{ "cutoffs": { "10": { "cells": 12, "completion_time_sources": { "reconstructed_from_latency": 37 }, "examples": 37, "headline": { "bootstrap": { "accuracy_delta_instrumentation_minus_outcome": { "ci95": [ -0.1333333333333333, 0.05555555555555547 ], "point": -0.027027027027027084 }, "brier_delta_instrumentation_minus_outcome": { "ci95": [ -0.047874993358699436, 0.04753434529263568 ], "point": -0.0026830647286030934 }, "replicates": 10000, "seed": 20260714, "semantics": "group bootstrap over cells; diagnostic confidence interval" }, "instrumentation_aware": { "classification": { "accuracy": 0.8918918918918919, "balanced_accuracy": 0.8531746031746033, "brier": 0.07268572313801988, "confusion": { "false_negative": 2, "false_positive": 2, "true_negative": 7, "true_positive": 26 }, "log_loss": 0.22136936001258795 }, "policies": [ { "abstain_continue_full": 3, "correctly_saved_h20_hours": 0.7938369600363888, "decision_coverage": 0.918918918918919, "early_accept": 27, "early_reject": 7, "false_accept": 1, "false_accept_examples": [ { "anchor": 0.49609375, "cell": "tp2_mns8", "label_feasible": false, "probability_feasible": 0.8623652659223954, "remaining_h20_hours": 0.007340132528333333 } ], "false_reject": 1, "false_reject_examples": [ { "anchor": 0.033717411016, "cell": "tp4_mns32", "label_feasible": true, "probability_feasible": 0.11022926282564288, "remaining_h20_hours": 0.018465968481111112 } ], "full_trial_h20_hours": 1.0669595034675, "invalidly_saved_h20_hours": 0.025806101009444443, "remaining_h20_hours_at_cutoff": 0.8475150590230554, "saved_h20_hours_if_decisions_used": 0.8196430610458332, "threshold": 0.8, "valid_cost_reduction_fraction": null, "valid_zero_error_policy": false }, { "abstain_continue_full": 8, "correctly_saved_h20_hours": 0.7340601279138887, "decision_coverage": 0.7837837837837838, "early_accept": 24, "early_reject": 5, "false_accept": 0, "false_accept_examples": [], "false_reject": 0, "false_reject_examples": [], "full_trial_h20_hours": 1.0669595034675, "invalidly_saved_h20_hours": 0.0, "remaining_h20_hours_at_cutoff": 0.8475150590230554, "saved_h20_hours_if_decisions_used": 0.7340601279138887, "threshold": 0.9, "valid_cost_reduction_fraction": 0.6879924922438714, "valid_zero_error_policy": true }, { "abstain_continue_full": 15, "correctly_saved_h20_hours": 0.620235621837222, "decision_coverage": 0.5945945945945946, "early_accept": 17, "early_reject": 5, "false_accept": 0, "false_accept_examples": [], "false_reject": 0, "false_reject_examples": [], "full_trial_h20_hours": 1.0669595034675, "invalidly_saved_h20_hours": 0.0, "remaining_h20_hours_at_cutoff": 0.8475150590230554, "saved_h20_hours_if_decisions_used": 0.620235621837222, "threshold": 0.95, "valid_cost_reduction_fraction": 0.5813113054633519, "valid_zero_error_policy": true } ] }, "outcome_only": { "classification": { "accuracy": 0.918918918918919, "balanced_accuracy": 0.8333333333333333, "brier": 0.07536878786662297, "confusion": { "false_negative": 0, "false_positive": 3, "true_negative": 6, "true_positive": 28 }, "log_loss": 0.23513650573912553 }, "policies": [ { "abstain_continue_full": 5, "correctly_saved_h20_hours": 0.7625537088033332, "decision_coverage": 0.8648648648648649, "early_accept": 27, "early_reject": 5, "false_accept": 2, "false_accept_examples": [ { "anchor": 0.49609375, "cell": "tp2_mns8", "label_feasible": false, "probability_feasible": 0.8486897200744762, "remaining_h20_hours": 0.007340132528333333 }, { "anchor": 0.034252608017, "cell": "tp4_mns8", "label_feasible": false, "probability_feasible": 0.8216385310591849, "remaining_h20_hours": 0.01680367415111111 } ], "false_reject": 0, "false_reject_examples": [], "full_trial_h20_hours": 1.0669595034675, "invalidly_saved_h20_hours": 0.024143806679444442, "remaining_h20_hours_at_cutoff": 0.8475150590230554, "saved_h20_hours_if_decisions_used": 0.7866975154827778, "threshold": 0.8, "valid_cost_reduction_fraction": null, "valid_zero_error_policy": false }, { "abstain_continue_full": 11, "correctly_saved_h20_hours": 0.7042463331286111, "decision_coverage": 0.7027027027027027, "early_accept": 22, "early_reject": 4, "false_accept": 0, "false_accept_examples": [], "false_reject": 0, "false_reject_examples": [], "full_trial_h20_hours": 1.0669595034675, "invalidly_saved_h20_hours": 0.0, "remaining_h20_hours_at_cutoff": 0.8475150590230554, "saved_h20_hours_if_decisions_used": 0.7042463331286111, "threshold": 0.9, "valid_cost_reduction_fraction": 0.6600497308847137, "valid_zero_error_policy": true }, { "abstain_continue_full": 19, "correctly_saved_h20_hours": 0.5876473660130556, "decision_coverage": 0.4864864864864865, "early_accept": 15, "early_reject": 3, "false_accept": 0, "false_accept_examples": [], "false_reject": 0, "false_reject_examples": [], "full_trial_h20_hours": 1.0669595034675, "invalidly_saved_h20_hours": 0.0, "remaining_h20_hours_at_cutoff": 0.8475150590230554, "saved_h20_hours_if_decisions_used": 0.5876473660130556, "threshold": 0.95, "valid_cost_reduction_fraction": 0.5507682007641966, "valid_zero_error_policy": true } ] }, "paired_correctness": { "both_correct": 32, "both_wrong": 2, "instrumentation_only_correct": 1, "mcnemar_exact_two_sided_p": 1.0, "outcome_only_correct": 2 } }, "headline_regularization": 1.0, "label_sanity": { "distinct_n": 2, "max": 1.0, "min": 0.0, "n": 37, "negative": 9, "positive": 28, "primary_adjudicated_disagreements": 3 }, "regularization_sensitivity": { "0.1": { "bootstrap": { "accuracy_delta_instrumentation_minus_outcome": { "ci95": [ -0.08571428571428574, 0.07317073170731703 ], "point": 0.0 }, "brier_delta_instrumentation_minus_outcome": { "ci95": [ -0.01721313553473963, 0.0846932121032576 ], "point": 0.019153359424323196 }, "replicates": 10000, "seed": 20260714, "semantics": "group bootstrap over cells; diagnostic confidence interval" }, "instrumentation_aware": { "classification": { "accuracy": 0.918918918918919, "balanced_accuracy": 0.871031746031746, "brier": 0.06720185291989068, "confusion": { "false_negative": 1, "false_positive": 2, "true_negative": 7, "true_positive": 27 }, "log_loss": 0.22428253113260074 }, "policies": [ { "abstain_continue_full": 2, "correctly_saved_h20_hours": 0.802005166048611, "decision_coverage": 0.9459459459459459, "early_accept": 27, "early_reject": 8, "false_accept": 1, "false_accept_examples": [ { "anchor": 0.49609375, "cell": "tp2_mns8", "label_feasible": false, "probability_feasible": 0.905553117931728, "remaining_h20_hours": 0.007340132528333333 } ], "false_reject": 1, "false_reject_examples": [ { "anchor": 0.033717411016, "cell": "tp4_mns32", "label_feasible": true, "probability_feasible": 0.02745974548325657, "remaining_h20_hours": 0.018465968481111112 } ], "full_trial_h20_hours": 1.0669595034675, "invalidly_saved_h20_hours": 0.025806101009444443, "remaining_h20_hours_at_cutoff": 0.8475150590230554, "saved_h20_hours_if_decisions_used": 0.8278112670580555, "threshold": 0.8, "valid_cost_reduction_fraction": null, "valid_zero_error_policy": false }, { "abstain_continue_full": 4, "correctly_saved_h20_hours": 0.7653433791469442, "decision_coverage": 0.8918918918918919, "early_accept": 26, "early_reject": 7, "false_accept": 1, "false_accept_examples": [ { "anchor": 0.49609375, "cell": "tp2_mns8", "label_feasible": false, "probability_feasible": 0.905553117931728, "remaining_h20_hours": 0.007340132528333333 } ], "false_reject": 1, "false_reject_examples": [ { "anchor": 0.033717411016, "cell": "tp4_mns32", "label_feasible": true, "probability_feasible": 0.02745974548325657, "remaining_h20_hours": 0.018465968481111112 } ], "full_trial_h20_hours": 1.0669595034675, "invalidly_saved_h20_hours": 0.025806101009444443, "remaining_h20_hours_at_cutoff": 0.8475150590230554, "saved_h20_hours_if_decisions_used": 0.7911494801563888, "threshold": 0.9, "valid_cost_reduction_fraction": null, "valid_zero_error_policy": false }, { "abstain_continue_full": 7, "correctly_saved_h20_hours": 0.7226741330227777, "decision_coverage": 0.8108108108108109, "early_accept": 23, "early_reject": 7, "false_accept": 0, "false_accept_examples": [], "false_reject": 1, "false_reject_examples": [ { "anchor": 0.033717411016, "cell": "tp4_mns32", "label_feasible": true, "probability_feasible": 0.02745974548325657, "remaining_h20_hours": 0.018465968481111112 } ], "full_trial_h20_hours": 1.0669595034675, "invalidly_saved_h20_hours": 0.018465968481111112, "remaining_h20_hours_at_cutoff": 0.8475150590230554, "saved_h20_hours_if_decisions_used": 0.7411401015038888, "threshold": 0.95, "valid_cost_reduction_fraction": null, "valid_zero_error_policy": false } ] }, "outcome_only": { "classification": { "accuracy": 0.918918918918919, "balanced_accuracy": 0.8333333333333333, "brier": 0.04804849349556749, "confusion": { "false_negative": 0, "false_positive": 3, "true_negative": 6, "true_positive": 28 }, "log_loss": 0.15081259765362642 }, "policies": [ { "abstain_continue_full": 4, "correctly_saved_h20_hours": 0.7836433176388888, "decision_coverage": 0.8918918918918919, "early_accept": 27, "early_reject": 6, "false_accept": 1, "false_accept_examples": [ { "anchor": 0.49609375, "cell": "tp2_mns8", "label_feasible": false, "probability_feasible": 0.8878008796435344, "remaining_h20_hours": 0.007340132528333333 } ], "false_reject": 0, "false_reject_examples": [], "full_trial_h20_hours": 1.0669595034675, "invalidly_saved_h20_hours": 0.007340132528333333, "remaining_h20_hours_at_cutoff": 0.8475150590230554, "saved_h20_hours_if_decisions_used": 0.7909834501672222, "threshold": 0.8, "valid_cost_reduction_fraction": null, "valid_zero_error_policy": false }, { "abstain_continue_full": 8, "correctly_saved_h20_hours": 0.718791861695, "decision_coverage": 0.7837837837837838, "early_accept": 24, "early_reject": 5, "false_accept": 0, "false_accept_examples": [], "false_reject": 0, "false_reject_examples": [], "full_trial_h20_hours": 1.0669595034675, "invalidly_saved_h20_hours": 0.0, "remaining_h20_hours_at_cutoff": 0.8475150590230554, "saved_h20_hours_if_decisions_used": 0.718791861695, "threshold": 0.9, "valid_cost_reduction_fraction": 0.6736824212718535, "valid_zero_error_policy": true }, { "abstain_continue_full": 10, "correctly_saved_h20_hours": 0.7046020469899998, "decision_coverage": 0.7297297297297297, "early_accept": 23, "early_reject": 4, "false_accept": 0, "false_accept_examples": [], "false_reject": 0, "false_reject_examples": [], "full_trial_h20_hours": 1.0669595034675, "invalidly_saved_h20_hours": 0.0, "remaining_h20_hours_at_cutoff": 0.8475150590230554, "saved_h20_hours_if_decisions_used": 0.7046020469899998, "threshold": 0.95, "valid_cost_reduction_fraction": 0.6603831211026486, "valid_zero_error_policy": true } ] }, "paired_correctness": { "both_correct": 33, "both_wrong": 2, "instrumentation_only_correct": 1, "mcnemar_exact_two_sided_p": 1.0, "outcome_only_correct": 1 } }, "1.0": { "bootstrap": { "accuracy_delta_instrumentation_minus_outcome": { "ci95": [ -0.1333333333333333, 0.05555555555555547 ], "point": -0.027027027027027084 }, "brier_delta_instrumentation_minus_outcome": { "ci95": [ -0.047874993358699436, 0.04753434529263568 ], "point": -0.0026830647286030934 }, "replicates": 10000, "seed": 20260714, "semantics": "group bootstrap over cells; diagnostic confidence interval" }, "instrumentation_aware": { "classification": { "accuracy": 0.8918918918918919, "balanced_accuracy": 0.8531746031746033, "brier": 0.07268572313801988, "confusion": { "false_negative": 2, "false_positive": 2, "true_negative": 7, "true_positive": 26 }, "log_loss": 0.22136936001258795 }, "policies": [ { "abstain_continue_full": 3, "correctly_saved_h20_hours": 0.7938369600363888, "decision_coverage": 0.918918918918919, "early_accept": 27, "early_reject": 7, "false_accept": 1, "false_accept_examples": [ { "anchor": 0.49609375, "cell": "tp2_mns8", "label_feasible": false, "probability_feasible": 0.8623652659223954, "remaining_h20_hours": 0.007340132528333333 } ], "false_reject": 1, "false_reject_examples": [ { "anchor": 0.033717411016, "cell": "tp4_mns32", "label_feasible": true, "probability_feasible": 0.11022926282564288, "remaining_h20_hours": 0.018465968481111112 } ], "full_trial_h20_hours": 1.0669595034675, "invalidly_saved_h20_hours": 0.025806101009444443, "remaining_h20_hours_at_cutoff": 0.8475150590230554, "saved_h20_hours_if_decisions_used": 0.8196430610458332, "threshold": 0.8, "valid_cost_reduction_fraction": null, "valid_zero_error_policy": false }, { "abstain_continue_full": 8, "correctly_saved_h20_hours": 0.7340601279138887, "decision_coverage": 0.7837837837837838, "early_accept": 24, "early_reject": 5, "false_accept": 0, "false_accept_examples": [], "false_reject": 0, "false_reject_examples": [], "full_trial_h20_hours": 1.0669595034675, "invalidly_saved_h20_hours": 0.0, "remaining_h20_hours_at_cutoff": 0.8475150590230554, "saved_h20_hours_if_decisions_used": 0.7340601279138887, "threshold": 0.9, "valid_cost_reduction_fraction": 0.6879924922438714, "valid_zero_error_policy": true }, { "abstain_continue_full": 15, "correctly_saved_h20_hours": 0.620235621837222, "decision_coverage": 0.5945945945945946, "early_accept": 17, "early_reject": 5, "false_accept": 0, "false_accept_examples": [], "false_reject": 0, "false_reject_examples": [], "full_trial_h20_hours": 1.0669595034675, "invalidly_saved_h20_hours": 0.0, "remaining_h20_hours_at_cutoff": 0.8475150590230554, "saved_h20_hours_if_decisions_used": 0.620235621837222, "threshold": 0.95, "valid_cost_reduction_fraction": 0.5813113054633519, "valid_zero_error_policy": true } ] }, "outcome_only": { "classification": { "accuracy": 0.918918918918919, "balanced_accuracy": 0.8333333333333333, "brier": 0.07536878786662297, "confusion": { "false_negative": 0, "false_positive": 3, "true_negative": 6, "true_positive": 28 }, "log_loss": 0.23513650573912553 }, "policies": [ { "abstain_continue_full": 5, "correctly_saved_h20_hours": 0.7625537088033332, "decision_coverage": 0.8648648648648649, "early_accept": 27, "early_reject": 5, "false_accept": 2, "false_accept_examples": [ { "anchor": 0.49609375, "cell": "tp2_mns8", "label_feasible": false, "probability_feasible": 0.8486897200744762, "remaining_h20_hours": 0.007340132528333333 }, { "anchor": 0.034252608017, "cell": "tp4_mns8", "label_feasible": false, "probability_feasible": 0.8216385310591849, "remaining_h20_hours": 0.01680367415111111 } ], "false_reject": 0, "false_reject_examples": [], "full_trial_h20_hours": 1.0669595034675, "invalidly_saved_h20_hours": 0.024143806679444442, "remaining_h20_hours_at_cutoff": 0.8475150590230554, "saved_h20_hours_if_decisions_used": 0.7866975154827778, "threshold": 0.8, "valid_cost_reduction_fraction": null, "valid_zero_error_policy": false }, { "abstain_continue_full": 11, "correctly_saved_h20_hours": 0.7042463331286111, "decision_coverage": 0.7027027027027027, "early_accept": 22, "early_reject": 4, "false_accept": 0, "false_accept_examples": [], "false_reject": 0, "false_reject_examples": [], "full_trial_h20_hours": 1.0669595034675, "invalidly_saved_h20_hours": 0.0, "remaining_h20_hours_at_cutoff": 0.8475150590230554, "saved_h20_hours_if_decisions_used": 0.7042463331286111, "threshold": 0.9, "valid_cost_reduction_fraction": 0.6600497308847137, "valid_zero_error_policy": true }, { "abstain_continue_full": 19, "correctly_saved_h20_hours": 0.5876473660130556, "decision_coverage": 0.4864864864864865, "early_accept": 15, "early_reject": 3, "false_accept": 0, "false_accept_examples": [], "false_reject": 0, "false_reject_examples": [], "full_trial_h20_hours": 1.0669595034675, "invalidly_saved_h20_hours": 0.0, "remaining_h20_hours_at_cutoff": 0.8475150590230554, "saved_h20_hours_if_decisions_used": 0.5876473660130556, "threshold": 0.95, "valid_cost_reduction_fraction": 0.5507682007641966, "valid_zero_error_policy": true } ] }, "paired_correctness": { "both_correct": 32, "both_wrong": 2, "instrumentation_only_correct": 1, "mcnemar_exact_two_sided_p": 1.0, "outcome_only_correct": 2 } }, "10.0": { "bootstrap": { "accuracy_delta_instrumentation_minus_outcome": { "ci95": [ 0.0, 0.0 ], "point": 0.0 }, "brier_delta_instrumentation_minus_outcome": { "ci95": [ -0.026808476987744526, 0.02087777396654144 ], "point": -0.005783107568599871 }, "replicates": 10000, "seed": 20260714, "semantics": "group bootstrap over cells; diagnostic confidence interval" }, "instrumentation_aware": { "classification": { "accuracy": 0.8378378378378378, "balanced_accuracy": 0.7420634920634921, "brier": 0.08901809268133021, "confusion": { "false_negative": 2, "false_positive": 4, "true_negative": 5, "true_positive": 26 }, "log_loss": 0.2787387373669301 }, "policies": [ { "abstain_continue_full": 5, "correctly_saved_h20_hours": 0.7770332858852778, "decision_coverage": 0.8648648648648649, "early_accept": 27, "early_reject": 5, "false_accept": 1, "false_accept_examples": [ { "anchor": 0.49609375, "cell": "tp2_mns8", "label_feasible": false, "probability_feasible": 0.8449371690787663, "remaining_h20_hours": 0.007340132528333333 } ], "false_reject": 0, "false_reject_examples": [], "full_trial_h20_hours": 1.0669595034675, "invalidly_saved_h20_hours": 0.007340132528333333, "remaining_h20_hours_at_cutoff": 0.8475150590230554, "saved_h20_hours_if_decisions_used": 0.7843734184136111, "threshold": 0.8, "valid_cost_reduction_fraction": null, "valid_zero_error_policy": false }, { "abstain_continue_full": 17, "correctly_saved_h20_hours": 0.6023881611255555, "decision_coverage": 0.5405405405405406, "early_accept": 16, "early_reject": 4, "false_accept": 0, "false_accept_examples": [], "false_reject": 0, "false_reject_examples": [], "full_trial_h20_hours": 1.0669595034675, "invalidly_saved_h20_hours": 0.0, "remaining_h20_hours_at_cutoff": 0.8475150590230554, "saved_h20_hours_if_decisions_used": 0.6023881611255555, "threshold": 0.9, "valid_cost_reduction_fraction": 0.564583903295168, "valid_zero_error_policy": true }, { "abstain_continue_full": 30, "correctly_saved_h20_hours": 0.28353444813305556, "decision_coverage": 0.1891891891891892, "early_accept": 6, "early_reject": 1, "false_accept": 0, "false_accept_examples": [], "false_reject": 0, "false_reject_examples": [], "full_trial_h20_hours": 1.0669595034675, "invalidly_saved_h20_hours": 0.0, "remaining_h20_hours_at_cutoff": 0.8475150590230554, "saved_h20_hours_if_decisions_used": 0.28353444813305556, "threshold": 0.95, "valid_cost_reduction_fraction": 0.2657405901644815, "valid_zero_error_policy": true } ] }, "outcome_only": { "classification": { "accuracy": 0.8378378378378378, "balanced_accuracy": 0.7420634920634921, "brier": 0.09480120024993008, "confusion": { "false_negative": 2, "false_positive": 4, "true_negative": 5, "true_positive": 26 }, "log_loss": 0.31278388302658006 }, "policies": [ { "abstain_continue_full": 8, "correctly_saved_h20_hours": 0.7566135917169443, "decision_coverage": 0.7837837837837838, "early_accept": 26, "early_reject": 3, "false_accept": 1, "false_accept_examples": [ { "anchor": 0.49609375, "cell": "tp2_mns8", "label_feasible": false, "probability_feasible": 0.8320431084673425, "remaining_h20_hours": 0.007340132528333333 } ], "false_reject": 0, "false_reject_examples": [], "full_trial_h20_hours": 1.0669595034675, "invalidly_saved_h20_hours": 0.007340132528333333, "remaining_h20_hours_at_cutoff": 0.8475150590230554, "saved_h20_hours_if_decisions_used": 0.7639537242452776, "threshold": 0.8, "valid_cost_reduction_fraction": null, "valid_zero_error_policy": false }, { "abstain_continue_full": 24, "correctly_saved_h20_hours": 0.5226926016833333, "decision_coverage": 0.35135135135135137, "early_accept": 12, "early_reject": 1, "false_accept": 0, "false_accept_examples": [], "false_reject": 0, "false_reject_examples": [], "full_trial_h20_hours": 1.0669595034675, "invalidly_saved_h20_hours": 0.0, "remaining_h20_hours_at_cutoff": 0.8475150590230554, "saved_h20_hours_if_decisions_used": 0.5226926016833333, "threshold": 0.9, "valid_cost_reduction_fraction": 0.48988982242028906, "valid_zero_error_policy": true }, { "abstain_continue_full": 34, "correctly_saved_h20_hours": 0.16934238506333332, "decision_coverage": 0.08108108108108109, "early_accept": 3, "early_reject": 0, "false_accept": 0, "false_accept_examples": [], "false_reject": 0, "false_reject_examples": [], "full_trial_h20_hours": 1.0669595034675, "invalidly_saved_h20_hours": 0.0, "remaining_h20_hours_at_cutoff": 0.8475150590230554, "saved_h20_hours_if_decisions_used": 0.16934238506333332, "threshold": 0.95, "valid_cost_reduction_fraction": 0.15871491327739184, "valid_zero_error_policy": true } ] }, "paired_correctness": { "both_correct": 31, "both_wrong": 6, "instrumentation_only_correct": 0, "mcnemar_exact_two_sided_p": 1.0, "outcome_only_correct": 0 } } }, "remaining_h20_hours": { "distinct_n": 37, "max": 0.057199826214444444, "min": 0.0012684118816666664, "n": 37 } }, "15": { "cells": 12, "completion_time_sources": { "reconstructed_from_latency": 36 }, "examples": 36, "headline": { "bootstrap": { "accuracy_delta_instrumentation_minus_outcome": { "ci95": [ -0.06896551724137923, 0.11428571428571421 ], "point": 0.02777777777777779 }, "brier_delta_instrumentation_minus_outcome": { "ci95": [ -0.07972316141025643, -0.004101962239149991 ], "point": -0.03744665409603411 }, "replicates": 10000, "seed": 20260714, "semantics": "group bootstrap over cells; diagnostic confidence interval" }, "instrumentation_aware": { "classification": { "accuracy": 0.9166666666666666, "balanced_accuracy": 0.8571428571428572, "brier": 0.03946379437502599, "confusion": { "false_negative": 1, "false_positive": 2, "true_negative": 6, "true_positive": 27 }, "log_loss": 0.12484297369457635 }, "policies": [ { "abstain_continue_full": 3, "correctly_saved_h20_hours": 0.7164359347680556, "decision_coverage": 0.9166666666666666, "early_accept": 27, "early_reject": 6, "false_accept": 0, "false_accept_examples": [], "false_reject": 0, "false_reject_examples": [], "full_trial_h20_hours": 1.0629133138080555, "invalidly_saved_h20_hours": 0.0, "remaining_h20_hours_at_cutoff": 0.7379133138080555, "saved_h20_hours_if_decisions_used": 0.7164359347680556, "threshold": 0.8, "valid_cost_reduction_fraction": 0.6740304458143536, "valid_zero_error_policy": true }, { "abstain_continue_full": 6, "correctly_saved_h20_hours": 0.6905889710727777, "decision_coverage": 0.8333333333333334, "early_accept": 25, "early_reject": 5, "false_accept": 0, "false_accept_examples": [], "false_reject": 0, "false_reject_examples": [], "full_trial_h20_hours": 1.0629133138080555, "invalidly_saved_h20_hours": 0.0, "remaining_h20_hours_at_cutoff": 0.7379133138080555, "saved_h20_hours_if_decisions_used": 0.6905889710727777, "threshold": 0.9, "valid_cost_reduction_fraction": 0.6497133511279798, "valid_zero_error_policy": true }, { "abstain_continue_full": 12, "correctly_saved_h20_hours": 0.5895315664361112, "decision_coverage": 0.6666666666666666, "early_accept": 20, "early_reject": 4, "false_accept": 0, "false_accept_examples": [], "false_reject": 0, "false_reject_examples": [], "full_trial_h20_hours": 1.0629133138080555, "invalidly_saved_h20_hours": 0.0, "remaining_h20_hours_at_cutoff": 0.7379133138080555, "saved_h20_hours_if_decisions_used": 