{ "f1_window": 0.04, "calibration_role": "ntdb/artifacts/experiments/routing_joint_v3_mmbert_static_actionable_benefit_seed42/split_indices/probability_calibration.npy", "policy_selection_role": "ntdb/artifacts/experiments/routing_joint_v3_mmbert_static_actionable_benefit_seed42/split_indices/policy_selection.npy", "evaluation_role": "unified_validation", "modes": { "l2_only": { "aggregation": "max", "f1_window": 0.04, "calibration_operating_points": { "best_f1": { "document_risk_margin_threshold": -0.9776157913729548, "macro_f1": 0.7736784202980337, "fpr": 0.28846153846153844, "fnr": 0.05627705627705628, "tp": 218, "fp": 30, "tn": 74, "fn": 13 }, "best_fpr_in_f1": { "document_risk_margin_threshold": 0.38240621984004974, "macro_f1": 0.7522394642414904, "fpr": 0.14423076923076922, "fnr": 0.17316017316017315, "tp": 191, "fp": 15, "tn": 89, "fn": 40 }, "best_fnr_in_f1": { "document_risk_margin_threshold": -0.9958261963911355, "macro_f1": 0.7393551718368689, "fpr": 0.4326923076923077, "fnr": 0.03896103896103896, "tp": 222, "fp": 45, "tn": 59, "fn": 9 } }, "sweep_threshold_count": 337, "policy_selection": { "best_f1": { "document_risk_margin_threshold": -0.9776157913729548, "macro_f1": 0.7736784202980337, "fpr": 0.28846153846153844, "fnr": 0.05627705627705628, "tp": 218, "fp": 30, "tn": 74, "fn": 13, "evaluation_metrics": { "macro_f1": 0.7293190887928167, "fpr": 0.36538461538461536, "fnr": 0.07327586206896551, "tp": 215, "fp": 38, "tn": 66, "fn": 17 } }, "best_fpr_in_f1": { "document_risk_margin_threshold": 0.38240621984004974, "macro_f1": 0.7522394642414904, "fpr": 0.14423076923076922, "fnr": 0.17316017316017315, "tp": 191, "fp": 15, "tn": 89, "fn": 40, "evaluation_metrics": { "macro_f1": 0.7532577189908589, "fpr": 0.16346153846153846, "fnr": 0.21982758620689655, "tp": 181, "fp": 17, "tn": 87, "fn": 51 } }, "best_fnr_in_f1": { "document_risk_margin_threshold": -0.9958261963911355, "macro_f1": 0.7393551718368689, "fpr": 0.4326923076923077, "fnr": 0.03896103896103896, "tp": 222, "fp": 45, "tn": 59, "fn": 9, "evaluation_metrics": { "macro_f1": 0.7071852920549851, "fpr": 0.4519230769230769, "fnr": 0.05172413793103448, "tp": 220, "fp": 47, "tn": 57, "fn": 12 } } }, "evaluation": { "unified_validation": { "best_f1": { "document_risk_margin_threshold": -0.9776157913729548, "macro_f1": 0.7736784202980337, "fpr": 0.28846153846153844, "fnr": 0.05627705627705628, "tp": 218, "fp": 30, "tn": 74, "fn": 13, "evaluation_metrics": { "macro_f1": 0.7563200679963783, "fpr": 0.35947712418300654, "fnr": 0.07807570977917981, "tp": 1169, "fp": 220, "tn": 392, "fn": 99 } }, "best_fpr_in_f1": { "document_risk_margin_threshold": 0.38240621984004974, "macro_f1": 0.7522394642414904, "fpr": 0.14423076923076922, "fnr": 0.17316017316017315, "tp": 191, "fp": 15, "tn": 89, "fn": 40, "evaluation_metrics": { "macro_f1": 0.7689902712116136, "fpr": 0.17973856209150327, "fnr": 0.1892744479495268, "tp": 1028, "fp": 110, "tn": 502, "fn": 240 } }, "best_fnr_in_f1": { "document_risk_margin_threshold": -0.9958261963911355, "macro_f1": 