Datasets

Trace-derived refund and return exceptions collected across the incident week to validate the recovery candidate against realistic edge cases.

Premium billing and chargeback-support cases used to validate escalation behavior and revenue-risk handling.

Failure cases where policy-heavy outputs drifted, over-fit, or hallucinated unsupported rules.

Shipping and logistics traces used to measure timeout recovery, latency, and operator-visible customer experience.

Executive-review cohort connecting customer-visible issues to budgets, SLAs, and experiment decisions.