{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-pure/iter-6/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "pure", "method": "pure OPD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-pure", "checkpoint_iteration": 6, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-pure/iter_0000006", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_pure_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_pure_measured_pass8_cap120.tsv": "82e41cff6187d02de9b8fe554f61dc30b1aea40aae793e1ec89e3b2ac24ab33c"}, "score_mean": 0.17387218045112782, "score_ci_low": 0.15228298782136945, "score_ci_high": 0.19780788707992675, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 51964.96428571428, "output_tokens_ci_low": 48173.170112781954, "output_tokens_ci_high": 55624.83364661654, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 77.96334586466165, "steps_ci_low": 72.92293233082707, "steps_ci_high": 82.93327067669173, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-pure/iter-13/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "pure", "method": "pure OPD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-pure", "checkpoint_iteration": 13, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-pure/iter_0000013", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_pure_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_pure_measured_pass8_cap120.tsv": "82e41cff6187d02de9b8fe554f61dc30b1aea40aae793e1ec89e3b2ac24ab33c"}, "score_mean": 0.22180451127819548, "score_ci_low": 0.19786614926032586, "score_ci_high": 0.2477445105978403, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 51597.14097744361, "output_tokens_ci_low": 46723.07330827068, "output_tokens_ci_high": 56586.34492481203, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 60.204887218045116, "steps_ci_low": 54.7265037593985, "steps_ci_high": 65.86748120300751, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-pure/iter-20/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "pure", "method": "pure OPD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-pure", "checkpoint_iteration": 20, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-pure/iter_0000020", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_pure_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_pure_measured_pass8_cap120.tsv": "82e41cff6187d02de9b8fe554f61dc30b1aea40aae793e1ec89e3b2ac24ab33c"}, "score_mean": 0.2716165413533835, "score_ci_low": 0.24574704810203443, "score_ci_high": 0.29912927063287914, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 38868.53947368421, "output_tokens_ci_low": 35007.6992481203, "output_tokens_ci_high": 42808.90695488722, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 59.36372180451128, "steps_ci_low": 53.76127819548872, "steps_ci_high": 65.07048872180451, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-pure/iter-29/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "pure", "method": "pure OPD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-pure", "checkpoint_iteration": 29, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-pure/iter_0000029", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_pure_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_pure_measured_pass8_cap120.tsv": "82e41cff6187d02de9b8fe554f61dc30b1aea40aae793e1ec89e3b2ac24ab33c"}, "score_mean": 0.2857142857142857, "score_ci_low": 0.2593782579984431, "score_ci_high": 0.3135921151355498, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 34733.767857142855, "output_tokens_ci_low": 30329.160714285714, "output_tokens_ci_high": 39276.996240601504, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 40.6343984962406, "steps_ci_low": 35.83176691729323, "steps_ci_high": 45.76973684210526, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-pure/iter-39/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "pure", "method": "pure OPD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-pure", "checkpoint_iteration": 39, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-pure/iter_0000039", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_pure_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_pure_measured_pass8_cap120.tsv": "82e41cff6187d02de9b8fe554f61dc30b1aea40aae793e1ec89e3b2ac24ab33c"}, "score_mean": 0.30545112781954886, "score_ci_low": 0.27851572031595856, "score_ci_high": 0.3337863289767455, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 34080.88533834586, "output_tokens_ci_low": 30149.65507518797, "output_tokens_ci_high": 38177.77631578947, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 42.579887218045116, "steps_ci_low": 37.840225563909776, "steps_ci_high": 47.5140977443609, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-pure/iter-49/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "pure", "method": "pure OPD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-pure", "checkpoint_iteration": 49, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-pure/iter_0000049", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_pure_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_pure_measured_pass8_cap120.tsv": "82e41cff6187d02de9b8fe554f61dc30b1aea40aae793e1ec89e3b2ac24ab33c"}, "score_mean": 0.3167293233082707, "score_ci_low": 0.2894783535943107, "score_ci_high": 0.34529893921765703, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 40409.046992481206, "output_tokens_ci_low": 35390.161654135336, "output_tokens_ci_high": 45806.13533834586, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 52.17293233082707, "steps_ci_low": 46.44548872180451, "steps_ci_high": 58.0, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-pure/iter-59/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "pure", "method": "pure OPD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-pure", "checkpoint_iteration": 59, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-pure/iter_0000059", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_pure_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_pure_measured_pass8_cap120.tsv": "82e41cff6187d02de9b8fe554f61dc30b1aea40aae793e1ec89e3b2ac24ab33c"}, "score_mean": 0.3524436090225564, "score_ci_low": 0.32431546506168313, "score_ci_high": 0.3816334322279524, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 33917.533834586466, "output_tokens_ci_low": 29813.211466165412, "output_tokens_ci_high": 38367.98684210526, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 44.0093984962406, "steps_ci_low": 39.19172932330827, "steps_ci_high": 49.099624060150376, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-pure/iter-69/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "pure", "method": "pure OPD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-pure", "checkpoint_iteration": 69, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-pure/iter_0000069", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_pure_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_pure_measured_pass8_cap120.tsv": "82e41cff6187d02de9b8fe554f61dc30b1aea40aae793e1ec89e3b2ac24ab33c"}, "score_mean": 0.325187969924812, "score_ci_low": 0.29771273834700984, "score_ci_high": 0.35392098710440545, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 41163.69736842105, "output_tokens_ci_low": 36375.80639097744, "output_tokens_ci_high": 46258.406015037595, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 48.80827067669173, "steps_ci_low": 43.54793233082707, "steps_ci_high": 54.16635338345865, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-pure/iter-79/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "pure", "method": "pure OPD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-pure", "checkpoint_iteration": 79, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-pure/iter_0000079", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_pure_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_pure_measured_pass8_cap120.tsv": "82e41cff6187d02de9b8fe554f61dc30b1aea40aae793e1ec89e3b2ac24ab33c"}, "score_mean": 0.3458646616541353, "score_ci_low": 0.317884583506165, "score_ci_high": 0.37495375506390005, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 38926.14097744361, "output_tokens_ci_low": 34266.563909774435, "output_tokens_ci_high": 43687.250939849626, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 45.39755639097744, "steps_ci_low": 40.56296992481203, "steps_ci_high": 50.5093984962406, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-pure/iter-90/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "pure", "method": "pure OPD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-pure", "checkpoint_iteration": 90, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-pure/iter_0000090", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_pure_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_pure_measured_pass8_cap120.tsv": "82e41cff6187d02de9b8fe554f61dc30b1aea40aae793e1ec89e3b2ac24ab33c"}, "score_mean": 0.3242481203007519, "score_ci_low": 0.2967972896419412, "score_ci_high": 0.3529634988495355, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 31818.9022556391, "output_tokens_ci_low": 27786.07612781955, "output_tokens_ci_high": 36325.73966165414, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 48.47744360902256, "steps_ci_low": 42.859022556390975, "steps_ci_high": 54.307330827067666, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-pure/iter-99/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "pure", "method": "pure OPD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-pure", "checkpoint_iteration": 99, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-pure/iter_0000099", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_pure_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_pure_measured_pass8_cap120.tsv": "82e41cff6187d02de9b8fe554f61dc30b1aea40aae793e1ec89e3b2ac24ab33c"}, "score_mean": 0.33364661654135336, "score_ci_low": 0.3059575159990184, "score_ci_high": 0.3625326420918445, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 32865.96992481203, "output_tokens_ci_low": 29219.049812030076, "output_tokens_ci_high": 36767.8204887218, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 46.74812030075188, "steps_ci_low": 41.31578947368421, "steps_ci_high": 52.51221804511278, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-pure/iter-109/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "pure", "method": "pure OPD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-pure", "checkpoint_iteration": 109, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-pure/iter_0000109", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_pure_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_pure_measured_pass8_cap120.tsv": "82e41cff6187d02de9b8fe554f61dc30b1aea40aae793e1ec89e3b2ac24ab33c"}, "score_mean": 0.34210526315789475, "score_ci_low": 0.3142124855933345, "score_ci_high": 0.3711341051369761, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 36037.357142857145, "output_tokens_ci_low": 31691.90037593985, "output_tokens_ci_high": 40490.92387218045, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 47.839285714285715, "steps_ci_low": 42.277255639097746, "steps_ci_high": 53.67575187969925, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-pure/iter-119/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "pure", "method": "pure OPD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-pure", "checkpoint_iteration": 119, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-pure/iter_0000119", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_pure_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_pure_measured_pass8_cap120.tsv": "82e41cff6187d02de9b8fe554f61dc30b1aea40aae793e1ec89e3b2ac24ab33c"}, "score_mean": 0.3298872180451128, "score_ci_low": 0.3022919005472217, "score_ci_high": 0.3587065097038867, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 37279.12312030075, "output_tokens_ci_low": 32950.98590225564, "output_tokens_ci_high": 41836.825187969924, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 43.16447368421053, "steps_ci_low": 38.32236842105263, "steps_ci_high": 48.181390977443606, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-pure/iter-129/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "pure", "method": "pure OPD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-pure", "checkpoint_iteration": 129, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-pure/iter_0000129", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_pure_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_pure_measured_pass8_cap120.tsv": "82e41cff6187d02de9b8fe554f61dc30b1aea40aae793e1ec89e3b2ac24ab33c"}, "score_mean": 0.34868421052631576, "score_ci_low": 0.32063994478107694, "score_ci_high": 0.377817204668804, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 35805.99154135338, "output_tokens_ci_low": 31390.735902255637, "output_tokens_ci_high": 40405.19078947369, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 44.7687969924812, "steps_ci_low": 39.64755639097744, "steps_ci_high": 50.1062030075188, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-pure/iter-139/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "pure", "method": "pure OPD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-pure", "checkpoint_iteration": 139, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-pure/iter_0000139", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_pure_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_pure_measured_pass8_cap120.tsv": "82e41cff6187d02de9b8fe554f61dc30b1aea40aae793e1ec89e3b2ac24ab33c"}, "score_mean": 0.34962406015037595, "score_ci_low": 0.3215586426001575, "score_ci_high": 0.37877144380966216, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 36047.26973684211, "output_tokens_ci_low": 31826.843045112782, "output_tokens_ci_high": 40385.94736842105, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 45.37593984962406, "steps_ci_low": 40.21898496240601, "steps_ci_high": 50.846804511278194, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2-pure/iter-6/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "aod", "method": "AOD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2-pure", "checkpoint_iteration": 6, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2-pure/iter_0000006", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_aod_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_aod_measured_pass8_cap120.tsv": "6947a6e2d1260bfad983f26aa7d5a345eb340bb992ba13ce2a8d6398903c9b99"}, "score_mean": 0.18703007518796994, "score_ci_low": 0.16474077417549524, "score_ci_high": 0.21157121816494176, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 29643.40037593985, "output_tokens_ci_low": 27431.605263157893, "output_tokens_ci_high": 31946.468984962405, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 59.38157894736842, "steps_ci_low": 55.07612781954887, "steps_ci_high": 63.66447368421053, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2-pure/iter-13/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "aod", "method": "AOD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2-pure", "checkpoint_iteration": 13, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2-pure/iter_0000013", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_aod_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_aod_measured_pass8_cap120.tsv": "6947a6e2d1260bfad983f26aa7d5a345eb340bb992ba13ce2a8d6398903c9b99"}, "score_mean": 0.21804511278195488, "score_ci_low": 0.1942720763553323, "score_ci_high": 0.24384683566307933, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 31001.82894736842, "output_tokens_ci_low": 28490.848684210527, "output_tokens_ci_high": 33516.00657894737, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 59.2734962406015, "steps_ci_low": 54.43045112781955, "steps_ci_high": 64.13063909774436, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2-pure/iter-20/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "aod", "method": "AOD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2-pure", "checkpoint_iteration": 20, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2-pure/iter_0000020", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_aod_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_aod_measured_pass8_cap120.tsv": "6947a6e2d1260bfad983f26aa7d5a345eb340bb992ba13ce2a8d6398903c9b99"}, "score_mean": 0.26597744360902253, "score_ci_low": 0.2403039114734244, "score_ci_high": 0.29333478550185726, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 24483.906015037595, "output_tokens_ci_low": 22362.25845864662, "output_tokens_ci_high": 26598.67387218045, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 41.79323308270677, "steps_ci_low": 38.401315789473685, "steps_ci_high": 45.18796992481203, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2-pure/iter-29/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "aod", "method": "AOD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2-pure", "checkpoint_iteration": 29, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2-pure/iter_0000029", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_aod_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_aod_measured_pass8_cap120.tsv": "6947a6e2d1260bfad983f26aa7d5a345eb340bb992ba13ce2a8d6398903c9b99"}, "score_mean": 0.24342105263157895, "score_ci_low": 0.21858754159440782, "score_ci_high": 0.2701006683423468, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 19135.85620300752, "output_tokens_ci_low": 17395.93045112782, "output_tokens_ci_high": 20912.93327067669, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 31.590225563909776, "steps_ci_low": 28.779135338345863, "steps_ci_high": 34.525375939849624, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2-pure/iter-39/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "aod", "method": "AOD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2-pure", "checkpoint_iteration": 39, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2-pure/iter_0000039", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_aod_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_aod_measured_pass8_cap120.tsv": "6947a6e2d1260bfad983f26aa7d5a345eb340bb992ba13ce2a8d6398903c9b99"}, "score_mean": 0.2781954887218045, "score_ci_low": 0.2521041817387488, "score_ci_high": 0.305882695715735, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 20794.668233082706, "output_tokens_ci_low": 18840.091165413534, "output_tokens_ci_high": 22740.99154135338, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 39.153195488721806, "steps_ci_low": 35.77161654135338, "steps_ci_high": 42.6578947368421, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2-pure/iter-49/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "aod", "method": "AOD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2-pure", "checkpoint_iteration": 49, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2-pure/iter_0000049", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_aod_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_aod_measured_pass8_cap120.tsv": "6947a6e2d1260bfad983f26aa7d5a345eb340bb992ba13ce2a8d6398903c9b99"}, "score_mean": 0.3157894736842105, "score_ci_low": 0.2885640727931766, "score_ci_high": 0.3443402830588524, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 22257.628759398496, "output_tokens_ci_low": 19885.983082706767, "output_tokens_ci_high": 24688.33552631579, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 38.535714285714285, "steps_ci_low": 35.25093984962406, "steps_ci_high": 41.8796992481203, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2-pure/iter-59/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "aod", "method": "AOD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2-pure", "checkpoint_iteration": 59, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2-pure/iter_0000059", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_aod_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_aod_measured_pass8_cap120.tsv": "6947a6e2d1260bfad983f26aa7d5a345eb340bb992ba13ce2a8d6398903c9b99"}, "score_mean": 0.29887218045112784, "score_ci_low": 0.2721297871459973, "score_ci_high": 0.32706170342713653, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 22585.712406015038, "output_tokens_ci_low": 20288.879699248122, "output_tokens_ci_high": 25004.656015037595, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 38.108082706766915, "steps_ci_low": 34.7312030075188, "steps_ci_high": 41.43609022556391, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2-pure/iter-69/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "aod", "method": "AOD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2-pure", "checkpoint_iteration": 69, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2-pure/iter_0000069", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_aod_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_aod_measured_pass8_cap120.tsv": "6947a6e2d1260bfad983f26aa7d5a345eb340bb992ba13ce2a8d6398903c9b99"}, "score_mean": 0.30639097744360905, "score_ci_low": 0.27942853902943515, "score_ci_high": 0.3347464472232076, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 22398.443609022557, "output_tokens_ci_low": 20006.915413533836, "output_tokens_ci_high": 24946.399436090225, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 40.598684210526315, "steps_ci_low": 37.07612781954887, "steps_ci_high": 44.204887218045116, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2-pure/iter-79/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "aod", "method": "AOD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2-pure", "checkpoint_iteration": 79, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2-pure/iter_0000079", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_aod_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_aod_measured_pass8_cap120.tsv": "6947a6e2d1260bfad983f26aa7d5a345eb340bb992ba13ce2a8d6398903c9b99"}, "score_mean": 0.3167293233082707, "score_ci_low": 0.2894783535943107, "score_ci_high": 0.34529893921765703, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 21416.74060150376, "output_tokens_ci_low": 19168.781954887218, "output_tokens_ci_high": 23829.293233082706, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 39.17763157894737, "steps_ci_low": 35.6437969924812, "steps_ci_high": 42.88815789473684, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2-pure/iter-90/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "aod", "method": "AOD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2-pure", "checkpoint_iteration": 90, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2-pure/iter_0000090", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_aod_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_aod_measured_pass8_cap120.tsv": "6947a6e2d1260bfad983f26aa7d5a345eb340bb992ba13ce2a8d6398903c9b99"}, "score_mean": 0.30545112781954886, "score_ci_low": 0.27851572031595856, "score_ci_high": 0.3337863289767455, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 18720.341165413534, "output_tokens_ci_low": 16770.04887218045, "output_tokens_ci_high": 20649.862781954886, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 34.00657894736842, "steps_ci_low": 31.05639097744361, "steps_ci_high": 37.12312030075188, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2-pure/iter-99/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "aod", "method": "AOD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2-pure", "checkpoint_iteration": 99, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2-pure/iter_0000099", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_aod_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_aod_measured_pass8_cap120.tsv": "6947a6e2d1260bfad983f26aa7d5a345eb340bb992ba13ce2a8d6398903c9b99"}, "score_mean": 0.30451127819548873, "score_ci_low": 0.27760303653302076, "score_ci_high": 0.33282607579974477, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 19251.456766917294, "output_tokens_ci_low": 17276.46522556391, "output_tokens_ci_high": 21357.84962406015, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 35.7593984962406, "steps_ci_low": 32.37875939849624, "steps_ci_high": 39.278195488721806, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2-pure/iter-109/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "aod", "method": "AOD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2-pure", "checkpoint_iteration": 109, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2-pure/iter_0000109", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_aod_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_aod_measured_pass8_cap120.tsv": "6947a6e2d1260bfad983f26aa7d5a345eb340bb992ba13ce2a8d6398903c9b99"}, "score_mean": 0.3157894736842105, "score_ci_low": 0.2885640727931766, "score_ci_high": 0.3443402830588524, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 19984.967105263157, "output_tokens_ci_low": 17923.049812030076, "output_tokens_ci_high": 22214.112781954886, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 35.59304511278196, "steps_ci_low": 32.409774436090224, "steps_ci_high": 38.79793233082707, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2-pure/iter-119/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "aod", "method": "AOD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2-pure", "checkpoint_iteration": 119, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2-pure/iter_0000119", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_aod_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_aod_measured_pass8_cap120.tsv": "6947a6e2d1260bfad983f26aa7d5a345eb340bb992ba13ce2a8d6398903c9b99"}, "score_mean": 0.325187969924812, "score_ci_low": 0.29771273834700984, "score_ci_high": 0.35392098710440545, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 20277.146616541355, "output_tokens_ci_low": 18008.333646616542, "output_tokens_ci_high": 22702.89097744361, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 39.42387218045113, "steps_ci_low": 35.545112781954884, "steps_ci_high": 43.43515037593985, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2-pure/iter-129/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "aod", "method": "AOD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2-pure", "checkpoint_iteration": 129, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2-pure/iter_0000129", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_aod_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_aod_measured_pass8_cap120.tsv": "6947a6e2d1260bfad983f26aa7d5a345eb340bb992ba13ce2a8d6398903c9b99"}, "score_mean": 0.3101503759398496, "score_ci_low": 0.28308115631917546, "score_ci_high": 0.3385855777732218, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 18221.76597744361, "output_tokens_ci_low": 16247.336466165414, "output_tokens_ci_high": 20262.504699248122, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 35.776315789473685, "steps_ci_low": 32.34586466165413, "steps_ci_high": 39.234022556390975, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2-pure/iter-139/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "aod", "method": "AOD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2-pure", "checkpoint_iteration": 139, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2-pure/iter_0000139", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_aod_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_aod_measured_pass8_cap120.tsv": "6947a6e2d1260bfad983f26aa7d5a345eb340bb992ba13ce2a8d6398903c9b99"}, "score_mean": 0.3242481203007519, "score_ci_low": 0.2967972896419412, "score_ci_high": 0.3529634988495355, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 21025.879699248122, "output_tokens_ci_low": 18770.42387218045, "output_tokens_ci_high": 23437.855263157893, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 35.56954887218045, "steps_ci_low": 32.287593984962406, "steps_ci_high": 39.00375939849624, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-pure/iter-6/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "aod2op", "method": "AOD on-policy", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-pure", "checkpoint_iteration": 6, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-pure/iter_0000006", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_aod2op_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_aod2op_measured_pass8_cap120.tsv": "0b11e635bc44abd4d07250f8675701617fb0c6eab660c4bdbf4532916f6cbbe6"}, "score_mean": 0.162593984962406, "score_ci_low": 0.14164243899379647, "score_ci_high": 0.18597319238823617, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 27934.888157894737, "output_tokens_ci_low": 25649.54887218045, "output_tokens_ci_high": 30247.76597744361, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 61.8937969924812, "steps_ci_low": 57.17763157894737, "steps_ci_high": 66.5234962406015, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-pure/iter-13/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "aod2op", "method": "AOD on-policy", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-pure", "checkpoint_iteration": 13, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-pure/iter_0000013", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_aod2op_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_aod2op_measured_pass8_cap120.tsv": "0b11e635bc44abd4d07250f8675701617fb0c6eab660c4bdbf4532916f6cbbe6"}, "score_mean": 0.2199248120300752, "score_ci_low": 0.19606874225959345, "score_ci_high": 0.24579604367869548, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 36270.17105263158, "output_tokens_ci_low": 33749.892857142855, "output_tokens_ci_high": 38749.5272556391, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 58.87781954887218, "steps_ci_low": 54.11372180451128, "steps_ci_high": 63.65601503759399, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-pure/iter-20/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "aod2op", "method": "AOD on-policy", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-pure", "checkpoint_iteration": 20, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-pure/iter_0000020", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_aod2op_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_aod2op_measured_pass8_cap120.tsv": "0b11e635bc44abd4d07250f8675701617fb0c6eab660c4bdbf4532916f6cbbe6"}, "score_mean": 0.27349624060150374, "score_ci_low": 0.24756263033895584, "score_ci_high": 0.30105956231583486, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 23683.14379699248, "output_tokens_ci_low": 21657.11560150376, "output_tokens_ci_high": 25756.255639097744, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 43.28947368421053, "steps_ci_low": 39.60338345864662, "steps_ci_high": 47.05451127819549, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-pure/iter-29/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "aod2op", "method": "AOD on-policy", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-pure", "checkpoint_iteration": 29, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-pure/iter_0000029", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_aod2op_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_aod2op_measured_pass8_cap120.tsv": "0b11e635bc44abd4d07250f8675701617fb0c6eab660c4bdbf4532916f6cbbe6"}, "score_mean": 0.3026315789473684, "score_ci_low": 0.27577807515792835, "score_ci_high": 0.33090516325495983, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 17587.985902255637, "output_tokens_ci_low": 16043.49060150376, "output_tokens_ci_high": 19203.342105263157, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 35.399436090225564, "steps_ci_low": 32.46052631578947, "steps_ci_high": 38.463345864661655, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-pure/iter-39/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "aod2op", "method": "AOD on-policy", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-pure", "checkpoint_iteration": 39, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-pure/iter_0000039", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_aod2op_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_aod2op_measured_pass8_cap120.tsv": "0b11e635bc44abd4d07250f8675701617fb0c6eab660c4bdbf4532916f6cbbe6"}, "score_mean": 0.2941729323308271, "score_ci_low": 0.2675725224767589, "score_ci_high": 0.3222542832966817, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 14302.868421052632, "output_tokens_ci_low": 12953.141917293233, "output_tokens_ci_high": 15696.166353383458, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 31.554511278195488, "steps_ci_low": 28.834586466165412, "steps_ci_high": 34.410714285714285, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-pure/iter-49/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "aod2op", "method": "AOD on-policy", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-pure", "checkpoint_iteration": 49, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-pure/iter_0000049", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_aod2op_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_aod2op_measured_pass8_cap120.tsv": "0b11e635bc44abd4d07250f8675701617fb0c6eab660c4bdbf4532916f6cbbe6"}, "score_mean": 0.2941729323308271, "score_ci_low": 0.2675725224767589, "score_ci_high": 0.3222542832966817, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 15009.526315789473, "output_tokens_ci_low": 13538.770676691729, "output_tokens_ci_high": 16663.976503759397, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 31.890037593984964, "steps_ci_low": 29.002819548872182, "steps_ci_high": 34.840225563909776, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-pure/iter-59/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "aod2op", "method": "AOD on-policy", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-pure", "checkpoint_iteration": 59, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-pure/iter_0000059", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_aod2op_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_aod2op_measured_pass8_cap120.tsv": "0b11e635bc44abd4d07250f8675701617fb0c6eab660c4bdbf4532916f6cbbe6"}, "score_mean": 0.31860902255639095, "score_ci_low": 0.2913073078308569, "score_ci_high": 0.34721585890098794, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 15297.626879699248, "output_tokens_ci_low": 13762.906015037594, "output_tokens_ci_high": 16930.68233082707, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 32.07142857142857, "steps_ci_low": 29.31296992481203, "steps_ci_high": 35.00093984962406, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-pure/iter-69/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "aod2op", "method": "AOD on-policy", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-pure", "checkpoint_iteration": 69, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-pure/iter_0000069", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_aod2op_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_aod2op_measured_pass8_cap120.tsv": "0b11e635bc44abd4d07250f8675701617fb0c6eab660c4bdbf4532916f6cbbe6"}, "score_mean": 0.31109022556390975, "score_ci_low": 0.2839946445603325, "score_ci_high": 0.33954502649200324, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 15496.293233082706, "output_tokens_ci_low": 13929.832706766918, "output_tokens_ci_high": 17161.962406015038, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 38.43045112781955, "steps_ci_low": 35.160714285714285, "steps_ci_high": 41.76127819548872, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-pure/iter-79/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "aod2op", "method": "AOD on-policy", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-pure", "checkpoint_iteration": 79, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-pure/iter_0000079", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_aod2op_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_aod2op_measured_pass8_cap120.tsv": "0b11e635bc44abd4d07250f8675701617fb0c6eab660c4bdbf4532916f6cbbe6"}, "score_mean": 0.2744360902255639, "score_ci_low": 0.2484706448984977, "score_ci_high": 0.3020244847162317, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 14458.013157894737, "output_tokens_ci_low": 13061.338345864662, "output_tokens_ci_high": 15956.421992481202, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 52.56954887218045, "steps_ci_low": 47.88533834586466, "steps_ci_high": 57.159774436090224, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-pure/iter-90/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "aod2op", "method": "AOD on-policy", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-pure", "checkpoint_iteration": 90, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-pure/iter_0000090", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_aod2op_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_aod2op_measured_pass8_cap120.tsv": "0b11e635bc44abd4d07250f8675701617fb0c6eab660c4bdbf4532916f6cbbe6"}, "score_mean": 0.3242481203007519, "score_ci_low": 0.2967972896419412, "score_ci_high": 0.3529634988495355, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 13658.75469924812, "output_tokens_ci_low": 12214.372180451128, "output_tokens_ci_high": 15252.046992481202, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 37.722744360902254, "steps_ci_low": 34.24812030075188, "steps_ci_high": 41.25657894736842, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-pure/iter-99/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "aod2op", "method": "AOD on-policy", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-pure", "checkpoint_iteration": 99, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-pure/iter_0000099", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_aod2op_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_aod2op_measured_pass8_cap120.tsv": "0b11e635bc44abd4d07250f8675701617fb0c6eab660c4bdbf4532916f6cbbe6"}, "score_mean": 0.3308270676691729, "score_ci_low": 0.3032081147585046, "score_ci_high": 0.3596632324525424, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 16380.598684210525, "output_tokens_ci_low": 14736.038533834586, "output_tokens_ci_high": 18125.929511278195, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 35.391917293233085, "steps_ci_low": 32.346804511278194, "steps_ci_high": 38.6390977443609, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-pure/iter-109/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "aod2op", "method": "AOD on-policy", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-pure", "checkpoint_iteration": 109, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-pure/iter_0000109", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_aod2op_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_aod2op_measured_pass8_cap120.tsv": "0b11e635bc44abd4d07250f8675701617fb0c6eab660c4bdbf4532916f6cbbe6"}, "score_mean": 0.34398496240601506, "score_ci_low": 0.3160482884596301, "score_ci_high": 0.3730441761905578, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 16041.456766917292, "output_tokens_ci_low": 14320.145676691729, "output_tokens_ci_high": 17819.627819548874, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 37.557330827067666, "steps_ci_low": 34.55075187969925, "steps_ci_high": 40.78947368421053, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-pure/iter-119/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "aod2op", "method": "AOD on-policy", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-pure", "checkpoint_iteration": 119, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-pure/iter_0000119", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_aod2op_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_aod2op_measured_pass8_cap120.tsv": "0b11e635bc44abd4d07250f8675701617fb0c6eab660c4bdbf4532916f6cbbe6"}, "score_mean": 0.32800751879699247, "score_ci_low": 0.3004598530619603, "score_ci_high": 0.3567926832692708, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 15244.981203007519, "output_tokens_ci_low": 13466.334586466166, "output_tokens_ci_high": 17152.461466165412, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 35.431390977443606, "steps_ci_low": 32.567669172932334, "steps_ci_high": 38.471804511278194, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-pure/iter-129/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "aod2op", "method": "AOD on-policy", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-pure", "checkpoint_iteration": 129, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-pure/iter_0000129", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_aod2op_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_aod2op_measured_pass8_cap120.tsv": "0b11e635bc44abd4d07250f8675701617fb0c6eab660c4bdbf4532916f6cbbe6"}, "score_mean": 0.3298872180451128, "score_ci_low": 0.3022919005472217, "score_ci_high": 0.3587065097038867, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 13256.86090225564, "output_tokens_ci_low": 11826.950187969926, "output_tokens_ci_high": 14699.791353383458, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 31.11466165413534, "steps_ci_low": 28.466165413533833, "steps_ci_high": 33.80921052631579, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-pure/iter-139/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "aod2op", "method": "AOD on-policy", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-pure", "checkpoint_iteration": 139, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-pure/iter_0000139", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_aod2op_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_aod2op_measured_pass8_cap120.tsv": "0b11e635bc44abd4d07250f8675701617fb0c6eab660c4bdbf4532916f6cbbe6"}, "score_mean": 0.33646616541353386, "score_ci_low": 0.30870804954688436, "score_ci_high": 0.3654009194237945, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 14209.71052631579, "output_tokens_ci_low": 12588.232142857143, "output_tokens_ci_high": 15979.093984962406, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 35.71052631578947, "steps_ci_low": 32.590225563909776, "steps_ci_high": 38.98872180451128, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter-6/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "aod2optail", "method": "AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure", "checkpoint_iteration": 6, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter_0000006", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_aod2optail_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb", "label_hill_climb"], "source_sha256": {"notes/figures/series/series_aod2optail_measured_pass8_cap120.tsv": "543f9ca6556cd58e0c144f3484375fcb2c5195156219708a70e1bb34c6e4419b"}, "score_mean": 0.20394736842105263, "score_ci_low": 0.1808214232300216, "score_ci_high": 0.22920343438931068, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 29877.649436090225, "output_tokens_ci_low": 27435.19172932331, "output_tokens_ci_high": 32449.16165413534, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 55.51973684210526, "steps_ci_low": 51.06203007518797, "steps_ci_high": 59.79887218045113, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter-13/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "aod2optail", "method": "AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure", "checkpoint_iteration": 13, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter_0000013", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_aod2optail_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb", "label_hill_climb"], "source_sha256": {"notes/figures/series/series_aod2optail_measured_pass8_cap120.tsv": "543f9ca6556cd58e0c144f3484375fcb2c5195156219708a70e1bb34c6e4419b"}, "score_mean": 0.27631578947368424, "score_ci_low": 0.25028711853039076, "score_ci_high": 0.30395388500421594, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 24958.573308270676, "output_tokens_ci_low": 22623.10902255639, "output_tokens_ci_high": 27429.891917293233, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 32.097744360902254, "steps_ci_low": 29.038533834586467, "steps_ci_high": 35.27161654135338, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter-20/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "aod2optail", "method": "AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure", "checkpoint_iteration": 20, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter_0000020", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_aod2optail_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb", "label_hill_climb"], "source_sha256": {"notes/figures/series/series_aod2optail_measured_pass8_cap120.tsv": "543f9ca6556cd58e0c144f3484375fcb2c5195156219708a70e1bb34c6e4419b"}, "score_mean": 0.3073308270676692, "score_ci_low": 0.2803414923276277, "score_ci_high": 0.3357064308849537, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 13045.054511278195, "output_tokens_ci_low": 11596.855263157895, "output_tokens_ci_high": 14646.929511278195, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 34.50375939849624, "steps_ci_low": 31.17763157894737, "steps_ci_high": 37.90883458646616, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter-29/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "aod2optail", "method": "AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure", "checkpoint_iteration": 29, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter_0000029", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_aod2optail_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb", "label_hill_climb"], "source_sha256": {"notes/figures/series/series_aod2optail_measured_pass8_cap120.tsv": "543f9ca6556cd58e0c144f3484375fcb2c5195156219708a70e1bb34c6e4419b"}, "score_mean": 0.32894736842105265, "score_ci_low": 0.30137581322322493, "score_ci_high": 0.35774966006794495, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 13176.018796992481, "output_tokens_ci_low": 11697.738721804511, "output_tokens_ci_high": 14740.727443609023, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 28.36842105263158, "steps_ci_low": 25.780075187969924, "steps_ci_high": 31.087406015037594, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter-39/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "aod2optail", "method": "AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure", "checkpoint_iteration": 39, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter_0000039", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_aod2optail_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb", "label_hill_climb"], "source_sha256": {"notes/figures/series/series_aod2optail_measured_pass8_cap120.tsv": "543f9ca6556cd58e0c144f3484375fcb2c5195156219708a70e1bb34c6e4419b"}, "score_mean": 0.35620300751879697, "score_ci_low": 0.327992921094081, "score_ci_high": 0.38544772403530897, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 13744.727443609023, "output_tokens_ci_low": 12204.924812030074, "output_tokens_ci_high": 15385.135338345865, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 29.898496240601503, "steps_ci_low": 27.14097744360902, "steps_ci_high": 32.755639097744364, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter-49/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "aod2optail", "method": "AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure", "checkpoint_iteration": 49, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter_0000049", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_aod2optail_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb", "label_hill_climb"], "source_sha256": {"notes/figures/series/series_aod2optail_measured_pass8_cap120.tsv": "543f9ca6556cd58e0c144f3484375fcb2c5195156219708a70e1bb34c6e4419b"}, "score_mean": 0.3656015037593985, "score_ci_low": 0.33719494079366086, "score_ci_high": 0.39497507393511533, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 13941.217105263158, "output_tokens_ci_low": 12194.08082706767, "output_tokens_ci_high": 15940.59962406015, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 28.337406015037594, "steps_ci_low": 25.81860902255639, "steps_ci_high": 31.031015037593985, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter-59/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "aod2optail", "method": "AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure", "checkpoint_iteration": 59, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter_0000059", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_aod2optail_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb", "label_hill_climb"], "source_sha256": {"notes/figures/series/series_aod2optail_measured_pass8_cap120.tsv": "543f9ca6556cd58e0c144f3484375fcb2c5195156219708a70e1bb34c6e4419b"}, "score_mean": 0.3731203007518797, "score_ci_low": 0.3445650700547768, "score_ci_high": 0.40258844035350844, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 12808.666353383458, "output_tokens_ci_low": 11147.390037593985, "output_tokens_ci_high": 14665.660714285714, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 26.92105263157895, "steps_ci_low": 24.401315789473685, "steps_ci_high": 29.577067669172934, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter-69/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "aod2optail", "method": "AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure", "checkpoint_iteration": 69, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter_0000069", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_aod2optail_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb", "label_hill_climb"], "source_sha256": {"notes/figures/series/series_aod2optail_measured_pass8_cap120.tsv": "543f9ca6556cd58e0c144f3484375fcb2c5195156219708a70e1bb34c6e4419b"}, "score_mean": 0.3768796992481203, "score_ci_low": 0.3482529394885122, "score_ci_high": 0.40639231875952764, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 13677.937969924813, "output_tokens_ci_low": 11910.881578947368, "output_tokens_ci_high": 15600.586466165414, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 27.169172932330827, "steps_ci_low": 24.664473684210527, "steps_ci_high": 29.827067669172934, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter-79/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "aod2optail", "method": "AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure", "checkpoint_iteration": 79, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter_0000079", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_aod2optail_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb", "label_hill_climb"], "source_sha256": {"notes/figures/series/series_aod2optail_measured_pass8_cap120.tsv": "543f9ca6556cd58e0c144f3484375fcb2c5195156219708a70e1bb34c6e4419b"}, "score_mean": 0.3684210526315789, "score_ci_low": 0.33995785867697487, "score_ci_high": 0.3978309669316173, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 13539.529135338345, "output_tokens_ci_low": 11618.552631578947, "output_tokens_ci_high": 15772.937969924813, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 27.1203007518797, "steps_ci_low": 24.597744360902254, "steps_ci_high": 29.74436090225564, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter-90/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "aod2optail", "method": "AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure", "checkpoint_iteration": 90, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter_0000090", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_aod2optail_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb", "label_hill_climb"], "source_sha256": {"notes/figures/series/series_aod2optail_measured_pass8_cap120.tsv": "543f9ca6556cd58e0c144f3484375fcb2c5195156219708a70e1bb34c6e4419b"}, "score_mean": 0.37781954887218044, "score_ci_low": 0.3491751970934815, "score_ci_high": 0.40734299811449687, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 12676.779135338345, "output_tokens_ci_low": 11145.312030075187, "output_tokens_ci_high": 14369.702067669174, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 28.987781954887218, "steps_ci_low": 26.06484962406015, "steps_ci_high": 32.036654135338345, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter-99/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "aod2optail", "method": "AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure", "checkpoint_iteration": 99, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter_0000099", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_aod2optail_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb", "label_hill_climb"], "source_sha256": {"notes/figures/series/series_aod2optail_measured_pass8_cap120.tsv": "543f9ca6556cd58e0c144f3484375fcb2c5195156219708a70e1bb34c6e4419b"}, "score_mean": 0.36654135338345867, "score_ci_low": 0.3381157955738945, "score_ci_high": 0.39592715611482043, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 12496.271616541353, "output_tokens_ci_low": 10761.980263157895, "output_tokens_ci_high": 14469.40037593985, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 29.223684210526315, "steps_ci_low": 26.292293233082706, "steps_ci_high": 32.295112781954884, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter-109/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "aod2optail", "method": "AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure", "checkpoint_iteration": 109, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter_0000109", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_aod2optail_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb", "label_hill_climb"], "source_sha256": {"notes/figures/series/series_aod2optail_measured_pass8_cap120.tsv": "543f9ca6556cd58e0c144f3484375fcb2c5195156219708a70e1bb34c6e4419b"}, "score_mean": 0.3684210526315789, "score_ci_low": 0.33995785867697487, "score_ci_high": 0.3978309669316173, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 14063.479323308271, "output_tokens_ci_low": 12019.40977443609, "output_tokens_ci_high": 16222.364661654137, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 27.68703007518797, "steps_ci_low": 25.06672932330827, "steps_ci_high": 30.341165413533833, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter-119/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "aod2optail", "method": "AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure", "checkpoint_iteration": 119, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter_0000119", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_aod2optail_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb", "label_hill_climb"], "source_sha256": {"notes/figures/series/series_aod2optail_measured_pass8_cap120.tsv": "543f9ca6556cd58e0c144f3484375fcb2c5195156219708a70e1bb34c6e4419b"}, "score_mean": 0.39097744360902253, "score_ci_low": 0.36209888376477495, "score_ci_high": 0.4206404288823442, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 12497.176691729323, "output_tokens_ci_low": 10831.947368421053, "output_tokens_ci_high": 14295.32142857143, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 28.356203007518797, "steps_ci_low": 25.578007518796994, "steps_ci_high": 31.2453007518797, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter-129/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "aod2optail", "method": "AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure", "checkpoint_iteration": 129, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter_0000129", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_aod2optail_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb", "label_hill_climb"], "source_sha256": {"notes/figures/series/series_aod2optail_measured_pass8_cap120.tsv": "543f9ca6556cd58e0c144f3484375fcb2c5195156219708a70e1bb34c6e4419b"}, "score_mean": 0.375, "score_ci_low": 0.3464087722286727, "score_ci_high": 0.4044906120994898, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 13051.141917293233, "output_tokens_ci_low": 11385.543233082706, "output_tokens_ci_high": 14860.773496240601, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 29.478383458646615, "steps_ci_low": 26.42105263157895, "steps_ci_high": 32.714285714285715, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter-139/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "aod2optail", "method": "AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure", "checkpoint_iteration": 139, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter_0000139", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_aod2optail_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb", "label_hill_climb"], "source_sha256": {"notes/figures/series/series_aod2optail_measured_pass8_cap120.tsv": "543f9ca6556cd58e0c144f3484375fcb2c5195156219708a70e1bb34c6e4419b"}, "score_mean": 0.3768796992481203, "score_ci_low": 0.3482529394885122, "score_ci_high": 0.40639231875952764, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 15588.517857142857, "output_tokens_ci_low": 13548.981203007519, "output_tokens_ci_high": 17811.771616541355, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 31.045112781954888, "steps_ci_low": 27.835526315789473, "steps_ci_high": 34.47838345864662, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-pure/iter-6/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "thinkonly", "method": "think-only", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-pure", "checkpoint_iteration": 6, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-pure/iter_0000006", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_thinkonly_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_thinkonly_measured_pass8_cap120.tsv": "a7c5e4c27973e9422f6d1dd0a91168b1d30804dd37d74fcbc66fc7ee0524a37d"}, "score_mean": 0.20676691729323307, "score_ci_low": 0.18350803925907969, "score_ci_high": 0.23213562924006845, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 38393.03947368421, "output_tokens_ci_low": 35586.3007518797, "output_tokens_ci_high": 41272.66917293233, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 75.09398496240601, "steps_ci_low": 69.84304511278195, "steps_ci_high": 80.14755639097744, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-pure/iter-13/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "thinkonly", "method": "think-only", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-pure", "checkpoint_iteration": 13, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-pure/iter_0000013", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_thinkonly_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_thinkonly_measured_pass8_cap120.tsv": "a7c5e4c27973e9422f6d1dd0a91168b1d30804dd37d74fcbc66fc7ee0524a37d"}, "score_mean": 0.23966165413533835, "score_ci_low": 0.2149772509710471, "score_ci_high": 0.266219211125953, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 39118.31296992481, "output_tokens_ci_low": 36060.2265037594, "output_tokens_ci_high": 42207.734962406015, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 67.32048872180451, "steps_ci_low": 61.795112781954884, "steps_ci_high": 72.85432330827068, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-pure/iter-20/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "thinkonly", "method": "think-only", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-pure", "checkpoint_iteration": 20, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-pure/iter_0000020", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_thinkonly_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_thinkonly_measured_pass8_cap120.tsv": "a7c5e4c27973e9422f6d1dd0a91168b1d30804dd37d74fcbc66fc7ee0524a37d"}, "score_mean": 