# Event E2E suite (maintainable)

Goal: every catalog event is covered by automation that stays cheap when new ids are added.

## Tracks

| Track | Scenario | What it proves | How new events join |
|-------|----------|----------------|---------------------|
| **Inventory audits** | nested in `event_coverage` | Live library vs authored catalog | Add JSON row / live seed; audit fails until matched |
| **A inject coverage** | `event_inject_coverage` / `ec30` | Every **camera-worthy** id can register an interest key + label | Auto via `EventCatalog.*.AuthoredIds` |
| **Live outcome** | `event_live_outcome_smoke` | Real API/Beh + `EventOutcome` confirm (curated) | Add a deterministic harness step when a patch exists |

Full suite: `event_suite` = `event_coverage` + `event_live_outcome_smoke`.
Regression nests `event_suite`.

## Commands

```bash
./scripts/harness-run.sh event_inject_coverage
./scripts/harness-run.sh event_live_outcome_smoke
./scripts/harness-run.sh event_suite
./scripts/harness-run.sh regression
```

Artifact: `IdleSpectator/.harness/event-inject-coverage.tsv`
(`domain`, `id`, `camera`, `result`, `key`, `notes`).

## Design rules

1. **Do not** hand-write one harness step per catalog id for inject - loop `AuthoredIds`.
2. B-only rows (`CreatesInterest=false`, Ambient happiness) are `skip_b`, not failures.
3. Inject coverage ≠ live outcome. Inject proves wiring; live proves confirm predicates.
4. Expand live outcome slowly (family, status apply, happiness apply, …) as wrappers exist.
5. Libraries: only camera-worthy live-seeded ids are injected (can be large; Signal-gated).
6. After each successful inject, remove the key from `InterestRegistry`. The registry caps at 96 candidates; leaving them in causes false `register_failed` on later ids.

## Code

- `EventInjectCoverageHarness.Run` - data-driven register loop
- Assert expect `event_inject_coverage` in `AgentHarness`
- Observe→confirm docs: `event-observation.md`