0.7393551718368689, "fpr": 0.4326923076923077, "fnr": 0.03896103896103896, "tp": 222, "fp": 45, "tn": 59, "fn": 9, "evaluation_metrics": { "macro_f1": 0.7359094556827233, "fpr": 0.4673202614379085, "fnr": 0.0473186119873817, "tp": 1208, "fp": 286, "tn": 326, "fn": 60 } } } }, "operating_points": { "best_f1": { "threshold": -0.9776157913729548, "threshold_semantics": "best_non_default_class_score_minus_default_class_score", "metrics": { "document_risk_margin_threshold": -0.9776157913729548, "macro_f1": 0.7736784202980337, "fpr": 0.28846153846153844, "fnr": 0.05627705627705628, "tp": 218, "fp": 30, "tn": 74, "fn": 13 }, "policy_selection_metrics": { "macro_f1": 0.7293190887928167, "fpr": 0.36538461538461536, "fnr": 0.07327586206896551, "tp": 215, "fp": 38, "tn": 66, "fn": 17 }, "evaluation_metrics": { "unified_validation": { "macro_f1": 0.7563200679963783, "fpr": 0.35947712418300654, "fnr": 0.07807570977917981, "tp": 1169, "fp": 220, "tn": 392, "fn": 99 } } }, "best_fpr_in_f1": { "threshold": 0.38240621984004974, "threshold_semantics": "best_non_default_class_score_minus_default_class_score", "metrics": { "document_risk_margin_threshold": 0.38240621984004974, "macro_f1": 0.7522394642414904, "fpr": 0.14423076923076922, "fnr": 0.17316017316017315, "tp": 191, "fp": 15, "tn": 89, "fn": 40 }, "policy_selection_metrics": { "macro_f1": 0.7532577189908589, "fpr": 0.16346153846153846, "fnr": 0.21982758620689655, "tp": 181, "fp": 17, "tn": 87, "fn": 51 }, "evaluation_metrics": { "unified_validation": { "macro_f1": 0.7689902712116136, "fpr": 0.17973856209150327, "fnr": 0.1892744479495268, "tp": 1028, "fp": 110, "tn": 502, "fn": 240 } } }, "best_fnr_in_f1": { "threshold": -0.9958261963911355, "threshold_semantics": "best_non_default_class_score_minus_default_class_score", "metrics": { "document_risk_margin_threshold": -0.9958261963911355, "macro_f1": 0.7393551718368689, "fpr": 0.4326923076923077, "fnr": 0.03896103896103896, "tp": 222, "fp": 45, "tn": 59, "fn": 9 }, "policy_selection_metrics": { "macro_f1": 0.7071852920549851, "fpr": 0.4519230769230769, "fnr": 0.05172413793103448, "tp": 220, "fp": 47, "tn": 57, "fn": 12 }, "evaluation_metrics": { "unified_validation": { "macro_f1": 0.7359094556827233, "fpr": 0.4673202614379085, "fnr": 0.0473186119873817, "tp": 1208, "fp": 286, "tn": 326, "fn": 60 } } } } }, "l3_only": { "aggregation": "max", "f1_window": 0.04, "calibration_operating_points": { "best_f1": { "document_risk_margin_threshold": -0.13476252555847168, "macro_f1": 0.9841403339027968, "fpr": 0.019230769230769232, "fnr": 0.0, "tp": 231, "fp": 2, "tn": 102, "fn": 0 }, "best_fpr_in_f1": { "document_risk_margin_threshold": 0.31495702266693115, "macro_f1": 0.9798308999503854, "fpr": 0.0, "fnr": 0.025974025974025976, "tp": 225, "fp": 0, "tn": 104, "fn": 6 }, "best_fnr_in_f1": { "document_risk_margin_threshold": -0.13476252555847168, "macro_f1": 0.9841403339027968, "fpr": 0.019230769230769232, "fnr": 0.0, "tp": 231, "fp": 2, "tn": 102, "fn": 0 } }, "sweep_threshold_count": 