0.2819548872180451, "score_ci_low": 0.2557400622822514, "score_ci_high": 0.30973856301198704, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 34696.10620300752, "output_tokens_ci_low": 31877.975563909775, "output_tokens_ci_high": 37663.13157894737, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 71.421992481203, "steps_ci_low": 66.03195488721805, "steps_ci_high": 76.90695488721805, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-pure/iter-29/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "thinkonly", "method": "think-only", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-pure", "checkpoint_iteration": 29, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-pure/iter_0000029", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_thinkonly_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_thinkonly_measured_pass8_cap120.tsv": "a7c5e4c27973e9422f6d1dd0a91168b1d30804dd37d74fcbc66fc7ee0524a37d"}, "score_mean": 0.2800751879699248, "score_ci_low": 0.2539218308558332, "score_ci_high": 0.307810920518528, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 27771.303571428572, "output_tokens_ci_low": 24836.919172932332, "output_tokens_ci_high": 30878.470864661653, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 52.66447368421053, "steps_ci_low": 46.96616541353384, "steps_ci_high": 58.45112781954887, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-pure/iter-39/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "thinkonly", "method": "think-only", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-pure", "checkpoint_iteration": 39, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-pure/iter_0000039", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_thinkonly_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_thinkonly_measured_pass8_cap120.tsv": "a7c5e4c27973e9422f6d1dd0a91168b1d30804dd37d74fcbc66fc7ee0524a37d"}, "score_mean": 0.30451127819548873, "score_ci_low": 0.27760303653302076, "score_ci_high": 0.33282607579974477, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 30979.122180451126, "output_tokens_ci_low": 27800.187969924813, "output_tokens_ci_high": 34305.265037593985, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 54.76973684210526, "steps_ci_low": 49.1296992481203, "steps_ci_high": 60.51973684210526, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-pure/iter-49/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "thinkonly", "method": "think-only", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-pure", "checkpoint_iteration": 49, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-pure/iter_0000049", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_thinkonly_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_thinkonly_measured_pass8_cap120.tsv": "a7c5e4c27973e9422f6d1dd0a91168b1d30804dd37d74fcbc66fc7ee0524a37d"}, "score_mean": 0.32048872180451127, "score_ci_low": 0.2931367835268187, "score_ci_high": 0.34913225712490337, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 28812.999060150374, "output_tokens_ci_low": 25986.541353383458, "output_tokens_ci_high": 31720.50845864662, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 59.524436090225564, "steps_ci_low": 53.83176691729323, "steps_ci_high": 65.3562030075188, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-pure/iter-59/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "thinkonly", "method": "think-only", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-pure", "checkpoint_iteration": 59, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-pure/iter_0000059", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_thinkonly_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_thinkonly_measured_pass8_cap120.tsv": "a7c5e4c27973e9422f6d1dd0a91168b1d30804dd37d74fcbc66fc7ee0524a37d"}, "score_mean": 0.31203007518796994, "score_ci_low": 0.2849082657035805, "score_ci_high": 0.340504342308694, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 25295.458646616542, "output_tokens_ci_low": 22549.29887218045, "output_tokens_ci_high": 28083.992481203008, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 51.74342105263158, "steps_ci_low": 46.453947368421055, "steps_ci_high": 57.01221804511278, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-pure/iter-69/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "thinkonly", "method": "think-only", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-pure", "checkpoint_iteration": 69, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-pure/iter_0000069", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_thinkonly_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_thinkonly_measured_pass8_cap120.tsv": "a7c5e4c27973e9422f6d1dd0a91168b1d30804dd37d74fcbc66fc7ee0524a37d"}, "score_mean": 0.31860902255639095, "score_ci_low": 0.2913073078308569, "score_ci_high": 0.34721585890098794, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 26744.486842105263, "output_tokens_ci_low": 23854.63157894737, "output_tokens_ci_high": 29827.487781954886, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 52.21052631578947, "steps_ci_low": 46.56109022556391, "steps_ci_high": 57.92481203007519, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-pure/iter-79/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "thinkonly", "method": "think-only", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-pure", "checkpoint_iteration": 79, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-pure/iter_0000079", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_thinkonly_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_thinkonly_measured_pass8_cap120.tsv": "a7c5e4c27973e9422f6d1dd0a91168b1d30804dd37d74fcbc66fc7ee0524a37d"}, "score_mean": 0.3308270676691729, "score_ci_low": 0.3032081147585046, "score_ci_high": 0.3596632324525424, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 29823.894736842107, "output_tokens_ci_low": 26405.015037593985, "output_tokens_ci_high": 33301.96240601504, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 52.55357142857143, "steps_ci_low": 47.25751879699248, "steps_ci_high": 58.03101503759399, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-pure/iter-90/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "thinkonly", "method": "think-only", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-pure", "checkpoint_iteration": 90, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-pure/iter_0000090", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_thinkonly_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_thinkonly_measured_pass8_cap120.tsv": "a7c5e4c27973e9422f6d1dd0a91168b1d30804dd37d74fcbc66fc7ee0524a37d"}, "score_mean": 0.33364661654135336, "score_ci_low": 0.3059575159990184, "score_ci_high": 0.3625326420918445, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 26845.27819548872, "output_tokens_ci_low": 23715.809210526317, "output_tokens_ci_high": 30167.071428571428, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 46.75, "steps_ci_low": 41.775375939849624, "steps_ci_high": 51.80451127819549, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-pure/iter-99/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "thinkonly", "method": "think-only", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-pure", "checkpoint_iteration": 99, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-pure/iter_0000099", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_thinkonly_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_thinkonly_measured_pass8_cap120.tsv": "a7c5e4c27973e9422f6d1dd0a91168b1d30804dd37d74fcbc66fc7ee0524a37d"}, "score_mean": 0.33364661654135336, "score_ci_low": 0.3059575159990184, "score_ci_high": 0.3625326420918445, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 27412.062969924813, "output_tokens_ci_low": 24203.91447368421, "output_tokens_ci_high": 30755.105263157893, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 52.295112781954884, "steps_ci_low": 46.90695488721804, "steps_ci_high": 57.72932330827068, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-pure/iter-109/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "thinkonly", "method": "think-only", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-pure", "checkpoint_iteration": 109, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-pure/iter_0000109", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_thinkonly_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_thinkonly_measured_pass8_cap120.tsv": "a7c5e4c27973e9422f6d1dd0a91168b1d30804dd37d74fcbc66fc7ee0524a37d"}, "score_mean": 0.3308270676691729, "score_ci_low": 0.3032081147585046, "score_ci_high": 0.3596632324525424, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 27435.89285714286, "output_tokens_ci_low": 24361.501879699248, "output_tokens_ci_high": 30673.408834586466, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 52.56015037593985, "steps_ci_low": 47.037593984962406, "steps_ci_high": 58.15413533834587, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-pure/iter-119/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "thinkonly", "method": "think-only", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-pure", "checkpoint_iteration": 119, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-pure/iter_0000119", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_thinkonly_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_thinkonly_measured_pass8_cap120.tsv": "a7c5e4c27973e9422f6d1dd0a91168b1d30804dd37d74fcbc66fc7ee0524a37d"}, "score_mean": 0.3308270676691729, "score_ci_low": 0.3032081147585046, "score_ci_high": 0.3596632324525424, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 26310.781015037595, "output_tokens_ci_low": 23210.454887218046, "output_tokens_ci_high": 29489.875939849626, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 51.078947368421055, "steps_ci_low": 45.54229323308271, "steps_ci_high": 56.87687969924812, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-pure/iter-129/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "thinkonly", "method": "think-only", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-pure", "checkpoint_iteration": 129, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-pure/iter_0000129", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_thinkonly_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_thinkonly_measured_pass8_cap120.tsv": "a7c5e4c27973e9422f6d1dd0a91168b1d30804dd37d74fcbc66fc7ee0524a37d"}, "score_mean": 0.35620300751879697, "score_ci_low": 0.327992921094081, "score_ci_high": 0.38544772403530897, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 28750.11842105263, "output_tokens_ci_low": 25434.089285714286, "output_tokens_ci_high": 32270.697368421053, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 50.17293233082707, "steps_ci_low": 44.83646616541353, "steps_ci_high": 55.67857142857143, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-pure/iter-139/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "thinkonly", "method": "think-only", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-pure", "checkpoint_iteration": 139, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-pure/iter_0000139", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_thinkonly_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_thinkonly_measured_pass8_cap120.tsv": "a7c5e4c27973e9422f6d1dd0a91168b1d30804dd37d74fcbc66fc7ee0524a37d"}, "score_mean": 0.3092105263157895, "score_ci_low": 0.28216780131053343, "score_ci_high": 0.3376259958219252, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 27387.980263157893, "output_tokens_ci_low": 24398.53477443609, "output_tokens_ci_high": 30494.670112781954, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 53.17951127819549, "steps_ci_low": 47.68327067669173, "steps_ci_high": 58.79135338345865, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-maskthink-pure/iter-139/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "maskthink", "method": "mask-think", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-maskthink-pure", "checkpoint_iteration": 139, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-maskthink-pure/iter_0000139", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_maskthink_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_maskthink_measured_pass8_cap120.tsv": "96991c8be89cf09c7a71c5aa8e9fa78873bf1dbf6dda88ccf74421223c38fb38"}, "score_mean": 0.287593984962406, "score_ci_low": 0.26119821541810373, "score_ci_high": 0.3155180316357664, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 31883.501879699248, "output_tokens_ci_low": 28573.205827067668, "output_tokens_ci_high": 35291.00563909775, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 43.31109022556391, "steps_ci_low": 38.35338345864662, "steps_ci_high": 48.4953007518797, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-first130-pure/iter-6/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "first130", "method": "first 130 response tokens", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-first130-pure", "checkpoint_iteration": 6, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-first130-pure/iter_0000006", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_first130_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_first130_measured_pass8_cap120.tsv": "c82baf5112476b8793aa26e15338352813ebf870dda64920e6891cb920d9dd6e"}, "score_mean": 0.19454887218045114, "score_ci_low": 0.17187927313354398, "score_ci_high": 0.21941621488640203, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 42772.43796992481, "output_tokens_ci_low": 39663.09868421053, "output_tokens_ci_high": 45871.07330827068, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 75.57424812030075, "steps_ci_low": 70.43233082706767, "steps_ci_high": 80.79135338345864, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-first130-pure/iter-13/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "first130", "method": "first 130 response tokens", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-first130-pure", "checkpoint_iteration": 13, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-first130-pure/iter_0000013", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_first130_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_first130_measured_pass8_cap120.tsv": "c82baf5112476b8793aa26e15338352813ebf870dda64920e6891cb920d9dd6e"}, "score_mean": 0.20958646616541354, "score_ci_low": 0.18619643850965706, "score_ci_high": 0.235066040869307, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 42665.14943609023, "output_tokens_ci_low": 39032.246240601504, "output_tokens_ci_high": 46371.017857142855, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 68.0234962406015, "steps_ci_low": 62.483082706766915, "steps_ci_high": 73.53665413533835, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-first130-pure/iter-20/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "first130", "method": "first 130 response tokens", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-first130-pure", "checkpoint_iteration": 20, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-first130-pure/iter_0000020", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_first130_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_first130_measured_pass8_cap120.tsv": "c82baf5112476b8793aa26e15338352813ebf870dda64920e6891cb920d9dd6e"}, "score_mean": 0.26033834586466165, "score_ci_low": 0.23486627469017043, "score_ci_high": 0.2875348005254795, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 37308.55263157895, "output_tokens_ci_low": 33981.77537593985, "output_tokens_ci_high": 40665.54605263158, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 58.943609022556394, "steps_ci_low": 54.017857142857146, "steps_ci_high": 64.02161654135338, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-first130-pure/iter-29/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "first130", "method": "first 130 response tokens", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-first130-pure", "checkpoint_iteration": 29, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-first130-pure/iter_0000029", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_first130_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_first130_measured_pass8_cap120.tsv": "c82baf5112476b8793aa26e15338352813ebf870dda64920e6891cb920d9dd6e"}, "score_mean": 0.2866541353383459, "score_ci_low": 0.26028816543016065, "score_ci_high": 0.314555144663771, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 30202.542293233084, "output_tokens_ci_low": 27205.515037593985, "output_tokens_ci_high": 33287.128759398496, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 48.580827067669176, "steps_ci_low": 43.421992481203006, "steps_ci_high": 53.97932330827068, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-first130-pure/iter-39/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "first130", "method": "first 130 response tokens", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-first130-pure", "checkpoint_iteration": 39, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-first130-pure/iter_0000039", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_first130_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_first130_measured_pass8_cap120.tsv": "c82baf5112476b8793aa26e15338352813ebf870dda64920e6891cb920d9dd6e"}, "score_mean": 0.2800751879699248, "score_ci_low": 0.2539218308558332, "score_ci_high": 0.307810920518528, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 33995.93703007519, "output_tokens_ci_low": 30637.43327067669, "output_tokens_ci_high": 37436.954887218046, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 46.75281954887218, "steps_ci_low": 42.170112781954884, "steps_ci_high": 51.55921052631579, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-first130-pure/iter-49/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "first130", "method": "first 130 response tokens", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-first130-pure", "checkpoint_iteration": 49, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-first130-pure/iter_0000049", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_first130_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_first130_measured_pass8_cap120.tsv": "c82baf5112476b8793aa26e15338352813ebf870dda64920e6891cb920d9dd6e"}, "score_mean": 0.2744360902255639, "score_ci_low": 0.2484706448984977, "score_ci_high": 0.3020244847162317, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 34196.88909774436, "output_tokens_ci_low": 30627.78947368421, "output_tokens_ci_high": 37891.76127819549, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 53.534774436090224, "steps_ci_low": 48.11278195488722, "steps_ci_high": 59.232142857142854, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-first130-pure/iter-59/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "first130", "method": "first 130 response tokens", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-first130-pure", "checkpoint_iteration": 59, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-first130-pure/iter_0000059", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_first130_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_first130_measured_pass8_cap120.tsv": "c82baf5112476b8793aa26e15338352813ebf870dda64920e6891cb920d9dd6e"}, "score_mean": 0.2866541353383459, "score_ci_low": 0.26028816543016065, "score_ci_high": 0.314555144663771, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 31787.448308270676, "output_tokens_ci_low": 28086.927631578947, "output_tokens_ci_high": 35662.88815789474, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 46.8796992481203, "steps_ci_low": 41.60432330827068, "steps_ci_high": 52.43609022556391, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-first130-pure/iter-69/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "first130", "method": "first 130 response tokens", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-first130-pure", "checkpoint_iteration": 69, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-first130-pure/iter_0000069", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_first130_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_first130_measured_pass8_cap120.tsv": "c82baf5112476b8793aa26e15338352813ebf870dda64920e6891cb920d9dd6e"}, "score_mean": 0.29887218045112784, "score_ci_low": 0.2721297871459973, "score_ci_high": 0.32706170342713653, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 32119.049812030076, "output_tokens_ci_low": 28203.401315789473, "output_tokens_ci_high": 36142.24154135338, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 48.13345864661654, "steps_ci_low": 43.03383458646616, "steps_ci_high": 53.556390977443606, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-first130-pure/iter-79/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "first130", "method": "first 130 response tokens", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-first130-pure", "checkpoint_iteration": 79, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-first130-pure/iter_0000079", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_first130_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_first130_measured_pass8_cap120.tsv": "c82baf5112476b8793aa26e15338352813ebf870dda64920e6891cb920d9dd6e"}, "score_mean": 0.2725563909774436, "score_ci_low": 0.24665476458067637, "score_ci_high": 0.30009449111417574, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 32383.289473684206, "output_tokens_ci_low": 28816.63157894737, "output_tokens_ci_high": 36025.03759398496, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 49.31015037593985, "steps_ci_low": 43.931390977443606, "steps_ci_high": 54.76033834586466, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-first130-pure/iter-90/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "first130", "method": "first 130 response tokens", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-first130-pure", "checkpoint_iteration": 90, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-first130-pure/iter_0000090", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_first130_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_first130_measured_pass8_cap120.tsv": "c82baf5112476b8793aa26e15338352813ebf870dda64920e6891cb920d9dd6e"}, "score_mean": 0.3223684210526316, "score_ci_low": 0.2949667782575094, "score_ci_high": 0.35104813631409004, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 32674.062030075183, "output_tokens_ci_low": 28833.431390977443, "output_tokens_ci_high": 36795.515037593985, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 49.69736842105263, "steps_ci_low": 44.38251879699248, "steps_ci_high": 55.17763157894737, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-first130-pure/iter-99/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "first130", "method": "first 130 response tokens", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-first130-pure", "checkpoint_iteration": 99, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-first130-pure/iter_0000099", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_first130_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_first130_measured_pass8_cap120.tsv": "c82baf5112476b8793aa26e15338352813ebf870dda64920e6891cb920d9dd6e"}, "score_mean": 0.3223684210526316, "score_ci_low": 0.2949667782575094, "score_ci_high": 0.35104813631409004, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 30428.70206766917, "output_tokens_ci_low": 26790.543233082706, "output_tokens_ci_high": 34307.80639097744, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 46.28947368421053, "steps_ci_low": 41.0921052631579, "steps_ci_high": 51.81578947368421, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-first130-pure/iter-109/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "first130", "method": "first 130 response tokens", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-first130-pure", "checkpoint_iteration": 109, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-first130-pure/iter_0000109", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_first130_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_first130_measured_pass8_cap120.tsv": "c82baf5112476b8793aa26e15338352813ebf870dda64920e6891cb920d9dd6e"}, "score_mean": 0.3233082706766917, "score_ci_low": 0.2958819695151336, "score_ci_high": 0.3520058820164043, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 31185.214285714286, "output_tokens_ci_low": 27470.871240601504, "output_tokens_ci_high": 34993.25469924812, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 44.74812030075188, "steps_ci_low": 39.56203007518797, "steps_ci_high": 50.21804511278196, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-first130-pure/iter-119/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "first130", "method": "first 130 response tokens", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-first130-pure", "checkpoint_iteration": 119, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-first130-pure/iter_0000119", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_first130_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_first130_measured_pass8_cap120.tsv": "c82baf5112476b8793aa26e15338352813ebf870dda64920e6891cb920d9dd6e"}, "score_mean": 0.33364661654135336, "score_ci_low": 0.3059575159990184, "score_ci_high": 0.3625326420918445, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 32016.893796992485, "output_tokens_ci_low": 27954.529135338347, "output_tokens_ci_high": 36182.0742481203, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 44.32048872180451, "steps_ci_low": 39.662593984962406, "steps_ci_high": 49.234022556390975, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-first130-pure/iter-129/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "first130", "method": "first 130 response tokens", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-first130-pure", "checkpoint_iteration": 129, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-first130-pure/iter_0000129", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_first130_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_first130_measured_pass8_cap120.tsv": "c82baf5112476b8793aa26e15338352813ebf870dda64920e6891cb920d9dd6e"}, "score_mean": 0.32142857142857145, "score_ci_low": 0.29405171616264975, "score_ci_high": 0.35009026144901106, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 35034.06954887218, "output_tokens_ci_low": 30769.387218045114, "output_tokens_ci_high": 39487.671992481206, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 45.346804511278194, "steps_ci_low": 40.54793233082707, "steps_ci_high": 50.524436090225564, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-first130-pure/iter-139/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "first130", "method": "first 130 response tokens", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-first130-pure", "checkpoint_iteration": 139, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-thinkdist-first130-pure/iter_0000139", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_first130_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_first130_measured_pass8_cap120.tsv": "c82baf5112476b8793aa26e15338352813ebf870dda64920e6891cb920d9dd6e"}, "score_mean": 0.32142857142857145, "score_ci_low": 0.29405171616264975, "score_ci_high": 0.35009026144901106, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 35964.10244360902, "output_tokens_ci_low": 31594.266917293233, "output_tokens_ci_high": 40629.36372180451, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 48.36466165413534, "steps_ci_low": 43.277255639097746, "steps_ci_high": 53.79793233082707, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-tthink4k-pure/iter-79/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "tthink4k", "method": "AOD on-policy tail (tthink4k)", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-tthink4k-pure", "checkpoint_iteration": 79, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-tthink4k-pure/iter_0000079", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_tthink4k_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_tthink4k_measured_pass8_cap120.tsv": "d8b7b43b14fe3d03c6e287d9525bdae3ff757c3f01d57b9cfb20f1e69bd010e2"}, "score_mean": 0.2640977443609023, "score_ci_low": 0.23849074927578395, "score_ci_high": 0.2914020737796206, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 41112.48402255639, "output_tokens_ci_low": 36693.625939849626, "output_tokens_ci_high": 45425.09868421053, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 67.94360902255639, "steps_ci_low": 61.68984962406015, "steps_ci_high": 74.52161654135338, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-tthink4k-pure/iter-90/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "tthink4k", "method": "AOD on-policy tail (tthink4k)", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-tthink4k-pure", "checkpoint_iteration": 90, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-tthink4k-pure/iter_0000090", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_tthink4k_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_tthink4k_measured_pass8_cap120.tsv": "d8b7b43b14fe3d03c6e287d9525bdae3ff757c3f01d57b9cfb20f1e69bd010e2"}, "score_mean": 0.25093984962406013, "score_ci_low": 0.22581608220756388, "score_ci_high": 0.27785562340869974, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 42799.387218045114, "output_tokens_ci_low": 38771.9727443609, "output_tokens_ci_high": 46927.421992481206, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 70.96146616541354, "steps_ci_low": 64.86654135338345, "steps_ci_high": 77.04605263157895, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-tthink4k-pure/iter-99/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "tthink4k", "method": "AOD on-policy tail (tthink4k)", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-tthink4k-pure", "checkpoint_iteration": 99, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-tthink4k-pure/iter_0000099", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_tthink4k_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_tthink4k_measured_pass8_cap120.tsv": "d8b7b43b14fe3d03c6e287d9525bdae3ff757c3f01d57b9cfb20f1e69bd010e2"}, "score_mean": 0.2819548872180451, "score_ci_low": 0.2557400622822514, "score_ci_high": 0.30973856301198704, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 32725.44078947368, "output_tokens_ci_low": 29590.781954887218, "output_tokens_ci_high": 35992.89191729324, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 64.8984962406015, "steps_ci_low": 58.69078947368421, "steps_ci_high": 71.14191729323308, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-tthink4k-pure/iter-109/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "tthink4k", "method": "AOD on-policy tail (tthink4k)", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-tthink4k-pure", "checkpoint_iteration": 109, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-tthink4k-pure/iter_0000109", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_tthink4k_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_tthink4k_measured_pass8_cap120.tsv": "d8b7b43b14fe3d03c6e287d9525bdae3ff757c3f01d57b9cfb20f1e69bd010e2"}, "score_mean": 0.26127819548872183, "score_ci_low": 0.23577216078455746, "score_ci_high": 0.28850185139103124, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 34575.10338345865, "output_tokens_ci_low": 30962.967105263157, "output_tokens_ci_high": 38364.34304511278, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 60.6437969924812, "steps_ci_low": 54.45676691729323, "steps_ci_high": 67.04041353383458, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-tthink4k-pure/iter-119/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "tthink4k", "method": "AOD on-policy tail (tthink4k)", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-tthink4k-pure", "checkpoint_iteration": 119, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-tthink4k-pure/iter_0000119", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_tthink4k_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_tthink4k_measured_pass8_cap120.tsv": "d8b7b43b14fe3d03c6e287d9525bdae3ff757c3f01d57b9cfb20f1e69bd010e2"}, "score_mean": 0.3082706766917293, "score_ci_low": 0.28125457986785213, "score_ci_high": 0.3366662803046678, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 34331.56954887218, "output_tokens_ci_low": 30488.760338345866, "output_tokens_ci_high": 38329.61842105263, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 54.6062030075188, "steps_ci_low": 49.15601503759399, "steps_ci_high": 60.1343984962406, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-tthink4k-pure/iter-129/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "tthink4k", "method": "AOD on-policy tail (tthink4k)", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-tthink4k-pure", "checkpoint_iteration": 129, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-tthink4k-pure/iter_0000129", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_tthink4k_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_tthink4k_measured_pass8_cap120.tsv": "d8b7b43b14fe3d03c6e287d9525bdae3ff757c3f01d57b9cfb20f1e69bd010e2"}, "score_mean": 0.29135338345864664, "score_ci_low": 0.2648398326776072, "score_ci_high": 0.31936816221601755, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 35871.299812030076, "output_tokens_ci_low": 32222.64285714286, "output_tokens_ci_high": 39743.14097744361, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 57.23966165413534, "steps_ci_low": 51.68984962406015, "steps_ci_high": 62.8796992481203, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-tthink4k-pure/iter-139/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "tthink4k", "method": "AOD on-policy tail (tthink4k)", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-tthink4k-pure", "checkpoint_iteration": 139, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-tthink4k-pure/iter_0000139", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_tthink4k_measured_pass8_cap120.tsv"], "figures": ["opd_hill_climb"], "source_sha256": {"notes/figures/series/series_tthink4k_measured_pass8_cap120.tsv": "d8b7b43b14fe3d03c6e287d9525bdae3ff757c3f01d57b9cfb20f1e69bd010e2"}, "score_mean": 0.25093984962406013, "score_ci_low": 0.22581608220756388, "score_ci_high": 0.27785562340869974, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 39265.561090225565, "output_tokens_ci_low": 35538.38815789474, "output_tokens_ci_high": 42957.453007518794, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 72.5093984962406, "steps_ci_low": 66.2265037593985, "steps_ci_high": 78.7875939849624, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-dr_grpo-full-2node/iter-9/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "drgrpo", "method": "Dr. GRPO", "model": "Qwen3.5-4B", "teacher": null, "training_run": "qwen3.5-4B-dr_grpo-full-2node", "checkpoint_iteration": 9, "checkpoint_folder": "checkpoints/qwen3.5-4B-dr_grpo-full-2node/iter_0000009", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_drgrpo_measured_pass8_cap120.tsv"], "figures": ["label_hill_climb"], "source_sha256": {"notes/figures/series/series_drgrpo_measured_pass8_cap120.tsv": "e5146b274b0c20541e934ce138941549b45c93aaa2d29c3e51c7d1726e5d3ba2"}, "score_mean": 0.17481203007518797, "score_ci_low": 0.15317130788522437, "score_ci_high": 0.19879250397601045, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 22294.208646616542, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 59.723684210526315, "steps_ci_low": null, "steps_ci_high": null, "steps_ci_level": null, "steps_ci_method": null, "steps_ci_status": "not_in_plot_source", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-dr_grpo-full-2node/iter-13/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "drgrpo", "method": "Dr. GRPO", "model": "Qwen3.5-4B", "teacher": null, "training_run": "qwen3.5-4B-dr_grpo-full-2node", "checkpoint_iteration": 13, "checkpoint_folder": "checkpoints/qwen3.5-4B-dr_grpo-full-2node/iter_0000013", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_drgrpo_measured_pass8_cap120.tsv"], "figures": ["label_hill_climb"], "source_sha256": {"notes/figures/series/series_drgrpo_measured_pass8_cap120.tsv": "e5146b274b0c20541e934ce138941549b45c93aaa2d29c3e51c7d1726e5d3ba2"}, "score_mean": 0.17669172932330826, "score_ci_low": 0.1549486663151132, "score_ci_high": 0.20076101946599886, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 19985.25657894737, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 60.72744360902256, "steps_ci_low": null, "steps_ci_high": null, "steps_ci_level": null, "steps_ci_method": null, "steps_ci_status": "not_in_plot_source", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-dr_grpo-full-2node/iter-19/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "drgrpo", "method": "Dr. GRPO", "model": "Qwen3.5-4B", "teacher": null, "training_run": "qwen3.5-4B-dr_grpo-full-2node", "checkpoint_iteration": 19, "checkpoint_folder": "checkpoints/qwen3.5-4B-dr_grpo-full-2node/iter_0000019", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_drgrpo_measured_pass8_cap120.tsv"], "figures": ["label_hill_climb"], "source_sha256": {"notes/figures/series/series_drgrpo_measured_pass8_cap120.tsv": "e5146b274b0c20541e934ce138941549b45c93aaa2d29c3e51c7d1726e5d3ba2"}, "score_mean": 0.17011278195488722, "score_ci_low": 0.1487321377543129, "score_ci_high": 0.19386698930722876, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 19586.96522556391, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 63.23966165413534, "steps_ci_low": null, "steps_ci_high": null, "steps_ci_level": null, "steps_ci_method": null, "steps_ci_status": "not_in_plot_source", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-dr_grpo-full-2node/iter-29/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "drgrpo", "method": "Dr. GRPO", "model": "Qwen3.5-4B", "teacher": null, "training_run": "qwen3.5-4B-dr_grpo-full-2node", "checkpoint_iteration": 29, "checkpoint_folder": "checkpoints/qwen3.5-4B-dr_grpo-full-2node/iter_0000029", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_drgrpo_measured_pass8_cap120.tsv"], "figures": ["label_hill_climb"], "source_sha256": {"notes/figures/series/series_drgrpo_measured_pass8_cap120.tsv": "e5146b274b0c20541e934ce138941549b45c93aaa2d29c3e51c7d1726e5d3ba2"}, "score_mean": 0.18703007518796994, "score_ci_low": 0.16474077417549524, "score_ci_high": 0.21157121816494176, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 14976.564849624061, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 59.38251879699248, "steps_ci_low": null, "steps_ci_high": null, "steps_ci_level": null, "steps_ci_method": null, "steps_ci_status": "not_in_plot_source", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-dr_grpo-full-2node/iter-39/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "drgrpo", "method": "Dr. GRPO", "model": "Qwen3.5-4B", "teacher": null, "training_run": "qwen3.5-4B-dr_grpo-full-2node", "checkpoint_iteration": 39, "checkpoint_folder": "checkpoints/qwen3.5-4B-dr_grpo-full-2node/iter_0000039", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_drgrpo_measured_pass8_cap120.tsv"], "figures": ["label_hill_climb"], "source_sha256": {"notes/figures/series/series_drgrpo_measured_pass8_cap120.tsv": "e5146b274b0c20541e934ce138941549b45c93aaa2d29c3e51c7d1726e5d3ba2"}, "score_mean": 0.26597744360902253, "score_ci_low": 0.2403039114734244, "score_ci_high": 0.29333478550185726, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 14017.83082706767, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 55.82048872180451, "steps_ci_low": null, "steps_ci_high": null, "steps_ci_level": null, "steps_ci_method": null, "steps_ci_status": "not_in_plot_source", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-dr_grpo-full-2node/iter-49/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "drgrpo", "method": "Dr. GRPO", "model": "Qwen3.5-4B", "teacher": null, "training_run": "qwen3.5-4B-dr_grpo-full-2node", "checkpoint_iteration": 49, "checkpoint_folder": "checkpoints/qwen3.5-4B-dr_grpo-full-2node/iter_0000049", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_drgrpo_measured_pass8_cap120.tsv"], "figures": ["label_hill_climb"], "source_sha256": {"notes/figures/series/series_drgrpo_measured_pass8_cap120.tsv": "e5146b274b0c20541e934ce138941549b45c93aaa2d29c3e51c7d1726e5d3ba2"}, "score_mean": 0.29229323308270677, "score_ci_low": 0.2657505892889406, "score_ci_high": 0.3203303425646228, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 14598.393796992481, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 42.57142857142857, "steps_ci_low": null, "steps_ci_high": null, "steps_ci_level": null, "steps_ci_method": null, "steps_ci_status": "not_in_plot_source", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-dr_grpo-full-2node/iter-59/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "drgrpo", "method": "Dr. GRPO", "model": "Qwen3.5-4B", "teacher": null, "training_run": "qwen3.5-4B-dr_grpo-full-2node", "checkpoint_iteration": 59, "checkpoint_folder": "checkpoints/qwen3.5-4B-dr_grpo-full-2node/iter_0000059", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_drgrpo_measured_pass8_cap120.tsv"], "figures": ["label_hill_climb"], "source_sha256": {"notes/figures/series/series_drgrpo_measured_pass8_cap120.tsv": "e5146b274b0c20541e934ce138941549b45c93aaa2d29c3e51c7d1726e5d3ba2"}, "score_mean": 0.35150375939849626, "score_ci_low": 0.3233964029642965, "score_ci_high": 0.3806795573654004, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 12042.508458646616, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 33.06672932330827, "steps_ci_low": null, "steps_ci_high": null, "steps_ci_level": null, "steps_ci_method": null, "steps_ci_status": "not_in_plot_source", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-dr_grpo-full-2node/iter-69/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "drgrpo", "method": "Dr. GRPO", "model": "Qwen3.5-4B", "teacher": null, "training_run": "qwen3.5-4B-dr_grpo-full-2node", "checkpoint_iteration": 69, "checkpoint_folder": "checkpoints/qwen3.5-4B-dr_grpo-full-2node/iter_0000069", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_drgrpo_measured_pass8_cap120.tsv"], "figures": ["label_hill_climb"], "source_sha256": {"notes/figures/series/series_drgrpo_measured_pass8_cap120.tsv": "e5146b274b0c20541e934ce138941549b45c93aaa2d29c3e51c7d1726e5d3ba2"}, "score_mean": 0.36466165413533835, "score_ci_low": 0.3362742041153868, "score_ci_high": 0.3940228736534509, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 9820.194548872181, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 26.877819548872182, "steps_ci_low": null, "steps_ci_high": null, "steps_ci_level": null, "steps_ci_method": null, "steps_ci_status": "not_in_plot_source", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-dr_grpo-full-2node/iter-79/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "drgrpo", "method": "Dr. GRPO", "model": "Qwen3.5-4B", "teacher": null, "training_run": "qwen3.5-4B-dr_grpo-full-2node", "checkpoint_iteration": 79, "checkpoint_folder": "checkpoints/qwen3.5-4B-dr_grpo-full-2node/iter_0000079", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_drgrpo_measured_pass8_cap120.tsv"], "figures": ["label_hill_climb"], "source_sha256": {"notes/figures/series/series_drgrpo_measured_pass8_cap120.tsv": "e5146b274b0c20541e934ce138941549b45c93aaa2d29c3e51c7d1726e5d3ba2"}, "score_mean": 0.35150375939849626, "score_ci_low": 0.3233964029642965, "score_ci_high": 0.3806795573654004, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 8013.6550751879695, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 25.274436090225564, "steps_ci_low": null, "steps_ci_high": null, "steps_ci_level": null, "steps_ci_method": null, "steps_ci_status": "not_in_plot_source", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-dr_grpo-full-2node/iter-89/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "drgrpo", "method": "Dr. GRPO", "model": "Qwen3.5-4B", "teacher": null, "training_run": "qwen3.5-4B-dr_grpo-full-2node", "checkpoint_iteration": 89, "checkpoint_folder": "checkpoints/qwen3.5-4B-dr_grpo-full-2node/iter_0000089", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_drgrpo_measured_pass8_cap120.tsv"], "figures": ["label_hill_climb"], "source_sha256": {"notes/figures/series/series_drgrpo_measured_pass8_cap120.tsv": "e5146b274b0c20541e934ce138941549b45c93aaa2d29c3e51c7d1726e5d3ba2"}, "score_mean": 0.3533834586466165, "score_ci_low": 0.32523464814097663, "score_ci_high": 0.3825871861085974, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 8754.586466165414, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 25.218045112781954, "steps_ci_low": null, "steps_ci_high": null, "steps_ci_level": null, "steps_ci_method": null, "steps_ci_status": "not_in_plot_source", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-dr_grpo-full-2node/iter-99/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "drgrpo", "method": "Dr. GRPO", "model": "Qwen3.5-4B", "teacher": null, "training_run": "qwen3.5-4B-dr_grpo-full-2node", "checkpoint_iteration": 99, "checkpoint_folder": "checkpoints/qwen3.5-4B-dr_grpo-full-2node/iter_0000099", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_drgrpo_measured_pass8_cap120.tsv"], "figures": ["label_hill_climb"], "source_sha256": {"notes/figures/series/series_drgrpo_measured_pass8_cap120.tsv": "e5146b274b0c20541e934ce138941549b45c93aaa2d29c3e51c7d1726e5d3ba2"}, "score_mean": 0.37406015037593987, "score_ci_low": 0.3454868629186761, "score_ci_high": 0.4035395844495479, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 10268.785714285714, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 26.633458646616543, "steps_ci_low": null, "steps_ci_high": null, "steps_ci_level": null, "steps_ci_method": null, "steps_ci_status": "not_in_plot_source", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-dr_grpo-full-2node/iter-109/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "drgrpo", "method": "Dr. GRPO", "model": "Qwen3.5-4B", "teacher": null, "training_run": "qwen3.5-4B-dr_grpo-full-2node", "checkpoint_iteration": 109, "checkpoint_folder": "checkpoints/qwen3.5-4B-dr_grpo-full-2node/iter_0000109", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_drgrpo_measured_pass8_cap120.tsv"], "figures": ["label_hill_climb"], "source_sha256": {"notes/figures/series/series_drgrpo_measured_pass8_cap120.tsv": "e5146b274b0c20541e934ce138941549b45c93aaa2d29c3e51c7d1726e5d3ba2"}, "score_mean": 0.4041353383458647, "score_ci_low": 0.37504483190984794, "score_ci_high": 0.4339155981764121, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 14404.40507518797, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 29.110902255639097, "steps_ci_low": null, "steps_ci_high": null, "steps_ci_level": null, "steps_ci_method": null, "steps_ci_status": "not_in_plot_source", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-dr_grpo-full-2node/iter-119/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "drgrpo", "method": "Dr. GRPO", "model": "Qwen3.5-4B", "teacher": null, "training_run": "qwen3.5-4B-dr_grpo-full-2node", "checkpoint_iteration": 119, "checkpoint_folder": "checkpoints/qwen3.5-4B-dr_grpo-full-2node/iter_0000119", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_drgrpo_measured_pass8_cap120.tsv"], "figures": ["label_hill_climb"], "source_sha256": {"notes/figures/series/series_drgrpo_measured_pass8_cap120.tsv": "e5146b274b0c20541e934ce138941549b45c93aaa2d29c3e51c7d1726e5d3ba2"}, "score_mean": 0.39097744360902253, "score_ci_low": 0.36209888376477495, "score_ci_high": 0.4206404288823442, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 15352.843984962406, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 30.841165413533833, "steps_ci_low": null, "steps_ci_high": null, "steps_ci_level": null, "steps_ci_method": null, "steps_ci_status": "not_in_plot_source", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-dr_grpo-full-2node/iter-129/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "drgrpo", "method": "Dr. GRPO", "model": "Qwen3.5-4B", "teacher": null, "training_run": "qwen3.5-4B-dr_grpo-full-2node", "checkpoint_iteration": 129, "checkpoint_folder": "checkpoints/qwen3.5-4B-dr_grpo-full-2node/iter_0000129", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_drgrpo_measured_pass8_cap120.tsv"], "figures": ["label_hill_climb"], "source_sha256": {"notes/figures/series/series_drgrpo_measured_pass8_cap120.tsv": "e5146b274b0c20541e934ce138941549b45c93aaa2d29c3e51c7d1726e5d3ba2"}, "score_mean": 0.39285714285714285, "score_ci_low": 0.3639469512829026, "score_ci_high": 0.4225382352840938, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 14659.841165413534, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 30.280075187969924, "steps_ci_low": null, "steps_ci_high": null, "steps_ci_level": null, "steps_ci_method": null, "steps_ci_status": "not_in_plot_source", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-dr_grpo-full-2node/iter-139/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "drgrpo", "method": "Dr. GRPO", "model": "Qwen3.5-4B", "teacher": null, "training_run": "qwen3.5-4B-dr_grpo-full-2node", "checkpoint_iteration": 139, "checkpoint_folder": "checkpoints/qwen3.5-4B-dr_grpo-full-2node/iter_0000139", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_drgrpo_measured_pass8_cap120.tsv"], "figures": ["label_hill_climb"], "source_sha256": {"notes/figures/series/series_drgrpo_measured_pass8_cap120.tsv": "e5146b274b0c20541e934ce138941549b45c93aaa2d29c3e51c7d1726e5d3ba2"}, "score_mean": 0.42387218045112784, "score_ci_low": 0.3945046967922136, "score_ci_high": 0.4537874094527576, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 13302.859022556391, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 28.154135338345863, "steps_ci_low": null, "steps_ci_high": null, "steps_ci_level": null, "steps_ci_method": null, "steps_ci_status": "not_in_plot_source", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-dr_grpo-full-2node-opdkl1.0-rbs16-ep20-aod2op-tail/iter-9/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "drgrpotail", "method": "Dr. GRPO + AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-dr_grpo-full-2node-opdkl1.0-rbs16-ep20-aod2op-tail", "checkpoint_iteration": 9, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-dr_grpo-full-2node-opdkl1.0-rbs16-ep20-aod2op-tail/iter_0000009", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_drgrpotail_measured_pass8_cap120.tsv"], "figures": ["label_hill_climb"], "source_sha256": {"notes/figures/series/series_drgrpotail_measured_pass8_cap120.tsv": "117c8a2e13464d47057d2145b3c40d565cc7770705147ba44c6c6274dc50d1db"}, "score_mean": 0.28101503759398494, "score_ci_low": 0.25483087400270843, "score_ci_high": 0.30877481433159126, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 36318.69830827068, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 29.73872180451128, "steps_ci_low": null, "steps_ci_high": null, "steps_ci_level": null, "steps_ci_method": null, "steps_ci_status": "not_in_plot_source", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-dr_grpo-full-2node-opdkl1.0-rbs16-ep20-aod2op-tail/iter-13/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "drgrpotail", "method": "Dr. GRPO + AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-dr_grpo-full-2node-opdkl1.0-rbs16-ep20-aod2op-tail", "checkpoint_iteration": 13, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-dr_grpo-full-2node-opdkl1.0-rbs16-ep20-aod2op-tail/iter_0000013", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_drgrpotail_measured_pass8_cap120.tsv"], "figures": ["label_hill_climb"], "source_sha256": {"notes/figures/series/series_drgrpotail_measured_pass8_cap120.tsv": "117c8a2e13464d47057d2145b3c40d565cc7770705147ba44c6c6274dc50d1db"}, "score_mean": 0.3167293233082707, "score_ci_low": 0.2894783535943107, "score_ci_high": 0.34529893921765703, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 16799.181390977443, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 33.64473684210526, "steps_ci_low": null, "steps_ci_high": null, "steps_ci_level": null, "steps_ci_method": null, "steps_ci_status": "not_in_plot_source", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-dr_grpo-full-2node-opdkl1.0-rbs16-ep20-aod2op-tail/iter-19/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "drgrpotail", "method": "Dr. GRPO + AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-dr_grpo-full-2node-opdkl1.0-rbs16-ep20-aod2op-tail", "checkpoint_iteration": 19, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-dr_grpo-full-2node-opdkl1.0-rbs16-ep20-aod2op-tail/iter_0000019", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_drgrpotail_measured_pass8_cap120.tsv"], "figures": ["label_hill_climb"], "source_sha256": {"notes/figures/series/series_drgrpotail_measured_pass8_cap120.tsv": "117c8a2e13464d47057d2145b3c40d565cc7770705147ba44c6c6274dc50d1db"}, "score_mean": 0.34022556390977443, "score_ci_low": 0.3123771768363375, "score_ci_high": 0.36922353997409585, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 11470.627819548872, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 27.94454887218045, "steps_ci_low": null, "steps_ci_high": null, "steps_ci_level": null, "steps_ci_method": null, "steps_ci_status": "not_in_plot_source", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-dr_grpo-full-2node-opdkl1.0-rbs16-ep20-aod2op-tail/iter-29/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "drgrpotail", "method": "Dr. GRPO + AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-dr_grpo-full-2node-opdkl1.0-rbs16-ep20-aod2op-tail", "checkpoint_iteration": 29, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-dr_grpo-full-2node-opdkl1.0-rbs16-ep20-aod2op-tail/iter_0000029", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_drgrpotail_measured_pass8_cap120.tsv"], "figures": ["label_hill_climb"], "source_sha256": {"notes/figures/series/series_drgrpotail_measured_pass8_cap120.tsv": "117c8a2e13464d47057d2145b3c40d565cc7770705147ba44c6c6274dc50d1db"}, "score_mean": 0.35620300751879697, "score_ci_low": 0.327992921094081, "score_ci_high": 0.38544772403530897, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 12855.426691729323, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 30.170112781954888, "steps_ci_low": null, "steps_ci_high": null, "steps_ci_level": null, "steps_ci_method": null, "steps_ci_status": "not_in_plot_source", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-dr_grpo-full-2node-opdkl1.0-rbs16-ep20-aod2op-tail/iter-39/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "drgrpotail", "method": "Dr. GRPO + AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-dr_grpo-full-2node-opdkl1.0-rbs16-ep20-aod2op-tail", "checkpoint_iteration": 39, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-dr_grpo-full-2node-opdkl1.0-rbs16-ep20-aod2op-tail/iter_0000039", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_drgrpotail_measured_pass8_cap120.tsv"], "figures": ["label_hill_climb"], "source_sha256": {"notes/figures/series/series_drgrpotail_measured_pass8_cap120.tsv": "117c8a2e13464d47057d2145b3c40d565cc7770705147ba44c6c6274dc50d1db"}, "score_mean": 0.38721804511278196, "score_ci_low": 0.35840411255487387, "score_ci_high": 0.4168434522524908, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 13339.53947368421, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 26.68515037593985, "steps_ci_low": null, "steps_ci_high": null, "steps_ci_level": null, "steps_ci_method": null, "steps_ci_status": "not_in_plot_source", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-dr_grpo-full-2node-opdkl1.0-rbs16-ep20-aod2op-tail/iter-49/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "drgrpotail", "method": "Dr. GRPO + AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-dr_grpo-full-2node-opdkl1.0-rbs16-ep20-aod2op-tail", "checkpoint_iteration": 49, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-dr_grpo-full-2node-opdkl1.0-rbs16-ep20-aod2op-tail/iter_0000049", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_drgrpotail_measured_pass8_cap120.tsv"], "figures": ["label_hill_climb"], "source_sha256": {"notes/figures/series/series_drgrpotail_measured_pass8_cap120.tsv": "117c8a2e13464d47057d2145b3c40d565cc7770705147ba44c6c6274dc50d1db"}, "score_mean": 0.3890977443609023, "score_ci_low": 0.36025127046177985, "score_ci_high": 0.4187421682654622, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 12444.693609022557, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 25.68703007518797, "steps_ci_low": null, "steps_ci_high": null, "steps_ci_level": null, "steps_ci_method": null, "steps_ci_status": "not_in_plot_source", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-dr_grpo-full-2node-opdkl1.0-rbs16-ep20-aod2op-tail/iter-59/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "drgrpotail", "method": "Dr. GRPO + AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-dr_grpo-full-2node-opdkl1.0-rbs16-ep20-aod2op-tail", "checkpoint_iteration": 59, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-dr_grpo-full-2node-opdkl1.0-rbs16-ep20-aod2op-tail/iter_0000059", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_drgrpotail_measured_pass8_cap120.tsv"], "figures": ["label_hill_climb"], "source_sha256": {"notes/figures/series/series_drgrpotail_measured_pass8_cap120.tsv": "117c8a2e13464d47057d2145b3c40d565cc7770705147ba44c6c6274dc50d1db"}, "score_mean": 0.38627819548872183, "score_ci_low": 0.3574807047509531, "score_ci_high": 0.41589392309647305, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 12992.072368421053, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 27.541353383458645, "steps_ci_low": null, "steps_ci_high": null, "steps_ci_level": null, "steps_ci_method": null, "steps_ci_status": "not_in_plot_source", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-dr_grpo-full-2node-opdkl1.0-rbs16-ep20-aod2op-tail/iter-69/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "drgrpotail", "method": "Dr. GRPO + AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-dr_grpo-full-2node-opdkl1.0-rbs16-ep20-aod2op-tail", "checkpoint_iteration": 69, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-dr_grpo-full-2node-opdkl1.0-rbs16-ep20-aod2op-tail/iter_0000069", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_drgrpotail_measured_pass8_cap120.tsv"], "figures": ["label_hill_climb"], "source_sha256": {"notes/figures/series/series_drgrpotail_measured_pass8_cap120.tsv": "117c8a2e13464d47057d2145b3c40d565cc7770705147ba44c6c6274dc50d1db"}, "score_mean": 0.3693609022556391, "score_ci_low": 0.3408790666228082, "score_ci_high": 0.3987826959457226, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 12479.562969924813, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 26.156954887218046, "steps_ci_low": null, "steps_ci_high": null, "steps_ci_level": null, "steps_ci_method": null, "steps_ci_status": "not_in_plot_source", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-dr_grpo-full-2node-opdkl1.0-rbs16-ep20-aod2op-tail/iter-79/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "drgrpotail", "method": "Dr. GRPO + AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-dr_grpo-full-2node-opdkl1.0-rbs16-ep20-aod2op-tail", "checkpoint_iteration": 79, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-dr_grpo-full-2node-opdkl1.0-rbs16-ep20-aod2op-tail/iter_0000079", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_drgrpotail_measured_pass8_cap120.tsv"], "figures": ["label_hill_climb"], "source_sha256": {"notes/figures/series/series_drgrpotail_measured_pass8_cap120.tsv": "117c8a2e13464d47057d2145b3c40d565cc7770705147ba44c6c6274dc50d1db"}, "score_mean": 