observability-scribelisted

# Observability Scribe Produces **`lab-notes/observability/observability-contract.md`**: what the harness emits for humans, transcripts, stream-json, and automation—grounded in code, not wishlists. ## When to use - Create or update `lab-notes/observability/observability-contract.md` - Inventory `KERNEL_EVENT_TYPES` and `STOP_REASONS` - Document stream-json / `--output-format` shapes - Map autopsy (`loop-autopsy.js`) and ledger fields for replay tooling ## When not to use - Policy citations (**anthropic-official**) - Parity adopt/skip matrix (**parity-archivist**) - Live bench runs (**oauth-evidence**) ## Charter Read [`lab-notes/agents/CHARTER.md`](../../lab-notes/agents/CHARTER.md). ## File ownership | File | Scope | | ---- | ----- | | `lab-notes/observability/observability-contract.md` | Event catalog, shapes, examples | ## Read first 1. `src/runner/kernel/contract.js` — canonical `STOP_REASONS`, `KERNEL_EVENT_TYPES` 2. `src/runner/event-bus.js` — validation against kernel types 3. `src/runner/human-log.js`, `src/runner/transcript.js` 4. `src/runner/loop-autopsy.js`, `src/runner/session-ledger.js` 5. Tests: `test/runner/harness-architecture.test.js`, `test/runner/session-ledger.test.js`, `test/runner/loop-autopsy.test.js` 6. `docs/runner-quickstart.html` (output flags section) ## Workflow 1. Extract enums from `contract.js` verbatim (do not paraphrase stop reason strings). 2. For each `KERNEL_EVENT_TYPE`, document: emitter, payload shape, redaction rules, strea