Atlas Blackbox // incident evidence

The minutes before failure, kept.

Every minute, the recorder keeps a bounded telemetry and event frame. A failure seals the rolling window before it can rotate away. Select an incident, scrub back from the trigger, and open its postmortem when a reviewed record has been published.

Live recorder

Incident replay deck

contacting recorder…

Read-only source: api.atlas-systems.uk/blackbox/incidents

Evidence boundary

A recorder, not a verdict.

01

Record

Telemetry and estate events are captured before a failure, while the evidence still exists.

02

Draft

atlas-postmortem reads sealed incidents locally and writes bounded drafts that remain outside production.

03

Review

Only a reviewed repository change attaches a postmortem to the public incident record.

Honesty rule

Unknown stays unknown.

Telemetry is sampled once a minute. Values between samples are visibly interpolated; event timestamps are not. A quiet graph cannot prove that an unrelated service was healthy.

A published postmortem may conclude that root cause is unresolved. The interface presents the reviewed evidence boundary instead of inventing certainty to fill the panel.