Definition
The proportion of audit trail entries generated by the agentic platform that would materially inform a governance, compliance, or debugging decision, versus entries that are procedural noise.
The Measurement Gap
There is no metric that tells you whether the agentic audit trail is useful for governance decisions or merely voluminous, which means audit completeness can increase while audit utility decreases.
Formula
Plain language: Sample a representative set of audit trail entries. Classify each as "actionable" (would change a decision if reviewed) or "procedural" (routine logging of expected behavior). Calculate the ratio. An audit trail that's 95% procedural noise and 5% actionable signal is functionally useless even if it's technically complete.
Notation:
ATAR = A_actionable / A_sampled × 100
Variables:
- A_actionable = audit trail entries in the sample classified as actionable (count, MANUAL: requires auditor classification)
- A_sampled = total entries in the review sample (count, MANUAL)
Classification criteria for "actionable":
- Entry records a decision that deviated from expected workflow (e.g., agent chose an unusual approach, checkpoint was overridden, security scan found and auto-resolved an issue)
- Entry records a parameter or configuration change that altered agent behavior
- Entry records an error, retry, or fallback to a different agent or model
- Entry would cause an auditor to ask a follow-up question
Classification criteria for "procedural":
- Entry records a routine step executing as expected (e.g., "code agent started," "test agent completed," "checkpoint approved")
- Entry contains no deviation, decision, or exception information
- Entry would not prompt any auditor question or action
Proposed formula, pending validation pilots.
At a Glance
Decision Supported
Is the audit trail a useful governance tool, or has it become a compliance artifact that nobody can meaningfully review because signal is buried in noise?