Prove automation outcomes with evidence you can audit

Completion rate

Successful eligible runs divided by attempted eligible runs, with exclusions documented.

Recovery time

Elapsed time from detection to restored service, with owner and severity recorded.

Cycle time

Start and stop events, comparison window, and manual work are defined before measurement.

Human effort

Review, correction, and exception-handling effort are included—not hidden as automation.

Critical-path coverage

Happy paths, known failure modes, fallbacks, and high-impact transitions receive explicit checks.

Change attribution

Model, prompt, policy, tool, schema, and dataset versions travel with evaluation evidence.

Unit cost

Model, tool, infrastructure, and human-review costs are tied to successful business units.

Trace completeness

Decision inputs, policy result, actor, outcome, and exception state are checked for completeness.