{
  "schema_version": "ironbench.release-site-data.v1",
  "episode": {
    "source": {
      "path": "runs/revised-v1-confirmatory-v8-k4/rl-offset-000500.json",
      "sha256": "cffc2509183e163dc74980c2b6b2dd7311060c55d6f0b72aa61aeded32b61d72",
      "checkpoint": "tinker://4dc55fff-d194-5bba-aae2-c9fb73a50554:train:0/sampler_weights/final",
      "temperature": 0.75
    },
    "task_id": "proc-design-20261003-34792-117410",
    "trajectory_index": 0,
    "total_reward": 1.1606760764321935,
    "spec": {
      "spec_id": "proc-design-20261003-34792-117410",
      "kva": 2500.0,
      "hv_kv_ll": 13.8,
      "lv_kv_ll": 4.16,
      "hv_bil_kv": 95.0,
      "lv_bil_kv": 60.0,
      "z_pct_target": 4.743431697125419,
      "max_no_load_loss_w": 2794.1953164090005,
      "max_load_loss_w": 12327.492953117759,
      "phases": 3,
      "freq_hz": 60.0,
      "connection": "Dyn",
      "z_pct_tol_of_value": 7.5,
      "ratio_tol_pct": 0.5,
      "max_avg_winding_rise_k": 93.7809579362155,
      "max_top_oil_rise_k": 97.93471924969283,
      "a_usd_per_w_no_load": 5.0,
      "b_usd_per_w_load": 1.5
    },
    "turns": [
      {
        "turn": 1,
        "kind": "call",
        "compiler": {
          "b_target_t": 1.3,
          "ducts": 1,
          "extra_gap_mm": 6.0,
          "grade": "M4",
          "hv_layers": 8,
          "j_target_a_mm2": 1.8,
          "lv_layers": 2,
          "margin_profile": "standard",
          "vpt_factor": 0.48
        },
        "changed": [],
        "result": {
          "compliant": false,
          "calls_remaining": 3,
          "constraint_pass_fraction": 0.90322581,
          "failed": [
            {
              "name": "impedance",
              "value": 5.65655129,
              "unit": "%",
              "lower_limit": 4.38767432,
              "upper_limit": 5.09918907,
              "normalized_margin": -1.74450457
            },
            {
              "name": "no_load_loss",
              "value": 2914.30924978,
              "unit": "W",
              "lower_limit": null,
              "upper_limit": 2794.19531641,
              "normalized_margin": -0.10035123
            },
            {
              "name": "leakage_numerics",
              "value": 0.0112205,
              "unit": "fraction",
              "lower_limit": null,
              "upper_limit": null,
              "normalized_margin": null
            }
          ],
          "tightest": [
            "impedance",
            "no_load_loss",
            "axial_fit_in_window",
            "tank_height_fit"
          ],
          "recommendation": {
            "changes": {
              "grade": "M3"
            },
            "predicted_compliant": false,
            "predicted_curriculum_progress": 0.69166164,
            "predicted_worst_constraint": "impedance"
          },
          "instruction": "The local recommendation improves one neighborhood direction but is explicitly predicted to remain noncompliant. Do not follow it blindly: change multiple coupled variables together using all failed and tight constraint margins, and do not revisit a prior design.",
          "metrics": {
            "no_load_loss_w": 2914.30924978,
            "load_loss_w": 11526.58876332,
            "flux_density_t": 1.3,
            "capex_usd": 53202.01878834,
            "max_temperature_rise_k": 73.54450902,
            "efficiency_100pct": 0.99425682
          }
        }
      },
      {
        "turn": 2,
        "kind": "call",
        "compiler": {
          "b_target_t": 1.2,
          "ducts": 1,
          "extra_gap_mm": 6.0,
          "grade": "M3",
          "hv_layers": 8,
          "j_target_a_mm2": 1.8,
          "lv_layers": 2,
          "margin_profile": "standard",
          "vpt_factor": 0.48
        },
        "changed": [
          "b_target_t",
          "grade"
        ],
        "result": {
          "compliant": false,
          "calls_remaining": 2,
          "constraint_pass_fraction": 0.90322581,
          "failed": [
            {
              "name": "impedance",
              "value": 5.79372897,
              "unit": "%",
              "lower_limit": 4.38767432,
              "upper_limit": 5.09918907,
              "normalized_margin": -2.13441035
            },
            {
              "name": "load_loss",
              "value": 11803.28879409,
              "unit": "W",
              "lower_limit": null,
              "upper_limit": 12327.49295312,
              "normalized_margin": -0.00535066
            },
            {
              "name": "leakage_numerics",
              "value": 0.0112205,
              "unit": "fraction",
              "lower_limit": null,
              "upper_limit": null,
              "normalized_margin": null
            }
          ],
          "tightest": [
            "impedance",
            "load_loss",
            "axial_fit_in_window",
            "tank_height_fit"
          ],
          "recommendation": {
            "changes": {
              "ducts": 0
            },
            "predicted_compliant": false,
            "predicted_curriculum_progress": 0.69137907,
            "predicted_worst_constraint": "impedance"
          },
          "instruction": "The local recommendation improves one neighborhood direction but is explicitly predicted to remain noncompliant. Do not follow it blindly: change multiple coupled variables together using all failed and tight constraint margins, and do not revisit a prior design.",
          "metrics": {
            "no_load_loss_w": 2450.85554368,
            "load_loss_w": 11803.28879409,
            "flux_density_t": 1.2,
            "capex_usd": 57789.94527937,
            "max_temperature_rise_k": 76.38460737,
            "efficiency_100pct": 0.99433067
          }
        }
      },
      {
        "turn": 3,
        "kind": "call",
        "compiler": {
          "b_target_t": 1.45,
          "ducts": 1,
          "extra_gap_mm": 6.0,
          "grade": "M3",
          "hv_layers": 8,
          "j_target_a_mm2": 1.8,
          "lv_layers": 2,
          "margin_profile": "standard",
          "vpt_factor": 0.48
        },
        "changed": [
          "b_target_t"
        ],
        "result": {
          "compliant": false,
          "calls_remaining": 1,
          "constraint_pass_fraction": 0.90322581,
          "failed": [
            {
              "name": "impedance",
              "value": 5.43646377,
              "unit": "%",
              "lower_limit": 4.38767432,
              "upper_limit": 5.09918907,
              "normalized_margin": -1.11894069
            },
            {
              "name": "no_load_loss",
              "value": 2882.98598172,
              "unit": "W",
              "lower_limit": null,
              "upper_limit": 2794.19531641,
              "normalized_margin": -0.08852455
            },
