{ "n_test": 13632, "n_calib": 5572, "traits": { "A": { "n": 1134, "category_acc": 0.8854, "category_macro_f1": 0.8722, "needs_tools_acc": 0.9836, "needs_tools_auc": 0.991339380909933, "needs_vision_acc": 0.9982, "needs_vision_auc": 0.9998792270531401, "needs_long_context_acc": 0.9947, "needs_long_context_auc": 0.9995456251605128, "follow_up_acc": 0.9991, "follow_up_auc": 0.9997983309485168, "difficulty_spearman": 0.7386881928498473, "category_acc_by_source": { "arena55k": 0.8889, "synthetic-s1": 0.9125, "routellm-gpt4": 0.8503 } }, "B": { "n": 1134, "category_acc": 0.8854, "category_macro_f1": 0.8734, "needs_tools_acc": 0.9918, "needs_tools_auc": 0.9915403841784209, "needs_vision_acc": 0.9974, "needs_vision_auc": 0.9992149758454106, "needs_long_context_acc": 0.9938, "needs_long_context_auc": 0.9994468480214939, "follow_up_acc": 0.9991, "follow_up_auc": 0.9995006290153748, "difficulty_spearman": 0.7361477126530993, "category_acc_by_source": { "arena55k": 0.8679, "synthetic-s1": 0.9125, "routellm-gpt4": 0.8822 } }, "laya": { "n": 1134, "category_acc": 0.0935, "category_macro_f1": 0.112, "needs_tools_acc": 0.8046, "needs_tools_auc": 0.6708527782147414, "needs_vision_acc": 0.9594, "needs_vision_auc": 0.972403381642512, "needs_long_context_acc": 0.9466, "needs_long_context_auc": 0.8954345206345443, "follow_up_acc": 0.8834, "follow_up_auc": 0.8828494876645764, "difficulty_spearman": 0.2967021811737586, "category_acc_by_source": { "arena55k": 0.065, "synthetic-s1": 0.1924, "routellm-gpt4": 0.0287 } }, "jev": { "n": 1134, "category_acc": 0.858, "category_macro_f1": 0.8427, "needs_tools_acc": 0.9598, "needs_tools_auc": 0.9917763445370807, "needs_vision_acc": 0.9974, "needs_vision_auc": 1.0, "needs_long_context_acc": 0.9867, "needs_long_context_auc": 0.9971552183962543, "follow_up_acc": 0.9947, "follow_up_auc": 0.9985114903342904, "difficulty_spearman": 0.7532382322770758, "category_acc_by_source": { "arena55k": 0.8889, "synthetic-s1": 0.8717, "routellm-gpt4": 0.7962 } } }, "router70": { "A": { "n_cat": 59, "category_acc": 0.9492, "n": 70, "difficulty_llm_spearman": 0.190796955646925, "difficulty_outcome_spearman": 0.29776922403989153 }, "B": { "n_cat": 59, "category_acc": 0.9831, "n": 70, "difficulty_llm_spearman": 0.06482372495844632, "difficulty_outcome_spearman": 0.17910943924416062 }, "laya": { "n_cat": 59, "category_acc": 0.339, "n": 70, "difficulty_llm_spearman": -0.06321406701076021, "difficulty_outcome_spearman": -0.06321406701076021 }, "jev": { "n_cat": 59, "category_acc": 0.7966, "n": 70, "difficulty_llm_spearman": 0.29941387455165774, "difficulty_outcome_spearman": 0.29941387455165774 } }, "succ": { "routerbench": { "prior (heuristic)": { "small": { "auc": 0.0, "brier": 0.1666 }, "mid": { "auc": 0.0, "brier": 0.0567 }, "strong": { "auc": 0.0, "brier": 0.1017 } }, "A succ-heads": { "small": { "auc": 0.5871859132317802, "brier": 0.158 }, "mid": { "auc": 0.7284847213431849, "brier": 0.0455 }, "strong": { "auc": 0.6929741161200729, "brier": 0.085 } }, "A traits+calib": { "small": { "auc": 0.5390554995519713, "brier": 0.1623 }, "mid": { "auc": 0.6121454121361675, "brier": 0.0524 }, "strong": { "auc": 0.5743148952005791, "brier": 0.0964 } }, "B succ-heads": { "small": { "auc": 0.6710606145459976, "brier": 0.1491 }, "mid": { "auc": 0.7469368017579167, "brier": 0.0432 }, "strong": { "auc": 0.7181409533491113, "brier": 0.0816 } }, "B traits+calib": { "small": { "auc": 0.5797094347371565, "brier": 0.1586 }, "mid": { "auc": 0.6881226843785779, "brier": 0.0474 }, "strong": { "auc": 0.6694183558392462, "brier": 0.0898 } }, "laya traits+calib": { "small": { "auc": 0.512134109916368, "brier": 0.1646 }, "mid": { "auc": 0.5518876981389693, "brier": 0.0563 }, "strong": { "auc": 0.5477457402898577, "brier": 0.1021 } }, "jev traits+calib": { "small": { "auc": 0.5369670325567503, "brier": 0.1632 }, "mid": { "auc": 0.6381834861080493, "brier": 0.0528 }, "strong": { "auc": 0.5912175884220295, "brier": 0.0943 } } }, "measured-s2": { "prior (heuristic)": { "small": { "auc": 0.0, "brier": 0.2594 }, "mid": { "auc": 0.0, "brier": 0.2539 }, "strong": { "auc": 0.0, "brier": 0.1296 } }, "A succ-heads": { "small": { "auc": 0.9344919786096256, "brier": 0.0994 }, "mid": { "auc": 0.9594594594594594, "brier": 0.0348 }, "strong": { "auc": 0.9094567404426559, "brier": 0.0647 } }, "A traits+calib": { "small": { "auc": 0.9144385026737968, "brier": 0.1897 }, "mid": { "auc": 0.8412162162162162, "brier": 0.1478 }, "strong": { "auc": 0.7927565392354124, "brier": 0.0998 } }, "B succ-heads": { "small": { "auc": 0.9358288770053476, "brier": 0.0957 }, "mid": { "auc": 0.9425675675675675, "brier": 0.0308 }, "strong": { "auc": 0.8933601609657947, "brier": 0.0637 } }, "B traits+calib": { "small": { "auc": 0.9512032085561497, "brier": 0.1558 }, "mid": { "auc": 0.8141891891891891, "brier": 0.1351 }, "strong": { "auc": 0.8189134808853119, "brier": 0.0919 } }, "laya traits+calib": { "small": { "auc": 0.554144385026738, "brier": 0.2552 }, "mid": { "auc": 0.7567567567567568, "brier": 0.2061 }, "strong": { "auc": 0.2676056338028169, "brier": 0.1168 } }, "jev traits+calib": { "small": { "auc": 0.6724598930481284, "brier": 0.2268 }, "mid": { "auc": 0.7837837837837838, "brier": 0.1897 }, "strong": { "auc": 0.579476861167002, "brier": 0.1396 } } } }, "replay": { "routerbench": { "prior (heuristic)": { "n": 1500, "always_small": [ 4.6e-05, 0.2845 ], "always_mid": [ 0.0001341, 0.4722 ], "always_strong": [ 0.0031709, 0.6862 ], "oracle_quality": 0.7723, "best_quality": 0.6862, "quality_at_10pct_cost": 