{ "f1_window": 0.04, "calibration_role": "ntdb/artifacts/experiments/sensitive_joint_v3_mmbert_static_actionable_benefit_v41_all_sensitive_aps_phase2_l3_seed42/split_indices/probability_calibration.npy", "policy_selection_role": "ntdb/artifacts/experiments/sensitive_joint_v3_mmbert_static_actionable_benefit_v41_all_sensitive_aps_phase2_l3_seed42/split_indices/policy_selection.npy", "evaluation_role": "unified_validation", "modes": { "l2_only": { "aggregation": "mean", "f1_window": 0.04, "calibration_operating_points": { "best_f1": { "document_risk_margin_threshold": 0.3269468620419502, "macro_f1": 0.8212544747840629, "fpr": 0.10416666666666667, "fnr": 0.0684931506849315, "tp": 340, "fp": 5, "tn": 43, "fn": 25 }, "best_fpr_in_f1": { "document_risk_margin_threshold": 0.2802514135837555, "macro_f1": 0.8208552098917432, "fpr": 0.10416666666666667, "fnr": 0.06575342465753424, "tp": 341, "fp": 5, "tn": 43, "fn": 24 }, "best_fnr_in_f1": { "document_risk_margin_threshold": -0.7966152094304562, "macro_f1": 0.8119645317052127, "fpr": 0.25, "fnr": 0.005479452054794521, "tp": 363, "fp": 12, "tn": 36, "fn": 2 } }, "sweep_threshold_count": 414, "policy_selection": { "best_f1": { "document_risk_margin_threshold": 0.3269468620419502, "macro_f1": 0.8212544747840629, "fpr": 0.10416666666666667, "fnr": 0.0684931506849315, "tp": 340, "fp": 5, "tn": 43, "fn": 25, "evaluation_metrics": { "macro_f1": 0.7908984301290851, "fpr": 0.2127659574468085, "fnr": 0.08719346049046321, "tp": 335, "fp": 10, "tn": 37, "fn": 32 } }, "best_fpr_in_f1": { "document_risk_margin_threshold": 0.2802514135837555, "macro_f1": 0.8208552098917432, "fpr": 0.10416666666666667, "fnr": 0.06575342465753424, "tp": 341, "fp": 5, "tn": 43, "fn": 24, "evaluation_metrics": { "macro_f1": 0.7908984301290851, "fpr": 0.2127659574468085, "fnr": 0.08719346049046321, "tp": 335, "fp": 10, "tn": 37, "fn": 32 } }, "best_fnr_in_f1": { "document_risk_margin_threshold": -0.7966152094304562, "macro_f1": 0.8119645317052127, "fpr": 0.25, "fnr": 0.005479452054794521, "tp": 363, "fp": 12, "tn": 36, "fn": 2, "evaluation_metrics": { "macro_f1": 0.7779886588170615, "fpr": 0.40425531914893614, "fnr": 0.02452316076294278, "tp": 358, "fp": 19, "tn": 28, "fn": 9 } } }, "evaluation": { "unified_validation": { "best_f1": { "document_risk_margin_threshold": 0.3269468620419502, "macro_f1": 0.8212544747840629, "fpr": 0.10416666666666667, "fnr": 0.0684931506849315, "tp": 340, "fp": 5, "tn": 43, "fn": 25, "evaluation_metrics": { "macro_f1": 0.8748585804941437, "fpr": 0.1459227467811159, "fnr": 0.030534351145038167, "tp": 1778, "fp": 34, "tn": 199, "fn": 56 } }, "best_fpr_in_f1": { "document_risk_margin_threshold": 0.2802514135837555, "macro_f1": 0.8208552098917432, "fpr": 0.10416666666666667, "fnr": 0.06575342465753424, "tp": 341, "fp": 5, "tn": 43, "fn": 24, "evaluation_metrics": { "macro_f1": 0.8752677517950971, "fpr": 0.1459227467811159, "fnr": 