0.3843984962406015, "score_ci_low": 0.3556342322082201, "score_ci_high": 0.4139945217193287, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 12730.704887218046, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 24.946428571428573, "steps_ci_low": null, "steps_ci_high": null, "steps_ci_level": null, "steps_ci_method": null, "steps_ci_status": "not_in_plot_source", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-dr_grpo-full-2node-opdkl1.0-rbs16-ep20-aod2op-tail/iter-89/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "drgrpotail", "method": "Dr. GRPO + AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-dr_grpo-full-2node-opdkl1.0-rbs16-ep20-aod2op-tail", "checkpoint_iteration": 89, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-dr_grpo-full-2node-opdkl1.0-rbs16-ep20-aod2op-tail/iter_0000089", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_drgrpotail_measured_pass8_cap120.tsv"], "figures": ["label_hill_climb"], "source_sha256": {"notes/figures/series/series_drgrpotail_measured_pass8_cap120.tsv": "117c8a2e13464d47057d2145b3c40d565cc7770705147ba44c6c6274dc50d1db"}, "score_mean": 0.3919172932330827, "score_ci_low": 0.36302286081977114, "score_ci_high": 0.42158938878728675, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 12479.944548872181, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 27.215225563909776, "steps_ci_low": null, "steps_ci_high": null, "steps_ci_level": null, "steps_ci_method": null, "steps_ci_status": "not_in_plot_source", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-dr_grpo-full-2node-opdkl1.0-rbs16-ep20-aod2op-tail/iter-99/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "drgrpotail", "method": "Dr. GRPO + AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-dr_grpo-full-2node-opdkl1.0-rbs16-ep20-aod2op-tail", "checkpoint_iteration": 99, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-dr_grpo-full-2node-opdkl1.0-rbs16-ep20-aod2op-tail/iter_0000099", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_drgrpotail_measured_pass8_cap120.tsv"], "figures": ["label_hill_climb"], "source_sha256": {"notes/figures/series/series_drgrpotail_measured_pass8_cap120.tsv": "117c8a2e13464d47057d2145b3c40d565cc7770705147ba44c6c6274dc50d1db"}, "score_mean": 0.3956766917293233, "score_ci_low": 0.36671990169195456, "score_ci_high": 0.4253840957548577, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 12784.425751879699, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 27.06390977443609, "steps_ci_low": null, "steps_ci_high": null, "steps_ci_level": null, "steps_ci_method": null, "steps_ci_status": "not_in_plot_source", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-dr_grpo-full-2node-opdkl1.0-rbs16-ep20-aod2op-tail/iter-109/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "drgrpotail", "method": "Dr. GRPO + AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-dr_grpo-full-2node-opdkl1.0-rbs16-ep20-aod2op-tail", "checkpoint_iteration": 109, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-dr_grpo-full-2node-opdkl1.0-rbs16-ep20-aod2op-tail/iter_0000109", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_drgrpotail_measured_pass8_cap120.tsv"], "figures": ["label_hill_climb"], "source_sha256": {"notes/figures/series/series_drgrpotail_measured_pass8_cap120.tsv": "117c8a2e13464d47057d2145b3c40d565cc7770705147ba44c6c6274dc50d1db"}, "score_mean": 0.4116541353383459, "score_ci_low": 0.38245236611607486, "score_ci_high": 0.44149155964969417, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 12887.086466165414, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 26.093984962406015, "steps_ci_low": null, "steps_ci_high": null, "steps_ci_level": null, "steps_ci_method": null, "steps_ci_status": "not_in_plot_source", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-dr_grpo-full-2node-opdkl1.0-rbs16-ep20-aod2op-tail/iter-119/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "drgrpotail", "method": "Dr. GRPO + AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-dr_grpo-full-2node-opdkl1.0-rbs16-ep20-aod2op-tail", "checkpoint_iteration": 119, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-dr_grpo-full-2node-opdkl1.0-rbs16-ep20-aod2op-tail/iter_0000119", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_drgrpotail_measured_pass8_cap120.tsv"], "figures": ["label_hill_climb"], "source_sha256": {"notes/figures/series/series_drgrpotail_measured_pass8_cap120.tsv": "117c8a2e13464d47057d2145b3c40d565cc7770705147ba44c6c6274dc50d1db"}, "score_mean": 0.3881578947368421, "score_ci_low": 0.3593276345090154, "score_ci_high": 0.4177928672582879, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 14262.131578947368, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 27.736842105263158, "steps_ci_low": null, "steps_ci_high": null, "steps_ci_level": null, "steps_ci_method": null, "steps_ci_status": "not_in_plot_source", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-dr_grpo-full-2node-opdkl1.0-rbs16-ep20-aod2op-tail/iter-129/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "drgrpotail", "method": "Dr. GRPO + AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-dr_grpo-full-2node-opdkl1.0-rbs16-ep20-aod2op-tail", "checkpoint_iteration": 129, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-dr_grpo-full-2node-opdkl1.0-rbs16-ep20-aod2op-tail/iter_0000129", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_drgrpotail_measured_pass8_cap120.tsv"], "figures": ["label_hill_climb"], "source_sha256": {"notes/figures/series/series_drgrpotail_measured_pass8_cap120.tsv": "117c8a2e13464d47057d2145b3c40d565cc7770705147ba44c6c6274dc50d1db"}, "score_mean": 0.40131578947368424, "score_ci_low": 0.37226884561694196, "score_ci_high": 0.43107277358950224, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 13929.738721804511, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 26.50093984962406, "steps_ci_low": null, "steps_ci_high": null, "steps_ci_level": null, "steps_ci_method": null, "steps_ci_status": "not_in_plot_source", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-dr_grpo-full-2node-opdkl1.0-rbs16-ep20-aod2op-tail/iter-139/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "drgrpotail", "method": "Dr. GRPO + AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-opd-dr_grpo-full-2node-opdkl1.0-rbs16-ep20-aod2op-tail", "checkpoint_iteration": 139, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-dr_grpo-full-2node-opdkl1.0-rbs16-ep20-aod2op-tail/iter_0000139", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_drgrpotail_measured_pass8_cap120.tsv"], "figures": ["label_hill_climb"], "source_sha256": {"notes/figures/series/series_drgrpotail_measured_pass8_cap120.tsv": "117c8a2e13464d47057d2145b3c40d565cc7770705147ba44c6c6274dc50d1db"}, "score_mean": 0.40601503759398494, "score_ci_low": 0.37689604794497256, "score_ci_high": 0.43581025606116464, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 12939.508458646616, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 26.169172932330827, "steps_ci_low": null, "steps_ci_high": null, "steps_ci_level": null, "steps_ci_method": null, "steps_ci_status": "not_in_plot_source", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-pure/iter-6/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q38pure", "method": "pure OPD", "model": "Qwen3.5-4B", "teacher": "Qwen3.8-27B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-pure", "checkpoint_iteration": 6, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-pure/iter_0000006", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q38pure_measured_pass8_cap120.tsv"], "figures": ["q38_hill_climb"], "source_sha256": {"notes/figures/series/series_q38pure_measured_pass8_cap120.tsv": "502ee00fe1a24bbc38d35c4e7c01772ef67e09ac0e7100f4e032dfede88940f5"}, "score_mean": 0.16165413533834586, "score_ci_low": 0.14075738489399342, "score_ci_high": 0.1849853095281006, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 78220.39003759398, "output_tokens_ci_low": 73038.80545112782, "output_tokens_ci_high": 83484.76879699249, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 68.03007518796993, "steps_ci_low": 63.19642857142857, "steps_ci_high": 72.8656015037594, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-pure/iter-13/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q38pure", "method": "pure OPD", "model": "Qwen3.5-4B", "teacher": "Qwen3.8-27B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-pure", "checkpoint_iteration": 13, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-pure/iter_0000013", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q38pure_measured_pass8_cap120.tsv"], "figures": ["q38_hill_climb"], "source_sha256": {"notes/figures/series/series_q38pure_measured_pass8_cap120.tsv": "502ee00fe1a24bbc38d35c4e7c01772ef67e09ac0e7100f4e032dfede88940f5"}, "score_mean": 0.2518796992481203, "score_ci_low": 0.2267203836778008, "score_ci_high": 0.2788242588984015, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 119020.2462406015, "output_tokens_ci_low": 109441.34868421052, "output_tokens_ci_high": 128802.36184210525, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 35.97932330827068, "steps_ci_low": 33.3796992481203, "steps_ci_high": 38.81484962406015, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-pure/iter-20/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q38pure", "method": "pure OPD", "model": "Qwen3.5-4B", "teacher": "Qwen3.8-27B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-pure", "checkpoint_iteration": 20, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-pure/iter_0000020", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q38pure_measured_pass8_cap120.tsv"], "figures": ["q38_hill_climb"], "source_sha256": {"notes/figures/series/series_q38pure_measured_pass8_cap120.tsv": "502ee00fe1a24bbc38d35c4e7c01772ef67e09ac0e7100f4e032dfede88940f5"}, "score_mean": 0.28383458646616544, "score_ci_low": 0.257558872485482, "score_ci_high": 0.3116656267286337, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 103875.03383458647, "output_tokens_ci_low": 92280.3947368421, "output_tokens_ci_high": 115866.71240601504, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 36.86842105263158, "steps_ci_low": 33.15507518796993, "steps_ci_high": 40.99248120300752, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-pure/iter-29/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q38pure", "method": "pure OPD", "model": "Qwen3.5-4B", "teacher": "Qwen3.8-27B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-pure", "checkpoint_iteration": 29, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-pure/iter_0000029", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q38pure_measured_pass8_cap120.tsv"], "figures": ["q38_hill_climb"], "source_sha256": {"notes/figures/series/series_q38pure_measured_pass8_cap120.tsv": "502ee00fe1a24bbc38d35c4e7c01772ef67e09ac0e7100f4e032dfede88940f5"}, "score_mean": 0.38721804511278196, "score_ci_low": 0.35840411255487387, "score_ci_high": 0.4168434522524908, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 83589.82894736843, "output_tokens_ci_low": 71493.0037593985, "output_tokens_ci_high": 96556.11466165414, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 35.442669172932334, "steps_ci_low": 31.047932330827066, "steps_ci_high": 40.359962406015036, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-pure/iter-39/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q38pure", "method": "pure OPD", "model": "Qwen3.5-4B", "teacher": "Qwen3.8-27B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-pure", "checkpoint_iteration": 39, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-pure/iter_0000039", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q38pure_measured_pass8_cap120.tsv"], "figures": ["q38_hill_climb"], "source_sha256": {"notes/figures/series/series_q38pure_measured_pass8_cap120.tsv": "502ee00fe1a24bbc38d35c4e7c01772ef67e09ac0e7100f4e032dfede88940f5"}, "score_mean": 0.35902255639097747, "score_ci_low": 0.3307522751352434, "score_ci_high": 0.38830718087396254, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 78039.91165413534, "output_tokens_ci_low": 65691.68139097745, "output_tokens_ci_high": 91400.71616541353, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 28.710526315789473, "steps_ci_low": 25.55169172932331, "steps_ci_high": 32.28101503759399, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-pure/iter-49/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q38pure", "method": "pure OPD", "model": "Qwen3.5-4B", "teacher": "Qwen3.8-27B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-pure", "checkpoint_iteration": 49, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-pure/iter_0000049", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q38pure_measured_pass8_cap120.tsv"], "figures": ["q38_hill_climb"], "source_sha256": {"notes/figures/series/series_q38pure_measured_pass8_cap120.tsv": "502ee00fe1a24bbc38d35c4e7c01772ef67e09ac0e7100f4e032dfede88940f5"}, "score_mean": 0.42951127819548873, "score_ci_low": 0.40007355728579014, "score_ci_high": 0.4594561707188128, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 82887.6992481203, "output_tokens_ci_low": 70067.93327067669, "output_tokens_ci_high": 96514.14755639098, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 29.972744360902254, "steps_ci_low": 26.51127819548872, "steps_ci_high": 33.891917293233085, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-pure/iter-59/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q38pure", "method": "pure OPD", "model": "Qwen3.5-4B", "teacher": "Qwen3.8-27B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-pure", "checkpoint_iteration": 59, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-pure/iter_0000059", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q38pure_measured_pass8_cap120.tsv"], "figures": ["q38_hill_climb"], "source_sha256": {"notes/figures/series/series_q38pure_measured_pass8_cap120.tsv": "502ee00fe1a24bbc38d35c4e7c01772ef67e09ac0e7100f4e032dfede88940f5"}, "score_mean": 0.424812030075188, "score_ci_low": 0.39543256738643984, "score_ci_high": 0.45473247581847, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 77660.49060150376, "output_tokens_ci_low": 64945.5620300752, "output_tokens_ci_high": 90886.43045112782, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 27.894736842105264, "steps_ci_low": 24.797932330827066, "steps_ci_high": 31.43515037593985, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-pure/iter-69/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q38pure", "method": "pure OPD", "model": "Qwen3.5-4B", "teacher": "Qwen3.8-27B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-pure", "checkpoint_iteration": 69, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-pure/iter_0000069", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q38pure_measured_pass8_cap120.tsv"], "figures": ["q38_hill_climb"], "source_sha256": {"notes/figures/series/series_q38pure_measured_pass8_cap120.tsv": "502ee00fe1a24bbc38d35c4e7c01772ef67e09ac0e7100f4e032dfede88940f5"}, "score_mean": 0.4398496240601504, "score_ci_low": 0.4102932945219264, "score_ci_high": 0.46983874004200143, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 75657.88345864661, "output_tokens_ci_low": 62241.8007518797, "output_tokens_ci_high": 89989.51503759398, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 27.87406015037594, "steps_ci_low": 24.55733082706767, "steps_ci_high": 31.576127819548873, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-pure/iter-79/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q38pure", "method": "pure OPD", "model": "Qwen3.5-4B", "teacher": "Qwen3.8-27B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-pure", "checkpoint_iteration": 79, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-pure/iter_0000079", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q38pure_measured_pass8_cap120.tsv"], "figures": ["q38_hill_climb"], "source_sha256": {"notes/figures/series/series_q38pure_measured_pass8_cap120.tsv": "502ee00fe1a24bbc38d35c4e7c01772ef67e09ac0e7100f4e032dfede88940f5"}, "score_mean": 0.46146616541353386, "score_ci_low": 0.43170392746335495, "score_ci_high": 0.4915056571791614, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 72559.13909774437, "output_tokens_ci_low": 59731.55733082707, "output_tokens_ci_high": 86048.98966165414, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 26.770676691729324, "steps_ci_low": 23.794172932330827, "steps_ci_high": 30.045112781954888, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-pure/iter-90/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q38pure", "method": "pure OPD", "model": "Qwen3.5-4B", "teacher": "Qwen3.8-27B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-pure", "checkpoint_iteration": 90, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-pure/iter_0000090", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q38pure_measured_pass8_cap120.tsv"], "figures": ["q38_hill_climb"], "source_sha256": {"notes/figures/series/series_q38pure_measured_pass8_cap120.tsv": "502ee00fe1a24bbc38d35c4e7c01772ef67e09ac0e7100f4e032dfede88940f5"}, "score_mean": 0.42387218045112784, "score_ci_low": 0.3945046967922136, "score_ci_high": 0.4537874094527576, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 61444.79041353383, "output_tokens_ci_low": 49809.49248120301, "output_tokens_ci_high": 73631.04511278195, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 24.097744360902254, "steps_ci_low": 21.38627819548872, "steps_ci_high": 27.012218045112782, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-pure/iter-99/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q38pure", "method": "pure OPD", "model": "Qwen3.5-4B", "teacher": "Qwen3.8-27B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-pure", "checkpoint_iteration": 99, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-pure/iter_0000099", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q38pure_measured_pass8_cap120.tsv"], "figures": ["q38_hill_climb"], "source_sha256": {"notes/figures/series/series_q38pure_measured_pass8_cap120.tsv": "502ee00fe1a24bbc38d35c4e7c01772ef67e09ac0e7100f4e032dfede88940f5"}, "score_mean": 0.43796992481203006, "score_ci_low": 0.40843418629888395, "score_ci_high": 0.46795197434516667, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 67698.43796992481, "output_tokens_ci_low": 55187.0977443609, "output_tokens_ci_high": 81342.43515037594, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 27.366541353383457, "steps_ci_low": 24.174812030075188, "steps_ci_high": 30.960526315789473, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-pure/iter-109/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q38pure", "method": "pure OPD", "model": "Qwen3.5-4B", "teacher": "Qwen3.8-27B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-pure", "checkpoint_iteration": 109, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-pure/iter_0000109", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q38pure_measured_pass8_cap120.tsv"], "figures": ["q38_hill_climb"], "source_sha256": {"notes/figures/series/series_q38pure_measured_pass8_cap120.tsv": "502ee00fe1a24bbc38d35c4e7c01772ef67e09ac0e7100f4e032dfede88940f5"}, "score_mean": 0.47274436090225563, "score_ci_low": 0.44289710659308595, "score_ci_high": 0.5027877215686939, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 58529.16635338346, "output_tokens_ci_low": 45924.01033834586, "output_tokens_ci_high": 71746.79793233082, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 23.217105263157894, "steps_ci_low": 20.382518796992482, "steps_ci_high": 26.351503759398497, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-pure/iter-119/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q38pure", "method": "pure OPD", "model": "Qwen3.5-4B", "teacher": "Qwen3.8-27B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-pure", "checkpoint_iteration": 119, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-pure/iter_0000119", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q38pure_measured_pass8_cap120.tsv"], "figures": ["q38_hill_climb"], "source_sha256": {"notes/figures/series/series_q38pure_measured_pass8_cap120.tsv": "502ee00fe1a24bbc38d35c4e7c01772ef67e09ac0e7100f4e032dfede88940f5"}, "score_mean": 0.41353383458646614, "score_ci_low": 0.38430535905685836, "score_ci_high": 0.4433844406287879, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 72622.5037593985, "output_tokens_ci_low": 59759.33928571428, "output_tokens_ci_high": 86121.50281954887, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 28.98308270676692, "steps_ci_low": 25.708646616541355, "steps_ci_high": 32.70864661654135, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-pure/iter-129/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q38pure", "method": "pure OPD", "model": "Qwen3.5-4B", "teacher": "Qwen3.8-27B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-pure", "checkpoint_iteration": 129, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-pure/iter_0000129", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q38pure_measured_pass8_cap120.tsv"], "figures": ["q38_hill_climb"], "source_sha256": {"notes/figures/series/series_q38pure_measured_pass8_cap120.tsv": "502ee00fe1a24bbc38d35c4e7c01772ef67e09ac0e7100f4e032dfede88940f5"}, "score_mean": 0.45300751879699247, "score_ci_low": 0.42331910640020554, "score_ci_high": 0.4830340456028631, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 65107.79229323309, "output_tokens_ci_low": 52141.424812030076, "output_tokens_ci_high": 78730.31203007519, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 23.553571428571427, "steps_ci_low": 21.036654135338345, "steps_ci_high": 26.582706766917294, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-pure/iter-139/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q38pure", "method": "pure OPD", "model": "Qwen3.5-4B", "teacher": "Qwen3.8-27B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-pure", "checkpoint_iteration": 139, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-pure/iter_0000139", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q38pure_measured_pass8_cap120.tsv"], "figures": ["q38_hill_climb"], "source_sha256": {"notes/figures/series/series_q38pure_measured_pass8_cap120.tsv": "502ee00fe1a24bbc38d35c4e7c01772ef67e09ac0e7100f4e032dfede88940f5"}, "score_mean": 0.45206766917293234, "score_ci_low": 0.42238799384141373, "score_ci_high": 0.4820922212017163, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 64262.44736842105, "output_tokens_ci_low": 51882.40131578947, "output_tokens_ci_high": 77027.64003759398, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 24.407894736842106, "steps_ci_low": 21.866541353383457, "steps_ci_high": 27.227443609022558, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-aod2op-tail-pure/iter-6/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q38tail", "method": "AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.8-27B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-aod2op-tail-pure", "checkpoint_iteration": 6, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-aod2op-tail-pure/iter_0000006", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q38tail_measured_pass8_cap120.tsv"], "figures": ["q38_hill_climb"], "source_sha256": {"notes/figures/series/series_q38tail_measured_pass8_cap120.tsv": "63634592904802f99b3efb6faf181b8e41065747eb6bd807d218515d538a2b5e"}, "score_mean": 0.16541353383458646, "score_ci_low": 0.14429916314148392, "score_ci_high": 0.1889352791203646, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 27166.359962406015, "output_tokens_ci_low": 25016.44172932331, "output_tokens_ci_high": 29303.059210526317, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 54.306390977443606, "steps_ci_low": 50.56109022556391, "steps_ci_high": 58.00375939849624, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-aod2op-tail-pure/iter-13/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q38tail", "method": "AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.8-27B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-aod2op-tail-pure", "checkpoint_iteration": 13, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-aod2op-tail-pure/iter_0000013", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q38tail_measured_pass8_cap120.tsv"], "figures": ["q38_hill_climb"], "source_sha256": {"notes/figures/series/series_q38tail_measured_pass8_cap120.tsv": "63634592904802f99b3efb6faf181b8e41065747eb6bd807d218515d538a2b5e"}, "score_mean": 0.22462406015037595, "score_ci_low": 0.2005636346718273, "score_ci_high": 0.2506658360661548, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 43723.85620300752, "output_tokens_ci_low": 41585.53947368421, "output_tokens_ci_high": 45831.411654135336, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 67.52067669172932, "steps_ci_low": 63.409774436090224, "steps_ci_high": 71.6015037593985, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-aod2op-tail-pure/iter-20/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q38tail", "method": "AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.8-27B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-aod2op-tail-pure", "checkpoint_iteration": 20, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-aod2op-tail-pure/iter_0000020", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q38tail_measured_pass8_cap120.tsv"], "figures": ["q38_hill_climb"], "source_sha256": {"notes/figures/series/series_q38tail_measured_pass8_cap120.tsv": "63634592904802f99b3efb6faf181b8e41065747eb6bd807d218515d538a2b5e"}, "score_mean": 0.39473684210526316, "score_ci_low": 0.36579547186035927, "score_ci_high": 0.4244355886265145, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 62614.30263157895, "output_tokens_ci_low": 55723.049812030076, "output_tokens_ci_high": 69628.50093984962, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 40.25751879699248, "steps_ci_low": 36.63627819548872, "steps_ci_high": 43.93327067669173, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-aod2op-tail-pure/iter-29/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q38tail", "method": "AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.8-27B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-aod2op-tail-pure", "checkpoint_iteration": 29, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-aod2op-tail-pure/iter_0000029", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q38tail_measured_pass8_cap120.tsv"], "figures": ["q38_hill_climb"], "source_sha256": {"notes/figures/series/series_q38tail_measured_pass8_cap120.tsv": "63634592904802f99b3efb6faf181b8e41065747eb6bd807d218515d538a2b5e"}, "score_mean": 0.38627819548872183, "score_ci_low": 0.3574807047509531, "score_ci_high": 0.41589392309647305, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 37092.31203007519, "output_tokens_ci_low": 31953.40037593985, "output_tokens_ci_high": 42539.968984962405, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 29.923872180451127, "steps_ci_low": 26.898496240601503, "steps_ci_high": 33.005639097744364, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-aod2op-tail-pure/iter-39/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q38tail", "method": "AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.8-27B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-aod2op-tail-pure", "checkpoint_iteration": 39, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-aod2op-tail-pure/iter_0000039", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q38tail_measured_pass8_cap120.tsv"], "figures": ["q38_hill_climb"], "source_sha256": {"notes/figures/series/series_q38tail_measured_pass8_cap120.tsv": "63634592904802f99b3efb6faf181b8e41065747eb6bd807d218515d538a2b5e"}, "score_mean": 0.4191729323308271, "score_ci_low": 0.38986698598012814, "score_ci_high": 0.44906043546514995, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 39917.515037593985, "output_tokens_ci_low": 34319.216165413534, "output_tokens_ci_high": 45809.27067669173, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 37.75, "steps_ci_low": 33.31109022556391, "steps_ci_high": 42.307330827067666, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-aod2op-tail-pure/iter-49/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q38tail", "method": "AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.8-27B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-aod2op-tail-pure", "checkpoint_iteration": 49, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-aod2op-tail-pure/iter_0000049", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q38tail_measured_pass8_cap120.tsv"], "figures": ["q38_hill_climb"], "source_sha256": {"notes/figures/series/series_q38tail_measured_pass8_cap120.tsv": "63634592904802f99b3efb6faf181b8e41065747eb6bd807d218515d538a2b5e"}, "score_mean": 0.3956766917293233, "score_ci_low": 0.36671990169195456, "score_ci_high": 0.4253840957548577, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 25974.59962406015, "output_tokens_ci_low": 22666.364661654134, "output_tokens_ci_high": 29358.51879699248, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 40.661654135338345, "steps_ci_low": 35.50187969924812, "steps_ci_high": 46.068609022556394, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-aod2op-tail-pure/iter-59/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q38tail", "method": "AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.8-27B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-aod2op-tail-pure", "checkpoint_iteration": 59, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-aod2op-tail-pure/iter_0000059", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q38tail_measured_pass8_cap120.tsv"], "figures": ["q38_hill_climb"], "source_sha256": {"notes/figures/series/series_q38tail_measured_pass8_cap120.tsv": "63634592904802f99b3efb6faf181b8e41065747eb6bd807d218515d538a2b5e"}, "score_mean": 0.39379699248120303, "score_ci_low": 0.3648711550104779, "score_ci_high": 0.42348696851645723, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 18222.908834586466, "output_tokens_ci_low": 16056.079887218048, "output_tokens_ci_high": 20502.13345864662, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 44.32142857142857, "steps_ci_low": 38.82236842105263, "steps_ci_high": 49.91635338345865, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-aod2op-tail-pure/iter-69/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q38tail", "method": "AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.8-27B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-aod2op-tail-pure", "checkpoint_iteration": 69, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-aod2op-tail-pure/iter_0000069", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q38tail_measured_pass8_cap120.tsv"], "figures": ["q38_hill_climb"], "source_sha256": {"notes/figures/series/series_q38tail_measured_pass8_cap120.tsv": "63634592904802f99b3efb6faf181b8e41065747eb6bd807d218515d538a2b5e"}, "score_mean": 0.4116541353383459, "score_ci_low": 0.38245236611607486, "score_ci_high": 0.44149155964969417, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 23631.59962406015, "output_tokens_ci_low": 20200.5507518797, "output_tokens_ci_high": 27340.04605263158, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 31.517857142857142, "steps_ci_low": 27.342105263157894, "steps_ci_high": 35.846804511278194, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-aod2op-tail-pure/iter-79/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q38tail", "method": "AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.8-27B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-aod2op-tail-pure", "checkpoint_iteration": 79, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-aod2op-tail-pure/iter_0000079", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q38tail_measured_pass8_cap120.tsv"], "figures": ["q38_hill_climb"], "source_sha256": {"notes/figures/series/series_q38tail_measured_pass8_cap120.tsv": "63634592904802f99b3efb6faf181b8e41065747eb6bd807d218515d538a2b5e"}, "score_mean": 0.46710526315789475, "score_ci_low": 0.43729860415428906, "score_ci_high": 0.497148602247859, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 26757.00939849624, "output_tokens_ci_low": 22318.24154135338, "output_tokens_ci_high": 31358.742481203008, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 30.05545112781955, "steps_ci_low": 26.007518796992482, "steps_ci_high": 34.30169172932331, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-aod2op-tail-pure/iter-90/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q38tail", "method": "AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.8-27B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-aod2op-tail-pure", "checkpoint_iteration": 90, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-aod2op-tail-pure/iter_0000090", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q38tail_measured_pass8_cap120.tsv"], "figures": ["q38_hill_climb"], "source_sha256": {"notes/figures/series/series_q38tail_measured_pass8_cap120.tsv": "63634592904802f99b3efb6faf181b8e41065747eb6bd807d218515d538a2b5e"}, "score_mean": 0.40789473684210525, "score_ci_low": 0.37874770963006305, "score_ci_high": 0.4377044682959514, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 21936.757518796992, "output_tokens_ci_low": 18335.870300751878, "output_tokens_ci_high": 25833.794172932332, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 28.647556390977442, "steps_ci_low": 25.06296992481203, "steps_ci_high": 32.44924812030075, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-aod2op-tail-pure/iter-99/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q38tail", "method": "AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.8-27B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-aod2op-tail-pure", "checkpoint_iteration": 99, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-aod2op-tail-pure/iter_0000099", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q38tail_measured_pass8_cap120.tsv"], "figures": ["q38_hill_climb"], "source_sha256": {"notes/figures/series/series_q38tail_measured_pass8_cap120.tsv": "63634592904802f99b3efb6faf181b8e41065747eb6bd807d218515d538a2b5e"}, "score_mean": 0.4144736842105263, "score_ci_low": 0.38523202130587186, "score_ci_high": 0.444330715339713, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 24121.01597744361, "output_tokens_ci_low": 20419.841165413534, "output_tokens_ci_high": 28101.78665413534, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 31.451127819548873, "steps_ci_low": 27.42669172932331, "steps_ci_high": 35.74906015037594, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-aod2op-tail-pure/iter-109/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q38tail", "method": "AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.8-27B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-aod2op-tail-pure", "checkpoint_iteration": 109, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-aod2op-tail-pure/iter_0000109", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q38tail_measured_pass8_cap120.tsv"], "figures": ["q38_hill_climb"], "source_sha256": {"notes/figures/series/series_q38tail_measured_pass8_cap120.tsv": "63634592904802f99b3efb6faf181b8e41065747eb6bd807d218515d538a2b5e"}, "score_mean": 0.4116541353383459, "score_ci_low": 0.38245236611607486, "score_ci_high": 0.44149155964969417, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 16028.461466165414, "output_tokens_ci_low": 13695.353383458647, "output_tokens_ci_high": 18387.821428571428, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 34.06390977443609, "steps_ci_low": 29.48402255639098, "steps_ci_high": 38.955827067669176, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-aod2op-tail-pure/iter-119/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q38tail", "method": "AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.8-27B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-aod2op-tail-pure", "checkpoint_iteration": 119, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-aod2op-tail-pure/iter_0000119", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q38tail_measured_pass8_cap120.tsv"], "figures": ["q38_hill_climb"], "source_sha256": {"notes/figures/series/series_q38tail_measured_pass8_cap120.tsv": "63634592904802f99b3efb6faf181b8e41065747eb6bd807d218515d538a2b5e"}, "score_mean": 0.40131578947368424, "score_ci_low": 0.37226884561694196, "score_ci_high": 0.43107277358950224, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 20208.727443609023, "output_tokens_ci_low": 17576.915413533836, "output_tokens_ci_high": 22886.140037593985, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 35.80263157894737, "steps_ci_low": 31.730263157894736, "steps_ci_high": 40.04793233082707, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-aod2op-tail-pure/iter-129/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q38tail", "method": "AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.8-27B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-aod2op-tail-pure", "checkpoint_iteration": 129, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-aod2op-tail-pure/iter_0000129", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q38tail_measured_pass8_cap120.tsv"], "figures": ["q38_hill_climb"], "source_sha256": {"notes/figures/series/series_q38tail_measured_pass8_cap120.tsv": "63634592904802f99b3efb6faf181b8e41065747eb6bd807d218515d538a2b5e"}, "score_mean": 0.39473684210526316, "score_ci_low": 0.36579547186035927, "score_ci_high": 0.4244355886265145, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 16889.443609022557, "output_tokens_ci_low": 14802.307330827067, "output_tokens_ci_high": 19020.91447368421, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 45.703947368421055, "steps_ci_low": 40.20864661654135, "steps_ci_high": 51.39661654135338, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-aod2op-tail-pure/iter-139/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q38tail", "method": "AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.8-27B", "training_run": "qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-aod2op-tail-pure", "checkpoint_iteration": 139, "checkpoint_folder": "checkpoints/qwen3.5-4B-opd-grpo-full-2node-opdkl1.0-rbs16-n1-ep20-tchq38_27b-aod2op-tail-pure/iter_0000139", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q38tail_measured_pass8_cap120.tsv"], "figures": ["q38_hill_climb"], "source_sha256": {"notes/figures/series/series_q38tail_measured_pass8_cap120.tsv": "63634592904802f99b3efb6faf181b8e41065747eb6bd807d218515d538a2b5e"}, "score_mean": 0.3768796992481203, "score_ci_low": 0.3482529394885122, "score_ci_high": 0.40639231875952764, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 14103.052631578947, "output_tokens_ci_low": 12493.029135338345, "output_tokens_ci_high": 15760.234022556391, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 49.045112781954884, "steps_ci_low": 43.53007518796993, "steps_ci_high": 54.682330827067666, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-pure/iter-6/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q9bpure", "method": "pure OPD", "model": "Qwen3.5-9B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-pure", "checkpoint_iteration": 6, "checkpoint_folder": "checkpoints/qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-pure/iter_0000006", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q9bpure_measured_pass8_cap120.tsv"], "figures": ["qwen9b_hill_climb"], "source_sha256": {"notes/figures/series/series_q9bpure_measured_pass8_cap120.tsv": "dcb16190f372437d8ff921aca400c4aea4054f5c65b174d52e846480e36bfd7a"}, "score_mean": 0.29135338345864664, "score_ci_low": 0.2648398326776072, "score_ci_high": 0.31936816221601755, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 42306.92857142857, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 66.20394736842105, "steps_ci_low": null, "steps_ci_high": null, "steps_ci_level": null, "steps_ci_method": null, "steps_ci_status": "not_in_plot_source", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-pure/iter-13/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q9bpure", "method": "pure OPD", "model": "Qwen3.5-9B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-pure", "checkpoint_iteration": 13, "checkpoint_folder": "checkpoints/qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-pure/iter_0000013", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q9bpure_measured_pass8_cap120.tsv"], "figures": ["qwen9b_hill_climb"], "source_sha256": {"notes/figures/series/series_q9bpure_measured_pass8_cap120.tsv": "dcb16190f372437d8ff921aca400c4aea4054f5c65b174d52e846480e36bfd7a"}, "score_mean": 0.34868421052631576, "score_ci_low": 0.32063994478107694, "score_ci_high": 0.377817204668804, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 39470.362781954886, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 56.54793233082707, "steps_ci_low": null, "steps_ci_high": null, "steps_ci_level": null, "steps_ci_method": null, "steps_ci_status": "not_in_plot_source", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-pure/iter-20/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q9bpure", "method": "pure OPD", "model": "Qwen3.5-9B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-pure", "checkpoint_iteration": 20, "checkpoint_folder": "checkpoints/qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-pure/iter_0000020", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q9bpure_measured_pass8_cap120.tsv"], "figures": ["qwen9b_hill_climb"], "source_sha256": {"notes/figures/series/series_q9bpure_measured_pass8_cap120.tsv": "dcb16190f372437d8ff921aca400c4aea4054f5c65b174d52e846480e36bfd7a"}, "score_mean": 0.3007518796992481, "score_ci_low": 0.27395365773467806, "score_ci_high": 0.3289837067583329, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 36606.16635338346, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 61.87687969924812, "steps_ci_low": null, "steps_ci_high": null, "steps_ci_level": null, "steps_ci_method": null, "steps_ci_status": "not_in_plot_source", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-pure/iter-29/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q9bpure", "method": "pure OPD", "model": "Qwen3.5-9B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-pure", "checkpoint_iteration": 29, "checkpoint_folder": "checkpoints/qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-pure/iter_0000029", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q9bpure_measured_pass8_cap120.tsv"], "figures": ["qwen9b_hill_climb"], "source_sha256": {"notes/figures/series/series_q9bpure_measured_pass8_cap120.tsv": "dcb16190f372437d8ff921aca400c4aea4054f5c65b174d52e846480e36bfd7a"}, "score_mean": 0.3449248120300752, "score_ci_low": 0.3169663745795445, "score_ci_high": 0.3739990270305819, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 29538.375, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 44.16729323308271, "steps_ci_low": null, "steps_ci_high": null, "steps_ci_level": null, "steps_ci_method": null, "steps_ci_status": "not_in_plot_source", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-pure/iter-39/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q9bpure", "method": "pure OPD", "model": "Qwen3.5-9B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-pure", "checkpoint_iteration": 39, "checkpoint_folder": "checkpoints/qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-pure/iter_0000039", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q9bpure_measured_pass8_cap120.tsv"], "figures": ["qwen9b_hill_climb"], "source_sha256": {"notes/figures/series/series_q9bpure_measured_pass8_cap120.tsv": "dcb16190f372437d8ff921aca400c4aea4054f5c65b174d52e846480e36bfd7a"}, "score_mean": 0.34774436090225563, "score_ci_low": 0.31972136883927876, "score_ci_high": 0.3768628436506635, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 31128.23120300752, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 48.70864661654135, "steps_ci_low": null, "steps_ci_high": null, "steps_ci_level": null, "steps_ci_method": null, "steps_ci_status": "not_in_plot_source", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-pure/iter-49/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q9bpure", "method": "pure OPD", "model": "Qwen3.5-9B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-pure", "checkpoint_iteration": 49, "checkpoint_folder": "checkpoints/qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-pure/iter_0000049", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q9bpure_measured_pass8_cap120.tsv"], "figures": ["qwen9b_hill_climb"], "source_sha256": {"notes/figures/series/series_q9bpure_measured_pass8_cap120.tsv": "dcb16190f372437d8ff921aca400c4aea4054f5c65b174d52e846480e36bfd7a"}, "score_mean": 0.37781954887218044, "score_ci_low": 0.3491751970934815, "score_ci_high": 0.40734299811449687, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 31346.144736842107, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 49.47932330827068, "steps_ci_low": null, "steps_ci_high": null, "steps_ci_level": null, "steps_ci_method": null, "steps_ci_status": "not_in_plot_source", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-pure/iter-59/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q9bpure", "method": "pure OPD", "model": "Qwen3.5-9B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-pure", "checkpoint_iteration": 59, "checkpoint_folder": "checkpoints/qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-pure/iter_0000059", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q9bpure_measured_pass8_cap120.tsv"], "figures": ["qwen9b_hill_climb"], "source_sha256": {"notes/figures/series/series_q9bpure_measured_pass8_cap120.tsv": "dcb16190f372437d8ff921aca400c4aea4054f5c65b174d52e846480e36bfd7a"}, "score_mean": 0.3430451127819549, "score_ci_low": 0.31513032538422237, "score_ci_high": 0.3720892023060268, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 34137.40507518797, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 52.87124060150376, "steps_ci_low": null, "steps_ci_high": null, "steps_ci_level": null, "steps_ci_method": null, "steps_ci_status": "not_in_plot_source", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-pure/iter-69/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q9bpure", "method": "pure OPD", "model": "Qwen3.5-9B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-pure", "checkpoint_iteration": 69, "checkpoint_folder": "checkpoints/qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-pure/iter_0000069", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q9bpure_measured_pass8_cap120.tsv"], "figures": ["qwen9b_hill_climb"], "source_sha256": {"notes/figures/series/series_q9bpure_measured_pass8_cap120.tsv": "dcb16190f372437d8ff921aca400c4aea4054f5c65b174d52e846480e36bfd7a"}, "score_mean": 0.3609022556390977, "score_ci_low": 0.3325924423130394, "score_ci_high": 0.39021288761604367, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 29547.07612781955, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 43.669172932330824, "steps_ci_low": null, "steps_ci_high": null, "steps_ci_level": null, "steps_ci_method": null, "steps_ci_status": "not_in_plot_source", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-pure/iter-79/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q9bpure", "method": "pure OPD", "model": "Qwen3.5-9B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-pure", "checkpoint_iteration": 79, "checkpoint_folder": "checkpoints/qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-pure/iter_0000079", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q9bpure_measured_pass8_cap120.tsv"], "figures": ["qwen9b_hill_climb"], "source_sha256": {"notes/figures/series/series_q9bpure_measured_pass8_cap120.tsv": "dcb16190f372437d8ff921aca400c4aea4054f5c65b174d52e846480e36bfd7a"}, "score_mean": 0.39097744360902253, "score_ci_low": 0.36209888376477495, "score_ci_high": 0.4206404288823442, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 30909.76597744361, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 43.65883458646616, "steps_ci_low": null, "steps_ci_high": null, "steps_ci_level": null, "steps_ci_method": null, "steps_ci_status": "not_in_plot_source", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-pure/iter-90/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q9bpure", "method": "pure OPD", "model": "Qwen3.5-9B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-pure", "checkpoint_iteration": 90, "checkpoint_folder": "checkpoints/qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-pure/iter_0000090", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q9bpure_measured_pass8_cap120.tsv"], "figures": ["qwen9b_hill_climb"], "source_sha256": {"notes/figures/series/series_q9bpure_measured_pass8_cap120.tsv": "dcb16190f372437d8ff921aca400c4aea4054f5c65b174d52e846480e36bfd7a"}, "score_mean": 0.3768796992481203, "score_ci_low": 0.3482529394885122, "score_ci_high": 0.40639231875952764, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 29468.31015037594, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 44.06484962406015, "steps_ci_low": null, "steps_ci_high": null, "steps_ci_level": null, "steps_ci_method": null, "steps_ci_status": "not_in_plot_source", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-pure/iter-99/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q9bpure", "method": "pure OPD", "model": "Qwen3.5-9B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-pure", "checkpoint_iteration": 99, "checkpoint_folder": "checkpoints/qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-pure/iter_0000099", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q9bpure_measured_pass8_cap120.tsv"], "figures": ["qwen9b_hill_climb"], "source_sha256": {"notes/figures/series/series_q9bpure_measured_pass8_cap120.tsv": "dcb16190f372437d8ff921aca400c4aea4054f5c65b174d52e846480e36bfd7a"}, "score_mean": 0.38627819548872183, "score_ci_low": 0.3574807047509531, "score_ci_high": 0.41589392309647305, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 30386.966165413534, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 48.12124060150376, "steps_ci_low": null, "steps_ci_high": null, "steps_ci_level": null, "steps_ci_method": null, "steps_ci_status": "not_in_plot_source", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-pure/iter-109/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q9bpure", "method": "pure OPD", "model": "Qwen3.5-9B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-pure", "checkpoint_iteration": 109, "checkpoint_folder": "checkpoints/qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-pure/iter_0000109", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q9bpure_measured_pass8_cap120.tsv"], "figures": ["qwen9b_hill_climb"], "source_sha256": {"notes/figures/series/series_q9bpure_measured_pass8_cap120.tsv": "dcb16190f372437d8ff921aca400c4aea4054f5c65b174d52e846480e36bfd7a"}, "score_mean": 0.3900375939849624, "score_ci_low": 0.36117502026316894, "score_ci_high": 0.4196913554240116, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 29722.60620300752, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 45.05357142857143, "steps_ci_low": null, "steps_ci_high": null, "steps_ci_level": null, "steps_ci_method": null, "steps_ci_status": "not_in_plot_source", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-pure/iter-119/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q9bpure", "method": "pure OPD", "model": "Qwen3.5-9B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-pure", "checkpoint_iteration": 119, "checkpoint_folder": "checkpoints/qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-pure/iter_0000119", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q9bpure_measured_pass8_cap120.tsv"], "figures": ["qwen9b_hill_climb"], "source_sha256": {"notes/figures/series/series_q9bpure_measured_pass8_cap120.tsv": "dcb16190f372437d8ff921aca400c4aea4054f5c65b174d52e846480e36bfd7a"}, "score_mean": 0.3900375939849624, "score_ci_low": 0.36117502026316894, "score_ci_high": 0.4196913554240116, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 28132.27819548872, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 41.31109022556391, "steps_ci_low": null, "steps_ci_high": null, "steps_ci_level": null, "steps_ci_method": null, "steps_ci_status": "not_in_plot_source", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-pure/iter-129/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q9bpure", "method": "pure OPD", "model": "Qwen3.5-9B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-pure", "checkpoint_iteration": 129, "checkpoint_folder": "checkpoints/qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-pure/iter_0000129", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q9bpure_measured_pass8_cap120.tsv"], "figures": ["qwen9b_hill_climb"], "source_sha256": {"notes/figures/series/series_q9bpure_measured_pass8_cap120.tsv": "dcb16190f372437d8ff921aca400c4aea4054f5c65b174d52e846480e36bfd7a"}, "score_mean": 0.37781954887218044, "score_ci_low": 0.3491751970934815, "score_ci_high": 0.40734299811449687, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 29687.648496240603, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 44.31578947368421, "steps_ci_low": null, "steps_ci_high": null, "steps_ci_level": null, "steps_ci_method": null, "steps_ci_status": "not_in_plot_source", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-pure/iter-139/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q9bpure", "method": "pure OPD", "model": "Qwen3.5-9B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-pure", "checkpoint_iteration": 139, "checkpoint_folder": "checkpoints/qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-pure/iter_0000139", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q9bpure_measured_pass8_cap120.tsv"], "figures": ["qwen9b_hill_climb"], "source_sha256": {"notes/figures/series/series_q9bpure_measured_pass8_cap120.tsv": "dcb16190f372437d8ff921aca400c4aea4054f5c65b174d52e846480e36bfd7a"}, "score_mean": 0.3890977443609023, "score_ci_low": 0.36025127046177985, "score_ci_high": 0.4187421682654622, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 30682.68233082707, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 45.94736842105263, "steps_ci_low": null, "steps_ci_high": null, "steps_ci_level": null, "steps_ci_method": null, "steps_ci_status": "not_in_plot_source", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter-6/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q9baod", "method": "AOD on-policy tail", "model": "Qwen3.5-9B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure", "checkpoint_iteration": 6, "checkpoint_folder": "checkpoints/qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter_0000006", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q9baod_measured_pass8_cap120.tsv"], "figures": ["qwen9b_hill_climb"], "source_sha256": {"notes/figures/series/series_q9baod_measured_pass8_cap120.tsv": "907e66ad8184e61dccc03b62e26e93aaea8d607a2cc957172a7e5760ec30e385"}, "score_mean": 0.22556390977443608, "score_ci_low": 0.2014631598431633, "score_ci_high": 0.25163924785475744, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 33893.39097744361, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 54.784774436090224, "steps_ci_low": null, "steps_ci_high": null, "steps_ci_level": null, "steps_ci_method": null, "steps_ci_status": "not_in_plot_source", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter-13/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q9baod", "method": "AOD on-policy tail", "model": "Qwen3.5-9B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure", "checkpoint_iteration": 13, "checkpoint_folder": "checkpoints/qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter_0000013", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q9baod_measured_pass8_cap120.tsv"], "figures": ["qwen9b_hill_climb"], "source_sha256": {"notes/figures/series/series_q9baod_measured_pass8_cap120.tsv": "907e66ad8184e61dccc03b62e26e93aaea8d607a2cc957172a7e5760ec30e385"}, "score_mean": 0.2857142857142857, "score_ci_low": 0.2593782579984431, "score_ci_high": 0.3135921151355498, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 39167.29605263158, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 47.79229323308271, "steps_ci_low": null, "steps_ci_high": null, "steps_ci_level": null, "steps_ci_method": null, "steps_ci_status": "not_in_plot_source", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter-20/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q9baod", "method": "AOD on-policy tail", "model": "Qwen3.5-9B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure", "checkpoint_iteration": 20, "checkpoint_folder": "checkpoints/qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter_0000020", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q9baod_measured_pass8_cap120.tsv"], "figures": ["qwen9b_hill_climb"], "source_sha256": {"notes/figures/series/series_q9baod_measured_pass8_cap120.tsv": "907e66ad8184e61dccc03b62e26e93aaea8d607a2cc957172a7e5760ec30e385"}, "score_mean": 0.35150375939849626, "score_ci_low": 0.3233964029642965, "score_ci_high": 0.3806795573654004, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 25188.56766917293, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 36.46898496240601, "steps_ci_low": null, "steps_ci_high": null, "steps_ci_level": null, "steps_ci_method": null, "steps_ci_status": "not_in_plot_source", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter-29/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q9baod", "method": "AOD on-policy tail", "model": "Qwen3.5-9B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure", "checkpoint_iteration": 29, "checkpoint_folder": "checkpoints/qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter_0000029", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q9baod_measured_pass8_cap120.tsv"], "figures": ["qwen9b_hill_climb"], "source_sha256": {"notes/figures/series/series_q9baod_measured_pass8_cap120.tsv": "907e66ad8184e61dccc03b62e26e93aaea8d607a2cc957172a7e5760ec30e385"}, "score_mean": 0.35902255639097747, "score_ci_low": 0.3307522751352434, "score_ci_high": 0.38830718087396254, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 21210.52537593985, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 27.922932330827066, "steps_ci_low": null, "steps_ci_high": null, "steps_ci_level": null, "steps_ci_method": null, "steps_ci_status": "not_in_plot_source", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter-39/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q9baod", "method": "AOD on-policy tail", "model": "Qwen3.5-9B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure", "checkpoint_iteration": 39, "checkpoint_folder": "checkpoints/qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter_0000039", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q9baod_measured_pass8_cap120.tsv"], "figures": ["qwen9b_hill_climb"], "source_sha256": {"notes/figures/series/series_q9baod_measured_pass8_cap120.tsv": "907e66ad8184e61dccc03b62e26e93aaea8d607a2cc957172a7e5760ec30e385"}, "score_mean": 0.37124060150375937, "score_ci_low": 0.3427218343742527, "score_ci_high": 0.40068580211415533, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 