            {
              "name": "leakage_numerics",
              "value": 0.0112205,
              "unit": "fraction",
              "lower_limit": null,
              "upper_limit": null,
              "normalized_margin": null
            }
          ],
          "tightest": [
            "impedance",
            "no_load_loss",
            "axial_fit_in_window",
            "tank_height_fit"
          ],
          "recommendation": {
            "changes": {
              "hv_layers": 10
            },
            "predicted_compliant": true,
            "predicted_curriculum_progress": 0.70822632,
            "predicted_worst_constraint": "axial_fit_in_window"
          },
          "instruction": "The local search predicts that recommended_compiler_change is compliant. Apply it, verify once, then submit the verified parameter set.",
          "metrics": {
            "no_load_loss_w": 2882.98598172,
            "load_loss_w": 11082.52909545,
            "flux_density_t": 1.45,
            "capex_usd": 50824.0063349,
            "max_temperature_rise_k": 74.31151824,
            "efficiency_100pct": 0.99444483
          }
        }
      },
      {
        "turn": 4,
        "kind": "call",
        "compiler": {
          "b_target_t": 1.45,
          "ducts": 1,
          "extra_gap_mm": 6.0,
          "grade": "M3",
          "hv_layers": 10,
          "j_target_a_mm2": 1.8,
          "lv_layers": 2,
          "margin_profile": "standard",
          "vpt_factor": 0.48
        },
        "changed": [
          "hv_layers"
        ],
        "result": {
          "compliant": true,
          "calls_remaining": 0,
          "constraint_pass_fraction": 1.0,
          "failed": [],
          "tightest": [
            "axial_fit_in_window",
            "tank_length_fit",
            "tank_height_fit",
            "no_load_loss"
          ],
          "recommendation": null,
          "instruction": "This design is compliant. Submit it now, or use a remaining call to lower candidate TOC while preserving margins.",
          "metrics": {
            "no_load_loss_w": 2627.22930854,
            "load_loss_w": 11183.1621533,
            "flux_density_t": 1.45,
            "capex_usd": 48938.50434219,
            "max_temperature_rise_k": 77.65507036,
            "efficiency_100pct": 0.99450619
          }
        }
      },
      {
        "turn": 5,
        "kind": "submit",
        "submission": {
          "b_target_t": 1.45,
          "ducts": 1,
          "extra_gap_mm": 6.0,
          "grade": "M3",
          "hv_layers": 10,
          "j_target_a_mm2": 1.8,
          "lv_layers": 2,
          "margin_profile": "standard",
          "vpt_factor": 0.48
        },
        "reward": 1.1606760764321935
      }
    ]
  },
  "sources": {
    "results": {
      "path": "data/rl/revised-run-results-v1.json",
      "sha256": "0f60d505c1a2e0b8e4caf69cc3dad29066c42961c25975513ef30673bd72bab9",
      "reported_at": "2026-08-30"
    },
    "rescore": {
      "path": "data/solver/rl-policy-detailed-rescore-audit.json",
      "sha256": "caa3c3569418d81ac5c1f16bba5b351c06c758b72438a4ff923a406b9234293c"
    },
    "search": {
      "path": "data/rl/test-time-search-results.json",
      "sha256": "3345dfa8ac2269de8d25afa2ea3e366828cd5306f25a5c868433cee7e813178c"
    },
    "exploratory": {
      "path": "data/rl/20mva-gsu-exploratory-policy-result.json",
      "sha256": "a31626331bac941b1bd5753d05f75b596f5208f25e39ebfd5334026c4ece213f"
    },
    "exploratory_audit": {
      "path": "data/solver/20mva-gsu-detailed-audit.json",
      "sha256": "c604364f9cca732c61e062b2cccd8a4ff9fe7ed7d747b57853c1c37551626fa9"
    },
    "inference": {
      "path": "data/inference/kimi-k26-modal-benchmark.json",
      "sha256": "4c687816da9b5677b436c857ccda23d4e6e82294256ef4abdcfa97fc2dbb08a3"
    }
  },
  "checkpoints": {
    "pre_rl_sampler": "tinker://4a34f4c2-5335-5c03-b5ff-79696b99d84f:train:0/sampler_weights/final",
    "coordinated_repair_sft_sampler": "tinker://c04b376a-aeb2-5192-b0c5-0b23e76fcd36:train:0/sampler_weights/final",
    "verified_trajectory_rft_sampler": "tinker://71cd5acf-b7c3-5d83-998b-58a89fca7fcd:train:0/sampler_weights/final",
    "selected_rl_state": "tinker://4dc55fff-d194-5bba-aae2-c9fb73a50554:train:0/weights/final",
    "selected_rl_sampler": "tinker://4dc55fff-d194-5bba-aae2-c9fb73a50554:train:0/sampler_weights/final"
  },
  "objective": {
    "held_out_rl_improvement_confirmed": true,
    "strong_sim_v1_policy_demonstrated": true,
    "physical_or_factory_ready_claim": false,
    "preregistered_composite_confirmation_flag": false,
    "composite_flag_note": "The aggregate mean reward among positive samples missed by 0.0000456082417735; a frozen task-paired diagnostic shows significantly better compliant reward on shared solved tasks. The original flag is preserved without post-hoc alteration."
  },
  "data_separation": {
    "training_tasks": 1800,
    "development_tasks": 600,
    "confirmatory_tasks": 900,
    "all_pairwise_task_id_intersections": 0
  },
  "integrity": {
    "training_contains_detected_file_descriptor_errors": false,
    "development_invalid_shards_resampled": 8,
    "development_final_infrastructure_errors": 0,
    "confirmatory_infrastructure_errors": 0,
    "open_file_soft_limit": 8192,
    "evaluation_shard_failure_policy": "reject and resample; never score host failures as policy failures"
  },
  "training": {
    "updates": 50,
    "rollouts_per_update": 256,
    "total_rollouts": 12800,
    "first_10_mean_hard_compliance_rate": 0.566796875,
    "last_10_mean_hard_compliance_rate": 0.71875,
    "hard_compliance_rate_delta": 0.15195312500000002,
    "first_10_mean_hard_reward": 0.5755215793261788,
    "last_10_mean_hard_reward": 0.7515880157531095,
    "mean_hard_reward_delta": 0.17606643642693076,
    "passes_training_gate": true,
    "curve": [
      {
        "update": 1,
        "compliance": 0.5,
        "mean_reward": 0.5055307591310338,
        "compliant": 128,
        "trajectories": 256,
        "pass_at_4": 0.8125,
        "solved_tasks": 52,
        "tasks": 64,
        "summary_sha256": "a348da2a5df49df4de1bc3272d0d34c2bc61a12552cb8b3641bf937523c3237b"
      },
      {
        "update": 2,
        "compliance": 0.4296875,
        "mean_reward": 0.4495346305371187,
        "compliant": 110,
        "trajectories": 256,
        "pass_at_4": 0.734375,
        "solved_tasks": 47,
        "tasks": 64,
        "summary_sha256": "c90a92baed21740f957c4670e7075f90d5150bf04ce3eeb58da8df2b60e768f4"
      },
      {
        "update": 3,
        "compliance": 0.49609375,
        "mean_reward": 0.509763540903148,
        "compliant": 127,
        "trajectories": 256,
        "pass_at_4": 0.859375,
        "solved_tasks": 55,
        "tasks": 64,
        "summary_sha256": "f77ff8860f288271eabd19d7fa9db40618b17a9dfdb5164c3d35b291a5be540f"