0.5895315664361112, "threshold": 0.95, "valid_cost_reduction_fraction": 0.5546374843344664, "valid_zero_error_policy": true } ] }, "outcome_only": { "classification": { "accuracy": 0.8888888888888888, "balanced_accuracy": 0.75, "brier": 0.0769104484710601, "confusion": { "false_negative": 0, "false_positive": 4, "true_negative": 4, "true_positive": 28 }, "log_loss": 0.23230947656220863 }, "policies": [ { "abstain_continue_full": 6, "correctly_saved_h20_hours": 0.6709655458816667, "decision_coverage": 0.8333333333333334, "early_accept": 26, "early_reject": 4, "false_accept": 1, "false_accept_examples": [ { "anchor": 0.5, "cell": "tp1_mns64", "label_feasible": false, "probability_feasible": 0.8578028168458919, "remaining_h20_hours": 0.0014445317586111114 } ], "false_reject": 0, "false_reject_examples": [], "full_trial_h20_hours": 1.0629133138080555, "invalidly_saved_h20_hours": 0.0014445317586111114, "remaining_h20_hours_at_cutoff": 0.7379133138080555, "saved_h20_hours_if_decisions_used": 0.6724100776402777, "threshold": 0.8, "valid_cost_reduction_fraction": null, "valid_zero_error_policy": false }, { "abstain_continue_full": 11, "correctly_saved_h20_hours": 0.626660078601389, "decision_coverage": 0.6944444444444444, "early_accept": 22, "early_reject": 3, "false_accept": 0, "false_accept_examples": [], "false_reject": 0, "false_reject_examples": [], "full_trial_h20_hours": 1.0629133138080555, "invalidly_saved_h20_hours": 0.0, "remaining_h20_hours_at_cutoff": 0.7379133138080555, "saved_h20_hours_if_decisions_used": 0.626660078601389, "threshold": 0.9, "valid_cost_reduction_fraction": 0.5895683782116529, "valid_zero_error_policy": true }, { "abstain_continue_full": 16, "correctly_saved_h20_hours": 0.5494646261305556, "decision_coverage": 0.5555555555555556, "early_accept": 17, "early_reject": 3, "false_accept": 0, "false_accept_examples": [], "false_reject": 0, "false_reject_examples": [], "full_trial_h20_hours": 1.0629133138080555, "invalidly_saved_h20_hours": 0.0, "remaining_h20_hours_at_cutoff": 0.7379133138080555, "saved_h20_hours_if_decisions_used": 0.5494646261305556, "threshold": 0.95, "valid_cost_reduction_fraction": 0.5169420864266074, "valid_zero_error_policy": true } ] }, "paired_correctness": { "both_correct": 31, "both_wrong": 2, "instrumentation_only_correct": 2, "mcnemar_exact_two_sided_p": 1.0, "outcome_only_correct": 1 } }, "headline_regularization": 1.0, "label_sanity": { "distinct_n": 2, "max": 1.0, "min": 0.0, "n": 36, "negative": 8, "positive": 28, "primary_adjudicated_disagreements": 3 }, "regularization_sensitivity": { "0.1": { "bootstrap": { "accuracy_delta_instrumentation_minus_outcome": { "ci95": [ 0.0, 0.13793103448275856 ], "point": 0.05555555555555558 }, "brier_delta_instrumentation_minus_outcome": { "ci95": [ -0.08924754562701061, 0.007148232838160279 ], "point": -0.03187976039254431 }, "replicates": 10000, "seed": 20260714, "semantics": "group bootstrap over cells; diagnostic confidence interval" }, "instrumentation_aware": { "classification": { "accuracy": 0.9444444444444444, "balanced_accuracy": 0.875, "brier": 0.03162164266499348, "confusion": { "false_negative": 0, "false_positive": 2, "true_negative": 6, "true_positive": 28 }, "log_loss": 0.08770869035960555 }, "policies": [ { "abstain_continue_full": 3, "correctly_saved_h20_hours": 0.7164359347680556, "decision_coverage": 0.9166666666666666, "early_accept": 27, "early_reject": 6, "false_accept": 0, "false_accept_examples": [], "false_reject": 0, "false_reject_examples": [], "full_trial_h20_hours": 1.0629133138080555, "invalidly_saved_h20_hours": 0.0, "remaining_h20_hours_at_cutoff": 0.7379133138080555, "saved_h20_hours_if_decisions_used": 0.7164359347680556, "threshold": 0.8, "valid_cost_reduction_fraction": 0.6740304458143536, "valid_zero_error_policy": true }, { "abstain_continue_full": 3, "correctly_saved_h20_hours": 0.7164359347680556, "decision_coverage": 0.9166666666666666, "early_accept": 27, "early_reject": 6, "false_accept": 0, "false_accept_examples": [], "false_reject": 0, "false_reject_examples": [], "full_trial_h20_hours": 1.0629133138080555, "invalidly_saved_h20_hours": 0.0, "remaining_h20_hours_at_cutoff": 0.7379133138080555, "saved_h20_hours_if_decisions_used": 0.7164359347680556, "threshold": 0.9, "valid_cost_reduction_fraction": 0.6740304458143536, "valid_zero_error_policy": true }, { "abstain_continue_full": 4, "correctly_saved_h20_hours": 0.6910240435036111, "decision_coverage": 0.8888888888888888, "early_accept": 26, "early_reject": 6, "false_accept": 0, "false_accept_examples": [], "false_reject": 0, "false_reject_examples": [], "full_trial_h20_hours": 1.0629133138080555, "invalidly_saved_h20_hours": 0.0, "remaining_h20_hours_at_cutoff": 0.7379133138080555, "saved_h20_hours_if_decisions_used": 0.6910240435036111, "threshold": 0.95, "valid_cost_reduction_fraction": 0.6501226718366222, "valid_zero_error_policy": true } ] }, "outcome_only": { "classification": { "accuracy": 0.8888888888888888, "balanced_accuracy": 0.7946428571428572, "brier": 0.06350140305753779, "confusion": { "false_negative": 1, "false_positive": 3, "true_negative": 5, "true_positive": 27 }, "log_loss": 0.2433253313244278 }, "policies": [ { "abstain_continue_full": 4, "correctly_saved_h20_hours": 0.689579511745, "decision_coverage": 0.8888888888888888, "early_accept": 27, "early_reject": 5, "false_accept": 1, "false_accept_examples": [ { "anchor": 0.5, "cell": "tp1_mns64", "label_feasible": false, "probability_feasible": 0.9922655130114992, "remaining_h20_hours": 0.0014445317586111114 } ], "false_reject": 0, "false_reject_examples": [], "full_trial_h20_hours": 1.0629133138080555, "invalidly_saved_h20_hours": 0.0014445317586111114, "remaining_h20_hours_at_cutoff": 0.7379133138080555, "saved_h20_hours_if_decisions_used": 0.6910240435036111, "threshold": 0.8, "valid_cost_reduction_fraction": null, "valid_zero_error_policy": false }, { "abstain_continue_full": 5, "correctly_saved_h20_hours": 0.6783313931494445, "decision_coverage": 0.8611111111111112, "early_accept": 27, "early_reject": 4, "false_accept": 1, "false_accept_examples": [ { "anchor": 0.5, "cell": "tp1_mns64", "label_feasible": false, "probability_feasible": 0.9922655130114992, "remaining_h20_hours": 0.0014445317586111114 } ], "false_reject": 0, "false_reject_examples": [], "full_trial_h20_hours": 1.0629133138080555, "invalidly_saved_h20_hours": 0.0014445317586111114, "remaining_h20_hours_at_cutoff": 0.7379133138080555, "saved_h20_hours_if_decisions_used": 0.6797759249080555, "threshold": 0.9, "valid_cost_reduction_fraction": null, "valid_zero_error_policy": false }, { "abstain_continue_full": 8, "correctly_saved_h20_hours": 0.6449625425236111, "decision_coverage": 0.7777777777777778, "early_accept": 24, "early_reject": 4, "false_accept": 1, "false_accept_examples": [ { "anchor": 0.5, "cell": "tp1_mns64", "label_feasible": false, "probability_feasible": 0.9922655130114992, "remaining_h20_hours": 0.0014445317586111114 } ], "false_reject": 0, "false_reject_examples": [], "full_trial_h20_hours": 1.0629133138080555, "invalidly_saved_h20_hours": 0.0014445317586111114, "remaining_h20_hours_at_cutoff": 0.7379133138080555, "saved_h20_hours_if_decisions_used": 0.6464070742822221, "threshold": 0.95, "valid_cost_reduction_fraction": null, "valid_zero_error_policy": false } ] }, "paired_correctness": { "both_correct": 32, "both_wrong": 2, "instrumentation_only_correct": 2, "mcnemar_exact_two_sided_p": 0.5, "outcome_only_correct": 0 } }, "1.0": { "bootstrap": { "accuracy_delta_instrumentation_minus_outcome": { "ci95": [ -0.06896551724137923, 0.11428571428571421 ], "point": 0.02777777777777779 }, "brier_delta_instrumentation_minus_outcome": { "ci95": [ -0.07972316141025643, -0.004101962239149991 ], "point": -0.03744665409603411 }, "replicates": 10000, "seed": 20260714, "semantics": "group bootstrap over cells; diagnostic confidence interval" }, "instrumentation_aware": { "classification": { "accuracy": 0.9166666666666666, "balanced_accuracy": 0.8571428571428572, "brier": 0.03946379437502599, "confusion": { "false_negative": 1, "false_positive": 2, "true_negative": 6, "true_positive": 27 }, "log_loss": 0.12484297369457635 }, "policies": [ { "abstain_continue_full": 3, "correctly_saved_h20_hours": 0.7164359347680556, "decision_coverage": 0.9166666666666666, "early_accept": 27, "early_reject": 6, "false_accept": 0, "false_accept_examples": [], "false_reject": 0, "false_reject_examples": [], "full_trial_h20_hours": 1.0629133138080555, "invalidly_saved_h20_hours": 0.0, "remaining_h20_hours_at_cutoff": 0.7379133138080555, "saved_h20_hours_if_decisions_used": 0.7164359347680556, "threshold": 0.8, "valid_cost_reduction_fraction": 0.6740304458143536, "valid_zero_error_policy": true }, { "abstain_continue_full": 6, "correctly_saved_h20_hours": 0.6905889710727777, "decision_coverage": 0.8333333333333334, "early_accept": 25, "early_reject": 5, "false_accept": 0, "false_accept_examples": [], "false_reject": 0, "false_reject_examples": [], "full_trial_h20_hours": 1.0629133138080555, "invalidly_saved_h20_hours": 0.0, "remaining_h20_hours_at_cutoff": 0.7379133138080555, "saved_h20_hours_if_decisions_used": 0.6905889710727777, "threshold": 0.9, "valid_cost_reduction_fraction": 0.6497133511279798, "valid_zero_error_policy": true }, { "abstain_continue_full": 12, "correctly_saved_h20_hours": 0.5895315664361112, "decision_coverage": 0.6666666666666666, "early_accept": 20, "early_reject": 4, "false_accept": 0, "false_accept_examples": [], "false_reject": 0, "false_reject_examples": [], "full_trial_h20_hours": 1.0629133138080555, "invalidly_saved_h20_hours": 0.0, "remaining_h20_hours_at_cutoff": 0.7379133138080555, "saved_h20_hours_if_decisions_used": 0.5895315664361112, "threshold": 0.95, "valid_cost_reduction_fraction": 0.5546374843344664, "valid_zero_error_policy": true } ] }, "outcome_only": { "classification": { "accuracy": 0.8888888888888888, "balanced_accuracy": 0.75, "brier": 0.0769104484710601, "confusion": { "false_negative": 0, "false_positive": 4, "true_negative": 4, "true_positive": 28 }, "log_loss": 0.23230947656220863 }, "policies": [ { "abstain_continue_full": 6, "correctly_saved_h20_hours": 0.6709655458816667, "decision_coverage": 0.8333333333333334, "early_accept": 26, "early_reject": 4, "false_accept": 1, "false_accept_examples": [ { "anchor": 0.5, "cell": "tp1_mns64", "label_feasible": false, "probability_feasible": 0.8578028168458919, "remaining_h20_hours": 0.0014445317586111114 } ], "false_reject": 0, "false_reject_examples": [], "full_trial_h20_hours": 1.0629133138080555, "invalidly_saved_h20_hours": 0.0014445317586111114, "remaining_h20_hours_at_cutoff": 0.7379133138080555, "saved_h20_hours_if_decisions_used": 0.6724100776402777, "threshold": 0.8, "valid_cost_reduction_fraction": null, "valid_zero_error_policy": false }, { "abstain_continue_full": 11, "correctly_saved_h20_hours": 0.626660078601389, "decision_coverage": 0.6944444444444444, "early_accept": 22, "early_reject": 3, "false_accept": 0, "false_accept_examples": [], "false_reject": 0, "false_reject_examples": [], "full_trial_h20_hours": 1.0629133138080555, "invalidly_saved_h20_hours": 0.0, "remaining_h20_hours_at_cutoff": 0.7379133138080555, "saved_h20_hours_if_decisions_used": 0.626660078601389, "threshold": 0.9, "valid_cost_reduction_fraction": 0.5895683782116529, "valid_zero_error_policy": true }, { "abstain_continue_full": 16, "correctly_saved_h20_hours": 0.5494646261305556, "decision_coverage": 0.5555555555555556, "early_accept": 17, "early_reject": 