fix(chunker): estimated-token hard cap — URL-dense/CJK-fallback chunks overflow strict embedding-server token limits

Takeover of #2847 (rebased onto current master). Fixes #2826. - cjk.ts: estimateEmbeddingTokens() — conservative per-char-class token estimate (CJK 1.0, other 0.75, whitespace 0.1 per code unit). - recursive.ts (MARKDOWN_CHUNKER_VERSION 3→4): countWords floored at ceil(nonWhitespaceChars/6); capByEstimatedTokens() final pass with ChunkOptions.maxTokens (default 1500). - code.ts (CHUNKER_VERSION 4→5): capCodeChunks() applies the same cap to AST-path chunks that splitLargeNode can't subdivide. Co-authored-by: paul-0320 <paul-0320@users.noreply.github.com> Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
2026-08-16 01:42:23 +00:00 · 2026-07-21 14:18:40 -07:00
44 changed files with 580 additions and 1264 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/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)
+- **[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)

 Then paste this into your agent:

@@ -131,9 +131,7 @@ 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. Harvest must preserve
-  the top-level OpenClaw-native plugin fields (`id`, `configSchema`, `contracts`)
-  because OpenClaw validates those before it can install the package.
+- `openclaw.plugin.json` updated with the new slug, sorted.
 - `--no-lint` bypasses the linter (after a manual editorial scrub).

 Use the `skillpack-harvest` skill (its companion editorial workflow)
@@ -233,14 +233,13 @@ 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/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:
+`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:

 ```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/proposed.md, updates best.md, and prints the proposal path.
+# → writes skills/meeting-prep/skillopt/best.md (the proposed rewrite), prints its path. Copy what you want.

 # 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/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)
+- **[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)

 Then paste this into your agent:

@@ -1,5 +1,4 @@
 {
-  "id": "gbrain-context-engine",
  "name": "gbrain",
  "version": "0.32.3.0",
  "description": "Personal knowledge brain with Postgres + pgvector hybrid search",
@@ -266,5 +266,4 @@ 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, without removing OpenClaw-native plugin fields
-  like `id`, `configSchema`, or `contracts`
+  array, sorted alphabetically
@@ -57,8 +57,6 @@ 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
@@ -74,7 +72,7 @@ This mode guarantees:
 ### Automation

 ```bash
-bun test test/skills-conformance.test.ts test/resolver.test.ts test/openclaw-plugin-manifest.test.ts
+bun test test/skills-conformance.test.ts test/resolver.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', 'bench', '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']);
+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', '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.
@@ -78,8 +78,6 @@ 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
@@ -106,9 +104,6 @@ const CLI_ONLY_SELF_HELP = new Set([
  // `gbrain connect --help` prints its own usage (flags + examples) from
  // runConnect; route around the generic one-line short-circuit.
  'connect',
-  // #1474: bench-publish ships its own detailed HELP (flags, exit codes,
-  // the export → publish → gate loop). Route around the generic stub.
-  'bench',
 ]);

 // v114 (#1941): alias -> operation lookup, kept separate from `cliOps` so
@@ -1432,29 +1427,6 @@ async function handleCliOnly(command: string, args: string[]) {
    return;
  }

-  // #1474: `gbrain bench publish` is pure file I/O (reads a captured
-  // eval-candidates NDJSON from `gbrain eval export`, writes a baseline
-  // NDJSON). No DB access; bypass connectEngine entirely so the documented
-  // export → publish → gate loop works on machines without a brain.
-  // The v0.41.1 wave shipped bench-publish.ts + docs/eval-bench.md but this
-  // dispatcher case was never added, so the command hit 'Unknown command'.
-  if (command === 'bench') {
-    if (args[0] === 'publish') {
-      const { runBenchPublish } = await import('./commands/bench-publish.ts');
-      await runBenchPublish(args.slice(1));
-      return;
-    }
-    if (args.length === 0 || args[0] === '--help' || args[0] === '-h') {
-      const { runBenchPublish } = await import('./commands/bench-publish.ts');
-      await runBenchPublish(['--help']);
-      return;
-    }
-    console.error(`Unknown bench subcommand: ${args[0]}`);
-    console.error('Usage: gbrain bench publish --from <captured.ndjson> --to <baseline.ndjson> [flags]');
-    console.error('  See docs/eval-bench.md for the full loop: eval export → bench publish → eval gate');
-    process.exit(2);
-  }
-
  // v0.42.x (#2390): `gbrain eval chronicle` is deterministic — brings its own
  // in-memory PGLite, no DB/gateway. CI fixture gate runs anywhere.
  if (command === 'eval' && args[0] === 'chronicle') {
@@ -1785,11 +1757,6 @@ 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.
@@ -2250,22 +2217,6 @@ async function connectEngine(opts?: { probeOnly?: boolean }): Promise<BrainEngin
      if (merged.embedding_image_ocr_model !== undefined) {
        process.env.GBRAIN_EMBEDDING_IMAGE_OCR_MODEL = merged.embedding_image_ocr_model;
      }
-      // #1475: stash the merged eval.* flags the same way. The capture gate
-      // (isEvalCaptureEnabled / isEvalScrubEnabled) runs against ctx.config,
-      // which is built from the sync file-plane loadConfig() in both the CLI
-      // op path and MCP dispatch — it never sees the DB plane directly. The
-      // gates consult this stash when the file plane is silent, so
-      // `gbrain config set eval.capture true` actually turns capture on.
-      // A pre-set env value wins over the DB plane (env-above-config, the
-      // incident escape hatch) — unlike GBRAIN_EMBEDDING_MULTIMODAL these
-      // keys have no loadConfig() env mapping, so without this guard the
-      // DB stash would silently clobber an operator's export.
-      if (process.env.GBRAIN_EVAL_CAPTURE === undefined && merged.eval?.capture !== undefined) {
-        process.env.GBRAIN_EVAL_CAPTURE = String(merged.eval.capture);
-      }
-      if (process.env.GBRAIN_EVAL_SCRUB_PII === undefined && merged.eval?.scrub_pii !== undefined) {
-        process.env.GBRAIN_EVAL_SCRUB_PII = String(merged.eval.scrub_pii);
-      }
      // Always re-configure with merged values when DB merge succeeded. The
      // trigger used to be field-name-gated (only when embedding_multimodal_model
      // was set); that coupled the gate to the field set and would silently
@@ -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 => preserveCodeMetadata(c, {
+  const updated: ChunkInput[] = chunks.map(c => ({
    chunk_index: c.chunk_index,
    chunk_text: c.chunk_text,
    chunk_source: c.chunk_source,
@@ -605,31 +605,6 @@ 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,
@@ -742,10 +717,8 @@ 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.
-      // 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, {
+      // Preserve ALL chunks, only update embeddings for stale ones
+      const updated: ChunkInput[] = chunks.map(c => ({
        chunk_index: c.chunk_index,
        chunk_text: c.chunk_text,
        chunk_source: c.chunk_source,
@@ -1039,10 +1012,7 @@ async function embedAllStale(
          for (let j = 0; j < stale.length; j++) {
            staleIdxToEmbedding.set(stale[j].chunk_index, embeddings[j]);
          }
-          // 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, {
+          const merged: ChunkInput[] = existing.map(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.
 */
-export async function extractStaleFromDB(
+async function extractStaleFromDB(
  engine: BrainEngine,
  opts: {
    dryRun: boolean;
@@ -1,224 +0,0 @@
-/**
- * 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.');
-  }
-}
@@ -15,11 +15,6 @@
 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 ---

