PinCite — evaluation results
PinCiteEvaluation
Measure the review system against labelled examples.
About this evaluation
Pilot-scale labelled set based on public briefs and court orders. It is not the full SPEC.md §12 target set, and results describe only the referenced persisted jobs; no pipeline work is rerun while evaluating.
Labelled citations: 14 across 2 briefs
Per-check precision and recall
| Check | Precision | Recall | F1 | Labelled items |
|---|---|---|---|---|
| existence | 50.0% | 33.3% | — | 14 |
| proposition | — | 0.0% | — | 12 |
| quote | 0.0% | 0.0% | — | 12 |
Retrieval ablation
Where available, this compares proposition-judge accuracy under each retrieval mode.
Retrieval ablation has not been recorded for this pilot yet.