0.4722, "quality_at_25pct_cost": 0.4722, "quality_at_50pct_cost": 0.4722, "cost_share_for_80pct_gap": 1.0, "cost_share_for_95pct_gap": 1.0, "cost_share_for_100pct_gap": 1.0, "apgr": 0.4939 }, "oracle": { "n": 1500, "always_small": [ 4.6e-05, 0.2845 ], "always_mid": [ 0.0001341, 0.4722 ], "always_strong": [ 0.0031709, 0.6862 ], "oracle_quality": 0.7723, "best_quality": 0.7723, "quality_at_10pct_cost": 0.6457, "quality_at_25pct_cost": 0.7073, "quality_at_50pct_cost": 0.7666, "cost_share_for_80pct_gap": 0.0578, "cost_share_for_95pct_gap": 0.1468, "cost_share_for_100pct_gap": 0.1713, "apgr": 1.1403 }, "A succ-heads": { "n": 1500, "always_small": [ 4.6e-05, 0.2845 ], "always_mid": [ 0.0001341, 0.4722 ], "always_strong": [ 0.0031709, 0.6862 ], "oracle_quality": 0.7723, "best_quality": 0.6867, "quality_at_10pct_cost": 0.5137, "quality_at_25pct_cost": 0.5904, "quality_at_50pct_cost": 0.6601, "cost_share_for_80pct_gap": 0.3136, "cost_share_for_95pct_gap": 0.6521, "cost_share_for_100pct_gap": 0.9987, "apgr": 0.8618 }, "A traits+calib": { "n": 1500, "always_small": [ 4.6e-05, 0.2845 ], "always_mid": [ 0.0001341, 0.4722 ], "always_strong": [ 0.0031709, 0.6862 ], "oracle_quality": 0.7723, "best_quality": 0.6862, "quality_at_10pct_cost": 0.4873, "quality_at_25pct_cost": 0.582, "quality_at_50pct_cost": 0.6586, "cost_share_for_80pct_gap": 0.4096, "cost_share_for_95pct_gap": 0.8807, "cost_share_for_100pct_gap": 1.0, "apgr": 0.8335 }, "B succ-heads": { "n": 1500, "always_small": [ 4.6e-05, 0.2845 ], "always_mid": [ 0.0001341, 0.4722 ], "always_strong": [ 0.0031709, 0.6862 ], "oracle_quality": 0.7723, "best_quality": 0.6862, "quality_at_10pct_cost": 0.5179, "quality_at_25pct_cost": 0.5895, "quality_at_50pct_cost": 0.6512, "cost_share_for_80pct_gap": 0.3039, "cost_share_for_95pct_gap": 0.6865, "cost_share_for_100pct_gap": 1.0, "apgr": 0.8638 }, "B traits+calib": { "n": 1500, "always_small": [ 4.6e-05, 0.2845 ], "always_mid": [ 0.0001341, 0.4722 ], "always_strong": [ 0.0031709, 0.6862 ], "oracle_quality": 0.7723, "best_quality": 0.6862, "quality_at_10pct_cost": 0.4978, "quality_at_25pct_cost": 0.5818, "quality_at_50pct_cost": 0.6593, "cost_share_for_80pct_gap": 0.3795, "cost_share_for_95pct_gap": 0.8564, "cost_share_for_100pct_gap": 1.0, "apgr": 0.8411 }, "laya traits+calib": { "n": 1500, "always_small": [ 4.6e-05, 0.2845 ], "always_mid": [ 0.0001341, 0.4722 ], "always_strong": [ 0.0031709, 0.6862 ], "oracle_quality": 0.7723, "best_quality": 0.6862, "quality_at_10pct_cost": 0.4892, "quality_at_25pct_cost": 0.5384, "quality_at_50pct_cost": 0.6349, "cost_share_for_80pct_gap": 0.3893, "cost_share_for_95pct_gap": 0.7946, "cost_share_for_100pct_gap": 1.0, "apgr": 0.8125 }, "jev traits+calib": { "n": 1500, "always_small": [ 4.6e-05, 