The expanded anomaly pipeline is live. API PR #84 activated the in-app five-minute scheduler on August 11, 2026. Production verification at the 20:20, 20:25, and 20:30 UTC boundaries observed exactly one elected runner per boundary, cycle times of 655 ms, 350 ms, and 490 ms, and no errors. The first cycle created 13 continuity incidents from existing evidence gaps; the next two created no duplicate incidents.
Detection semantics
Reliability baselines are source-specific: proxy and OTLP facts are not blended. Error-rate evaluation requires at least 10 baseline requests and 5 recent requests. Latency evaluation requires at least 10 baseline samples, 5 recent samples, and at least 50% duration coverage in both windows. Missing or invalid duration samples do not prevent error-rate evaluation. When latency is eligible,details.latency_statistic is mean_of_exact_5m_bucket_p95. MeshAI computes an exact p95 within each five-minute bucket, then compares the mean of those bucket values across the baseline and recent windows. This is not an exact percentile across all requests in the full window. details also reports source, sample counts, coverage, and whether error-rate and latency evidence were available.

