{ "hacked_run_dir": "/root/workspace/Shaer/grpo/outputs/train/shaer_grpo_20260413_071842", "hacked_step": 1800, "train_manifest_path": "/root/workspace/Shaer/grpo/outputs/curated_meter_count_easyfirst_short_drop_trio/cap_1000/selected_manifest.csv", "prompt_file": "/root/workspace/Shaer/grpo/prompts/judge_quality_v5.yaml", "golden_summary": { "count": 30, "reward_total": { "n": 30, "mean": 0.6777313252915126, "median": 0.6208921169221314, "min": 0.4973053566057063, "p10": 0.5762708849947799, "p25": 0.6152646022737026, "p75": 0.7479642466276728, "p90": 0.9166323143562357, "max": 0.9459829156707615 }, "meter": { "n": 30, "mean": 0.9576655101608114, "median": 0.9898881312815053, "min": 0.5395800608267627, "p10": 0.848249732191195, "p25": 0.9805628310292759, "p75": 0.9948783055038647, "p90": 0.9973563416849837, "max": 0.9981962209330739 }, "count_adherence": { "n": 30, "mean": 1.0, "median": 1.0, "min": 1.0, "p10": 1.0, "p25": 1.0, "p75": 1.0, "p90": 1.0, "max": 1.0 }, "judge_quality": { "n": 30, "mean": 0.5276666666666667, "median": 0.45, "min": 0.25, "p10": 0.42, "p25": 0.42, "p75": 0.6525000000000001, "p90": 0.884, "max": 0.92 }, "repeat_soft": { "n": 30, "mean": 0.9838626053754798, "median": 0.9943360071301247, "min": 0.8383064516129032, "p10": 0.9625, "p25": 0.9832476452251293, "p75": 1.0, "p90": 1.0, "max": 1.0 }, "hard_gate_pass_rate": 1.0, "exact_count_rate": 1.0, "latin_leakage_rate": 0.0, "digit_leakage_rate": 0.0, "artifact_leakage_rate": 0.0 }, "hacked_summary": { "count": 30, "reward_total": { "n": 30, "mean": 0.5478744505176345, "median": 0.5930548397227667, "min": 0.38773352667297567, "p10": 0.44052989232171735, "p25": 0.466384036403926, "p75": 0.6101410305654957, "p90": 0.6210252001960789, "max": 0.6360540489105433 }, "meter": { "n": 30, "mean": 0.9527989474726047, "median": 0.9781341266840117, "min": 0.8002345370782145, "p10": 0.8762595567428119, "p25": 0.9301577503277665, "p75": 0.9908177612440305, "p90": 0.992913118491257, "max": 0.9974696070423849 }, "count_adherence": { "n": 30, "mean": 0.8711111111111111, "median": 1.0, "min": 0.5, "p10": 0.5, "p25": 0.75, "p75": 1.0, "p90": 1.0, "max": 1.0 }, "judge_quality": { "n": 30, "mean": 0.38899999999999996, "median": 0.42, "min": 0.15, "p10": 0.18, "p25": 0.42, "p75": 0.42, "p90": 0.42, "max": 0.45 }, "repeat_soft": { "n": 30, "mean": 0.8893664909084346, "median": 0.9064090507253031, "min": 0.6886338864019518, "p10": 0.7672463653804329, "p25": 0.8352720649042025, "p75": 0.9500494071146245, "p90": 1.0, "max": 1.0 }, "hard_gate_pass_rate": 1.0, "exact_count_rate": 0.6666666666666666, "latin_leakage_rate": 0.0, "digit_leakage_rate": 0.0, "artifact_leakage_rate": 0.0 }, "health": { "healthy": false, "checks": { "golden_total_gt_hacked_total_by_0.15": false, "golden_judge_gt_hacked_judge_by_0.15": false, "golden_judge_not_too_harsh": true, "golden_meter_not_broken": true, "hacked_total_sufficiently_low": false }, "failed_checks": [ "golden_total_gt_hacked_total_by_0.15", "golden_judge_gt_hacked_judge_by_0.15", "hacked_total_sufficiently_low" ] } }