0.2845 ], "always_mid": [ 0.0001341, 0.4722 ], "always_strong": [ 0.0031709, 0.6862 ], "oracle_quality": 0.7723, "best_quality": 0.6862, "quality_at_10pct_cost": 0.503, "quality_at_25pct_cost": 0.596, "quality_at_50pct_cost": 0.6555, "cost_share_for_80pct_gap": 0.2817, "cost_share_for_95pct_gap": 0.8623, "cost_share_for_100pct_gap": 1.0, "apgr": 0.8499 } }, "measured-s2": { "prior (heuristic)": { "n": 78, "always_small": [ 1.14e-05, 0.4359 ], "always_mid": [ 0.0001846, 0.9487 ], "always_strong": [ 0.0003176, 0.9103 ], "oracle_quality": 0.9872, "best_quality": 0.9487, "quality_at_10pct_cost": 0.4359, "quality_at_25pct_cost": 0.4359, "quality_at_50pct_cost": 0.4359, "cost_share_for_80pct_gap": 0.5814, "cost_share_for_95pct_gap": 0.5814, "cost_share_for_100pct_gap": 0.5814, "apgr": 0.4865 }, "oracle": { "n": 78, "always_small": [ 1.14e-05, 0.4359 ], "always_mid": [ 0.0001846, 0.9487 ], "always_strong": [ 0.0003176, 0.9103 ], "oracle_quality": 0.9872, "best_quality": 0.9872, "quality_at_10pct_cost": 0.4359, "quality_at_25pct_cost": 0.4359, "quality_at_50pct_cost": 0.9872, "cost_share_for_80pct_gap": 0.3877, "cost_share_for_95pct_gap": 0.3877, "cost_share_for_100pct_gap": 0.3877, "apgr": 0.7514 }, "A succ-heads": { "n": 78, "always_small": [ 1.14e-05, 0.4359 ], "always_mid": [ 0.0001846, 0.9487 ], "always_strong": [ 0.0003176, 0.9103 ], "oracle_quality": 0.9872, "best_quality": 0.9872, "quality_at_10pct_cost": 0.5385, "quality_at_25pct_cost": 0.7179, "quality_at_50pct_cost": 0.9231, "cost_share_for_80pct_gap": 0.3784, "cost_share_for_95pct_gap": 0.4781, "cost_share_for_100pct_gap": 0.4913, "apgr": 0.8986 }, "A traits+calib": { "n": 78, "always_small": [ 1.14e-05, 0.4359 ], "always_mid": [ 0.0001846, 0.9487 ], "always_strong": [ 0.0003176, 0.9103 ], "oracle_quality": 0.9872, "best_quality": 0.9487, "quality_at_10pct_cost": 0.5128, "quality_at_25pct_cost": 0.5128, "quality_at_50pct_cost": 0.6667, "cost_share_for_80pct_gap": 0.5814, "cost_share_for_95pct_gap": 0.5814, "cost_share_for_100pct_gap": 0.5814, "apgr": 0.6608 }, "B succ-heads": { "n": 78, "always_small": [ 1.14e-05, 0.4359 ], "always_mid": [ 0.0001846, 0.9487 ], "always_strong": [ 0.0003176, 0.9103 ], "oracle_quality": 0.9872, "best_quality": 0.9872, "quality_at_10pct_cost": 0.5513, "quality_at_25pct_cost": 0.6795, "quality_at_50pct_cost": 0.9103, "cost_share_for_80pct_gap": 0.3551, "cost_share_for_95pct_gap": 0.4647, "cost_share_for_100pct_gap": 0.495, "apgr": 0.8959 }, "B traits+calib": { "n": 78, "always_small": [ 1.14e-05, 0.4359 ], "always_mid": [ 0.0001846, 0.9487 ], "always_strong": [ 0.0003176, 0.9103 ], "oracle_quality": 0.9872, "best_quality": 0.9487, "quality_at_10pct_cost": 0.5256, "quality_at_25pct_cost": 0.5769, "quality_at_50pct_cost": 