0.028353326063249727, "tp": 1782, "fp": 34, "tn": 199, "fn": 52 } }, "best_fnr_in_f1": { "document_risk_margin_threshold": -0.7966152094304562, "macro_f1": 0.8119645317052127, "fpr": 0.25, "fnr": 0.005479452054794521, "tp": 363, "fp": 12, "tn": 36, "fn": 2, "evaluation_metrics": { "macro_f1": 0.8693104068945013, "fpr": 0.296137339055794, "fnr": 0.008178844056706653, "tp": 1819, "fp": 69, "tn": 164, "fn": 15 } } } }, "operating_points": { "best_f1": { "threshold": 0.3269468620419502, "threshold_semantics": "best_non_default_class_score_minus_default_class_score", "metrics": { "document_risk_margin_threshold": 0.3269468620419502, "macro_f1": 0.8212544747840629, "fpr": 0.10416666666666667, "fnr": 0.0684931506849315, "tp": 340, "fp": 5, "tn": 43, "fn": 25 }, "policy_selection_metrics": { "macro_f1": 0.7908984301290851, "fpr": 0.2127659574468085, "fnr": 0.08719346049046321, "tp": 335, "fp": 10, "tn": 37, "fn": 32 }, "evaluation_metrics": { "unified_validation": { "macro_f1": 0.8748585804941437, "fpr": 0.1459227467811159, "fnr": 0.030534351145038167, "tp": 1778, "fp": 34, "tn": 199, "fn": 56 } } }, "best_fpr_in_f1": { "threshold": 0.2802514135837555, "threshold_semantics": "best_non_default_class_score_minus_default_class_score", "metrics": { "document_risk_margin_threshold": 0.2802514135837555, "macro_f1": 0.8208552098917432, "fpr": 0.10416666666666667, "fnr": 0.06575342465753424, "tp": 341, "fp": 5, "tn": 43, "fn": 24 }, "policy_selection_metrics": { "macro_f1": 0.7908984301290851, "fpr": 0.2127659574468085, "fnr": 0.08719346049046321, "tp": 335, "fp": 10, "tn": 37, "fn": 32 }, "evaluation_metrics": { "unified_validation": { "macro_f1": 0.8752677517950971, "fpr": 0.1459227467811159, "fnr": 0.028353326063249727, "tp": 1782, "fp": 34, "tn": 199, "fn": 52 } } }, "best_fnr_in_f1": { "threshold": -0.7966152094304562, "threshold_semantics": "best_non_default_class_score_minus_default_class_score", "metrics": { "document_risk_margin_threshold": -0.7966152094304562, "macro_f1": 0.8119645317052127, "fpr": 0.25, "fnr": 0.005479452054794521, "tp": 363, "fp": 12, "tn": 36, "fn": 2 }, "policy_selection_metrics": { "macro_f1": 0.7779886588170615, "fpr": 0.40425531914893614, "fnr": 0.02452316076294278, "tp": 358, "fp": 19, "tn": 28, "fn": 9 }, "evaluation_metrics": { "unified_validation": { "macro_f1": 0.8693104068945013, "fpr": 0.296137339055794, "fnr": 0.008178844056706653, "tp": 1819, "fp": 69, "tn": 164, "fn": 15 } } } } }, "l3_only": { "aggregation": "smoothmax", "f1_window": 0.04, "calibration_operating_points": { "best_f1": { "document_risk_margin_threshold": -0.22447717189788818, "macro_f1": 0.995257067546521, "fpr": 0.0, "fnr": 0.0, "tp": 365, "fp": 0, "tn": 48, "fn": 0 }, "best_fpr_in_f1": { "document_risk_margin_threshold": -0.22447717189788818, "macro_f1": 0.995257067546521, "fpr": 0.0, "fnr": 0.0, "tp": 365, "fp": 0, "tn": 48, "fn": 0 }, "best_fnr_in_f1": { "document_risk_margin_threshold": -0.22447717189788818, "macro_f1": 0.995257067546521, "fpr": 0.0, "fnr": 0.0, "tp": 365, "fp": 0, "tn": 48, "fn": 0 } }, "sweep_threshold_count": 415, "policy_selection": { "best_f1": { "document_risk_margin_threshold": -0.22447717189788818, "macro_f1": 0.995257067546521, "fpr": 0.0, "fnr": 0.0, "tp": 365, "fp": 0, "tn": 48, "fn": 0, "evaluation_metrics": { "macro_f1": 0.9879752997730525, "fpr": 0.0, "fnr": 0.0, "tp": 367, "fp": 0, "tn": 47, "fn": 0 } }, "best_fpr_in_f1": { "document_risk_margin_threshold": -0.22447717189788818, "macro_f1": 0.995257067546521, "fpr": 0.0, "fnr": 0.0, "tp": 365, "fp": 0, "tn": 48, "fn": 0, "evaluation_metrics": { "macro_f1": 0.9879752997730525, "fpr": 0.0, "fnr": 0.0, "tp": 367, "fp": 0, "tn": 47, "fn": 0 } }, "best_fnr_in_f1": { "document_risk_margin_threshold": -0.22447717189788818, "macro_f1": 0.995257067546521, "fpr": 0.0, "fnr": 0.0, "tp": 365, "fp": 0, "tn": 48, "fn": 0, "evaluation_metrics": { "macro_f1": 0.9879752997730525, "fpr": 0.0, "fnr": 0.0, "tp": 367, "fp": 0, "tn": 47, "fn": 0 } } }, "evaluation": { "unified_validation": { "best_f1": { "document_risk_margin_threshold": -0.22447717189788818, "macro_f1": 0.995257067546521, "fpr": 0.0, "fnr": 0.0, "tp": 365, "fp": 0, "tn": 48, "fn": 0, "evaluation_metrics": { "macro_f1": 0.9680697934798839, "fpr": 0.060085836909871244, "fnr": 0.006543075245365322, "tp": 1822, "fp": 14, "tn": 219, "fn": 12 } }, "best_fpr_in_f1": { "document_risk_margin_threshold": -0.22447717189788818, "macro_f1": 0.995257067546521, "fpr": 0.0, "fnr": 0.0, "tp": 365, "fp": 0, "tn": 48, "fn": 0, "evaluation_metrics": { "macro_f1": 0.9680697934798839, "fpr": 0.060085836909871244, "fnr": 0.006543075245365322, "tp": 1822, "fp": 14, "tn": 219, "fn": 12 } }, "best_fnr_in_f1": { "document_risk_margin_threshold": -0.22447717189788818, "macro_f1": 0.995257067546521, "fpr": 0.0, "fnr": 0.0, "tp": 365, "fp": 0, "tn": 48, "fn": 0, "evaluation_metrics": { "macro_f1": 0.9680697934798839, "fpr": 0.060085836909871244, "fnr": 0.006543075245365322, "tp": 1822, "fp": 14, "tn": 219, "fn": 12 } } } }, "operating_points": { "best_f1": { "threshold": -0.22447717189788818, "threshold_semantics": "best_non_default_class_score_minus_default_class_score", "metrics": { "document_risk_margin_threshold": -0.22447717189788818, "macro_f1": 0.995257067546521, "fpr": 0.0, "fnr": 0.0, "tp": 365, "fp": 0, "tn": 48, "fn": 0 }, "policy_selection_metrics": { "macro_f1": 0.9879752997730525, "fpr": 0.0, "fnr": 0.0, "tp": 367, "fp": 0, "tn": 47, "fn": 0 }, "evaluation_metrics": { "unified_validation": { "macro_f1": 0.9680697934798839, "fpr": 0.060085836909871244, "fnr": 0.006543075245365322, "tp": 1822, "fp": 14, "tn": 219, "fn": 12 } } }, "best_fpr_in_f1": { "threshold": -0.22447717189788818, "threshold_semantics": "best_non_default_class_score_minus_default_class_score", "metrics": { "document_risk_margin_threshold": -0.22447717189788818, "macro_f1": 