Decision log

The AI-specific record that captures, for each consequential model output, the inputs the model saw, the model version and configuration, the output produced, and where available the confidence or salient features. It is the record that lets a reconstruction establish not just what the system did but why the model produced the output it did.

A living record of the reasoning behind an organization's major AI decisions (why this vendor, what was tried and killed, how the rollout was sequenced), which converts a leader's private judgment into organizational memory and is the highest-value key-person-risk mitigation available at small scale.

A short, dated record of the significant governance decisions and who made them, held near the front of the dossier, so a reader sees not only what the evidence is but the sequence of judgment that produced it. It is what turns a static evidence file into an account of a functioning governance process over time.

The per-decision record within an audit trail, capturing the AI's recommendation, the human's action (including overrides), the human's identity, and, on high-impact decisions, the rationale. The override field is what actually proves human-in-the-loop.