ChaosProbe v2 — C2 replication-rescue campaign (online-boutique)

Raw data for the C2 replication-rescue campaign (workload: online-boutique) of the ChaosProbe v2 pre-registered study, testing V2-H3 (replication rescues availability under node-drain only when replicas do not share the failure domain). 24 node-drain sessions: three cells (r=1 packed, r=3 packed, r=3 anti-affine) x 8 replicate sessions, nominal cross-node fraction f=0.50, capacity-feasible round-robin packed assignment, host-side Locust on the user route. Collected on a strict-clean tree at commit e533d5b; all 24 sessions accepted, untainted, and pass archive_run.py --strict. Registered confirmatory test: ART ANOVA (factors r x mode), both-must-pass conjunction over EndpointSlice trough depth and user-route error rate; verdict CONJUNCTION not met (anti-affine error-rescue strong and significant; depth-rescue below the registered 1-pod margin; both packed-approx-r1 controls pass). Per-hypothesis p-values are uncorrected pending Holm across the confirmatory family. Each session directory carries summary.json, the f-050.json raw, charts/, and an artifact-manifest.json; a root SHA256SUMS verifies the archive. See the frozen pre-registration (10.5281/zenodo.20690836) and the C1 campaign (10.5281/zenodo.20690737).