@@ -37,14 +32,65 @@ 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, overrides?: OrphanPolicyOverrides): boolean {
-  return shouldExcludeFromOrphanReporting(slug, overrides);
+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;
 }

 /**
@@ -110,7 +156,6 @@ 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,
@@ -139,7 +184,7 @@ export async function findOrphans(
    total = liveRows.length;
    excludedAll = includePseudo
      ? 0
-      : liveRows.reduce((n, r) => n + (shouldExclude(r.slug, overrides) ? 1 : 0), 0);
+      : liveRows.reduce((n, r) => n + (shouldExclude(r.slug) ? 1 : 0), 0);
  } finally {
    stopHb();
    progress.finish();
@@ -147,7 +192,7 @@ export async function findOrphans(

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

  const orphans: OrphanPage[] = filtered.map(row => ({
    slug: row.slug,
@@ -18,8 +18,9 @@
 * at runtime.
 */

-import { chunkText as recursiveChunk } from './recursive.ts';
+import { chunkText as recursiveChunk, capByEstimatedTokens, DEFAULT_MAX_EST_TOKENS } from './recursive.ts';
 import { buildQualifiedName } from './qualified-names.ts';
+import { estimateEmbeddingTokens } from '../cjk.ts';

 // Embed the tree-sitter runtime + per-language grammars as files.
 // `with { type: 'file' }` returns a path (string) at runtime. Bun bundles
@@ -111,7 +112,15 @@ import G_ZIG from '../../assets/wasm/grammars/tree-sitter-zig.wasm' with { type:
 // chunks get the new columns populated. Without this, the v28 backfill
 // gives every existing chunk a search_vector but subsequent Layer 5 AST
 // work would silently no-op.
-export const CHUNKER_VERSION = 4;
+//
+// v5: estimated-token hard cap on AST-path chunks (capCodeChunks). A node
+// splitLargeNode can't subdivide (giant single-statement function, huge
+// literal) previously shipped WHOLE regardless of size and could overflow
+// strict per-request embedding-token limits (local llama-server crashes
+// past ~2,050 tokens, measured). Mirrors the markdown
+// chunker's v4 cap; fallback-path chunks are already capped inside
+// recursiveChunk.
+export const CHUNKER_VERSION = 5;

 // Lazy-loaded tree-sitter module (v0.22.x API: Parser is default export)
 let Parser: typeof import('web-tree-sitter') | null = null;
@@ -708,7 +717,7 @@ export async function chunkCodeTextFull(
    if (chunks.length === 0) {
      return { chunks: fallbackChunks(source, filePath, language, opts), edges: rawEdges };
    }
-    return { chunks: mergeSmallSiblings(chunks, chunkTarget), edges: rawEdges };
+    return { chunks: capCodeChunks(mergeSmallSiblings(chunks, chunkTarget)), edges: rawEdges };
  } catch {
    return { chunks: fallbackChunks(source, filePath, language, opts), edges: [] };
  } finally {
@@ -791,6 +800,33 @@ function mergeSmallSiblings(chunks: CodeChunk[], chunkTarget: number): CodeChunk
  return merged;
 }

+/**
+ * v5 final safety pass for AST-path chunks: split any chunk whose
+ * ESTIMATED embedding tokens (conservative per-char-class heuristic,
+ * cjk.ts) exceed DEFAULT_MAX_EST_TOKENS. Reaches chunks the AST logic
+ * can't subdivide — splitLargeNode returns [] for nodes with < 2 body
+ * children (giant single-statement functions, huge literals), which
+ * previously shipped whole at any size.
+ *
+ * Split pieces inherit the source chunk's metadata verbatim; start/end
+ * lines become approximate for pieces after the first. Acceptable —
+ * these chunks exist for embedding + retrieval, and the alternative was
+ * an embedding request the server rejects (or worse, crashes on).
+ */
+function capCodeChunks(chunks: CodeChunk[]): CodeChunk[] {
+  if (chunks.every((c) => estimateEmbeddingTokens(c.text) <= DEFAULT_MAX_EST_TOKENS)) {
+    return chunks;
+  }
+  const out: CodeChunk[] = [];
+  for (const c of chunks) {
+    const pieces = capByEstimatedTokens(c.text, DEFAULT_MAX_EST_TOKENS);
+    for (const piece of pieces) {
+      out.push({ ...c, text: piece, index: out.length, metadata: { ...c.metadata } });
+    }
+  }
+  return out;
+}
+
 function buildMergedChunk(group: CodeChunk[], index: number): CodeChunk {
  const first = group[0]!;
  const last = group[group.length - 1]!;
@@ -17,7 +17,13 @@
 * Lossless invariant: non-overlapping portions reassemble to original.
 */

-import { countCJKAwareWords, CJK_SENTENCE_DELIMITERS, CJK_CLAUSE_DELIMITERS } from '../cjk.ts';
+import {
+  countCJKAwareWords,
+  CJK_SENTENCE_DELIMITERS,
+  CJK_CLAUSE_DELIMITERS,
+  charEmbedTokenWeight,
+  estimateEmbeddingTokens,
+} from '../cjk.ts';

 /**
 * Markdown chunker version. Folded into the per-page chunker_version column
@@ -33,8 +39,20 @@ import { countCJKAwareWords, CJK_SENTENCE_DELIMITERS, CJK_CLAUSE_DELIMITERS } fr
 * re-embed (not re-chunk) so existing pages pick up the wrapper on the
 * post-upgrade reembed sweep. See
 * `src/core/contextual-retrieval-service.ts`.
+ *
+ * v4: estimated-token hard cap + whitespace-word undercount fix. The word
+ * pipeline counted a 150-char URL as ONE whitespace word, so URL/phone/
+ * email-dense docs (CJK density < 0.30 → whitespace fallback) produced
+ * 3-4K-char chunks that overflow strict per-request embedding-token
+ * limits (measured: local llama-server crashes past ~2,050 tokens; URL
+ * soup tokenizes at ~1.6 chars/token). Two changes:
+ *   1. countWords() floors the count at ceil(nonWhitespaceChars/6) so a
+ *      URL counts roughly per-character, not as one word.
+ *   2. capByEstimatedTokens() final pass guarantees every chunk fits
+ *      `maxTokens` (default 1500) under a conservative per-char-class
+ *      token estimate, regardless of how word counting misjudged it.
 */
-export const MARKDOWN_CHUNKER_VERSION = 3;
+export const MARKDOWN_CHUNKER_VERSION = 4;

 const DELIMITERS: string[][] = [
  ['\n\n'],                          // L0: paragraphs
@@ -48,8 +66,20 @@ export interface ChunkOptions {
  chunkSize?: number;    // target words per chunk (default 300)
  chunkOverlap?: number; // overlap words (default 50)
  maxChars?: number;     // hard cap on any chunk's char length (default 6000)
+  /**
+   * v4: hard cap on any chunk's ESTIMATED embedding tokens (default 1500).
+   * Estimate = conservative per-char-class weights (see cjk.ts
+   * estimateEmbeddingTokens) — deliberately high, so the real tokenizer
+   * count stays below this value. Default leaves headroom for the
+   * contextual-retrieval wrapper (≤ ~630 chars) under a ~2,050-token
+   * per-request embedding server limit.
+   */
+  maxTokens?: number;
 }

