
+    expect(basenameLinks).toEqual(['notes/struktura']);
+  });
+
  test('flag OFF → no basename edges via DB path (back-compat)', async () => {
    await engine.putPage('projects/struktura', {
      type: 'project', title: 'Struktura',
@@ -140,6 +140,35 @@ describeE2E('scanIntegrity batch parity (E2E, Postgres-only)', () => {
    });
  });

+  describe('code pages', () => {
+    test('type:code page is skipped on both paths (#1143)', async () => {
+      const engine = getEngine();
+
+      await engine.putPage('people/alice', {
+        type: 'person',
+        title: 'Alice',
+        compiled_truth: 'Alice tweeted about something.',
+        timeline: '',
+        frontmatter: {},
+      });
+      await engine.putPage('code/src/bot.py', {
+        type: 'code',
+        title: 'src/bot.py (python)',
+        compiled_truth: '# the user tweeted about this feature\ndef send_tweet():\n    pass',
+        timeline: '',
+        frontmatter: { language: 'python', file: 'src/bot.py' },
+      });
+
+      const batchResult = await scanIntegrity(engine, { limit: 100, batchLoad: true });
+      const seqResult = await scanIntegrity(engine, { limit: 100, batchLoad: false });
+
+      expect(batchResult.pagesScanned).toBe(seqResult.pagesScanned);
+      expect(batchResult.pagesScanned).toBe(1);
+      expect(batchResult.bareHits.map(h => h.slug)).not.toContain('code/src/bot.py');
+      expect(seqResult.bareHits.map(h => h.slug)).not.toContain('code/src/bot.py');
+    });
+  });
+
  describe('topPages', () => {
    test('topPages ordering matches between paths', async () => {
      const engine = getEngine();
@@ -39,6 +39,7 @@ import { runSkillOpt } from '../../src/core/skillopt/orchestrator.ts';
 import {
  bestPath,
  loadHistory,
+  proposedPath,
  skillPath,
 } from '../../src/core/skillopt/version-store.ts';
 import { loadRejectedBuffer } from '../../src/core/skillopt/rejected-buffer.ts';
@@ -741,7 +742,7 @@ describe('skillopt T3 — F11 held-out gate, ablation opts, no-DB-pollution', ()
    } finally { fixture.cleanup(); }
  });

-  test('--no-mutate writes proposed.md (best.md), leaves SKILL.md untouched', async () => {
+  test('--no-mutate writes proposed.md and best.md, leaves SKILL.md untouched', async () => {
    const fixture = setupFixture(SKILL_PEOPLE_ONLY, CITATIONS_BENCHMARK);
    try {
      installStub({
@@ -753,10 +754,9 @@ describe('skillopt T3 — F11 held-out gate, ablation opts, no-DB-pollution', ()
          const result = await runOnce(fixture, { noMutate: true });
          expect(result.outcome).toBe('accepted');
          expect(result.mutatedSkillFile).toBe(false);
-          expect(result.proposedPath).toBeDefined();
-          // proposed.md (best.md) exists and carries the improvement.
-          expect(fs.existsSync(result.proposedPath!)).toBe(true);
+          expect(result.proposedPath).toBe(proposedPath(fixture.skillsDir, SKILL));
          expect(fs.readFileSync(result.proposedPath!, 'utf8')).toContain('## Citations');
+          expect(fs.readFileSync(bestPath(fixture.skillsDir, SKILL), 'utf8')).toContain('## Citations');
          // SKILL.md on disk is UNCHANGED (still People-only).
          const skill = fs.readFileSync(skillPath(fixture.skillsDir, SKILL), 'utf8');
          expect(skill).not.toContain('## Citations');
@@ -803,3 +803,107 @@ describe('embedAllStale --source threading (D7)', () => {
    expect((firstCallOpts as { sourceId?: string }).sourceId).toBe('media-corpus');
  });
 });
+
+// ────────────────────────────────────────────────────────────────
+// Code metadata preservation across re-embed (regression for #769)
+// ────────────────────────────────────────────────────────────────
+//
+// gbrain v0.30.1 and earlier silently clobbered code-chunk metadata
+// (language, symbol_name, symbol_type, start_line, end_line,
+// parent_symbol_path, doc_comment, symbol_name_qualified) on every
+// re-embed pass. The chunker populated those columns at import time,
+// but embed.ts loaded chunks via getChunks then mapped them to a
+// stripped ChunkInput carrying only 5 fields. upsertChunks then
+// OVERWROTE (not COALESCEd) the metadata columns from EXCLUDED, so
+// re-embed wiped them to NULL. End result on a real brain: 4875 code
+// pages, 47866 chunks, all with NULL language/symbol_name/symbol_type;
+// code-def returned 0 hits across every indexed repo.
+//
+// All three runEmbed paths (--stale autopilot, --all, --slugs) must
+// thread metadata through the re-upsert. Tests below assert that the
+// engine.upsertChunks call carries the same metadata it loaded.
+
+describe('runEmbed preserves code-chunk metadata across re-embed (regression for #769)', () => {
+  const fullCodeChunk = {
+    chunk_index: 0,
+    chunk_text: '[Java] foo/Bar.java:10-20 method baz',
+    chunk_source: 'compiled_truth' as const,
+    embedded_at: null,
+    token_count: 12,
+    language: 'java',
+    symbol_name: 'baz',
+    symbol_type: 'function',
+    start_line: 10,
+    end_line: 20,
+    parent_symbol_path: ['Bar'],
+    doc_comment: 'does the thing',
+    symbol_name_qualified: 'Bar.baz',
+  };
+
+  function metadataOf(chunk: any) {
+    return {
+      language: chunk.language,
+      symbol_name: chunk.symbol_name,
+      symbol_type: chunk.symbol_type,
+      start_line: chunk.start_line,
+      end_line: chunk.end_line,
+      parent_symbol_path: chunk.parent_symbol_path,
+      doc_comment: chunk.doc_comment,
+      symbol_name_qualified: chunk.symbol_name_qualified,
+    };
+  }
+
+  test('--stale (autopilot path) carries code metadata into upsertChunks', async () => {
+    const stale = [{
+      slug: 'code-page',
+      chunk_index: 0,
+      chunk_text: fullCodeChunk.chunk_text,
+      chunk_source: 'compiled_truth',
+      model: null,
+      token_count: 12,
+    }];
+    let upsertChunkArgs: any[] | null = null;
+    const engine = mockEngine({
+      countStaleChunks: async () => 1,
+      listStaleChunks: async () => stale,
+      getChunks: async () => [fullCodeChunk],
+      upsertChunks: async (_slug: string, chunks: any[]) => { upsertChunkArgs = chunks; },
+    });
+
+    await runEmbed(engine, ['--stale']);
+
+    expect(upsertChunkArgs).not.toBeNull();
+    expect(upsertChunkArgs!).toHaveLength(1);
+    expect(metadataOf(upsertChunkArgs![0])).toEqual(metadataOf(fullCodeChunk));
+  });
+
+  test('--all (full re-embed) carries code metadata into upsertChunks', async () => {
+    let upsertChunkArgs: any[] | null = null;
+    const engine = mockEngine({
+      listPages: async () => [{ slug: 'code-page' }],
+      getChunks: async () => [fullCodeChunk],
+      upsertChunks: async (_slug: string, chunks: any[]) => { upsertChunkArgs = chunks; },
+    });
+
+    await runEmbed(engine, ['--all']);
+
+    expect(upsertChunkArgs).not.toBeNull();
+    expect(upsertChunkArgs!).toHaveLength(1);
+    expect(metadataOf(upsertChunkArgs![0])).toEqual(metadataOf(fullCodeChunk));
+  });
+
+  test('--slugs (per-page embed) carries code metadata into upsertChunks', async () => {
+    let upsertChunkArgs: any[] | null = null;
+    const engine = mockEngine({
+      getPage: async () => ({ slug: 'code-page', compiled_truth: 'x', timeline: '' }),
+      getChunks: async () => [fullCodeChunk],
+      upsertChunks: async (_slug: string, chunks: any[]) => { upsertChunkArgs = chunks; },
+    });
+
+    await runEmbed(engine, ['--slugs', 'code-page']);
+
+    expect(upsertChunkArgs).not.toBeNull();
+    expect(upsertChunkArgs!).toHaveLength(1);
+    expect(metadataOf(upsertChunkArgs![0])).toEqual(metadataOf(fullCodeChunk));
+  });
+});
@@ -304,9 +304,7 @@ describe('runExtractFacts — empty-fence guard (Codex R2-#7)', () => {
    expect(r.legacyRowsPending).toBe(1);
    expect(r.factsInserted).toBe(0);
    expect(r.factsDeleted).toBe(0);
-    // #1867: the remedy hint points at the re-runnable backfill command,
-    // not the one-shot v0_32_2 migration (which the ledger never re-runs).
-    expect(r.warnings.some(w => w.includes('gbrain facts fence-backfill'))).toBe(true);
+    expect(r.warnings.some(w => w.includes('apply-migrations'))).toBe(true);