20999.03947368421, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 29.170112781954888, "steps_ci_low": null, "steps_ci_high": null, "steps_ci_level": null, "steps_ci_method": null, "steps_ci_status": "not_in_plot_source", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter-49/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q9baod", "method": "AOD on-policy tail", "model": "Qwen3.5-9B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure", "checkpoint_iteration": 49, "checkpoint_folder": "checkpoints/qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter_0000049", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q9baod_measured_pass8_cap120.tsv"], "figures": ["qwen9b_hill_climb"], "source_sha256": {"notes/figures/series/series_q9baod_measured_pass8_cap120.tsv": "907e66ad8184e61dccc03b62e26e93aaea8d607a2cc957172a7e5760ec30e385"}, "score_mean": 0.37969924812030076, "score_ci_low": 0.3510200594042683, "score_ci_high": 0.40924400972358743, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 18698.495300751878, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 28.64097744360902, "steps_ci_low": null, "steps_ci_high": null, "steps_ci_level": null, "steps_ci_method": null, "steps_ci_status": "not_in_plot_source", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter-59/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q9baod", "method": "AOD on-policy tail", "model": "Qwen3.5-9B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure", "checkpoint_iteration": 59, "checkpoint_folder": "checkpoints/qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter_0000059", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q9baod_measured_pass8_cap120.tsv"], "figures": ["qwen9b_hill_climb"], "source_sha256": {"notes/figures/series/series_q9baod_measured_pass8_cap120.tsv": "907e66ad8184e61dccc03b62e26e93aaea8d607a2cc957172a7e5760ec30e385"}, "score_mean": 0.3900375939849624, "score_ci_low": 0.36117502026316894, "score_ci_high": 0.4196913554240116, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 21148.619360902256, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 28.829887218045112, "steps_ci_low": null, "steps_ci_high": null, "steps_ci_level": null, "steps_ci_method": null, "steps_ci_status": "not_in_plot_source", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter-69/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q9baod", "method": "AOD on-policy tail", "model": "Qwen3.5-9B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure", "checkpoint_iteration": 69, "checkpoint_folder": "checkpoints/qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter_0000069", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q9baod_measured_pass8_cap120.tsv"], "figures": ["qwen9b_hill_climb"], "source_sha256": {"notes/figures/series/series_q9baod_measured_pass8_cap120.tsv": "907e66ad8184e61dccc03b62e26e93aaea8d607a2cc957172a7e5760ec30e385"}, "score_mean": 0.37218045112781956, "score_ci_low": 0.34364339381377873, "score_ci_high": 0.4016371796345679, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 21080.73120300752, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 29.49342105263158, "steps_ci_low": null, "steps_ci_high": null, "steps_ci_level": null, "steps_ci_method": null, "steps_ci_status": "not_in_plot_source", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter-79/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q9baod", "method": "AOD on-policy tail", "model": "Qwen3.5-9B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure", "checkpoint_iteration": 79, "checkpoint_folder": "checkpoints/qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter_0000079", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q9baod_measured_pass8_cap120.tsv"], "figures": ["qwen9b_hill_climb"], "source_sha256": {"notes/figures/series/series_q9baod_measured_pass8_cap120.tsv": "907e66ad8184e61dccc03b62e26e93aaea8d607a2cc957172a7e5760ec30e385"}, "score_mean": 0.3900375939849624, "score_ci_low": 0.36117502026316894, "score_ci_high": 0.4196913554240116, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 19099.867481203008, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 29.160714285714285, "steps_ci_low": null, "steps_ci_high": null, "steps_ci_level": null, "steps_ci_method": null, "steps_ci_status": "not_in_plot_source", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter-90/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q9baod", "method": "AOD on-policy tail", "model": "Qwen3.5-9B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure", "checkpoint_iteration": 90, "checkpoint_folder": "checkpoints/qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter_0000090", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q9baod_measured_pass8_cap120.tsv"], "figures": ["qwen9b_hill_climb"], "source_sha256": {"notes/figures/series/series_q9baod_measured_pass8_cap120.tsv": "907e66ad8184e61dccc03b62e26e93aaea8d607a2cc957172a7e5760ec30e385"}, "score_mean": 0.40225563909774437, "score_ci_low": 0.3731940625119931, "score_ci_high": 0.4320204936543896, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 22687.470864661653, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 29.87406015037594, "steps_ci_low": null, "steps_ci_high": null, "steps_ci_level": null, "steps_ci_method": null, "steps_ci_status": "not_in_plot_source", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter-99/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q9baod", "method": "AOD on-policy tail", "model": "Qwen3.5-9B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure", "checkpoint_iteration": 99, "checkpoint_folder": "checkpoints/qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter_0000099", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q9baod_measured_pass8_cap120.tsv"], "figures": ["qwen9b_hill_climb"], "source_sha256": {"notes/figures/series/series_q9baod_measured_pass8_cap120.tsv": "907e66ad8184e61dccc03b62e26e93aaea8d607a2cc957172a7e5760ec30e385"}, "score_mean": 0.38627819548872183, "score_ci_low": 0.3574807047509531, "score_ci_high": 0.41589392309647305, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 17894.323308270676, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 28.14191729323308, "steps_ci_low": null, "steps_ci_high": null, "steps_ci_level": null, "steps_ci_method": null, "steps_ci_status": "not_in_plot_source", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter-109/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q9baod", "method": "AOD on-policy tail", "model": "Qwen3.5-9B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure", "checkpoint_iteration": 109, "checkpoint_folder": "checkpoints/qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter_0000109", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q9baod_measured_pass8_cap120.tsv"], "figures": ["qwen9b_hill_climb"], "source_sha256": {"notes/figures/series/series_q9baod_measured_pass8_cap120.tsv": "907e66ad8184e61dccc03b62e26e93aaea8d607a2cc957172a7e5760ec30e385"}, "score_mean": 0.37875939849624063, "score_ci_low": 0.3500975704547792, "score_ci_high": 0.4082935617131379, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 17514.14379699248, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 28.295112781954888, "steps_ci_low": null, "steps_ci_high": null, "steps_ci_level": null, "steps_ci_method": null, "steps_ci_status": "not_in_plot_source", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter-119/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q9baod", "method": "AOD on-policy tail", "model": "Qwen3.5-9B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure", "checkpoint_iteration": 119, "checkpoint_folder": "checkpoints/qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter_0000119", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q9baod_measured_pass8_cap120.tsv"], "figures": ["qwen9b_hill_climb"], "source_sha256": {"notes/figures/series/series_q9baod_measured_pass8_cap120.tsv": "907e66ad8184e61dccc03b62e26e93aaea8d607a2cc957172a7e5760ec30e385"}, "score_mean": 0.3890977443609023, "score_ci_low": 0.36025127046177985, "score_ci_high": 0.4187421682654622, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 15355.082706766918, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 30.132518796992482, "steps_ci_low": null, "steps_ci_high": null, "steps_ci_level": null, "steps_ci_method": null, "steps_ci_status": "not_in_plot_source", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter-129/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q9baod", "method": "AOD on-policy tail", "model": "Qwen3.5-9B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure", "checkpoint_iteration": 129, "checkpoint_folder": "checkpoints/qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter_0000129", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q9baod_measured_pass8_cap120.tsv"], "figures": ["qwen9b_hill_climb"], "source_sha256": {"notes/figures/series/series_q9baod_measured_pass8_cap120.tsv": "907e66ad8184e61dccc03b62e26e93aaea8d607a2cc957172a7e5760ec30e385"}, "score_mean": 0.36278195488721804, "score_ci_low": 0.3344330858391748, "score_ci_high": 0.3921181180097855, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 14229.432330827067, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 31.409774436090224, "steps_ci_low": null, "steps_ci_high": null, "steps_ci_level": null, "steps_ci_method": null, "steps_ci_status": "not_in_plot_source", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter-139/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q9baod", "method": "AOD on-policy tail", "model": "Qwen3.5-9B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure", "checkpoint_iteration": 139, "checkpoint_folder": "checkpoints/qwen3.5-9B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter_0000139", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q9baod_measured_pass8_cap120.tsv"], "figures": ["qwen9b_hill_climb"], "source_sha256": {"notes/figures/series/series_q9baod_measured_pass8_cap120.tsv": "907e66ad8184e61dccc03b62e26e93aaea8d607a2cc957172a7e5760ec30e385"}, "score_mean": 0.3533834586466165, "score_ci_low": 0.32523464814097663, "score_ci_high": 0.3825871861085974, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 16417.977443609023, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 34.80921052631579, "steps_ci_low": null, "steps_ci_high": null, "steps_ci_level": null, "steps_ci_method": null, "steps_ci_status": "not_in_plot_source", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-27B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-pure/iter-49/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q27bpure", "method": "pure OPD", "model": "Qwen3.5-27B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-27B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-pure", "checkpoint_iteration": 49, "checkpoint_folder": "checkpoints/qwen3.5-27B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-pure/iter_0000049", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q27b_pure_measured_pass8_cap120.tsv"], "figures": ["qwen27b_hill_climb"], "source_sha256": {"notes/figures/series/series_q27b_pure_measured_pass8_cap120.tsv": "e06c115058cb1e33f9e2f4f121e71e2635746fdd496aaf74ba3f7f86c7bda1f6"}, "score_mean": 0.5046992481203008, "score_ci_low": 0.47469392513933, "score_ci_high": 0.534670759660363, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 24027.764097744363, "output_tokens_ci_low": 21199.916353383458, "output_tokens_ci_high": 27076.00845864662, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 41.05545112781955, "steps_ci_low": 36.390037593984964, "steps_ci_high": 46.03101503759399, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-27B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-pure/iter-69/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q27bpure", "method": "pure OPD", "model": "Qwen3.5-27B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-27B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-pure", "checkpoint_iteration": 69, "checkpoint_folder": "checkpoints/qwen3.5-27B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-pure/iter_0000069", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q27b_pure_measured_pass8_cap120.tsv"], "figures": ["qwen27b_hill_climb"], "source_sha256": {"notes/figures/series/series_q27b_pure_measured_pass8_cap120.tsv": "e06c115058cb1e33f9e2f4f121e71e2635746fdd496aaf74ba3f7f86c7bda1f6"}, "score_mean": 0.4718045112781955, "score_ci_low": 0.44196375752486383, "score_ci_high": 0.5018481336769773, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 23317.167293233084, "output_tokens_ci_low": 20500.217105263157, "output_tokens_ci_high": 26166.603383458645, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 40.818609022556394, "steps_ci_low": 36.18421052631579, "steps_ci_high": 46.06390977443609, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-27B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-pure/iter-79/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q27bpure", "method": "pure OPD", "model": "Qwen3.5-27B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-27B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-pure", "checkpoint_iteration": 79, "checkpoint_folder": "checkpoints/qwen3.5-27B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-pure/iter_0000079", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q27b_pure_measured_pass8_cap120.tsv"], "figures": ["qwen27b_hill_climb"], "source_sha256": {"notes/figures/series/series_q27b_pure_measured_pass8_cap120.tsv": "e06c115058cb1e33f9e2f4f121e71e2635746fdd496aaf74ba3f7f86c7bda1f6"}, "score_mean": 0.45112781954887216, "score_ci_low": 0.4214569883303125, "score_ci_high": 0.4811502897528788, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 26257.658834586466, "output_tokens_ci_low": 23121.90507518797, "output_tokens_ci_high": 29477.43233082707, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 43.55263157894737, "steps_ci_low": 38.54041353383459, "steps_ci_high": 48.63815789473684, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-27B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-pure/iter-90/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q27bpure", "method": "pure OPD", "model": "Qwen3.5-27B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-27B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-pure", "checkpoint_iteration": 90, "checkpoint_folder": "checkpoints/qwen3.5-27B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-pure/iter_0000090", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q27b_pure_measured_pass8_cap120.tsv"], "figures": ["qwen27b_hill_climb"], "source_sha256": {"notes/figures/series/series_q27b_pure_measured_pass8_cap120.tsv": "e06c115058cb1e33f9e2f4f121e71e2635746fdd496aaf74ba3f7f86c7bda1f6"}, "score_mean": 0.47368421052631576, "score_ci_low": 0.443830561712662, "score_ci_high": 0.5037272034090564, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 24241.098684210527, "output_tokens_ci_low": 21332.40319548872, "output_tokens_ci_high": 27275.34962406015, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 38.96616541353384, "steps_ci_low": 34.47086466165413, "steps_ci_high": 43.71804511278196, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-27B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-pure/iter-99/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q27bpure", "method": "pure OPD", "model": "Qwen3.5-27B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-27B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-pure", "checkpoint_iteration": 99, "checkpoint_folder": "checkpoints/qwen3.5-27B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-pure/iter_0000099", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q27b_pure_measured_pass8_cap120.tsv"], "figures": ["qwen27b_hill_climb"], "source_sha256": {"notes/figures/series/series_q27b_pure_measured_pass8_cap120.tsv": "e06c115058cb1e33f9e2f4f121e71e2635746fdd496aaf74ba3f7f86c7bda1f6"}, "score_mean": 0.4934210526315789, "score_ci_low": 0.46345757042959906, "score_ci_high": 0.5234318708508305, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 24450.464285714286, "output_tokens_ci_low": 21411.93984962406, "output_tokens_ci_high": 27708.08834586466, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 42.12124060150376, "steps_ci_low": 37.09492481203007, "steps_ci_high": 47.41353383458647, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-27B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-pure/iter-109/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q27bpure", "method": "pure OPD", "model": "Qwen3.5-27B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-27B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-pure", "checkpoint_iteration": 109, "checkpoint_folder": "checkpoints/qwen3.5-27B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-pure/iter_0000109", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q27b_pure_measured_pass8_cap120.tsv"], "figures": ["qwen27b_hill_climb"], "source_sha256": {"notes/figures/series/series_q27b_pure_measured_pass8_cap120.tsv": "e06c115058cb1e33f9e2f4f121e71e2635746fdd496aaf74ba3f7f86c7bda1f6"}, "score_mean": 0.45864661654135336, "score_ci_low": 0.4289080266121757, "score_ci_high": 0.4886827471505247, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 24589.380639097744, "output_tokens_ci_low": 21708.957706766916, "output_tokens_ci_high": 27423.227443609023, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 38.850563909774436, "steps_ci_low": 34.443609022556394, "steps_ci_high": 43.57048872180451, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-27B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-pure/iter-119/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q27bpure", "method": "pure OPD", "model": "Qwen3.5-27B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-27B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-pure", "checkpoint_iteration": 119, "checkpoint_folder": "checkpoints/qwen3.5-27B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-pure/iter_0000119", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q27b_pure_measured_pass8_cap120.tsv"], "figures": ["qwen27b_hill_climb"], "source_sha256": {"notes/figures/series/series_q27b_pure_measured_pass8_cap120.tsv": "e06c115058cb1e33f9e2f4f121e71e2635746fdd496aaf74ba3f7f86c7bda1f6"}, "score_mean": 0.4934210526315789, "score_ci_low": 0.46345757042959906, "score_ci_high": 0.5234318708508305, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 25807.78665413534, "output_tokens_ci_low": 22830.25845864662, "output_tokens_ci_high": 28936.958646616542, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 40.70958646616541, "steps_ci_low": 36.06296992481203, "steps_ci_high": 45.58552631578947, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-27B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-pure/iter-129/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q27bpure", "method": "pure OPD", "model": "Qwen3.5-27B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-27B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-pure", "checkpoint_iteration": 129, "checkpoint_folder": "checkpoints/qwen3.5-27B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-pure/iter_0000129", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q27b_pure_measured_pass8_cap120.tsv"], "figures": ["qwen27b_hill_climb"], "source_sha256": {"notes/figures/series/series_q27b_pure_measured_pass8_cap120.tsv": "e06c115058cb1e33f9e2f4f121e71e2635746fdd496aaf74ba3f7f86c7bda1f6"}, "score_mean": 0.48214285714285715, "score_ci_low": 0.45223642519320273, "score_ci_high": 0.5121777725679634, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 24088.50657894737, "output_tokens_ci_low": 21192.71334586466, "output_tokens_ci_high": 27176.019736842107, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 40.5328947368421, "steps_ci_low": 35.892857142857146, "steps_ci_high": 45.33176691729323, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-27B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-pure/iter-139/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q27bpure", "method": "pure OPD", "model": "Qwen3.5-27B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-27B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-pure", "checkpoint_iteration": 139, "checkpoint_folder": "checkpoints/qwen3.5-27B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-pure/iter_0000139", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q27b_pure_measured_pass8_cap120.tsv"], "figures": ["qwen27b_hill_climb"], "source_sha256": {"notes/figures/series/series_q27b_pure_measured_pass8_cap120.tsv": "e06c115058cb1e33f9e2f4f121e71e2635746fdd496aaf74ba3f7f86c7bda1f6"}, "score_mean": 0.46616541353383456, "score_ci_low": 0.4363658921303548, "score_ci_high": 0.4962083773118545, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 25255.707706766916, "output_tokens_ci_low": 22181.34962406015, "output_tokens_ci_high": 28403.936090225565, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 42.74906015037594, "steps_ci_low": 37.94078947368421, "steps_ci_high": 47.734962406015036, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-27B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter-6/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q27baod", "method": "AOD on-policy tail", "model": "Qwen3.5-27B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-27B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure", "checkpoint_iteration": 6, "checkpoint_folder": "checkpoints/qwen3.5-27B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter_0000006", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q27b_aod_measured_pass8_cap120.tsv"], "figures": ["qwen27b_hill_climb"], "source_sha256": {"notes/figures/series/series_q27b_aod_measured_pass8_cap120.tsv": "e19ebd514f695c7192c411f02583e4c633b8fc91a8d3408dd7d4c526dae3bbe2"}, "score_mean": 0.4830827067669173, "score_ci_low": 0.4531709392531354, "score_ci_high": 0.5131161954679694, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 17301.828007518798, "output_tokens_ci_low": 15153.29887218045, "output_tokens_ci_high": 19582.46334586466, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 34.6156015037594, "steps_ci_low": 30.655075187969924, "steps_ci_high": 38.81484962406015, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-27B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter-13/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q27baod", "method": "AOD on-policy tail", "model": "Qwen3.5-27B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-27B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure", "checkpoint_iteration": 13, "checkpoint_folder": "checkpoints/qwen3.5-27B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter_0000013", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q27b_aod_measured_pass8_cap120.tsv"], "figures": ["qwen27b_hill_climb"], "source_sha256": {"notes/figures/series/series_q27b_aod_measured_pass8_cap120.tsv": "e19ebd514f695c7192c411f02583e4c633b8fc91a8d3408dd7d4c526dae3bbe2"}, "score_mean": 0.47462406015037595, "score_ci_low": 0.44476412285158284, "score_ci_high": 0.5046665792300742, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 25935.502819548874, "output_tokens_ci_low": 22587.91447368421, "output_tokens_ci_high": 29518.82894736842, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 26.159774436090224, "steps_ci_low": 23.19266917293233, "steps_ci_high": 29.338345864661655, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-27B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter-20/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q27baod", "method": "AOD on-policy tail", "model": "Qwen3.5-27B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-27B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure", "checkpoint_iteration": 20, "checkpoint_folder": "checkpoints/qwen3.5-27B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter_0000020", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q27b_aod_measured_pass8_cap120.tsv"], "figures": ["qwen27b_hill_climb"], "source_sha256": {"notes/figures/series/series_q27b_aod_measured_pass8_cap120.tsv": "e19ebd514f695c7192c411f02583e4c633b8fc91a8d3408dd7d4c526dae3bbe2"}, "score_mean": 0.5385338345864662, "score_ci_low": 0.5084943428208387, "score_ci_high": 0.5682960725366452, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 23430.707706766916, "output_tokens_ci_low": 20537.255639097744, "output_tokens_ci_high": 26691.329887218046, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 26.1156015037594, "steps_ci_low": 23.68515037593985, "steps_ci_high": 28.632518796992482, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-27B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter-29/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q27baod", "method": "AOD on-policy tail", "model": "Qwen3.5-27B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-27B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure", "checkpoint_iteration": 29, "checkpoint_folder": "checkpoints/qwen3.5-27B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter_0000029", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q27b_aod_measured_pass8_cap120.tsv"], "figures": ["qwen27b_hill_climb"], "source_sha256": {"notes/figures/series/series_q27b_aod_measured_pass8_cap120.tsv": "e19ebd514f695c7192c411f02583e4c633b8fc91a8d3408dd7d4c526dae3bbe2"}, "score_mean": 0.5112781954887218, "score_ci_low": 0.48125548748033736, "score_ci_high": 0.541219756038926, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 13840.656015037594, "output_tokens_ci_low": 12048.248120300752, "output_tokens_ci_high": 15890.161654135338, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 22.291353383458645, "steps_ci_low": 20.397556390977442, "steps_ci_high": 24.25657894736842, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-27B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter-39/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q27baod", "method": "AOD on-policy tail", "model": "Qwen3.5-27B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-27B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure", "checkpoint_iteration": 39, "checkpoint_folder": "checkpoints/qwen3.5-27B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter_0000039", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q27b_aod_measured_pass8_cap120.tsv"], "figures": ["qwen27b_hill_climb"], "source_sha256": {"notes/figures/series/series_q27b_aod_measured_pass8_cap120.tsv": "e19ebd514f695c7192c411f02583e4c633b8fc91a8d3408dd7d4c526dae3bbe2"}, "score_mean": 0.5178571428571429, "score_ci_low": 0.4878222274320367, "score_ci_high": 0.5477635748067974, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 10807.075187969926, "output_tokens_ci_low": 9385.65507518797, "output_tokens_ci_high": 12357.563909774437, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 24.290413533834588, "steps_ci_low": 22.015037593984964, "steps_ci_high": 26.73402255639098, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-27B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter-49/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q27baod", "method": "AOD on-policy tail", "model": "Qwen3.5-27B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-27B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure", "checkpoint_iteration": 49, "checkpoint_folder": "checkpoints/qwen3.5-27B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter_0000049", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q27b_aod_measured_pass8_cap120.tsv"], "figures": ["qwen27b_hill_climb"], "source_sha256": {"notes/figures/series/series_q27b_aod_measured_pass8_cap120.tsv": "e19ebd514f695c7192c411f02583e4c633b8fc91a8d3408dd7d4c526dae3bbe2"}, "score_mean": 0.5028195488721805, "score_ci_low": 0.4728201435091982, "score_ci_high": 0.5327986673706179, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 10726.276315789473, "output_tokens_ci_low": 9245.700187969926, "output_tokens_ci_high": 12393.660714285714, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 23.945488721804512, "steps_ci_low": 21.770676691729324, "steps_ci_high": 26.237781954887218, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-27B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter-59/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q27baod", "method": "AOD on-policy tail", "model": "Qwen3.5-27B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-27B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure", "checkpoint_iteration": 59, "checkpoint_folder": "checkpoints/qwen3.5-27B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter_0000059", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q27b_aod_measured_pass8_cap120.tsv"], "figures": ["qwen27b_hill_climb"], "source_sha256": {"notes/figures/series/series_q27b_aod_measured_pass8_cap120.tsv": "e19ebd514f695c7192c411f02583e4c633b8fc91a8d3408dd7d4c526dae3bbe2"}, "score_mean": 0.5234962406015038, "score_ci_low": 0.4934549870635784, "score_ci_high": 0.5533684369348874, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 10538.73590225564, "output_tokens_ci_low": 9169.227443609023, "output_tokens_ci_high": 12056.986842105263, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 23.616541353383457, "steps_ci_low": 21.469924812030076, "steps_ci_high": 25.856203007518797, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-27B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter-69/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q27baod", "method": "AOD on-policy tail", "model": "Qwen3.5-27B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-27B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure", "checkpoint_iteration": 69, "checkpoint_folder": "checkpoints/qwen3.5-27B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter_0000069", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q27b_aod_measured_pass8_cap120.tsv"], "figures": ["qwen27b_hill_climb"], "source_sha256": {"notes/figures/series/series_q27b_aod_measured_pass8_cap120.tsv": "e19ebd514f695c7192c411f02583e4c633b8fc91a8d3408dd7d4c526dae3bbe2"}, "score_mean": 0.5272556390977443, "score_ci_low": 0.49721227843130605, "score_ci_high": 0.557102893406914, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 10949.011278195489, "output_tokens_ci_low": 9374.375939849624, "output_tokens_ci_high": 12684.855263157895, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 25.231203007518797, "steps_ci_low": 22.875, "steps_ci_high": 27.69830827067669, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-27B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter-79/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q27baod", "method": "AOD on-policy tail", "model": "Qwen3.5-27B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-27B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure", "checkpoint_iteration": 79, "checkpoint_folder": "checkpoints/qwen3.5-27B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter_0000079", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q27b_aod_measured_pass8_cap120.tsv"], "figures": ["qwen27b_hill_climb"], "source_sha256": {"notes/figures/series/series_q27b_aod_measured_pass8_cap120.tsv": "e19ebd514f695c7192c411f02583e4c633b8fc91a8d3408dd7d4c526dae3bbe2"}, "score_mean": 0.5018796992481203, "score_ci_low": 0.4718834110746723, "score_ci_high": 0.531862462845205, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 11175.94172932331, "output_tokens_ci_low": 9698.916353383458, "output_tokens_ci_high": 12855.801691729323, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 24.777255639097746, "steps_ci_low": 22.35808270676692, "steps_ci_high": 27.343045112781954, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-27B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter-90/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q27baod", "method": "AOD on-policy tail", "model": "Qwen3.5-27B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-27B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure", "checkpoint_iteration": 90, "checkpoint_folder": "checkpoints/qwen3.5-27B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter_0000090", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q27b_aod_measured_pass8_cap120.tsv"], "figures": ["qwen27b_hill_climb"], "source_sha256": {"notes/figures/series/series_q27b_aod_measured_pass8_cap120.tsv": "e19ebd514f695c7192c411f02583e4c633b8fc91a8d3408dd7d4c526dae3bbe2"}, "score_mean": 0.5112781954887218, "score_ci_low": 0.48125548748033736, "score_ci_high": 0.541219756038926, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 11114.109022556391, "output_tokens_ci_low": 9694.856203007519, "output_tokens_ci_high": 12697.591165413534, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 24.781015037593985, "steps_ci_low": 22.51691729323308, "steps_ci_high": 27.17575187969925, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-27B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter-99/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q27baod", "method": "AOD on-policy tail", "model": "Qwen3.5-27B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-27B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure", "checkpoint_iteration": 99, "checkpoint_folder": "checkpoints/qwen3.5-27B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter_0000099", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q27b_aod_measured_pass8_cap120.tsv"], "figures": ["qwen27b_hill_climb"], "source_sha256": {"notes/figures/series/series_q27b_aod_measured_pass8_cap120.tsv": "e19ebd514f695c7192c411f02583e4c633b8fc91a8d3408dd7d4c526dae3bbe2"}, "score_mean": 0.5169172932330827, "score_ci_low": 0.4868838045320306, "score_ci_high": 0.5468290607468645, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 11705.99060150376, "output_tokens_ci_low": 10212.934210526315, "output_tokens_ci_high": 13338.002819548872, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 26.67669172932331, "steps_ci_low": 24.1296992481203, "steps_ci_high": 29.2546992481203, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-27B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter-109/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q27baod", "method": "AOD on-policy tail", "model": "Qwen3.5-27B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-27B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure", "checkpoint_iteration": 109, "checkpoint_folder": "checkpoints/qwen3.5-27B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter_0000109", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q27b_aod_measured_pass8_cap120.tsv"], "figures": ["qwen27b_hill_climb"], "source_sha256": {"notes/figures/series/series_q27b_aod_measured_pass8_cap120.tsv": "e19ebd514f695c7192c411f02583e4c633b8fc91a8d3408dd7d4c526dae3bbe2"}, "score_mean": 0.4859022556390977, "score_ci_low": 0.4559751159254703, "score_ci_high": 0.5159308296754503, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 15158.78007518797, "output_tokens_ci_low": 13678.081766917294, "output_tokens_ci_high": 16776.771616541355, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 32.8656015037594, "steps_ci_low": 30.06954887218045, "steps_ci_high": 35.82518796992481, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-27B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter-119/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q27baod", "method": "AOD on-policy tail", "model": "Qwen3.5-27B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-27B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure", "checkpoint_iteration": 119, "checkpoint_folder": "checkpoints/qwen3.5-27B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter_0000119", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q27b_aod_measured_pass8_cap120.tsv"], "figures": ["qwen27b_hill_climb"], "source_sha256": {"notes/figures/series/series_q27b_aod_measured_pass8_cap120.tsv": "e19ebd514f695c7192c411f02583e4c633b8fc91a8d3408dd7d4c526dae3bbe2"}, "score_mean": 0.4107142857142857, "score_ci_low": 0.3815260356483884, "score_ci_high": 0.440544953157442, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 17144.656015037595, "output_tokens_ci_low": 15651.805451127819, "output_tokens_ci_high": 18722.18984962406, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 39.534774436090224, "steps_ci_low": 36.474624060150376, "steps_ci_high": 42.587406015037594, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-27B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter-129/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q27baod", "method": "AOD on-policy tail", "model": "Qwen3.5-27B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-27B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure", "checkpoint_iteration": 129, "checkpoint_folder": "checkpoints/qwen3.5-27B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter_0000129", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q27b_aod_measured_pass8_cap120.tsv"], "figures": ["qwen27b_hill_climb"], "source_sha256": {"notes/figures/series/series_q27b_aod_measured_pass8_cap120.tsv": "e19ebd514f695c7192c411f02583e4c633b8fc91a8d3408dd7d4c526dae3bbe2"}, "score_mean": 0.4483082706766917, "score_ci_low": 0.4186646146770528, "score_ci_high": 0.47832385252632265, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 20496.88627819549, "output_tokens_ci_low": 18685.898496240603, "output_tokens_ci_high": 22336.703007518798, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 37.3515037593985, "steps_ci_low": 34.30357142857143, "steps_ci_high": 40.57236842105263, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "officeqa/qwen3.5-27B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter-139/attempts8-cap120", "dataset": "officeqa", "eval_split": "held-out-133", "source_kind": "offline_saved_checkpoint", "family": "q27baod", "method": "AOD on-policy tail", "model": "Qwen3.5-27B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-27B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure", "checkpoint_iteration": 139, "checkpoint_folder": "checkpoints/qwen3.5-27B-opd-grpo-full-4node-opdkl1.0-rbs16-n1-ep20-aod2op-tail-pure/iter_0000139", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 133, "protocol_episode_count": 1064, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "pass_at_1", "source_files": ["notes/figures/series/series_q27b_aod_measured_pass8_cap120.tsv"], "figures": ["qwen27b_hill_climb"], "source_sha256": {"notes/figures/series/series_q27b_aod_measured_pass8_cap120.tsv": "e19ebd514f695c7192c411f02583e4c633b8fc91a8d3408dd7d4c526dae3bbe2"}, "score_mean": 0.4266917293233083, "score_ci_low": 0.3972886362379541, "score_ci_high": 0.456622280886833, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 25291.11654135338, "output_tokens_ci_low": 23099.329887218046, "output_tokens_ci_high": 27613.60902255639, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 33.65507518796993, "steps_ci_low": 30.835526315789473, "steps_ci_high": 36.588345864661655, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter-9/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "pure", "method": "pure OPD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-pure-rbs16-n1-ep5", "checkpoint_iteration": 9, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter_0000009", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.4203125, "score_ci_low": 0.4033153584, "score_ci_high": 0.4375007419, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 58307.17813, "output_tokens_ci_low": 54853.47864, "output_tokens_ci_high": 61760.87761, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 34.0946875, "steps_ci_low": 32.97589146, "steps_ci_high": 35.21348354, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter-19/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "pure", "method": "pure OPD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-pure-rbs16-n1-ep5", "checkpoint_iteration": 19, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter_0000019", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.4584375, "score_ci_low": 0.4412334579, "score_ci_high": 0.475741214, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 19980.50469, "output_tokens_ci_low": 18344.23964, "output_tokens_ci_high": 21616.76974, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 22.608125, "steps_ci_low": 21.41275144, "steps_ci_high": 23.80349856, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter-29/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "pure", "method": "pure OPD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-pure-rbs16-n1-ep5", "checkpoint_iteration": 29, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter_0000029", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.491875, "score_ci_low": 0.474573299, "score_ci_high": 0.5091961857, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 29279.13375, "output_tokens_ci_low": 26930.36059, "output_tokens_ci_high": 31627.90691, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 23.774375, "steps_ci_low": 22.5577946, "steps_ci_high": 24.9909554, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter-39/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "pure", "method": "pure OPD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-pure-rbs16-n1-ep5", "checkpoint_iteration": 39, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter_0000039", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.474375, "score_ci_low": 0.4571147247, "score_ci_high": 0.4916967272, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 24416.2325, "output_tokens_ci_low": 22347.88452, "output_tokens_ci_high": 26484.58048, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 22.1575, "steps_ci_low": 20.95310704, "steps_ci_high": 23.36189296, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter-49/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "pure", "method": "pure OPD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-pure-rbs16-n1-ep5", "checkpoint_iteration": 49, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter_0000049", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.454375, "score_ci_low": 0.4371881262, "score_ci_high": 0.471671288, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 24226.9675, "output_tokens_ci_low": 22208.52063, "output_tokens_ci_high": 26245.41437, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 21.870625, "steps_ci_low": 20.70210105, "steps_ci_high": 23.03914895, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter-59/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "pure", "method": "pure OPD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-pure-rbs16-n1-ep5", "checkpoint_iteration": 59, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter_0000059", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.543125, "score_ci_low": 0.5258240055, "score_ci_high": 0.5603225756, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 23530.04781, "output_tokens_ci_low": 21472.560869999998, "output_tokens_ci_high": 25587.53475, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 21.28, "steps_ci_low": 20.07161627, "steps_ci_high": 22.48838373, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter-69/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "pure", "method": "pure OPD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-pure-rbs16-n1-ep5", "checkpoint_iteration": 69, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter_0000069", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.5525, "score_ci_low": 0.535218914, "score_ci_high": 0.5696551847, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 23365.523119999998, "output_tokens_ci_low": 21241.41806, "output_tokens_ci_high": 25489.62819, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 20.38625, "steps_ci_low": 19.21138383, "steps_ci_high": 21.56111617, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter-79/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "pure", "method": "pure OPD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-pure-rbs16-n1-ep5", "checkpoint_iteration": 79, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter_0000079", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.549375, "score_ci_low": 0.5320865927, "score_ci_high": 0.566545, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 21954.39125, "output_tokens_ci_low": 19968.56576, "output_tokens_ci_high": 23940.21674, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 20.0125, "steps_ci_low": 18.87837602, "steps_ci_high": 21.14662398, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter-89/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "pure", "method": "pure OPD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-pure-rbs16-n1-ep5", "checkpoint_iteration": 89, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter_0000089", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.54625, "score_ci_low": 0.528954957, "score_ci_high": 0.5634341299, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 25174.74094, "output_tokens_ci_low": 22919.08133, "output_tokens_ci_high": 27430.400550000002, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 21.5571875, "steps_ci_low": 20.34213205, "steps_ci_high": 22.77224295, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter-99/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "pure", "method": "pure OPD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-pure-rbs16-n1-ep5", "checkpoint_iteration": 99, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter_0000099", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.5546875, "score_ci_low": 0.5374119475, "score_ci_high": 0.5718319053, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 20289.78938, "output_tokens_ci_low": 18429.42876, "output_tokens_ci_high": 22150.149989999998, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 20.174375, "steps_ci_low": 18.97865746, "steps_ci_high": 21.37009254, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter-109/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "pure", "method": "pure OPD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-pure-rbs16-n1-ep5", "checkpoint_iteration": 109, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter_0000109", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.5546875, "score_ci_low": 0.5374119475, "score_ci_high": 0.5718319053, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 21557.05312, "output_tokens_ci_low": 19639.29164, "output_tokens_ci_high": 23474.814609999998, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 20.3375, "steps_ci_low": 19.18777816, "steps_ci_high": 21.48722184, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter-119/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "pure", "method": "pure OPD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-pure-rbs16-n1-ep5", "checkpoint_iteration": 119, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter_0000119", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.5690625, "score_ci_low": 0.5518317187, "score_ci_high": 0.5861276611, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 24274.64406, "output_tokens_ci_low": 22158.05415, "output_tokens_ci_high": 26391.233969999997, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 20.1109375, "steps_ci_low": 18.98383006, "steps_ci_high": 21.23804494, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter-129/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "pure", "method": "pure OPD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-pure-rbs16-n1-ep5", "checkpoint_iteration": 129, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter_0000129", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.5614254455, "score_ci_low": 0.5441663519, "score_ci_high": 0.5785371874, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 20815.97589, "output_tokens_ci_low": 18959.799010000002, "output_tokens_ci_high": 22672.15278, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 19.96623945, "steps_ci_low": 18.82611012, "steps_ci_high": 21.11755059, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter-139/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "pure", "method": "pure OPD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-pure-rbs16-n1-ep5", "checkpoint_iteration": 139, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter_0000139", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.57375, "score_ci_low": 0.5565369908, "score_ci_high": 0.5907861477, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 22221.225619999997, "output_tokens_ci_low": 20256.576689999998, "output_tokens_ci_high": 24185.87456, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 20.3496875, "steps_ci_low": 19.20338741, "steps_ci_high": 21.49598759, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter-154/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "pure", "method": "pure OPD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-pure-rbs16-n1-ep5", "checkpoint_iteration": 154, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter_0000154", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.5553125, "score_ci_low": 0.5380385903, "score_ci_high": 0.5724537636, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 19503.318440000003, "output_tokens_ci_low": 17720.186869999998, "output_tokens_ci_high": 21286.449999999997, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 18.6671875, "steps_ci_low": 17.58858734, "steps_ci_high": 19.74578766, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2-pure-rbs16-n1-ep5/iter-9/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "aod2", "method": "AOD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2-pure-rbs16-n1-ep5", "checkpoint_iteration": 9, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2-pure-rbs16-n1-ep5/iter_0000009", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.46625, "score_ci_low": 0.4490161821, "score_ci_high": 0.4835647545, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 21682.91188, "output_tokens_ci_low": 20338.443949999997, "output_tokens_ci_high": 23027.3798, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 27.06, "steps_ci_low": 25.86210373, "steps_ci_high": 28.25789627, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2-pure-rbs16-n1-ep5/iter-19/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "aod2", "method": "AOD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2-pure-rbs16-n1-ep5", "checkpoint_iteration": 19, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2-pure-rbs16-n1-ep5/iter_0000019", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.4871875, "score_ci_low": 0.4698948148, "score_ci_high": 0.5045109111, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 18937.44, "output_tokens_ci_low": 17619.410519999998, "output_tokens_ci_high": 20255.46948, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 23.2803125, "steps_ci_low": 22.1345659, "steps_ci_high": 24.4260591, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2-pure-rbs16-n1-ep5/iter-29/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "aod2", "method": "AOD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2-pure-rbs16-n1-ep5", "checkpoint_iteration": 29, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2-pure-rbs16-n1-ep5/iter_0000029", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.5146875, "score_ci_low": 0.4973636247, "score_ci_high": 0.5319761529, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 17662.13844, "output_tokens_ci_low": 16354.11453, "output_tokens_ci_high": 18970.162340000003, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 20.3684375, "steps_ci_low": 19.32143455, "steps_ci_high": 21.41544045, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2-pure-rbs16-n1-ep5/iter-39/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "aod2", "method": "AOD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2-pure-rbs16-n1-ep5", "checkpoint_iteration": 39, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2-pure-rbs16-n1-ep5/iter_0000039", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.5215625, "score_ci_low": 0.5042390065, "score_ci_high": 0.538834284, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 18735.29156, "output_tokens_ci_low": 17304.93019, "output_tokens_ci_high": 20165.65293, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 21.9840625, "steps_ci_low": 20.83080755, "steps_ci_high": 23.13731745, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2-pure-rbs16-n1-ep5/iter-49/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "aod2", "method": "AOD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2-pure-rbs16-n1-ep5", "checkpoint_iteration": 49, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2-pure-rbs16-n1-ep5/iter_0000049", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.523125, "score_ci_low": 0.5058020503, "score_ci_high": 0.5403924932, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 20695.075630000003, "output_tokens_ci_low": 19241.51474, "output_tokens_ci_high": 22148.63651, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 23.5340625, "steps_ci_low": 22.37618159, "steps_ci_high": 24.69194341, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2-pure-rbs16-n1-ep5/iter-59/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "aod2", "method": "AOD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2-pure-rbs16-n1-ep5", "checkpoint_iteration": 59, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2-pure-rbs16-n1-ep5/iter_0000059", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.531875, "score_ci_low": 0.5145582288, "score_ci_high": 0.5491153311, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 19316.756250000002, "output_tokens_ci_low": 17879.277299999998, "output_tokens_ci_high": 20754.2352, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 22.7015625, "steps_ci_low": 21.55263175, "steps_ci_high": 23.85049325, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2-pure-rbs16-n1-ep5/iter-69/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "aod2", "method": "AOD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2-pure-rbs16-n1-ep5", "checkpoint_iteration": 69, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2-pure-rbs16-n1-ep5/iter_0000069", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.5334375, "score_ci_low": 0.5161223924, "score_ci_high": 0.5506724205, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 19430.40375, "output_tokens_ci_low": 17993.143969999997, "output_tokens_ci_high": 20867.66353, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 22.1921875, "steps_ci_low": 21.072371, "steps_ci_high": 23.312004, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2-pure-rbs16-n1-ep5/iter-79/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "aod2", "method": "AOD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2-pure-rbs16-n1-ep5", "checkpoint_iteration": 79, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2-pure-rbs16-n1-ep5/iter_0000079", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.5346875, "score_ci_low": 0.5173738456, "score_ci_high": 0.5519179696, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 18621.94, "output_tokens_ci_low": 17245.48999, "output_tokens_ci_high": 19998.390010000003, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 22.3853125, "steps_ci_low": 21.23176999, "steps_ci_high": 23.53885501, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2-pure-rbs16-n1-ep5/iter-89/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "aod2", "method": "AOD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2-pure-rbs16-n1-ep5", "checkpoint_iteration": 89, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2-pure-rbs16-n1-ep5/iter_0000089", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.515, "score_ci_low": 0.497676071, "score_ci_high": 0.5322879572, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 17423.530310000002, "output_tokens_ci_low": 16115.928549999999, "output_tokens_ci_high": 18731.13208, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 21.016875, "steps_ci_low": 19.91916966, "steps_ci_high": 22.11458034, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2-pure-rbs16-n1-ep5/iter-99/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "aod2", "method": "AOD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2-pure-rbs16-n1-ep5", "checkpoint_iteration": 99, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2-pure-rbs16-n1-ep5/iter_0000099", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.510625, "score_ci_low": 0.4933024382, "score_ci_high": 0.5279220818, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 18952.69531, "output_tokens_ci_low": 17541.72277, "output_tokens_ci_high": 20363.667849999998, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 21.560625, "steps_ci_low": 20.47613439, "steps_ci_high": 22.64511561, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2-pure-rbs16-n1-ep5/iter-109/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "aod2", "method": "AOD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2-pure-rbs16-n1-ep5", "checkpoint_iteration": 109, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2-pure-rbs16-n1-ep5/iter_0000109", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.513125, "score_ci_low": 0.4958014946, "score_ci_high": 0.53041703, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 17037.44437, "output_tokens_ci_low": 15732.88997, "output_tokens_ci_high": 18341.99878, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 19.6896875, "steps_ci_low": 18.64181957, "steps_ci_high": 20.73755543, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2-pure-rbs16-n1-ep5/iter-119/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "aod2", "method": "AOD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2-pure-rbs16-n1-ep5", "checkpoint_iteration": 119, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2-pure-rbs16-n1-ep5/iter_0000119", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.51875, "score_ci_low": 0.5014259543, "score_ci_high": 0.5360290809, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 18631.219380000002, "output_tokens_ci_low": 17256.61362, "output_tokens_ci_high": 20005.82513, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 21.8428125, "steps_ci_low": 20.71546841, "steps_ci_high": 22.97015659, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2-pure-rbs16-n1-ep5/iter-129/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "aod2", "method": "AOD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2-pure-rbs16-n1-ep5", "checkpoint_iteration": 129, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2-pure-rbs16-n1-ep5/iter_0000129", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.5125, "score_ci_low": 0.4951766899, "score_ci_high": 0.5297933335, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 17311.216559999997, "output_tokens_ci_low": 15976.706460000001, "output_tokens_ci_high": 18645.72666, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 20.621875, "steps_ci_low": 19.51891325, "steps_ci_high": 21.72483675, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2-pure-rbs16-n1-ep5/iter-139/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "aod2", "method": "AOD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2-pure-rbs16-n1-ep5", "checkpoint_iteration": 139, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2-pure-rbs16-n1-ep5/iter_0000139", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.525, "score_ci_low": 0.5076779265, "score_ci_high": 0.5422621205, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 15899.73094, "output_tokens_ci_low": 