      },
      {
        "update": 4,
        "compliance": 0.61328125,
        "mean_reward": 0.627506261495524,
        "compliant": 157,
        "trajectories": 256,
        "pass_at_4": 0.859375,
        "solved_tasks": 55,
        "tasks": 64,
        "summary_sha256": "a19334de2a4ec533b26638678201a88aa2a9ed9cfb272c9abf785c2a9892ba63"
      },
      {
        "update": 5,
        "compliance": 0.57421875,
        "mean_reward": 0.5853475474411609,
        "compliant": 147,
        "trajectories": 256,
        "pass_at_4": 0.8125,
        "solved_tasks": 52,
        "tasks": 64,
        "summary_sha256": "0eac0e54655e23246480810de5a1348554005412ffb3394fd8508adf006bbd30"
      },
      {
        "update": 6,
        "compliance": 0.49609375,
        "mean_reward": 0.5076670971195446,
        "compliant": 127,
        "trajectories": 256,
        "pass_at_4": 0.796875,
        "solved_tasks": 51,
        "tasks": 64,
        "summary_sha256": "caf64dcd2568bab20711fc876da73d107fa4e0c29828d666e5f679bfed14803f"
      },
      {
        "update": 7,
        "compliance": 0.6171875,
        "mean_reward": 0.6248828362935968,
        "compliant": 158,
        "trajectories": 256,
        "pass_at_4": 0.796875,
        "solved_tasks": 51,
        "tasks": 64,
        "summary_sha256": "1cb219379529d448a1a7fd14475806051d06c184d376aabe904cebe9f6bd4de2"
      },
      {
        "update": 8,
        "compliance": 0.578125,
        "mean_reward": 0.573169654202195,
        "compliant": 148,
        "trajectories": 256,
        "pass_at_4": 0.828125,
        "solved_tasks": 53,
        "tasks": 64,
        "summary_sha256": "5a421a48f0f0dc122fc2a5c893a8e56ca354a47c6b52ef772c80e726411d6d54"
      },
      {
        "update": 9,
        "compliance": 0.72265625,
        "mean_reward": 0.7293620990487657,
        "compliant": 185,
        "trajectories": 256,
        "pass_at_4": 0.890625,
        "solved_tasks": 57,
        "tasks": 64,
        "summary_sha256": "7a96d34bee7d234f0b082f7444f11c3ea6b738569f6b2eaedbc1b8099f5bb560"
      },
      {
        "update": 10,
        "compliance": 0.640625,
        "mean_reward": 0.6424513670896997,
        "compliant": 164,
        "trajectories": 256,
        "pass_at_4": 0.796875,
        "solved_tasks": 51,
        "tasks": 64,
        "summary_sha256": "3c5d14d57cfc02e2f6674f882921378c7e9f33e4abc03c089cb824b597cd90c0"
      },
      {
        "update": 11,
        "compliance": 0.73046875,
        "mean_reward": 0.750794074755495,
        "compliant": 187,
        "trajectories": 256,
        "pass_at_4": 0.921875,
        "solved_tasks": 59,
        "tasks": 64,
        "summary_sha256": "da0dfa455bce68f0ce3af7d37333005a83fbd0125e5e0b03f1dd2cabd0bd98a9"
      },
      {
        "update": 12,
        "compliance": 0.69140625,
        "mean_reward": 0.6922441328043978,
        "compliant": 177,
        "trajectories": 256,
        "pass_at_4": 0.859375,
        "solved_tasks": 55,
        "tasks": 64,
        "summary_sha256": "e1ceab901ffdfaa347ce23de06ea55c0763b9f7408c7fe2118720d80b2d3fc8a"
      },
      {
        "update": 13,
        "compliance": 0.72265625,
        "mean_reward": 0.7372993969441416,
        "compliant": 185,
        "trajectories": 256,
        "pass_at_4": 0.90625,
        "solved_tasks": 58,
        "tasks": 64,
        "summary_sha256": "57a9083def4fff6c858dfedcd6678e5a1d07652bf2266c906aaea14de2b23032"
      },
      {
        "update": 14,
        "compliance": 0.6171875,
        "mean_reward": 0.6412383465764443,
        "compliant": 158,
        "trajectories": 256,
        "pass_at_4": 0.828125,
        "solved_tasks": 53,
        "tasks": 64,
        "summary_sha256": "94dea57d1ce32622c903757be77cfc577d1d5f1ff5d68367d972ae0b489ecac6"
      },
      {
        "update": 15,
        "compliance": 0.68359375,
        "mean_reward": 0.6887788748475494,
        "compliant": 175,
        "trajectories": 256,
        "pass_at_4": 0.890625,
        "solved_tasks": 57,
        "tasks": 64,
        "summary_sha256": "55ea25dd75befa9d9b43bac504f79c23a65a689e470f0f13be27add93a9a6485"
      },
      {
        "update": 16,
        "compliance": 0.6328125,
        "mean_reward": 0.6472823403148037,
        "compliant": 162,
        "trajectories": 256,
        "pass_at_4": 0.8125,
        "solved_tasks": 52,
        "tasks": 64,
        "summary_sha256": "5931c20f68bab0c6e7aa9d12abba6d0f59659a3199f4782787b322d738a62c78"
      },
      {
        "update": 17,
        "compliance": 0.71484375,
        "mean_reward": 0.7425900325895113,
        "compliant": 183,
        "trajectories": 256,
        "pass_at_4": 0.90625,
        "solved_tasks": 58,
        "tasks": 64,
        "summary_sha256": "235a65f5de4cbdfcae78e1681b35efcfa2f9db23d7f0dc0b6207fcd8d7443522"
      },
      {
        "update": 18,
        "compliance": 0.59375,
        "mean_reward": 0.625820749543503,
        "compliant": 152,
        "trajectories": 256,
        "pass_at_4": 0.84375,
        "solved_tasks": 54,
        "tasks": 64,
        "summary_sha256": "acb1177fcb4129ba7d5cc967793c93ae8c423292d0b4c60a2579618dc1277b56"
      },
      {
        "update": 19,
        "compliance": 0.64453125,
        "mean_reward": 0.6616031609037711,
        "compliant": 165,
        "trajectories": 256,
        "pass_at_4": 0.875,
        "solved_tasks": 56,
        "tasks": 64,
        "summary_sha256": "16e3648ee4e6db156a8116ffb711b9e6e2e9d6a58b82215b8c1c8c3681582921"
      },
      {
        "update": 20,
        "compliance": 0.59765625,
        "mean_reward": 0.6242814161077991,
        "compliant": 153,
        "trajectories": 256,
        "pass_at_4": 0.8125,
        "solved_tasks": 52,
        "tasks": 64,
        "summary_sha256": "caef8a3a0fcd58752344b2baea4d93cc229188216a6beb1daf485b89899ca8da"
      },
      {
        "update": 21,
        "compliance": 0.6796875,
        "mean_reward": 0.7131624432861136,
        "compliant": 174,
        "trajectories": 256,
        "pass_at_4": 0.90625,
        "solved_tasks": 58,
        "tasks": 64,
        "summary_sha256": "c9a1ba9b15dd0f0548f72a5e3ce0349c35596b4392b5795789a9fc53b6882542"
      },
      {
        "update": 22,
        "compliance": 0.73046875,
        "mean_reward": 0.7593845730929678,
        "compliant": 187,
        "trajectories": 256,
        "pass_at_4": 0.921875,
        "solved_tasks": 59,
        "tasks": 64,
        "summary_sha256": "da0260887d66854ad3f535a5ccc6a14296090757e96b0962c14cfbe165822c99"
      },
      {
        "update": 23,
        "compliance": 0.79296875,
        "mean_reward": 0.8249202800437341,
        "compliant": 203,
        "trajectories": 256,
        "pass_at_4": 0.953125,
        "solved_tasks": 61,
        "tasks": 64,
        "summary_sha256": "0914631cc3cb1a1d8ce382b4269f74ff1a926a3bb3c1edc79e2cea1efa29e253"