    // Legacy row was NOT touched.
    // eslint-disable-next-line @typescript-eslint/no-explicit-any
@@ -1,133 +0,0 @@
-/**
- * #1867 — `gbrain facts fence-backfill` command tests.
- *
- * The command re-runs the (idempotent) v0_32_2 phase B on demand so
- * row_num-NULL backlogs — remote extract_facts deposits that predate
- * the fence-write backstop — can be cleared without re-running the
- * one-shot migration. Real PGLite + real tempdir filesystem.
- */
-
-import { describe, test, expect, beforeAll, afterAll, beforeEach } from 'bun:test';
-import { mkdtempSync, rmSync, existsSync, readFileSync } from 'node:fs';
-import { tmpdir } from 'node:os';
-import { join } from 'node:path';
-
-import { PGLiteEngine } from '../src/core/pglite-engine.ts';
-import { runFactsCommand } from '../src/commands/facts.ts';
-import { phaseBFenceFacts } from '../src/commands/migrations/v0_32_2.ts';
-
-let engine: PGLiteEngine;
-let brainDir: string;
-
-beforeAll(async () => {
-  engine = new PGLiteEngine();
-  await engine.connect({});
-  await engine.initSchema();
-});
-
-afterAll(async () => {
-  await engine.disconnect();
-  try {
-    if (brainDir) rmSync(brainDir, { recursive: true, force: true });
-  } catch { /* best-effort */ }
-});
-
-beforeEach(async () => {
-  brainDir = mkdtempSync(join(tmpdir(), 'facts-backfill-cmd-test-'));
-  // eslint-disable-next-line @typescript-eslint/no-explicit-any
-  await (engine as any).db.query('DELETE FROM facts');
-  // eslint-disable-next-line @typescript-eslint/no-explicit-any
-  await (engine as any).db.query(
-    `UPDATE sources SET local_path = $1 WHERE id = 'default'`,
-    [brainDir],
-  );
-});
-
-async function seedLegacyFact(fact: string): Promise<void> {
-  // The row_num-NULL shape a remote extract_facts deposit leaves behind
-  // when it lands via the legacy DB-only path.
-  // eslint-disable-next-line @typescript-eslint/no-explicit-any
-  await (engine as any).db.query(
-    `INSERT INTO facts (source_id, entity_slug, fact, kind, visibility, notability,
-                        valid_from, source, confidence)
-     VALUES ('default', 'people/alice', $1, 'fact', 'private', 'medium',
-             now(), 'mcp:extract_facts', 1.0)`,
-    [fact],
-  );
-}
-
-describe('gbrain facts fence-backfill', () => {
-  test('fences row_num-NULL rows and stamps the DB', async () => {
-    await seedLegacyFact('Deposited remotely');
-
-    await runFactsCommand(engine, ['fence-backfill']);
-
-    // The fence exists on disk with the claim.
-    const filePath = join(brainDir, 'people/alice.md');
-    expect(existsSync(filePath)).toBe(true);
-    expect(readFileSync(filePath, 'utf-8')).toContain('Deposited remotely');
-
-    // The backlog is cleared: no row_num-NULL rows remain.
-    // eslint-disable-next-line @typescript-eslint/no-explicit-any
-    const rows = await (engine as any).db.query(
-      'SELECT row_num, source_markdown_slug FROM facts',
-    );
-    expect(rows.rows).toHaveLength(1);
-    expect(rows.rows[0].row_num).toBe(1);
-    expect(rows.rows[0].source_markdown_slug).toBe('people/alice');
-  });
-
-  test('re-run is a no-op (idempotent)', async () => {
-    await seedLegacyFact('Deposited remotely');
-    await runFactsCommand(engine, ['fence-backfill']);
-    await runFactsCommand(engine, ['fence-backfill']);
-
-    // eslint-disable-next-line @typescript-eslint/no-explicit-any
-    const rows = await (engine as any).db.query('SELECT id FROM facts');
-    expect(rows.rows).toHaveLength(1);
-    const body = readFileSync(join(brainDir, 'people/alice.md'), 'utf-8');
-    expect(body.match(/Deposited remotely/g)).toHaveLength(1);
-  });
-
-  test('--dry-run reports without writing', async () => {
-    await seedLegacyFact('Deposited remotely');
-    await runFactsCommand(engine, ['fence-backfill', '--dry-run']);
-
-    expect(existsSync(join(brainDir, 'people/alice.md'))).toBe(false);
-    // eslint-disable-next-line @typescript-eslint/no-explicit-any
-    const rows = await (engine as any).db.query(
-      'SELECT row_num FROM facts',
-    );
-    expect(rows.rows[0].row_num).toBeNull();
-  });
-
-  test('diverged page: appends past the DB row_num max instead of colliding (#2044 class)', async () => {
-    // The #2044 divergence shape the backfill must survive: the DB already
-    // holds stamped rows 1..3 for the slug (legacy wrong-path fence write),
-    // but the fence at the resolved path is missing. Fence-max+1 (= 1) would
-    // collide with the stamped rows on the post-rename UPDATE, failing the
-    // page and leaving the renamed fence disagreeing with the DB.
-    for (let n = 1; n <= 3; n++) {
-      // eslint-disable-next-line @typescript-eslint/no-explicit-any
-      await (engine as any).db.query(
-        `INSERT INTO facts (source_id, entity_slug, fact, kind, visibility, notability,
-                            valid_from, source, confidence, row_num, source_markdown_slug)
-         VALUES ('default', 'people/alice', $1, 'fact', 'private', 'medium',
-                 now(), 'mcp:extract_facts', 1.0, $2, 'people/alice')`,
-        [`stamped ${n}`, n],
-      );
-    }
-    await seedLegacyFact('Deposited remotely');
-
-    const result = await phaseBFenceFacts(engine, { dryRun: false });
-    expect(result.status).toBe('complete');
-
-    // eslint-disable-next-line @typescript-eslint/no-explicit-any
-    const rows = await (engine as any).db.query(
-      `SELECT row_num FROM facts WHERE fact = 'Deposited remotely'`,
-    );
-    expect(rows.rows[0].row_num).toBe(4);
-    const body = readFileSync(join(brainDir, 'people/alice.md'), 'utf-8');
-    expect(body).toContain('| 4 | Deposited remotely |');
-  });
-});
@@ -292,49 +292,6 @@ describe('lookupSourceLocalPath', () => {
  });
 });