3, "false_accept": 0, "false_accept_examples": [], "false_reject": 0, "false_reject_examples": [], "full_trial_h20_hours": 1.0629133138080555, "invalidly_saved_h20_hours": 0.0, "remaining_h20_hours_at_cutoff": 0.7379133138080555, "saved_h20_hours_if_decisions_used": 0.5494646261305556, "threshold": 0.95, "valid_cost_reduction_fraction": 0.5169420864266074, "valid_zero_error_policy": true } ] }, "paired_correctness": { "both_correct": 31, "both_wrong": 2, "instrumentation_only_correct": 2, "mcnemar_exact_two_sided_p": 1.0, "outcome_only_correct": 1 } }, "10.0": { "bootstrap": { "accuracy_delta_instrumentation_minus_outcome": { "ci95": [ 0.0, 0.17142857142857137 ], "point": 0.08333333333333326 }, "brier_delta_instrumentation_minus_outcome": { "ci95": [ -0.051136314558473485, -0.009942733239401492 ], "point": -0.029809575955846893 }, "replicates": 10000, "seed": 20260714, "semantics": "group bootstrap over cells; diagnostic confidence interval" }, "instrumentation_aware": { "classification": { "accuracy": 0.8888888888888888, "balanced_accuracy": 0.8392857142857143, "brier": 0.07023194254174596, "confusion": { "false_negative": 2, "false_positive": 2, "true_negative": 6, "true_positive": 26 }, "log_loss": 0.23287532021698182 }, "policies": [ { "abstain_continue_full": 5, "correctly_saved_h20_hours": 0.6924315406702778, "decision_coverage": 0.8611111111111112, "early_accept": 27, "early_reject": 4, "false_accept": 1, "false_accept_examples": [ { "anchor": 0.75, "cell": "tp2_mns64", "label_feasible": false, "probability_feasible": 0.801686701254931, "remaining_h20_hours": 0.004004611363888889 } ], "false_reject": 0, "false_reject_examples": [], "full_trial_h20_hours": 1.0629133138080555, "invalidly_saved_h20_hours": 0.004004611363888889, "remaining_h20_hours_at_cutoff": 0.7379133138080555, "saved_h20_hours_if_decisions_used": 0.6964361520341666, "threshold": 0.8, "valid_cost_reduction_fraction": null, "valid_zero_error_policy": false }, { "abstain_continue_full": 14, "correctly_saved_h20_hours": 0.5771830138255556, "decision_coverage": 0.6111111111111112, "early_accept": 19, "early_reject": 3, "false_accept": 0, "false_accept_examples": [], "false_reject": 0, "false_reject_examples": [], "full_trial_h20_hours": 1.0629133138080555, "invalidly_saved_h20_hours": 0.0, "remaining_h20_hours_at_cutoff": 0.7379133138080555, "saved_h20_hours_if_decisions_used": 0.5771830138255556, "threshold": 0.9, "valid_cost_reduction_fraction": 0.5430198364509199, "valid_zero_error_policy": true }, { "abstain_continue_full": 26, "correctly_saved_h20_hours": 0.30550869967055555, "decision_coverage": 0.2777777777777778, "early_accept": 8, "early_reject": 2, "false_accept": 0, "false_accept_examples": [], "false_reject": 0, "false_reject_examples": [], "full_trial_h20_hours": 1.0629133138080555, "invalidly_saved_h20_hours": 0.0, "remaining_h20_hours_at_cutoff": 0.7379133138080555, "saved_h20_hours_if_decisions_used": 0.30550869967055555, "threshold": 0.95, "valid_cost_reduction_fraction": 0.2874257907034979, "valid_zero_error_policy": true } ] }, "outcome_only": { "classification": { "accuracy": 0.8055555555555556, "balanced_accuracy": 0.6517857142857143, "brier": 0.10004151849759285, "confusion": { "false_negative": 2, "false_positive": 5, "true_negative": 3, "true_positive": 26 }, "log_loss": 0.32055333394171265 }, "policies": [ { "abstain_continue_full": 7, "correctly_saved_h20_hours": 0.6877041925958333, "decision_coverage": 0.8055555555555556, "early_accept": 27, "early_reject": 2, "false_accept": 1, "false_accept_examples": [ { "anchor": 0.49609375, "cell": "tp2_mns8", "label_feasible": false, "probability_feasible": 0.8140722809073944, "remaining_h20_hours": 0.004562354750555555 } ], "false_reject": 0, "false_reject_examples": [], "full_trial_h20_hours": 1.0629133138080555, "invalidly_saved_h20_hours": 0.004562354750555555, "remaining_h20_hours_at_cutoff": 0.7379133138080555, "saved_h20_hours_if_decisions_used": 0.6922665473463888, "threshold": 0.8, "valid_cost_reduction_fraction": null, "valid_zero_error_policy": false }, { "abstain_continue_full": 24, "correctly_saved_h20_hours": 0.4570601465275, "decision_coverage": 0.3333333333333333, "early_accept": 11, "early_reject": 1, "false_accept": 0, "false_accept_examples": [], "false_reject": 0, "false_reject_examples": [], "full_trial_h20_hours": 1.0629133138080555, "invalidly_saved_h20_hours": 0.0, "remaining_h20_hours_at_cutoff": 0.7379133138080555, "saved_h20_hours_if_decisions_used": 0.4570601465275, "threshold": 0.9, "valid_cost_reduction_fraction": 0.43000698230978923, "valid_zero_error_policy": true }, { "abstain_continue_full": 33, "correctly_saved_h20_hours": 0.15267571839666666, "decision_coverage": 0.08333333333333333, "early_accept": 3, "early_reject": 0, "false_accept": 0, "false_accept_examples": [], "false_reject": 0, "false_reject_examples": [], "full_trial_h20_hours": 1.0629133138080555, "invalidly_saved_h20_hours": 0.0, "remaining_h20_hours_at_cutoff": 0.7379133138080555, "saved_h20_hours_if_decisions_used": 0.15267571839666666, "threshold": 0.95, "valid_cost_reduction_fraction": 0.14363891806913368, "valid_zero_error_policy": true } ] }, "paired_correctness": { "both_correct": 29, "both_wrong": 4, "instrumentation_only_correct": 3, "mcnemar_exact_two_sided_p": 0.25, "outcome_only_correct": 0 } } }, "remaining_h20_hours": { "distinct_n": 36, "max": 0.05164427065888889, "min": 0.0014445317586111114, "n": 36 } }, "20": { "cells": 12, "completion_time_sources": { "reconstructed_from_latency": 36 }, "examples": 36, "headline": { "bootstrap": { "accuracy_delta_instrumentation_minus_outcome": { "ci95": [ 0.0, 0.12195121951219512 ], "point": 0.05555555555555547 }, "brier_delta_instrumentation_minus_outcome": { "ci95": [ -0.09816256823569453, -0.009602741441109449 ], "point": -0.04807041919163683 }, "replicates": 10000, "seed": 20260714, "semantics": "group bootstrap over cells; diagnostic confidence interval" }, "instrumentation_aware": { "classification": { "accuracy": 0.9166666666666666, "balanced_accuracy": 0.8571428571428572, "brier": 0.05902138516195827, "confusion": { "false_negative": 1, "false_positive": 2, "true_negative": 6, "true_positive": 27 }, "log_loss": 0.24132948833387438 }, "policies": [ { "abstain_continue_full": 2, "correctly_saved_h20_hours": 0.6191580696761111, "decision_coverage": 0.9444444444444444, "early_accept": 28, "early_reject": 6, "false_accept": 1, "false_accept_examples": [ { "anchor": 0.49609375, "cell": "tp2_mns8", "label_feasible": false, "probability_feasible": 0.8399807697230681, "remaining_h20_hours": 0.0017845769727777771 } ], "false_reject": 1, "false_reject_examples": [ { "anchor": 0.033717411016, "cell": "tp4_mns32", "label_feasible": true, "probability_feasible": 0.008919034978436878, "remaining_h20_hours": 0.0073548573700000005 } ], "full_trial_h20_hours": 1.0629133138080555, "invalidly_saved_h20_hours": 0.009139434342777777, "remaining_h20_hours_at_cutoff": 0.6295799804747222, "saved_h20_hours_if_decisions_used": 0.6282975040188888, "threshold": 0.8, "valid_cost_reduction_fraction": null, "valid_zero_error_policy": false }, { "abstain_continue_full": 5, "correctly_saved_h20_hours": 0.6101058238716667, "decision_coverage": 0.8611111111111112, "early_accept": 27, "early_reject": 4, "false_accept": 0, "false_accept_examples": [], "false_reject": 1, "false_reject_examples": [ { "anchor": 0.033717411016, "cell": "tp4_mns32", "label_feasible": true, "probability_feasible": 0.008919034978436878, "remaining_h20_hours": 0.0073548573700000005 } ], "full_trial_h20_hours": 1.0629133138080555, "invalidly_saved_h20_hours": 0.0073548573700000005, "remaining_h20_hours_at_cutoff": 0.6295799804747222, "saved_h20_hours_if_decisions_used": 0.6174606812416666, "threshold": 0.9, "valid_cost_reduction_fraction": null, "valid_zero_error_policy": false }, { "abstain_continue_full": 9, "correctly_saved_h20_hours": 0.5390392875763889, "decision_coverage": 0.75, "early_accept": 23, "early_reject": 4, "false_accept": 0, "false_accept_examples": [], "false_reject": 1, "false_reject_examples": [ { "anchor": 0.033717411016, "cell": "tp4_mns32", "label_feasible": true, "probability_feasible": 0.008919034978436878, "remaining_h20_hours": 0.0073548573700000005 } ], "full_trial_h20_hours": 1.0629133138080555, "invalidly_saved_h20_hours": 0.0073548573700000005, "remaining_h20_hours_at_cutoff": 0.6295799804747222, "saved_h20_hours_if_decisions_used": 0.5463941449463888, "threshold": 0.95, "valid_cost_reduction_fraction": null, "valid_zero_error_policy": false } ] }, "outcome_only": { "classification": { "accuracy": 0.8611111111111112, "balanced_accuracy": 0.7321428571428572, "brier": 0.1070918043535951, "confusion": { "false_negative": 1, "false_positive": 4, "true_negative": 4, "true_positive": 27 }, "log_loss": 0.437208492871466 }, "policies": [ { "abstain_continue_full": 4, "correctly_saved_h20_hours": 0.5864084509805555, "decision_coverage": 0.8888888888888888, "early_accept": 28, "early_reject": 4, "false_accept": 3, "false_accept_examples": [ { "anchor": 0.5, "cell": "tp1_mns64", "label_feasible": false, "probability_feasible": 0.9419008835359353, "remaining_h20_hours": 5.564286972222258e-05 }, { "anchor": 0.75, "cell": "tp2_mns64", "label_feasible": false, "probability_feasible": 0.8529942983175343, "remaining_h20_hours": 0.0012268335861111111 }, { "anchor": 0.49609375, "cell": "tp2_mns8", "label_feasible": false, "probability_feasible": 0.8268198678404454, "remaining_h20_hours": 0.0017845769727777771 } ], "false_reject": 1, "false_reject_examples": [ { "anchor": 0.033717411016, "cell": "tp4_mns32", "label_feasible": true, "probability_feasible": 0.0017338236710752416, "remaining_h20_hours": 0.0073548573700000005 } ], "full_trial_h20_hours": 1.0629133138080555, "invalidly_saved_h20_hours": 0.01042191079861111, "remaining_h20_hours_at_cutoff": 0.6295799804747222, "saved_h20_hours_if_decisions_used": 0.5968303617791667, "threshold": 0.8, "valid_cost_reduction_fraction": null, "valid_zero_error_policy": false }, { "abstain_continue_full": 8, "correctly_saved_h20_hours": 0.5682670563286111, "decision_coverage": 0.7777777777777778, "early_accept": 25, "early_reject": 3, "false_accept": 1, "false_accept_examples": [ { "anchor": 0.5, "cell": "tp1_mns64", "label_feasible": false, "probability_feasible": 0.9419008835359353, "remaining_h20_hours": 5.564286972222258e-05 } ], "false_reject": 1, "false_reject_examples": [ { "anchor": 0.033717411016, "cell": "tp4_mns32", "label_feasible": true, "probability_feasible": 0.0017338236710752416, "remaining_h20_hours": 0.0073548573700000005 } ], "full_trial_h20_hours": 1.0629133138080555, "invalidly_saved_h20_hours": 0.007410500239722223, "remaining_h20_hours_at_cutoff": 0.6295799804747222, "saved_h20_hours_if_decisions_used": 0.5756775565683333, "threshold": 0.9, "valid_cost_reduction_fraction": null, "valid_zero_error_policy": false }, { "abstain_continue_full": 17, "correctly_saved_h20_hours": 0.40106617994777777, "decision_coverage": 0.5277777777777778, "early_accept": 16, "early_reject": 3, "false_accept": 0, "false_accept_examples": [], "false_reject": 1, "false_reject_examples": [ { "anchor": 0.033717411016, "cell": "tp4_mns32", "label_feasible": true, "probability_feasible": 0.0017338236710752416, "remaining_h20_hours": 0.0073548573700000005 } ], "full_trial_h20_hours": 1.0629133138080555, "invalidly_saved_h20_hours": 0.0073548573700000005, "remaining_h20_hours_at_cutoff": 0.6295799804747222, "saved_h20_hours_if_decisions_used": 0.4084210373177778, "threshold": 