14617.26182, "output_tokens_ci_high": 17182.20005, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 19.246875, "steps_ci_low": 18.20852969, "steps_ci_high": 20.28522031, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2-pure-rbs16-n1-ep5/iter-154/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "aod2", "method": "AOD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2-pure-rbs16-n1-ep5", "checkpoint_iteration": 154, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2-pure-rbs16-n1-ep5/iter_0000154", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.520625, "score_ci_low": 0.5033012615, "score_ci_high": 0.5378992773, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 15731.21656, "output_tokens_ci_low": 14528.61753, "output_tokens_ci_high": 16933.81559, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 19.793125, "steps_ci_low": 18.77375478, "steps_ci_high": 20.81249522, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-pure-rbs16-n1-ep5/iter-9/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "aod2op", "method": "AOD on-policy", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-pure-rbs16-n1-ep5", "checkpoint_iteration": 9, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-pure-rbs16-n1-ep5/iter_0000009", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.4578125, "score_ci_low": 0.440611024, "score_ci_high": 0.4751151467, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 20442.16062, "output_tokens_ci_low": 19182.83406, "output_tokens_ci_high": 21701.487190000003, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 25.8909375, "steps_ci_low": 24.76447516, "steps_ci_high": 27.01739984, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-pure-rbs16-n1-ep5/iter-19/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "aod2op", "method": "AOD on-policy", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-pure-rbs16-n1-ep5", "checkpoint_iteration": 19, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-pure-rbs16-n1-ep5/iter_0000019", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.4990625, "score_ci_low": 0.4817499278, "score_ci_high": 0.5163773204, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 20102.13063, "output_tokens_ci_low": 18783.111999999997, "output_tokens_ci_high": 21421.14925, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 22.10375, "steps_ci_low": 21.02810747, "steps_ci_high": 23.17939253, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-pure-rbs16-n1-ep5/iter-29/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "aod2op", "method": "AOD on-policy", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-pure-rbs16-n1-ep5", "checkpoint_iteration": 29, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-pure-rbs16-n1-ep5/iter_0000029", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.5284375, "score_ci_low": 0.511117667, "score_ci_high": 0.5456891364, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 13688.543749999999, "output_tokens_ci_low": 12684.032630000002, "output_tokens_ci_high": 14693.05487, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 17.545, "steps_ci_low": 16.60913795, "steps_ci_high": 18.48086205, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-pure-rbs16-n1-ep5/iter-39/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "aod2op", "method": "AOD on-policy", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-pure-rbs16-n1-ep5", "checkpoint_iteration": 39, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-pure-rbs16-n1-ep5/iter_0000039", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.5459375, "score_ci_low": 0.528641831, "score_ci_high": 0.5631230053, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 12554.76094, "output_tokens_ci_low": 11626.839090000001, "output_tokens_ci_high": 13482.68279, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 16.2365625, "steps_ci_low": 15.37289469, "steps_ci_high": 17.10023031, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-pure-rbs16-n1-ep5/iter-49/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "aod2op", "method": "AOD on-policy", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-pure-rbs16-n1-ep5", "checkpoint_iteration": 49, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-pure-rbs16-n1-ep5/iter_0000049", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.546875, "score_ci_low": 0.5295812293, "score_ci_high": 0.5640563587, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 13273.00406, "output_tokens_ci_low": 12301.16364, "output_tokens_ci_high": 14244.84449, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 17.580625, "steps_ci_low": 16.63965195, "steps_ci_high": 18.52159805, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-pure-rbs16-n1-ep5/iter-59/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "aod2op", "method": "AOD on-policy", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-pure-rbs16-n1-ep5", "checkpoint_iteration": 59, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-pure-rbs16-n1-ep5/iter_0000059", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.5315625, "score_ci_low": 0.5142454165, "score_ci_high": 0.5488038928, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 15151.218439999999, "output_tokens_ci_low": 14073.9646, "output_tokens_ci_high": 16228.47227, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 19.199375, "steps_ci_low": 18.19581364, "steps_ci_high": 20.20293636, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-pure-rbs16-n1-ep5/iter-69/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "aod2op", "method": "AOD on-policy", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-pure-rbs16-n1-ep5", "checkpoint_iteration": 69, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-pure-rbs16-n1-ep5/iter_0000069", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.5209375, "score_ci_low": 0.5036138364, "score_ci_high": 0.538210953, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 15916.77813, "output_tokens_ci_low": 14783.053020000001, "output_tokens_ci_high": 17050.503230000002, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 17.3959375, "steps_ci_low": 16.52129831, "steps_ci_high": 18.27057669, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-pure-rbs16-n1-ep5/iter-79/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "aod2op", "method": "AOD on-policy", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-pure-rbs16-n1-ep5", "checkpoint_iteration": 79, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-pure-rbs16-n1-ep5/iter_0000079", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.49, "score_ci_low": 0.4727017229, "score_ci_high": 0.5073222584, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 12725.147500000001, "output_tokens_ci_low": 11776.49531, "output_tokens_ci_high": 13673.79969, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 15.3153125, "steps_ci_low": 14.52445051, "steps_ci_high": 16.10617449, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-pure-rbs16-n1-ep5/iter-89/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "aod2op", "method": "AOD on-policy", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-pure-rbs16-n1-ep5", "checkpoint_iteration": 89, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-pure-rbs16-n1-ep5/iter_0000089", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.469375, "score_ci_low": 0.4521304628, "score_ci_high": 0.4866929797, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 9985.877187, "output_tokens_ci_low": 9305.10062, "output_tokens_ci_high": 10666.653760000001, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 15.8984375, "steps_ci_low": 15.04588446, "steps_ci_high": 16.75099054, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-pure-rbs16-n1-ep5/iter-99/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "aod2op", "method": "AOD on-policy", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-pure-rbs16-n1-ep5", "checkpoint_iteration": 99, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-pure-rbs16-n1-ep5/iter_0000099", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.474375, "score_ci_low": 0.4571147247, "score_ci_high": 0.4916967272, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 10506.51, "output_tokens_ci_low": 9785.350451999999, "output_tokens_ci_high": 11227.66955, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 16.3053125, "steps_ci_low": 15.41848832, "steps_ci_high": 17.19213668, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-pure-rbs16-n1-ep5/iter-109/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "aod2op", "method": "AOD on-policy", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-pure-rbs16-n1-ep5", "checkpoint_iteration": 109, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-pure-rbs16-n1-ep5/iter_0000109", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.483125, "score_ci_low": 0.4658413591, "score_ci_high": 0.5004491092, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 12153.94437, "output_tokens_ci_low": 11310.977420000001, "output_tokens_ci_high": 12996.911329999999, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 14.618125, "steps_ci_low": 13.86843096, "steps_ci_high": 15.36781904, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-pure-rbs16-n1-ep5/iter-119/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "aod2op", "method": "AOD on-policy", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-pure-rbs16-n1-ep5", "checkpoint_iteration": 119, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-pure-rbs16-n1-ep5/iter_0000119", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.5, "score_ci_low": 0.4826862733, "score_ci_high": 0.5173137267, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 12336.22469, "output_tokens_ci_low": 11496.35414, "output_tokens_ci_high": 13176.09524, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 13.840625, "steps_ci_low": 13.1857351, "steps_ci_high": 14.4955149, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-pure-rbs16-n1-ep5/iter-129/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "aod2op", "method": "AOD on-policy", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-pure-rbs16-n1-ep5", "checkpoint_iteration": 129, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-pure-rbs16-n1-ep5/iter_0000129", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.475625, "score_ci_low": 0.4583610615, "score_ci_high": 0.4929473927, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 10986.85438, "output_tokens_ci_low": 10253.57646, "output_tokens_ci_high": 11720.13229, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 13.2765625, "steps_ci_low": 12.6403923, "steps_ci_high": 13.9127327, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-pure-rbs16-n1-ep5/iter-139/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "aod2op", "method": "AOD on-policy", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-pure-rbs16-n1-ep5", "checkpoint_iteration": 139, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-pure-rbs16-n1-ep5/iter_0000139", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.4946875, "score_ci_low": 0.4773811195, "score_ci_high": 0.5120066206, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 10700.18281, "output_tokens_ci_low": 9985.383046, "output_tokens_ci_high": 11414.98258, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 12.8475, "steps_ci_low": 12.27482261, "steps_ci_high": 13.42017739, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-pure-rbs16-n1-ep5/iter-154/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "aod2op", "method": "AOD on-policy", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-pure-rbs16-n1-ep5", "checkpoint_iteration": 154, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-pure-rbs16-n1-ep5/iter_0000154", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.5028125, "score_ci_low": 0.4854956745, "score_ci_high": 0.5201225808, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 12811.555, "output_tokens_ci_low": 11934.67296, "output_tokens_ci_high": 13688.43704, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 13.0146875, "steps_ci_low": 12.44442382, "steps_ci_high": 13.58495118, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5/iter-9/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "aod2optail", "method": "AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5", "checkpoint_iteration": 9, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5/iter_0000009", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci", "ssb1_label_hill_climb"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.494375, "score_ci_low": 0.4770691124, "score_ci_high": 0.5116943771, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 24098.65313, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 25.35, "steps_ci_low": 24.12173278, "steps_ci_high": 26.57826722, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5/iter-19/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "aod2optail", "method": "AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5", "checkpoint_iteration": 19, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5/iter_0000019", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci", "ssb1_label_hill_climb"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.509375, "score_ci_low": 0.4920530722, "score_ci_high": 0.5266744455, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 13642.82781, "output_tokens_ci_low": 12596.20923, "output_tokens_ci_high": 14689.446390000001, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 17.831875, "steps_ci_low": 16.86104856, "steps_ci_high": 18.80270144, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5/iter-29/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "aod2optail", "method": "AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5", "checkpoint_iteration": 29, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5/iter_0000029", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci", "ssb1_label_hill_climb"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.520625, "score_ci_low": 0.5033012615, "score_ci_high": 0.5378992773, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 11214.16688, "output_tokens_ci_low": 10383.50138, "output_tokens_ci_high": 12044.83237, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 17.788125, "steps_ci_low": 16.84148083, "steps_ci_high": 18.73476917, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5/iter-39/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "aod2optail", "method": "AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5", "checkpoint_iteration": 39, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5/iter_0000039", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci", "ssb1_label_hill_climb"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.55875, "score_ci_low": 0.5414856184, "score_ci_high": 0.575873492, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 11779.4475, "output_tokens_ci_low": 10865.349569999998, "output_tokens_ci_high": 12693.54543, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 15.5903125, "steps_ci_low": 14.74967787, "steps_ci_high": 16.43094713, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5/iter-49/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "aod2optail", "method": "AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5", "checkpoint_iteration": 49, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5/iter_0000049", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci", "ssb1_label_hill_climb"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.5771875, "score_ci_low": 0.5599885217, "score_ci_high": 0.5942013733, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 10369.51687, "output_tokens_ci_low": 9549.184822000001, "output_tokens_ci_high": 11189.84893, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 15.8815625, "steps_ci_low": 15.02511097, "steps_ci_high": 16.73801403, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5/iter-59/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "aod2optail", "method": "AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5", "checkpoint_iteration": 59, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5/iter_0000059", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci", "ssb1_label_hill_climb"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.568125, "score_ci_low": 0.5508908522, "score_ci_high": 0.5851957758, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 10770.72875, "output_tokens_ci_low": 9903.793192000001, "output_tokens_ci_high": 11637.66431, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 16.5221875, "steps_ci_low": 15.61069429, "steps_ci_high": 17.43368071, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5/iter-69/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "aod2optail", "method": "AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5", "checkpoint_iteration": 69, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5/iter_0000069", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci", "ssb1_label_hill_climb"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.573125, "score_ci_low": 0.5559095306, "score_ci_high": 0.5901651068, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 10102.38906, "output_tokens_ci_low": 9297.182896, "output_tokens_ci_high": 10907.59523, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 15.2596875, "steps_ci_low": 14.40654273, "steps_ci_high": 16.11283227, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5/iter-79/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "aod2optail", "method": "AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5", "checkpoint_iteration": 79, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5/iter_0000079", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci", "ssb1_label_hill_climb"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.5940625, "score_ci_low": 0.5769447476, "score_ci_high": 0.6109546791, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 9605.710000000001, "output_tokens_ci_low": 8810.928004000001, "output_tokens_ci_high": 10400.492, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 15.5040625, "steps_ci_low": 14.60929291, "steps_ci_high": 16.39883209, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5/iter-89/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "aod2optail", "method": "AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5", "checkpoint_iteration": 89, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5/iter_0000089", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci", "ssb1_label_hill_climb"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.5796875, "score_ci_low": 0.5624992581, "score_ci_high": 0.5966846416, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 9540.224063, "output_tokens_ci_low": 8771.734999999999, "output_tokens_ci_high": 10308.71313, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 15.765, "steps_ci_low": 14.86317577, "steps_ci_high": 16.66682423, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5/iter-99/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "aod2optail", "method": "AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5", "checkpoint_iteration": 99, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5/iter_0000099", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci", "ssb1_label_hill_climb"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.5978125, "score_ci_low": 0.5807156076, "score_ci_high": 0.6146748262, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 9041.899687000001, "output_tokens_ci_low": 8294.39414, "output_tokens_ci_high": 9789.405235, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 15.25125, "steps_ci_low": 14.39849163, "steps_ci_high": 16.10400837, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5/iter-109/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "aod2optail", "method": "AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5", "checkpoint_iteration": 109, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5/iter_0000109", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci", "ssb1_label_hill_climb"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.5603125, "score_ci_low": 0.5430527251, "score_ci_high": 0.5774276382, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 8684.965936999999, "output_tokens_ci_low": 7953.3549920000005, "output_tokens_ci_high": 9416.576883, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 15.2921875, "steps_ci_low": 14.43221575, "steps_ci_high": 16.15215925, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5/iter-119/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "aod2optail", "method": "AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5", "checkpoint_iteration": 119, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5/iter_0000119", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci", "ssb1_label_hill_climb"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.5809375, "score_ci_low": 0.5637547948, "score_ci_high": 0.5979261073, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 9197.949062, "output_tokens_ci_low": 8421.931738, "output_tokens_ci_high": 9973.966387, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 14.7740625, "steps_ci_low": 13.97940227, "steps_ci_high": 15.56872273, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5/iter-129/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "aod2optail", "method": "AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5", "checkpoint_iteration": 129, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5/iter_0000129", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci", "ssb1_label_hill_climb"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.6128125, "score_ci_low": 0.5958094097, "score_ci_high": 0.6295450522, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 9273.560937999999, "output_tokens_ci_low": 8513.723954, "output_tokens_ci_high": 10033.397920000001, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 16.283125, "steps_ci_low": 15.34574757, "steps_ci_high": 17.22050243, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5/iter-139/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "aod2optail", "method": "AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5", "checkpoint_iteration": 139, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5/iter_0000139", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci", "ssb1_label_hill_climb"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.604375, "score_ci_low": 0.5873170969, "score_ci_high": 0.6211825992, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 8475.15625, "output_tokens_ci_low": 7765.152766, "output_tokens_ci_high": 9185.159734, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 14.061875, "steps_ci_low": 13.2576534, "steps_ci_high": 14.8660966, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5/iter-154/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "aod2optail", "method": "AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5", "checkpoint_iteration": 154, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5/iter_0000154", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci", "ssb1_label_hill_climb"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.6028125, "score_ci_low": 0.5857450259, "score_ci_high": 0.6196334173, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 9143.093125, "output_tokens_ci_low": 8414.139223, "output_tokens_ci_high": 9872.047027, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 15.81125, "steps_ci_low": 14.95522979, "steps_ci_high": 16.66727021, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-pure-rbs16-n1-ep5/iter-9/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "thinkonly", "method": "think-only", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-pure-rbs16-n1-ep5", "checkpoint_iteration": 9, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-pure-rbs16-n1-ep5/iter_0000009", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.4459375, "score_ci_low": 0.4287899805, "score_ci_high": 0.4632146679, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 35282.54, "output_tokens_ci_low": 33259.99426, "output_tokens_ci_high": 37305.08574, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 31.3778125, "steps_ci_low": 30.21301914, "steps_ci_high": 32.54260586, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-pure-rbs16-n1-ep5/iter-19/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "thinkonly", "method": "think-only", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-pure-rbs16-n1-ep5", "checkpoint_iteration": 19, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-pure-rbs16-n1-ep5/iter_0000019", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.4671875, "score_ci_low": 0.4499503949, "score_ci_high": 0.4845032934, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 19189.08094, "output_tokens_ci_low": 17798.86628, "output_tokens_ci_high": 20579.29559, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 24.4028125, "steps_ci_low": 23.25680009, "steps_ci_high": 25.54882491, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-pure-rbs16-n1-ep5/iter-29/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "thinkonly", "method": "think-only", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-pure-rbs16-n1-ep5", "checkpoint_iteration": 29, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-pure-rbs16-n1-ep5/iter_0000029", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.5140625, "score_ci_low": 0.4967387524, "score_ci_high": 0.531352524, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 22055.27844, "output_tokens_ci_low": 20468.355750000002, "output_tokens_ci_high": 23642.201119999998, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 23.0859375, "steps_ci_low": 21.96788522, "steps_ci_high": 24.20398978, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-pure-rbs16-n1-ep5/iter-39/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "thinkonly", "method": "think-only", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-pure-rbs16-n1-ep5", "checkpoint_iteration": 39, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-pure-rbs16-n1-ep5/iter_0000039", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.5021875, "score_ci_low": 0.4848713159, "score_ci_high": 0.5194984382, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 18531.539689999998, "output_tokens_ci_low": 17170.3073, "output_tokens_ci_high": 19892.77207, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 21.59625, "steps_ci_low": 20.54240469, "steps_ci_high": 22.65009531, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-pure-rbs16-n1-ep5/iter-49/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "thinkonly", "method": "think-only", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-pure-rbs16-n1-ep5", "checkpoint_iteration": 49, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-pure-rbs16-n1-ep5/iter_0000049", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.534375, "score_ci_low": 0.5170609721, "score_ci_high": 0.5516065925, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 20544.76469, "output_tokens_ci_low": 19006.017969999997, "output_tokens_ci_high": 22083.5114, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 22.08125, "steps_ci_low": 20.99523125, "steps_ci_high": 23.16726875, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-pure-rbs16-n1-ep5/iter-59/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "thinkonly", "method": "think-only", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-pure-rbs16-n1-ep5", "checkpoint_iteration": 59, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-pure-rbs16-n1-ep5/iter_0000059", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.5446875, "score_ci_low": 0.5273893958, "score_ci_high": 0.5618784382, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 17110.33156, "output_tokens_ci_low": 15745.872309999999, "output_tokens_ci_high": 18474.79082, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 19.75125, "steps_ci_low": 18.7183546, "steps_ci_high": 20.7841454, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-pure-rbs16-n1-ep5/iter-69/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "thinkonly", "method": "think-only", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-pure-rbs16-n1-ep5", "checkpoint_iteration": 69, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-pure-rbs16-n1-ep5/iter_0000069", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.5528125, "score_ci_low": 0.5355321839, "score_ci_high": 0.5699661654, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 17673.020940000002, "output_tokens_ci_low": 16250.67465, "output_tokens_ci_high": 19095.36722, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 19.768125, "steps_ci_low": 18.75964846, "steps_ci_high": 20.77660154, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-pure-rbs16-n1-ep5/iter-79/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "thinkonly", "method": "think-only", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-pure-rbs16-n1-ep5", "checkpoint_iteration": 79, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-pure-rbs16-n1-ep5/iter_0000079", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.5325, "score_ci_low": 0.5151838739, "score_ci_high": 0.5497381872, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 18213.465, "output_tokens_ci_low": 16785.19879, "output_tokens_ci_high": 19641.731209999998, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 20.7496875, "steps_ci_low": 19.69809184, "steps_ci_high": 21.80128316, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-pure-rbs16-n1-ep5/iter-89/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "thinkonly", "method": "think-only", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-pure-rbs16-n1-ep5", "checkpoint_iteration": 89, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-pure-rbs16-n1-ep5/iter_0000089", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.546875, "score_ci_low": 0.5295812293, "score_ci_high": 0.5640563587, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 16410.647810000002, "output_tokens_ci_low": 15069.64761, "output_tokens_ci_high": 17751.64802, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 18.705, "steps_ci_low": 17.72282486, "steps_ci_high": 19.68717514, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-pure-rbs16-n1-ep5/iter-99/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "thinkonly", "method": "think-only", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-pure-rbs16-n1-ep5", "checkpoint_iteration": 99, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-pure-rbs16-n1-ep5/iter_0000099", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.555, "score_ci_low": 0.5377252655, "score_ci_high": 0.5721428379, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 16827.29188, "output_tokens_ci_low": 15403.64936, "output_tokens_ci_high": 18250.934390000002, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 19.51875, "steps_ci_low": 18.46614126, "steps_ci_high": 20.57135874, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-pure-rbs16-n1-ep5/iter-109/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "thinkonly", "method": "think-only", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-pure-rbs16-n1-ep5", "checkpoint_iteration": 109, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-pure-rbs16-n1-ep5/iter_0000109", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.5646875, "score_ci_low": 0.5474415429, "score_ci_high": 0.5817783286, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 16302.61156, "output_tokens_ci_low": 14929.75904, "output_tokens_ci_high": 17675.46408, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 18.8278125, "steps_ci_low": 17.82211211, "steps_ci_high": 19.83351289, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-pure-rbs16-n1-ep5/iter-119/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "thinkonly", "method": "think-only", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-pure-rbs16-n1-ep5", "checkpoint_iteration": 119, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-pure-rbs16-n1-ep5/iter_0000119", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.548125, "score_ci_low": 0.5308338562, "score_ci_high": 0.5653007342, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 17213.132810000003, "output_tokens_ci_low": 15749.487930000001, "output_tokens_ci_high": 18676.777690000003, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 19.1496875, "steps_ci_low": 18.12178176, "steps_ci_high": 20.17759324, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-pure-rbs16-n1-ep5/iter-129/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "thinkonly", "method": "think-only", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-pure-rbs16-n1-ep5", "checkpoint_iteration": 129, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-pure-rbs16-n1-ep5/iter_0000129", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.578125, "score_ci_low": 0.5609299953, "score_ci_high": 0.5951326515, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 16515.155310000002, "output_tokens_ci_low": 15094.594009999999, "output_tokens_ci_high": 17935.71661, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 19.6021875, "steps_ci_low": 18.52381327, "steps_ci_high": 20.68056173, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-pure-rbs16-n1-ep5/iter-139/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "thinkonly", "method": "think-only", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-pure-rbs16-n1-ep5", "checkpoint_iteration": 139, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-pure-rbs16-n1-ep5/iter_0000139", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.564375, "score_ci_low": 0.547128011, "score_ci_high": 0.58146761, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 14672.055629999999, "output_tokens_ci_low": 13401.17472, "output_tokens_ci_high": 15942.93653, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 17.5046875, "steps_ci_low": 16.53436596, "steps_ci_high": 18.47500904, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-pure-rbs16-n1-ep5/iter-154/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "thinkonly", "method": "think-only", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-pure-rbs16-n1-ep5", "checkpoint_iteration": 154, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-pure-rbs16-n1-ep5/iter_0000154", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.5790625, "score_ci_low": 0.561871532, "score_ci_high": 0.5960638666, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 16682.88562, "output_tokens_ci_low": 15285.17495, "output_tokens_ci_high": 18080.5963, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 18.56375, "steps_ci_low": 17.57211594, "steps_ci_high": 19.55538406, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-maskthink-pure-rbs16-n1-ep5/iter-9/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "maskthink", "method": "mask-think", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-maskthink-pure-rbs16-n1-ep5", "checkpoint_iteration": 9, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-maskthink-pure-rbs16-n1-ep5/iter_0000009", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.450625, "score_ci_low": 0.433455, "score_ci_high": 0.4679134073, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 21582.56156, "output_tokens_ci_low": 20146.29207, "output_tokens_ci_high": 23018.83106, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 23.5503125, "steps_ci_low": 22.4777934, "steps_ci_high": 24.6228316, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-maskthink-pure-rbs16-n1-ep5/iter-19/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "maskthink", "method": "mask-think", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-maskthink-pure-rbs16-n1-ep5", "checkpoint_iteration": 19, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-maskthink-pure-rbs16-n1-ep5/iter_0000019", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.4753125, "score_ci_low": 0.4580494671, "score_ci_high": 0.4926347365, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 23032.87156, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 22.4578125, "steps_ci_low": 21.34363038, "steps_ci_high": 23.57199462, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-maskthink-pure-rbs16-n1-ep5/iter-29/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "maskthink", "method": "mask-think", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-maskthink-pure-rbs16-n1-ep5", "checkpoint_iteration": 29, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-maskthink-pure-rbs16-n1-ep5/iter_0000029", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.4734375, "score_ci_low": 0.4561800432, "score_ci_high": 0.4907586568, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 19017.81625, "output_tokens_ci_low": 17482.70618, "output_tokens_ci_high": 20552.926320000002, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 19.4171875, "steps_ci_low": 18.38406829, "steps_ci_high": 20.45030671, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-maskthink-pure-rbs16-n1-ep5/iter-39/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "maskthink", "method": "mask-think", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-maskthink-pure-rbs16-n1-ep5", "checkpoint_iteration": 39, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-maskthink-pure-rbs16-n1-ep5/iter_0000039", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.48125, "score_ci_low": 0.4639709191, "score_ci_high": 0.4985740457, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 22837.59875, "output_tokens_ci_low": 21051.26315, "output_tokens_ci_high": 24623.93435, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 21.65875, "steps_ci_low": 20.55239942, "steps_ci_high": 22.76510058, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-maskthink-pure-rbs16-n1-ep5/iter-49/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "maskthink", "method": "mask-think", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-maskthink-pure-rbs16-n1-ep5", "checkpoint_iteration": 49, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-maskthink-pure-rbs16-n1-ep5/iter_0000049", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.4959375, "score_ci_low": 0.4786292153, "score_ci_high": 0.5132555271, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 19967.13906, "output_tokens_ci_low": 18302.79581, "output_tokens_ci_high": 21631.48231, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 19.5878125, "steps_ci_low": 18.54275547, "steps_ci_high": 20.63286953, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-maskthink-pure-rbs16-n1-ep5/iter-59/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "maskthink", "method": "mask-think", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-maskthink-pure-rbs16-n1-ep5", "checkpoint_iteration": 59, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-maskthink-pure-rbs16-n1-ep5/iter_0000059", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.5, "score_ci_low": 0.4826862733, "score_ci_high": 0.5173137267, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 19384.7475, "output_tokens_ci_low": 17731.98297, "output_tokens_ci_high": 21037.512029999998, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 18.844375, "steps_ci_low": 17.80935933, "steps_ci_high": 19.87939067, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-maskthink-pure-rbs16-n1-ep5/iter-69/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "maskthink", "method": "mask-think", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-maskthink-pure-rbs16-n1-ep5", "checkpoint_iteration": 69, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-maskthink-pure-rbs16-n1-ep5/iter_0000069", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.514375, "score_ci_low": 0.4970511851, "score_ci_high": 0.5316643419, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 21589.58469, "output_tokens_ci_low": 19838.12432, "output_tokens_ci_high": 23341.04506, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 20.725625, "steps_ci_low": 19.63747755, "steps_ci_high": 21.81377245, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-maskthink-pure-rbs16-n1-ep5/iter-79/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "maskthink", "method": "mask-think", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-maskthink-pure-rbs16-n1-ep5", "checkpoint_iteration": 79, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-maskthink-pure-rbs16-n1-ep5/iter_0000079", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.5165625, "score_ci_low": 0.499238404, "score_ci_high": 0.5338468772, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 19774.401560000002, "output_tokens_ci_low": 18070.07379, "output_tokens_ci_high": 21478.72934, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 18.6553125, "steps_ci_low": 17.65458927, "steps_ci_high": 19.65603573, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-maskthink-pure-rbs16-n1-ep5/iter-89/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "maskthink", "method": "mask-think", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-maskthink-pure-rbs16-n1-ep5", "checkpoint_iteration": 89, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-maskthink-pure-rbs16-n1-ep5/iter_0000089", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.5475, "score_ci_low": 0.5302075291, "score_ci_high": 0.5646785602, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 21770.66594, "output_tokens_ci_low": 19989.49941, "output_tokens_ci_high": 23551.83246, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 19.97875, "steps_ci_low": 18.9457107, "steps_ci_high": 21.0117893, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-maskthink-pure-rbs16-n1-ep5/iter-99/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "maskthink", "method": "mask-think", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-maskthink-pure-rbs16-n1-ep5", "checkpoint_iteration": 99, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-maskthink-pure-rbs16-n1-ep5/iter_0000099", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.5325, "score_ci_low": 0.5151838739, "score_ci_high": 0.5497381872, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 19057.8425, "output_tokens_ci_low": 17415.72397, "output_tokens_ci_high": 20699.961030000002, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 18.3259375, "steps_ci_low": 17.31801086, "steps_ci_high": 19.33386414, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-maskthink-pure-rbs16-n1-ep5/iter-109/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "maskthink", "method": "mask-think", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-maskthink-pure-rbs16-n1-ep5", "checkpoint_iteration": 109, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-maskthink-pure-rbs16-n1-ep5/iter_0000109", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.5321875, "score_ci_low": 0.5148710479, "score_ci_high": 0.5494267625, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 20135.41938, "output_tokens_ci_low": 18388.50894, "output_tokens_ci_high": 21882.329810000003, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 18.54875, "steps_ci_low": 17.54709935, "steps_ci_high": 19.55040065, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-maskthink-pure-rbs16-n1-ep5/iter-119/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "maskthink", "method": "mask-think", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-maskthink-pure-rbs16-n1-ep5", "checkpoint_iteration": 119, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-maskthink-pure-rbs16-n1-ep5/iter_0000119", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.5146875, "score_ci_low": 0.4973636247, "score_ci_high": 0.5319761529, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 17612.01313, "output_tokens_ci_low": 16054.62066, "output_tokens_ci_high": 19169.40559, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 17.0678125, "steps_ci_low": 16.145886, "steps_ci_high": 17.989739, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-maskthink-pure-rbs16-n1-ep5/iter-129/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "maskthink", "method": "mask-think", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-maskthink-pure-rbs16-n1-ep5", "checkpoint_iteration": 129, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-maskthink-pure-rbs16-n1-ep5/iter_0000129", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.5415625, "score_ci_low": 0.524258786, "score_ci_high": 0.5587665421, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 18917.670000000002, "output_tokens_ci_low": 17327.0782, "output_tokens_ci_high": 20508.2618, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 18.026875, "steps_ci_low": 17.10192726, "steps_ci_high": 18.95182274, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-maskthink-pure-rbs16-n1-ep5/iter-139/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "maskthink", "method": "mask-think", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-maskthink-pure-rbs16-n1-ep5", "checkpoint_iteration": 139, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-maskthink-pure-rbs16-n1-ep5/iter_0000139", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.5353125, "score_ci_low": 0.517999613, "score_ci_high": 0.5525407033, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 15341.244379999998, "output_tokens_ci_low": 13961.5716, "output_tokens_ci_high": 16720.917149999997, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 15.893125, "steps_ci_low": 15.0417253, "steps_ci_high": 16.7445247, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-maskthink-pure-rbs16-n1-ep5/iter-154/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "maskthink", "method": "mask-think", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-maskthink-pure-rbs16-n1-ep5", "checkpoint_iteration": 154, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-maskthink-pure-rbs16-n1-ep5/iter_0000154", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.5515625, "score_ci_low": 0.5342791456, "score_ci_high": 0.5687222013, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 19347.535, "output_tokens_ci_low": 17658.99728, "output_tokens_ci_high": 21036.07272, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 16.8046875, "steps_ci_low": 15.92720403, "steps_ci_high": 17.68217097, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-first280-pure-rbs16-n1-ep5/iter-9/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "first280", "method": "first 280 response tokens", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-first280-pure-rbs16-n1-ep5", "checkpoint_iteration": 9, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-first280-pure-rbs16-n1-ep5/iter_0000009", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.503125, "score_ci_low": 0.485807864, "score_ci_high": 0.5204346419, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 36814.744380000004, "output_tokens_ci_low": 34368.31779, "output_tokens_ci_high": 39261.17096, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 28.839375, "steps_ci_low": 27.7041712, "steps_ci_high": 29.9745788, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-first280-pure-rbs16-n1-ep5/iter-19/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "first280", "method": "first 280 response tokens", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-first280-pure-rbs16-n1-ep5", "checkpoint_iteration": 19, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-first280-pure-rbs16-n1-ep5/iter_0000019", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.481875, "score_ci_low": 0.464594372, "score_ci_high": 0.4991990939, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 25667.62375, "output_tokens_ci_low": 23631.00391, "output_tokens_ci_high": 27704.243590000002, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 22.4559375, "steps_ci_low": 21.40635152, "steps_ci_high": 23.50552348, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-first280-pure-rbs16-n1-ep5/iter-29/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "first280", "method": "first 280 response tokens", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-first280-pure-rbs16-n1-ep5", "checkpoint_iteration": 29, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-first280-pure-rbs16-n1-ep5/iter_0000029", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.518125, "score_ci_low": 0.5008009061, "score_ci_high": 0.535405628, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 26197.659689999997, "output_tokens_ci_low": 24041.46069, "output_tokens_ci_high": 28353.85869, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 21.338125, "steps_ci_low": 20.35362387, "steps_ci_high": 22.32262613, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-first280-pure-rbs16-n1-ep5/iter-39/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "first280", "method": "first 280 response tokens", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-first280-pure-rbs16-n1-ep5", "checkpoint_iteration": 39, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-first280-pure-rbs16-n1-ep5/iter_0000039", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.52875, "score_ci_low": 0.5114304114, "score_ci_high": 0.5460006426, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 22655.05281, "output_tokens_ci_low": 20593.036529999998, "output_tokens_ci_high": 24717.0691, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 18.1278125, "steps_ci_low": 17.21865596, "steps_ci_high": 19.03696904, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-first280-pure-rbs16-n1-ep5/iter-49/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "first280", "method": "first 280 response tokens", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-first280-pure-rbs16-n1-ep5", "checkpoint_iteration": 49, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-first280-pure-rbs16-n1-ep5/iter_0000049", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.523125, "score_ci_low": 0.5058020503, "score_ci_high": 0.5403924932, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 25283.186560000002, "output_tokens_ci_low": 23211.422769999997, "output_tokens_ci_high": 27354.95035, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 18.1603125, "steps_ci_low": 17.3159484, "steps_ci_high": 19.0046766, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-first280-pure-rbs16-n1-ep5/iter-59/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "first280", "method": "first 280 response tokens", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-first280-pure-rbs16-n1-ep5", "checkpoint_iteration": 59, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-first280-pure-rbs16-n1-ep5/iter_0000059", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.5519074422, "score_ci_low": 0.5346194988, "score_ci_high": 0.5690708274, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 25083.6479, "output_tokens_ci_low": 22812.5848, "output_tokens_ci_high": 27354.711000000003, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 17.57254534, "steps_ci_low": 16.73042343, "steps_ci_high": 18.42457657, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-first280-pure-rbs16-n1-ep5/iter-69/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "first280", "method": "first 280 response tokens", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-first280-pure-rbs16-n1-ep5", "checkpoint_iteration": 69, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-first280-pure-rbs16-n1-ep5/iter_0000069", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.5475, "score_ci_low": 0.5302075291, "score_ci_high": 0.5646785602, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 24182.264369999997, "output_tokens_ci_low": 21899.50801, "output_tokens_ci_high": 26465.02074, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 17.7378125, "steps_ci_low": 16.85506421, "steps_ci_high": 18.62056079, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-first280-pure-rbs16-n1-ep5/iter-79/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "first280", "method": "first 280 response tokens", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-first280-pure-rbs16-n1-ep5", "checkpoint_iteration": 79, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-first280-pure-rbs16-n1-ep5/iter_0000079", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.5528710386, "score_ci_low": 0.5355555724, "score_ci_high": 0.5700591971, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 30567.42473, "output_tokens_ci_low": 27535.427630000002, "output_tokens_ci_high": 33599.421839999995, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 17.61060559, "steps_ci_low": 16.77042428, "steps_ci_high": 18.53954596, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-first280-pure-rbs16-n1-ep5/iter-89/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "first280", "method": "first 280 response tokens", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-first280-pure-rbs16-n1-ep5", "checkpoint_iteration": 89, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-first280-pure-rbs16-n1-ep5/iter_0000089", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.5634375, "score_ci_low": 0.5461874566, "score_ci_high": 0.5805354126, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 22978.7175, "output_tokens_ci_low": 20821.97837, "output_tokens_ci_high": 25135.45663, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 18.0878125, "steps_ci_low": 17.16306685, "steps_ci_high": 19.01255815, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-first280-pure-rbs16-n1-ep5/iter-99/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "first280", "method": "first 280 response tokens", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-first280-pure-rbs16-n1-ep5", "checkpoint_iteration": 99, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-first280-pure-rbs16-n1-ep5/iter_0000099", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.5702220832, "score_ci_low": 0.5529874419, "score_ci_high": 0.5872881656, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 25749.95942, "output_tokens_ci_low": 23345.168950000003, "output_tokens_ci_high": 28154.749890000003, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 17.51454489, "steps_ci_low": 16.65835946, "steps_ci_high": 18.37887268, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-first280-pure-rbs16-n1-ep5/iter-109/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "first280", "method": "first 280 response tokens", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-first280-pure-rbs16-n1-ep5", "checkpoint_iteration": 109, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-first280-pure-rbs16-n1-ep5/iter_0000109", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.5664270084, "score_ci_low": 0.5491842188, "score_ci_high": 0.5835104484, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 23616.900579999998, "output_tokens_ci_low": 21298.85491, "output_tokens_ci_high": 25934.94625, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 16.71991247, "steps_ci_low": 15.85071, "steps_ci_high": 17.59804, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-first280-pure-rbs16-n1-ep5/iter-119/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "first280", "method": "first 280 response tokens", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-first280-pure-rbs16-n1-ep5", "checkpoint_iteration": 119, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-first280-pure-rbs16-n1-ep5/iter_0000119", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.5785356696, "score_ci_low": 0.5613316232, "score_ci_high": 0.5955511425, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 25446.14085, "output_tokens_ci_low": 23059.11277, "output_tokens_ci_high": 27833.16893, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 17.30600751, "steps_ci_low": 16.44363439, "steps_ci_high": 18.19770489, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-first280-pure-rbs16-n1-ep5/iter-129/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "first280", "method": "first 280 response tokens", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-first280-pure-rbs16-n1-ep5", "checkpoint_iteration": 129, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-first280-pure-rbs16-n1-ep5/iter_0000129", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.5939356049, "score_ci_low": 0.5768144747, "score_ci_high": 0.6108313957, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 21454.45737, "output_tokens_ci_low": 19355.641920000002, "output_tokens_ci_high": 23553.27282, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 16.16098781, "steps_ci_low": 15.35195782, "steps_ci_high": 16.96831004, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-first280-pure-rbs16-n1-ep5/iter-139/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "first280", "method": "first 280 response tokens", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-first280-pure-rbs16-n1-ep5", "checkpoint_iteration": 139, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-first280-pure-rbs16-n1-ep5/iter_0000139", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.5828125, "score_ci_low": 0.5656383106, "score_ci_high": 0.599788095, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 21061.71563, "output_tokens_ci_low": 19088.21162, "output_tokens_ci_high": 23035.21963, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 16.5609375, "steps_ci_low": 15.70244029, "steps_ci_high": 17.41943471, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-first280-pure-rbs16-n1-ep5/iter-154/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "first280", "method": "first 280 response tokens", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-first280-pure-rbs16-n1-ep5", "checkpoint_iteration": 154, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-thinkdist-first280-pure-rbs16-n1-ep5/iter_0000154", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_offline_pass8.tsv": "e69556ba74df893c9be1b7db1c80a1ccc439c7ee3f7517cb90030375f96c9aa3", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.566875, "score_ci_low": 0.5496364608, "score_ci_high": 0.5839531649, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 20640.07594, "output_tokens_ci_low": 18696.1781, "output_tokens_ci_high": 22583.97377, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 16.22875, "steps_ci_low": 15.41008867, "steps_ci_high": 17.04741133, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-tthink4k-pure-rbs16-n1-ep5/iter-19/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "tthink4k", "method": "AOD on-policy tail (tthink4k)", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-tthink4k-pure-rbs16-n1-ep5", "checkpoint_iteration": 19, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-tthink4k-pure-rbs16-n1-ep5/iter_0000019", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_tthink4k_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_tthink4k_offline_pass8.tsv": "1bc5908ee5a0423438a0ede02f40cd1b87b41d97778dc6b972ad64396690bb3f", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.4203125, "score_ci_low": 0.4033153584, "score_ci_high": 0.4375007419, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 28278.510000000002, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 27.3728125, "steps_ci_low": 26.0880751, "steps_ci_high": 28.6575499, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-tthink4k-pure-rbs16-n1-ep5/iter-39/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "tthink4k", "method": "AOD on-policy tail (tthink4k)", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-tthink4k-pure-rbs16-n1-ep5", "checkpoint_iteration": 39, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-tthink4k-pure-rbs16-n1-ep5/iter_0000039", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_tthink4k_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_tthink4k_offline_pass8.tsv": "1bc5908ee5a0423438a0ede02f40cd1b87b41d97778dc6b972ad64396690bb3f", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.4953125, "score_ci_low": 0.4780051539, "score_ci_high": 0.5126310873, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 27701.450940000002, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 25.6871875, "steps_ci_low": 24.40226377, "steps_ci_high": 26.97211123, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-tthink4k-pure-rbs16-n1-ep5/iter-49/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "tthink4k", "method": "AOD on-policy tail (tthink4k)", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-tthink4k-pure-rbs16-n1-ep5", "checkpoint_iteration": 49, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-tthink4k-pure-rbs16-n1-ep5/iter_0000049", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_tthink4k_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_tthink4k_offline_pass8.tsv": "1bc5908ee5a0423438a0ede02f40cd1b87b41d97778dc6b972ad64396690bb3f", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.4659375, "score_ci_low": 0.4487047914, "score_ci_high": 0.4832518946, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 29978.145, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 32.6909375, "steps_ci_low": 31.38595768, "steps_ci_high": 33.99591732, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-tthink4k-pure-rbs16-n1-ep5/iter-59/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "tthink4k", "method": "AOD on-policy tail (tthink4k)", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-tthink4k-pure-rbs16-n1-ep5", "checkpoint_iteration": 59, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-tthink4k-pure-rbs16-n1-ep5/iter_0000059", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_tthink4k_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_tthink4k_offline_pass8.tsv": "1bc5908ee5a0423438a0ede02f40cd1b87b41d97778dc6b972ad64396690bb3f", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.434375, "score_ci_low": 0.4172895571, "score_ci_high": 0.4516178196, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 18146.424369999997, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 21.78375, "steps_ci_low": 20.58606049, "steps_ci_high": 22.98143951, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-tthink4k-pure-rbs16-n1-ep5/iter-69/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "tthink4k", "method": "AOD on-policy tail (tthink4k)", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-tthink4k-pure-rbs16-n1-ep5", "checkpoint_iteration": 69, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-tthink4k-pure-rbs16-n1-ep5/iter_0000069", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_tthink4k_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_tthink4k_offline_pass8.tsv": "1bc5908ee5a0423438a0ede02f40cd1b87b41d97778dc6b972ad64396690bb3f", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.4515625, "score_ci_low": 0.4343881891, "score_ci_high": 0.4688529699, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 22422.79469, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 28.6715625, "steps_ci_low": 27.47989267, "steps_ci_high": 29.86323233, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-tthink4k-pure-rbs16-n1-ep5/iter-79/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "tthink4k", "method": "AOD on-policy tail (tthink4k)", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-tthink4k-pure-rbs16-n1-ep5", "checkpoint_iteration": 79, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-tthink4k-pure-rbs16-n1-ep5/iter_0000079", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_tthink4k_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_tthink4k_offline_pass8.tsv": "1bc5908ee5a0423438a0ede02f40cd1b87b41d97778dc6b972ad64396690bb3f", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.479375, "score_ci_low": 0.4621007227, "score_ci_high": 0.4966987385, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 17064.34375, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 23.193125, "steps_ci_low": 21.94681857, "steps_ci_high": 24.43943143, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-tthink4k-pure-rbs16-n1-ep5/iter-89/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "tthink4k", "method": "AOD on-policy tail (tthink4k)", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-tthink4k-pure-rbs16-n1-ep5", "checkpoint_iteration": 89, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-tthink4k-pure-rbs16-n1-ep5/iter_0000089", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_tthink4k_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_tthink4k_offline_pass8.tsv": "1bc5908ee5a0423438a0ede02f40cd1b87b41d97778dc6b972ad64396690bb3f", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.508125, "score_ci_low": 0.4908038143, "score_ci_high": 0.525426701, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 24871.91031, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 31.789375, "steps_ci_low": 30.48261839, "steps_ci_high": 33.09613161, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-tthink4k-pure-rbs16-n1-ep5/iter-99/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "tthink4k", "method": "AOD on-policy tail (tthink4k)", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-tthink4k-pure-rbs16-n1-ep5", "checkpoint_iteration": 99, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-tthink4k-pure-rbs16-n1-ep5/iter_0000099", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_tthink4k_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_tthink4k_offline_pass8.tsv": "1bc5908ee5a0423438a0ede02f40cd1b87b41d97778dc6b972ad64396690bb3f", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.4971875, "score_ci_low": 0.4798774192, "score_ci_high": 0.5145043255, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 17195.170619999997, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 21.0028125, "steps_ci_low": 19.84282651, "steps_ci_high": 22.16279849, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-tthink4k-pure-rbs16-n1-ep5/iter-109/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "tthink4k", "method": "AOD on-policy tail (tthink4k)", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-tthink4k-pure-rbs16-n1-ep5", "checkpoint_iteration": 109, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-tthink4k-pure-rbs16-n1-ep5/iter_0000109", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_tthink4k_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_tthink4k_offline_pass8.tsv": "1bc5908ee5a0423438a0ede02f40cd1b87b41d97778dc6b972ad64396690bb3f", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.4921875, "score_ci_low": 0.4748852521, "score_ci_high": 0.5095084833, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 20703.01281, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 27.1965625, "steps_ci_low": 25.82090183, "steps_ci_high": 28.57222317, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-tthink4k-pure-rbs16-n1-ep5/iter-119/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "tthink4k", "method": "AOD on-policy tail (tthink4k)", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-tthink4k-pure-rbs16-n1-ep5", "checkpoint_iteration": 119, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-tthink4k-pure-rbs16-n1-ep5/iter_0000119", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_tthink4k_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_tthink4k_offline_pass8.tsv": "1bc5908ee5a0423438a0ede02f40cd1b87b41d97778dc6b972ad64396690bb3f", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.4878125, "score_ci_low": 0.4705185248, "score_ci_high": 0.5051357023, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 17994.981249999997, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 27.1021875, "steps_ci_low": 25.71865052, "steps_ci_high": 28.48572448, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-tthink4k-pure-rbs16-n1-ep5/iter-129/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "tthink4k", "method": "AOD on-policy tail (tthink4k)", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-tthink4k-pure-rbs16-n1-ep5", "checkpoint_iteration": 129, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-tthink4k-pure-rbs16-n1-ep5/iter_0000129", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_tthink4k_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_tthink4k_offline_pass8.tsv": "1bc5908ee5a0423438a0ede02f40cd1b87b41d97778dc6b972ad64396690bb3f", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.5175, "score_ci_low": 0.5001758849, "score_ci_high": 0.5347821479, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 20401.34406, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 31.4696875, "steps_ci_low": 30.09762343, "steps_ci_high": 32.84175157, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-tthink4k-pure-rbs16-n1-ep5/iter-139/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "tthink4k", "method": "AOD on-policy tail (tthink4k)", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-tthink4k-pure-rbs16-n1-ep5", "checkpoint_iteration": 139, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-tthink4k-pure-rbs16-n1-ep5/iter_0000139", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_tthink4k_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_tthink4k_offline_pass8.tsv": "1bc5908ee5a0423438a0ede02f40cd1b87b41d97778dc6b972ad64396690bb3f", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.4865625, "score_ci_low": 0.4692711319, "score_ci_high": 0.5038860929, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 13850.89062, "output_tokens_ci_low": 12829.3557, "output_tokens_ci_high": 14872.425550000002, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 23.5484375, "steps_ci_low": 22.2579821, "steps_ci_high": 24.8388929, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-tthink4k-pure-rbs16-n1-ep5/iter-154/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "tthink4k", "method": "AOD on-policy tail (tthink4k)", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-tthink4k-pure-rbs16-n1-ep5", "checkpoint_iteration": 154, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-grpo-full-2node-opdkl1.0-pt-aod2op-tail-tthink4k-pure-rbs16-n1-ep5/iter_0000154", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_tthink4k_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_hill_climb", "ssb1_hill_climb_ci"], "source_sha256": {"notes/figures/series/series_ssb1_tthink4k_offline_pass8.tsv": "1bc5908ee5a0423438a0ede02f40cd1b87b41d97778dc6b972ad64396690bb3f", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.4975, "score_ci_low": 0.4801894871, "score_ci_high": 0.5148165082, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 16316.209690000002, "output_tokens_ci_low": 15116.825139999999, "output_tokens_ci_high": 17515.594230000002, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 24.3115625, "steps_ci_low": 23.05005591, "steps_ci_high": 25.57306909, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-dr_grpo-full-2node-rbs16-ep5/iter-9/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "drgrpo", "method": "Dr. GRPO", "model": "Qwen3.5-4B", "teacher": null, "training_run": "qwen3.5-4B-ssb1-dr_grpo-full-2node-rbs16-ep5", "checkpoint_iteration": 9, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-dr_grpo-full-2node-rbs16-ep5/iter_0000009", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_drgrpo_offline_pass8.tsv"], "figures": ["ssb1_label_hill_climb"], "source_sha256": {"notes/figures/series/series_ssb1_drgrpo_offline_pass8.tsv": "fdf23f76d82cca22ef660c8ae1007807235a3998295f54a54e6f15e027d504a4"}, "score_mean": 0.3221875, "score_ci_low": 0.3062173971, "score_ci_high": 0.3385840188, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 14365.135, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 25.7559375, "steps_ci_low": 24.76697631, "steps_ci_high": 26.74489869, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-dr_grpo-full-2node-rbs16-ep5/iter-19/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "drgrpo", "method": "Dr. GRPO", "model": "Qwen3.5-4B", "teacher": null, "training_run": "qwen3.5-4B-ssb1-dr_grpo-full-2node-rbs16-ep5", "checkpoint_iteration": 19, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-dr_grpo-full-2node-rbs16-ep5/iter_0000019", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_drgrpo_offline_pass8.tsv"], "figures": ["ssb1_label_hill_climb"], "source_sha256": {"notes/figures/series/series_ssb1_drgrpo_offline_pass8.tsv": "fdf23f76d82cca22ef660c8ae1007807235a3998295f54a54e6f15e027d504a4"}, "score_mean": 0.3403125, "score_ci_low": 0.3240958784, "score_ci_high": 0.3569120716, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 14673.79406, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 23.2809375, "steps_ci_low": 22.36457316, "steps_ci_high": 24.19730184, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-dr_grpo-full-2node-rbs16-ep5/iter-29/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "drgrpo", "method": "Dr. GRPO", "model": "Qwen3.5-4B", "teacher": null, "training_run": "qwen3.5-4B-ssb1-dr_grpo-full-2node-rbs16-ep5", "checkpoint_iteration": 29, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-dr_grpo-full-2node-rbs16-ep5/iter_0000029", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_drgrpo_offline_pass8.tsv"], "figures": ["ssb1_label_hill_climb"], "source_sha256": {"notes/figures/series/series_ssb1_drgrpo_offline_pass8.tsv": "fdf23f76d82cca22ef660c8ae1007807235a3998295f54a54e6f15e027d504a4"}, "score_mean": 0.3653125, "score_ci_low": 0.3487994853, "score_ci_high": 0.3821485116, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 18254.710939999997, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 23.4896875, "steps_ci_low": 22.60911025, "steps_ci_high": 24.37026475, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-dr_grpo-full-2node-rbs16-ep5/iter-39/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "drgrpo", "method": "Dr. GRPO", "model": "Qwen3.5-4B", "teacher": null, "training_run": "qwen3.5-4B-ssb1-dr_grpo-full-2node-rbs16-ep5", "checkpoint_iteration": 39, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-dr_grpo-full-2node-rbs16-ep5/iter_0000039", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_drgrpo_offline_pass8.tsv"], "figures": ["ssb1_label_hill_climb"], "source_sha256": {"notes/figures/series/series_ssb1_drgrpo_offline_pass8.tsv": "fdf23f76d82cca22ef660c8ae1007807235a3998295f54a54e6f15e027d504a4"}, "score_mean": 0.4065625, "score_ci_low": 0.3896654453, "score_ci_high": 0.4236836291, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 22575.52187, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 24.726875, "steps_ci_low": 23.7684529, "steps_ci_high": 25.6852971, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-dr_grpo-full-2node-rbs16-ep5/iter-49/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "drgrpo", "method": "Dr. GRPO", "model": "Qwen3.5-4B", "teacher": null, "training_run": "qwen3.5-4B-ssb1-dr_grpo-full-2node-rbs16-ep5", "checkpoint_iteration": 49, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-dr_grpo-full-2node-rbs16-ep5/iter_0000049", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_drgrpo_offline_pass8.tsv"], "figures": ["ssb1_label_hill_climb"], "source_sha256": {"notes/figures/series/series_ssb1_drgrpo_offline_pass8.tsv": "fdf23f76d82cca22ef660c8ae1007807235a3998295f54a54e6f15e027d504a4"}, "score_mean": 0.443125, "score_ci_low": 0.4259917111, "score_ci_high": 0.4603946821, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 24946.59344, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 25.3015625, "steps_ci_low": 24.31119506, "steps_ci_high": 26.29192994, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-dr_grpo-full-2node-rbs16-ep5/iter-59/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "drgrpo", "method": "Dr. GRPO", "model": "Qwen3.5-4B", "teacher": null, "training_run": "qwen3.5-4B-ssb1-dr_grpo-full-2node-rbs16-ep5", "checkpoint_iteration": 59, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-dr_grpo-full-2node-rbs16-ep5/iter_0000059", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_drgrpo_offline_pass8.tsv"], "figures": ["ssb1_label_hill_climb"], "source_sha256": {"notes/figures/series/series_ssb1_drgrpo_offline_pass8.tsv": "fdf23f76d82cca22ef660c8ae1007807235a3998295f54a54e6f15e027d504a4"}, "score_mean": 0.4696875, "score_ci_low": 0.4524419283, "score_ci_high": 0.4870057648, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 30551.99781, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 26.858125, "steps_ci_low": 25.85791994, "steps_ci_high": 27.85833006, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-dr_grpo-full-2node-rbs16-ep5/iter-69/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "drgrpo", "method": "Dr. GRPO", "model": "Qwen3.5-4B", "teacher": null, "training_run": "qwen3.5-4B-ssb1-dr_grpo-full-2node-rbs16-ep5", "checkpoint_iteration": 69, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-dr_grpo-full-2node-rbs16-ep5/iter_0000069", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_drgrpo_offline_pass8.tsv"], "figures": ["ssb1_label_hill_climb"], "source_sha256": {"notes/figures/series/series_ssb1_drgrpo_offline_pass8.tsv": "fdf23f76d82cca22ef660c8ae1007807235a3998295f54a54e6f15e027d504a4"}, "score_mean": 0.51125, "score_ci_low": 0.4939271617, "score_ci_high": 0.5285458594, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 31948.257810000003, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 22.9425, "steps_ci_low": 22.10190851, "steps_ci_high": 23.78309149, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-dr_grpo-full-2node-rbs16-ep5/iter-79/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "drgrpo", "method": "Dr. GRPO", "model": "Qwen3.5-4B", "teacher": null, "training_run": "qwen3.5-4B-ssb1-dr_grpo-full-2node-rbs16-ep5", "checkpoint_iteration": 79, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-dr_grpo-full-2node-rbs16-ep5/iter_0000079", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_drgrpo_offline_pass8.tsv"], "figures": ["ssb1_label_hill_climb"], "source_sha256": {"notes/figures/series/series_ssb1_drgrpo_offline_pass8.tsv": "fdf23f76d82cca22ef660c8ae1007807235a3998295f54a54e6f15e027d504a4"}, "score_mean": 0.5221875, "score_ci_low": 0.5048642037, "score_ci_high": 0.539457588, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 28134.05281, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 20.1584375, "steps_ci_low": 19.39012073, "steps_ci_high": 20.92675427, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-dr_grpo-full-2node-rbs16-ep5/iter-89/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "drgrpo", "method": "Dr. GRPO", "model": "Qwen3.5-4B", "teacher": null, "training_run": "qwen3.5-4B-ssb1-dr_grpo-full-2node-rbs16-ep5", "checkpoint_iteration": 89, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-dr_grpo-full-2node-rbs16-ep5/iter_0000089", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_drgrpo_offline_pass8.tsv"], "figures": ["ssb1_label_hill_climb"], "source_sha256": {"notes/figures/series/series_ssb1_drgrpo_offline_pass8.tsv": "fdf23f76d82cca22ef660c8ae1007807235a3998295f54a54e6f15e027d504a4"}, "score_mean": 0.5, "score_ci_low": 0.4826862733, "score_ci_high": 0.5173137267, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 28696.00687, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 19.7959375, "steps_ci_low": 19.04680348, "steps_ci_high": 20.54507152, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-dr_grpo-full-2node-rbs16-ep5/iter-99/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "drgrpo", "method": "Dr. GRPO", "model": "Qwen3.5-4B", "teacher": null, "training_run": "qwen3.5-4B-ssb1-dr_grpo-full-2node-rbs16-ep5", "checkpoint_iteration": 99, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-dr_grpo-full-2node-rbs16-ep5/iter_0000099", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_drgrpo_offline_pass8.tsv"], "figures": ["ssb1_label_hill_climb"], "source_sha256": {"notes/figures/series/series_ssb1_drgrpo_offline_pass8.tsv": "fdf23f76d82cca22ef660c8ae1007807235a3998295f54a54e6f15e027d504a4"}, "score_mean": 0.4967177243, "score_ci_low": 0.4794056045, "score_ci_high": 0.5140377178, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 28186.053770000002, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 19.21381682, "steps_ci_low": 18.50728583, "steps_ci_high": 19.92226774, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-dr_grpo-full-2node-rbs16-ep5/iter-109/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "drgrpo", "method": "Dr. GRPO", "model": "Qwen3.5-4B", "teacher": null, "training_run": "qwen3.5-4B-ssb1-dr_grpo-full-2node-rbs16-ep5", "checkpoint_iteration": 109, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-dr_grpo-full-2node-rbs16-ep5/iter_0000109", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_drgrpo_offline_pass8.tsv"], "figures": ["ssb1_label_hill_climb"], "source_sha256": {"notes/figures/series/series_ssb1_drgrpo_offline_pass8.tsv": "fdf23f76d82cca22ef660c8ae1007807235a3998295f54a54e6f15e027d504a4"}, "score_mean": 0.519375, "score_ci_low": 0.5020510296, "score_ci_high": 0.5366525068, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 38024.73906, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 20.0409375, "steps_ci_low": 19.29758443, "steps_ci_high": 20.78429057, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-dr_grpo-full-2node-rbs16-ep5/iter-119/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "drgrpo", "method": "Dr. GRPO", "model": "Qwen3.5-4B", "teacher": null, "training_run": "qwen3.5-4B-ssb1-dr_grpo-full-2node-rbs16-ep5", "checkpoint_iteration": 119, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-dr_grpo-full-2node-rbs16-ep5/iter_0000119", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_drgrpo_offline_pass8.tsv"], "figures": ["ssb1_label_hill_climb"], "source_sha256": {"notes/figures/series/series_ssb1_drgrpo_offline_pass8.tsv": "fdf23f76d82cca22ef660c8ae1007807235a3998295f54a54e6f15e027d504a4"}, "score_mean": 0.51438399, "score_ci_low": 0.4970547582, "score_ci_high": 0.5316787057, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 45406.670730000005, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 19.72357724, "steps_ci_low": 18.98926944, "steps_ci_high": 20.46090913, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-dr_grpo-full-2node-rbs16-ep5/iter-129/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "drgrpo", "method": "Dr. GRPO", "model": "Qwen3.5-4B", "teacher": null, "training_run": "qwen3.5-4B-ssb1-dr_grpo-full-2node-rbs16-ep5", "checkpoint_iteration": 129, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-dr_grpo-full-2node-rbs16-ep5/iter_0000129", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_drgrpo_offline_pass8.tsv"], "figures": ["ssb1_label_hill_climb"], "source_sha256": {"notes/figures/series/series_ssb1_drgrpo_offline_pass8.tsv": "fdf23f76d82cca22ef660c8ae1007807235a3998295f54a54e6f15e027d504a4"}, "score_mean": 0.5570490778, "score_ci_low": 0.539777171, "score_ci_high": 0.5741841314, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 46773.52016, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 20.10659581, "steps_ci_low": 19.40136855, "steps_ci_high": 20.81122073, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-dr_grpo-full-2node-rbs16-ep5/iter-139/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "drgrpo", "method": "Dr. GRPO", "model": "Qwen3.5-4B", "teacher": null, "training_run": "qwen3.5-4B-ssb1-dr_grpo-full-2node-rbs16-ep5", "checkpoint_iteration": 139, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-dr_grpo-full-2node-rbs16-ep5/iter_0000139", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_drgrpo_offline_pass8.tsv"], "figures": ["ssb1_label_hill_climb"], "source_sha256": {"notes/figures/series/series_ssb1_drgrpo_offline_pass8.tsv": "fdf23f76d82cca22ef660c8ae1007807235a3998295f54a54e6f15e027d504a4"}, "score_mean": 0.5546875, "score_ci_low": 0.5374119475, "score_ci_high": 0.5718319053, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 39744.12875, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 16.9759375, "steps_ci_low": 16.28201662, "steps_ci_high": 17.66985838, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-dr_grpo-full-2node-rbs16-ep5/iter-154/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "drgrpo", "method": "Dr. GRPO", "model": "Qwen3.5-4B", "teacher": null, "training_run": "qwen3.5-4B-ssb1-dr_grpo-full-2node-rbs16-ep5", "checkpoint_iteration": 154, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-dr_grpo-full-2node-rbs16-ep5/iter_0000154", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_drgrpo_offline_pass8.tsv"], "figures": ["ssb1_label_hill_climb"], "source_sha256": {"notes/figures/series/series_ssb1_drgrpo_offline_pass8.tsv": "fdf23f76d82cca22ef660c8ae1007807235a3998295f54a54e6f15e027d504a4"}, "score_mean": 0.5765625, "score_ci_low": 0.5593609077, "score_ci_high": 0.5935804861, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 35342.204690000006, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 17.9228125, "steps_ci_low": 17.21125607, "steps_ci_high": 18.63436893, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-dr_grpo-full-2node-opdkl1.0-pt-aod2op-tail-rbs16-ep5/iter-9/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "drgrpotail", "method": "Dr. GRPO + AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-dr_grpo-full-2node-opdkl1.0-pt-aod2op-tail-rbs16-ep5", "checkpoint_iteration": 9, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-dr_grpo-full-2node-opdkl1.0-pt-aod2op-tail-rbs16-ep5/iter_0000009", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_drgrpotail_offline_pass8.tsv"], "figures": ["ssb1_label_hill_climb"], "source_sha256": {"notes/figures/series/series_ssb1_drgrpotail_offline_pass8.tsv": "8de7d099a5ff1f1444548ff0b1b75169842630fe9ff6f9af412190f4c566c07c"}, "score_mean": 0.529375, "score_ci_low": 0.5120559205, "score_ci_high": 0.5466236346, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 34950.481250000004, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 19.1390625, "steps_ci_low": 18.20088015, "steps_ci_high": 20.07724485, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-dr_grpo-full-2node-opdkl1.0-pt-aod2op-tail-rbs16-ep5/iter-19/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "drgrpotail", "method": "Dr. GRPO + AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-dr_grpo-full-2node-opdkl1.0-pt-aod2op-tail-rbs16-ep5", "checkpoint_iteration": 19, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-dr_grpo-full-2node-opdkl1.0-pt-aod2op-tail-rbs16-ep5/iter_0000019", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_drgrpotail_offline_pass8.tsv"], "figures": ["ssb1_label_hill_climb"], "source_sha256": {"notes/figures/series/series_ssb1_drgrpotail_offline_pass8.tsv": "8de7d099a5ff1f1444548ff0b1b75169842630fe9ff6f9af412190f4c566c07c"}, "score_mean": 0.5278125, "score_ci_low": 0.5104921986, "score_ci_high": 0.5450661037, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 10464.625310000001, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 17.9884375, "steps_ci_low": 17.04632683, "steps_ci_high": 18.93054817, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-dr_grpo-full-2node-opdkl1.0-pt-aod2op-tail-rbs16-ep5/iter-29/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "drgrpotail", "method": "Dr. GRPO + AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-dr_grpo-full-2node-opdkl1.0-pt-aod2op-tail-rbs16-ep5", "checkpoint_iteration": 29, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-dr_grpo-full-2node-opdkl1.0-pt-aod2op-tail-rbs16-ep5/iter_0000029", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_drgrpotail_offline_pass8.tsv"], "figures": ["ssb1_label_hill_climb"], "source_sha256": {"notes/figures/series/series_ssb1_drgrpotail_offline_pass8.tsv": "8de7d099a5ff1f1444548ff0b1b75169842630fe9ff6f9af412190f4c566c07c"}, "score_mean": 0.5734375, "score_ci_low": 0.5562232572, "score_ci_high": 0.5904756307, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 11175.95375, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 15.2115625, "steps_ci_low": 14.40250467, "steps_ci_high": 16.02062033, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-dr_grpo-full-2node-opdkl1.0-pt-aod2op-tail-rbs16-ep5/iter-39/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "drgrpotail", "method": "Dr. GRPO + AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-dr_grpo-full-2node-opdkl1.0-pt-aod2op-tail-rbs16-ep5", "checkpoint_iteration": 39, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-dr_grpo-full-2node-opdkl1.0-pt-aod2op-tail-rbs16-ep5/iter_0000039", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_drgrpotail_offline_pass8.tsv"], "figures": ["ssb1_label_hill_climb"], "source_sha256": {"notes/figures/series/series_ssb1_drgrpotail_offline_pass8.tsv": "8de7d099a5ff1f1444548ff0b1b75169842630fe9ff6f9af412190f4c566c07c"}, "score_mean": 0.589375, "score_ci_low": 0.572232616, "score_ci_high": 0.606303052, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 10077.92375, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 15.90125, "steps_ci_low": 15.04391792, "steps_ci_high": 16.75858208, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-dr_grpo-full-2node-opdkl1.0-pt-aod2op-tail-rbs16-ep5/iter-49/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "drgrpotail", "method": "Dr. GRPO + AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-dr_grpo-full-2node-opdkl1.0-pt-aod2op-tail-rbs16-ep5", "checkpoint_iteration": 49, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-dr_grpo-full-2node-opdkl1.0-pt-aod2op-tail-rbs16-ep5/iter_0000049", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_drgrpotail_offline_pass8.tsv"], "figures": ["ssb1_label_hill_climb"], "source_sha256": {"notes/figures/series/series_ssb1_drgrpotail_offline_pass8.tsv": "8de7d099a5ff1f1444548ff0b1b75169842630fe9ff6f9af412190f4c566c07c"}, "score_mean": 0.591875, "score_ci_low": 0.5747455537, "score_ci_high": 0.6087841189, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 9695.715625, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 15.95625, "steps_ci_low": 15.03031807, "steps_ci_high": 16.88218193, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-dr_grpo-full-2node-opdkl1.0-pt-aod2op-tail-rbs16-ep5/iter-59/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "drgrpotail", "method": "Dr. GRPO + AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-dr_grpo-full-2node-opdkl1.0-pt-aod2op-tail-rbs16-ep5", "checkpoint_iteration": 59, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-dr_grpo-full-2node-opdkl1.0-pt-aod2op-tail-rbs16-ep5/iter_0000059", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_drgrpotail_offline_pass8.tsv"], "figures": ["ssb1_label_hill_climb"], "source_sha256": {"notes/figures/series/series_ssb1_drgrpotail_offline_pass8.tsv": "8de7d099a5ff1f1444548ff0b1b75169842630fe9ff6f9af412190f4c566c07c"}, "score_mean": 0.604375, "score_ci_low": 0.5873170969, "score_ci_high": 0.6211825992, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 8970.888125, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 14.6125, "steps_ci_low": 13.79599346, "steps_ci_high": 15.42900654, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-dr_grpo-full-2node-opdkl1.0-pt-aod2op-tail-rbs16-ep5/iter-69/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "drgrpotail", "method": "Dr. GRPO + AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-dr_grpo-full-2node-opdkl1.0-pt-aod2op-tail-rbs16-ep5", "checkpoint_iteration": 69, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-dr_grpo-full-2node-opdkl1.0-pt-aod2op-tail-rbs16-ep5/iter_0000069", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_drgrpotail_offline_pass8.tsv"], "figures": ["ssb1_label_hill_climb"], "source_sha256": {"notes/figures/series/series_ssb1_drgrpotail_offline_pass8.tsv": "8de7d099a5ff1f1444548ff0b1b75169842630fe9ff6f9af412190f4c566c07c"}, "score_mean": 0.6228125, "score_ci_low": 0.605881275, "score_ci_high": 0.6394492058, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 8890.404688, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 15.2134375, "steps_ci_low": 14.37360799, "steps_ci_high": 16.05326701, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-dr_grpo-full-2node-opdkl1.0-pt-aod2op-tail-rbs16-ep5/iter-79/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "drgrpotail", "method": "Dr. GRPO + AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-dr_grpo-full-2node-opdkl1.0-pt-aod2op-tail-rbs16-ep5", "checkpoint_iteration": 79, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-dr_grpo-full-2node-opdkl1.0-pt-aod2op-tail-rbs16-ep5/iter_0000079", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_drgrpotail_offline_pass8.tsv"], "figures": ["ssb1_label_hill_climb"], "source_sha256": {"notes/figures/series/series_ssb1_drgrpotail_offline_pass8.tsv": "8de7d099a5ff1f1444548ff0b1b75169842630fe9ff6f9af412190f4c566c07c"}, "score_mean": 0.61625, "score_ci_low": 0.5992707641, "score_ci_high": 0.6329504543, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 8748.440625000001, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 14.7709375, "steps_ci_low": 13.95930553, "steps_ci_high": 15.58256947, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-dr_grpo-full-2node-opdkl1.0-pt-aod2op-tail-rbs16-ep5/iter-89/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "drgrpotail", "method": "Dr. GRPO + AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-dr_grpo-full-2node-opdkl1.0-pt-aod2op-tail-rbs16-ep5", "checkpoint_iteration": 89, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-dr_grpo-full-2node-opdkl1.0-pt-aod2op-tail-rbs16-ep5/iter_0000089", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_drgrpotail_offline_pass8.tsv"], "figures": ["ssb1_label_hill_climb"], "source_sha256": {"notes/figures/series/series_ssb1_drgrpotail_offline_pass8.tsv": "8de7d099a5ff1f1444548ff0b1b75169842630fe9ff6f9af412190f4c566c07c"}, "score_mean": 0.621875, "score_ci_low": 0.6049367168, "score_ci_high": 0.6385210122, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 8700.650938, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 15.0053125, "steps_ci_low": 14.17241403, "steps_ci_high": 15.83821097, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-dr_grpo-full-2node-opdkl1.0-pt-aod2op-tail-rbs16-ep5/iter-99/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "drgrpotail", "method": "Dr. GRPO + AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-dr_grpo-full-2node-opdkl1.0-pt-aod2op-tail-rbs16-ep5", "checkpoint_iteration": 99, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-dr_grpo-full-2node-opdkl1.0-pt-aod2op-tail-rbs16-ep5/iter_0000099", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_drgrpotail_offline_pass8.tsv"], "figures": ["ssb1_label_hill_climb"], "source_sha256": {"notes/figures/series/series_ssb1_drgrpotail_offline_pass8.tsv": "8de7d099a5ff1f1444548ff0b1b75169842630fe9ff6f9af412190f4c566c07c"}, "score_mean": 0.6209375, "score_ci_low": 0.6039922253, "score_ci_high": 0.6375927519, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 8850.397188, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 15.3775, "steps_ci_low": 14.5222055, "steps_ci_high": 16.2327945, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-dr_grpo-full-2node-opdkl1.0-pt-aod2op-tail-rbs16-ep5/iter-109/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "drgrpotail", "method": "Dr. GRPO + AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-dr_grpo-full-2node-opdkl1.0-pt-aod2op-tail-rbs16-ep5", "checkpoint_iteration": 109, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-dr_grpo-full-2node-opdkl1.0-pt-aod2op-tail-rbs16-ep5/iter_0000109", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_drgrpotail_offline_pass8.tsv"], "figures": ["ssb1_label_hill_climb"], "source_sha256": {"notes/figures/series/series_ssb1_drgrpotail_offline_pass8.tsv": "8de7d099a5ff1f1444548ff0b1b75169842630fe9ff6f9af412190f4c566c07c"}, "score_mean": 0.63625, "score_ci_low": 0.6194273247, "score_ci_high": 0.6527459313, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 8859.996562999999, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 15.0078125, "steps_ci_low": 14.14780576, "steps_ci_high": 15.86781924, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-dr_grpo-full-2node-opdkl1.0-pt-aod2op-tail-rbs16-ep5/iter-119/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "drgrpotail", "method": "Dr. GRPO + AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-dr_grpo-full-2node-opdkl1.0-pt-aod2op-tail-rbs16-ep5", "checkpoint_iteration": 119, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-dr_grpo-full-2node-opdkl1.0-pt-aod2op-tail-rbs16-ep5/iter_0000119", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_drgrpotail_offline_pass8.tsv"], "figures": ["ssb1_label_hill_climb"], "source_sha256": {"notes/figures/series/series_ssb1_drgrpotail_offline_pass8.tsv": "8de7d099a5ff1f1444548ff0b1b75169842630fe9ff6f9af412190f4c566c07c"}, "score_mean": 0.6246875, "score_ci_low": 0.6077705915, "score_ci_high": 0.6413053928, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 8504.010624999999, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 14.39875, "steps_ci_low": 13.60642282, "steps_ci_high": 15.19107718, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-dr_grpo-full-2node-opdkl1.0-pt-aod2op-tail-rbs16-ep5/iter-129/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "drgrpotail", "method": "Dr. GRPO + AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-dr_grpo-full-2node-opdkl1.0-pt-aod2op-tail-rbs16-ep5", "checkpoint_iteration": 129, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-dr_grpo-full-2node-opdkl1.0-pt-aod2op-tail-rbs16-ep5/iter_0000129", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_drgrpotail_offline_pass8.tsv"], "figures": ["ssb1_label_hill_climb"], "source_sha256": {"notes/figures/series/series_ssb1_drgrpotail_offline_pass8.tsv": "8de7d099a5ff1f1444548ff0b1b75169842630fe9ff6f9af412190f4c566c07c"}, "score_mean": 0.64125, "score_ci_low": 0.6244712773, "score_ci_high": 0.6576899881, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 9162.037187, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 14.6021875, "steps_ci_low": 13.79975721, "steps_ci_high": 15.40461779, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-dr_grpo-full-2node-opdkl1.0-pt-aod2op-tail-rbs16-ep5/iter-139/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "drgrpotail", "method": "Dr. GRPO + AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-dr_grpo-full-2node-opdkl1.0-pt-aod2op-tail-rbs16-ep5", "checkpoint_iteration": 139, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-dr_grpo-full-2node-opdkl1.0-pt-aod2op-tail-rbs16-ep5/iter_0000139", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_drgrpotail_offline_pass8.tsv"], "figures": ["ssb1_label_hill_climb"], "source_sha256": {"notes/figures/series/series_ssb1_drgrpotail_offline_pass8.tsv": "8de7d099a5ff1f1444548ff0b1b75169842630fe9ff6f9af412190f4c566c07c"}, "score_mean": 0.6325, "score_ci_low": 0.6156456351, "score_ci_high": 0.6490366139, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 9419.479063, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 15.0078125, "steps_ci_low": 14.19943173, "steps_ci_high": 15.81619327, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-4B-ssb1-dr_grpo-full-2node-opdkl1.0-pt-aod2op-tail-rbs16-ep5/iter-154/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "drgrpotail", "method": "Dr. GRPO + AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-ssb1-dr_grpo-full-2node-opdkl1.0-pt-aod2op-tail-rbs16-ep5", "checkpoint_iteration": 154, "checkpoint_folder": "checkpoints/qwen3.5-4B-ssb1-dr_grpo-full-2node-opdkl1.0-pt-aod2op-tail-rbs16-ep5/iter_0000154", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_ssb1_drgrpotail_offline_pass8.tsv"], "figures": ["ssb1_label_hill_climb"], "source_sha256": {"notes/figures/series/series_ssb1_drgrpotail_offline_pass8.tsv": "8de7d099a5ff1f1444548ff0b1b75169842630fe9ff6f9af412190f4c566c07c"}, "score_mean": 0.6459375, "score_ci_low": 0.6292017616, "score_ci_high": 0.6623232626, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 9895.249375000001, "output_tokens_ci_low": null, "output_tokens_ci_high": null, "output_tokens_ci_level": null, "output_tokens_ci_method": null, "output_tokens_ci_status": "not_in_plot_source", "steps_mean": 15.433125, "steps_ci_low": 14.61886348, "steps_ci_high": 16.24738652, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter-19/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "q9bssb1pure", "method": "pure OPD", "model": "Qwen3.5-9B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-pure-rbs16-n1-ep5", "checkpoint_iteration": 19, "checkpoint_folder": "checkpoints/qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter_0000019", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_q9bssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_qwen9b_hill_climb"], "source_sha256": {"notes/figures/series/series_q9bssb1_offline_pass8.tsv": "793ed06da5cbd5c423bc0b0276b6b4a67a4de579ad2bf31cabd5edb5141c8625", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.5403125, "score_ci_low": 0.5230067332, "score_ci_high": 0.5575215925, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 19590.881559999998, "output_tokens_ci_low": 17749.842380000002, "output_tokens_ci_high": 21431.920739999998, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 19.8525, "steps_ci_low": 18.72016534, "steps_ci_high": 20.98483466, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter-39/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "q9bssb1pure", "method": "pure OPD", "model": "Qwen3.5-9B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-pure-rbs16-n1-ep5", "checkpoint_iteration": 39, "checkpoint_folder": "checkpoints/qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter_0000039", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_q9bssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_qwen9b_hill_climb"], "source_sha256": {"notes/figures/series/series_q9bssb1_offline_pass8.tsv": "793ed06da5cbd5c423bc0b0276b6b4a67a4de579ad2bf31cabd5edb5141c8625", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.5659375, "score_ci_low": 0.5486957401, "score_ci_high": 0.5830211338, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 19298.730630000002, "output_tokens_ci_low": 17520.84486, "output_tokens_ci_high": 21076.616390000003, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 18.3303125, "steps_ci_low": 17.2965436, "steps_ci_high": 19.3640814, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter-59/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "q9bssb1pure", "method": "pure OPD", "model": "Qwen3.5-9B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-pure-rbs16-n1-ep5", "checkpoint_iteration": 59, "checkpoint_folder": "checkpoints/qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter_0000059", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_q9bssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_qwen9b_hill_climb"], "source_sha256": {"notes/figures/series/series_q9bssb1_offline_pass8.tsv": "793ed06da5cbd5c423bc0b0276b6b4a67a4de579ad2bf31cabd5edb5141c8625", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.5965625, "score_ci_low": 0.5794585399, "score_ci_high": 0.6134348916, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 19229.57781, "output_tokens_ci_low": 17376.57351, "output_tokens_ci_high": 21082.58211, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 17.81875, "steps_ci_low": 16.7844946, "steps_ci_high": 18.8530054, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter-69/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "q9bssb1pure", "method": "pure OPD", "model": "Qwen3.5-9B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-pure-rbs16-n1-ep5", "checkpoint_iteration": 69, "checkpoint_folder": "checkpoints/qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter_0000069", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_q9bssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_qwen9b_hill_climb"], "source_sha256": {"notes/figures/series/series_q9bssb1_offline_pass8.tsv": "793ed06da5cbd5c423bc0b0276b6b4a67a4de579ad2bf31cabd5edb5141c8625", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.61375, "score_ci_low": 0.5967533276, "score_ci_high": 0.6304738861, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 16434.256250000002, "output_tokens_ci_low": 14851.67893, "output_tokens_ci_high": 18016.83357, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 17.748125, "steps_ci_low": 16.68520811, "steps_ci_high": 18.81104189, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter-79/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "q9bssb1pure", "method": "pure OPD", "model": "Qwen3.5-9B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-pure-rbs16-n1-ep5", "checkpoint_iteration": 79, "checkpoint_folder": "checkpoints/qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter_0000079", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_q9bssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_qwen9b_hill_climb"], "source_sha256": {"notes/figures/series/series_q9bssb1_offline_pass8.tsv": "793ed06da5cbd5c423bc0b0276b6b4a67a4de579ad2bf31cabd5edb5141c8625", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.611875, "score_ci_low": 0.5948655576, "score_ci_high": 0.6286161526, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 17475.861559999998, "output_tokens_ci_low": 15824.152460000001, "output_tokens_ci_high": 19127.57067, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 17.574375, "steps_ci_low": 16.54694155, "steps_ci_high": 18.60180845, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter-89/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "q9bssb1pure", "method": "pure OPD", "model": "Qwen3.5-9B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-pure-rbs16-n1-ep5", "checkpoint_iteration": 89, "checkpoint_folder": "checkpoints/qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter_0000089", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_q9bssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_qwen9b_hill_climb"], "source_sha256": {"notes/figures/series/series_q9bssb1_offline_pass8.tsv": "793ed06da5cbd5c423bc0b0276b6b4a67a4de579ad2bf31cabd5edb5141c8625", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.616875, "score_ci_low": 0.5999001966, "score_ci_high": 0.633569523, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 19119.52594, "output_tokens_ci_low": 17394.85131, "output_tokens_ci_high": 20844.20057, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 18.685, "steps_ci_low": 17.62232517, "steps_ci_high": 19.74767483, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter-99/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "q9bssb1pure", "method": "pure OPD", "model": "Qwen3.5-9B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-pure-rbs16-n1-ep5", "checkpoint_iteration": 99, "checkpoint_folder": "checkpoints/qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter_0000099", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_q9bssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_qwen9b_hill_climb"], "source_sha256": {"notes/figures/series/series_q9bssb1_offline_pass8.tsv": "793ed06da5cbd5c423bc0b0276b6b4a67a4de579ad2bf31cabd5edb5141c8625", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.6009375, "score_ci_low": 0.5838587784, "score_ci_high": 0.6177741612, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 15568.33406, "output_tokens_ci_low": 14029.3239, "output_tokens_ci_high": 17107.344220000003, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 16.2721875, "steps_ci_low": 15.30001888, "steps_ci_high": 17.24435612, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter-109/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "q9bssb1pure", "method": "pure OPD", "model": "Qwen3.5-9B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-pure-rbs16-n1-ep5", "checkpoint_iteration": 109, "checkpoint_folder": "checkpoints/qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter_0000109", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_q9bssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_qwen9b_hill_climb"], "source_sha256": {"notes/figures/series/series_q9bssb1_offline_pass8.tsv": "793ed06da5cbd5c423bc0b0276b6b4a67a4de579ad2bf31cabd5edb5141c8625", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.6096875, "score_ci_low": 0.5926634911, "score_ci_high": 0.626448465, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 15686.63156, "output_tokens_ci_low": 14230.04845, "output_tokens_ci_high": 17143.21468, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 17.4884375, "steps_ci_low": 16.47586735, "steps_ci_high": 18.50100765, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter-119/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "q9bssb1pure", "method": "pure OPD", "model": "Qwen3.5-9B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-pure-rbs16-n1-ep5", "checkpoint_iteration": 119, "checkpoint_folder": "checkpoints/qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter_0000119", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_q9bssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_qwen9b_hill_climb"], "source_sha256": {"notes/figures/series/series_q9bssb1_offline_pass8.tsv": "793ed06da5cbd5c423bc0b0276b6b4a67a4de579ad2bf31cabd5edb5141c8625", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.624375, "score_ci_low": 0.6074556869, "score_ci_high": 0.6409960468, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 17769.44062, "output_tokens_ci_low": 16079.456310000001, "output_tokens_ci_high": 19459.42494, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 17.199375, "steps_ci_low": 16.18323711, "steps_ci_high": 18.21551289, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter-129/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "q9bssb1pure", "method": "pure OPD", "model": "Qwen3.5-9B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-pure-rbs16-n1-ep5", "checkpoint_iteration": 129, "checkpoint_folder": "checkpoints/qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter_0000129", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_q9bssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_qwen9b_hill_climb"], "source_sha256": {"notes/figures/series/series_q9bssb1_offline_pass8.tsv": "793ed06da5cbd5c423bc0b0276b6b4a67a4de579ad2bf31cabd5edb5141c8625", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.6184375, "score_ci_low": 0.6014739064, "score_ci_high": 0.6351170661, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 15162.84094, "output_tokens_ci_low": 13663.92063, "output_tokens_ci_high": 16661.76124, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 17.3771875, "steps_ci_low": 16.31564172, "steps_ci_high": 18.43873328, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter-139/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "q9bssb1pure", "method": "pure OPD", "model": "Qwen3.5-9B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-pure-rbs16-n1-ep5", "checkpoint_iteration": 139, "checkpoint_folder": "checkpoints/qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter_0000139", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_q9bssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_qwen9b_hill_climb"], "source_sha256": {"notes/figures/series/series_q9bssb1_offline_pass8.tsv": "793ed06da5cbd5c423bc0b0276b6b4a67a4de579ad2bf31cabd5edb5141c8625", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.618125, "score_ci_low": 0.6011591497, "score_ci_high": 0.6348075722, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 18437.52781, "output_tokens_ci_low": 16700.549049999998, "output_tokens_ci_high": 20174.506569999998, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 18.0146875, "steps_ci_low": 16.96217908, "steps_ci_high": 19.06719592, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter-154/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "q9bssb1pure", "method": "pure OPD", "model": "Qwen3.5-9B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-pure-rbs16-n1-ep5", "checkpoint_iteration": 154, "checkpoint_folder": "checkpoints/qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter_0000154", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_q9bssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_qwen9b_hill_climb"], "source_sha256": {"notes/figures/series/series_q9bssb1_offline_pass8.tsv": "793ed06da5cbd5c423bc0b0276b6b4a67a4de579ad2bf31cabd5edb5141c8625", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.6284375, "score_ci_low": 0.6115500287, "score_ci_high": 0.6450169626, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 15798.047499999999, "output_tokens_ci_low": 14248.625479999999, "output_tokens_ci_high": 17347.469520000002, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 16.7690625, "steps_ci_low": 15.77044193, "steps_ci_high": 17.76768307, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5/iter-9/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "q9bssb1aod", "method": "AOD on-policy tail", "model": "Qwen3.5-9B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5", "checkpoint_iteration": 9, "checkpoint_folder": "checkpoints/qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5/iter_0000009", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_q9bssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_qwen9b_hill_climb"], "source_sha256": {"notes/figures/series/series_q9bssb1_offline_pass8.tsv": "793ed06da5cbd5c423bc0b0276b6b4a67a4de579ad2bf31cabd5edb5141c8625", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.5865625, "score_ci_low": 0.5694061034, "score_ci_high": 0.6035113092, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 17607.892809999998, "output_tokens_ci_low": 16298.751070000002, "output_tokens_ci_high": 18917.03456, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 19.5546875, "steps_ci_low": 18.54752999, "steps_ci_high": 20.56184501, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5/iter-19/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "q9bssb1aod", "method": "AOD on-policy tail", "model": "Qwen3.5-9B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5", "checkpoint_iteration": 19, "checkpoint_folder": "checkpoints/qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5/iter_0000019", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_q9bssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_qwen9b_hill_climb"], "source_sha256": {"notes/figures/series/series_q9bssb1_offline_pass8.tsv": "793ed06da5cbd5c423bc0b0276b6b4a67a4de579ad2bf31cabd5edb5141c8625", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.6153125, "score_ci_low": 0.5983266705, "score_ci_high": 0.6320217962, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 14646.58031, "output_tokens_ci_low": 13394.77833, "output_tokens_ci_high": 15898.382290000001, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 15.66875, "steps_ci_low": 14.81122829, "steps_ci_high": 16.52627171, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5/iter-29/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "q9bssb1aod", "method": "AOD on-policy tail", "model": "Qwen3.5-9B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5", "checkpoint_iteration": 29, "checkpoint_folder": "checkpoints/qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5/iter_0000029", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_q9bssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_qwen9b_hill_climb"], "source_sha256": {"notes/figures/series/series_q9bssb1_offline_pass8.tsv": "793ed06da5cbd5c423bc0b0276b6b4a67a4de579ad2bf31cabd5edb5141c8625", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.6228125, "score_ci_low": 0.605881275, "score_ci_high": 0.6394492058, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 12069.960630000001, "output_tokens_ci_low": 11099.337329999998, "output_tokens_ci_high": 13040.58392, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 17.0115625, "steps_ci_low": 16.06804448, "steps_ci_high": 17.95508052, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5/iter-39/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "q9bssb1aod", "method": "AOD on-policy tail", "model": "Qwen3.5-9B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5", "checkpoint_iteration": 39, "checkpoint_folder": "checkpoints/qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5/iter_0000039", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_q9bssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_qwen9b_hill_climb"], "source_sha256": {"notes/figures/series/series_q9bssb1_offline_pass8.tsv": "793ed06da5cbd5c423bc0b0276b6b4a67a4de579ad2bf31cabd5edb5141c8625", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.628125, "score_ci_low": 0.6112350345, "score_ci_high": 0.6447077062, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 12388.98437, "output_tokens_ci_low": 11405.80142, "output_tokens_ci_high": 13372.16733, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 15.6540625, "steps_ci_low": 14.80449671, "steps_ci_high": 16.50362829, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5/iter-49/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "q9bssb1aod", "method": "AOD on-policy tail", "model": "Qwen3.5-9B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5", "checkpoint_iteration": 49, "checkpoint_folder": "checkpoints/qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5/iter_0000049", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_q9bssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_qwen9b_hill_climb"], "source_sha256": {"notes/figures/series/series_q9bssb1_offline_pass8.tsv": "793ed06da5cbd5c423bc0b0276b6b4a67a4de579ad2bf31cabd5edb5141c8625", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.619375, "score_ci_low": 0.6024182207, "score_ci_high": 0.6360455036, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 11402.20312, "output_tokens_ci_low": 10517.33396, "output_tokens_ci_high": 12287.07229, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 15.0496875, "steps_ci_low": 14.25957248, "steps_ci_high": 15.83980252, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5/iter-59/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "q9bssb1aod", "method": "AOD on-policy tail", "model": "Qwen3.5-9B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5", "checkpoint_iteration": 59, "checkpoint_folder": "checkpoints/qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5/iter_0000059", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_q9bssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_qwen9b_hill_climb"], "source_sha256": {"notes/figures/series/series_q9bssb1_offline_pass8.tsv": "793ed06da5cbd5c423bc0b0276b6b4a67a4de579ad2bf31cabd5edb5141c8625", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.619375, "score_ci_low": 0.6024182207, "score_ci_high": 0.6360455036, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 11528.007810000001, "output_tokens_ci_low": 10652.326410000001, "output_tokens_ci_high": 12403.68922, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 15.519375, "steps_ci_low": 14.70736439, "steps_ci_high": 16.33138561, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5/iter-69/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "q9bssb1aod", "method": "AOD on-policy