+/** v4 default for ChunkOptions.maxTokens — see the field doc above. */
+export const DEFAULT_MAX_EST_TOKENS = 1500;
+
 export interface TextChunk {
  text: string;
  index: number;
@@ -73,6 +103,7 @@ export function chunkText(text: string, opts?: ChunkOptions): TextChunk[] {
  const chunkSize = opts?.chunkSize || 300;
  const chunkOverlap = opts?.chunkOverlap || 50;
  const maxChars = opts?.maxChars || 6000;
+  const maxTokens = opts?.maxTokens || DEFAULT_MAX_EST_TOKENS;

  if (!text || text.trim().length === 0) return [];

@@ -89,8 +120,9 @@ export function chunkText(text: string, opts?: ChunkOptions): TextChunk[] {

  const wordCount = countWords(stripped);
  if (wordCount <= chunkSize) {
-    // Single-chunk path: still apply the maxChars cap.
-    const capped = capByChars(stripped.trim(), maxChars);
+    // Single-chunk path: still apply the maxChars + maxTokens caps.
+    const capped = capByChars(stripped.trim(), maxChars)
+      .flatMap((t) => capByEstimatedTokens(t, maxTokens));
    return capped.map((t, i) => ({ text: t, index: i }));
  }

@@ -101,9 +133,14 @@ export function chunkText(text: string, opts?: ChunkOptions): TextChunk[] {
  // v0.32.7: hard char cap. Catches pathological CJK + whitespace-less text
  // that the word-level pipeline can't bound (a single Chinese paragraph can
  // exceed 8192 OpenAI embedding tokens at any word count).
+  // v4: estimated-token cap on top — the char cap alone passes token-dense
+  // content (URL soup at ~1.6 chars/token) that overflows strict embedding
+  // server limits.
  const capped: string[] = [];
  for (const chunk of withOverlap) {
-    capped.push(...capByChars(chunk.trim(), maxChars));
+    for (const piece of capByChars(chunk.trim(), maxChars)) {
+      capped.push(...capByEstimatedTokens(piece, maxTokens));
+    }
  }
  return capped.map((t, i) => ({ text: t, index: i }));
 }
@@ -132,6 +169,68 @@ function capByChars(text: string, maxChars: number): string[] {
  return out;
 }

+/**
+ * How far back (in chars) the token cap looks for a friendly cut point
+ * before falling back to a hard cut. 300 covers typical rollup/list line
+ * lengths so forced splits land at line starts, not mid-URL.
+ */
+const TOKEN_CAP_CUT_LOOKBACK = 300;
+
+/**
+ * v4: hard-cap a chunk's ESTIMATED embedding tokens. Final safety pass —
+ * runs after capByChars on every chunk, so no upstream miscounting
+ * (whitespace-word fallback, overlap inflation, char-cap survivors) can
+ * emit a chunk past `maxTokens`.
+ *
+ * Cut placement prefers, within the last TOKEN_CAP_CUT_LOOKBACK chars of
+ * the window: a newline, then any whitespace, then a hard cut. This keeps
+ * forced splits off mid-line/mid-URL positions for list-shaped content
+ * and inside code fences. No overlap is added (pieces stay lossless
+ * modulo the trims the char cap already applies).
+ *
+ * @internal exported for the code chunker (code.ts) and tests.
+ */
+export function capByEstimatedTokens(text: string, maxTokens: number): string[] {
+  if (text.length === 0) return [];
+  if (estimateEmbeddingTokens(text) <= maxTokens) return [text];
+
+  const out: string[] = [];
+  let start = 0;
+  while (start < text.length) {
+    // Greedily extend the window until the next char would break the cap.
+    // Always take at least one char so the loop makes forward progress.
+    let est = 0;
+    let end = start;
+    while (end < text.length) {
+      const w = charEmbedTokenWeight(text.charCodeAt(end));
+      if (est + w > maxTokens && end > start) break;
+      est += w;
+      end++;
+    }
+
+    if (end < text.length) {
+      const windowStart = Math.max(start + 1, end - TOKEN_CAP_CUT_LOOKBACK);
+      let cut = text.lastIndexOf('\n', end - 1);
+      if (cut < windowStart) {
+        cut = -1;
+        for (let i = end - 1; i >= windowStart; i--) {
+          const code = text.charCodeAt(i);
+          if (code === 0x20 || (code >= 0x09 && code <= 0x0d)) {
+            cut = i;
+            break;
+          }
+        }
+      }
+      if (cut >= windowStart) end = cut + 1;
+    }
+
+    const slice = text.slice(start, end).trim();
+    if (slice.length > 0) out.push(slice);
+    start = end;
+  }
+  return out;
+}
+
 function recursiveSplit(text: string, level: number, target: number): string[] {
  if (level >= DELIMITERS.length) {
    // Level 4: split on whitespace
@@ -317,7 +416,19 @@ function extractTrailingContext(text: string, targetWords: number): string {
 * Delegated to src/core/cjk.ts so the slugify whitelist, expansion
 * detection, and PGLite keyword fallback all agree on what "CJK enough"
 * means.
+ *
+ * v4: floored at ceil(nonWhitespaceChars/6). The whitespace fallback
+ * counts a 150-char URL as ONE word, so URL/phone/email-dense docs
+ * (whose ASCII mass pushes CJK density below the 0.30 threshold) were
+ * sized at a fraction of their real bulk and merged into 3-4K-char
+ * chunks. The floor makes long whitespace-less runs count roughly
+ * per-character while leaving normal Latin prose untouched (average
+ * English word ≈ 5 chars < 6, so the whitespace count still wins).
+ * Kept local to the chunker — search/expansion.ts keeps the original
+ * countCJKAwareWords semantics for its query-length check.
 */
 function countWords(text: string): number {
-  return countCJKAwareWords(text);
+  const cjkAware = countCJKAwareWords(text);
+  const nonWhitespace = text.replace(/\s/g, '').length;
+  return Math.max(cjkAware, Math.ceil(nonWhitespace / 6));
 }
@@ -65,3 +65,64 @@ export function countCJKAwareWords(s: string): number {
 export function escapeLikePattern(s: string): string {
  return s.replace(/\\/g, '\\\\').replace(/%/g, '\\%').replace(/_/g, '\\_');
 }
+
+/**
+ * Conservative per-char-class embedding-token weights (markdown chunker v4).
+ *
+ * Why this exists: the chunker's "word" counting drastically UNDER-counts
+ * whitespace-less ASCII runs (a 150-char URL = 1 whitespace word), so
+ * word-based size targets can emit chunks that overflow an embedding
+ * server's per-request token limit. Measured on a local Qwen3-embedding
+ * llama-server stack:
+ *   - URL/phone/email-dense text tokenizes at ~1.6 chars/token
+ *   - base64-ish / minified blobs approach ~1.3 chars/token (worst case)
+ *   - Korean prose tokenizes NO WORSE than 1 char/token in practice
+ *
+ * Weights are deliberately HIGH (tokens are overestimated) so any cap
+ * based on this estimate is safe against real tokenizers:
+ *   - CJK char        → 1.0 token  (real CJK prose is cheaper)
+ *   - other non-space → 0.75 token (≈1.33 chars/token, covers base64)
+ *   - whitespace      → 0.1 token  (mostly folds into neighbor tokens)
+ */
+export const EMBED_TOKEN_WEIGHT_CJK = 1.0;
+export const EMBED_TOKEN_WEIGHT_OTHER = 0.75;
+export const EMBED_TOKEN_WEIGHT_WS = 0.1;
+
+/** BMP CJK check by UTF-16 code unit — same ranges as CJK_SLUG_CHARS. */
+export function isCJKCodeUnit(code: number): boolean {
+  return (
+    (code >= 0x4e00 && code <= 0x9fff) || // Han
+    (code >= 0x3040 && code <= 0x309f) || // Hiragana
+    (code >= 0x30a0 && code <= 0x30ff) || // Katakana
+    (code >= 0xac00 && code <= 0xd7af)    // Hangul Syllables
+  );
+}
+
+/**
+ * Per-code-unit token weight. Unrecognized whitespace (exotic Unicode
+ * spaces) intentionally falls into OTHER — that only overestimates.
+ */
+export function charEmbedTokenWeight(code: number): number {
+  if (isCJKCodeUnit(code)) return EMBED_TOKEN_WEIGHT_CJK;
+  if (
+    code === 0x20 || (code >= 0x09 && code <= 0x0d) ||
+    code === 0xa0 || code === 0x3000
+  ) {
+    return EMBED_TOKEN_WEIGHT_WS;
+  }
+  return EMBED_TOKEN_WEIGHT_OTHER;
+}
+
+/**
+ * Tokenizer-free embedding-token estimate (conservative overestimate).
+ * See weight docs above. Astral chars count as 2 OTHER code units —
+ * another overestimate, which is the safe direction.
+ */
+export function estimateEmbeddingTokens(s: string): number {
+  if (s.length === 0) return 0;
+  let est = 0;
+  for (let i = 0; i < s.length; i++) {
+    est += charEmbedTokenWeight(s.charCodeAt(i));
+  }
+  return Math.ceil(est);
+}
@@ -816,25 +816,6 @@ export async function loadConfigWithEngine(
    merged.dream = mergedDream;
  }