-describe('writeFactsToFence — fence/DB divergence (#2044)', () => {
-  test('seeds row_num past the DB max when the fence lags the DB', async () => {
-    // Simulate the divergence class from #2044: DB rows were stamped with
-    // row_nums against a fence written at a path this code no longer reads
-    // (e.g. the pre-"Local patch 2026-06-11" wrong-path writes). The page
-    // on disk has NO fence, but the DB already holds row_num 1..3 for the
-    // slug. Pre-fix, the next deposit re-assigned row_num=1 from fence
-    // text alone and the whole insertFacts batch failed on
-    // idx_facts_fence_key.
-    for (let n = 1; n <= 3; n++) {
-      // eslint-disable-next-line @typescript-eslint/no-explicit-any
-      await (engine as any).db.query(
-        `INSERT INTO facts (source_id, entity_slug, fact, kind, visibility, notability,
-                            valid_from, source, confidence, row_num, source_markdown_slug)
-         VALUES ('default', 'people/dana', $1, 'fact', 'private', 'medium',
-                 now(), 'mcp:extract_facts', 1.0, $2, 'people/dana')`,
-        [`old claim ${n}`, n],
-      );
-    }
-
-    const result = await writeFactsToFence(
-      engine,
-      { sourceId: 'default', localPath: brainDir, slug: 'people/dana' },
-      [baseInput({ fact: 'second deposit' })],
-    );
-
-    expect(result.fenceWriteFailed).toBeUndefined();
-    expect(result.inserted).toBe(1);
-
-    // The new row landed PAST the DB max, not at fence-max+1 (= 1).
-    // eslint-disable-next-line @typescript-eslint/no-explicit-any
-    const rows = await (engine as any).db.query(
-      'SELECT row_num FROM facts WHERE id = $1',
-      [result.ids[0]],
-    );
-    expect(rows.rows[0].row_num).toBe(4);
-
-    // And the on-disk fence carries the same row_num — fence and DB agree.
-    const body = readFileSync(join(brainDir, 'people/dana.md'), 'utf-8');
-    expect(body).toContain('| 4 | second deposit |');
-  });
-});
-
 // Cleanup any leftover tempdirs after the whole suite.
 afterAll(() => {
  // No-op: each test cleaned up via the beforeEach; this is a safety net.
@@ -5,7 +5,7 @@
 *   - Batch insert N rows persists row_num + source_markdown_slug
 *   - Empty batch is a no-op
 *   - Returns ids in input-order
- *   - v51 partial UNIQUE collision skips only the colliding row (#2044)
+ *   - v51 partial UNIQUE index rolls back the whole batch on a collision
 *   - deleteFactsForPage scopes by (source_id, source_markdown_slug);
 *     never touches other pages or pre-v51 NULL-source_markdown_slug rows
 *   - deleteFactsForPage on an empty page returns deleted:0 (idempotent)
@@ -135,29 +135,30 @@ describe('engine.insertFacts — batch insert', () => {
    });
  });

-  test('v51 partial UNIQUE collision skips ONLY the colliding row (#2044)', async () => {
+  test('v51 partial UNIQUE index rolls back the whole batch on collision', async () => {
    // Seed row #1 first.
    await engine.insertFacts([fixtureFact(1, { fact: 'seeded' })], { source_id: 'default' });