0.995257067546521, "fpr": 0.0, "fnr": 0.0, "tp": 365, "fp": 0, "tn": 48, "fn": 0 }, "policy_selection_metrics": { "macro_f1": 0.9879752997730525, "fpr": 0.0, "fnr": 0.0, "tp": 367, "fp": 0, "tn": 47, "fn": 0 }, "evaluation_metrics": { "unified_validation": { "macro_f1": 0.9680697934798839, "fpr": 0.060085836909871244, "fnr": 0.006543075245365322, "tp": 1822, "fp": 14, "tn": 219, "fn": 12 } } }, "best_fnr_in_f1": { "threshold": -0.22447717189788818, "threshold_semantics": "best_non_default_class_score_minus_default_class_score", "metrics": { "document_risk_margin_threshold": -0.22447717189788818, "macro_f1": 0.995257067546521, "fpr": 0.0, "fnr": 0.0, "tp": 365, "fp": 0, "tn": 48, "fn": 0 }, "policy_selection_metrics": { "macro_f1": 0.9879752997730525, "fpr": 0.0, "fnr": 0.0, "tp": 367, "fp": 0, "tn": 47, "fn": 0 }, "evaluation_metrics": { "unified_validation": { "macro_f1": 0.9680697934798839, "fpr": 0.060085836909871244, "fnr": 0.006543075245365322, "tp": 1822, "fp": 14, "tn": 219, "fn": 12 } } } } }, "union_l2_l3": { "aggregation": "mean", "f1_window": 0.04, "calibration_operating_points": { "best_f1": { "document_risk_margin_threshold": -0.19841083884239197, "macro_f1": 0.9810854243276995, "fpr": 0.0, "fnr": 0.0, "tp": 365, "fp": 0, "tn": 48, "fn": 0 }, "best_fpr_in_f1": { "document_risk_margin_threshold": -0.19841083884239197, "macro_f1": 0.9810854243276995, "fpr": 0.0, "fnr": 0.0, "tp": 365, "fp": 0, "tn": 48, "fn": 0 }, "best_fnr_in_f1": { "document_risk_margin_threshold": -0.19841083884239197, "macro_f1": 0.9810854243276995, "fpr": 0.0, "fnr": 0.0, "tp": 365, "fp": 0, "tn": 48, "fn": 0 } }, "sweep_threshold_count": 414, "policy_selection": { "best_f1": { "document_risk_margin_threshold": -0.19841083884239197, "macro_f1": 0.9810854243276995, "fpr": 0.0, "fnr": 0.0, "tp": 365, "fp": 0, "tn": 48, "fn": 0, "evaluation_metrics": { "macro_f1": 0.9736874663611368, "fpr": 0.0425531914893617, "fnr": 0.005449591280653951, "tp": 365, "fp": 2, "tn": 45, "fn": 2 } }, "best_fpr_in_f1": { "document_risk_margin_threshold": -0.19841083884239197, "macro_f1": 0.9810854243276995, "fpr": 0.0, "fnr": 0.0, "tp": 365, "fp": 0, "tn": 48, "fn": 0, "evaluation_metrics": { "macro_f1": 0.9736874663611368, "fpr": 0.0425531914893617, "fnr": 0.005449591280653951, "tp": 365, "fp": 2, "tn": 45, "fn": 2 } }, "best_fnr_in_f1": { "document_risk_margin_threshold": -0.19841083884239197, "macro_f1": 0.9810854243276995, "fpr": 0.0, "fnr": 0.0, "tp": 365, "fp": 0, "tn": 48, "fn": 0, "evaluation_metrics": { "macro_f1": 0.9736874663611368, "fpr": 0.0425531914893617, "fnr": 0.005449591280653951, "tp": 365, "fp": 2, "tn": 45, "fn": 2 } } }, "evaluation": { "unified_validation": { "best_f1": { "document_risk_margin_threshold": -0.19841083884239197, "macro_f1": 0.9810854243276995, "fpr": 0.0, "fnr": 0.0, "tp": 365, "fp": 0, "tn": 48, "fn": 0, "evaluation_metrics": { "macro_f1": 