Skip to main content
Operators extend trust to automation in proportion to how completely they can answer four questions after the fact: what happened, why did it happen, what changed, and who authorized it? ChangeGuard AI’s answer is an append-only audit trail that records every step of every remediation, surfaced directly in the dashboard as an activity timeline on each incident.

​
The activity timeline

Expand any incident in the dashboard and the timeline shows the complete lifecycle in order, each entry with a state-aware badge, the actor, detail, and timestamp:
  • Detection - the incident opened, with the symptom
  • Analysis - root cause attached, incident analyzed
  • Queued - a remediation was created (proposed or policy-approved)
  • Approval - who approved: a named human, or agent-policy under Auto (with the policy id and version)
  • Permit - permit_issued when the authority was bound to a short-lived Permit; the Permit’s own events (issued, consumed, used, expired, revoked with its reason) sit on the change record, correlated by Permit id and attempt id
  • Execution - the Executor consumed the Permit and applied the fix
  • Verification - verified, or verify_failed after the 20-minute window
  • Rollback (if any) - rollback_initiated, then rollback_verified, rollback_failed, or rollback_policy_denied
  • Resolution or escalation - resolved with attribution, or escalated to a human
The same data is available as ordered JSON from GET /api/timeline?incident_id=... for export or tooling.

​
Properties of the record

  • Append-only. Events are only ever added, each stamped with actor, timestamp, and detail. Nothing in the product edits or deletes an audit event - state transitions add entries rather than rewriting history.
  • Attribution is total. Every execution traces to a named human approval or to agent-policy, and a policy approval implies the policy’s terms at that moment (namespace, confidence, cap) were satisfied - there is no anonymous path to a cluster write. Every write also traces to one consumed Permit and one attempt id; with the Cloud Executor the Permit id is the AWS session name in your own CloudTrail.
  • Failure is as visible as success. verify_failed, rollback_failed, rollback_policy_denied, and drift refusals are first-class events. The timeline of an escalated incident shows precisely where automation stopped and why it handed control to you.
  • Crash-consistent. Controller crash injection during validation confirmed every lifecycle event lands exactly once - no lost events, no duplicates - so the trail remains trustworthy evidence even across control-plane restarts.

​
Reading a timeline: a failed fix, done right

A healthy escalation reads like this - every hop attributable, terminating with a human:
Audit events and timelines are tenant-scoped like all ChangeGuard AI data. For platform-level authentication and API-access auditing, see Compliance & Audit.