Rule-based tiers, retrospective model evaluation, and (below) live triage.
The saved triage model (e.g. random forest) is run on every case currently in the database — same implementation as live triage (triage_bundle.joblib). Tiers are computed on demand.
Same pipeline as scripts/test_triage.py / train_triage_model.py:
rule-based priority scores from triage_cases are binned into tiers; a random forest or shallow tree
learns structured tags to predict those tiers on a held-out 20% test set (80% train), with a fixed split so
results are comparable between runs.
Labels are quantile bins of normalized rule scores (low / medium / high). This is a research view of how well extracted features recover the current rules—not investigator ground truth.
Paste one or more case narratives below. Start each case with Case N : case summary; the next Case line starts the following case. You can also paste a single narrative with no headers; it will be treated as one case.
Case 1 : First narrative text…
Case 2 : Second narrative…
Case 3 : …