# Chaos Harness

## Scenarios
- Node churn (random kill/restart orchestrator replicas).
- Message duplication and reordering on HMMM.
- License server throttling (grace + degrade paths).
- Index swap mid-formation (pin via content hash).

## Metrics to Watch
- state.advance_latency{region}
- retry_count{transition}
- quarantine_rate
- recovery_time_from_churn
- pending_transitions