      },
      {
        "update": 24,
        "compliance": 0.70703125,
        "mean_reward": 0.7360282400928124,
        "compliant": 181,
        "trajectories": 256,
        "pass_at_4": 0.890625,
        "solved_tasks": 57,
        "tasks": 64,
        "summary_sha256": "85955d5c530529c878a1e5d712c360247d77b5da38eac1531d99a3405b81935f"
      },
      {
        "update": 25,
        "compliance": 0.71875,
        "mean_reward": 0.7454788551485926,
        "compliant": 184,
        "trajectories": 256,
        "pass_at_4": 0.890625,
        "solved_tasks": 57,
        "tasks": 64,
        "summary_sha256": "4600aee85ae0976adf36bb9b94e582f5f0462e5a669a2128c0babf98307ce309"
      },
      {
        "update": 26,
        "compliance": 0.7578125,
        "mean_reward": 0.7644239937564128,
        "compliant": 194,
        "trajectories": 256,
        "pass_at_4": 0.953125,
        "solved_tasks": 61,
        "tasks": 64,
        "summary_sha256": "625a7895c89150ae6eac9beeb790104e72dd59b4be428ab844c8872ee62b5dc1"
      },
      {
        "update": 27,
        "compliance": 0.75,
        "mean_reward": 0.7772200238083664,
        "compliant": 192,
        "trajectories": 256,
        "pass_at_4": 0.9375,
        "solved_tasks": 60,
        "tasks": 64,
        "summary_sha256": "081ac828541613c76c0bf67a475c30cb924f9325aacdf643f5655cbcde136dd7"
      },
      {
        "update": 28,
        "compliance": 0.76953125,
        "mean_reward": 0.783573606808513,
        "compliant": 197,
        "trajectories": 256,
        "pass_at_4": 0.9375,
        "solved_tasks": 60,
        "tasks": 64,
        "summary_sha256": "22998f4e45c17ff1cc51dbee10cf7f5e0a5d0f51d1a0d8d959ebe3003cdbc315"
      },
      {
        "update": 29,
        "compliance": 0.80078125,
        "mean_reward": 0.8218530782345714,
        "compliant": 205,
        "trajectories": 256,
        "pass_at_4": 0.953125,
        "solved_tasks": 61,
        "tasks": 64,
        "summary_sha256": "4ef8e9269c32051ddf2c4c0860c5082ecf073a35842e651fe975ba85cd79d3c0"
      },
      {
        "update": 30,
        "compliance": 0.7890625,
        "mean_reward": 0.8134885467250251,
        "compliant": 202,
        "trajectories": 256,
        "pass_at_4": 0.953125,
        "solved_tasks": 61,
        "tasks": 64,
        "summary_sha256": "19bca0a788935d3c5f85ac2449faadf72cf5a4bb3f8a8b1cdbf90e8aa1bc253c"
      },
      {
        "update": 31,
        "compliance": 0.66015625,
        "mean_reward": 0.6905760120498522,
        "compliant": 169,
        "trajectories": 256,
        "pass_at_4": 0.84375,
        "solved_tasks": 54,
        "tasks": 64,
        "summary_sha256": "5d47340733d5f59c4de0587375f074d19c5d1492f274780809cb41bb58c9e837"
      },
      {
        "update": 32,
        "compliance": 0.76953125,
        "mean_reward": 0.7929442664354103,
        "compliant": 197,
        "trajectories": 256,
        "pass_at_4": 0.9375,
        "solved_tasks": 60,
        "tasks": 64,
        "summary_sha256": "f73cfb9220635d8e980acfa35a2948ee63765f5c544eeea4cc4949100e1e0ce9"
      },
      {
        "update": 33,
        "compliance": 0.80078125,
        "mean_reward": 0.8326832862425382,
        "compliant": 205,
        "trajectories": 256,
        "pass_at_4": 0.90625,
        "solved_tasks": 58,
        "tasks": 64,
        "summary_sha256": "ca4c9274e6711bd0c739d2e1e947c88de14e9b2a1db961b9c781a2c547b577ae"
      },
      {
        "update": 34,
        "compliance": 0.73828125,
        "mean_reward": 0.7476678380884042,
        "compliant": 189,
        "trajectories": 256,
        "pass_at_4": 0.9375,
        "solved_tasks": 60,
        "tasks": 64,
        "summary_sha256": "ccdb80bc7e9a7bfab5c42f00acee1683c9662031b5856d0a7105953401ccff19"
      },
      {
        "update": 35,
        "compliance": 0.8203125,
        "mean_reward": 0.8477356324260731,
        "compliant": 210,
        "trajectories": 256,
        "pass_at_4": 0.90625,
        "solved_tasks": 58,
        "tasks": 64,
        "summary_sha256": "367e25c5c76ce571ef72ebf48861ca997bb9bef6b1923b847d1be384b038c020"
      },
      {
        "update": 36,
        "compliance": 0.69140625,
        "mean_reward": 0.7125303281227449,
        "compliant": 177,
        "trajectories": 256,
        "pass_at_4": 0.875,
        "solved_tasks": 56,
        "tasks": 64,
        "summary_sha256": "0deec4e2c28b7b09e81b4f74c589e50546d67c3d83b0b53e71a4f7dbc98fdbbf"
      },
      {
        "update": 37,
        "compliance": 0.73046875,
        "mean_reward": 0.7453887662053374,
        "compliant": 187,
        "trajectories": 256,
        "pass_at_4": 0.921875,
        "solved_tasks": 59,
        "tasks": 64,
        "summary_sha256": "3ce1c6b8383ce09be4c36f1653f84adef5e505616d77f4b1b716c94b99b65692"
      },
      {
        "update": 38,
        "compliance": 0.77734375,
        "mean_reward": 0.8102951702026373,
        "compliant": 199,
        "trajectories": 256,
        "pass_at_4": 0.953125,
        "solved_tasks": 61,
        "tasks": 64,
        "summary_sha256": "d5c7ce9b6ea236727bd13e1f169559637ab9c532757425f3a1eb4b41660cbed9"
      },
      {
        "update": 39,
        "compliance": 0.76953125,
        "mean_reward": 0.800499075965806,
        "compliant": 197,
        "trajectories": 256,
        "pass_at_4": 0.90625,
        "solved_tasks": 58,
        "tasks": 64,
        "summary_sha256": "e5dc7e936f6df1559280fc14650969ea73f424ceee55df8efd8d1f94ffadc9b7"
      },
      {
        "update": 40,
        "compliance": 0.8046875,
        "mean_reward": 0.8196113030750497,
        "compliant": 206,
        "trajectories": 256,
        "pass_at_4": 0.953125,
        "solved_tasks": 61,
        "tasks": 64,
        "summary_sha256": "6bf676bd556f38eaadb6bfcb4ab7aa9b2af815bc2db1081799cf2655e7e060dd"
      },
      {
        "update": 41,
        "compliance": 0.7578125,
        "mean_reward": 0.7837794831290923,
        "compliant": 194,
        "trajectories": 256,
        "pass_at_4": 0.953125,
        "solved_tasks": 61,
        "tasks": 64,
        "summary_sha256": "0c1d36c3a67d0eb3f893cce39911141b2a1d5d973e086768adf6e30ec6a52989"
      },
      {
        "update": 42,
        "compliance": 0.7109375,
        "mean_reward": 0.7492006692709474,
        "compliant": 182,
        "trajectories": 256,
        "pass_at_4": 0.921875,
        "solved_tasks": 59,
        "tasks": 64,
        "summary_sha256": "ab5fd11194bedc6aebf0dc7563a6f730be1c14b8dba5a7a3478aac5df4533c7b"
      },
      {
        "update": 43,
        "compliance": 0.72265625,
        "mean_reward": 0.7507122754267275,
        "compliant": 185,
        "trajectories": 256,
        "pass_at_4": 0.890625,
        "solved_tasks": 57,
        "tasks": 64,
        "summary_sha256": "d7cc8da5da34da848c1e47ecd35b61445da4db99850b850b62d205705e614745"
      },
      {
        "update": 44,
