Impact
Impact Analysis
Impact is computed from evidence bags and exact ID joins. Nothing is estimated from titles, timestamps alone, or similarity heuristics.
Evidence only
Analysis pulls reliability facts (monitors, alerts) and telemetry aggregates (traces, logs, browser) only when correlation IDs are present. Each dimension reports whether data is available and why not when it is missing.
Correlation rules
- Allowed:
trace_id,span_id,session_id,request_id/correlation_idwhen persisted. - Forbidden: timestamp-only joins, title/URL similarity, inventing catalog links from monitor names.
Related: Correlation.
Time window
For incidents, the default window is started → resolved (or now if still open), clamped by a hard maximum window. Responses include whether the requested range was clamped.
Unavailable vs zero
Missing user-impact evidence is unavailable, never presented as zero affected users. Quiet telemetry is empty, not a fake all-clear score.
Limitations
- Alerts without attached telemetry IDs still return reliability facts; telemetry dimensions stay unavailable.
- ClickHouse or store failures mark that dimension unavailable — they do not invent spans.