PRESERVED SYNTHETIC EVALUATION - not rerun here
Developer classification: 9/20 (45.0%)
Developer keyword baseline: 90.0%
Independent classification: 15/40 (37.5%)
Independent keyword baseline: 24/40 (60.0%)
Answered 8; abstained 31; rejected 1
Duplicate pairs: TP 7; FP 1; missed 1
Duplicate precision 87.5%; recall 87.5%
Nearest positive query matches: 15/16
Broad classification was worse than keyword baselines.
No real-world benchmark or human time saving measured.