tail", "model": "Qwen3.5-9B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5", "checkpoint_iteration": 69, "checkpoint_folder": "checkpoints/qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5/iter_0000069", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_q9bssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_qwen9b_hill_climb"], "source_sha256": {"notes/figures/series/series_q9bssb1_offline_pass8.tsv": "793ed06da5cbd5c423bc0b0276b6b4a67a4de579ad2bf31cabd5edb5141c8625", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.6421875, "score_ci_low": 0.625417236, "score_ci_high": 0.6586167812, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 10691.150309999999, "output_tokens_ci_low": 9870.556098, "output_tokens_ci_high": 11511.74453, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 15.3428125, "steps_ci_low": 14.53392173, "steps_ci_high": 16.15170327, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5/iter-79/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "q9bssb1aod", "method": "AOD on-policy tail", "model": "Qwen3.5-9B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5", "checkpoint_iteration": 79, "checkpoint_folder": "checkpoints/qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5/iter_0000079", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_q9bssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_qwen9b_hill_climb"], "source_sha256": {"notes/figures/series/series_q9bssb1_offline_pass8.tsv": "793ed06da5cbd5c423bc0b0276b6b4a67a4de579ad2bf31cabd5edb5141c8625", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.6496875, "score_ci_low": 0.6329873988, "score_ci_high": 0.6660286324, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 10802.960000000001, "output_tokens_ci_low": 9964.393351, "output_tokens_ci_high": 11641.52665, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 14.8159375, "steps_ci_low": 14.02283411, "steps_ci_high": 15.60904089, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5/iter-89/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "q9bssb1aod", "method": "AOD on-policy tail", "model": "Qwen3.5-9B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5", "checkpoint_iteration": 89, "checkpoint_folder": "checkpoints/qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5/iter_0000089", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_q9bssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_qwen9b_hill_climb"], "source_sha256": {"notes/figures/series/series_q9bssb1_offline_pass8.tsv": "793ed06da5cbd5c423bc0b0276b6b4a67a4de579ad2bf31cabd5edb5141c8625", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.6525, "score_ci_low": 0.6358273611, "score_ci_high": 0.6688069255, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 9900.611874999999, "output_tokens_ci_low": 9133.157333, "output_tokens_ci_high": 10668.066420000001, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 14.7584375, "steps_ci_low": 13.97426694, "steps_ci_high": 15.54260806, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5/iter-99/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "q9bssb1aod", "method": "AOD on-policy tail", "model": "Qwen3.5-9B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5", "checkpoint_iteration": 99, "checkpoint_folder": "checkpoints/qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5/iter_0000099", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_q9bssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_qwen9b_hill_climb"], "source_sha256": {"notes/figures/series/series_q9bssb1_offline_pass8.tsv": "793ed06da5cbd5c423bc0b0276b6b4a67a4de579ad2bf31cabd5edb5141c8625", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.6584375, "score_ci_low": 0.6418249212, "score_ci_high": 0.6746701265, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 10098.427810000001, "output_tokens_ci_low": 9320.846151, "output_tokens_ci_high": 10876.009469999999, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 14.5615625, "steps_ci_low": 13.75697435, "steps_ci_high": 15.36615065, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5/iter-109/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "q9bssb1aod", "method": "AOD on-policy tail", "model": "Qwen3.5-9B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5", "checkpoint_iteration": 109, "checkpoint_folder": "checkpoints/qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5/iter_0000109", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_q9bssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_qwen9b_hill_climb"], "source_sha256": {"notes/figures/series/series_q9bssb1_offline_pass8.tsv": "793ed06da5cbd5c423bc0b0276b6b4a67a4de579ad2bf31cabd5edb5141c8625", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.6571875, "score_ci_low": 0.6405620407, "score_ci_high": 0.6734360046, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 9450.057812000001, "output_tokens_ci_low": 8742.415987999999, "output_tokens_ci_high": 10157.69964, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 14.76, "steps_ci_low": 13.97368892, "steps_ci_high": 15.54631108, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5/iter-119/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "q9bssb1aod", "method": "AOD on-policy tail", "model": "Qwen3.5-9B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5", "checkpoint_iteration": 119, "checkpoint_folder": "checkpoints/qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5/iter_0000119", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_q9bssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_qwen9b_hill_climb"], "source_sha256": {"notes/figures/series/series_q9bssb1_offline_pass8.tsv": "793ed06da5cbd5c423bc0b0276b6b4a67a4de579ad2bf31cabd5edb5141c8625", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.65625, "score_ci_low": 0.6396149632, "score_ci_high": 0.6725103304, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 9471.862812, "output_tokens_ci_low": 8765.323158, "output_tokens_ci_high": 10178.402469999999, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 14.4640625, "steps_ci_low": 13.709133, "steps_ci_high": 15.218992, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5/iter-129/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "q9bssb1aod", "method": "AOD on-policy tail", "model": "Qwen3.5-9B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5", "checkpoint_iteration": 129, "checkpoint_folder": "checkpoints/qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5/iter_0000129", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_q9bssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_qwen9b_hill_climb"], "source_sha256": {"notes/figures/series/series_q9bssb1_offline_pass8.tsv": "793ed06da5cbd5c423bc0b0276b6b4a67a4de579ad2bf31cabd5edb5141c8625", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.654375, "score_ci_low": 0.6377210209, "score_ci_high": 0.6706587692, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 9285.39875, "output_tokens_ci_low": 8592.869574, "output_tokens_ci_high": 9977.927926, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 15.02375, "steps_ci_low": 14.24108197, "steps_ci_high": 15.80641803, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5/iter-139/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "q9bssb1aod", "method": "AOD on-policy tail", "model": "Qwen3.5-9B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5", "checkpoint_iteration": 139, "checkpoint_folder": "checkpoints/qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5/iter_0000139", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_q9bssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_qwen9b_hill_climb"], "source_sha256": {"notes/figures/series/series_q9bssb1_offline_pass8.tsv": "793ed06da5cbd5c423bc0b0276b6b4a67a4de579ad2bf31cabd5edb5141c8625", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.66875, "score_ci_low": 0.6522485496, "score_ci_high": 0.6848467675, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 9365.771875, "output_tokens_ci_low": 8660.015224, "output_tokens_ci_high": 10071.52853, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 15.763125, "steps_ci_low": 14.91406531, "steps_ci_high": 16.61218469, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5/iter-154/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "q9bssb1aod", "method": "AOD on-policy tail", "model": "Qwen3.5-9B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5", "checkpoint_iteration": 154, "checkpoint_folder": "checkpoints/qwen3.5-9B-ssb1-grpo-full-4node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5/iter_0000154", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_q9bssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_qwen9b_hill_climb"], "source_sha256": {"notes/figures/series/series_q9bssb1_offline_pass8.tsv": "793ed06da5cbd5c423bc0b0276b6b4a67a4de579ad2bf31cabd5edb5141c8625", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.653125, "score_ci_low": 0.6364585497, "score_ci_high": 0.669424238, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 9213.578438, "output_tokens_ci_low": 8557.338639, "output_tokens_ci_high": 9869.818236000001, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 14.88625, "steps_ci_low": 14.11047886, "steps_ci_high": 15.66202114, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-27B-ssb1-grpo-full-4node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter-19/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "q27bssb1pure", "method": "pure OPD", "model": "Qwen3.5-27B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-27B-ssb1-grpo-full-4node-opdkl1.0-pt-pure-rbs16-n1-ep5", "checkpoint_iteration": 19, "checkpoint_folder": "checkpoints/qwen3.5-27B-ssb1-grpo-full-4node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter_0000019", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_q27bssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_qwen27b_hill_climb"], "source_sha256": {"notes/figures/series/series_q27bssb1_offline_pass8.tsv": "a3b4b05aa85044776531315a77c4b172d98cdc909fa919df690c27e8e21ea054", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.734375, "score_ci_low": 0.718797639, "score_ci_high": 0.7493903014, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 9760.644375, "output_tokens_ci_low": 8691.734813000001, "output_tokens_ci_high": 10829.55394, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 13.68625, "steps_ci_low": 12.84277434, "steps_ci_high": 14.52972566, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-27B-ssb1-grpo-full-4node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter-29/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "q27bssb1pure", "method": "pure OPD", "model": "Qwen3.5-27B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-27B-ssb1-grpo-full-4node-opdkl1.0-pt-pure-rbs16-n1-ep5", "checkpoint_iteration": 29, "checkpoint_folder": "checkpoints/qwen3.5-27B-ssb1-grpo-full-4node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter_0000029", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_q27bssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_qwen27b_hill_climb"], "source_sha256": {"notes/figures/series/series_q27bssb1_offline_pass8.tsv": "a3b4b05aa85044776531315a77c4b172d98cdc909fa919df690c27e8e21ea054", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.73375, "score_ci_low": 0.718161939, "score_ci_high": 0.7487775002, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 10963.826879999999, "output_tokens_ci_low": 9758.875253, "output_tokens_ci_high": 12168.7785, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 13.5284375, "steps_ci_low": 12.72736791, "steps_ci_high": 14.32950709, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-27B-ssb1-grpo-full-4node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter-39/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "q27bssb1pure", "method": "pure OPD", "model": "Qwen3.5-27B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-27B-ssb1-grpo-full-4node-opdkl1.0-pt-pure-rbs16-n1-ep5", "checkpoint_iteration": 39, "checkpoint_folder": "checkpoints/qwen3.5-27B-ssb1-grpo-full-4node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter_0000039", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_q27bssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_qwen27b_hill_climb"], "source_sha256": {"notes/figures/series/series_q27bssb1_offline_pass8.tsv": "a3b4b05aa85044776531315a77c4b172d98cdc909fa919df690c27e8e21ea054", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.7484375, "score_ci_low": 0.7331114207, "score_ci_high": 0.7631677961, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 11981.30938, "output_tokens_ci_low": 10709.049719999999, "output_tokens_ci_high": 13253.569029999999, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 14.853125, "steps_ci_low": 13.94469042, "steps_ci_high": 15.76155958, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-27B-ssb1-grpo-full-4node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter-49/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "q27bssb1pure", "method": "pure OPD", "model": "Qwen3.5-27B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-27B-ssb1-grpo-full-4node-opdkl1.0-pt-pure-rbs16-n1-ep5", "checkpoint_iteration": 49, "checkpoint_folder": "checkpoints/qwen3.5-27B-ssb1-grpo-full-4node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter_0000049", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_q27bssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_qwen27b_hill_climb"], "source_sha256": {"notes/figures/series/series_q27bssb1_offline_pass8.tsv": "a3b4b05aa85044776531315a77c4b172d98cdc909fa919df690c27e8e21ea054", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.740625, "score_ci_low": 0.7251568086, "score_ci_high": 0.7555161435, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 9941.391563, "output_tokens_ci_low": 8861.970648, "output_tokens_ci_high": 11020.81248, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 12.766875, "steps_ci_low": 12.0085285, "steps_ci_high": 13.5252215, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-27B-ssb1-grpo-full-4node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter-59/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "q27bssb1pure", "method": "pure OPD", "model": "Qwen3.5-27B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-27B-ssb1-grpo-full-4node-opdkl1.0-pt-pure-rbs16-n1-ep5", "checkpoint_iteration": 59, "checkpoint_folder": "checkpoints/qwen3.5-27B-ssb1-grpo-full-4node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter_0000059", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_q27bssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_qwen27b_hill_climb"], "source_sha256": {"notes/figures/series/series_q27bssb1_offline_pass8.tsv": "a3b4b05aa85044776531315a77c4b172d98cdc909fa919df690c27e8e21ea054", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.760625, "score_ci_low": 0.7455335863, "score_ci_high": 0.7750914034, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 10728.20687, "output_tokens_ci_low": 9590.049858, "output_tokens_ci_high": 11866.36389, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 13.5284375, "steps_ci_low": 12.71882967, "steps_ci_high": 14.33804533, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-27B-ssb1-grpo-full-4node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter-69/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "q27bssb1pure", "method": "pure OPD", "model": "Qwen3.5-27B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-27B-ssb1-grpo-full-4node-opdkl1.0-pt-pure-rbs16-n1-ep5", "checkpoint_iteration": 69, "checkpoint_folder": "checkpoints/qwen3.5-27B-ssb1-grpo-full-4node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter_0000069", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_q27bssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_qwen27b_hill_climb"], "source_sha256": {"notes/figures/series/series_q27bssb1_offline_pass8.tsv": "a3b4b05aa85044776531315a77c4b172d98cdc909fa919df690c27e8e21ea054", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.745, "score_ci_low": 0.7296106075, "score_ci_high": 0.7598018529, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 9424.77125, "output_tokens_ci_low": 8429.342609000001, "output_tokens_ci_high": 10420.19989, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 12.6534375, "steps_ci_low": 11.90990732, "steps_ci_high": 13.39696768, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-27B-ssb1-grpo-full-4node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter-79/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "q27bssb1pure", "method": "pure OPD", "model": "Qwen3.5-27B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-27B-ssb1-grpo-full-4node-opdkl1.0-pt-pure-rbs16-n1-ep5", "checkpoint_iteration": 79, "checkpoint_folder": "checkpoints/qwen3.5-27B-ssb1-grpo-full-4node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter_0000079", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_q27bssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_qwen27b_hill_climb"], "source_sha256": {"notes/figures/series/series_q27bssb1_offline_pass8.tsv": "a3b4b05aa85044776531315a77c4b172d98cdc909fa919df690c27e8e21ea054", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.7409375, "score_ci_low": 0.7254748715, "score_ci_high": 0.7558223312, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 10028.60844, "output_tokens_ci_low": 8951.942567, "output_tokens_ci_high": 11105.27431, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 12.8496875, "steps_ci_low": 12.08560729, "steps_ci_high": 13.61376771, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-27B-ssb1-grpo-full-4node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter-89/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "q27bssb1pure", "method": "pure OPD", "model": "Qwen3.5-27B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-27B-ssb1-grpo-full-4node-opdkl1.0-pt-pure-rbs16-n1-ep5", "checkpoint_iteration": 89, "checkpoint_folder": "checkpoints/qwen3.5-27B-ssb1-grpo-full-4node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter_0000089", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_q27bssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_qwen27b_hill_climb"], "source_sha256": {"notes/figures/series/series_q27bssb1_offline_pass8.tsv": "a3b4b05aa85044776531315a77c4b172d98cdc909fa919df690c27e8e21ea054", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.7525, "score_ci_low": 0.7372503581, "score_ci_high": 0.7671441163, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 9771.698750000001, "output_tokens_ci_low": 8719.978028, "output_tokens_ci_high": 10823.419469999999, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 12.9484375, "steps_ci_low": 12.18717841, "steps_ci_high": 13.70969659, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-27B-ssb1-grpo-full-4node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter-99/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "q27bssb1pure", "method": "pure OPD", "model": "Qwen3.5-27B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-27B-ssb1-grpo-full-4node-opdkl1.0-pt-pure-rbs16-n1-ep5", "checkpoint_iteration": 99, "checkpoint_folder": "checkpoints/qwen3.5-27B-ssb1-grpo-full-4node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter_0000099", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_q27bssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_qwen27b_hill_climb"], "source_sha256": {"notes/figures/series/series_q27bssb1_offline_pass8.tsv": "a3b4b05aa85044776531315a77c4b172d98cdc909fa919df690c27e8e21ea054", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.74625, "score_ci_low": 0.7308834869, "score_ci_high": 0.7610259758, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 9875.199063, "output_tokens_ci_low": 8770.985308, "output_tokens_ci_high": 10979.41282, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 12.6878125, "steps_ci_low": 11.9176941, "steps_ci_high": 13.4579309, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-27B-ssb1-grpo-full-4node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter-109/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "q27bssb1pure", "method": "pure OPD", "model": "Qwen3.5-27B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-27B-ssb1-grpo-full-4node-opdkl1.0-pt-pure-rbs16-n1-ep5", "checkpoint_iteration": 109, "checkpoint_folder": "checkpoints/qwen3.5-27B-ssb1-grpo-full-4node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter_0000109", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_q27bssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_qwen27b_hill_climb"], "source_sha256": {"notes/figures/series/series_q27bssb1_offline_pass8.tsv": "a3b4b05aa85044776531315a77c4b172d98cdc909fa919df690c27e8e21ea054", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.765, "score_ci_low": 0.7499968115, "score_ci_high": 0.7793676864, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 10059.543440000001, "output_tokens_ci_low": 8979.336888, "output_tokens_ci_high": 11139.74999, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 12.77625, "steps_ci_low": 12.02807163, "steps_ci_high": 13.52442837, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-27B-ssb1-grpo-full-4node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter-119/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "q27bssb1pure", "method": "pure OPD", "model": "Qwen3.5-27B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-27B-ssb1-grpo-full-4node-opdkl1.0-pt-pure-rbs16-n1-ep5", "checkpoint_iteration": 119, "checkpoint_folder": "checkpoints/qwen3.5-27B-ssb1-grpo-full-4node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter_0000119", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_q27bssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_qwen27b_hill_climb"], "source_sha256": {"notes/figures/series/series_q27bssb1_offline_pass8.tsv": "a3b4b05aa85044776531315a77c4b172d98cdc909fa919df690c27e8e21ea054", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.7584375, "score_ci_low": 0.7433027746, "score_ci_high": 0.772952461, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 10496.586560000002, "output_tokens_ci_low": 9367.329066, "output_tokens_ci_high": 11625.844060000001, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 13.5371875, "steps_ci_low": 12.69805133, "steps_ci_high": 14.37632367, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-27B-ssb1-grpo-full-4node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter-129/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "q27bssb1pure", "method": "pure OPD", "model": "Qwen3.5-27B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-27B-ssb1-grpo-full-4node-opdkl1.0-pt-pure-rbs16-n1-ep5", "checkpoint_iteration": 129, "checkpoint_folder": "checkpoints/qwen3.5-27B-ssb1-grpo-full-4node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter_0000129", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_q27bssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_qwen27b_hill_climb"], "source_sha256": {"notes/figures/series/series_q27bssb1_offline_pass8.tsv": "a3b4b05aa85044776531315a77c4b172d98cdc909fa919df690c27e8e21ea054", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.759375, "score_ci_low": 0.7442587719, "score_ci_high": 0.7738692154, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 9514.1375, "output_tokens_ci_low": 8480.85642, "output_tokens_ci_high": 10547.41858, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 12.588125, "steps_ci_low": 11.83370002, "steps_ci_high": 13.34254998, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-27B-ssb1-grpo-full-4node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter-139/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "q27bssb1pure", "method": "pure OPD", "model": "Qwen3.5-27B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-27B-ssb1-grpo-full-4node-opdkl1.0-pt-pure-rbs16-n1-ep5", "checkpoint_iteration": 139, "checkpoint_folder": "checkpoints/qwen3.5-27B-ssb1-grpo-full-4node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter_0000139", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_q27bssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_qwen27b_hill_climb"], "source_sha256": {"notes/figures/series/series_q27bssb1_offline_pass8.tsv": "a3b4b05aa85044776531315a77c4b172d98cdc909fa919df690c27e8e21ea054", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.7628125, "score_ci_low": 0.7477649303, "score_ci_high": 0.7772298135, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 10570.375, "output_tokens_ci_low": 9457.072801, "output_tokens_ci_high": 11683.6772, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 13.240625, "steps_ci_low": 12.45595854, "steps_ci_high": 14.02529146, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-27B-ssb1-grpo-full-4node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter-154/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "q27bssb1pure", "method": "pure OPD", "model": "Qwen3.5-27B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-27B-ssb1-grpo-full-4node-opdkl1.0-pt-pure-rbs16-n1-ep5", "checkpoint_iteration": 154, "checkpoint_folder": "checkpoints/qwen3.5-27B-ssb1-grpo-full-4node-opdkl1.0-pt-pure-rbs16-n1-ep5/iter_0000154", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_q27bssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_qwen27b_hill_climb"], "source_sha256": {"notes/figures/series/series_q27bssb1_offline_pass8.tsv": "a3b4b05aa85044776531315a77c4b172d98cdc909fa919df690c27e8e21ea054", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.7428125, "score_ci_low": 0.7273834604, "score_ci_high": 0.7576592458, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 9719.492813, "output_tokens_ci_low": 8678.458783, "output_tokens_ci_high": 10760.52684, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 12.8934375, "steps_ci_low": 12.15296627, "steps_ci_high": 13.63390873, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-27B-ssb1-grpo-full-4node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5/iter-109/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "q27bssb1aod", "method": "AOD on-policy tail", "model": "Qwen3.5-27B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-27B-ssb1-grpo-full-4node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5", "checkpoint_iteration": 109, "checkpoint_folder": "checkpoints/qwen3.5-27B-ssb1-grpo-full-4node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5/iter_0000109", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_q27bssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_qwen27b_hill_climb"], "source_sha256": {"notes/figures/series/series_q27bssb1_offline_pass8.tsv": "a3b4b05aa85044776531315a77c4b172d98cdc909fa919df690c27e8e21ea054", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.77125, "score_ci_low": 0.7563766222, "score_ci_high": 0.7854728875, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 6184.8815620000005, "output_tokens_ci_low": 5695.864723, "output_tokens_ci_high": 6673.898402, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 11.9865625, "steps_ci_low": 11.35535117, "steps_ci_high": 12.61777383, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-27B-ssb1-grpo-full-4node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5/iter-119/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "q27bssb1aod", "method": "AOD on-policy tail", "model": "Qwen3.5-27B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-27B-ssb1-grpo-full-4node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5", "checkpoint_iteration": 119, "checkpoint_folder": "checkpoints/qwen3.5-27B-ssb1-grpo-full-4node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5/iter_0000119", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_q27bssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_qwen27b_hill_climb"], "source_sha256": {"notes/figures/series/series_q27bssb1_offline_pass8.tsv": "a3b4b05aa85044776531315a77c4b172d98cdc909fa919df690c27e8e21ea054", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.7575, "score_ci_low": 0.7423468741, "score_ci_high": 0.7720356097, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 6649.839688, "output_tokens_ci_low": 6139.17592, "output_tokens_ci_high": 7160.503455, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 12.0684375, "steps_ci_low": 11.46286272, "steps_ci_high": 12.67401228, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-27B-ssb1-grpo-full-4node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5/iter-129/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "q27bssb1aod", "method": "AOD on-policy tail", "model": "Qwen3.5-27B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-27B-ssb1-grpo-full-4node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5", "checkpoint_iteration": 129, "checkpoint_folder": "checkpoints/qwen3.5-27B-ssb1-grpo-full-4node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5/iter_0000129", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_q27bssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_qwen27b_hill_climb"], "source_sha256": {"notes/figures/series/series_q27bssb1_offline_pass8.tsv": "a3b4b05aa85044776531315a77c4b172d98cdc909fa919df690c27e8e21ea054", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.77375, "score_ci_low": 0.7589298145, "score_ci_high": 0.7879136999, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 7628.86125, "output_tokens_ci_low": 7007.6921170000005, "output_tokens_ci_high": 8250.030383000001, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 12.788125, "steps_ci_low": 12.1316912, "steps_ci_high": 13.4445588, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-27B-ssb1-grpo-full-4node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5/iter-139/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "q27bssb1aod", "method": "AOD on-policy tail", "model": "Qwen3.5-27B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-27B-ssb1-grpo-full-4node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5", "checkpoint_iteration": 139, "checkpoint_folder": "checkpoints/qwen3.5-27B-ssb1-grpo-full-4node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5/iter_0000139", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_q27bssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_qwen27b_hill_climb"], "source_sha256": {"notes/figures/series/series_q27bssb1_offline_pass8.tsv": "a3b4b05aa85044776531315a77c4b172d98cdc909fa919df690c27e8e21ea054", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.759375, "score_ci_low": 0.7442587719, "score_ci_high": 0.7738692154, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 9479.845313, "output_tokens_ci_low": 8781.044661, "output_tokens_ci_high": 10178.645960000002, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 14.6959375, "steps_ci_low": 13.99004548, "steps_ci_high": 15.40182952, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "ssb1/qwen3.5-27B-ssb1-grpo-full-4node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5/iter-154/attempts8-cap50", "dataset": "ssb1", "eval_split": "Verified-400", "source_kind": "offline_saved_checkpoint", "family": "q27bssb1aod", "method": "AOD on-policy tail", "model": "Qwen3.5-27B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-27B-ssb1-grpo-full-4node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5", "checkpoint_iteration": 154, "checkpoint_folder": "checkpoints/qwen3.5-27B-ssb1-grpo-full-4node-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep5/iter_0000154", "temperature": 1.0, "step_cap": 50, "attempts_per_task": 8, "protocol_task_count": 400, "protocol_episode_count": 3200, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": null, "score_metric": "hard_restriction_pass_at_1", "source_files": ["notes/figures/series/series_q27bssb1_offline_pass8.tsv", "notes/figures/series/series_ssb1_token_ci.tsv"], "figures": ["ssb1_qwen27b_hill_climb"], "source_sha256": {"notes/figures/series/series_q27bssb1_offline_pass8.tsv": "a3b4b05aa85044776531315a77c4b172d98cdc909fa919df690c27e8e21ea054", "notes/figures/series/series_ssb1_token_ci.tsv": "87747bf6911808cc546059ff55f6247d5bc20a44db9f4d962958eba4b8a8c4e6"}, "score_mean": 0.788125, "score_ci_low": 0.7736252542, "score_ci_high": 0.8019337872, "score_ci_level": 0.95, "score_ci_method": "Wilson, z=1.96, over attempts (not task-cluster adjusted)", "score_ci_status": "available", "output_tokens_mean": 6198.74625, "output_tokens_ci_low": 5682.878233, "output_tokens_ci_high": 6714.614267, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "output_tokens_ci_status": "available", "steps_mean": 11.4590625, "steps_ci_low": 10.82316633, "steps_ci_high": 12.09495867, "steps_ci_level": 0.95, "steps_ci_method": "normal interval over task means, mean +/- 1.96*SE, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": null, "no_reward_fraction_ci_low": null, "no_reward_fraction_ci_high": null, "no_reward_fraction_ci_level": null, "no_reward_fraction_ci_method": null, "no_reward_fraction_ci_status": null}
{"evaluation_id": "jobbench_easy/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pure-rbs16-n1-ep20-ctx128k/iter-3/attempts8-cap120", "dataset": "jobbench_easy", "eval_split": "easy", "source_kind": "offline_saved_checkpoint", "family": "pure", "method": "pure OPD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pure-rbs16-n1-ep20-ctx128k", "checkpoint_iteration": 3, "checkpoint_folder": "checkpoints/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pure-rbs16-n1-ep20-ctx128k/iter_0000003", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 63, "protocol_episode_count": 504, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": "stage", "score_metric": "normalized_rubric_score", "source_files": ["notes/figures/series/series_jobbenchmt_pure_offline_pass8_cap120.tsv"], "figures": ["jobbench_mtrain_hill_climb"], "source_sha256": {"notes/figures/series/series_jobbenchmt_pure_offline_pass8_cap120.tsv": "0525131a60debee698514a2ddaeecf657914382e5a96d63fea3029a8151f578c"}, "score_mean": 0.3195226190476191, "score_ci_low": 0.2717950396825397, "score_ci_high": 0.36710178571428564, "score_ci_level": 0.95, "score_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "score_ci_status": "available", "output_tokens_mean": 35006.22222222222, "output_tokens_ci_low": 30604.928571428572, "output_tokens_ci_high": 39803.27976190476, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 45.68849206349206, "steps_ci_low": 41.13492063492063, "steps_ci_high": 50.53174603174603, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": 0.047619047619047616, "no_reward_fraction_ci_low": 0.011904761904761904, "no_reward_fraction_ci_high": 0.09126984126984126, "no_reward_fraction_ci_level": 0.95, "no_reward_fraction_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "no_reward_fraction_ci_status": "available"}
{"evaluation_id": "jobbench_easy/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pure-rbs16-n1-ep20-ctx128k/iter-7/attempts8-cap120", "dataset": "jobbench_easy", "eval_split": "easy", "source_kind": "offline_saved_checkpoint", "family": "pure", "method": "pure OPD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pure-rbs16-n1-ep20-ctx128k", "checkpoint_iteration": 7, "checkpoint_folder": "checkpoints/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pure-rbs16-n1-ep20-ctx128k/iter_0000007", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 63, "protocol_episode_count": 504, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": "stage", "score_metric": "normalized_rubric_score", "source_files": ["notes/figures/series/series_jobbenchmt_pure_offline_pass8_cap120.tsv"], "figures": ["jobbench_mtrain_hill_climb"], "source_sha256": {"notes/figures/series/series_jobbenchmt_pure_offline_pass8_cap120.tsv": "0525131a60debee698514a2ddaeecf657914382e5a96d63fea3029a8151f578c"}, "score_mean": 0.33489503968253964, "score_ci_low": 0.2858932539682539, "score_ci_high": 0.38511765873015863, "score_ci_level": 0.95, "score_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "score_ci_status": "available", "output_tokens_mean": 35683.90079365079, "output_tokens_ci_low": 31104.007936507936, "output_tokens_ci_high": 40672.75, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 46.839285714285715, "steps_ci_low": 42.160714285714285, "steps_ci_high": 51.63492063492063, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": 0.06349206349206349, "no_reward_fraction_ci_low": 0.023809523809523808, "no_reward_fraction_ci_high": 0.11507936507936507, "no_reward_fraction_ci_level": 0.95, "no_reward_fraction_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "no_reward_fraction_ci_status": "available"}
{"evaluation_id": "jobbench_easy/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pure-rbs16-n1-ep20-ctx128k/iter-9/attempts8-cap120", "dataset": "jobbench_easy", "eval_split": "easy", "source_kind": "offline_saved_checkpoint", "family": "pure", "method": "pure OPD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pure-rbs16-n1-ep20-ctx128k", "checkpoint_iteration": 9, "checkpoint_folder": "checkpoints/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pure-rbs16-n1-ep20-ctx128k/iter_0000009", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 63, "protocol_episode_count": 504, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": "stage", "score_metric": "normalized_rubric_score", "source_files": ["notes/figures/series/series_jobbenchmt_pure_offline_pass8_cap120.tsv"], "figures": ["jobbench_mtrain_hill_climb"], "source_sha256": {"notes/figures/series/series_jobbenchmt_pure_offline_pass8_cap120.tsv": "0525131a60debee698514a2ddaeecf657914382e5a96d63fea3029a8151f578c"}, "score_mean": 0.3242857142857143, "score_ci_low": 0.2777218253968254, "score_ci_high": 0.3720128968253969, "score_ci_level": 0.95, "score_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "score_ci_status": "available", "output_tokens_mean": 34277.083333333336, "output_tokens_ci_low": 29090.617063492064, "output_tokens_ci_high": 40056.482142857145, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 43.56944444444444, "steps_ci_low": 39.10515873015873, "steps_ci_high": 48.375, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": 0.02976190476190476, "no_reward_fraction_ci_low": 0.005952380952380952, "no_reward_fraction_ci_high": 0.06547619047619048, "no_reward_fraction_ci_level": 0.95, "no_reward_fraction_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "no_reward_fraction_ci_status": "available"}
{"evaluation_id": "jobbench_easy/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pure-rbs16-n1-ep20-ctx128k/iter-11/attempts8-cap120", "dataset": "jobbench_easy", "eval_split": "easy", "source_kind": "offline_saved_checkpoint", "family": "pure", "method": "pure OPD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pure-rbs16-n1-ep20-ctx128k", "checkpoint_iteration": 11, "checkpoint_folder": "checkpoints/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pure-rbs16-n1-ep20-ctx128k/iter_0000011", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 63, "protocol_episode_count": 504, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": "stage", "score_metric": "normalized_rubric_score", "source_files": ["notes/figures/series/series_jobbenchmt_pure_offline_pass8_cap120.tsv"], "figures": ["jobbench_mtrain_hill_climb"], "source_sha256": {"notes/figures/series/series_jobbenchmt_pure_offline_pass8_cap120.tsv": "0525131a60debee698514a2ddaeecf657914382e5a96d63fea3029a8151f578c"}, "score_mean": 0.3181406746031748, "score_ci_low": 0.27097658730158725, "score_ci_high": 0.3667412698412699, "score_ci_level": 0.95, "score_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "score_ci_status": "available", "output_tokens_mean": 34593.62301587302, "output_tokens_ci_low": 29792.009920634922, "output_tokens_ci_high": 39864.24801587302, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 42.964285714285715, "steps_ci_low": 38.50595238095238, "steps_ci_high": 47.67460317460318, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": 0.04563492063492063, "no_reward_fraction_ci_low": 0.011904761904761904, "no_reward_fraction_ci_high": 0.09126984126984126, "no_reward_fraction_ci_level": 0.95, "no_reward_fraction_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "no_reward_fraction_ci_status": "available"}
{"evaluation_id": "jobbench_easy/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pure-rbs16-n1-ep20-ctx128k/iter-15/attempts8-cap120", "dataset": "jobbench_easy", "eval_split": "easy", "source_kind": "offline_saved_checkpoint", "family": "pure", "method": "pure OPD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pure-rbs16-n1-ep20-ctx128k", "checkpoint_iteration": 15, "checkpoint_folder": "checkpoints/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pure-rbs16-n1-ep20-ctx128k/iter_0000015", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 63, "protocol_episode_count": 504, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": "stage", "score_metric": "normalized_rubric_score", "source_files": ["notes/figures/series/series_jobbenchmt_pure_offline_pass8_cap120.tsv"], "figures": ["jobbench_mtrain_hill_climb"], "source_sha256": {"notes/figures/series/series_jobbenchmt_pure_offline_pass8_cap120.tsv": "0525131a60debee698514a2ddaeecf657914382e5a96d63fea3029a8151f578c"}, "score_mean": 0.3197031746031746, "score_ci_low": 0.2732952380952381, "score_ci_high": 0.36684345238095245, "score_ci_level": 0.95, "score_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "score_ci_status": "available", "output_tokens_mean": 34763.77182539683, "output_tokens_ci_low": 30182.39880952381, "output_tokens_ci_high": 39849.7996031746, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 42.083333333333336, "steps_ci_low": 38.02579365079365, "steps_ci_high": 46.317460317460316, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": 0.047619047619047616, "no_reward_fraction_ci_low": 0.00992063492063492, "no_reward_fraction_ci_high": 0.09722222222222222, "no_reward_fraction_ci_level": 0.95, "no_reward_fraction_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "no_reward_fraction_ci_status": "available"}
{"evaluation_id": "jobbench_easy/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pure-rbs16-n1-ep20-ctx128k/iter-19/attempts8-cap120", "dataset": "jobbench_easy", "eval_split": "easy", "source_kind": "offline_saved_checkpoint", "family": "pure", "method": "pure OPD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pure-rbs16-n1-ep20-ctx128k", "checkpoint_iteration": 19, "checkpoint_folder": "checkpoints/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pure-rbs16-n1-ep20-ctx128k/iter_0000019", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 63, "protocol_episode_count": 504, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": "stage", "score_metric": "normalized_rubric_score", "source_files": ["notes/figures/series/series_jobbenchmt_pure_offline_pass8_cap120.tsv"], "figures": ["jobbench_mtrain_hill_climb"], "source_sha256": {"notes/figures/series/series_jobbenchmt_pure_offline_pass8_cap120.tsv": "0525131a60debee698514a2ddaeecf657914382e5a96d63fea3029a8151f578c"}, "score_mean": 0.3268126984126985, "score_ci_low": 0.2801690476190476, "score_ci_high": 0.37604107142857135, "score_ci_level": 0.95, "score_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "score_ci_status": "available", "output_tokens_mean": 31255.64484126984, "output_tokens_ci_low": 27373.126984126986, "output_tokens_ci_high": 35653.60317460317, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 37.94642857142857, "steps_ci_low": 34.46825396825397, "steps_ci_high": 41.61904761904762, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": 0.047619047619047616, "no_reward_fraction_ci_low": 0.011904761904761904, "no_reward_fraction_ci_high": 0.09722222222222222, "no_reward_fraction_ci_level": 0.95, "no_reward_fraction_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "no_reward_fraction_ci_status": "available"}
{"evaluation_id": "jobbench_easy/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pure-rbs16-n1-ep20-ctx128k/iter-23/attempts8-cap120", "dataset": "jobbench_easy", "eval_split": "easy", "source_kind": "offline_saved_checkpoint", "family": "pure", "method": "pure OPD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pure-rbs16-n1-ep20-ctx128k", "checkpoint_iteration": 23, "checkpoint_folder": "checkpoints/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pure-rbs16-n1-ep20-ctx128k/iter_0000023", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 63, "protocol_episode_count": 504, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": "stage", "score_metric": "normalized_rubric_score", "source_files": ["notes/figures/series/series_jobbenchmt_pure_offline_pass8_cap120.tsv"], "figures": ["jobbench_mtrain_hill_climb"], "source_sha256": {"notes/figures/series/series_jobbenchmt_pure_offline_pass8_cap120.tsv": "0525131a60debee698514a2ddaeecf657914382e5a96d63fea3029a8151f578c"}, "score_mean": 0.3537349206349208, "score_ci_low": 0.3040079365079364, "score_ci_high": 0.4043432539682538, "score_ci_level": 0.95, "score_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "score_ci_status": "available", "output_tokens_mean": 29452.31150793651, "output_tokens_ci_low": 25723.27380952381, "output_tokens_ci_high": 33368.32142857143, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 33.93849206349206, "steps_ci_low": 30.930555555555557, "steps_ci_high": 37.10912698412698, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": 0.03968253968253968, "no_reward_fraction_ci_low": 0.007936507936507936, "no_reward_fraction_ci_high": 0.08531746031746032, "no_reward_fraction_ci_level": 0.95, "no_reward_fraction_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "no_reward_fraction_ci_status": "available"}
{"evaluation_id": "jobbench_easy/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pure-rbs16-n1-ep20-ctx128k/iter-27/attempts8-cap120", "dataset": "jobbench_easy", "eval_split": "easy", "source_kind": "offline_saved_checkpoint", "family": "pure", "method": "pure OPD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pure-rbs16-n1-ep20-ctx128k", "checkpoint_iteration": 27, "checkpoint_folder": "checkpoints/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pure-rbs16-n1-ep20-ctx128k/iter_0000027", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 63, "protocol_episode_count": 504, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": "stage", "score_metric": "normalized_rubric_score", "source_files": ["notes/figures/series/series_jobbenchmt_pure_offline_pass8_cap120.tsv"], "figures": ["jobbench_mtrain_hill_climb"], "source_sha256": {"notes/figures/series/series_jobbenchmt_pure_offline_pass8_cap120.tsv": "0525131a60debee698514a2ddaeecf657914382e5a96d63fea3029a8151f578c"}, "score_mean": 0.343456746031746, "score_ci_low": 0.295772619047619, "score_ci_high": 0.3933420634920634, "score_ci_level": 0.95, "score_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "score_ci_status": "available", "output_tokens_mean": 28088.6626984127, "output_tokens_ci_low": 24565.37896825397, "output_tokens_ci_high": 31858.579365079364, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 33.51984126984127, "steps_ci_low": 30.533730158730158, "steps_ci_high": 36.68055555555556, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": 0.037698412698412696, "no_reward_fraction_ci_low": 0.00992063492063492, "no_reward_fraction_ci_high": 0.07738095238095238, "no_reward_fraction_ci_level": 0.95, "no_reward_fraction_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "no_reward_fraction_ci_status": "available"}
{"evaluation_id": "jobbench_easy/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pure-rbs16-n1-ep20-ctx128k/iter-29/attempts8-cap120", "dataset": "jobbench_easy", "eval_split": "easy", "source_kind": "offline_saved_checkpoint", "family": "pure", "method": "pure OPD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pure-rbs16-n1-ep20-ctx128k", "checkpoint_iteration": 29, "checkpoint_folder": "checkpoints/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pure-rbs16-n1-ep20-ctx128k/iter_0000029", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 63, "protocol_episode_count": 504, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": "stage", "score_metric": "normalized_rubric_score", "source_files": ["notes/figures/series/series_jobbenchmt_pure_offline_pass8_cap120.tsv"], "figures": ["jobbench_mtrain_hill_climb"], "source_sha256": {"notes/figures/series/series_jobbenchmt_pure_offline_pass8_cap120.tsv": "0525131a60debee698514a2ddaeecf657914382e5a96d63fea3029a8151f578c"}, "score_mean": 0.3519837301587301, "score_ci_low": 0.3057640873015874, "score_ci_high": 0.3983206349206348, "score_ci_level": 0.95, "score_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "score_ci_status": "available", "output_tokens_mean": 28071.484126984127, "output_tokens_ci_low": 24977.680555555555, "output_tokens_ci_high": 31386.86111111111, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 35.29365079365079, "steps_ci_low": 32.34722222222222, "steps_ci_high": 38.470238095238095, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": 0.04365079365079365, "no_reward_fraction_ci_low": 0.011904761904761904, "no_reward_fraction_ci_high": 0.0873015873015873, "no_reward_fraction_ci_level": 0.95, "no_reward_fraction_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "no_reward_fraction_ci_status": "available"}
{"evaluation_id": "jobbench_easy/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pure-rbs16-n1-ep20-ctx128k/iter-31/attempts8-cap120", "dataset": "jobbench_easy", "eval_split": "easy", "source_kind": "offline_saved_checkpoint", "family": "pure", "method": "pure OPD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pure-rbs16-n1-ep20-ctx128k", "checkpoint_iteration": 31, "checkpoint_folder": "checkpoints/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pure-rbs16-n1-ep20-ctx128k/iter_0000031", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 63, "protocol_episode_count": 504, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": "stage", "score_metric": "normalized_rubric_score", "source_files": ["notes/figures/series/series_jobbenchmt_pure_offline_pass8_cap120.tsv"], "figures": ["jobbench_mtrain_hill_climb"], "source_sha256": {"notes/figures/series/series_jobbenchmt_pure_offline_pass8_cap120.tsv": "0525131a60debee698514a2ddaeecf657914382e5a96d63fea3029a8151f578c"}, "score_mean": 0.35084027777777776, "score_ci_low": 0.30338551587301593, "score_ci_high": 0.40111845238095245, "score_ci_level": 0.95, "score_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "score_ci_status": "available", "output_tokens_mean": 28605.68650793651, "output_tokens_ci_low": 25195.853174603173, "output_tokens_ci_high": 32219.22619047619, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 36.214285714285715, "steps_ci_low": 32.94642857142857, "steps_ci_high": 39.698412698412696, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": 0.041666666666666664, "no_reward_fraction_ci_low": 0.011904761904761904, "no_reward_fraction_ci_high": 0.08333333333333333, "no_reward_fraction_ci_level": 0.95, "no_reward_fraction_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "no_reward_fraction_ci_status": "available"}
{"evaluation_id": "jobbench_easy/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pure-rbs16-n1-ep20-ctx128k/iter-35/attempts8-cap120", "dataset": "jobbench_easy", "eval_split": "easy", "source_kind": "offline_saved_checkpoint", "family": "pure", "method": "pure OPD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pure-rbs16-n1-ep20-ctx128k", "checkpoint_iteration": 35, "checkpoint_folder": "checkpoints/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pure-rbs16-n1-ep20-ctx128k/iter_0000035", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 63, "protocol_episode_count": 504, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": "stage", "score_metric": "normalized_rubric_score", "source_files": ["notes/figures/series/series_jobbenchmt_pure_offline_pass8_cap120.tsv"], "figures": ["jobbench_mtrain_hill_climb"], "source_sha256": {"notes/figures/series/series_jobbenchmt_pure_offline_pass8_cap120.tsv": "0525131a60debee698514a2ddaeecf657914382e5a96d63fea3029a8151f578c"}, "score_mean": 0.3467051587301587, "score_ci_low": 0.2958099206349206, "score_ci_high": 0.40098373015873023, "score_ci_level": 0.95, "score_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "score_ci_status": "available", "output_tokens_mean": 30027.009920634922, "output_tokens_ci_low": 26648.68650793651, "output_tokens_ci_high": 33563.65476190476, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 37.74404761904762, "steps_ci_low": 34.361111111111114, "steps_ci_high": 41.351190476190474, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": 0.015873015873015872, "no_reward_fraction_ci_low": 0.001984126984126984, "no_reward_fraction_ci_high": 0.03571428571428571, "no_reward_fraction_ci_level": 0.95, "no_reward_fraction_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "no_reward_fraction_ci_status": "available"}
{"evaluation_id": "jobbench_easy/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pure-rbs16-n1-ep20-ctx128k/iter-39/attempts8-cap120", "dataset": "jobbench_easy", "eval_split": "easy", "source_kind": "offline_saved_checkpoint", "family": "pure", "method": "pure OPD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pure-rbs16-n1-ep20-ctx128k", "checkpoint_iteration": 39, "checkpoint_folder": "checkpoints/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pure-rbs16-n1-ep20-ctx128k/iter_0000039", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 63, "protocol_episode_count": 504, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": "stage", "score_metric": "normalized_rubric_score", "source_files": ["notes/figures/series/series_jobbenchmt_pure_offline_pass8_cap120.tsv"], "figures": ["jobbench_mtrain_hill_climb"], "source_sha256": {"notes/figures/series/series_jobbenchmt_pure_offline_pass8_cap120.tsv": "0525131a60debee698514a2ddaeecf657914382e5a96d63fea3029a8151f578c"}, "score_mean": 0.3524222222222223, "score_ci_low": 0.30347281746031757, "score_ci_high": 0.4028142857142857, "score_ci_level": 0.95, "score_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "score_ci_status": "available", "output_tokens_mean": 29008.345238095237, "output_tokens_ci_low": 25515.771825396827, "output_tokens_ci_high": 32653.351190476194, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 35.892857142857146, "steps_ci_low": 32.595238095238095, "steps_ci_high": 39.263888888888886, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": 0.0, "no_reward_fraction_ci_low": 0.0, "no_reward_fraction_ci_high": 0.0, "no_reward_fraction_ci_level": 0.95, "no_reward_fraction_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "no_reward_fraction_ci_status": "available"}
{"evaluation_id": "jobbench_easy/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pure-rbs16-n1-ep20-ctx128k/iter-43/attempts8-cap120", "dataset": "jobbench_easy", "eval_split": "easy", "source_kind": "offline_saved_checkpoint", "family": "pure", "method": "pure OPD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pure-rbs16-n1-ep20-ctx128k", "checkpoint_iteration": 43, "checkpoint_folder": "checkpoints/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pure-rbs16-n1-ep20-ctx128k/iter_0000043", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 63, "protocol_episode_count": 504, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": "stage", "score_metric": "normalized_rubric_score", "source_files": ["notes/figures/series/series_jobbenchmt_pure_offline_pass8_cap120.tsv"], "figures": ["jobbench_mtrain_hill_climb"], "source_sha256": {"notes/figures/series/series_jobbenchmt_pure_offline_pass8_cap120.tsv": "0525131a60debee698514a2ddaeecf657914382e5a96d63fea3029a8151f578c"}, "score_mean": 0.34374563492063487, "score_ci_low": 0.2940162698412698, "score_ci_high": 0.39375277777777795, "score_ci_level": 0.95, "score_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "score_ci_status": "available", "output_tokens_mean": 26887.555555555555, "output_tokens_ci_low": 23697.718253968254, "output_tokens_ci_high": 30409.630952380954, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 34.04563492063492, "steps_ci_low": 31.067460317460316, "steps_ci_high": 37.166666666666664, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": 0.0, "no_reward_fraction_ci_low": 0.0, "no_reward_fraction_ci_high": 0.0, "no_reward_fraction_ci_level": 0.95, "no_reward_fraction_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "no_reward_fraction_ci_status": "available"}
{"evaluation_id": "jobbench_easy/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pure-rbs16-n1-ep20-ctx128k/iter-47/attempts8-cap120", "dataset": "jobbench_easy", "eval_split": "easy", "source_kind": "offline_saved_checkpoint", "family": "pure", "method": "pure OPD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pure-rbs16-n1-ep20-ctx128k", "checkpoint_iteration": 47, "checkpoint_folder": "checkpoints/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pure-rbs16-n1-ep20-ctx128k/iter_0000047", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 63, "protocol_episode_count": 504, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": "stage", "score_metric": "normalized_rubric_score", "source_files": ["notes/figures/series/series_jobbenchmt_pure_offline_pass8_cap120.tsv"], "figures": ["jobbench_mtrain_hill_climb"], "source_sha256": {"notes/figures/series/series_jobbenchmt_pure_offline_pass8_cap120.tsv": "0525131a60debee698514a2ddaeecf657914382e5a96d63fea3029a8151f578c"}, "score_mean": 0.34286031746031753, "score_ci_low": 0.29326111111111125, "score_ci_high": 0.3948236111111111, "score_ci_level": 0.95, "score_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "score_ci_status": "available", "output_tokens_mean": 29597.930555555555, "output_tokens_ci_low": 26120.23015873016, "output_tokens_ci_high": 33277.38690476191, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 36.06547619047619, "steps_ci_low": 32.70436507936508, "steps_ci_high": 39.59325396825397, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": 0.0, "no_reward_fraction_ci_low": 0.0, "no_reward_fraction_ci_high": 0.0, "no_reward_fraction_ci_level": 0.95, "no_reward_fraction_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "no_reward_fraction_ci_status": "available"}
{"evaluation_id": "jobbench_easy/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pure-rbs16-n1-ep20-ctx128k/iter-49/attempts8-cap120", "dataset": "jobbench_easy", "eval_split": "easy", "source_kind": "offline_saved_checkpoint", "family": "pure", "method": "pure OPD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pure-rbs16-n1-ep20-ctx128k", "checkpoint_iteration": 49, "checkpoint_folder": "checkpoints/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pure-rbs16-n1-ep20-ctx128k/iter_0000049", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 63, "protocol_episode_count": 504, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": "stage", "score_metric": "normalized_rubric_score", "source_files": ["notes/figures/series/series_jobbenchmt_pure_offline_pass8_cap120.tsv"], "figures": ["jobbench_mtrain_hill_climb"], "source_sha256": {"notes/figures/series/series_jobbenchmt_pure_offline_pass8_cap120.tsv": "0525131a60debee698514a2ddaeecf657914382e5a96d63fea3029a8151f578c"}, "score_mean": 0.34481884920634914, "score_ci_low": 0.2965037698412698, "score_ci_high": 0.39519285714285723, "score_ci_level": 0.95, "score_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "score_ci_status": "available", "output_tokens_mean": 29306.92261904762, "output_tokens_ci_low": 26045.178571428572, "output_tokens_ci_high": 32938.90873015873, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 37.1468253968254, "steps_ci_low": 33.90873015873016, "steps_ci_high": 40.625, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": 0.0, "no_reward_fraction_ci_low": 0.0, "no_reward_fraction_ci_high": 0.0, "no_reward_fraction_ci_level": 0.95, "no_reward_fraction_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "no_reward_fraction_ci_status": "available"}