0.95, "valid_cost_reduction_fraction": null, "valid_zero_error_policy": false } ] }, "paired_correctness": { "both_correct": 31, "both_wrong": 3, "instrumentation_only_correct": 2, "mcnemar_exact_two_sided_p": 0.5, "outcome_only_correct": 0 } }, "headline_regularization": 1.0, "label_sanity": { "distinct_n": 2, "max": 1.0, "min": 0.0, "n": 36, "negative": 8, "positive": 28, "primary_adjudicated_disagreements": 3 }, "regularization_sensitivity": { "0.1": { "bootstrap": { "accuracy_delta_instrumentation_minus_outcome": { "ci95": [ 0.0, 0.1842105263157895 ], "point": 0.08333333333333326 }, "brier_delta_instrumentation_minus_outcome": { "ci95": [ -0.1289106028344674, -0.00039980805027801426 ], "point": -0.0571263639718715 }, "replicates": 10000, "seed": 20260714, "semantics": "group bootstrap over cells; diagnostic confidence interval" }, "instrumentation_aware": { "classification": { "accuracy": 0.9444444444444444, "balanced_accuracy": 0.9196428571428572, "brier": 0.05738846072877349, "confusion": { "false_negative": 1, "false_positive": 1, "true_negative": 7, "true_positive": 27 }, "log_loss": 0.3180853188579326 }, "policies": [ { "abstain_continue_full": 2, "correctly_saved_h20_hours": 0.6191580696761111, "decision_coverage": 0.9444444444444444, "early_accept": 28, "early_reject": 6, "false_accept": 1, "false_accept_examples": [ { "anchor": 0.49609375, "cell": "tp2_mns8", "label_feasible": false, "probability_feasible": 0.8655614945720641, "remaining_h20_hours": 0.0017845769727777771 } ], "false_reject": 1, "false_reject_examples": [ { "anchor": 0.033717411016, "cell": "tp4_mns32", "label_feasible": true, "probability_feasible": 0.00028842593944516384, "remaining_h20_hours": 0.0073548573700000005 } ], "full_trial_h20_hours": 1.0629133138080555, "invalidly_saved_h20_hours": 0.009139434342777777, "remaining_h20_hours_at_cutoff": 0.6295799804747222, "saved_h20_hours_if_decisions_used": 0.6282975040188888, "threshold": 0.8, "valid_cost_reduction_fraction": null, "valid_zero_error_policy": false }, { "abstain_continue_full": 3, "correctly_saved_h20_hours": 0.6191580696761111, "decision_coverage": 0.9166666666666666, "early_accept": 27, "early_reject": 6, "false_accept": 0, "false_accept_examples": [], "false_reject": 1, "false_reject_examples": [ { "anchor": 0.033717411016, "cell": "tp4_mns32", "label_feasible": true, "probability_feasible": 0.00028842593944516384, "remaining_h20_hours": 0.0073548573700000005 } ], "full_trial_h20_hours": 1.0629133138080555, "invalidly_saved_h20_hours": 0.0073548573700000005, "remaining_h20_hours_at_cutoff": 0.6295799804747222, "saved_h20_hours_if_decisions_used": 0.6265129270461112, "threshold": 0.9, "valid_cost_reduction_fraction": null, "valid_zero_error_policy": false }, { "abstain_continue_full": 4, "correctly_saved_h20_hours": 0.6074562703719444, "decision_coverage": 0.8888888888888888, "early_accept": 26, "early_reject": 6, "false_accept": 0, "false_accept_examples": [], "false_reject": 1, "false_reject_examples": [ { "anchor": 0.033717411016, "cell": "tp4_mns32", "label_feasible": true, "probability_feasible": 0.00028842593944516384, "remaining_h20_hours": 0.0073548573700000005 } ], "full_trial_h20_hours": 1.0629133138080555, "invalidly_saved_h20_hours": 0.0073548573700000005, "remaining_h20_hours_at_cutoff": 0.6295799804747222, "saved_h20_hours_if_decisions_used": 0.6148111277419444, "threshold": 0.95, "valid_cost_reduction_fraction": null, "valid_zero_error_policy": false } ] }, "outcome_only": { "classification": { "accuracy": 0.8611111111111112, "balanced_accuracy": 0.7767857142857143, "brier": 0.11451482470064499, "confusion": { "false_negative": 2, "false_positive": 3, "true_negative": 5, "true_positive": 26 }, "log_loss": 0.7747412375287079 }, "policies": [ { "abstain_continue_full": 3, "correctly_saved_h20_hours": 0.5848221568852777, "decision_coverage": 0.9166666666666666, "early_accept": 27, "early_reject": 6, "false_accept": 2, "false_accept_examples": [ { "anchor": 0.5, "cell": "tp1_mns64", "label_feasible": false, "probability_feasible": 0.9999405808371276, "remaining_h20_hours": 5.564286972222258e-05 }, { "anchor": 0.75, "cell": "tp2_mns64", "label_feasible": false, "probability_feasible": 0.9739951015698483, "remaining_h20_hours": 0.0012268335861111111 } ], "false_reject": 1, "false_reject_examples": [ { "anchor": 0.033717411016, "cell": "tp4_mns32", "label_feasible": true, "probability_feasible": 1.6376668165938014e-05, "remaining_h20_hours": 0.0073548573700000005 } ], "full_trial_h20_hours": 1.0629133138080555, "invalidly_saved_h20_hours": 0.008637333825833335, "remaining_h20_hours_at_cutoff": 0.6295799804747222, "saved_h20_hours_if_decisions_used": 0.5934594907111111, "threshold": 0.8, "valid_cost_reduction_fraction": null, "valid_zero_error_policy": false }, { "abstain_continue_full": 4, "correctly_saved_h20_hours": 0.5791295938452777, "decision_coverage": 0.8888888888888888, "early_accept": 27, "early_reject": 5, "false_accept": 2, "false_accept_examples": [ { "anchor": 0.5, "cell": "tp1_mns64", "label_feasible": false, "probability_feasible": 0.9999405808371276, "remaining_h20_hours": 5.564286972222258e-05 }, { "anchor": 0.75, "cell": "tp2_mns64", "label_feasible": false, "probability_feasible": 0.9739951015698483, "remaining_h20_hours": 0.0012268335861111111 } ], "false_reject": 1, "false_reject_examples": [ { "anchor": 0.033717411016, "cell": "tp4_mns32", "label_feasible": true, "probability_feasible": 1.6376668165938014e-05, "remaining_h20_hours": 0.0073548573700000005 } ], "full_trial_h20_hours": 1.0629133138080555, "invalidly_saved_h20_hours": 0.008637333825833335, "remaining_h20_hours_at_cutoff": 0.6295799804747222, "saved_h20_hours_if_decisions_used": 0.5877669276711112, "threshold": 0.9, "valid_cost_reduction_fraction": null, "valid_zero_error_policy": false }, { "abstain_continue_full": 5, "correctly_saved_h20_hours": 0.5676638636697222, "decision_coverage": 0.8611111111111112, "early_accept": 26, "early_reject": 5, "false_accept": 2, "false_accept_examples": [ { "anchor": 0.5, "cell": "tp1_mns64", "label_feasible": false, "probability_feasible": 0.9999405808371276, "remaining_h20_hours": 5.564286972222258e-05 }, { "anchor": 0.75, "cell": "tp2_mns64", "label_feasible": false, "probability_feasible": 0.9739951015698483, "remaining_h20_hours": 0.0012268335861111111 } ], "false_reject": 1, "false_reject_examples": [ { "anchor": 0.033717411016, "cell": "tp4_mns32", "label_feasible": true, "probability_feasible": 1.6376668165938014e-05, "remaining_h20_hours": 0.0073548573700000005 } ], "full_trial_h20_hours": 1.0629133138080555, "invalidly_saved_h20_hours": 0.008637333825833335, "remaining_h20_hours_at_cutoff": 0.6295799804747222, "saved_h20_hours_if_decisions_used": 0.5763011974955555, "threshold": 0.95, "valid_cost_reduction_fraction": null, "valid_zero_error_policy": false } ] }, "paired_correctness": { "both_correct": 31, "both_wrong": 2, "instrumentation_only_correct": 3, "mcnemar_exact_two_sided_p": 0.25, "outcome_only_correct": 0 } }, "1.0": { "bootstrap": { "accuracy_delta_instrumentation_minus_outcome": { "ci95": [ 0.0, 0.12195121951219512 ], "point": 0.05555555555555547 }, "brier_delta_instrumentation_minus_outcome": { "ci95": [ -0.09816256823569453, -0.009602741441109449 ], "point": -0.04807041919163683 }, "replicates": 10000, "seed": 20260714, "semantics": "group bootstrap over cells; diagnostic confidence interval" }, "instrumentation_aware": { "classification": { "accuracy": 0.9166666666666666, "balanced_accuracy": 0.8571428571428572, "brier": 0.05902138516195827, "confusion": { "false_negative": 1, "false_positive": 2, "true_negative": 6, "true_positive": 27 }, "log_loss": 0.24132948833387438 }, "policies": [ { "abstain_continue_full": 2, "correctly_saved_h20_hours": 0.6191580696761111, "decision_coverage": 0.9444444444444444, "early_accept": 28, "early_reject": 6, "false_accept": 1, "false_accept_examples": [ { "anchor": 0.49609375, "cell": "tp2_mns8", "label_feasible": false, "probability_feasible": 0.8399807697230681, "remaining_h20_hours": 0.0017845769727777771 } ], "false_reject": 1, "false_reject_examples": [ { "anchor": 0.033717411016, "cell": "tp4_mns32", "label_feasible": true, "probability_feasible": 0.008919034978436878, "remaining_h20_hours": 0.0073548573700000005 } ], "full_trial_h20_hours": 1.0629133138080555, "invalidly_saved_h20_hours": 0.009139434342777777, "remaining_h20_hours_at_cutoff": 0.6295799804747222, "saved_h20_hours_if_decisions_used": 0.6282975040188888, "threshold": 0.8, "valid_cost_reduction_fraction": null, "valid_zero_error_policy": false }, { "abstain_continue_full": 5, "correctly_saved_h20_hours": 0.6101058238716667, "decision_coverage": 0.8611111111111112, "early_accept": 27, "early_reject": 4, "false_accept": 0, "false_accept_examples": [], "false_reject": 1, "false_reject_examples": [ { "anchor": 0.033717411016, "cell": "tp4_mns32", "label_feasible": true, "probability_feasible": 0.008919034978436878, "remaining_h20_hours": 0.0073548573700000005 } ], "full_trial_h20_hours": 1.0629133138080555, "invalidly_saved_h20_hours": 0.0073548573700000005, "remaining_h20_hours_at_cutoff": 0.6295799804747222, "saved_h20_hours_if_decisions_used": 0.6174606812416666, "threshold": 0.9, "valid_cost_reduction_fraction": null, "valid_zero_error_policy": false }, { "abstain_continue_full": 9, "correctly_saved_h20_hours": 0.5390392875763889, "decision_coverage": 0.75, "early_accept": 23, "early_reject": 4, "false_accept": 0, "false_accept_examples": [], "false_reject": 1, "false_reject_examples": [ { "anchor": 0.033717411016, "cell": "tp4_mns32", "label_feasible": true, "probability_feasible": 0.008919034978436878, "remaining_h20_hours": 0.0073548573700000005 } ], "full_trial_h20_hours": 1.0629133138080555, "invalidly_saved_h20_hours": 0.0073548573700000005, "remaining_h20_hours_at_cutoff": 0.6295799804747222, "saved_h20_hours_if_decisions_used": 0.5463941449463888, "threshold": 0.95, "valid_cost_reduction_fraction": null, "valid_zero_error_policy": false } ] }, "outcome_only": { "classification": { "accuracy": 0.8611111111111112, "balanced_accuracy": 0.7321428571428572, "brier": 0.1070918043535951, "confusion": { "false_negative": 1, "false_positive": 4, "true_negative": 4, "true_positive": 27 }, "log_loss": 0.437208492871466 }, "policies": [ { "abstain_continue_full": 4, "correctly_saved_h20_hours": 0.5864084509805555, "decision_coverage": 0.8888888888888888, "early_accept": 28, "early_reject": 4, "false_accept": 3, "false_accept_examples": [ { "anchor": 0.5, "cell": "tp1_mns64", "label_feasible": false, "probability_feasible": 0.9419008835359353, "remaining_h20_hours": 5.564286972222258e-05 }, { "anchor": 0.75, "cell": "tp2_mns64", "label_feasible": false, "probability_feasible": 0.8529942983175343, "remaining_h20_hours": 0.0012268335861111111 }, { "anchor": 0.49609375, "cell": "tp2_mns8", "label_feasible": false, "probability_feasible": 0.8268198678404454, "remaining_h20_hours": 0.0017845769727777771 } ], "false_reject": 1, "false_reject_examples": [ { "anchor": 0.033717411016, "cell": "tp4_mns32", "label_feasible": true, "probability_feasible": 0.0017338236710752416, "remaining_h20_hours": 0.0073548573700000005 } ], "full_trial_h20_hours": 1.0629133138080555, "invalidly_saved_h20_hours": 0.01042191079861111, "remaining_h20_hours_at_cutoff": 0.6295799804747222, "saved_h20_hours_if_decisions_used": 0.5968303617791667, "threshold": 0.8, "valid_cost_reduction_fraction": null, "valid_zero_error_policy": false }, { "abstain_continue_full": 