Agent objective
- Interact with the supplied stateful environment.
- Produce verifier-checkable actions or artifacts.
- Maximise scalar reward under the package contract.
A time-boxed production-operations incident in a deterministic Kubernetes-behavior simulator with continuous traffic. Each episode combines latent fault classes and may require rollback, canary-template repair, multi-resource configuration repair, or an evidence-supported no-write hold. Agents must diagnose safely, preserve four stable plus one canary endpoint, and stay within a two-mutation budget.
A time-boxed production-operations incident in a deterministic Kubernetes-behavior simulator with continuous traffic. Each episode combines latent fault classes and may require rollback, canary-template repair, multi-resource configuration repair, or an evidence-supported no-write hold. Agents must diagnose safely, preserve four stable plus one canary endpoint, and stay within a two-mutation budget.
Enough detail to understand the intellectual terrain; generated instances, hidden mechanisms, and solution paths remain inside the private package.
| Environment | Mathematical or technical frontier | Adaptive research problem |
|---|---|---|
| Diagnosis and provenance | Kubernetes objects, logs, metrics, releases | Correlate two interacting causes and verify approved images, configuration digests, ABI, init, and evidence coverage. |
| Business behavior | Canary and shared-Service semantics | Validate quotes, refunds, thresholds, rounding, jurisdictions, coupons, malformed requests, and every backend. |
| Lifecycle and rollout | Probes, drain, surge, disruption | Separate startup/readiness/liveness, preserve zero-unavailable rollout, survive restart and node eviction, and verify convergence. |
| Security and resources | Pod shape and least privilege | Preserve non-root, read-only, capability, token, volume, quota, and topology contracts. |
| Minimal evidence-based change | Two-mutation safety budget | Use diagnostics and dry runs, alter only justified resources, and submit exact causes, evidence, changes, and decision. |
A recorded GPT-5.6 Pro solution finished in 33 of 45 minutes and earned 1.000. It passed all 23 checks, including 37 hidden canary assertions and 185 shared-Service assertions, with one of two allowed mutations and no critical safety violation. Reference calibration reaches 1.0 across 64 seeds while inappropriate blind rollback remains low. This is a simulator, not a live cluster, and hosted scoring is the stronger trust boundary.
We publish aggregate behavior and task structure, while withholding generated instances, hidden labels, exact successful probes, private checks, and solution trajectories.
Shown with its provenance and limitations; it is not a performance guarantee.
Solved in 33 minutes. The report records 37/37 hidden canary assertions and 185/185 hidden service assertions passed.
As identified by the supplied artifact.
1 reported runs.
notes.md
Machine-readable provenance and the exact displayed metric are available in results.json.
The paid ZIP will live in a private R2 bucket. Vercel authorizes the buyer and issues a 2–5 minute object URL; R2 serves the bytes directly.
Authenticated buyer + entitlement check
+ private R2 object + 2–5 minute signed URL
= direct, auditable download
Package SHA-256
3d5a45484a48d74ad0228403c6983bc434de6ef88a0ff61fa8c62b4ac7074781One purchase licenses this identified item to one legal organisation for worldwide, perpetual commercial model training, evaluation, research and development. Redistribution and resale of the package are not permitted.