Supersedes #1613 (re-homed from the fork so CI can run with secrets; real names scrubbed for the privacy rule; orphan_ratio --source scoping bundled in).

What ships (two fixes, 3 bisect-friendly commits)

1. bold-name-no-time conversation parser builtin (the 14th)

Parses **Speaker:** text transcripts with no per-line timestamp — the shape Circleback / Granola / Zoom emit. Every prior builtin required a time anchor, so this shape matched nothing: a production brain had 104 conversation pages + 3,423 eligible pages silently extracting zero conversation-facts. This is the unlock for whole-brain conversation-facts extraction.

• Messages anchor at T00:00:00Z of the frontmatter date (no fabricated wall-clock; line order preserves sequence), same convention as irc-classic.
• regex /^\*\*(?!\[)(.+?):\*\*\s*(.*)$/: the colon-inside-bold (not declaration order) prevents shadowing bold-paren-time; the (?!\[) lookahead rejects telegram-bracket **[18:37] Name:** so disabling telegram-bracket yields an honest no_match instead of speaker="[18:37] Name".
• new optional PatternEntry.score_full_body: **Label:** text is a common prose idiom, so a notes page with bold labels clustered in its first 10 lines scored 0.3 on the head pass (NOT < SCORING_HEAD_TRIGGER_THRESHOLD, so the full-body fallback never fired) and cleared the 0.05 floor. parse.ts now recomputes the winner over the full body before the floor, so such a page stays no_match.
• scrubbed pre-existing real names from bold-paren-time test samples (privacy rule).

2. orphan_ratio / find_orphans source scoping

gbrain doctor --source <id> and gbrain orphans --source <id> now scope to that source instead of reporting brain-wide.

• findOrphanPages(opts?: { sourceId?, sourceIds? }) on both engines scopes the candidate set (scalar = $1 / federated = ANY($1::text[])). Cross-source inbound links still count, so a page in X linked from Y is reachable (not an orphan of X).
• Corrected the total_linkable denominator: excluded pages (templates/, scratch/) that have inbound links no longer inflate it and suppress warnings. Changes orphan_ratio output for every brain, in the accurate direction.
• The find_orphans MCP op threads sourceScopeOpts(ctx), closing a cross-source read leak for source-bound OAuth clients (v0.34.1 source-isolation class).
• Under explicit --source below 100 entity pages, orphan_ratio reports the ratio with a low-scale caveat instead of a vacuous "ok". Thin-client doctor --source deferred (TODOS.md).

Review

• /plan-eng-review (cleared) + /codex outside-voice (8 findings, 7 actioned + 1 confirmed-good). Codex caught the score_full_body floor gap, the bracket-timestamp mis-capture, the pre-existing denominator bug, and the find_orphans MCP leak.

Tests

• bun run verify — 29 checks green (typecheck, privacy, conversation-parser eval).
• conversation-parser + back-compat: 135 pass. orphan-area (incl. doctor-orphan-ratio + orphan-reduction E2E): 73 pass. engine-parity on real Postgres (scalar + federated parity): 11 pass.

🤖 Generated with Claude Code