Loading documentation
Autonomous systems / 05
Evidence / Correlated
Operational evidence links mission intent to agent reasoning, human approvals, platform commands, telemetry acknowledgements, and outcomes without collapsing them into a generic activity log.
Every event should answer who, what, why, under which policy, and with what effect.
Health metrics must reflect mission risk, not only infrastructure availability.
An alert is useful only when it names the required operational response.
Classify whether the condition affects safety, authority, mission outcome, or service quality.
Attach mission context, current platform state, last trusted evidence, and recommended containment.
Notify the role that owns the decision, with a separate technical owner where needed.
Escalate on missed acknowledgement; do not treat delivery as resolution.
Replay is an analytical tool, not permission to issue the same external action twice.
Freeze the original release, policies, inputs, model references, and event order.
Reproduce deterministic stages and mark non-deterministic agent outputs explicitly.
Stub or isolate external effects while preserving their recorded acknowledgements.
Compare expected and observed decisions, interventions, timing, and mission outcomes.