-  // #1475: eval.* DB-plane merge. `gbrain config set eval.capture true`
-  // writes the DB plane (both keys are in KNOWN_CONFIG_KEYS, so `set`
-  // accepts them silently), but the capture gate (isEvalCaptureEnabled)
-  // reads the merged config. Without this merge the DB value was written
-  // and never read — capture only fired via GBRAIN_CONTRIBUTOR_MODE=1.
-  // Sparse per-key merge: file/env wins per key, DB fills the gaps.
-  const dbEvalCapture = await dbBool('eval.capture');
-  const dbEvalScrub = await dbBool('eval.scrub_pii');
-  const mergedEval: NonNullable<GBrainConfig['eval']> = { ...(merged.eval ?? {}) };
-  if (mergedEval.capture === undefined && dbEvalCapture !== undefined) {
-    mergedEval.capture = dbEvalCapture;
-  }
-  if (mergedEval.scrub_pii === undefined && dbEvalScrub !== undefined) {
-    mergedEval.scrub_pii = dbEvalScrub;
-  }
-  if (Object.keys(mergedEval).length > 0) {
-    merged.eval = mergedEval;
-  }
-
  return merged;
 }

@@ -54,7 +54,6 @@ import {
 import {
  generatePerChunkSynopsis,
  SYNOPSIS_PROMPT_VERSION,
-  SYNOPSIS_DOC_MAX_CHARS,
  type GeneratePerChunkSynopsisResult,
 } from './page-summary.ts';
 import {
@@ -104,17 +103,8 @@ 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 {
-  const h = createHash('sha256')
+  return createHash('sha256')
    .update(args.crMode)
    .update('|')
    .update(String(SYNOPSIS_PROMPT_VERSION))
@@ -123,11 +113,9 @@ export function computeCorpusGeneration(args: {
    .update('|')
    .update(String(TITLE_WRAPPER_VERSION))
    .update('|')
-    .update(getEmbeddingModelTag());
-  if (args.synopsisDocMaxChars !== undefined) {
-    h.update('|doc_cap=').update(String(args.synopsisDocMaxChars));
-  }
-  return h.digest('hex').slice(0, 16);
+    .update(getEmbeddingModelTag())
+    .digest('hex')
+    .slice(0, 16);
 }

 /**
@@ -265,11 +253,7 @@ export async function reembedPageWithContextualRetrieval(
      args.pageSlug,
      args.sourceId,
      resolution.mode,
-      computeCorpusGeneration({
-        crMode: resolution.mode,
-        haikuModel: args.haikuModel ?? DEFAULT_HAIKU_MODEL,
-        synopsisDocMaxChars: resolution.mode === 'per_chunk_synopsis' ? SYNOPSIS_DOC_MAX_CHARS : undefined,
-      }),
+      computeCorpusGeneration({ crMode: resolution.mode, haikuModel: args.haikuModel ?? DEFAULT_HAIKU_MODEL }),
    );
    return { kind: 'skipped', reason: 'no_chunks' };
  }
@@ -298,7 +282,6 @@ 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 ───────────────────────────
@@ -251,14 +251,6 @@ registerBackgroundWorkDrainer({
 export function isEvalCaptureEnabled(config: GBrainConfig | null | undefined): boolean {
  if (config?.eval?.capture === true) return true;
  if (config?.eval?.capture === false) return false;
-  // #1475: DB-plane stash. `gbrain config set eval.capture true` lands in the
-  // config table; connectEngine stamps the merged value here because
-  // ctx.config is the sync file-plane load and never sees the DB plane.
-  // Explicit per-key setting (file above, DB here) beats the broad
-  // CONTRIBUTOR_MODE flag, matching how file-plane `false` already wins.
-  // Doubles as a direct operator env knob.
-  if (process.env.GBRAIN_EVAL_CAPTURE === 'true') return true;
-  if (process.env.GBRAIN_EVAL_CAPTURE === 'false') return false;
  return process.env.GBRAIN_CONTRIBUTOR_MODE === '1';
 }

@@ -271,8 +263,5 @@ export function isEvalCaptureEnabled(config: GBrainConfig | null | undefined): b
 * have explicit `capture: true`.
 */
 export function isEvalScrubEnabled(config: GBrainConfig | null | undefined): boolean {
-  if (config?.eval?.scrub_pii === false) return false;
-  if (config?.eval?.scrub_pii === true) return true;
-  // #1475: DB-plane stash — see isEvalCaptureEnabled. Default stays true.
-  return process.env.GBRAIN_EVAL_SCRUB_PII !== 'false';
+  return config?.eval?.scrub_pii !== false;
 }
@@ -733,11 +733,6 @@ 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
@@ -489,22 +489,7 @@ 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.
-      //
-      // 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(`/${ref.slug}`));
-      }
+      const matches = await resolver.resolveBasenameMatches(ref.slug);
      if (matches.length === 0) continue;
      const idx = content.indexOf(ref.slug);
      const context = idx >= 0 ? excerpt(content, idx, 240) : ref.name;
@@ -1,116 +0,0 @@
-/**
- * Shared orphan-reporting exclusion policy.
- *
- * These are pages where "no inbound links" is expected and should not count
- * against health. Keep this in core so the CLI orphan report and engine health
- * dashboard cannot drift.
- *
- * Defaults are GBrain-wide conventions only. Brain-specific exclusions
- * (private folder names, one-off fixture slugs) belong in the brain's own
- * config, not here:
- *
- *   gbrain config set orphans.exclude_prefixes "my-private-folder/,archive/"
- *   gbrain config set orphans.exclude_slugs "some-one-off-page"
- */
-
-const AUTO_SUFFIX_PATTERNS = ['/_index', '/log'];
-
-const PSEUDO_SLUGS = new Set(['_atlas', '_index', '_stats', '_orphans', '_scratch', 'claude']);
-
-const RAW_SEGMENT = '/raw/';
-
-const DENY_PREFIXES = [
-  'output/',
-  'dashboards/',
-  'scripts/',
-  'templates/',
-  '_templates/',
-  'openclaw/config/',
-  'extracts/',
-];
-
-const FIRST_SEGMENT_EXCLUSIONS = new Set([
-  'scratch',
-  'thoughts',
-  'catalog',
-  'entities',
-  'raw',
-  'atoms',
-  'skills',
-  'dreaming',
-  'daily',
-]);
-
-const ROOT_DATE_SLUG = /^\d{4}-\d{2}-\d{2}(?:-.+)?$/;
-
-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;
-}
@@ -44,33 +44,6 @@ 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
@@ -215,19 +188,11 @@ 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>',
-    trimmedDoc,
+    documentText,
    '</full_document>',
    '',
    '<chunk>',
@@ -57,8 +57,6 @@ 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,
@@ -2324,10 +2322,6 @@ 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
@@ -2351,14 +2345,14 @@ export class PGLiteEngine implements BrainEngine {
                THEN EXCLUDED.embedded_at
           ELSE content_chunks.embedded_at
         END,
-         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,
+         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,
         modality = EXCLUDED.modality,
         embedding_image = COALESCE(EXCLUDED.embedding_image, content_chunks.embedding_image)`,
      params
@@ -5213,10 +5207,15 @@ export class PGLiteEngine implements BrainEngine {
        (SELECT count(*) FROM pages) as 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,
-        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 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,
        (SELECT count(*) FROM links l
         WHERE NOT EXISTS (SELECT 1 FROM pages p WHERE p.id = l.to_page_id)
        ) as dead_links,
@@ -5241,20 +5240,10 @@ export class PGLiteEngine implements BrainEngine {
      LIMIT 5
    `);