        "compliance": 0.7421875,
        "mean_reward": 0.786782557966046,
        "compliant": 190,
        "trajectories": 256,
        "pass_at_4": 0.96875,
        "solved_tasks": 62,
        "tasks": 64,
        "summary_sha256": "2a85b5796fd25a24435e18be02c194fa3dbc87f133c991baf5b291b1de688918"
      },
      {
        "update": 45,
        "compliance": 0.73046875,
        "mean_reward": 0.7582063749425202,
        "compliant": 187,
        "trajectories": 256,
        "pass_at_4": 0.921875,
        "solved_tasks": 59,
        "tasks": 64,
        "summary_sha256": "deb271fef90cfb77bc619002b66fedc469bd1218cd7bd036725fee08c49cd61d"
      },
      {
        "update": 46,
        "compliance": 0.734375,
        "mean_reward": 0.7671808298441225,
        "compliant": 188,
        "trajectories": 256,
        "pass_at_4": 0.96875,
        "solved_tasks": 62,
        "tasks": 64,
        "summary_sha256": "d1618eeb977e04c4dd2e2db432591b27e1d45e3cecdee3603fab6eb877ac8bda"
      },
      {
        "update": 47,
        "compliance": 0.74609375,
        "mean_reward": 0.7872472149226115,
        "compliant": 191,
        "trajectories": 256,
        "pass_at_4": 0.90625,
        "solved_tasks": 58,
        "tasks": 64,
        "summary_sha256": "13df5a4fd2e47145843d8fb9f699cf4e7257c86dec0ba38e7601f14520febfe2"
      },
      {
        "update": 48,
        "compliance": 0.6953125,
        "mean_reward": 0.718787879905262,
        "compliant": 178,
        "trajectories": 256,
        "pass_at_4": 0.84375,
        "solved_tasks": 54,
        "tasks": 64,
        "summary_sha256": "a7b087f9ba4d60af1c7d8219542c81f1360e2cd8a6d4b2b14562831abe29db19"
      },
      {
        "update": 49,
        "compliance": 0.7265625,
        "mean_reward": 0.7711967509086999,
        "compliant": 186,
        "trajectories": 256,
        "pass_at_4": 0.90625,
        "solved_tasks": 58,
        "tasks": 64,
        "summary_sha256": "5110b53e3c353e5e90422f25b42c18e6a558b5619211dec35325638db4855118"
      },
      {
        "update": 50,
        "compliance": 0.62109375,
        "mean_reward": 0.6427861212150656,
        "compliant": 159,
        "trajectories": 256,
        "pass_at_4": 0.890625,
        "solved_tasks": 57,
        "tasks": 64,
        "summary_sha256": "64aaf818d9cf98e3715bf3ae9a917672ec8e51cb7ed22b2edcf5a234cdff93b3"
      }
    ],
    "curve_source_sha256": "c1140327e492f644b732a82a019f733cdabe42fad8e4855954ca7ed9bfd94684",
    "window_updates": 10,
    "pass_at_4": {
      "samples_per_task": 4,
      "tasks_per_update": 64,
      "first_window_mean": 0.81875,
      "last_window_mean": 0.9171875,
      "source": {
        "path": "runs/tinker-kimi-k26-rft-rl-v1b",
        "config_sha256": "96e0ecffb6a3f754b5a9071a808a02a89ca5476c8349790f51e3ce18d5e4dfc3"
      }
    }
  },
  "splits": {
    "development": {
      "name": "development",
      "tasks": 600,
      "samples_per_task": 4,
      "temperature": 0.75,
      "rows": [
        {
          "key": "pre_rl",
          "label": "Pre-RL",
          "note": "Kimi K2.6 base policy with the verifier tool",
          "pass_rate": 0.40541666666666665,
          "mean_reward": 0.4207,
          "pass_at_4": 0.715,
          "repeat_rate": 0.1713
        },
        {
          "key": "coordinated_repair_sft",
          "label": "Repair SFT",
          "note": "Coordinated-repair supervised fine-tune (rejected)",
          "pass_rate": 0.3258333333333333,
          "mean_reward": 0.3374,
          "pass_at_4": 0.7067,
          "repeat_rate": 0.1188
        },
        {
          "key": "verified_trajectory_rft",
          "label": "RFT",
          "note": "Rejection fine-tune on verified compliant trajectories",
          "pass_rate": 0.43625,
          "mean_reward": 0.4505,
          "pass_at_4": 0.7583,
          "repeat_rate": 0.16
        },
        {
          "key": "rl",
          "label": "RFT + RL",
          "note": "50 policy-gradient updates from the RFT initialization",
          "pass_rate": 0.7,
          "mean_reward": 0.7271,
          "pass_at_4": 0.9067,
          "repeat_rate": 0.0192
        }
      ],
      "deltas": {
        "pre_rl": {
          "pass_delta": 0.2945833333333333,
          "ci95": [
            0.26416666666666666,
            0.32458333333333333
          ]
        },
        "coordinated_repair_sft": {
          "pass_delta": 0.3742,
          "ci95": [
            0.345,
            0.4038
          ]
        },
        "verified_trajectory_rft": {
          "pass_delta": 0.26375,
          "ci95": [
            0.23458333333333334,
            0.29291666666666666
          ]
        }
      }
    },
    "confirmatory": {
      "name": "confirmatory",
      "tasks": 900,
      "samples_per_task": 4,
      "temperature": 0.75,
      "rows": [
        {
          "key": "pre_rl",
          "label": "Pre-RL",
          "note": "Kimi K2.6 base policy with the verifier tool",
          "pass_rate": 0.42694444444444446,
          "mean_reward": 0.444521546052866,
          "pass_at_4": 0.7455555555555555,
          "repeat_rate": 0.17277777777777778
        },
        {
          "key": "coordinated_repair_sft",
          "label": "Repair SFT",
          "note": "Coordinated-repair supervised fine-tune (rejected)",
          "pass_rate": 0.3411111111111111,
          "mean_reward": 0.3523130886463108,
          "pass_at_4": 0.7022222222222222,
          "repeat_rate": 0.11916666666666667
        },
        {
          "key": "verified_trajectory_rft",
          "label": "RFT",
          "note": "Rejection fine-tune on verified compliant trajectories",
          "pass_rate": 0.46055555555555555,
          "mean_reward": 0.47737861866245934,
          "pass_at_4": 0.7722222222222223,
          "repeat_rate": 0.15888888888888889
        },
        {
          "key": "rl",
          "label": "RFT + RL",
          "note": "50 policy-gradient updates from the RFT initialization",
          "pass_rate": 0.7444444444444445,
          "mean_reward": 0.7750589186497627,
          "pass_at_4": 0.93,
          "repeat_rate": 0.014444444444444444
        }
      ],
      "deltas": {
        "pre_rl": {
          "pass_delta": 0.3175,
          "ci95": [
            0.2936111111111111,
            0.3408333333333333
          ],
          "reward_delta": 0.3305373725968967,
          "reward_ci95": [
            0.30581747782581603,
            0.3547122589260015
          ]
        },
        "coordinated_repair_sft": {
          "pass_delta": 0.4033333333333333,
          "ci95": [
            0.37972222222222224,
            0.42694444444444446
          ],
          "reward_delta": 0.4227458300034519,
          "reward_ci95": [
            0.3979280689849089,
            0.44731546478607453
          ]
        },
        "verified_trajectory_rft": {