-    // Batch-insert rows that include a colliding row_num=1. Pre-#2044 this
-    // threw and rolled back the whole batch, making a second remote
-    // extract_facts deposit to an already-fenced page a hard failure. Now
-    // ON CONFLICT DO NOTHING skips the colliding row and keeps the rest.
-    const result = await engine.insertFacts(
-      [
-        fixtureFact(2, { fact: 'second' }),
-        fixtureFact(1, { fact: 'collides' }),  // row_num=1 on same (source_id, source_markdown_slug)
-        fixtureFact(3, { fact: 'third' }),
-      ],
-      { source_id: 'default' },
-    );
-    expect(result.inserted).toBe(2);
-    expect(result.ids).toHaveLength(2);
+    // Now try to batch-insert rows that include a colliding row_num=1.
+    let threw = false;
+    try {
+      await engine.insertFacts(
+        [
+          fixtureFact(2, { fact: 'second' }),
+          fixtureFact(1, { fact: 'collides' }),  // row_num=1 on same (source_id, source_markdown_slug)
+          fixtureFact(3, { fact: 'third' }),
+        ],
+        { source_id: 'default' },
+      );
+    } catch {
+      threw = true;
+    }
+    expect(threw).toBe(true);

-    // The seeded row survives untouched; the colliding claim is skipped.
+    // Verify the transaction rolled back — only the seeded row should remain.
    // eslint-disable-next-line @typescript-eslint/no-explicit-any
    const rows = await (engine as any).db.query('SELECT fact FROM facts ORDER BY id');
-    expect(rows.rows.map((r: { fact: string }) => r.fact)).toEqual(['seeded', 'second', 'third']);
+    expect(rows.rows.map((r: { fact: string }) => r.fact)).toEqual(['seeded']);
  });

  test('different source_markdown_slug values DO NOT collide on the same row_num', async () => {
@@ -201,6 +201,15 @@ describe('scanIntegrity', () => {
      timeline: '',
      frontmatter: { validate: false },
    });
+    // Indexed source file — 'tweeted about' in a comment must NOT be flagged
+    // as a bare-tweet citation gap (#1143: auto-repair would corrupt source).
+    await engine.putPage('code/src/bot.py', {
+      type: 'code',
+      title: 'src/bot.py (python)',
+      compiled_truth: '# the user tweeted about this feature\ndef send_tweet():\n    pass',
+      timeline: '',
+      frontmatter: { language: 'python', file: 'src/bot.py' },
+    });
  }, 60_000);

  afterAll(async () => {
@@ -223,6 +232,13 @@ describe('scanIntegrity', () => {
    expect(slugs).not.toContain('people/legacy');
  });

+  test('skips type:code pages (#1143 — bare-tweet false positives on source files)', async () => {
+    const res = await scanIntegrity(engine);
+    const slugs = res.bareHits.map(h => h.slug);
+    expect(slugs).not.toContain('code/src/bot.py');
+    expect(res.pagesScanned).toBe(2);
+  });
+
  test('honors limit', async () => {
    const res = await scanIntegrity(engine, { limit: 1 });
    expect(res.pagesScanned).toBe(1);
@@ -403,6 +403,77 @@ describe('extractPageLinks', () => {
    expect(candidates).toEqual([]);
  });

