Signal to incident in four moves.
No configuration. The engine evaluates every project, every minute — and only promotes what crosses a threshold.
Six heuristics, 60s cycle
New high-volume groups, uptime failures, error-rate spikes, latency anomalies, resource saturation, deploy correlation — evaluated every minute.
Same key, same incident
Signals sharing an environment, top frame, host, monitor, route, or release SHA merge into one incident instead of paging you six times.
Deterministic likely cause
Backtrace overlap, time correlation, and similar prior incidents feed a signal-weighted confidence score. No LLM involved.
AI only when you ask
The evidence panel is complete without AI. Want prose? Hit Summarize — it runs behind an explicit action, never automatically.
Six heuristics. Real thresholds.
Each condition is documented, deterministic, and tuned to fire on incidents — not noise. Uptime failures enter as critical; everything else is scored by severity.
- Evaluated on a 60-second cycle, per project
- Manual incidents for what the heuristics don’t catch
- Open → Acknowledged → Resolved lifecycle
A confidence score, not a guess.
The likely-cause panel runs the moment evidence loads. Each contributing signal is shown with its weight — you can audit the score, not just trust it.
- Signal-weighted confidence, contributing signals shown
- AI summary only behind an explicit Summarize → action
Evidence merges. Pages don’t multiply.
Six merge keys decide whether a new signal joins an open incident or opens a new one. One bad deploy is one incident — not an error page, an uptime page, and a latency page.
- Keys: environment, top frame, host, monitor, route, release SHA
- Evidence timeline keeps every merged signal inspectable