Monitor prompt iteration near operating FPR

gpt-6-luna, high, test split · 194 positives of 1,643 cases · single sample per case, averaged over 3 runs

Optimized prompt (pAUC@10% 0.88)Baseline (pAUC@10% 0.41)
Monitor prompt iteration near operating FPR00.250.50.751True positive rate00.0250.050.0750.1False positive rate