337, "policy_selection": { "best_f1": { "document_risk_margin_threshold": -0.13476252555847168, "macro_f1": 0.9841403339027968, "fpr": 0.019230769230769232, "fnr": 0.0, "tp": 231, "fp": 2, "tn": 102, "fn": 0, "evaluation_metrics": { "macro_f1": 0.9970136144049189, "fpr": 0.009615384615384616, "fnr": 0.0, "tp": 232, "fp": 1, "tn": 103, "fn": 0 } }, "best_fpr_in_f1": { "document_risk_margin_threshold": 0.31495702266693115, "macro_f1": 0.9798308999503854, "fpr": 0.0, "fnr": 0.025974025974025976, "tp": 225, "fp": 0, "tn": 104, "fn": 6, "evaluation_metrics": { "macro_f1": 0.9786755075124012, "fpr": 0.0, "fnr": 0.034482758620689655, "tp": 224, "fp": 0, "tn": 104, "fn": 8 } }, "best_fnr_in_f1": { "document_risk_margin_threshold": -0.13476252555847168, "macro_f1": 0.9841403339027968, "fpr": 0.019230769230769232, "fnr": 0.0, "tp": 231, "fp": 2, "tn": 102, "fn": 0, "evaluation_metrics": { "macro_f1": 0.9970136144049189, "fpr": 0.009615384615384616, "fnr": 0.0, "tp": 232, "fp": 1, "tn": 103, "fn": 0 } } }, "evaluation": { "unified_validation": { "best_f1": { "document_risk_margin_threshold": -0.13476252555847168, "macro_f1": 0.9841403339027968, "fpr": 0.019230769230769232, "fnr": 0.0, "tp": 231, "fp": 2, "tn": 102, "fn": 0, "evaluation_metrics": { "macro_f1": 0.929221185069775, "fpr": 0.07679738562091504, "fnr": 0.04258675078864353, "tp": 1214, "fp": 47, "tn": 565, "fn": 54 } }, "best_fpr_in_f1": { "document_risk_margin_threshold": 0.31495702266693115, "macro_f1": 0.9798308999503854, "fpr": 0.0, "fnr": 0.025974025974025976, "tp": 225, "fp": 0, "tn": 104, "fn": 6, "evaluation_metrics": { "macro_f1": 0.922069102861841, "fpr": 0.04738562091503268, "fnr": 0.0749211356466877, "tp": 1173, "fp": 29, "tn": 583, "fn": 95 } }, "best_fnr_in_f1": { "document_risk_margin_threshold": -0.13476252555847168, "macro_f1": 0.9841403339027968, "fpr": 0.019230769230769232, "fnr": 0.0, "tp": 231, "fp": 2, "tn": 102, "fn": 0, "evaluation_metrics": { "macro_f1": 0.929221185069775, "fpr": 0.07679738562091504, "fnr": 0.04258675078864353, "tp": 1214, "fp": 47, "tn": 565, "fn": 54 } } } }, "operating_points": { "best_f1": { "threshold": -0.13476252555847168, "threshold_semantics": "best_non_default_class_score_minus_default_class_score", "metrics": { "document_risk_margin_threshold": -0.13476252555847168, "macro_f1": 0.9841403339027968, "fpr": 0.019230769230769232, "fnr": 0.0, "tp": 231, "fp": 2, "tn": 102, "fn": 0 }, "policy_selection_metrics": { "macro_f1": 0.9970136144049189, "fpr": 0.009615384615384616, "fnr": 0.0, "tp": 232, "fp": 1, "tn": 103, "fn": 0 }, "evaluation_metrics": { "unified_validation": { "macro_f1": 0.929221185069775, "fpr": 0.07679738562091504, "fnr": 0.04258675078864353, "tp": 1214, "fp": 47, "tn": 565, "fn": 54 } } }, "best_fpr_in_f1": { "threshold": 0.31495702266693115, "threshold_semantics": "best_non_default_class_score_minus_default_class_score", "metrics": { "document_risk_margin_threshold": 0.31495702266693115, "macro_f1": 0.9798308999503854, "fpr": 0.0, "fnr": 0.025974025974025976, "tp": 225, "fp": 0, "tn": 104, "fn": 6 }, "policy_selection_metrics": { "macro_f1": 