          "pass_delta": 0.2838888888888889,
          "ci95": [
            0.2608333333333333,
            0.30777777777777776
          ],
          "reward_delta": 0.29768029998730344,
          "reward_ci95": [
            0.2737594805429024,
            0.32245235973775144
          ]
        }
      },
      "paired_compliant_reward": {
        "tasks": 651,
        "baseline_only_passing_tasks": 20,
        "rl_only_passing_tasks": 186,
        "mean_positive_reward_delta": 0.009393813622188643,
        "mean_positive_reward_delta_ci95": [
          0.0042961823145805204,
          0.014776310088082962
        ],
        "best_of_4_reward_delta": 0.019470063092651847,
        "best_of_4_reward_delta_ci95": [
          0.013146823728594324,
          0.02587952690357338
        ]
      },
      "literal_aggregate_mean_positive_reward": {
        "pre_rl": 1.0411695288160816,
        "rl": 1.0411239205743081,
        "delta": -4.56082417735e-05
      }
    }
  },
  "rescore": {
    "summary": {
      "candidate_count": 15,
      "all_stages_completed": true,
      "all_solver_numerics_converged": false,
      "fast_compliant_count": 15,
      "counterfactual_detailed_compliant_count": 5,
      "compliance_reversal_count": 10,
      "paired_task_count": 6,
      "paired_winner_change_count": 5,
      "fast_vs_detailed_reward_spearman": -0.33594217189442416,
      "rl_only_success_count": 3,
      "rl_only_detailed_compliant_count": 2,
      "material_incremental_signal": true
    },
    "observables": {
      "impedance": {
        "count": 15,
        "mean_signed_relative_change": -0.042527420044412526,
        "minimum_signed_relative_change": -0.11502167892658509,
        "maximum_signed_relative_change": -0.005716357525760741
      },
      "no_load_loss": {
        "count": 15,
        "mean_signed_relative_change": 0.008735529587883196,
        "minimum_signed_relative_change": -0.040659897354912954,
        "maximum_signed_relative_change": 0.05189329773781433
      },
      "load_loss": {
        "count": 15,
        "mean_signed_relative_change": -0.00041367234186090847,
        "minimum_signed_relative_change": -0.022398654115699353,
        "maximum_signed_relative_change": 0.028820576308620158
      }
    },
    "stages": {
      "elmer_core_loss": {
        "completed_count": 15,
        "numerically_converged_count": 15,
        "full_stage_ok_count": 15,
        "candidate_count": 15
      },
      "elmer_harmonic_loss": {
        "completed_count": 15,
        "numerically_converged_count": 15,
        "full_stage_ok_count": 15,
        "candidate_count": 15
      },
      "elmer_magnetic": {
        "completed_count": 15,
        "numerically_converged_count": 15,
        "full_stage_ok_count": 15,
        "candidate_count": 15
      },
      "elmer_short_circuit_thermal": {
        "completed_count": 15,
        "numerically_converged_count": 15,
        "full_stage_ok_count": 15,
        "candidate_count": 15
      },
      "getdp_dielectric": {
        "completed_count": 15,
        "numerically_converged_count": 15,
        "full_stage_ok_count": 15,
        "candidate_count": 15
      },
      "getdp_leakage": {
        "completed_count": 15,
        "numerically_converged_count": 15,
        "full_stage_ok_count": 13,
        "candidate_count": 15
      },
      "openfoam_onan": {
        "completed_count": 15,
        "numerically_converged_count": 0,
        "full_stage_ok_count": 0,
        "candidate_count": 15
      }
    },
    "records": [
      {
        "task_id": "proc-design-20261003-3513-11607",
        "policy": "rft",
        "role": "paired_best_of_four",
        "model_hash": "e0e96d2cfeb5",
        "fast_reward": 0.9878509340255821,
        "detailed_reward": 0.0,
        "fast_compliant": true,
        "detailed_compliant": false,
        "failed_checks": [
          "no_load_loss"
        ],
        "impedance": {
          "fast_value": 3.617248703050203,
          "detailed_value": 3.466558149998143,
          "signed_relative_change": -0.041658886469429675,
          "lower_limit": 3.169764717693771,
          "upper_limit": 3.683780617860328,
          "counterfactual_ok": true
        },
        "no_load_loss": {
          "fast_value": 1748.7978981964393,
          "detailed_value": 1790.0294250085337,
          "signed_relative_change": 0.023577067913117415,
          "upper_limit": 1885.453397844989,
          "counterfactual_ok": false
        },
        "load_loss": {
          "fast_value": 8846.531303327904,
          "detailed_value": 8768.322845487883,
          "signed_relative_change": -0.00884057888435892,
          "upper_limit": 11668.436362004008,
          "counterfactual_ok": true
        }
      },
      {
        "task_id": "proc-design-20261003-3513-11607",
        "policy": "rl",
        "role": "paired_best_of_four",
        "model_hash": "50b5ab2af8c7",
        "fast_reward": 1.1702796462949432,
        "detailed_reward": 0.0,
        "fast_compliant": true,
        "detailed_compliant": false,
        "failed_checks": [
          "no_load_loss"
        ],
        "impedance": {
          "fast_value": 3.630344197669486,
          "detailed_value": 3.4987903537079093,
          "signed_relative_change": -0.03623729233333533,
          "lower_limit": 3.169764717693771,
          "upper_limit": 3.683780617860328,
          "counterfactual_ok": true
        },
        "no_load_loss": {
          "fast_value": 1758.4153194242062,
          "detailed_value": 1832.4421894666477,
          "signed_relative_change": 0.0420986266581672,
          "upper_limit": 1885.453397844989,
          "counterfactual_ok": false
        },
        "load_loss": {
          "fast_value": 10256.51641644931,
          "detailed_value": 10172.143252019961,
          "signed_relative_change": -0.008226298384706177,
          "upper_limit": 11668.436362004008,
          "counterfactual_ok": true
        }
      },
      {
        "task_id": "proc-design-20261003-3402-11271",
        "policy": "rft",
        "role": "paired_best_of_four",
        "model_hash": "6b7abbda9cd6",
        "fast_reward": 1.0861081772147898,
        "detailed_reward": 0.0,
        "fast_compliant": true,
        "detailed_compliant": false,
        "failed_checks": [
          "impedance"
        ],
        "impedance": {
          "fast_value": 7.33983110052756,
          "detailed_value": 6.727231185675335,
          "signed_relative_change": -0.08346239940156022,
          "lower_limit": 6.879068088206456,
          "upper_limit": 7.994592643050746,
          "counterfactual_ok": false
        },
        "no_load_loss": {
          "fast_value": 1852.2742791247533,
          "detailed_value": 1873.8984452935363,
          "signed_relative_change": 0.011674386678306043,
          "upper_limit": 2334.3716275644383,
          "counterfactual_ok": true
