Explanation workflows

Audit logs

Use explanation results as records for transparency, accountability, and reconstruction.
View as Markdown

An audit log records the process and context behind a model prediction.

For an Umnai model, the explanation output for a prediction contains the information needed to understand how that prediction was produced. This makes the explanation result useful as an audit record.

Why audit logs matter

Audit logs support transparency, accountability, and compliance.

They provide the information needed to reconstruct and explain a prediction after it has been made. This is important when predictions influence operational workflows, regulated decisions, customer outcomes, or internal review processes.

A useful audit log should capture:

InformationPurpose
Prediction contextShows what was submitted and what the model returned.
Explanation dataShows how the prediction was formed.
Model and dataset identifiersConnects the result to the model and dataset used.
TimestampsSupports traceability over time.
Verification informationSupports integrity and anti-tampering checks.

Results view as an audit log

The Results view produced by a query through an Explanation Structure Model (ESM) already contains the prediction and real-time explanation data.

Because it captures the prediction and the explanation together, the Results view can be stored in a system of record for future reference.

This allows teams to reconstruct and explain the prediction later, without relying only on the final model output.

Verification

A Unique Verification Code, or UVC, is available for explanation results.

The UVC enables anti-tampering verification checks. These checks can be performed locally or through a distributed ledger system, depending on the application’s verification requirements.