0.9786755075124012, "fpr": 0.0, "fnr": 0.034482758620689655, "tp": 224, "fp": 0, "tn": 104, "fn": 8 }, "evaluation_metrics": { "unified_validation": { "macro_f1": 0.922069102861841, "fpr": 0.04738562091503268, "fnr": 0.0749211356466877, "tp": 1173, "fp": 29, "tn": 583, "fn": 95 } } }, "best_fnr_in_f1": { "threshold": -0.13476252555847168, "threshold_semantics": "best_non_default_class_score_minus_default_class_score", "metrics": { "document_risk_margin_threshold": -0.13476252555847168, "macro_f1": 0.9841403339027968, "fpr": 0.019230769230769232, "fnr": 0.0, "tp": 231, "fp": 2, "tn": 102, "fn": 0 }, "policy_selection_metrics": { "macro_f1": 0.9970136144049189, "fpr": 0.009615384615384616, "fnr": 0.0, "tp": 232, "fp": 1, "tn": 103, "fn": 0 }, "evaluation_metrics": { "unified_validation": { "macro_f1": 0.929221185069775, "fpr": 0.07679738562091504, "fnr": 0.04258675078864353, "tp": 1214, "fp": 47, "tn": 565, "fn": 54 } } } } }, "union_l2_l3": { "aggregation": "max", "f1_window": 0.04, "calibration_operating_points": { "best_f1": { "document_risk_margin_threshold": -0.7704736590385437, "macro_f1": 0.8881845323533561, "fpr": 0.10576923076923077, "fnr": 0.06060606060606061, "tp": 217, "fp": 11, "tn": 93, "fn": 14 }, "best_fpr_in_f1": { "document_risk_margin_threshold": 0.6007874235510826, "macro_f1": 0.8734210789256979, "fpr": 0.08653846153846154, "fnr": 0.10822510822510822, "tp": 206, "fp": 9, "tn": 95, "fn": 25 }, "best_fnr_in_f1": { "document_risk_margin_threshold": -0.9776157913729548, "macro_f1": 0.8594848792230071, "fpr": 0.23076923076923078, "fnr": 0.05194805194805195, "tp": 219, "fp": 24, "tn": 80, "fn": 12 } }, "sweep_threshold_count": 337, "policy_selection": { "best_f1": { "document_risk_margin_threshold": -0.7704736590385437, "macro_f1": 0.8881845323533561, "fpr": 0.10576923076923077, "fnr": 0.06060606060606061, "tp": 217, "fp": 11, "tn": 93, "fn": 14, "evaluation_metrics": { "macro_f1": 0.8991596445533936, "fpr": 0.10576923076923077, "fnr": 0.07327586206896551, "tp": 215, "fp": 11, "tn": 93, "fn": 17 } }, "best_fpr_in_f1": { "document_risk_margin_threshold": 0.6007874235510826, "macro_f1": 0.8734210789256979, "fpr": 0.08653846153846154, "fnr": 0.10822510822510822, "tp": 206, "fp": 9, "tn": 95, "fn": 25, "evaluation_metrics": { "macro_f1": 0.8932623821471027, "fpr": 0.07692307692307693, "fnr": 0.11206896551724138, "tp": 206, "fp": 8, "tn": 96, "fn": 26 } }, "best_fnr_in_f1": { "document_risk_margin_threshold": -0.9776157913729548, "macro_f1": 0.8594848792230071, "fpr": 0.23076923076923078, "fnr": 0.05194805194805195, "tp": 219, "fp": 24, "tn": 80, "fn": 12, "evaluation_metrics": { "macro_f1": 0.8441721159673253, "fpr": 0.3076923076923077, "fnr": 0.06465517241379311, "tp": 217, "fp": 32, "tn": 72, "fn": 15 } } }, "evaluation": { "unified_validation": { "best_f1": { "document_risk_margin_threshold": -0.7704736590385437, "macro_f1": 0.8881845323533561, "fpr": 0.10576923076923077, "fnr": 0.06060606060606061, "tp": 217, "fp": 11, "tn": 93, "fn": 14, "evaluation_metrics": { "macro_f1": 