+  test('path-qualified wikilink outside DIR_PATTERN queries by final segment', async () => {
+    // `[[notes/struktura]]` (dir not in DIR_PATTERN) falls to the generic
+    // pass. The resolver's basename index is keyed by final path segments,
+    // so the lookup must strip the dirname — mirroring the FS path
+    // (resolveSlugAll). Regression: the raw literal was passed through,
+    // which never matched, so these links silently dropped.
+    const seen: string[] = [];
+    const resolver: SlugResolver = {
+      resolve: async () => null,
+      resolveBasenameMatches: async (name) => {
+        seen.push(name);
+        return name === 'struktura' ? ['notes/struktura'] : [];
+      },
+    };
+    const { candidates } = await extractPageLinks(
+      'concepts/x', 'See [[notes/struktura]].',
+      {}, 'concept', resolver, { globalBasename: true },
+    );
+    expect(seen).toContain('struktura');
+    expect(seen).not.toContain('notes/struktura');
+    expect(candidates.map(c => c.targetSlug)).toEqual(['notes/struktura']);
+    expect(candidates[0].linkType).toBe('wikilink_basename');
+    expect(candidates[0].linkSource).toBe('wikilink-resolved');
+  });
+
+  test('path-qualified wikilink keeps only matches ending with the written path', async () => {
+    // The written path disambiguates: `[[notes/struktura]]` must never
+    // attach to `wiki/struktura` even though both share the basename.
+    const resolver: SlugResolver = {
+      resolve: async () => null,
+      resolveBasenameMatches: async (name) =>
+        name === 'struktura' ? ['notes/struktura', 'wiki/struktura'] : [],
+    };
+    const { candidates } = await extractPageLinks(
+      'concepts/x', 'See [[notes/struktura]].',
+      {}, 'concept', resolver, { globalBasename: true },
+    );
+    expect(candidates.map(c => c.targetSlug)).toEqual(['notes/struktura']);
+  });
+
+  test('path-qualified wikilink matches a deeper real slug by path suffix', async () => {
+    // The page lives at vault/notes/struktura; the author wrote the shorter
+    // tail `[[notes/struktura]]`. Suffix matching connects them, while the
+    // basename-only sibling `wiki/struktura` stays excluded.
+    const resolver: SlugResolver = {
+      resolve: async () => null,
+      resolveBasenameMatches: async (name) =>
+        name === 'struktura' ? ['vault/notes/struktura', 'wiki/struktura'] : [],
+    };
+    const { candidates } = await extractPageLinks(
+      'concepts/x', 'See [[notes/struktura]].',
+      {}, 'concept', resolver, { globalBasename: true },
+    );
+    expect(candidates.map(c => c.targetSlug)).toEqual(['vault/notes/struktura']);
+  });
+
+  test('path-qualified self-link is dropped like the bare form', async () => {
+    // `[[notes/struktura]]` written on notes/struktura itself must not
+    // produce a self-loop (same guard as the bare `[[own-tail]]` case).
+    const resolver: SlugResolver = {
+      resolve: async () => null,
+      resolveBasenameMatches: async (name) =>
+        name === 'struktura' ? ['notes/struktura'] : [],
+    };
+    const { candidates } = await extractPageLinks(
+      'notes/struktura', 'See [[notes/struktura]].',
+      {}, 'concept', resolver, { globalBasename: true },
+    );
+    expect(candidates).toEqual([]);
+  });
+
  test('bare wikilink resolution does not interfere with DIR_PATTERN wikilinks', async () => {
    // 2b refs (people/alice) take the verb-inferred type;
    // 2c refs (struktura) take wikilink_basename. Same call.
@@ -0,0 +1,62 @@
+import { describe, expect, test } from 'bun:test';
+import {
+  extractCycleFreshnessSourceIds,
+  parseMaintainArgs,
+} from '../src/commands/maintain.ts';
+import type { Check } from '../src/commands/doctor.ts';
+
+describe('maintain args', () => {
+  test('defaults to dry-run unless --safe is explicit', () => {
+    expect(parseMaintainArgs([])).toMatchObject({
+      safe: false,
+      dryRun: true,
+      json: false,
+    });
+  });
+
+  test('--safe enables mutating safe mode', () => {
+    expect(parseMaintainArgs(['--safe', '--json'])).toMatchObject({
+      safe: true,
+      dryRun: false,
+      json: true,
+    });
+  });
+
+  test('--dry-run wins over --safe', () => {
+    expect(parseMaintainArgs(['--safe', '--dry-run'])).toMatchObject({
+      safe: true,
+      dryRun: true,
+    });
+  });
+});
+
+describe('cycle freshness source extraction', () => {
+  test('extracts stale source ids from doctor messages', () => {
+    const checks: Check[] = [
+      {
+        name: 'cycle_freshness',
+        status: 'fail',
+        message: "Source 'brain-sync-remote-teffur' last cycled 40h ago. Run `gbrain dream --source <id>`.",
+      },
+      {
+        name: 'cycle_freshness',
+        status: 'fail',
+        message: "Source 'wiki' last cycled 25h ago. Source 'wiki' last cycled 25h ago.",
+      },
+    ];
+
+    expect(extractCycleFreshnessSourceIds(checks)).toEqual([
+      'brain-sync-remote-teffur',
+      'wiki',
+    ]);
+  });
+
+  test('ignores ok and unrelated checks', () => {
+    const checks: Check[] = [
+      { name: 'cycle_freshness', status: 'ok', message: "Source 'fresh' last cycled recently." },
+      { name: 'frontmatter_integrity', status: 'warn', message: "Source 'wiki' has frontmatter issues." },
+    ];
+
+    expect(extractCycleFreshnessSourceIds(checks)).toEqual([]);
+  });
+});
@@ -0,0 +1,17 @@
+import { describe, expect, it } from 'bun:test';
+import { readFileSync } from 'fs';
+import { join } from 'path';
+
+describe('root OpenClaw plugin manifest', () => {
+  it('declares the id required by OpenClaw plugin installs', () => {
+    const manifest = JSON.parse(readFileSync(join(import.meta.dir, '..', 'openclaw.plugin.json'), 'utf8'));
+    const entrySource = readFileSync(join(import.meta.dir, '..', 'src', 'openclaw-context-engine.ts'), 'utf8');
+    const entryId = entrySource.match(/id:\s*'([^']+)'/)?.[1];
+
+    expect(manifest.id).toBe(entryId);
+    expect(manifest.configSchema).toBeDefined();
+    expect(typeof manifest.configSchema).toBe('object');
+    expect(manifest.contracts?.contextEngines).toContain('gbrain-context');
+    expect(entrySource).toContain('export function register');
+  });
+});
@@ -186,11 +186,67 @@ describe('shouldExclude — orphan filter regression (preserve curation)', () =>
    expect(shouldExclude('entities/anonymous')).toBe(true);
    expect(shouldExclude('atoms/fact-123')).toBe(true);
    expect(shouldExclude('skills/gbrain-operations')).toBe(true);
+    expect(shouldExclude('dreaming/light/2026-07-20')).toBe(true);
+    expect(shouldExclude('daily/2026-07-20')).toBe(true);
+    expect(shouldExclude('agent-openclaw/daily/2026-07-20')).toBe(true);
+  });
+
+  test('workspace convention slugs are excluded', () => {
+    expect(shouldExclude('_brain-conventions')).toBe(true);
+    expect(shouldExclude('_templates/decision')).toBe(true);
+    expect(shouldExclude('extracts/2026-06-30/takes.proposed/round-single')).toBe(true);
+    expect(shouldExclude('2026-07-20')).toBe(true);
+    expect(shouldExclude('2026-07-20-qa-sweep')).toBe(true);
+    expect(shouldExclude('agents/arya/identity')).toBe(true);
+    expect(shouldExclude('agents/arya/memory/dreaming/deep/2026-07-20')).toBe(true);
  });

