Lab · Tuple evaluation

Precision, recall and omission explorer

Withholding uncertain content makes a report more precise and less complete. This explorer makes that cost visible: every total is recomputed from the stored candidate tuples and their authored admission scores as the threshold moves. Nothing is interpolated and no study is being rerun.

Synthetic demonstration mode only. Twenty-four authored reports across two fictional cohorts, generated from seed 42. The owner-supplied manuscript results for this method are not release-ready and are therefore not shown on this site. The evaluator rules match the clinical-tuple-eval repository: one-to-one exact matching over the selected fields, null for zero denominators, report states computed as a union.