8, "correctly_saved_h20_hours": 0.5682670563286111, "decision_coverage": 0.7777777777777778, "early_accept": 25, "early_reject": 3, "false_accept": 1, "false_accept_examples": [ { "anchor": 0.5, "cell": "tp1_mns64", "label_feasible": false, "probability_feasible": 0.9419008835359353, "remaining_h20_hours": 5.564286972222258e-05 } ], "false_reject": 1, "false_reject_examples": [ { "anchor": 0.033717411016, "cell": "tp4_mns32", "label_feasible": true, "probability_feasible": 0.0017338236710752416, "remaining_h20_hours": 0.0073548573700000005 } ], "full_trial_h20_hours": 1.0629133138080555, "invalidly_saved_h20_hours": 0.007410500239722223, "remaining_h20_hours_at_cutoff": 0.6295799804747222, "saved_h20_hours_if_decisions_used": 0.5756775565683333, "threshold": 0.9, "valid_cost_reduction_fraction": null, "valid_zero_error_policy": false }, { "abstain_continue_full": 17, "correctly_saved_h20_hours": 0.40106617994777777, "decision_coverage": 0.5277777777777778, "early_accept": 16, "early_reject": 3, "false_accept": 0, "false_accept_examples": [], "false_reject": 1, "false_reject_examples": [ { "anchor": 0.033717411016, "cell": "tp4_mns32", "label_feasible": true, "probability_feasible": 0.0017338236710752416, "remaining_h20_hours": 0.0073548573700000005 } ], "full_trial_h20_hours": 1.0629133138080555, "invalidly_saved_h20_hours": 0.0073548573700000005, "remaining_h20_hours_at_cutoff": 0.6295799804747222, "saved_h20_hours_if_decisions_used": 0.4084210373177778, "threshold": 0.95, "valid_cost_reduction_fraction": null, "valid_zero_error_policy": false } ] }, "paired_correctness": { "both_correct": 31, "both_wrong": 3, "instrumentation_only_correct": 2, "mcnemar_exact_two_sided_p": 0.5, "outcome_only_correct": 0 } }, "10.0": { "bootstrap": { "accuracy_delta_instrumentation_minus_outcome": { "ci95": [ 0.0, 0.17142857142857137 ], "point": 0.08333333333333326 }, "brier_delta_instrumentation_minus_outcome": { "ci95": [ -0.047457441513371, -0.01422543161645023 ], "point": -0.029913894255621076 }, "replicates": 10000, "seed": 20260714, "semantics": "group bootstrap over cells; diagnostic confidence interval" }, "instrumentation_aware": { "classification": { "accuracy": 0.9166666666666666, "balanced_accuracy": 0.8571428571428572, "brier": 0.07346598213380073, "confusion": { "false_negative": 1, "false_positive": 2, "true_negative": 6, "true_positive": 27 }, "log_loss": 0.24820863982463115 }, "policies": [ { "abstain_continue_full": 5, "correctly_saved_h20_hours": 0.6026586119891666, "decision_coverage": 0.8611111111111112, "early_accept": 27, "early_reject": 4, "false_accept": 1, "false_accept_examples": [ { "anchor": 0.49609375, "cell": "tp2_mns8", "label_feasible": false, "probability_feasible": 0.8171600256181173, "remaining_h20_hours": 0.0017845769727777771 } ], "false_reject": 1, "false_reject_examples": [ { "anchor": 0.033717411016, "cell": "tp4_mns32", "label_feasible": true, "probability_feasible": 0.08736122237187642, "remaining_h20_hours": 0.0073548573700000005 } ], "full_trial_h20_hours": 1.0629133138080555, "invalidly_saved_h20_hours": 0.009139434342777777, "remaining_h20_hours_at_cutoff": 0.6295799804747222, "saved_h20_hours_if_decisions_used": 0.6117980463319445, "threshold": 0.8, "valid_cost_reduction_fraction": null, "valid_zero_error_policy": false }, { "abstain_continue_full": 8, "correctly_saved_h20_hours": 0.5909011698152777, "decision_coverage": 0.7777777777777778, "early_accept": 25, "early_reject": 3, "false_accept": 0, "false_accept_examples": [], "false_reject": 1, "false_reject_examples": [ { "anchor": 0.033717411016, "cell": "tp4_mns32", "label_feasible": true, "probability_feasible": 0.08736122237187642, "remaining_h20_hours": 0.0073548573700000005 } ], "full_trial_h20_hours": 1.0629133138080555, "invalidly_saved_h20_hours": 0.0073548573700000005, "remaining_h20_hours_at_cutoff": 0.6295799804747222, "saved_h20_hours_if_decisions_used": 0.5982560271852778, "threshold": 0.9, "valid_cost_reduction_fraction": null, "valid_zero_error_policy": false }, { "abstain_continue_full": 27, "correctly_saved_h20_hours": 0.27948134858055557, "decision_coverage": 0.25, "early_accept": 9, "early_reject": 0, "false_accept": 0, "false_accept_examples": [], "false_reject": 0, "false_reject_examples": [], "full_trial_h20_hours": 1.0629133138080555, "invalidly_saved_h20_hours": 0.0, "remaining_h20_hours_at_cutoff": 0.6295799804747222, "saved_h20_hours_if_decisions_used": 0.27948134858055557, "threshold": 0.95, "valid_cost_reduction_fraction": 0.26293898566315754, "valid_zero_error_policy": true } ] }, "outcome_only": { "classification": { "accuracy": 0.8333333333333334, "balanced_accuracy": 0.7142857142857143, "brier": 0.10337987638942181, "confusion": { "false_negative": 2, "false_positive": 4, "true_negative": 4, "true_positive": 26 }, "log_loss": 0.34599307331049317 }, "policies": [ { "abstain_continue_full": 7, "correctly_saved_h20_hours": 0.6007090416925, "decision_coverage": 0.8055555555555556, "early_accept": 27, "early_reject": 2, "false_accept": 1, "false_accept_examples": [ { "anchor": 0.49609375, "cell": "tp2_mns8", "label_feasible": false, "probability_feasible": 0.8140697427804628, "remaining_h20_hours": 0.0017845769727777771 } ], "false_reject": 1, "false_reject_examples": [ { "anchor": 0.033717411016, "cell": "tp4_mns32", "label_feasible": true, "probability_feasible": 0.0444153615651932, "remaining_h20_hours": 0.0073548573700000005 } ], "full_trial_h20_hours": 1.0629133138080555, "invalidly_saved_h20_hours": 0.009139434342777777, "remaining_h20_hours_at_cutoff": 0.6295799804747222, "saved_h20_hours_if_decisions_used": 0.6098484760352777, "threshold": 0.8, "valid_cost_reduction_fraction": null, "valid_zero_error_policy": false }, { "abstain_continue_full": 12, "correctly_saved_h20_hours": 0.5199227922302778, "decision_coverage": 0.6666666666666666, "early_accept": 23, "early_reject": 1, "false_accept": 0, "false_accept_examples": [], "false_reject": 1, "false_reject_examples": [ { "anchor": 0.033717411016, "cell": "tp4_mns32", "label_feasible": true, "probability_feasible": 0.0444153615651932, "remaining_h20_hours": 0.0073548573700000005 } ], "full_trial_h20_hours": 1.0629133138080555, "invalidly_saved_h20_hours": 0.0073548573700000005, "remaining_h20_hours_at_cutoff": 0.6295799804747222, "saved_h20_hours_if_decisions_used": 0.5272776496002778, "threshold": 0.9, "valid_cost_reduction_fraction": null, "valid_zero_error_policy": false }, { "abstain_continue_full": 33, "correctly_saved_h20_hours": 0.08992033662666665, "decision_coverage": 0.08333333333333333, "early_accept": 2, "early_reject": 1, "false_accept": 0, "false_accept_examples": [], "false_reject": 1, "false_reject_examples": [ { "anchor": 0.033717411016, "cell": "tp4_mns32", "label_feasible": true, "probability_feasible": 0.0444153615651932, "remaining_h20_hours": 0.0073548573700000005 } ], "full_trial_h20_hours": 1.0629133138080555, "invalidly_saved_h20_hours": 0.0073548573700000005, "remaining_h20_hours_at_cutoff": 0.6295799804747222, "saved_h20_hours_if_decisions_used": 0.09727519399666666, "threshold": 0.95, "valid_cost_reduction_fraction": null, "valid_zero_error_policy": false } ] }, "paired_correctness": { "both_correct": 30, "both_wrong": 3, "instrumentation_only_correct": 3, "mcnemar_exact_two_sided_p": 0.25, "outcome_only_correct": 0 } } }, "remaining_h20_hours": { "distinct_n": 36, "max": 0.04608871510333333, "min": 5.564286972222258e-05, "n": 36 } }, "5": { "cells": 12, "completion_time_sources": { "none_completed": 2, "reconstructed_from_latency": 35 }, "examples": 37, "headline": { "bootstrap": { "accuracy_delta_instrumentation_minus_outcome": { "ci95": [ 0.02777777777777779, 0.19999999999999996 ], "point": 0.10810810810810811 }, "brier_delta_instrumentation_minus_outcome": { "ci95": [ -0.08739451923629489, 0.010803161364314168 ], "point": -0.03964626050589401 }, "replicates": 10000, "seed": 20260714, "semantics": "group bootstrap over cells; diagnostic confidence interval" }, "instrumentation_aware": { "classification": { "accuracy": 0.8918918918918919, "balanced_accuracy": 0.8154761904761905, "brier": 0.09006542265758688, "confusion": { "false_negative": 1, "false_positive": 3, "true_negative": 6, "true_positive": 27 }, "log_loss": 0.2730810945072346 }, "policies": [ { "abstain_continue_full": 9, "correctly_saved_h20_hours": 0.7828422774630556, "decision_coverage": 0.7567567567567568, "early_accept": 23, "early_reject": 5, "false_accept": 1, "false_accept_examples": [ { "anchor": 0.49609375, "cell": "tp2_mns8", "label_feasible": false, "probability_feasible": 0.918171067470718, "remaining_h20_hours": 0.010117910306111111 } ], "false_reject": 0, "false_reject_examples": [], "full_trial_h20_hours": 1.0669595034675, "invalidly_saved_h20_hours": 0.010117910306111111, "remaining_h20_hours_at_cutoff": 0.957237281245278, "saved_h20_hours_if_decisions_used": 0.7929601877691667, "threshold": 0.8, "valid_cost_reduction_fraction": null, "valid_zero_error_policy": false }, { "abstain_continue_full": 12, "correctly_saved_h20_hours": 0.7136471349655555, "decision_coverage": 0.6756756756756757, "early_accept": 21, "early_reject": 4, "false_accept": 1, "false_accept_examples": [ { "anchor": 0.49609375, "cell": "tp2_mns8", "label_feasible": false, "probability_feasible": 0.918171067470718, "remaining_h20_hours": 0.010117910306111111 } ], "false_reject": 0, "false_reject_examples": [], "full_trial_h20_hours": 1.0669595034675, "invalidly_saved_h20_hours": 0.010117910306111111, "remaining_h20_hours_at_cutoff": 0.957237281245278, "saved_h20_hours_if_decisions_used": 0.7237650452716666, "threshold": 0.9, "valid_cost_reduction_fraction": null, "valid_zero_error_policy": false }, { "abstain_continue_full": 15, "correctly_saved_h20_hours": 0.6518691238655556, "decision_coverage": 0.5945945945945946, "early_accept": 18, "early_reject": 4, "false_accept": 0, "false_accept_examples": [], "false_reject": 0, "false_reject_examples": [], "full_trial_h20_hours": 1.0669595034675, "invalidly_saved_h20_hours": 0.0, "remaining_h20_hours_at_cutoff": 0.957237281245278, "saved_h20_hours_if_decisions_used": 0.6518691238655556, "threshold": 0.95, "valid_cost_reduction_fraction": 0.6109595741422738, "valid_zero_error_policy": true } ] }, "outcome_only": { "classification": { "accuracy": 0.7837837837837838, "balanced_accuracy": 0.7063492063492063, "brier": 0.1297116831634809, "confusion": { "false_negative": 4, "false_positive": 4, "true_negative": 5, "true_positive": 24 }, "log_loss": 0.38484417997476916 }, "policies": [ { "abstain_continue_full": 9, "correctly_saved_h20_hours": 0.745153891865, "decision_coverage": 0.7567567567567568, "early_accept": 24, "early_reject": 4, "false_accept": 3, "false_accept_examples": [ { "anchor": 0.75, "cell": "tp2_mns64", "label_feasible": false, "probability_feasible": 0.8051358474368825, "remaining_h20_hours": 0.009560166919444444 }, { "anchor": 0.49609375, "cell": "tp2_mns8", "label_feasible": false, "probability_feasible": 0.9221804334720108, "remaining_h20_hours": 0.010117910306111111 }, { "anchor": 0.034252608017, "cell": "tp4_mns8", "label_feasible": false, "probability_feasible": 0.8933807379877649, "remaining_h20_hours": 0.022359229706666667 } ], "false_reject": 0, "false_reject_examples": [], "full_trial_h20_hours": 1.0669595034675, "invalidly_saved_h20_hours": 0.042037306932222226, "remaining_h20_hours_at_cutoff": 0.957237281245278, "saved_h20_hours_if_decisions_used": 0.7871911987972221, "threshold": 0.8, "valid_cost_reduction_fraction": null, "valid_zero_error_policy": false }, { "abstain_continue_full": 14, "correctly_saved_h20_hours": 0.6362291990002777, "decision_coverage": 0.6216216216216216, "early_accept": 19, "early_reject": 4, "false_accept": 1, "false_accept_examples": [ { "anchor": 0.49609375, "cell": "tp2_mns8", "label_feasible": false, "probability_feasible": 0.9221804334720108, "remaining_h20_hours": 0.010117910306111111 } ], "false_reject": 0, "false_reject_examples": [], "full_trial_h20_hours": 1.0669595034675, "invalidly_saved_h20_hours": 0.010117910306111111, "remaining_h20_hours_at_cutoff": 0.957237281245278, "saved_h20_hours_if_decisions_used": 0.6463471093063888, "threshold": 0.9, "valid_cost_reduction_fraction": null, "valid_zero_error_policy": false }, { "abstain_continue_full": 21, "correctly_saved_h20_hours": 0.3878475556194444, "decision_coverage": 0.43243243243243246, "early_accept": 13, "early_reject": 3, "false_accept": 0, "false_accept_examples": [], "false_reject": 0, "false_reject_examples": [], "full_trial_h20_hours": 1.0669595034675, "invalidly_saved_h20_hours": 0.0, "remaining_h20_hours_at_cutoff": 0.957237281245278, "saved_h20_hours_if_decisions_used": 0.3878475556194444, "threshold": 0.95, "valid_cost_reduction_fraction": 0.3635072881013599, "valid_zero_error_policy": true } ] }, "paired_correctness": { "both_correct": 29, "both_wrong": 4, "instrumentation_only_correct": 4, "mcnemar_exact_two_sided_p": 0.125, "outcome_only_correct": 0 } }, "headline_regularization": 1.0, "label_sanity": { "distinct_n": 2, "max": 1.0, "min": 0.0, "n": 37, "negative": 9, "positive": 28, "primary_adjudicated_disagreements": 3 }, "regularization_sensitivity": { "0.1": { "bootstrap": { "accuracy_delta_instrumentation_minus_outcome": { "ci95": [ -0.06980897009966774, 0.21875 ], "point": 0.08108108108108103 }, "brier_delta_instrumentation_minus_outcome": { "ci95": [ -0.1649967051920624, 0.06447487686406284 ], "point": -0.05737762784013967 }, "replicates": 10000, "seed": 20260714, "semantics": "group bootstrap over cells; diagnostic confidence interval" }, "instrumentation_aware": { "classification": { "accuracy": 0.8378378378378378, "balanced_accuracy": 0.7797619047619048, "brier": 0.13521691422103324, "confusion": { "false_negative": 3, "false_positive": 3, "true_negative": 6, "true_positive": 25 }, "log_loss": 0.4511157566254769 }, "policies": [ { "abstain_continue_full": 3, "correctly_saved_h20_hours": 0.8122652196319445, "decision_coverage": 0.918918918918919, "early_accept": 26, "early_reject": 8, "false_accept": 3, "false_accept_examples": [ { "anchor": 0.75, "cell": "tp2_mns64", "label_feasible": false, "probability_feasible": 0.9405657729239719, "remaining_h20_hours": 0.009560166919444444 }, { "anchor": 0.49609375, "cell": "tp2_mns8", "label_feasible": false, "probability_feasible": 0.9783420846880035, "remaining_h20_hours": 0.010117910306111111 }, { "anchor": 0.033717411016, "cell": "tp4_mns16", "label_feasible": false, "probability_feasible": 0.9635702290378427, "remaining_h20_hours": 0.023106262014444445 } ], "false_reject": 2, "false_reject_examples": [ { "anchor": 0.21875, "cell": "tp1_mns8", "label_feasible": true, "probability_feasible": 0.07387488532278061, "remaining_h20_hours": 0.015631309493055554 }, { "anchor": 0.49609375, "cell": "tp2_mns16", "label_feasible": true, "probability_feasible": 0.17417966056624304, "remaining_h20_hours": 0.03127135866722222 } ], "full_trial_h20_hours": 1.0669595034675, "invalidly_saved_h20_hours": 0.08968700740027778, "remaining_h20_hours_at_cutoff": 0.957237281245278, "saved_h20_hours_if_decisions_used": 0.9019522270322223, "threshold": 0.8, "valid_cost_reduction_fraction": null, "valid_zero_error_policy": false }, { "abstain_continue_full": 6, "correctly_saved_h20_hours": 0.765429306841111, "decision_coverage": 0.8378378378378378, "early_accept": 24, "early_reject": 7, "false_accept": 3, "false_accept_examples": [ { "anchor": 0.75, "cell": "tp2_mns64", "label_feasible": false, "probability_feasible": 0.9405657729239719, "remaining_h20_hours": 0.009560166919444444 }, { "anchor": 0.49609375, "cell": "tp2_mns8", "label_feasible": false, "probability_feasible": 0.9783420846880035, "remaining_h20_hours": 0.010117910306111111 }, { "anchor": 0.033717411016, "cell": "tp4_mns16", "label_feasible": false, "probability_feasible": 0.9635702290378427, "remaining_h20_hours": 0.023106262014444445 } ], "false_reject": 1, "false_reject_examples": [ { "anchor": 0.21875, "cell": "tp1_mns8", "label_feasible": true, "probability_feasible": 0.07387488532278061, "remaining_h20_hours": 0.015631309493055554 } ], "full_trial_h20_hours": 1.0669595034675, "invalidly_saved_h20_hours": 0.05841564873305555, "remaining_h20_hours_at_cutoff": 0.957237281245278, "saved_h20_hours_if_decisions_used": 0.8238449555741666, "threshold": 0.9, "valid_cost_reduction_fraction": null, "valid_zero_error_policy": false }, { "abstain_continue_full": 9, "correctly_saved_h20_hours": 0.7469523484622221, "decision_coverage": 0.7567567567567568, "early_accept": 22, "early_reject": 6, "false_accept": 2, "false_accept_examples": [ { "anchor": 0.49609375, "cell": "tp2_mns8", "label_feasible": false, "probability_feasible": 0.9783420846880035, "remaining_h20_hours": 0.010117910306111111 }, { "anchor": 0.033717411016, "cell": "tp4_mns16", "label_feasible": false, "probability_feasible": 0.9635702290378427, "remaining_h20_hours": 0.023106262014444445 } ], "false_reject": 0, "false_reject_examples": [], "full_trial_h20_hours": 1.0669595034675, "invalidly_saved_h20_hours": 0.03322417232055556, "remaining_h20_hours_at_cutoff": 0.957237281245278, "saved_h20_hours_if_decisions_used": 0.7801765207827777, "threshold": 0.95, "valid_cost_reduction_fraction": null, "valid_zero_error_policy": false } ] }, "outcome_only": { "classification": { "accuracy": 0.7567567567567568, "balanced_accuracy": 0.7261904761904762, "brier": 0.1925945420611729, "confusion": { "false_negative": 6, "false_positive": 3, "true_negative": 6, "true_positive": 22 }, "log_loss": 0.7361068786964186 }, "policies": [ { "abstain_continue_full": 4, "correctly_saved_h20_hours": 0.7745768340338888, "decision_coverage": 0.8918918918918919, "early_accept": 25, "early_reject": 8, "false_accept": 3, "false_accept_examples": [ { "anchor": 0.75, "cell": "tp2_mns64", "label_feasible": false, "probability_feasible": 0.977871407650825, "remaining_h20_hours": 0.009560166919444444 }, { "anchor": 0.49609375, "cell": "tp2_mns8", "label_feasible": false, "probability_feasible": 0.9783309849310482, "remaining_h20_hours": 0.010117910306111111 }, { "anchor": 0.034252608017, "cell": "tp4_mns8", "label_feasible": false, "probability_feasible": 0.9900156648758626, "remaining_h20_hours": 0.022359229706666667 } ], "false_reject": 3, "false_reject_examples": [ { "anchor": 0.21875, "cell": "tp1_mns8", "label_feasible": true, "probability_feasible": 0.012882187111638552, "remaining_h20_hours": 0.015631309493055554 }, { "anchor": 0.2265625, "cell": "tp1_mns8", "label_feasible": true, "probability_feasible": 0.016042537020255383, "remaining_h20_hours": 0.01563306773861111 }, { "anchor": 0.23046875, "cell": "tp1_mns8", "label_feasible": true, "probability_feasible": 0.15321211351284938, "remaining_h20_hours": 0.015631133334166665 } ], "full_trial_h20_hours": 1.0669595034675, "invalidly_saved_h20_hours": 0.08893281749805555, "remaining_h20_hours_at_cutoff": 0.957237281245278, "saved_h20_hours_if_decisions_used": 0.8635096515319445, "threshold": 0.8, "valid_cost_reduction_fraction": null, "valid_zero_error_policy": false }, { "abstain_continue_full": 8, "correctly_saved_h20_hours": 0.6784465414575, "decision_coverage": 0.7837837837837838, "early_accept": 22, "early_reject": 7, "false_accept": 3, "false_accept_examples": [ { "anchor": 0.75, "cell": "tp2_mns64", "label_feasible": false, "probability_feasible": 0.977871407650825, "remaining_h20_hours": 0.009560166919444444 }, { "anchor": 0.49609375, "cell": "tp2_mns8", "label_feasible": false, "probability_feasible": 0.9783309849310482, "remaining_h20_hours": 0.010117910306111111 }, { "anchor": 0.034252608017, "cell": "tp4_mns8", "label_feasible": false, "probability_feasible": 0.9900156648758626, "remaining_h20_hours": 0.022359229706666667 } ], "false_reject": 2, "false_reject_examples": [ { "anchor": 0.21875, "cell": "tp1_mns8", "label_feasible": true, "probability_feasible": 0.012882187111638552, "remaining_h20_hours": 0.015631309493055554 }, { "anchor": 0.2265625, "cell": "tp1_mns8", "label_feasible": true, "probability_feasible": 0.016042537020255383, "remaining_h20_hours": 0.01563306773861111 } ], "full_trial_h20_hours": 1.0669595034675, "invalidly_saved_h20_hours": 0.0733016841638889, "remaining_h20_hours_at_cutoff": 0.957237281245278, "saved_h20_hours_if_decisions_used": 0.7517482256213891, "threshold": 0.9, "valid_cost_reduction_fraction": null, "valid_zero_error_policy": false }, { "abstain_continue_full": 9, "correctly_saved_h20_hours": 0.6471751827902779, "decision_coverage": 0.7567567567567568, "early_accept": 21, "early_reject": 7, "false_accept": 3, "false_accept_examples": [ { "anchor": 0.75, "cell": "tp2_mns64", "label_feasible": false, "probability_feasible": 0.977871407650825, "remaining_h20_hours": 0.009560166919444444 }, { "anchor": 0.49609375, "cell": "tp2_mns8", "label_feasible": false, "probability_feasible": 0.9783309849310482, "remaining_h20_hours": 0.010117910306111111 }, { "anchor": 0.034252608017, "cell": "tp4_mns8", "label_feasible": false, "probability_feasible": 0.9900156648758626, "remaining_h20_hours": 0.022359229706666667 } ], "false_reject": 2, "false_reject_examples": [ { "anchor": 0.21875, "cell": "tp1_mns8", "label_feasible": true, "probability_feasible": 0.012882187111638552, "remaining_h20_hours": 0.015631309493055554 }, { "anchor": 0.2265625, "cell": "tp1_mns8", "label_feasible": true, "probability_feasible": 0.016042537020255383, "remaining_h20_hours": 0.01563306773861111 } ], "full_trial_h20_hours": 1.0669595034675, "invalidly_saved_h20_hours": 0.0733016841638889, "remaining_h20_hours_at_cutoff": 0.957237281245278, "saved_h20_hours_if_decisions_used": 0.7204768669541667, "threshold": 0.95, "valid_cost_reduction_fraction": null, "valid_zero_error_policy": false } ] }, "paired_correctness": { "both_correct": 26, "both_wrong": 4, "instrumentation_only_correct": 5, "mcnemar_exact_two_sided_p": 0.453125, "outcome_only_correct": 2 } }, "1.0": { "bootstrap": { "accuracy_delta_instrumentation_minus_outcome": { "ci95": [ 0.02777777777777779, 0.19999999999999996 ], "point": 0.10810810810810811 }, "brier_delta_instrumentation_minus_outcome": { "ci95": [ -0.08739451923629489, 0.010803161364314168 ], "point": -0.03964626050589401 }, "replicates": 10000, "seed": 20260714, "semantics": "group bootstrap over cells; diagnostic confidence interval" }, "instrumentation_aware": { "classification": { "accuracy": 0.8918918918918919, "balanced_accuracy": 0.8154761904761905, "brier": 0.09006542265758688, "confusion": { "false_negative": 1, "false_positive": 3, "true_negative": 6, "true_positive": 27 }, "log_loss": 0.2730810945072346 }, "policies": [ { "abstain_continue_full": 9, "correctly_saved_h20_hours": 0.7828422774630556, "decision_coverage": 0.7567567567567568, "early_accept": 23, "early_reject": 5, "false_accept": 1, "false_accept_examples": [ { "anchor": 0.49609375, "cell": "tp2_mns8", "label_feasible": false, "probability_feasible": 0.918171067470718, "remaining_h20_hours": 0.010117910306111111 } ], "false_reject": 0, "false_reject_examples": [], "full_trial_h20_hours": 