{ "config": { "task": "evidence", "models": { "Qwen/Qwen3.5-4B": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a", "Qwen/Qwen3.6-35B-A3B": "995ad96eacd98c81ed38be0c5b274b04031597b0" }, "rates": { "Qwen/Qwen3.5-4B": { "train": 0.737, "prefill": 0.33, "sample": 1.005 }, "Qwen/Qwen3.6-35B-A3B": { "train": 1.177, "prefill": 0.45, "sample": 1.335 } }, "rows_sha256": "b3c542237dba50153d88d9358db3c05351bdfbd99fa6b7b87362b9751de5bde3", "epochs": 2, "seed": 2026110607, "rank": 16, "learning_rate": 8e-05, "batch_size": 16, "max_output": 160, "train_tokens_per_epoch": 1233757, "reserved_usd": 5.5, "scope": "Source-indexed passage selection on existing paper-disjoint annotations. Exact support-set agreement is a proxy, not human citation-entailment validation. Existing teacher labels can omit other valid passages. No full-paper or general agent claim.", "fixed_schedule": true }, "usage": { "train_tokens": 2467514, "sampling": { "Qwen/Qwen3.5-4B": { "input_tokens": 1197132, "output_tokens": 16190 }, "Qwen/Qwen3.6-35B-A3B": { "input_tokens": 598566, "output_tokens": 10810 } } }, "release_gate_passed": false, "requires_independent_hf_reload": true, "conditions": { "base": { "validation": { "n": 101, "accuracy": 0.45544554455445546, "balanced_accuracy": 0.5981566820276498, "by_category": { "abstain": 0.967741935483871, "answer": 0.22857142857142856 }, "leaks": 0, "mean_tokens": 41.12871287128713 }, "test": { "n": 156, "accuracy": 0.48717948717948717, "balanced_accuracy": 0.6095746709898913, "by_category": { "abstain": 0.9387755102040817, "answer": 0.2803738317757009 }, "leaks": 0, "mean_tokens": 39.18589743589744 } }, "challenger": { "validation": { "n": 101, "accuracy": 0.5742574257425742, "balanced_accuracy": 0.6928571428571428, "by_category": { "abstain": 1.0, "answer": 0.38571428571428573 }, "leaks": 0, "mean_tokens": 43.75247524752475 }, "test": { "n": 156, "accuracy": 0.6602564102564102, "balanced_accuracy": 0.7523364485981308, "by_category": { "abstain": 1.0, "answer": 0.5046728971962616 }, "leaks": 0, "mean_tokens": 40.967948717948715 } }, "candidate": { "validation": { "n": 101, "accuracy": 0.900990099009901, "balanced_accuracy": 0.9285714285714286, "by_category": { "abstain": 1.0, "answer": 0.8571428571428571 }, "leaks": 0, "mean_tokens": 23.633663366336634 }, "test": { "n": 156, "accuracy": 0.9551282051282052, "balanced_accuracy": 0.9672897196261683, "by_category": { "abstain": 1.0, "answer": 0.9345794392523364 }, "leaks": 0, "mean_tokens": 22.666666666666668 } }, "rules": { "validation": { "n": 101, "accuracy": 0.26732673267326734, "balanced_accuracy": 0.19285714285714287, "by_category": { "abstain": 0.0, "answer": 0.38571428571428573 }, "leaks": 0, "mean_tokens": 0.0 }, "test": { "n": 156, "accuracy": 0.2692307692307692, "balanced_accuracy": 0.19626168224299065, "by_category": { "abstain": 0.0, "answer": 0.3925233644859813 }, "leaks": 0, "mean_tokens": 0.0 } } }, "comparisons": { "base": { "groups": 103, "delta": 0.5275080906148868, "interval95": [ 0.4482200647249191, 0.6100323624595468 ] }, "challenger": { "groups": 103, "delta": 0.34012944983818777, "interval95": [ 0.2576051779935275, 0.42556634304207114 ] }, "rules": { "groups": 103, "delta": 0.6181229773462783, "interval95": [ 0.5307443365695793, 0.7022653721682848 ] } }, "per_request_token_cost_usd": { "base": 0.0007986441346153846, "challenger": 0.0010900499038461538, "candidate": 0.0007820423076923078, "rules": 0.0 }, "native_quality_gate_passed": true, "superior": true, "noninferior": true, "cheaper": false, "estimated_tokens_usd": 2.513668378 }