-    const { rows: islandedRows } = await this.db.query(`
-      SELECT p.slug
-      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)
-    `);
-
    const r = h as Record<string, unknown>;
    const pageCount = Number(r.page_count);
    const embedCoverage = Number(r.embed_coverage);
-    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 orphanPages = Number(r.orphan_pages);
    const deadLinks = Number(r.dead_links);
    const linkCount = Number(r.link_count);
    const pagesWithTimeline = Number(r.pages_with_timeline);
@@ -5282,7 +5271,7 @@ export class PGLiteEngine implements BrainEngine {
    return {
      page_count: pageCount,
      embed_coverage: embedCoverage,
-      stale_pages: stalePages,
+      stale_pages: Number(r.stale_pages),
      orphan_pages: orphanPages,
      missing_embeddings: Number(r.missing_embeddings),
      brain_score: brainScore,
@@ -67,8 +67,6 @@ 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, "''");
@@ -2475,13 +2473,6 @@ 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
@@ -2505,14 +2496,14 @@ export class PostgresEngine implements BrainEngine {
                THEN EXCLUDED.embedded_at
           ELSE content_chunks.embedded_at
         END,
-         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,
+         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,
         modality = EXCLUDED.modality,
         embedding_image = COALESCE(EXCLUDED.embedding_image, content_chunks.embedding_image)`,
      params as Parameters<typeof sql.unsafe>[1],
@@ -5322,9 +5313,11 @@ 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). The raw islanded list is filtered through the same
-    // policy as `gbrain orphans` so convention pages do not count against
-    // dashboard health.
+    // 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.
    const [h] = await sql`
      WITH entity_pages AS (
        SELECT id, slug FROM pages WHERE type IN ('person', 'company')
@@ -5333,8 +5326,13 @@ export class PostgresEngine implements BrainEngine {
        (SELECT count(*) FROM pages) as 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,
-        0 as stale_pages,
-        0 as orphan_pages,
+        (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,
        (SELECT count(*) FROM links l
         WHERE NOT EXISTS (SELECT 1 FROM pages p WHERE p.id = l.to_page_id)
        ) as dead_links,
@@ -5358,18 +5356,9 @@ export class PostgresEngine implements BrainEngine {
      LIMIT 5
    `;

-    const islandedRows = await sql<{ slug: string }[]>`
-      SELECT p.slug
-      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)
-    `;
-
    const pageCount = Number(h.page_count);
    const embedCoverage = Number(h.embed_coverage);
-    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 orphanPages = Number(h.orphan_pages);
    const deadLinks = Number(h.dead_links);
    const linkCount = Number(h.link_count);
    const pagesWithTimeline = Number(h.pages_with_timeline);
@@ -5397,7 +5386,7 @@ export class PostgresEngine implements BrainEngine {
    return {
      page_count: pageCount,
      embed_coverage: embedCoverage,
-      stale_pages: stalePages,
+      stale_pages: Number(h.stale_pages),
      orphan_pages: orphanPages,
      missing_embeddings: Number(h.missing_embeddings),
      brain_score: brainScore,
@@ -93,13 +93,7 @@ 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,
-  proposedPath as proposedFilePath,
-  revertAllPending,
-  skillPath,
-  writeProposed,
-} from './version-store.ts';
+import { acceptCandidate, bestPath, 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';
@@ -708,9 +702,9 @@ 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') {
-    // writeProposed() emitted both the best pointer and the stable review
-    // artifact in the accept branch. SKILL.md remains untouched.
-    proposedPath = proposedFilePath(skillsDir, skillName);
+    // 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);
  } else if (mutateDecision.mutate) {
    mutatedSkillFile = finalOutcome === 'accepted';
  }
@@ -23,7 +23,6 @@
 *
 *   history.json
 *   best.md
- *   proposed.md
 *   versions/
 *     v0001_e1_s1.md
 *     v0002_e1_s2.md
@@ -53,10 +52,6 @@ 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');
 }
@@ -176,18 +171,17 @@ export function acceptCandidate(input: AcceptInput): AcceptResult {
 }

 /**
- * 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).
+ * 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).
 */
 export function writeProposed(skillsDir: string, skillName: string, candidateText: string): string {
-  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;
+  const p = bestPath(skillsDir, skillName);
+  fs.mkdirSync(path.dirname(p), { recursive: true });
+  atomicWrite(p, candidateText);
+  return p;
 }

 /**
@@ -63,26 +63,25 @@ 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,
+
+  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,
+      });
+    });
+  },
 };