  test('regular slugs are NOT excluded', () => {
    expect(shouldExclude('people/alice')).toBe(false);
    expect(shouldExclude('companies/acme')).toBe(false);
    expect(shouldExclude('writing/post-1')).toBe(false);
+    expect(shouldExclude('agents/arya/qa-reports/launch-review')).toBe(false);
+  });
+});
+
+describe('getHealth orphan_pages uses shared exclusion policy', () => {
+  test('excluded convention islands do not count against health', async () => {
+    await engine.putPage('_templates/decision', {
+      type: 'template', title: 'Decision', compiled_truth: 'template', timeline: '', frontmatter: {},
+    });
+    await engine.putPage('skills/arya/source-check', {
+      type: 'concept', title: 'Skill', compiled_truth: 'skill', timeline: '', frontmatter: {},
+    });
+    await engine.putPage('agents/arya/identity', {
+      type: 'note', title: 'Identity', compiled_truth: 'identity', timeline: '', frontmatter: {},
+    });
+    await engine.putPage('people/alice', {
+      type: 'person', title: 'Alice', compiled_truth: 'real island', timeline: '', frontmatter: {},
+    });
+
+    const health = await engine.getHealth();
+
+    expect(health.orphan_pages).toBe(1);
+  });
+
+  test('per-brain config overrides (orphans.exclude_*) also apply to health', async () => {
+    await engine.putPage('my-private-folder/secret-ref', {
+      type: 'note', title: 'Ref', compiled_truth: 'ref', timeline: '', frontmatter: {},
+    });
+    await engine.putPage('one-off-fixture-page', {
+      type: 'note', title: 'Fixture', compiled_truth: 'fixture', timeline: '', frontmatter: {},
+    });
+    await engine.putPage('people/alice', {
+      type: 'person', title: 'Alice', compiled_truth: 'real island', timeline: '', frontmatter: {},
+    });
+
+    expect((await engine.getHealth()).orphan_pages).toBe(3);
+
+    await engine.setConfig('orphans.exclude_prefixes', 'my-private-folder/');
+    await engine.setConfig('orphans.exclude_slugs', 'one-off-fixture-page');
+    expect((await engine.getHealth()).orphan_pages).toBe(1);
+
+    await engine.unsetConfig('orphans.exclude_prefixes');
+    await engine.unsetConfig('orphans.exclude_slugs');
  });
 });
@@ -66,6 +66,10 @@ describe('shouldExclude', () => {
    expect(shouldExclude('templates/meeting-note')).toBe(true);
  });

+  test('excludes deny-prefix: _templates/', () => {
+    expect(shouldExclude('_templates/meeting-note')).toBe(true);
+  });
+
  test('excludes deny-prefix: openclaw/config/', () => {
    expect(shouldExclude('openclaw/config/agent')).toBe(true);
  });
@@ -86,10 +90,44 @@ describe('shouldExclude', () => {
    expect(shouldExclude('entities/product-hunt')).toBe(true);
  });

+  test('excludes first-segment: skills, dreaming, and daily', () => {
+    expect(shouldExclude('skills/arya/source-check')).toBe(true);
+    expect(shouldExclude('dreaming/light/2026-07-20')).toBe(true);
+    expect(shouldExclude('daily/2026-07-20')).toBe(true);
+    expect(shouldExclude('agent-openclaw/daily/2026-07-20')).toBe(true);
+  });
+
+  test('excludes root date logs and agent workspace conventions', () => {
+    expect(shouldExclude('_brain-conventions')).toBe(true);
+    expect(shouldExclude('2026-07-20')).toBe(true);
+    expect(shouldExclude('2026-07-20-qa-sweep')).toBe(true);
+    expect(shouldExclude('agents/arya/identity')).toBe(true);
+    expect(shouldExclude('agents/arya/memory/dreaming/deep/2026-07-20')).toBe(true);
+  });
+
+  test('excludes generated extracts', () => {
+    expect(shouldExclude('extracts/2026-06-30/takes.proposed/round-single')).toBe(true);
+  });
+
+  test('brain-specific exclusions come from config overrides, not global defaults', () => {
+    // No baked-in defaults for these:
+    expect(shouldExclude('my-private-folder/some-secret-ref.md')).toBe(false);
+    expect(shouldExclude('one-off-fixture-page')).toBe(false);
+    // The per-brain config plane (orphans.exclude_prefixes / exclude_slugs):
+    const overrides = {
+      excludePrefixes: ['my-private-folder/'],
+      excludeSlugs: ['one-off-fixture-page'],
+    };
+    expect(shouldExclude('my-private-folder/some-secret-ref.md', overrides)).toBe(true);
+    expect(shouldExclude('one-off-fixture-page', overrides)).toBe(true);
+    expect(shouldExclude('people/jane-doe', overrides)).toBe(false);
+  });
+
  test('does NOT exclude a normal content page', () => {
    expect(shouldExclude('companies/acme')).toBe(false);
    expect(shouldExclude('people/jane-doe')).toBe(false);
    expect(shouldExclude('projects/gbrain')).toBe(false);
+    expect(shouldExclude('agents/arya/qa-reports/launch-review')).toBe(false);
  });

  test('does NOT exclude a page ending with log-like text that is not /log', () => {
@@ -0,0 +1,34 @@
+/**
+ * #1098 — ClawVisor health checks must hit /ready (surfaces db/vault
+ * degradation), never /health (bare liveness, masks those failure modes).
+ * #1099 — the credential-gateway recipe tells users to activate Google
+ * Contacts; the canonical consumer recipe must actually ship.
+ */
+
+import { describe, test, expect } from 'bun:test';
+import { readFileSync, existsSync } from 'fs';
+import { join } from 'path';
+
+const RECIPES = join(import.meta.dir, '../recipes');
+
+describe('#1098 — ClawVisor health checks use /ready', () => {
+  for (const f of ['email-to-brain.md', 'calendar-to-brain.md', 'credential-gateway.md', 'contacts-to-brain.md']) {
+    test(`${f} references $CLAWVISOR_URL/ready, never /health`, () => {
+      const text = readFileSync(join(RECIPES, f), 'utf-8');
+      expect(text).not.toContain('$CLAWVISOR_URL/health');
+      expect(text).toContain('$CLAWVISOR_URL/ready');
+    });
+  }
+});
+
+describe('#1099 — canonical contacts-to-brain recipe ships', () => {
+  test('recipes/contacts-to-brain.md exists with the sibling frontmatter shape', () => {
+    const p = join(RECIPES, 'contacts-to-brain.md');
+    expect(existsSync(p)).toBe(true);
+    const text = readFileSync(p, 'utf-8');
+    expect(text).toContain('id: contacts-to-brain');
+    expect(text).toContain('requires: [credential-gateway]');
+    expect(text).toContain('health_checks:');
+    expect(text).toContain('contacts.readonly');
+  });
+});
@@ -12,9 +12,11 @@ import {
  bestPath,
  historyPath,
  loadHistory,
+  proposedPath,
  revertAllPending,
  skillPath,
  versionsDir,
+  writeProposed,
 } from '../../src/core/skillopt/version-store.ts';

 let tmpDir: string;
@@ -79,6 +81,19 @@ describe('acceptCandidate (D8 two-phase commit)', () => {
  });
 });