        },
        "load_loss": {
          "fast_value": 10114.057006644565,
          "detailed_value": 10006.349686457019,
          "signed_relative_change": -0.01064926963698016,
          "upper_limit": 11906.52295095116,
          "counterfactual_ok": true
        }
      },
      {
        "task_id": "proc-design-20261003-3402-11271",
        "policy": "rl",
        "role": "paired_best_of_four",
        "model_hash": "43fd900a8158",
        "fast_reward": 1.1241383811411079,
        "detailed_reward": 0.0,
        "fast_compliant": true,
        "detailed_compliant": false,
        "failed_checks": [
          "impedance"
        ],
        "impedance": {
          "fast_value": 7.19037313035022,
          "detailed_value": 6.587804393886081,
          "signed_relative_change": -0.08380215122922129,
          "lower_limit": 6.879068088206456,
          "upper_limit": 7.994592643050746,
          "counterfactual_ok": false
        },
        "no_load_loss": {
          "fast_value": 1948.450660449386,
          "detailed_value": 2049.562190699527,
          "signed_relative_change": 0.05189329773781433,
          "upper_limit": 2334.3716275644383,
          "counterfactual_ok": true
        },
        "load_loss": {
          "fast_value": 9908.425602864178,
          "detailed_value": 9800.520935894112,
          "signed_relative_change": -0.010890192982715048,
          "upper_limit": 11906.52295095116,
          "counterfactual_ok": true
        }
      },
      {
        "task_id": "proc-design-20261003-27639-93364",
        "policy": "rft",
        "role": "paired_best_of_four",
        "model_hash": "7ed355a9ed2f",
        "fast_reward": 1.0154202938777628,
        "detailed_reward": 1.0238865707607538,
        "fast_compliant": true,
        "detailed_compliant": true,
        "failed_checks": [],
        "impedance": {
          "fast_value": 3.5296485801075095,
          "detailed_value": 3.411830783162623,
          "signed_relative_change": -0.03337946944885317,
          "lower_limit": 3.147523930516237,
          "upper_limit": 3.6579332165458966,
          "counterfactual_ok": true
        },
        "no_load_loss": {
          "fast_value": 2269.812776700032,
          "detailed_value": 2270.454714907118,
          "signed_relative_change": 0.0002828154875485031,
          "upper_limit": 3065.09387233179,
          "counterfactual_ok": true
        },
        "load_loss": {
          "fast_value": 9011.96636196447,
          "detailed_value": 8824.672656310648,
          "signed_relative_change": -0.020782779044128157,
          "upper_limit": 9612.775458309274,
          "counterfactual_ok": true
        }
      },
      {
        "task_id": "proc-design-20261003-27639-93364",
        "policy": "rl",
        "role": "paired_best_of_four",
        "model_hash": "eea7ad489856",
        "fast_reward": 1.092567140279146,
        "detailed_reward": 0.0,
        "fast_compliant": true,
        "detailed_compliant": false,
        "failed_checks": [
          "no_load_loss"
        ],
        "impedance": {
          "fast_value": 3.6092179824224346,
          "detailed_value": 3.4866505162693797,
          "signed_relative_change": -0.03395956319346223,
          "lower_limit": 3.147523930516237,
          "upper_limit": 3.6579332165458966,
          "counterfactual_ok": true
        },
        "no_load_loss": {
          "fast_value": 2892.254069278902,
          "detailed_value": 3001.7420365224834,
          "signed_relative_change": 0.03785558412953636,
          "upper_limit": 3065.09387233179,
          "counterfactual_ok": false
        },
        "load_loss": {
          "fast_value": 8524.96605150615,
          "detailed_value": 8392.351295493452,
          "signed_relative_change": -0.015556045057712303,
          "upper_limit": 9612.775458309274,
          "counterfactual_ok": true
        }
      },
      {
        "task_id": "proc-design-20261003-30543-102972",
        "policy": "rft",
        "role": "paired_best_of_four",
        "model_hash": "c1ee597a2a5c",
        "fast_reward": 1.0256102112944685,
        "detailed_reward": 1.0211919593695793,
        "fast_compliant": true,
        "detailed_compliant": true,
        "failed_checks": [],
        "impedance": {
          "fast_value": 7.6473780501923105,
          "detailed_value": 7.479761851641625,
          "signed_relative_change": -0.021918126376199068,
          "lower_limit": 6.933850386610989,
          "upper_limit": 8.05825855741277,
          "counterfactual_ok": true
        },
        "no_load_loss": {
          "fast_value": 1938.6219169647845,
          "detailed_value": 1908.9569141988438,
          "signed_relative_change": -0.015302108423691967,
          "upper_limit": 2199.951918158839,
          "counterfactual_ok": true
        },
        "load_loss": {
          "fast_value": 13239.024559319643,
          "detailed_value": 13439.012393501973,
          "signed_relative_change": 0.015105934223949126,
          "upper_limit": 14143.639363781373,
          "counterfactual_ok": true
        }
      },
      {
        "task_id": "proc-design-20261003-30543-102972",
        "policy": "rl",
        "role": "paired_best_of_four",
        "model_hash": "4f6067b75ada",
        "fast_reward": 1.0147967289093094,
        "detailed_reward": 1.0115941072534107,
        "fast_compliant": true,
        "detailed_compliant": true,
        "failed_checks": [],
        "impedance": {
          "fast_value": 7.902509608841004,
          "detailed_value": 7.7908468245953895,
          "signed_relative_change": -0.014130040932906986,
          "lower_limit": 6.933850386610989,
          "upper_limit": 8.05825855741277,
          "counterfactual_ok": true
        },
        "no_load_loss": {
          "fast_value": 1964.0822591130732,
          "detailed_value": 1903.6409248587279,
          "signed_relative_change": -0.030773321216005983,
          "upper_limit": 2199.951918158839,
          "counterfactual_ok": true
        },
        "load_loss": {
          "fast_value": 12249.500679500028,
          "detailed_value": 12525.003448665972,
          "signed_relative_change": 0.022490938722670304,
          "upper_limit": 14143.639363781373,
          "counterfactual_ok": true
        }
      },
      {
        "task_id": "proc-design-20261003-50844-170612",
        "policy": "rft",
        "role": "paired_best_of_four",
        "model_hash": "2636f7d77d41",
        "fast_reward": 1.1283602399509185,
        "detailed_reward": 0.0,
        "fast_compliant": true,
        "detailed_compliant": false,
        "failed_checks": [
          "impedance"
        ],
        "impedance": {
          "fast_value": 3.228726676063897,
          "detailed_value": 3.1305005888703525,
          "signed_relative_change": -0.030422546424180626,
