> ## Documentation Index
> Fetch the complete documentation index at: https://docs.doblier.io/llms.txt
> Use this file to discover all available pages before exploring further.

# Chaos Test Suite (Gate 6)

> Seven mandatory fault experiments against production-sized state — recovery is measured, never assumed.

One SRE week embedded inside shadow mode. Seven faults are injected against **production-sized state**; documented recovery times become the on-call runbook. **All 7 must pass**, signed off by the SRE lead — this is go-live Gate 6. No ORCA deployment has skipped it.

| # | Experiment                | Pass criteria                    |
| - | ------------------------- | -------------------------------- |
| 1 | Kill Redis master         | Failover \< 10s, zero event loss |
| 2 | Kill Neo4j replica        | Enrichment continues, +\< 100ms  |
| 3 | ServiceNow webhook outage | Fallback poll \< 60s             |
| 4 | Crash Flink task manager  | Checkpoint recovery \< 60s       |
| 5 | Take Schema Registry down | Local cache serves 30 min        |
| 6 | Kill ClickHouse node      | HA failover, 0 records lost      |
| 7 | Kill Kafka broker         | Rebalance \< 30s, 0 message loss |

<Warning>Vendor failover claims (e.g., Redis "\< 5s") are never accepted at face value — they are verified against your own dedup-state size and client configuration before being trusted in production. If a vendor number survives contact with your state, it earns its place in your runbook; if not, you find out in shadow mode, not at 2 a.m.</Warning>

The injection scripts, preconditions, and evidence-capture procedure ship with the release package and are executed jointly with your SRE team during the embedded week.

<Snippet file="cta-architect.mdx" />