1.0669595034675, "invalidly_saved_h20_hours": 0.010117910306111111, "remaining_h20_hours_at_cutoff": 0.957237281245278, "saved_h20_hours_if_decisions_used": 0.7929601877691667, "threshold": 0.8, "valid_cost_reduction_fraction": null, "valid_zero_error_policy": false }, { "abstain_continue_full": 12, "correctly_saved_h20_hours": 0.7136471349655555, "decision_coverage": 0.6756756756756757, "early_accept": 21, "early_reject": 4, "false_accept": 1, "false_accept_examples": [ { "anchor": 0.49609375, "cell": "tp2_mns8", "label_feasible": false, "probability_feasible": 0.918171067470718, "remaining_h20_hours": 0.010117910306111111 } ], "false_reject": 0, "false_reject_examples": [], "full_trial_h20_hours": 1.0669595034675, "invalidly_saved_h20_hours": 0.010117910306111111, "remaining_h20_hours_at_cutoff": 0.957237281245278, "saved_h20_hours_if_decisions_used": 0.7237650452716666, "threshold": 0.9, "valid_cost_reduction_fraction": null, "valid_zero_error_policy": false }, { "abstain_continue_full": 15, "correctly_saved_h20_hours": 0.6518691238655556, "decision_coverage": 0.5945945945945946, "early_accept": 18, "early_reject": 4, "false_accept": 0, "false_accept_examples": [], "false_reject": 0, "false_reject_examples": [], "full_trial_h20_hours": 1.0669595034675, "invalidly_saved_h20_hours": 0.0, "remaining_h20_hours_at_cutoff": 0.957237281245278, "saved_h20_hours_if_decisions_used": 0.6518691238655556, "threshold": 0.95, "valid_cost_reduction_fraction": 0.6109595741422738, "valid_zero_error_policy": true } ] }, "outcome_only": { "classification": { "accuracy": 0.7837837837837838, "balanced_accuracy": 0.7063492063492063, "brier": 0.1297116831634809, "confusion": { "false_negative": 4, "false_positive": 4, "true_negative": 5, "true_positive": 24 }, "log_loss": 0.38484417997476916 }, "policies": [ { "abstain_continue_full": 9, "correctly_saved_h20_hours": 0.745153891865, "decision_coverage": 0.7567567567567568, "early_accept": 24, "early_reject": 4, "false_accept": 3, "false_accept_examples": [ { "anchor": 0.75, "cell": "tp2_mns64", "label_feasible": false, "probability_feasible": 0.8051358474368825, "remaining_h20_hours": 0.009560166919444444 }, { "anchor": 0.49609375, "cell": "tp2_mns8", "label_feasible": false, "probability_feasible": 0.9221804334720108, "remaining_h20_hours": 0.010117910306111111 }, { "anchor": 0.034252608017, "cell": "tp4_mns8", "label_feasible": false, "probability_feasible": 0.8933807379877649, "remaining_h20_hours": 0.022359229706666667 } ], "false_reject": 0, "false_reject_examples": [], "full_trial_h20_hours": 1.0669595034675, "invalidly_saved_h20_hours": 0.042037306932222226, "remaining_h20_hours_at_cutoff": 0.957237281245278, "saved_h20_hours_if_decisions_used": 0.7871911987972221, "threshold": 0.8, "valid_cost_reduction_fraction": null, "valid_zero_error_policy": false }, { "abstain_continue_full": 14, "correctly_saved_h20_hours": 0.6362291990002777, "decision_coverage": 0.6216216216216216, "early_accept": 19, "early_reject": 4, "false_accept": 1, "false_accept_examples": [ { "anchor": 0.49609375, "cell": "tp2_mns8", "label_feasible": false, "probability_feasible": 0.9221804334720108, "remaining_h20_hours": 0.010117910306111111 } ], "false_reject": 0, "false_reject_examples": [], "full_trial_h20_hours": 1.0669595034675, "invalidly_saved_h20_hours": 0.010117910306111111, "remaining_h20_hours_at_cutoff": 0.957237281245278, "saved_h20_hours_if_decisions_used": 0.6463471093063888, "threshold": 0.9, "valid_cost_reduction_fraction": null, "valid_zero_error_policy": false }, { "abstain_continue_full": 21, "correctly_saved_h20_hours": 0.3878475556194444, "decision_coverage": 0.43243243243243246, "early_accept": 13, "early_reject": 3, "false_accept": 0, "false_accept_examples": [], "false_reject": 0, "false_reject_examples": [], "full_trial_h20_hours": 1.0669595034675, "invalidly_saved_h20_hours": 0.0, "remaining_h20_hours_at_cutoff": 0.957237281245278, "saved_h20_hours_if_decisions_used": 0.3878475556194444, "threshold": 0.95, "valid_cost_reduction_fraction": 0.3635072881013599, "valid_zero_error_policy": true } ] }, "paired_correctness": { "both_correct": 29, "both_wrong": 4, "instrumentation_only_correct": 4, "mcnemar_exact_two_sided_p": 0.125, "outcome_only_correct": 0 } }, "10.0": { "bootstrap": { "accuracy_delta_instrumentation_minus_outcome": { "ci95": [ -0.08698849104859337, 0.07499999999999996 ], "point": 0.0 }, "brier_delta_instrumentation_minus_outcome": { "ci95": [ -0.0419671863122155, 0.010410939632453053 ], "point": -0.015389088872407489 }, "replicates": 10000, "seed": 20260714, "semantics": "group bootstrap over cells; diagnostic confidence interval" }, "instrumentation_aware": { "classification": { "accuracy": 0.8378378378378378, "balanced_accuracy": 0.7420634920634921, "brier": 0.09695183324390907, "confusion": { "false_negative": 2, "false_positive": 4, "true_negative": 5, "true_positive": 26 }, "log_loss": 0.31480154295615903 }, "policies": [ { "abstain_continue_full": 9, "correctly_saved_h20_hours": 0.7917544064236111, "decision_coverage": 0.7567567567567568, "early_accept": 24, "early_reject": 4, "false_accept": 1, "false_accept_examples": [ { "anchor": 0.49609375, "cell": "tp2_mns8", "label_feasible": false, "probability_feasible": 0.8905819069991598, "remaining_h20_hours": 0.010117910306111111 } ], "false_reject": 0, "false_reject_examples": [], "full_trial_h20_hours": 1.0669595034675, "invalidly_saved_h20_hours": 0.010117910306111111, "remaining_h20_hours_at_cutoff": 0.957237281245278, "saved_h20_hours_if_decisions_used": 0.8018723167297224, "threshold": 0.8, "valid_cost_reduction_fraction": null, "valid_zero_error_policy": false }, { "abstain_continue_full": 21, "correctly_saved_h20_hours": 0.5056768371675, "decision_coverage": 0.43243243243243246, "early_accept": 13, "early_reject": 3, "false_accept": 0, "false_accept_examples": [], "false_reject": 0, "false_reject_examples": [], "full_trial_h20_hours": 1.0669595034675, "invalidly_saved_h20_hours": 0.0, "remaining_h20_hours_at_cutoff": 0.957237281245278, "saved_h20_hours_if_decisions_used": 0.5056768371675, "threshold": 0.9, "valid_cost_reduction_fraction": 0.4739419214357306, "valid_zero_error_policy": true }, { "abstain_continue_full": 30, "correctly_saved_h20_hours": 0.19128570447222223, "decision_coverage": 0.1891891891891892, "early_accept": 6, "early_reject": 1, "false_accept": 0, "false_accept_examples": [], "false_reject": 0, "false_reject_examples": [], "full_trial_h20_hours": 1.0669595034675, "invalidly_saved_h20_hours": 0.0, "remaining_h20_hours_at_cutoff": 0.957237281245278, "saved_h20_hours_if_decisions_used": 0.19128570447222223, "threshold": 0.95, "valid_cost_reduction_fraction": 0.17928112908743485, "valid_zero_error_policy": true } ] }, "outcome_only": { "classification": { "accuracy": 0.8378378378378378, "balanced_accuracy": 0.7043650793650793, "brier": 0.11234092211631656, "confusion": { "false_negative": 1, "false_positive": 5, "true_negative": 4, "true_positive": 27 }, "log_loss": 0.36773394130409404 }, "policies": [ { "abstain_continue_full": 15, "correctly_saved_h20_hours": 0.6981008277897223, "decision_coverage": 0.5945945945945946, "early_accept": 21, "early_reject": 1, "false_accept": 1, "false_accept_examples": [ { "anchor": 0.49609375, "cell": "tp2_mns8", "label_feasible": false, "probability_feasible": 0.8520031544496152, "remaining_h20_hours": 0.010117910306111111 } ], "false_reject": 0, "false_reject_examples": [], "full_trial_h20_hours": 1.0669595034675, "invalidly_saved_h20_hours": 0.010117910306111111, "remaining_h20_hours_at_cutoff": 0.957237281245278, "saved_h20_hours_if_decisions_used": 0.7082187380958334, "threshold": 0.8, "valid_cost_reduction_fraction": null, "valid_zero_error_policy": false }, { "abstain_continue_full": 31, "correctly_saved_h20_hours": 0.23290921672555553, "decision_coverage": 0.16216216216216217, "early_accept": 6, "early_reject": 0, "false_accept": 0, "false_accept_examples": [], "false_reject": 0, "false_reject_examples": [], "full_trial_h20_hours": 1.0669595034675, "invalidly_saved_h20_hours": 0.0, "remaining_h20_hours_at_cutoff": 0.957237281245278, "saved_h20_hours_if_decisions_used": 0.23290921672555553, "threshold": 0.9, "valid_cost_reduction_fraction": 0.21829246186816503, "valid_zero_error_policy": true }, { "abstain_continue_full": 34, "correctly_saved_h20_hours": 0.13888648644666668, "decision_coverage": 0.08108108108108109, "early_accept": 3, "early_reject": 0, "false_accept": 0, "false_accept_examples": [], "false_reject": 0, "false_reject_examples": [], "full_trial_h20_hours": 1.0669595034675, "invalidly_saved_h20_hours": 0.0, "remaining_h20_hours_at_cutoff": 0.957237281245278, "saved_h20_hours_if_decisions_used": 0.13888648644666668, "threshold": 0.95, "valid_cost_reduction_fraction": 0.13017034479312573, "valid_zero_error_policy": true } ] }, "paired_correctness": { "both_correct": 30, "both_wrong": 5, "instrumentation_only_correct": 1, "mcnemar_exact_two_sided_p": 1.0, "outcome_only_correct": 1 } } }, "remaining_h20_hours": { "distinct_n": 37, "max": 0.06275538177, "min": 0.0026573007705555556, "n": 37 } } }, "cutoffs_s": [ 5.0, 10.0, 15.0, 20.0 ], "decision": { "contribution_established": false, "reason": "This dataset contains one workload and reconstructed rather than exact request completion times. Three TP4 primary trials also disagree with their 2-of-3 labels. It can reject a missing-signal premise but cannot establish generalization or a paper-facing cost reduction." }, "estimand": "2-of-3 adjudicated anchor feasibility from the first primary trial's identical short real prefix", "headline_incremental_deltas": { "10": { "accuracy": -0.027027027027027084, "brier": -0.0026830647286030934 }, "15": { "accuracy": 0.02777777777777779, "brier": -0.03744665409603411 }, "20": { "accuracy": 0.05555555555555547, "brier": -0.04807041919163683 }, "5": { "accuracy": 0.10810810810810811, "brier": -0.03964626050589401 } }, "instrumentation_features": [ "model_steps_per_second", "waiting_mean", "waiting_max", "waiting_nonzero_share", "running_mean", "running_max", "decode_batch_mean", "decode_batch_max", "decode_batch_cv", "kv_usage_mean", "kv_usage_max", "kv_usage_end_minus_start", "graph_none_share", "graph_full_share", "padding_fraction", "prefill_token_fraction", "preemptions" ], "model": "same L2 logistic model and folds; instrumentation model appends Layer-1 features", "outcome_features": [ "log_offered_rate_per_gpu", "log2_tp", "log2_max_num_seqs", "admitted_fraction", "completed_over_admitted", "completed_pass_rate", "completed_fail_fraction_of_total", "outstanding_over_admitted", "ttft_max_over_slo_max", "ttft_mean_over_slo_max", "tpot_max_over_slo", "tpot_mean_over_slo", "admitted_input_tokens_mean_over_limit" ], "provenance": { "phase6_metrics": "/home/gahow/phd/aituner/runs/opprof-phase6/phase6/metrics.json", "phase6_metrics_sha256": "290ba7fcb8727291166de7e4d47afdc84e230052495c81dd087db0ace9f93a16", "raw_root": "/home/gahow/phd/aituner/runs/opprof-phase6/phase6/solo-authoritative/cells" }, "sanity": { "cutoff_count": 4, "invariants": { "cutoffs_positive": true, "full_trial_label_not_used_as_feature": true, "paired_same_model_family": true, "probabilities_checked_in_unit_interval": true, "records_strictly_prefix_sliced": true }, "red_flags": [] }, "schema": "fidelity-prefix-v1", "scope": "retrospective single-workload prefix diagnostic; model selection, threshold choice, and contribution claims require held-out prospective tasks", "split": "leave-one-configuration-cell-out", "status": "PASS" }