 export default entry;
@@ -1,67 +0,0 @@
-// #1474: the v0.41.1 wave shipped bench-publish.ts + docs/eval-bench.md
-// advertising `gbrain bench publish`, but the cli.ts dispatcher case was never
-// added — the documented command hit 'Unknown command'. These tests spawn the
-// real CLI (no DB needed; bench publish is pure file I/O) and fail on any
-// regression of the dispatcher wiring.
-import { describe, expect, test } from 'bun:test';
-import { mkdtempSync, writeFileSync, existsSync, rmSync } from 'node:fs';
-import { tmpdir } from 'node:os';
-import { join } from 'node:path';
-import { spawnSync } from 'node:child_process';
-
-function runCli(args: string[]): { stdout: string; stderr: string; code: number } {
-  const result = spawnSync(process.execPath, ['run', 'src/cli.ts', 'bench', ...args], {
-    encoding: 'utf8',
-    cwd: process.cwd(),
-    env: { ...process.env },
-  });
-  return { stdout: result.stdout ?? '', stderr: result.stderr ?? '', code: result.status ?? -1 };
-}
-
-describe('gbrain bench dispatcher (#1474)', () => {
-  test('bench --help reaches bench-publish help without a DB (was: Unknown command)', () => {
-    const { stdout, stderr, code } = runCli(['--help']);
-    expect(stderr).not.toContain('Unknown command');
-    expect(code).toBe(0);
-    expect(stdout).toContain('gbrain bench publish');
-    expect(stdout).toContain('--from');
-  });
-
-  test('unknown bench subcommand exits 2 with usage', () => {
-    const { stderr, code } = runCli(['bogus']);
-    expect(code).toBe(2);
-    expect(stderr).toContain('Unknown bench subcommand');
-    expect(stderr).toContain('bench publish');
-  });
-
-  test('bench publish roundtrip: captured NDJSON in, baseline file out', () => {
-    const tmp = mkdtempSync(join(tmpdir(), 'bench-cli-'));
-    try {
-      const row = {
-        tool_name: 'query',
-        query: 'hello world',
-        retrieved_slugs: ['slug-a'],
-        retrieved_chunk_ids: [1],
-        source_ids: ['default'],
-        expand_enabled: false,
-        detail: 'medium',
-        detail_resolved: 'medium',
-        vector_enabled: true,
-        expansion_applied: false,
-        latency_ms: 100,
-        remote: false,
-        job_id: null,
-        subagent_id: null,
-      };
-      const from = join(tmp, 'captured.ndjson');
-      const to = join(tmp, 'personal.baseline.ndjson');
-      writeFileSync(from, `${JSON.stringify(row)}\n`);
-      const { code, stderr } = runCli(['publish', '--from', from, '--to', to]);
-      expect(stderr).not.toContain('Unknown command');
-      expect(code).toBe(0);
-      expect(existsSync(to)).toBe(true);
-    } finally {
-      rmSync(tmp, { recursive: true, force: true });
-    }
-  });
-});
@@ -15,19 +15,22 @@ import { describe, test, expect } from 'bun:test';
 import { CHUNKER_VERSION } from '../src/core/chunkers/code.ts';

 describe('Layer 12 — CHUNKER_VERSION constant', () => {
-  test('bumped to 4 for Cathedral II', () => {
+  test('bumped to 5 for the estimated-token hard cap', () => {
    // v3: v0.19.0 Chonkie parity (tokenizer + small-sibling merge).
    // v4: v0.20.0 Cathedral II (qualified names + parent scope + doc_comment
    //     + fence extraction + chunk-grain FTS). Folded into content_hash
    //     so any bump forces clean re-chunks on next sync.
-    expect(CHUNKER_VERSION).toBe(4);
+    // v5: estimated-token hard cap on AST-path chunks (capCodeChunks) so
+    //     un-subdividable giant nodes can't overflow strict embedding
+    //     server token limits.
+    expect(CHUNKER_VERSION).toBe(5);
  });

  test('is stable across imports (not recomputed at call time)', async () => {
    const a = (await import('../src/core/chunkers/code.ts')).CHUNKER_VERSION;
    const b = (await import('../src/core/chunkers/code.ts')).CHUNKER_VERSION;
    expect(a).toBe(b);
-    expect(a).toBe(4);
+    expect(a).toBe(5);
  });
 });

@@ -10,8 +10,8 @@ import { describe, test, expect } from 'bun:test';
 import { chunkCodeText, detectCodeLanguage, CHUNKER_VERSION } from '../../src/core/chunkers/code.ts';

 describe('CHUNKER_VERSION', () => {
-  test('v0.20.0 Cathedral II Layer 12 bumped to 4', () => {
-    expect(CHUNKER_VERSION).toBe(4);
+  test('v5: estimated-token hard cap on AST-path chunks', () => {
+    expect(CHUNKER_VERSION).toBe(5);
  });
 });

@@ -135,13 +135,14 @@ describe('Recursive Text Chunker', () => {
 });

 describe('CJK chunking (v0.32.7)', () => {
-  test('MARKDOWN_CHUNKER_VERSION is 3', async () => {
+  test('MARKDOWN_CHUNKER_VERSION is 4', async () => {
    // v0.40.3.0: bumped 2→3 to signal the post-upgrade reembed sweep that
-    // contextual retrieval wrapping is now applied at embed time. Chunk
-    // boundaries themselves are unchanged; the bump forces re-embed for
-    // pages where chunker_version < 3.
+    // contextual retrieval wrapping is now applied at embed time.
+    // v4: estimated-token hard cap + whitespace-word undercount floor
+    // (URL-dense docs produced chunks past strict embedding server token
+    // limits). Boundary change → forces re-chunk for chunker_version < 4.
    const mod = await import('../../src/core/chunkers/recursive.ts');
-    expect(mod.MARKDOWN_CHUNKER_VERSION).toBe(3);
+    expect(mod.MARKDOWN_CHUNKER_VERSION).toBe(4);
  });

