Clinical Note Audit

Inspect response availability, evidence conditions and reference agreement separately.

Scheduled responses—One first attempt per selected slot
Response coverage—
Pipeline Brier loss—Invalid responses use q = 0.5
Valid-response Brier loss—Conditional on output availability
These are recorded outputs on mock consultations. Lower Brier is better. Contract validity does not establish clinical correctness. References are physician risk grades 1 versus 2–4; results are not pooled with PriMock.

First failed contract check

Paired evidence comparison

Paired by consultation, model and contract. Selecting multiple settings increases response pairs, not the number of independent consultations. No new inferential tests are computed.

Inspect an individual response

Select a row to view the source, generated note and unchanged model response.