Latency and availability anomaly detection without thresholds

Thresholds are guesses that go stale. PulseTrace learns what normal looks like for each monitor and signal — latency by region, availability, flapping — and opens an anomaly episode when a value steps outside its own baseline.

Baselines are per monitor and per signal, so a chatty endpoint's normal never hides a regression on a quiet one. An anomaly is tracked as one episode from open to close, with its score and confidence, rather than a stream of point alerts; when it ends you get one recovery, not fifty.

Latency baselines are on Team; Pro adds availability and per-signal detection plus flapping detection, and sensitivity is adjustable per monitor for the cases where you know better than the model.

• Per-signal baselines — latency, availability, flapping.
• Episodes, not noise with score, confidence and severity.
• Tunable sensitivity per monitor.
• Incident rules can open incidents from anomalies.