Mickai Training Academy: Merebrook policy assistant exercise

All records, queries, policies and results are fictional. No patient data or model service is used.
Use Python 3: python evaluate.py
Or tally queries.csv against candidate_results.csv by hand.
The expected column is a practice answer key, not an exam bank. Keep it separate from any genuine evaluation candidate.
The candidate results are deliberately authored illustrative dispositions, not measured model output.
The script checks only case identity and disposition accuracy. It does not validate prose, clinical safety, permissions, identity or real handover delivery.
Expected: 52/60 correct, HOLD, eight errors; claim support and handover delivery unmeasured.
For a new practice run, create another results CSV and pass its path as the first argument. Never place real patient information in this exercise.
The Python script writes no files and makes no network requests.