+describe('writeProposed', () => {
+  test('writes distinct best and proposed artifacts without mutating SKILL.md (#2635)', () => {
+    const candidate = '---\nname: test\n---\nproposed body\n';
+
+    const written = writeProposed(tmpDir, SKILL, candidate);
+
+    expect(written).toBe(proposedPath(tmpDir, SKILL));
+    expect(fs.readFileSync(bestPath(tmpDir, SKILL), 'utf8')).toBe(candidate);
+    expect(fs.readFileSync(proposedPath(tmpDir, SKILL), 'utf8')).toBe(candidate);
+    expect(fs.readFileSync(skillPath(tmpDir, SKILL), 'utf8')).toContain('baseline body');
+  });
+});
+
 describe('revertAllPending (D8 crash recovery)', () => {
  test('no-op when no pending rows', () => {
    const reverted = revertAllPending(tmpDir, SKILL);
Author	SHA1	Message	Date
Garry TanandClaude Fable 5	7e1c4f312f	Merge origin/master — reconcile #1144 composition-aware brain_score with #3023 orphan-policy getHealth rewrite Master's #3023 moved orphan_pages out of SQL into a TS filter over islanded rows (shared orphan-reporting policy). Kept that structure and threaded the #1144 narrative exclusion into the islanded query instead: code/calendar-index pages no longer enter the orphan numerator, matching the narrative_pages denominator. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>	2026-07-22 11:12:32 -07:00
Time Attakc Garry Tan Filip Claude Fable 5	d9eb027bdd	fix(openclaw): declare gbrain plugin manifest entry (takeover of #2551 ) (#3185 ) Add the OpenClaw-required top-level id to openclaw.plugin.json, export a direct register(api) entrypoint from src/openclaw-context-engine.ts, add a manifest regression test, and document that skillpack harvest must preserve OpenClaw-native manifest fields (id, configSchema, contracts). llms bundles regenerated (bun run build:llms) — no content drift. Co-authored-by: Garry Tan <garrytan@gmail.com> Co-authored-by: Filip <FilipHarald@users.noreply.github.com> Co-authored-by: Claude Fable 5 <noreply@anthropic.com>	2026-07-22 01:32:11 -07:00
Time Attakc Garry Tan Ziyang Guo Claude Fable 5	62e009d192	fix(skillopt): emit proposed.md in no-mutate mode (#2635 ) (#3182 ) Takeover of #2719 (fork head; rebased onto origin/master). - writeProposed now writes both best.md (current-best pointer) and proposed.md (stable human-review artifact); returns the proposal path. - Orchestrator reports the real proposed.md path for accepted --no-mutate runs. - Tutorial updated; llms bundles regenerated (no content drift — tutorial is not inlined in the bundle). Co-authored-by: Garry Tan <garrytan@gmail.com> Co-authored-by: Ziyang Guo <121015044+RerankerGuo@users.noreply.github.com> Co-authored-by: Claude Fable 5 <noreply@anthropic.com>	2026-07-22 00:23:10 -07:00
Time Attakc Garry Tan jessems Claude Fable 5	314fefa560	fix(readme): correct broken OpenClaw and Hermes project links (#1961 ) (#3179 ) Point the OpenClaw and Hermes anchors in the "Have your agent install it" section at their real upstream repos; the previous openclawagents org URLs 404. Regenerated llms-full.txt to match. Takeover of #1961 (fork branch) rebased onto current master. Co-authored-by: Garry Tan <garrytan@gmail.com> Co-authored-by: jessems <jessems@users.noreply.github.com> Co-authored-by: Claude Fable 5 <noreply@anthropic.com>	2026-07-21 20:22:27 -07:00
Time Attakc Garry Tan jdewoski-cmd Claude Fable 5	7f841fae7f	feat(maintain): safe maintenance automation + shared orphan-exclusion policy (#3015 ) (#3023 ) Ports #3015 by @jdewoski-cmd onto current master: - src/core/orphan-policy.ts centralizes the orphan-reporting exclusion convention so `gbrain orphans`, doctor's orphan_ratio, and both engines' getHealth orphan_pages can no longer drift. - getHealth stale_pages now uses the link-extractor stale watermark (countStalePagesForExtraction) so health agrees with what `gbrain extract --stale` will actually process. - New `gbrain maintain` command: dry-run by default, `--safe` applies only the conservative runbook actions (DB-backed stale extraction + source-scoped dream cycles for doctor cycle_freshness findings), `--json` for structured before/action/after reports. Frontmatter mutations, schema-pack upgrades, and semantic hub links stay review-only by design. Changed from the original PR: the shared defaults carried slugs specific to the contributor's own brain ('josa-secrets/', '-ga4-property-id.md', '-josa-test', literal 'welcome'/'untitled' fixtures). Global defaults now carry only GBrain-wide conventions; brain-specific exclusions move to a new per-brain config plane the policy reads through loadOrphanPolicyOverrides: gbrain config set orphans.exclude_prefixes "my-private-folder/,archive/" gbrain config set orphans.exclude_slugs "some-one-off-page" Both engines' getHealth and the orphans command thread the overrides; tests cover the neutral defaults, the override plane, and health parity. Also registered `maintain` in CLI_ONLY_SELF_HELP so `gbrain maintain --help` reaches the command's own usage block. Co-authored-by: Garry Tan <garrytan@gmail.com> Co-authored-by: jdewoski-cmd <jdewoski@gmail.com> Co-authored-by: Claude Fable 5 <noreply@anthropic.com>	2026-07-21 19:52:18 -07:00
paul-0320 YMYD Claude Fable 5 Time Attakc	1fabbb9849	fix(links): resolve path-qualified wikilinks outside DIR_PATTERN in the DB/put_page path (#2866 ) The generic wikilink pass (issue #972) forwarded the raw literal to resolveBasenameMatches, whose index is keyed by final path segments — so [[notes/struktura]] (any dir outside DIR_PATTERN) silently produced zero edges from `extract links --source db` and put_page auto-link, while the FS extractor resolves the identical content (resolveSlugAll strips the dirname before its basename lookup). Query by the literal's final segment, then keep only matches whose slug ends with the written path — [[notes/struktura]] can resolve to vault/notes/struktura but never attach to wiki/struktura. Bare literals are untouched. Flag-gated by link_resolution.global_basename as before. Co-authored-by: YMYD <53603073+OJ-OnJourney@users.noreply.github.com> Co-authored-by: Claude Fable 5 <noreply@anthropic.com> Co-authored-by: Time Attakc <89218912+time-attack@users.noreply.github.com>	2026-07-21 18:45:23 -07:00