{"evaluation_id": "jobbench_easy/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pure-rbs16-n1-ep20-ctx128k/iter-51/attempts8-cap120", "dataset": "jobbench_easy", "eval_split": "easy", "source_kind": "offline_saved_checkpoint", "family": "pure", "method": "pure OPD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pure-rbs16-n1-ep20-ctx128k", "checkpoint_iteration": 51, "checkpoint_folder": "checkpoints/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pure-rbs16-n1-ep20-ctx128k/iter_0000051", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 63, "protocol_episode_count": 504, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": "stage", "score_metric": "normalized_rubric_score", "source_files": ["notes/figures/series/series_jobbenchmt_pure_offline_pass8_cap120.tsv"], "figures": ["jobbench_mtrain_hill_climb"], "source_sha256": {"notes/figures/series/series_jobbenchmt_pure_offline_pass8_cap120.tsv": "0525131a60debee698514a2ddaeecf657914382e5a96d63fea3029a8151f578c"}, "score_mean": 0.3512392857142857, "score_ci_low": 0.3009128968253968, "score_ci_high": 0.4037085317460316, "score_ci_level": 0.95, "score_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "score_ci_status": "available", "output_tokens_mean": 28387.18849206349, "output_tokens_ci_low": 25397.27380952381, "output_tokens_ci_high": 31583.428571428572, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 36.232142857142854, "steps_ci_low": 33.154761904761905, "steps_ci_high": 39.458333333333336, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": 0.0, "no_reward_fraction_ci_low": 0.0, "no_reward_fraction_ci_high": 0.0, "no_reward_fraction_ci_level": 0.95, "no_reward_fraction_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "no_reward_fraction_ci_status": "available"}
{"evaluation_id": "jobbench_easy/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pure-rbs16-n1-ep20-ctx128k/iter-55/attempts8-cap120", "dataset": "jobbench_easy", "eval_split": "easy", "source_kind": "offline_saved_checkpoint", "family": "pure", "method": "pure OPD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pure-rbs16-n1-ep20-ctx128k", "checkpoint_iteration": 55, "checkpoint_folder": "checkpoints/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pure-rbs16-n1-ep20-ctx128k/iter_0000055", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 63, "protocol_episode_count": 504, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": "stage", "score_metric": "normalized_rubric_score", "source_files": ["notes/figures/series/series_jobbenchmt_pure_offline_pass8_cap120.tsv"], "figures": ["jobbench_mtrain_hill_climb"], "source_sha256": {"notes/figures/series/series_jobbenchmt_pure_offline_pass8_cap120.tsv": "0525131a60debee698514a2ddaeecf657914382e5a96d63fea3029a8151f578c"}, "score_mean": 0.33604742063492077, "score_ci_low": 0.28665753968253965, "score_ci_high": 0.38602976190476196, "score_ci_level": 0.95, "score_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "score_ci_status": "available", "output_tokens_mean": 29818.222222222223, "output_tokens_ci_low": 26625.39484126984, "output_tokens_ci_high": 33094.005952380954, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 36.3968253968254, "steps_ci_low": 33.21230158730159, "steps_ci_high": 39.79365079365079, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": 0.0, "no_reward_fraction_ci_low": 0.0, "no_reward_fraction_ci_high": 0.0, "no_reward_fraction_ci_level": 0.95, "no_reward_fraction_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "no_reward_fraction_ci_status": "available"}
{"evaluation_id": "jobbench_easy/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pure-rbs16-n1-ep20-ctx128k/iter-59/attempts8-cap120", "dataset": "jobbench_easy", "eval_split": "easy", "source_kind": "offline_saved_checkpoint", "family": "pure", "method": "pure OPD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pure-rbs16-n1-ep20-ctx128k", "checkpoint_iteration": 59, "checkpoint_folder": "checkpoints/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pure-rbs16-n1-ep20-ctx128k/iter_0000059", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 63, "protocol_episode_count": 504, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": "stage", "score_metric": "normalized_rubric_score", "source_files": ["notes/figures/series/series_jobbenchmt_pure_offline_pass8_cap120.tsv"], "figures": ["jobbench_mtrain_hill_climb"], "source_sha256": {"notes/figures/series/series_jobbenchmt_pure_offline_pass8_cap120.tsv": "0525131a60debee698514a2ddaeecf657914382e5a96d63fea3029a8151f578c"}, "score_mean": 0.34689960317460317, "score_ci_low": 0.2963759920634921, "score_ci_high": 0.3989623015873018, "score_ci_level": 0.95, "score_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "score_ci_status": "available", "output_tokens_mean": 28142.799603174604, "output_tokens_ci_low": 25021.291666666668, "output_tokens_ci_high": 31521.630952380954, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 35.029761904761905, "steps_ci_low": 31.982142857142858, "steps_ci_high": 38.19444444444444, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": 0.0, "no_reward_fraction_ci_low": 0.0, "no_reward_fraction_ci_high": 0.0, "no_reward_fraction_ci_level": 0.95, "no_reward_fraction_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "no_reward_fraction_ci_status": "available"}
{"evaluation_id": "jobbench_easy/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pure-rbs16-n1-ep20-ctx128k/iter-63/attempts8-cap120", "dataset": "jobbench_easy", "eval_split": "easy", "source_kind": "offline_saved_checkpoint", "family": "pure", "method": "pure OPD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pure-rbs16-n1-ep20-ctx128k", "checkpoint_iteration": 63, "checkpoint_folder": "checkpoints/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pure-rbs16-n1-ep20-ctx128k/iter_0000063", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 63, "protocol_episode_count": 504, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": "stage", "score_metric": "normalized_rubric_score", "source_files": ["notes/figures/series/series_jobbenchmt_pure_offline_pass8_cap120.tsv"], "figures": ["jobbench_mtrain_hill_climb"], "source_sha256": {"notes/figures/series/series_jobbenchmt_pure_offline_pass8_cap120.tsv": "0525131a60debee698514a2ddaeecf657914382e5a96d63fea3029a8151f578c"}, "score_mean": 0.353296626984127, "score_ci_low": 0.3035984126984127, "score_ci_high": 0.4049966269841268, "score_ci_level": 0.95, "score_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "score_ci_status": "available", "output_tokens_mean": 26293.09126984127, "output_tokens_ci_low": 23397.48214285714, "output_tokens_ci_high": 29314.16865079365, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 33.18452380952381, "steps_ci_low": 30.142857142857142, "steps_ci_high": 36.37301587301587, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": 0.0, "no_reward_fraction_ci_low": 0.0, "no_reward_fraction_ci_high": 0.0, "no_reward_fraction_ci_level": 0.95, "no_reward_fraction_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "no_reward_fraction_ci_status": "available"}
{"evaluation_id": "jobbench_easy/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pure-rbs16-n1-ep20-ctx128k/iter-67/attempts8-cap120", "dataset": "jobbench_easy", "eval_split": "easy", "source_kind": "offline_saved_checkpoint", "family": "pure", "method": "pure OPD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pure-rbs16-n1-ep20-ctx128k", "checkpoint_iteration": 67, "checkpoint_folder": "checkpoints/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pure-rbs16-n1-ep20-ctx128k/iter_0000067", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 63, "protocol_episode_count": 504, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": "stage", "score_metric": "normalized_rubric_score", "source_files": ["notes/figures/series/series_jobbenchmt_pure_offline_pass8_cap120.tsv"], "figures": ["jobbench_mtrain_hill_climb"], "source_sha256": {"notes/figures/series/series_jobbenchmt_pure_offline_pass8_cap120.tsv": "0525131a60debee698514a2ddaeecf657914382e5a96d63fea3029a8151f578c"}, "score_mean": 0.3580357142857144, "score_ci_low": 0.3094228174603174, "score_ci_high": 0.40982063492063503, "score_ci_level": 0.95, "score_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "score_ci_status": "available", "output_tokens_mean": 27330.92261904762, "output_tokens_ci_low": 24432.097222222223, "output_tokens_ci_high": 30462.698412698413, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 33.904761904761905, "steps_ci_low": 30.89484126984127, "steps_ci_high": 37.176587301587304, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": 0.0, "no_reward_fraction_ci_low": 0.0, "no_reward_fraction_ci_high": 0.0, "no_reward_fraction_ci_level": 0.95, "no_reward_fraction_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "no_reward_fraction_ci_status": "available"}
{"evaluation_id": "jobbench_easy/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pure-rbs16-n1-ep20-ctx128k/iter-69/attempts8-cap120", "dataset": "jobbench_easy", "eval_split": "easy", "source_kind": "offline_saved_checkpoint", "family": "pure", "method": "pure OPD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pure-rbs16-n1-ep20-ctx128k", "checkpoint_iteration": 69, "checkpoint_folder": "checkpoints/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pure-rbs16-n1-ep20-ctx128k/iter_0000069", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 63, "protocol_episode_count": 504, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": "stage", "score_metric": "normalized_rubric_score", "source_files": ["notes/figures/series/series_jobbenchmt_pure_offline_pass8_cap120.tsv"], "figures": ["jobbench_mtrain_hill_climb"], "source_sha256": {"notes/figures/series/series_jobbenchmt_pure_offline_pass8_cap120.tsv": "0525131a60debee698514a2ddaeecf657914382e5a96d63fea3029a8151f578c"}, "score_mean": 0.35556250000000006, "score_ci_low": 0.30699365079365076, "score_ci_high": 0.4057200396825398, "score_ci_level": 0.95, "score_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "score_ci_status": "available", "output_tokens_mean": 28918.384920634922, "output_tokens_ci_low": 25759.50396825397, "output_tokens_ci_high": 32350.17261904762, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 34.84325396825397, "steps_ci_low": 31.71825396825397, "steps_ci_high": 38.04563492063492, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": 0.0, "no_reward_fraction_ci_low": 0.0, "no_reward_fraction_ci_high": 0.0, "no_reward_fraction_ci_level": 0.95, "no_reward_fraction_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "no_reward_fraction_ci_status": "available"}
{"evaluation_id": "jobbench_easy/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pure-rbs16-n1-ep20-ctx128k/iter-71/attempts8-cap120", "dataset": "jobbench_easy", "eval_split": "easy", "source_kind": "offline_saved_checkpoint", "family": "pure", "method": "pure OPD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pure-rbs16-n1-ep20-ctx128k", "checkpoint_iteration": 71, "checkpoint_folder": "checkpoints/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pure-rbs16-n1-ep20-ctx128k/iter_0000071", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 63, "protocol_episode_count": 504, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": "stage", "score_metric": "normalized_rubric_score", "source_files": ["notes/figures/series/series_jobbenchmt_pure_offline_pass8_cap120.tsv"], "figures": ["jobbench_mtrain_hill_climb"], "source_sha256": {"notes/figures/series/series_jobbenchmt_pure_offline_pass8_cap120.tsv": "0525131a60debee698514a2ddaeecf657914382e5a96d63fea3029a8151f578c"}, "score_mean": 0.35922162698412696, "score_ci_low": 0.3082367063492063, "score_ci_high": 0.4108621031746033, "score_ci_level": 0.95, "score_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "score_ci_status": "available", "output_tokens_mean": 28642.541666666668, "output_tokens_ci_low": 25558.23611111111, "output_tokens_ci_high": 31837.31349206349, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 34.442460317460316, "steps_ci_low": 31.466269841269842, "steps_ci_high": 37.60515873015873, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": 0.0, "no_reward_fraction_ci_low": 0.0, "no_reward_fraction_ci_high": 0.0, "no_reward_fraction_ci_level": 0.95, "no_reward_fraction_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "no_reward_fraction_ci_status": "available"}
{"evaluation_id": "jobbench_easy/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pure-rbs16-n1-ep20-ctx128k/iter-75/attempts8-cap120", "dataset": "jobbench_easy", "eval_split": "easy", "source_kind": "offline_saved_checkpoint", "family": "pure", "method": "pure OPD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pure-rbs16-n1-ep20-ctx128k", "checkpoint_iteration": 75, "checkpoint_folder": "checkpoints/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pure-rbs16-n1-ep20-ctx128k/iter_0000075", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 63, "protocol_episode_count": 504, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": "stage", "score_metric": "normalized_rubric_score", "source_files": ["notes/figures/series/series_jobbenchmt_pure_offline_pass8_cap120.tsv"], "figures": ["jobbench_mtrain_hill_climb"], "source_sha256": {"notes/figures/series/series_jobbenchmt_pure_offline_pass8_cap120.tsv": "0525131a60debee698514a2ddaeecf657914382e5a96d63fea3029a8151f578c"}, "score_mean": 0.3547906746031744, "score_ci_low": 0.303069246031746, "score_ci_high": 0.40717460317460324, "score_ci_level": 0.95, "score_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "score_ci_status": "available", "output_tokens_mean": 29052.988095238095, "output_tokens_ci_low": 25853.319444444445, "output_tokens_ci_high": 32509.74007936508, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 34.36507936507937, "steps_ci_low": 31.517857142857142, "steps_ci_high": 37.388888888888886, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": 0.0, "no_reward_fraction_ci_low": 0.0, "no_reward_fraction_ci_high": 0.0, "no_reward_fraction_ci_level": 0.95, "no_reward_fraction_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "no_reward_fraction_ci_status": "available"}
{"evaluation_id": "jobbench_easy/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pure-rbs16-n1-ep20-ctx128k/iter-79/attempts8-cap120", "dataset": "jobbench_easy", "eval_split": "easy", "source_kind": "offline_saved_checkpoint", "family": "pure", "method": "pure OPD", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pure-rbs16-n1-ep20-ctx128k", "checkpoint_iteration": 79, "checkpoint_folder": "checkpoints/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pure-rbs16-n1-ep20-ctx128k/iter_0000079", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 63, "protocol_episode_count": 504, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": "stage", "score_metric": "normalized_rubric_score", "source_files": ["notes/figures/series/series_jobbenchmt_pure_offline_pass8_cap120.tsv"], "figures": ["jobbench_mtrain_hill_climb"], "source_sha256": {"notes/figures/series/series_jobbenchmt_pure_offline_pass8_cap120.tsv": "0525131a60debee698514a2ddaeecf657914382e5a96d63fea3029a8151f578c"}, "score_mean": 0.35431091269841264, "score_ci_low": 0.30297797619047606, "score_ci_high": 0.4056565476190475, "score_ci_level": 0.95, "score_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "score_ci_status": "available", "output_tokens_mean": 27716.654761904763, "output_tokens_ci_low": 24622.15873015873, "output_tokens_ci_high": 30987.05753968254, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 33.81547619047619, "steps_ci_low": 31.04563492063492, "steps_ci_high": 36.654761904761905, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": 0.0, "no_reward_fraction_ci_low": 0.0, "no_reward_fraction_ci_high": 0.0, "no_reward_fraction_ci_level": 0.95, "no_reward_fraction_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "no_reward_fraction_ci_status": "available"}
{"evaluation_id": "jobbench_easy/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep20-ctx128k-mf050/iter-3/attempts8-cap120", "dataset": "jobbench_easy", "eval_split": "easy", "source_kind": "offline_saved_checkpoint", "family": "aod2optail", "method": "AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep20-ctx128k-mf050", "checkpoint_iteration": 3, "checkpoint_folder": "checkpoints/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep20-ctx128k-mf050/iter_0000003", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 63, "protocol_episode_count": 504, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": "stage", "score_metric": "normalized_rubric_score", "source_files": ["notes/figures/series/series_jobbenchmt_aod2optail_offline_pass8_cap120.tsv"], "figures": ["jobbench_mtrain_hill_climb"], "source_sha256": {"notes/figures/series/series_jobbenchmt_aod2optail_offline_pass8_cap120.tsv": "9fb7a2c8de1fa2a340091612fdcdcfd362961e42b883e880536fbbd2c9da41a0"}, "score_mean": 0.30568571428571434, "score_ci_low": 0.25759305555555556, "score_ci_high": 0.355981746031746, "score_ci_level": 0.95, "score_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "score_ci_status": "available", "output_tokens_mean": 30182.529761904763, "output_tokens_ci_low": 26424.06150793651, "output_tokens_ci_high": 34455.25396825397, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 42.68055555555556, "steps_ci_low": 38.28769841269841, "steps_ci_high": 47.45436507936508, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": 0.04563492063492063, "no_reward_fraction_ci_low": 0.00992063492063492, "no_reward_fraction_ci_high": 0.09126984126984126, "no_reward_fraction_ci_level": 0.95, "no_reward_fraction_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "no_reward_fraction_ci_status": "available"}
{"evaluation_id": "jobbench_easy/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep20-ctx128k-mf050/iter-7/attempts8-cap120", "dataset": "jobbench_easy", "eval_split": "easy", "source_kind": "offline_saved_checkpoint", "family": "aod2optail", "method": "AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep20-ctx128k-mf050", "checkpoint_iteration": 7, "checkpoint_folder": "checkpoints/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep20-ctx128k-mf050/iter_0000007", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 63, "protocol_episode_count": 504, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": "stage", "score_metric": "normalized_rubric_score", "source_files": ["notes/figures/series/series_jobbenchmt_aod2optail_offline_pass8_cap120.tsv"], "figures": ["jobbench_mtrain_hill_climb"], "source_sha256": {"notes/figures/series/series_jobbenchmt_aod2optail_offline_pass8_cap120.tsv": "9fb7a2c8de1fa2a340091612fdcdcfd362961e42b883e880536fbbd2c9da41a0"}, "score_mean": 0.31669384920634913, "score_ci_low": 0.2692150793650793, "score_ci_high": 0.3645634920634922, "score_ci_level": 0.95, "score_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "score_ci_status": "available", "output_tokens_mean": 29021.75, "output_tokens_ci_low": 25444.119047619046, "output_tokens_ci_high": 32896.52380952381, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 41.94444444444444, "steps_ci_low": 37.791666666666664, "steps_ci_high": 46.30357142857143, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": 0.041666666666666664, "no_reward_fraction_ci_low": 0.005952380952380952, "no_reward_fraction_ci_high": 0.0873015873015873, "no_reward_fraction_ci_level": 0.95, "no_reward_fraction_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "no_reward_fraction_ci_status": "available"}
{"evaluation_id": "jobbench_easy/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep20-ctx128k-mf050/iter-9/attempts8-cap120", "dataset": "jobbench_easy", "eval_split": "easy", "source_kind": "offline_saved_checkpoint", "family": "aod2optail", "method": "AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep20-ctx128k-mf050", "checkpoint_iteration": 9, "checkpoint_folder": "checkpoints/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep20-ctx128k-mf050/iter_0000009", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 63, "protocol_episode_count": 504, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": "stage", "score_metric": "normalized_rubric_score", "source_files": ["notes/figures/series/series_jobbenchmt_aod2optail_offline_pass8_cap120.tsv"], "figures": ["jobbench_mtrain_hill_climb"], "source_sha256": {"notes/figures/series/series_jobbenchmt_aod2optail_offline_pass8_cap120.tsv": "9fb7a2c8de1fa2a340091612fdcdcfd362961e42b883e880536fbbd2c9da41a0"}, "score_mean": 0.31044265873015886, "score_ci_low": 0.2653617063492064, "score_ci_high": 0.35788928571428597, "score_ci_level": 0.95, "score_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "score_ci_status": "available", "output_tokens_mean": 32214.454365079364, "output_tokens_ci_low": 28085.51785714286, "output_tokens_ci_high": 36578.24007936508, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 44.47222222222222, "steps_ci_low": 40.38095238095238, "steps_ci_high": 48.79761904761905, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": 0.047619047619047616, "no_reward_fraction_ci_low": 0.011904761904761904, "no_reward_fraction_ci_high": 0.09325396825396826, "no_reward_fraction_ci_level": 0.95, "no_reward_fraction_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "no_reward_fraction_ci_status": "available"}
{"evaluation_id": "jobbench_easy/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep20-ctx128k-mf050/iter-11/attempts8-cap120", "dataset": "jobbench_easy", "eval_split": "easy", "source_kind": "offline_saved_checkpoint", "family": "aod2optail", "method": "AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep20-ctx128k-mf050", "checkpoint_iteration": 11, "checkpoint_folder": "checkpoints/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep20-ctx128k-mf050/iter_0000011", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 63, "protocol_episode_count": 504, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": "stage", "score_metric": "normalized_rubric_score", "source_files": ["notes/figures/series/series_jobbenchmt_aod2optail_offline_pass8_cap120.tsv"], "figures": ["jobbench_mtrain_hill_climb"], "source_sha256": {"notes/figures/series/series_jobbenchmt_aod2optail_offline_pass8_cap120.tsv": "9fb7a2c8de1fa2a340091612fdcdcfd362961e42b883e880536fbbd2c9da41a0"}, "score_mean": 0.30186527777777794, "score_ci_low": 0.25343849206349206, "score_ci_high": 0.3540345238095239, "score_ci_level": 0.95, "score_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "score_ci_status": "available", "output_tokens_mean": 33514.843253968254, "output_tokens_ci_low": 28935.680555555555, "output_tokens_ci_high": 38681.57142857143, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 43.80357142857143, "steps_ci_low": 39.476190476190474, "steps_ci_high": 48.333333333333336, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": 0.05357142857142857, "no_reward_fraction_ci_low": 0.013888888888888888, "no_reward_fraction_ci_high": 0.10317460317460317, "no_reward_fraction_ci_level": 0.95, "no_reward_fraction_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "no_reward_fraction_ci_status": "available"}
{"evaluation_id": "jobbench_easy/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep20-ctx128k-mf050/iter-15/attempts8-cap120", "dataset": "jobbench_easy", "eval_split": "easy", "source_kind": "offline_saved_checkpoint", "family": "aod2optail", "method": "AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep20-ctx128k-mf050", "checkpoint_iteration": 15, "checkpoint_folder": "checkpoints/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep20-ctx128k-mf050/iter_0000015", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 63, "protocol_episode_count": 504, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": "stage", "score_metric": "normalized_rubric_score", "source_files": ["notes/figures/series/series_jobbenchmt_aod2optail_offline_pass8_cap120.tsv"], "figures": ["jobbench_mtrain_hill_climb"], "source_sha256": {"notes/figures/series/series_jobbenchmt_aod2optail_offline_pass8_cap120.tsv": "9fb7a2c8de1fa2a340091612fdcdcfd362961e42b883e880536fbbd2c9da41a0"}, "score_mean": 0.3165478174603175, "score_ci_low": 0.2688918650793651, "score_ci_high": 0.3647996031746032, "score_ci_level": 0.95, "score_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "score_ci_status": "available", "output_tokens_mean": 31302.39880952381, "output_tokens_ci_low": 26825.718253968254, "output_tokens_ci_high": 36491.07142857143, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 40.56547619047619, "steps_ci_low": 36.91468253968254, "steps_ci_high": 44.40079365079365, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": 0.047619047619047616, "no_reward_fraction_ci_low": 0.013888888888888888, "no_reward_fraction_ci_high": 0.09126984126984126, "no_reward_fraction_ci_level": 0.95, "no_reward_fraction_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "no_reward_fraction_ci_status": "available"}
{"evaluation_id": "jobbench_easy/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep20-ctx128k-mf050/iter-19/attempts8-cap120", "dataset": "jobbench_easy", "eval_split": "easy", "source_kind": "offline_saved_checkpoint", "family": "aod2optail", "method": "AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep20-ctx128k-mf050", "checkpoint_iteration": 19, "checkpoint_folder": "checkpoints/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep20-ctx128k-mf050/iter_0000019", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 63, "protocol_episode_count": 504, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": "stage", "score_metric": "normalized_rubric_score", "source_files": ["notes/figures/series/series_jobbenchmt_aod2optail_offline_pass8_cap120.tsv"], "figures": ["jobbench_mtrain_hill_climb"], "source_sha256": {"notes/figures/series/series_jobbenchmt_aod2optail_offline_pass8_cap120.tsv": "9fb7a2c8de1fa2a340091612fdcdcfd362961e42b883e880536fbbd2c9da41a0"}, "score_mean": 0.336939880952381, "score_ci_low": 0.2890170634920635, "score_ci_high": 0.3870128968253969, "score_ci_level": 0.95, "score_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "score_ci_status": "available", "output_tokens_mean": 30108.170634920636, "output_tokens_ci_low": 25988.07738095238, "output_tokens_ci_high": 34282.09722222222, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 37.573412698412696, "steps_ci_low": 34.18849206349206, "steps_ci_high": 41.232142857142854, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": 0.05555555555555555, "no_reward_fraction_ci_low": 0.017857142857142856, "no_reward_fraction_ci_high": 0.10515873015873016, "no_reward_fraction_ci_level": 0.95, "no_reward_fraction_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "no_reward_fraction_ci_status": "available"}
{"evaluation_id": "jobbench_easy/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep20-ctx128k-mf050/iter-23/attempts8-cap120", "dataset": "jobbench_easy", "eval_split": "easy", "source_kind": "offline_saved_checkpoint", "family": "aod2optail", "method": "AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep20-ctx128k-mf050", "checkpoint_iteration": 23, "checkpoint_folder": "checkpoints/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep20-ctx128k-mf050/iter_0000023", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 63, "protocol_episode_count": 504, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": "stage", "score_metric": "normalized_rubric_score", "source_files": ["notes/figures/series/series_jobbenchmt_aod2optail_offline_pass8_cap120.tsv"], "figures": ["jobbench_mtrain_hill_climb"], "source_sha256": {"notes/figures/series/series_jobbenchmt_aod2optail_offline_pass8_cap120.tsv": "9fb7a2c8de1fa2a340091612fdcdcfd362961e42b883e880536fbbd2c9da41a0"}, "score_mean": 0.33915337301587306, "score_ci_low": 0.28954265873015866, "score_ci_high": 0.3884841269841268, "score_ci_level": 0.95, "score_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "score_ci_status": "available", "output_tokens_mean": 27100.164682539682, "output_tokens_ci_low": 23730.79761904762, "output_tokens_ci_high": 30749.51984126984, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 34.71031746031746, "steps_ci_low": 31.751984126984127, "steps_ci_high": 37.90079365079365, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": 0.03373015873015873, "no_reward_fraction_ci_low": 0.003968253968253968, "no_reward_fraction_ci_high": 0.07539682539682539, "no_reward_fraction_ci_level": 0.95, "no_reward_fraction_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "no_reward_fraction_ci_status": "available"}
{"evaluation_id": "jobbench_easy/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep20-ctx128k-mf050/iter-27/attempts8-cap120", "dataset": "jobbench_easy", "eval_split": "easy", "source_kind": "offline_saved_checkpoint", "family": "aod2optail", "method": "AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep20-ctx128k-mf050", "checkpoint_iteration": 27, "checkpoint_folder": "checkpoints/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep20-ctx128k-mf050/iter_0000027", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 63, "protocol_episode_count": 504, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": "stage", "score_metric": "normalized_rubric_score", "source_files": ["notes/figures/series/series_jobbenchmt_aod2optail_offline_pass8_cap120.tsv"], "figures": ["jobbench_mtrain_hill_climb"], "source_sha256": {"notes/figures/series/series_jobbenchmt_aod2optail_offline_pass8_cap120.tsv": "9fb7a2c8de1fa2a340091612fdcdcfd362961e42b883e880536fbbd2c9da41a0"}, "score_mean": 0.33394563492063495, "score_ci_low": 0.2857053571428571, "score_ci_high": 0.38146765873015875, "score_ci_level": 0.95, "score_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "score_ci_status": "available", "output_tokens_mean": 23550.61111111111, "output_tokens_ci_low": 20713.20238095238, "output_tokens_ci_high": 26435.946428571428, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 33.632936507936506, "steps_ci_low": 30.865079365079364, "steps_ci_high": 36.59920634920635, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": 0.037698412698412696, "no_reward_fraction_ci_low": 0.00992063492063492, "no_reward_fraction_ci_high": 0.07738095238095238, "no_reward_fraction_ci_level": 0.95, "no_reward_fraction_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "no_reward_fraction_ci_status": "available"}
{"evaluation_id": "jobbench_easy/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep20-ctx128k-mf050/iter-29/attempts8-cap120", "dataset": "jobbench_easy", "eval_split": "easy", "source_kind": "offline_saved_checkpoint", "family": "aod2optail", "method": "AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep20-ctx128k-mf050", "checkpoint_iteration": 29, "checkpoint_folder": "checkpoints/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep20-ctx128k-mf050/iter_0000029", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 63, "protocol_episode_count": 504, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": "stage", "score_metric": "normalized_rubric_score", "source_files": ["notes/figures/series/series_jobbenchmt_aod2optail_offline_pass8_cap120.tsv"], "figures": ["jobbench_mtrain_hill_climb"], "source_sha256": {"notes/figures/series/series_jobbenchmt_aod2optail_offline_pass8_cap120.tsv": "9fb7a2c8de1fa2a340091612fdcdcfd362961e42b883e880536fbbd2c9da41a0"}, "score_mean": 0.34104087301587316, "score_ci_low": 0.2933063492063492, "score_ci_high": 0.38974246031746035, "score_ci_level": 0.95, "score_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "score_ci_status": "available", "output_tokens_mean": 23231.117063492064, "output_tokens_ci_low": 20672.37896825397, "output_tokens_ci_high": 25859.992063492064, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 34.43849206349206, "steps_ci_low": 31.599206349206348, "steps_ci_high": 37.37896825396825, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": 0.041666666666666664, "no_reward_fraction_ci_low": 0.011904761904761904, "no_reward_fraction_ci_high": 0.08333333333333333, "no_reward_fraction_ci_level": 0.95, "no_reward_fraction_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "no_reward_fraction_ci_status": "available"}
{"evaluation_id": "jobbench_easy/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep20-ctx128k-mf050/iter-31/attempts8-cap120", "dataset": "jobbench_easy", "eval_split": "easy", "source_kind": "offline_saved_checkpoint", "family": "aod2optail", "method": "AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep20-ctx128k-mf050", "checkpoint_iteration": 31, "checkpoint_folder": "checkpoints/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep20-ctx128k-mf050/iter_0000031", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 63, "protocol_episode_count": 504, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": "stage", "score_metric": "normalized_rubric_score", "source_files": ["notes/figures/series/series_jobbenchmt_aod2optail_offline_pass8_cap120.tsv"], "figures": ["jobbench_mtrain_hill_climb"], "source_sha256": {"notes/figures/series/series_jobbenchmt_aod2optail_offline_pass8_cap120.tsv": "9fb7a2c8de1fa2a340091612fdcdcfd362961e42b883e880536fbbd2c9da41a0"}, "score_mean": 0.34010952380952386, "score_ci_low": 0.2919011904761903, "score_ci_high": 0.3901188492063492, "score_ci_level": 0.95, "score_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "score_ci_status": "available", "output_tokens_mean": 24232.700396825396, "output_tokens_ci_low": 21307.51785714286, "output_tokens_ci_high": 27347.36111111111, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 35.851190476190474, "steps_ci_low": 33.007936507936506, "steps_ci_high": 38.95436507936508, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": 0.041666666666666664, "no_reward_fraction_ci_low": 0.00992063492063492, "no_reward_fraction_ci_high": 0.0873015873015873, "no_reward_fraction_ci_level": 0.95, "no_reward_fraction_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "no_reward_fraction_ci_status": "available"}
{"evaluation_id": "jobbench_easy/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep20-ctx128k-mf050/iter-35/attempts8-cap120", "dataset": "jobbench_easy", "eval_split": "easy", "source_kind": "offline_saved_checkpoint", "family": "aod2optail", "method": "AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep20-ctx128k-mf050", "checkpoint_iteration": 35, "checkpoint_folder": "checkpoints/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep20-ctx128k-mf050/iter_0000035", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 63, "protocol_episode_count": 504, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": "stage", "score_metric": "normalized_rubric_score", "source_files": ["notes/figures/series/series_jobbenchmt_aod2optail_offline_pass8_cap120.tsv"], "figures": ["jobbench_mtrain_hill_climb"], "source_sha256": {"notes/figures/series/series_jobbenchmt_aod2optail_offline_pass8_cap120.tsv": "9fb7a2c8de1fa2a340091612fdcdcfd362961e42b883e880536fbbd2c9da41a0"}, "score_mean": 0.33567996031746034, "score_ci_low": 0.28907619047619043, "score_ci_high": 0.38282043650793646, "score_ci_level": 0.95, "score_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "score_ci_status": "available", "output_tokens_mean": 25682.39285714286, "output_tokens_ci_low": 22590.2123015873, "output_tokens_ci_high": 28940.847222222223, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 37.48015873015873, "steps_ci_low": 34.285714285714285, "steps_ci_high": 40.77579365079365, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": 0.04563492063492063, "no_reward_fraction_ci_low": 0.007936507936507936, "no_reward_fraction_ci_high": 0.09325396825396826, "no_reward_fraction_ci_level": 0.95, "no_reward_fraction_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "no_reward_fraction_ci_status": "available"}
{"evaluation_id": "jobbench_easy/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep20-ctx128k-mf050/iter-39/attempts8-cap120", "dataset": "jobbench_easy", "eval_split": "easy", "source_kind": "offline_saved_checkpoint", "family": "aod2optail", "method": "AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep20-ctx128k-mf050", "checkpoint_iteration": 39, "checkpoint_folder": "checkpoints/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep20-ctx128k-mf050/iter_0000039", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 63, "protocol_episode_count": 504, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": "stage", "score_metric": "normalized_rubric_score", "source_files": ["notes/figures/series/series_jobbenchmt_aod2optail_offline_pass8_cap120.tsv"], "figures": ["jobbench_mtrain_hill_climb"], "source_sha256": {"notes/figures/series/series_jobbenchmt_aod2optail_offline_pass8_cap120.tsv": "9fb7a2c8de1fa2a340091612fdcdcfd362961e42b883e880536fbbd2c9da41a0"}, "score_mean": 0.33475714285714286, "score_ci_low": 0.2853982142857142, "score_ci_high": 0.3854079365079366, "score_ci_level": 0.95, "score_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "score_ci_status": "available", "output_tokens_mean": 25704.34126984127, "output_tokens_ci_low": 22866.4126984127, "output_tokens_ci_high": 28586.075396825396, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 35.279761904761905, "steps_ci_low": 32.09722222222222, "steps_ci_high": 38.79365079365079, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": 0.0, "no_reward_fraction_ci_low": 0.0, "no_reward_fraction_ci_high": 0.0, "no_reward_fraction_ci_level": 0.95, "no_reward_fraction_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "no_reward_fraction_ci_status": "available"}
{"evaluation_id": "jobbench_easy/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep20-ctx128k-mf050/iter-43/attempts8-cap120", "dataset": "jobbench_easy", "eval_split": "easy", "source_kind": "offline_saved_checkpoint", "family": "aod2optail", "method": "AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep20-ctx128k-mf050", "checkpoint_iteration": 43, "checkpoint_folder": "checkpoints/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep20-ctx128k-mf050/iter_0000043", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 63, "protocol_episode_count": 504, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": "stage", "score_metric": "normalized_rubric_score", "source_files": ["notes/figures/series/series_jobbenchmt_aod2optail_offline_pass8_cap120.tsv"], "figures": ["jobbench_mtrain_hill_climb"], "source_sha256": {"notes/figures/series/series_jobbenchmt_aod2optail_offline_pass8_cap120.tsv": "9fb7a2c8de1fa2a340091612fdcdcfd362961e42b883e880536fbbd2c9da41a0"}, "score_mean": 0.3269682539682541, "score_ci_low": 0.2795210317460317, "score_ci_high": 0.37700257936507936, "score_ci_level": 0.95, "score_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "score_ci_status": "available", "output_tokens_mean": 26307.406746031746, "output_tokens_ci_low": 22781.103174603173, "output_tokens_ci_high": 30329.345238095237, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 34.17063492063492, "steps_ci_low": 30.96825396825397, "steps_ci_high": 37.61507936507937, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": 0.0, "no_reward_fraction_ci_low": 0.0, "no_reward_fraction_ci_high": 0.0, "no_reward_fraction_ci_level": 0.95, "no_reward_fraction_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "no_reward_fraction_ci_status": "available"}
{"evaluation_id": "jobbench_easy/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep20-ctx128k-mf050/iter-47/attempts8-cap120", "dataset": "jobbench_easy", "eval_split": "easy", "source_kind": "offline_saved_checkpoint", "family": "aod2optail", "method": "AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep20-ctx128k-mf050", "checkpoint_iteration": 47, "checkpoint_folder": "checkpoints/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep20-ctx128k-mf050/iter_0000047", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 63, "protocol_episode_count": 504, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": "stage", "score_metric": "normalized_rubric_score", "source_files": ["notes/figures/series/series_jobbenchmt_aod2optail_offline_pass8_cap120.tsv"], "figures": ["jobbench_mtrain_hill_climb"], "source_sha256": {"notes/figures/series/series_jobbenchmt_aod2optail_offline_pass8_cap120.tsv": "9fb7a2c8de1fa2a340091612fdcdcfd362961e42b883e880536fbbd2c9da41a0"}, "score_mean": 0.3360609126984128, "score_ci_low": 0.2862640873015873, "score_ci_high": 0.388121626984127, "score_ci_level": 0.95, "score_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "score_ci_status": "available", "output_tokens_mean": 24075.285714285714, "output_tokens_ci_low": 21317.021825396827, "output_tokens_ci_high": 27030.299603174604, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 32.90079365079365, "steps_ci_low": 30.069444444444443, "steps_ci_high": 35.882936507936506, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": 0.0, "no_reward_fraction_ci_low": 0.0, "no_reward_fraction_ci_high": 0.0, "no_reward_fraction_ci_level": 0.95, "no_reward_fraction_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "no_reward_fraction_ci_status": "available"}
{"evaluation_id": "jobbench_easy/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep20-ctx128k-mf050/iter-49/attempts8-cap120", "dataset": "jobbench_easy", "eval_split": "easy", "source_kind": "offline_saved_checkpoint", "family": "aod2optail", "method": "AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep20-ctx128k-mf050", "checkpoint_iteration": 49, "checkpoint_folder": "checkpoints/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep20-ctx128k-mf050/iter_0000049", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 63, "protocol_episode_count": 504, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": "stage", "score_metric": "normalized_rubric_score", "source_files": ["notes/figures/series/series_jobbenchmt_aod2optail_offline_pass8_cap120.tsv"], "figures": ["jobbench_mtrain_hill_climb"], "source_sha256": {"notes/figures/series/series_jobbenchmt_aod2optail_offline_pass8_cap120.tsv": "9fb7a2c8de1fa2a340091612fdcdcfd362961e42b883e880536fbbd2c9da41a0"}, "score_mean": 0.3396242063492064, "score_ci_low": 0.2896232142857144, "score_ci_high": 0.39054980158730174, "score_ci_level": 0.95, "score_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "score_ci_status": "available", "output_tokens_mean": 23543.134920634922, "output_tokens_ci_low": 20727.73214285714, "output_tokens_ci_high": 26653.277777777777, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 33.45039682539682, "steps_ci_low": 30.51388888888889, "steps_ci_high": 36.69047619047619, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": 0.0, "no_reward_fraction_ci_low": 0.0, "no_reward_fraction_ci_high": 0.0, "no_reward_fraction_ci_level": 0.95, "no_reward_fraction_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "no_reward_fraction_ci_status": "available"}
{"evaluation_id": "jobbench_easy/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep20-ctx128k-mf050/iter-51/attempts8-cap120", "dataset": "jobbench_easy", "eval_split": "easy", "source_kind": "offline_saved_checkpoint", "family": "aod2optail", "method": "AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep20-ctx128k-mf050", "checkpoint_iteration": 51, "checkpoint_folder": "checkpoints/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep20-ctx128k-mf050/iter_0000051", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 63, "protocol_episode_count": 504, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": "stage", "score_metric": "normalized_rubric_score", "source_files": ["notes/figures/series/series_jobbenchmt_aod2optail_offline_pass8_cap120.tsv"], "figures": ["jobbench_mtrain_hill_climb"], "source_sha256": {"notes/figures/series/series_jobbenchmt_aod2optail_offline_pass8_cap120.tsv": "9fb7a2c8de1fa2a340091612fdcdcfd362961e42b883e880536fbbd2c9da41a0"}, "score_mean": 0.348130753968254, "score_ci_low": 0.2975204365079365, "score_ci_high": 0.40116527777777766, "score_ci_level": 0.95, "score_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "score_ci_status": "available", "output_tokens_mean": 24021.083333333332, "output_tokens_ci_low": 20852.81349206349, "output_tokens_ci_high": 27471.626984126986, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 33.90277777777778, "steps_ci_low": 30.9265873015873, "steps_ci_high": 37.07142857142857, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": 0.0, "no_reward_fraction_ci_low": 0.0, "no_reward_fraction_ci_high": 0.0, "no_reward_fraction_ci_level": 0.95, "no_reward_fraction_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "no_reward_fraction_ci_status": "available"}
{"evaluation_id": "jobbench_easy/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep20-ctx128k-mf050/iter-55/attempts8-cap120", "dataset": "jobbench_easy", "eval_split": "easy", "source_kind": "offline_saved_checkpoint", "family": "aod2optail", "method": "AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep20-ctx128k-mf050", "checkpoint_iteration": 55, "checkpoint_folder": "checkpoints/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep20-ctx128k-mf050/iter_0000055", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 63, "protocol_episode_count": 504, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": "stage", "score_metric": "normalized_rubric_score", "source_files": ["notes/figures/series/series_jobbenchmt_aod2optail_offline_pass8_cap120.tsv"], "figures": ["jobbench_mtrain_hill_climb"], "source_sha256": {"notes/figures/series/series_jobbenchmt_aod2optail_offline_pass8_cap120.tsv": "9fb7a2c8de1fa2a340091612fdcdcfd362961e42b883e880536fbbd2c9da41a0"}, "score_mean": 0.34471289682539674, "score_ci_low": 0.297481746031746, "score_ci_high": 0.39391924603174605, "score_ci_level": 0.95, "score_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "score_ci_status": "available", "output_tokens_mean": 25499.244047619046, "output_tokens_ci_low": 22458.434523809523, "output_tokens_ci_high": 28772.58134920635, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 35.64087301587302, "steps_ci_low": 32.56349206349206, "steps_ci_high": 38.8968253968254, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": 0.0, "no_reward_fraction_ci_low": 0.0, "no_reward_fraction_ci_high": 0.0, "no_reward_fraction_ci_level": 0.95, "no_reward_fraction_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "no_reward_fraction_ci_status": "available"}
{"evaluation_id": "jobbench_easy/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep20-ctx128k-mf050/iter-59/attempts8-cap120", "dataset": "jobbench_easy", "eval_split": "easy", "source_kind": "offline_saved_checkpoint", "family": "aod2optail", "method": "AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep20-ctx128k-mf050", "checkpoint_iteration": 59, "checkpoint_folder": "checkpoints/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep20-ctx128k-mf050/iter_0000059", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 63, "protocol_episode_count": 504, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": "stage", "score_metric": "normalized_rubric_score", "source_files": ["notes/figures/series/series_jobbenchmt_aod2optail_offline_pass8_cap120.tsv"], "figures": ["jobbench_mtrain_hill_climb"], "source_sha256": {"notes/figures/series/series_jobbenchmt_aod2optail_offline_pass8_cap120.tsv": "9fb7a2c8de1fa2a340091612fdcdcfd362961e42b883e880536fbbd2c9da41a0"}, "score_mean": 0.3419418650793652, "score_ci_low": 0.2931113095238096, "score_ci_high": 0.39238511904761914, "score_ci_level": 0.95, "score_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "score_ci_status": "available", "output_tokens_mean": 26097.16865079365, "output_tokens_ci_low": 22812.140873015873, "output_tokens_ci_high": 29730.765873015873, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 34.46230158730159, "steps_ci_low": 31.511904761904763, "steps_ci_high": 37.67460317460318, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": 0.0, "no_reward_fraction_ci_low": 0.0, "no_reward_fraction_ci_high": 0.0, "no_reward_fraction_ci_level": 0.95, "no_reward_fraction_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "no_reward_fraction_ci_status": "available"}
{"evaluation_id": "jobbench_easy/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep20-ctx128k-mf050/iter-63/attempts8-cap120", "dataset": "jobbench_easy", "eval_split": "easy", "source_kind": "offline_saved_checkpoint", "family": "aod2optail", "method": "AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep20-ctx128k-mf050", "checkpoint_iteration": 63, "checkpoint_folder": "checkpoints/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep20-ctx128k-mf050/iter_0000063", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 63, "protocol_episode_count": 504, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": "stage", "score_metric": "normalized_rubric_score", "source_files": ["notes/figures/series/series_jobbenchmt_aod2optail_offline_pass8_cap120.tsv"], "figures": ["jobbench_mtrain_hill_climb"], "source_sha256": {"notes/figures/series/series_jobbenchmt_aod2optail_offline_pass8_cap120.tsv": "9fb7a2c8de1fa2a340091612fdcdcfd362961e42b883e880536fbbd2c9da41a0"}, "score_mean": 0.3449103174603174, "score_ci_low": 0.29672023809523795, "score_ci_high": 0.39599999999999985, "score_ci_level": 0.95, "score_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "score_ci_status": "available", "output_tokens_mean": 25371.50396825397, "output_tokens_ci_low": 22465.801587301587, "output_tokens_ci_high": 28257.009920634922, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 33.607142857142854, "steps_ci_low": 31.001984126984127, "steps_ci_high": 36.351190476190474, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": 0.0, "no_reward_fraction_ci_low": 0.0, "no_reward_fraction_ci_high": 0.0, "no_reward_fraction_ci_level": 0.95, "no_reward_fraction_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "no_reward_fraction_ci_status": "available"}
{"evaluation_id": "jobbench_easy/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep20-ctx128k-mf050/iter-67/attempts8-cap120", "dataset": "jobbench_easy", "eval_split": "easy", "source_kind": "offline_saved_checkpoint", "family": "aod2optail", "method": "AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep20-ctx128k-mf050", "checkpoint_iteration": 67, "checkpoint_folder": "checkpoints/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep20-ctx128k-mf050/iter_0000067", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 63, "protocol_episode_count": 504, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": "stage", "score_metric": "normalized_rubric_score", "source_files": ["notes/figures/series/series_jobbenchmt_aod2optail_offline_pass8_cap120.tsv"], "figures": ["jobbench_mtrain_hill_climb"], "source_sha256": {"notes/figures/series/series_jobbenchmt_aod2optail_offline_pass8_cap120.tsv": "9fb7a2c8de1fa2a340091612fdcdcfd362961e42b883e880536fbbd2c9da41a0"}, "score_mean": 0.3568863095238094, "score_ci_low": 0.3067478174603174, "score_ci_high": 0.407822222222222, "score_ci_level": 0.95, "score_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "score_ci_status": "available", "output_tokens_mean": 24112.5873015873, "output_tokens_ci_low": 21221.91865079365, "output_tokens_ci_high": 27211.08134920635, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 33.48809523809524, "steps_ci_low": 30.803571428571427, "steps_ci_high": 36.279761904761905, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": 0.0, "no_reward_fraction_ci_low": 0.0, "no_reward_fraction_ci_high": 0.0, "no_reward_fraction_ci_level": 0.95, "no_reward_fraction_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "no_reward_fraction_ci_status": "available"}
{"evaluation_id": "jobbench_easy/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep20-ctx128k-mf050/iter-69/attempts8-cap120", "dataset": "jobbench_easy", "eval_split": "easy", "source_kind": "offline_saved_checkpoint", "family": "aod2optail", "method": "AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep20-ctx128k-mf050", "checkpoint_iteration": 69, "checkpoint_folder": "checkpoints/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep20-ctx128k-mf050/iter_0000069", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 63, "protocol_episode_count": 504, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": "stage", "score_metric": "normalized_rubric_score", "source_files": ["notes/figures/series/series_jobbenchmt_aod2optail_offline_pass8_cap120.tsv"], "figures": ["jobbench_mtrain_hill_climb"], "source_sha256": {"notes/figures/series/series_jobbenchmt_aod2optail_offline_pass8_cap120.tsv": "9fb7a2c8de1fa2a340091612fdcdcfd362961e42b883e880536fbbd2c9da41a0"}, "score_mean": 0.3511109126984129, "score_ci_low": 0.2982142857142857, "score_ci_high": 0.40376845238095244, "score_ci_level": 0.95, "score_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "score_ci_status": "available", "output_tokens_mean": 23887.45238095238, "output_tokens_ci_low": 21074.9126984127, "output_tokens_ci_high": 26947.777777777777, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 34.101190476190474, "steps_ci_low": 31.311507936507937, "steps_ci_high": 37.07936507936508, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": 0.0, "no_reward_fraction_ci_low": 0.0, "no_reward_fraction_ci_high": 0.0, "no_reward_fraction_ci_level": 0.95, "no_reward_fraction_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "no_reward_fraction_ci_status": "available"}
{"evaluation_id": "jobbench_easy/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep20-ctx128k-mf050/iter-71/attempts8-cap120", "dataset": "jobbench_easy", "eval_split": "easy", "source_kind": "offline_saved_checkpoint", "family": "aod2optail", "method": "AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep20-ctx128k-mf050", "checkpoint_iteration": 71, "checkpoint_folder": "checkpoints/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep20-ctx128k-mf050/iter_0000071", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 63, "protocol_episode_count": 504, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": "stage", "score_metric": "normalized_rubric_score", "source_files": ["notes/figures/series/series_jobbenchmt_aod2optail_offline_pass8_cap120.tsv"], "figures": ["jobbench_mtrain_hill_climb"], "source_sha256": {"notes/figures/series/series_jobbenchmt_aod2optail_offline_pass8_cap120.tsv": "9fb7a2c8de1fa2a340091612fdcdcfd362961e42b883e880536fbbd2c9da41a0"}, "score_mean": 0.3514496031746032, "score_ci_low": 0.30038650793650795, "score_ci_high": 0.40378194444444454, "score_ci_level": 0.95, "score_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "score_ci_status": "available", "output_tokens_mean": 23077.130952380954, "output_tokens_ci_low": 20520.751984126986, "output_tokens_ci_high": 25826.625, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 34.14484126984127, "steps_ci_low": 31.353174603174605, "steps_ci_high": 37.20238095238095, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": 0.0, "no_reward_fraction_ci_low": 0.0, "no_reward_fraction_ci_high": 0.0, "no_reward_fraction_ci_level": 0.95, "no_reward_fraction_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "no_reward_fraction_ci_status": "available"}
{"evaluation_id": "jobbench_easy/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep20-ctx128k-mf050/iter-75/attempts8-cap120", "dataset": "jobbench_easy", "eval_split": "easy", "source_kind": "offline_saved_checkpoint", "family": "aod2optail", "method": "AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep20-ctx128k-mf050", "checkpoint_iteration": 75, "checkpoint_folder": "checkpoints/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep20-ctx128k-mf050/iter_0000075", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 63, "protocol_episode_count": 504, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": "stage", "score_metric": "normalized_rubric_score", "source_files": ["notes/figures/series/series_jobbenchmt_aod2optail_offline_pass8_cap120.tsv"], "figures": ["jobbench_mtrain_hill_climb"], "source_sha256": {"notes/figures/series/series_jobbenchmt_aod2optail_offline_pass8_cap120.tsv": "9fb7a2c8de1fa2a340091612fdcdcfd362961e42b883e880536fbbd2c9da41a0"}, "score_mean": 0.3496940476190476, "score_ci_low": 0.3006654761904762, "score_ci_high": 0.4002537698412698, "score_ci_level": 0.95, "score_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "score_ci_status": "available", "output_tokens_mean": 23048.373015873014, "output_tokens_ci_low": 20531.0873015873, "output_tokens_ci_high": 25634.095238095237, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 33.31150793650794, "steps_ci_low": 30.700396825396826, "steps_ci_high": 36.03968253968254, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": 0.0, "no_reward_fraction_ci_low": 0.0, "no_reward_fraction_ci_high": 0.0, "no_reward_fraction_ci_level": 0.95, "no_reward_fraction_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "no_reward_fraction_ci_status": "available"}
{"evaluation_id": "jobbench_easy/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep20-ctx128k-mf050/iter-79/attempts8-cap120", "dataset": "jobbench_easy", "eval_split": "easy", "source_kind": "offline_saved_checkpoint", "family": "aod2optail", "method": "AOD on-policy tail", "model": "Qwen3.5-4B", "teacher": "Qwen3.5-397B", "training_run": "qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep20-ctx128k-mf050", "checkpoint_iteration": 79, "checkpoint_folder": "checkpoints/qwen3.5-4B-jobbench-mtrain-sfstage-opdkl1.0-pt-aod2op-tail-pure-rbs16-n1-ep20-ctx128k-mf050/iter_0000079", "temperature": 1.0, "step_cap": 120, "attempts_per_task": 8, "protocol_task_count": 63, "protocol_episode_count": 504, "counts_basis": "declared plotting protocol; not a fresh rollout census", "search_file_policy": "stage", "score_metric": "normalized_rubric_score", "source_files": ["notes/figures/series/series_jobbenchmt_aod2optail_offline_pass8_cap120.tsv"], "figures": ["jobbench_mtrain_hill_climb"], "source_sha256": {"notes/figures/series/series_jobbenchmt_aod2optail_offline_pass8_cap120.tsv": "9fb7a2c8de1fa2a340091612fdcdcfd362961e42b883e880536fbbd2c9da41a0"}, "score_mean": 0.3550285714285714, "score_ci_low": 0.3042323412698412, "score_ci_high": 0.40637896825396824, "score_ci_level": 0.95, "score_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "score_ci_status": "available", "output_tokens_mean": 24147.597222222223, "output_tokens_ci_low": 21464.303571428572, "output_tokens_ci_high": 26987.575396825396, "output_tokens_ci_level": 0.95, "output_tokens_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "output_tokens_ci_status": "available", "steps_mean": 33.75595238095238, "steps_ci_low": 31.003968253968253, "steps_ci_high": 36.638888888888886, "steps_ci_level": 0.95, "steps_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "steps_ci_status": "available", "no_reward_fraction_mean": 0.0, "no_reward_fraction_ci_low": 0.0, "no_reward_fraction_ci_high": 0.0, "no_reward_fraction_ci_level": 0.95, "no_reward_fraction_ci_method": "task-cluster percentile bootstrap, 10000 draws, 95%", "no_reward_fraction_ci_status": "available"}