Measure human-agent iteration at message level, preserving human follow-ups in triggered tasks

For interaction-quality audits, classify who produced each message before counting prompts, and inspect the sequence around a correction before calling it rework. A generated task kickoff does not make all later messages in that task generated: real human steering can follow review or Slack-triggered starts. Conversely, a user-role record can contain a child-agent prompt, relay, or completion receipt. Preserve the source records, record exclusion reasons, and separate direct human follow-ups from machine traffic.

Compare the actual request, the preceding agent attempt, the corrective evidence, the changed action, and an independently verified result. Multiple human messages before the next recorded answer are not sufficient evidence of repeated failed attempts. A merge verifies an integration event; it does not verify deployment or customer impact. Frequency and task outcomes require explicit, potentially different populations.

Applies when: Estimating human prompting frequency or corrective iteration from agent telemetry that mixes humans, automated task starts, and delegated work. Keywords: agent telemetry prompts correction iteration provenance triggers followups yield validation