0.8418848849143761, "fpr": 0.2107843137254902, "fnr": 0.07570977917981073, "tp": 1172, "fp": 129, "tn": 483, "fn": 96 } }, "best_fpr_in_f1": { "document_risk_margin_threshold": 0.6007874235510826, "macro_f1": 0.8734210789256979, "fpr": 0.08653846153846154, "fnr": 0.10822510822510822, "tp": 206, "fp": 9, "tn": 95, "fn": 25, "evaluation_metrics": { "macro_f1": 0.8351503663007138, "fpr": 0.1323529411764706, "fnr": 0.13958990536277602, "tp": 1091, "fp": 81, "tn": 531, "fn": 177 } }, "best_fnr_in_f1": { "document_risk_margin_threshold": -0.9776157913729548, "macro_f1": 0.8594848792230071, "fpr": 0.23076923076923078, "fnr": 0.05194805194805195, "tp": 219, "fp": 24, "tn": 80, "fn": 12, "evaluation_metrics": { "macro_f1": 0.8076357183011587, "fpr": 0.3464052287581699, "fnr": 0.0638801261829653, "tp": 1187, "fp": 212, "tn": 400, "fn": 81 } } } }, "union_routing": { "promoter": "factorized_actionable", "promoter_threshold": 0.20513465149941823, "calibration_target_promotion_rate": 0.25 }, "operating_points": { "best_f1": { "threshold": -0.7704736590385437, "threshold_semantics": "best_non_default_class_score_minus_default_class_score", "metrics": { "document_risk_margin_threshold": -0.7704736590385437, "macro_f1": 0.8881845323533561, "fpr": 0.10576923076923077, "fnr": 0.06060606060606061, "tp": 217, "fp": 11, "tn": 93, "fn": 14 }, "policy_selection_metrics": { "macro_f1": 0.8991596445533936, "fpr": 0.10576923076923077, "fnr": 0.07327586206896551, "tp": 215, "fp": 11, "tn": 93, "fn": 17 }, "evaluation_metrics": { "unified_validation": { "macro_f1": 0.8418848849143761, "fpr": 0.2107843137254902, "fnr": 0.07570977917981073, "tp": 1172, "fp": 129, "tn": 483, "fn": 96 } } }, "best_fpr_in_f1": { "threshold": 0.6007874235510826, "threshold_semantics": "best_non_default_class_score_minus_default_class_score", "metrics": { "document_risk_margin_threshold": 0.6007874235510826, "macro_f1": 0.8734210789256979, "fpr": 0.08653846153846154, "fnr": 0.10822510822510822, "tp": 206, "fp": 9, "tn": 95, "fn": 25 }, "policy_selection_metrics": { "macro_f1": 0.8932623821471027, "fpr": 0.07692307692307693, "fnr": 0.11206896551724138, "tp": 206, "fp": 8, "tn": 96, "fn": 26 }, "evaluation_metrics": { "unified_validation": { "macro_f1": 0.8351503663007138, "fpr": 0.1323529411764706, "fnr": 0.13958990536277602, "tp": 1091, "fp": 81, "tn": 531, "fn": 177 } } }, "best_fnr_in_f1": { "threshold": -0.9776157913729548, "threshold_semantics": "best_non_default_class_score_minus_default_class_score", "metrics": { "document_risk_margin_threshold": -0.9776157913729548, "macro_f1": 0.8594848792230071, "fpr": 0.23076923076923078, "fnr": 0.05194805194805195, "tp": 219, "fp": 24, "tn": 80, "fn": 12 }, "policy_selection_metrics": { "macro_f1": 0.8441721159673253, "fpr": 0.3076923076923077, "fnr": 0.06465517241379311, "tp": 217, "fp": 32, "tn": 72, "fn": 15 }, "evaluation_metrics": { "unified_validation": { "macro_f1": 0.8076357183011587, "fpr": 0.3464052287581699, "fnr": 0.0638801261829653, "tp": 1187, "fp": 212, "tn": 400, "fn": 81 } } } } } } }