  test('long pure-Chinese paragraph splits into multiple chunks', () => {
@@ -0,0 +1,195 @@
+/**
+ * Markdown chunker v4 / code chunker v5 — estimated-token hard cap
+ * regression tests.
+ *
+ * Reproduces a field failure: a local llama-server embedding backend
+ * (`-ub 2048`) crashes deterministically (trace/BPT trap → EOF at the
+ * client) when a single chunk exceeds ~2,050 real tokens. Two content
+ * shapes triggered it:
+ *
+ *   1. Korean docs carrying one long source URL per line.
+ *      The URLs' ASCII mass pushes CJK density below 0.30, flipping
+ *      countCJKAwareWords to whitespace counting, where a 150-char URL
+ *      counts as ONE word → chunks ballooned to 3-4K chars ≈ 2,000+
+ *      real tokens (URL soup tokenizes at ~1.6 chars/token).
+ *
+ *   2. Large JSON code blocks (~7K chars) that the word pipeline
+ *      undercounts the same way (few whitespace tokens).
+ *
+ * The fix: every emitted chunk must satisfy
+ *   estimateEmbeddingTokens(chunk) <= maxTokens (default 1500)
+ * where the estimate deliberately OVERSTATES real tokenizer counts.
+ */
+
+import { describe, test, expect } from 'bun:test';
+import { chunkText, capByEstimatedTokens, DEFAULT_MAX_EST_TOKENS } from '../../src/core/chunkers/recursive.ts';
+import { chunkCodeText } from '../../src/core/chunkers/code.ts';
+import { estimateEmbeddingTokens } from '../../src/core/cjk.ts';
+
+/** Synthesize the failing shape: Korean rollup lines each ending in a long Notion URL. */
+function urlDenseKoreanRollup(lines: number): string {
+  const out: string[] = ['# 링크가 줄마다 붙는 한국어 예시 문서', ''];
+  for (let i = 0; i < lines; i++) {
+    const hex32 = (i * 2654435761 >>> 0).toString(16).padStart(8, '0').repeat(4);
+    out.push(
+      `- **항목 ${i}**: 이 줄은 청커 동작 검증을 위한 의미 없는 한국어 예시 문장입니다 · 전화 000-0000-${String(1000 + i)} · ` +
+      `이메일 user${i}@example.com · 링크: https://docs.example.com/pages/${hex32}?v=abcdef0123456789&ref=sample`,
+    );
+  }
+  return out.join('\n');
+}
+
+/** Synthesize a large pretty-printed JSON block with CJK values. */
+function bigJsonBlock(targetChars: number): string {
+  const entries: string[] = [];
+  let i = 0;
+  let len = 0;
+  while (len < targetChars) {
+    const row =
+      `  "item_${i}": { "name": "예시-${i}", "url": "https://example.com/api/v2/items/${i}?token=abc${i}def", "qty": ${i % 100}, "memo": "한국어 값이 섞인 예시 데이터" }`;
+    entries.push(row);
+    len += row.length;
+    i++;
+  }
+  return `{\n${entries.join(',\n')}\n}`;
+}
+
+describe('v4 estimated-token cap — URL-dense Korean doc (field-failure shape)', () => {
+  test('every chunk stays under the estimated-token cap', () => {
+    const md = urlDenseKoreanRollup(60);
+    const chunks = chunkText(md);
+    expect(chunks.length).toBeGreaterThan(0);
+    for (const c of chunks) {
+      expect(estimateEmbeddingTokens(c.text)).toBeLessThanOrEqual(DEFAULT_MAX_EST_TOKENS);
+    }
+  });
+
+  test('no chunk reaches the measured 3K-char danger zone for URL soup', () => {
+    const md = urlDenseKoreanRollup(60);
+    const chunks = chunkText(md);
+    // 1500 est tokens at the OTHER weight (0.75/char) bounds chunks to
+    // ~2,000 chars for pure ASCII — well under the ~3,300 chars where
+    // URL-dense content crosses ~2,050 real tokens (1.6 chars/token).
+    for (const c of chunks) {
+      expect(c.text.length).toBeLessThanOrEqual(2600);
+    }
+  });
+
+  test('content is preserved (no lines dropped by the cap)', () => {
+    const md = urlDenseKoreanRollup(60);
+    const chunks = chunkText(md);
+    const joined = chunks.map((c) => c.text).join('\n');
+    // Spot-check first / middle / last rollup lines survive chunking.
+    for (const marker of ['항목 0', '항목 30', '항목 59']) {
+      expect(joined).toContain(marker);
+    }
+  });
+});
+
+describe('v4 estimated-token cap — large JSON blocks', () => {
+  test('7K-char pretty JSON through the prose path stays under the cap', () => {
+    const md = `설정 파일 원문 보존:\n\n\`\`\`\n${bigJsonBlock(7000)}\n\`\`\`\n`;
+    const chunks = chunkText(md);
+    expect(chunks.length).toBeGreaterThan(1);
+    for (const c of chunks) {
+      expect(estimateEmbeddingTokens(c.text)).toBeLessThanOrEqual(DEFAULT_MAX_EST_TOKENS);
+    }
+  });
+
+  test('7K-char minified JSON (single whitespace-less token) stays under the cap', () => {
+    const minified = bigJsonBlock(7000).replace(/\n\s*/g, '');
+    const chunks = chunkText(minified);
+    expect(chunks.length).toBeGreaterThan(1);
+    for (const c of chunks) {
+      expect(estimateEmbeddingTokens(c.text)).toBeLessThanOrEqual(DEFAULT_MAX_EST_TOKENS);
+    }
+  });
+
+  test('json fence via the code chunker stays under the cap (+header slack)', async () => {
+    const chunks = await chunkCodeText(bigJsonBlock(7000), 'fence.json');
+    expect(chunks.length).toBeGreaterThan(0);
+    for (const c of chunks) {
+      // buildChunk prepends a short "[JSON] fence.json:…" header AFTER the
+      // body-level cap; allow ~60 est tokens of header slack. Real-token
+      // safety margin (2,050 − overestimated 1,500) absorbs this easily.
+      expect(estimateEmbeddingTokens(c.text)).toBeLessThanOrEqual(DEFAULT_MAX_EST_TOKENS + 60);
+    }
+  });
+});
+
+describe('v4 word-count floor — behavior preserved for normal content', () => {
+  test('Latin prose chunking is unchanged by the floor (avg word < 6 chars)', () => {
+    const prose = Array.from({ length: 120 }, (_, i) =>
+      `This is sentence number ${i} and it talks about ordinary things in plain words.`,
+    ).join(' ');
+    const chunks = chunkText(prose);
+    // Historical behavior: ~1,560 whitespace words → multiple ~300-word chunks.
+    expect(chunks.length).toBeGreaterThan(3);
+    for (const c of chunks) {
+      const words = c.text.split(/\s+/).length;
+      expect(words).toBeLessThanOrEqual(300 * 1.5 + 50); // merge cap + overlap
+    }
+  });
+
+  test('Korean prose (CJK-dense, no URLs) never triggers the token cap', () => {
+    const prose = Array.from({ length: 80 }, (_, i) =>
+      `이 문장은 순수 한국어 산문의 청킹 동작을 확인하기 위한 ${i}번째 예시 문장입니다.`,
+    ).join(' ');
+    const chunks = chunkText(prose);
+    expect(chunks.length).toBeGreaterThan(1);
+    for (const c of chunks) {
+      // CJK-dense chunks are char-counted (≈450 max) — nowhere near 1500.
+      expect(estimateEmbeddingTokens(c.text)).toBeLessThanOrEqual(700);
+    }
+  });
+});
+
+describe('capByEstimatedTokens unit behavior', () => {
+  test('returns input unchanged when under the cap', () => {
+    expect(capByEstimatedTokens('short text', 1500)).toEqual(['short text']);
+    expect(capByEstimatedTokens('', 1500)).toEqual([]);
+  });
+
+  test('prefers newline cut points within the lookback window', () => {
+    const line = 'x'.repeat(100);
+    const text = Array.from({ length: 40 }, () => line).join('\n');
+    const pieces = capByEstimatedTokens(text, 1000);
+    expect(pieces.length).toBeGreaterThan(1);
+    for (const p of pieces) {
+      // Every piece should be whole lines (multiples of the 100-char line).
+      for (const l of p.split('\n')) {
+        expect(l).toBe(line);
+      }
+    }
+  });
+
+  test('makes forward progress on whitespace-less input (hard cut)', () => {
+    const blob = 'a'.repeat(10_000);
+    const pieces = capByEstimatedTokens(blob, 1000);
+    expect(pieces.length).toBeGreaterThan(1);
+    expect(pieces.join('')).toBe(blob);
+    for (const p of pieces) {
+      expect(estimateEmbeddingTokens(p)).toBeLessThanOrEqual(1000);
+    }
+  });
+});
+
+describe('estimateEmbeddingTokens — weight sanity', () => {
+  test('overestimates URL-dense ASCII (0.75/char ≥ measured ~0.63/char)', () => {
+    const url = 'https://docs.example.com/pages/a1b2c3d4e5f6a1b2c3d4e5f6a1b2c3d4?v=abc&ref=sample';
+    const est = estimateEmbeddingTokens(url);
+    expect(est).toBeGreaterThanOrEqual(Math.floor(url.length * 0.7));
+  });
+
+  test('counts CJK at 1 token/char', () => {
+    expect(estimateEmbeddingTokens('가나다라마')).toBe(5);
+  });
+
+  test('whitespace is nearly free', () => {
+    expect(estimateEmbeddingTokens('   \n\t  ')).toBeLessThanOrEqual(1);
+  });
+
+  test('empty string is 0', () => {
+    expect(estimateEmbeddingTokens('')).toBe(0);
+  });
+});
@@ -172,59 +172,6 @@ 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',
@@ -39,7 +39,6 @@ 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';
@@ -742,7 +741,7 @@ describe('skillopt T3 — F11 held-out gate, ablation opts, no-DB-pollution', ()
    } finally { fixture.cleanup(); }
  });