0.6795, "cost_share_for_80pct_gap": 0.5814, "cost_share_for_95pct_gap": 0.5814, "cost_share_for_100pct_gap": 0.5814, "apgr": 0.6797 }, "laya traits+calib": { "n": 78, "always_small": [ 1.14e-05, 0.4359 ], "always_mid": [ 0.0001846, 0.9487 ], "always_strong": [ 0.0003176, 0.9103 ], "oracle_quality": 0.9872, "best_quality": 0.9487, "quality_at_10pct_cost": 0.4615, "quality_at_25pct_cost": 0.5385, "quality_at_50pct_cost": 0.6282, "cost_share_for_80pct_gap": 0.5814, "cost_share_for_95pct_gap": 0.5814, "cost_share_for_100pct_gap": 0.5814, "apgr": 0.6446 }, "jev traits+calib": { "n": 78, "always_small": [ 1.14e-05, 0.4359 ], "always_mid": [ 0.0001846, 0.9487 ], "always_strong": [ 0.0003176, 0.9103 ], "oracle_quality": 0.9872, "best_quality": 0.9487, "quality_at_10pct_cost": 0.4487, "quality_at_25pct_cost": 0.5256, "quality_at_50pct_cost": 0.8077, "cost_share_for_80pct_gap": 0.5113, "cost_share_for_95pct_gap": 0.5814, "cost_share_for_100pct_gap": 0.5814, "apgr": 0.7311 } } }, "route": { "A": { "route_head": { "n": 6586, "oracle_agreement": 0.6731, "mean_regret_usd": 0.0502961 }, "econ[prior (heuristic)]": { "n": 6586, "oracle_agreement": 0.741, "mean_regret_usd": 0.0350319 }, "econ[oracle]": { "n": 6586, "oracle_agreement": 0.9368, "mean_regret_usd": 0.0216345 }, "econ[A succ-heads]": { "n": 6586, "oracle_agreement": 0.7475, "mean_regret_usd": 0.0348688 }, "econ[A traits+calib]": { "n": 6586, "oracle_agreement": 0.7426, "mean_regret_usd": 0.0347834 }, "econ[laya traits+calib]": { "n": 3156, "oracle_agreement": 0.7335, "mean_regret_usd": 0.0371752 }, "econ[jev traits+calib]": { "n": 3156, "oracle_agreement": 0.7411, "mean_regret_usd": 0.0369512 }, "always_strong": { "n": 6586, "oracle_agreement": 0.136, "mean_regret_usd": 0.1993001 }, "always_small": { "n": 6586, "oracle_agreement": 0.5742, "mean_regret_usd": 0.069763 } }, "B": { "route_head": { "n": 6586, "oracle_agreement": 0.6828, "mean_regret_usd": 0.0488386 }, "econ[prior (heuristic)]": { "n": 6586, "oracle_agreement": 0.741, "mean_regret_usd": 0.0350319 }, "econ[oracle]": { "n": 6586, "oracle_agreement": 0.9368, "mean_regret_usd": 0.0216345 }, "econ[B succ-heads]": { "n": 6586, "oracle_agreement": 0.7589, "mean_regret_usd": 0.0346441 }, "econ[B traits+calib]": { "n": 6586, "oracle_agreement": 0.7431, "mean_regret_usd": 0.0347052 }, "econ[laya traits+calib]": { "n": 3156, "oracle_agreement": 0.7335, "mean_regret_usd": 0.0371752 }, "econ[jev traits+calib]": { "n": 3156, "oracle_agreement": 0.7411, "mean_regret_usd": 0.0369512 }, "always_strong": { "n": 6586, "oracle_agreement": 0.136, "mean_regret_usd": 0.1993001 }, "always_small": { "n": 6586, "oracle_agreement": 0.5742, "mean_regret_usd": 0.069763 } } }, "latency_ms_batched_gpu": { "A": 6.219, "B": 10.11, "laya": 24.631, "jev": 274.5 } }