brainlisted

# /brain — second-brain health at one glance > 🧒 When reporting to a non-technical operator, end with a short plain-words recap (rule `eli5-always`). Catches SILENT breakage of the memory/RAG stack (the things that kept breaking month after month: the reindex, the local search server, the memory pilot). Read-only, 0 tokens. ## What it does (one script) `python $IMPORTS_ROOT/brain_health.py` Checks 5 things and colors them 🟢/🟡/🔴: 1. **Search server** on its local port — alive or not (if it's down, auto-recall silently disappears); also shows whether the graph-assisted mode is on. 2. **Index** `_brain_e5.npy` — how fresh (🟡 if the reindex lags >48h). 3. **TurnState ledger** — how many turns recorded (memory Phase 1), when the last one was. 4. **Nightly distillation** — candidates in quarantine + when the last run happened. 5. **A/B direct-vs-graph recall** — how many runs + the operator's verdicts (👍/👎). Writes a dashboard `$OBSIDIAN_VAULT/_Dashboards/Brain-Health.html` (the operator reads with their eyes, [[prefer-visual-dashboards]]). Exit code 0/1/2 = ok/warn/red (for scripts). ## When to fix (on 🔴/🟡) - **🔴 search server down** → run the restart script (as admin, see [[always-on-memory-pilot]]); or reboot (the at-logon task brings it up). - **🟡 index lagging** → `gpu_check.py [--kill]`, then `brain_embed_update.py [--wait-gpu 10]` ([[reindex-routine]]). - **🟡 ledger empty** → fine if memory Phase 1 was just enabled (it fills from the next sessions on). ##