-  test('--no-mutate writes proposed.md and best.md, leaves SKILL.md untouched', async () => {
+  test('--no-mutate writes proposed.md (best.md), leaves SKILL.md untouched', async () => {
    const fixture = setupFixture(SKILL_PEOPLE_ONLY, CITATIONS_BENCHMARK);
    try {
      installStub({
@@ -754,9 +753,10 @@ 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).toBe(proposedPath(fixture.skillsDir, SKILL));
+          expect(result.proposedPath).toBeDefined();
+          // proposed.md (best.md) exists and carries the improvement.
+          expect(fs.existsSync(result.proposedPath!)).toBe(true);
          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');
@@ -803,107 +803,3 @@ 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));
-  });
-});
@@ -309,62 +309,3 @@ describe('isEvalCaptureEnabled / isEvalScrubEnabled (CONTRIBUTOR_MODE-gated)', (
    } finally { restore(); }
  });
 });
-
-describe('DB-plane stash (#1475): GBRAIN_EVAL_CAPTURE / GBRAIN_EVAL_SCRUB_PII', () => {
-  // connectEngine stamps `gbrain config set eval.capture` (DB plane) onto
-  // these env vars because ctx.config is the sync file-plane load. Without
-  // the stash check the DB value was written and never read.
-  const origCapture = process.env.GBRAIN_EVAL_CAPTURE;
-  const origScrub = process.env.GBRAIN_EVAL_SCRUB_PII;
-  const origMode = process.env.GBRAIN_CONTRIBUTOR_MODE;
-  const restore = () => {
-    if (origCapture === undefined) delete process.env.GBRAIN_EVAL_CAPTURE;
-    else process.env.GBRAIN_EVAL_CAPTURE = origCapture;
-    if (origScrub === undefined) delete process.env.GBRAIN_EVAL_SCRUB_PII;
-    else process.env.GBRAIN_EVAL_SCRUB_PII = origScrub;
-    if (origMode === undefined) delete process.env.GBRAIN_CONTRIBUTOR_MODE;
-    else process.env.GBRAIN_CONTRIBUTOR_MODE = origMode;
-  };
-
-  test('stash=true turns capture on when file plane is silent (the #1475 repro)', () => {
-    delete process.env.GBRAIN_CONTRIBUTOR_MODE;
-    process.env.GBRAIN_EVAL_CAPTURE = 'true';
-    try {
-      expect(isEvalCaptureEnabled(null)).toBe(true);
-      // eslint-disable-next-line @typescript-eslint/no-explicit-any
-      const noEval: any = { engine: 'pglite' };
-      expect(isEvalCaptureEnabled(noEval)).toBe(true);
-    } finally { restore(); }
-  });
-
-  test('stash=false wins over CONTRIBUTOR_MODE=1 (explicit per-key beats broad flag)', () => {
-    process.env.GBRAIN_CONTRIBUTOR_MODE = '1';
-    process.env.GBRAIN_EVAL_CAPTURE = 'false';
-    try {
-      expect(isEvalCaptureEnabled(null)).toBe(false);
-    } finally { restore(); }
-  });
-
-  test('file-plane explicit value still wins over the stash', () => {
-    process.env.GBRAIN_EVAL_CAPTURE = 'true';
-    try {
-      // eslint-disable-next-line @typescript-eslint/no-explicit-any
-      const disabled: any = { engine: 'pglite', eval: { capture: false } };
-      expect(isEvalCaptureEnabled(disabled)).toBe(false);
-    } finally { restore(); }
-  });
-
-  test('scrub stash: false disables, file plane wins, default stays true', () => {
-    process.env.GBRAIN_EVAL_SCRUB_PII = 'false';
-    try {
-      expect(isEvalScrubEnabled(null)).toBe(false);
-      // eslint-disable-next-line @typescript-eslint/no-explicit-any
-      const fileWins: any = { engine: 'pglite', eval: { scrub_pii: true } };
-      expect(isEvalScrubEnabled(fileWins)).toBe(true);
-    } finally { restore(); }
-    delete process.env.GBRAIN_EVAL_SCRUB_PII;
-    try {
-      expect(isEvalScrubEnabled(null)).toBe(true);
-    } finally { restore(); }
-  });
-});
@@ -403,77 +403,6 @@ 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.
@@ -302,29 +302,4 @@ describe('loadConfigWithEngine (Phase 4 / F3)', () => {
      expect(merged?.engine).toBe('pglite');
    });
  });
-
-  describe('eval.* DB-plane merge (#1475)', () => {
-    test('gbrain config set eval.capture true reaches the merged config', async () => {
-      // The #1475 repro: DB plane has eval.capture=true, file plane silent.
-      // Pre-fix the merge skipped eval.* entirely and capture never fired.
-      const base: GBrainConfig = { engine: 'pglite' };
-      const engine = makeEngine({ 'eval.capture': 'true', 'eval.scrub_pii': 'false' });
-      const merged = await loadConfigWithEngine(engine, base);
-      expect(merged?.eval?.capture).toBe(true);
-      expect(merged?.eval?.scrub_pii).toBe(false);
-    });
-
-    test('file plane wins per key; DB fills only the gaps', async () => {
-      const base: GBrainConfig = { engine: 'pglite', eval: { capture: false } };
-      const engine = makeEngine({ 'eval.capture': 'true', 'eval.scrub_pii': 'false' });
-      const merged = await loadConfigWithEngine(engine, base);
-      expect(merged?.eval?.capture).toBe(false); // file wins
-      expect(merged?.eval?.scrub_pii).toBe(false); // DB fills the gap
-    });
-
-    test('no eval keys anywhere leaves cfg.eval undefined', async () => {
-      const merged = await loadConfigWithEngine(makeEngine({}), { engine: 'pglite' });
-      expect(merged?.eval).toBeUndefined();
-    });
-  });
 });
@@ -1,62 +0,0 @@
-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([]);
-  });
-});
@@ -1,17 +0,0 @@
-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');
-  });
-});
@@ -186,67 +186,11 @@ 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');
  });
 });
@@ -66,10 +66,6 @@ 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);
  });
@@ -90,44 +86,10 @@ 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', () => {
@@ -218,21 +218,15 @@ describe('progress reporter', () => {
  test('only one process-level signal handler installed across many reporters', () => {
    // Baseline: one handler already installed by prior tests in this file.
    const installedBefore = __signalHandlerInstalledForTest();
-    // liveReporters is process-global: earlier test files in the same shard
-    // can leave a live entry behind (e.g. a production path that skips
-    // finish() on an error branch). Assert NET-zero leak from THIS test's
-    // lifecycles, not an absolute zero we don't control — same tolerance
-    // the handler assertion below already applies via `installedBefore`.
-    const liveBefore = __liveReporterCountForTest();
    const { stream } = sink(false);
    for (let i = 0; i < 50; i++) {
      const p = createProgress({ mode: 'json', stream, minIntervalMs: 0, minItems: 1 });
      p.start(`phase_${i}`, 1);
      p.finish();
    }
-    // After 50 reporter lifecycles, still exactly one handler and zero NEWLY leaked live entries.
+    // After 50 reporter lifecycles, still exactly one handler and zero leaked live entries.
    expect(__signalHandlerInstalledForTest()).toBe(installedBefore || true);
-    expect(__liveReporterCountForTest()).toBe(liveBefore);
+    expect(__liveReporterCountForTest()).toBe(0);
  });

  test('startHeartbeat() fires heartbeats and stop() clears', async () => {
@@ -12,11 +12,9 @@ import {
  bestPath,
  historyPath,
  loadHistory,
-  proposedPath,
  revertAllPending,
  skillPath,
  versionsDir,
-  writeProposed,
 } from '../../src/core/skillopt/version-store.ts';

 let tmpDir: string;
@@ -81,19 +79,6 @@ 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);