Ryan Ayers Claude Opus 4.7 Time Attakc	64920f83c9	fix(embed): preserve code-chunk metadata across re-embed (#769 ) (#1232 ) Closes #769. Every re-embed pass clobbered code-chunk metadata (language, symbol_name, symbol_type, start_line, end_line, parent_symbol_path, doc_comment, symbol_name_qualified) to NULL, disabling code-def queries across thousands of indexed chunks. Two complementary fixes: embed.ts — three re-upsert call sites (embedPage, embedAll non-stale, embedAllStale autopilot path) build ChunkInputs from loaded chunks; they were stripping the 8 metadata fields. New preserveCodeMetadata helper threads those fields through consistently. Integrated cleanly with v0.34.4.0's cursor-paginated --stale hardening — the wrap sits inside the worker function between embedBatchWithBackoff and engine.upsertChunks. postgres-engine.ts + pglite-engine.ts — upsertChunks ON CONFLICT clause OVERWROTE metadata columns from EXCLUDED. Asymmetric vs the embedding/embedded_at columns which already used a chunk_text-gated CASE pattern (re-chunk → trust EXCLUDED, re-embed → COALESCE preserve). Applied the same pattern to all 8 metadata columns. Three regression tests in test/embed.serial.test.ts cover --stale (autopilot), --all, and --slugs paths. Each loads a chunk with full metadata, runs runEmbed, and asserts engine.upsertChunks receives the metadata round-tripped. Coexists with master's D5 embedBatchWithBackoff test block. Backfill required after deploy: \`gbrain sync --strategy code --force --source <id>\` per code source to re-populate metadata via the chunker. Without backfill, existing NULL columns stay NULL — re-embed alone never produces metadata, only the chunker does. Originally landed as part of PR #768 (the wave that bundled #767 + fix; this PR carries the #769 fix alone with no scope overlap. Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com> Co-authored-by: Time Attakc <89218912+time-attack@users.noreply.github.com>	2026-07-21 18:13:45 -07:00
Benjamin D. SmithandTime Attakc	e861b92da7	feat(synopsis): tail-truncate documentText for small-model chat handlers (#1427 ) * feat(synopsis): tail-truncate documentText for small-model chat handlers Small local chat models (Gemma 4 E2B, Qwen3 4B) get dramatically slower on long contexts even at 131K declared windows. A 73K-char page synopsis on Gemma 4 E2B takes 60-120s, exceeding the worker's default 30s `lockDuration` and tripping `lock-lost` errors. Add `SYNOPSIS_DOC_MAX_CHARS` env-overridable cap (default 32768 chars, ~8K tokens) applied in `buildUserPrompt`. Truncate the TAIL so the head (title, frontmatter, intro) preserves the document-level anchor the synopsis needs. Anthropic Haiku is unaffected at this cap; bump via `GBRAIN_SYNOPSIS_DOC_MAX_CHARS` for frontier models that want richer document anchoring. Belt-and-suspenders companion to commit 0aaff691 (--lock-duration flag on the worker). Combined: bumping lock TTL gives the handler more time, AND truncating doc cap makes the handler complete faster. Either alone helps; both together get the synopsis backfill running reliably on small local LLMs. Verified: real 1383-chunk personal brain backfill at GBRAIN_SYNOPSIS_MODEL=lmstudio:google/gemma-4-e2b + GBRAIN_SYNOPSIS_DOC_MAX_CHARS=16384 + `gbrain jobs work --concurrency 4 --lock-duration 300000` transitions from "lock-lost on every transcript page" to "no deaths, no stalls, steady throughput." RECOVERY REBUILD 2026-05-26 of original ac213aa6. * fix: fold SYNOPSIS_DOC_MAX_CHARS into corpus_generation hash Codex review of #1427 flagged that changing GBRAIN_SYNOPSIS_DOC_MAX_CHARS shifts the synopsis prompt + downstream embeddings for long documents but was NOT folded into the computeCorpusGeneration hash. Pages re-embedded with a different cap would retain the same corpus_generation, defeating the v0.40.3.0 D27 P1-5 cache invalidation contract. Three changes: 1. Export SYNOPSIS_DOC_MAX_CHARS from src/core/page-summary.ts 2. computeCorpusGeneration accepts optional synopsisDocMaxChars param. When set, folded into hash via '\|doc_cap=<N>'. Omitted for non-synopsis modes (title / none don't consult the cap) so existing pre-PR caches stay valid for those. 3. Service-layer call sites (2 in contextual-retrieval-service.ts) pass SYNOPSIS_DOC_MAX_CHARS when attemptMode/resolution.mode is per_chunk_synopsis, undefined otherwise. 4. import-file.ts inline path passes undefined (per_chunk_synopsis refused upstream there). One-time effect: per_chunk_synopsis pages re-embedded post-PR get a NEW corpus_generation including the cap. v0.40.3.0 query_cache.page_generations contract auto-invalidates cached query results on first re-embed. Future cap changes track correctly. Addresses codex review P2 on PR #1427. --------- Co-authored-by: Time Attakc <89218912+time-attack@users.noreply.github.com>	2026-07-21 16:50:43 -07:00
Garry TanandClaude Fable 5	e5b3e7aba7	fix(health): composition-aware brain_score + integrity/recipe fixes (#1098 #1099 #1143 #1144 ) - recipes: ClawVisor health checks hit /ready (surfaces db/vault degradation) instead of bare-liveness /health across email/calendar/credential-gateway (#1098) - recipes: ship canonical contacts-to-brain.md, the Google Contacts consumer the credential-gateway recipe already promises (#1099) - integrity: skip type:code pages in both scan paths and in integrity auto so 'tweet' in source identifiers/comments can't be flagged or auto-repaired (#1143) - health: exclude code / calendar-index pages from the orphan, link-density and timeline denominators of brain_score in BOTH engines (engine parity), so bulk code/calendar imports no longer crater the score (#1144) Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>	2026-07-21 14:35:01 -07:00