0.9518323126346822, "fpr": 0.09871244635193133, "fnr": 0.007088331515812432, "tp": 1821, "fp": 23, "tn": 210, "fn": 13 } }, "best_fpr_in_f1": { "document_risk_margin_threshold": -0.19841083884239197, "macro_f1": 0.9810854243276995, "fpr": 0.0, "fnr": 0.0, "tp": 365, "fp": 0, "tn": 48, "fn": 0, "evaluation_metrics": { "macro_f1": 0.9518323126346822, "fpr": 0.09871244635193133, "fnr": 0.007088331515812432, "tp": 1821, "fp": 23, "tn": 210, "fn": 13 } }, "best_fnr_in_f1": { "document_risk_margin_threshold": -0.19841083884239197, "macro_f1": 0.9810854243276995, "fpr": 0.0, "fnr": 0.0, "tp": 365, "fp": 0, "tn": 48, "fn": 0, "evaluation_metrics": { "macro_f1": 0.9518323126346822, "fpr": 0.09871244635193133, "fnr": 0.007088331515812432, "tp": 1821, "fp": 23, "tn": 210, "fn": 13 } } } }, "union_routing": { "promoter": "direct_actionable", "promoter_threshold": 0.004311159716500321, "calibration_target_promotion_rate": 0.25 }, "operating_points": { "best_f1": { "threshold": -0.19841083884239197, "threshold_semantics": "best_non_default_class_score_minus_default_class_score", "metrics": { "document_risk_margin_threshold": -0.19841083884239197, "macro_f1": 0.9810854243276995, "fpr": 0.0, "fnr": 0.0, "tp": 365, "fp": 0, "tn": 48, "fn": 0 }, "policy_selection_metrics": { "macro_f1": 0.9736874663611368, "fpr": 0.0425531914893617, "fnr": 0.005449591280653951, "tp": 365, "fp": 2, "tn": 45, "fn": 2 }, "evaluation_metrics": { "unified_validation": { "macro_f1": 0.9518323126346822, "fpr": 0.09871244635193133, "fnr": 0.007088331515812432, "tp": 1821, "fp": 23, "tn": 210, "fn": 13 } } }, "best_fpr_in_f1": { "threshold": -0.19841083884239197, "threshold_semantics": "best_non_default_class_score_minus_default_class_score", "metrics": { "document_risk_margin_threshold": -0.19841083884239197, "macro_f1": 0.9810854243276995, "fpr": 0.0, "fnr": 0.0, "tp": 365, "fp": 0, "tn": 48, "fn": 0 }, "policy_selection_metrics": { "macro_f1": 0.9736874663611368, "fpr": 0.0425531914893617, "fnr": 0.005449591280653951, "tp": 365, "fp": 2, "tn": 45, "fn": 2 }, "evaluation_metrics": { "unified_validation": { "macro_f1": 0.9518323126346822, "fpr": 0.09871244635193133, "fnr": 0.007088331515812432, "tp": 1821, "fp": 23, "tn": 210, "fn": 13 } } }, "best_fnr_in_f1": { "threshold": -0.19841083884239197, "threshold_semantics": "best_non_default_class_score_minus_default_class_score", "metrics": { "document_risk_margin_threshold": -0.19841083884239197, "macro_f1": 0.9810854243276995, "fpr": 0.0, "fnr": 0.0, "tp": 365, "fp": 0, "tn": 48, "fn": 0 }, "policy_selection_metrics": { "macro_f1": 0.9736874663611368, "fpr": 0.0425531914893617, "fnr": 0.005449591280653951, "tp": 365, "fp": 2, "tn": 45, "fn": 2 }, "evaluation_metrics": { "unified_validation": { "macro_f1": 0.9518323126346822, "fpr": 0.09871244635193133, "fnr": 0.007088331515812432, "tp": 1821, "fp": 23, "tn": 210, "fn": 13 } } } } } } }