Goal
Deliver a judge-ready live-cluster proof that uses observed Argus/Falco evidence, an explicitly approved Phoenix/Chaos Mesh fault, verified Kubernetes recovery, continuous availability probes, and Sentinel cross-agent correlation.
Acceptance criteria
- Read-only preflight validates the k3s security, chaos, agent, and SOG stack.
- Live execution requires exact-context and explicit fault confirmation.
- All mutable resources are isolated to a disposable namespace.
- One of two replicas is targeted; replacement and full readiness are verified.
- Availability and recovery time are measured, not inferred.
- Evidence provenance distinguishes observed/live chaos from replay/simulator.
- One command, tests, cleanup behavior, and judge documentation are included.
- The final video phase is tracked separately with a frame-by-frame two-minute plan.
Goal
Deliver a judge-ready live-cluster proof that uses observed Argus/Falco evidence, an explicitly approved Phoenix/Chaos Mesh fault, verified Kubernetes recovery, continuous availability probes, and Sentinel cross-agent correlation.
Acceptance criteria