          "lower_limit": 3.184884523001192,
          "upper_limit": 3.7013522834878714,
          "counterfactual_ok": false
        },
        "no_load_loss": {
          "fast_value": 3508.655304213271,
          "detailed_value": 3687.6765262710996,
          "signed_relative_change": 0.05102274419571967,
          "upper_limit": 4154.909993694911,
          "counterfactual_ok": true
        },
        "load_loss": {
          "fast_value": 10762.212336778262,
          "detailed_value": 10521.153265127054,
          "signed_relative_change": -0.022398654115699353,
          "upper_limit": 13138.02777483165,
          "counterfactual_ok": true
        }
      },
      {
        "task_id": "proc-design-20261003-50844-170612",
        "policy": "rl",
        "role": "paired_best_of_four",
        "model_hash": "d61138961862",
        "fast_reward": 1.1105307864804534,
        "detailed_reward": 0.0,
        "fast_compliant": true,
        "detailed_compliant": false,
        "failed_checks": [
          "impedance",
          "no_load_loss"
        ],
        "impedance": {
          "fast_value": 3.228726676063897,
          "detailed_value": 3.1305005888703525,
          "signed_relative_change": -0.030422546424180626,
          "lower_limit": 3.184884523001192,
          "upper_limit": 3.7013522834878714,
          "counterfactual_ok": false
        },
        "no_load_loss": {
          "fast_value": 3890.7860799196674,
          "detailed_value": 4083.27514556174,
          "signed_relative_change": 0.049473052922520695,
          "upper_limit": 4154.909993694911,
          "counterfactual_ok": false
        },
        "load_loss": {
          "fast_value": 10762.212336778262,
          "detailed_value": 10521.153265127054,
          "signed_relative_change": -0.022398654115699353,
          "upper_limit": 13138.02777483165,
          "counterfactual_ok": true
        }
      },
      {
        "task_id": "proc-design-20261003-22292-75107",
        "policy": "rft",
        "role": "paired_best_of_four",
        "model_hash": "448565dd5f8a",
        "fast_reward": 0.893579040841198,
        "detailed_reward": 0.0,
        "fast_compliant": true,
        "detailed_compliant": false,
        "failed_checks": [
          "impedance"
        ],
        "impedance": {
          "fast_value": 6.946088121595958,
          "detailed_value": 6.668537463213026,
          "signed_relative_change": -0.039957837206240465,
          "lower_limit": 6.771886625779557,
          "upper_limit": 7.870030402933,
          "counterfactual_ok": false
        },
        "no_load_loss": {
          "fast_value": 2337.984362011243,
          "detailed_value": 2268.800640379517,
          "signed_relative_change": -0.029591182368820787,
          "upper_limit": 2813.2237269367142,
          "counterfactual_ok": true
        },
        "load_loss": {
          "fast_value": 11556.091880681577,
          "detailed_value": 11824.389446888386,
          "signed_relative_change": 0.023216981050083612,
          "upper_limit": 14430.260355207873,
          "counterfactual_ok": true
        }
      },
      {
        "task_id": "proc-design-20261003-22292-75107",
        "policy": "rl",
        "role": "paired_best_of_four",
        "model_hash": "c99c543d9248",
        "fast_reward": 0.9415812885484112,
        "detailed_reward": 0.0,
        "fast_compliant": true,
        "detailed_compliant": false,
        "failed_checks": [
          "impedance"
        ],
        "impedance": {
          "fast_value": 7.36300230907173,
          "detailed_value": 6.516097421541977,
          "signed_relative_change": -0.11502167892658509,
          "lower_limit": 6.771886625779557,
          "upper_limit": 7.870030402933,
          "counterfactual_ok": false
        },
        "no_load_loss": {
          "fast_value": 2205.2688302438614,
          "detailed_value": 2115.602825966157,
          "signed_relative_change": -0.040659897354912954,
          "upper_limit": 2813.2237269367142,
          "counterfactual_ok": true
        },
        "load_loss": {
          "fast_value": 12363.364558009154,
          "detailed_value": 12703.334460132299,
          "signed_relative_change": 0.02749817014033671,
          "upper_limit": 14430.260355207873,
          "counterfactual_ok": true
        }
      },
      {
        "task_id": "proc-design-20261003-8440-27977",
        "policy": "rl",
        "role": "rl_only_success_best_of_four",
        "model_hash": "9b333212841c",
        "fast_reward": 1.0445866651313342,
        "detailed_reward": 1.0167488850428663,
        "fast_compliant": true,
        "detailed_compliant": true,
        "failed_checks": [],
        "impedance": {
          "fast_value": 6.748903612750864,
          "detailed_value": 6.693214581041891,
          "signed_relative_change": -0.00825156720326511,
          "lower_limit": 6.565866247434841,
          "upper_limit": 7.630601314586436,
          "counterfactual_ok": true
        },
        "no_load_loss": {
          "fast_value": 1445.2428417306926,
          "detailed_value": 1487.498633622665,
          "signed_relative_change": 0.029237848942652914,
          "upper_limit": 2455.550702345,
          "counterfactual_ok": true
        },
        "load_loss": {
          "fast_value": 12472.934144583418,
          "detailed_value": 12832.411294889778,
          "signed_relative_change": 0.028820576308620158,
          "upper_limit": 13817.254690229778,
          "counterfactual_ok": true
        }
      },
      {
        "task_id": "proc-design-20261003-5322-17672",
        "policy": "rl",
        "role": "rl_only_success_best_of_four",
        "model_hash": "0e356d4b4e1b",
        "fast_reward": 1.0998018662982672,
        "detailed_reward": 0.0,
        "fast_compliant": true,
        "detailed_compliant": false,
        "failed_checks": [
          "load_loss"
        ],
        "impedance": {
          "fast_value": 6.5212730500808,
          "detailed_value": 6.48399512180343,
          "signed_relative_change": -0.005716357525760741,
          "lower_limit": 6.330581572721172,
          "upper_limit": 7.357162368297578,
          "counterfactual_ok": true
        },
        "no_load_loss": {
          "fast_value": 2075.657257369088,
          "detailed_value": 1994.762869527729,
          "signed_relative_change": -0.03897290246458758,
          "upper_limit": 2917.2164459330907,
          "counterfactual_ok": true
        },
        "load_loss": {
          "fast_value": 11550.792048570846,
          "detailed_value": 11764.762890995999,
          "signed_relative_change": 0.018524343744170096,
          "upper_limit": 12284.528214679769,
          "counterfactual_ok": false
        }
      },
      {
        "task_id": "proc-design-20261003-35631-120234",
        "policy": "rl",
        "role": "rl_only_success_best_of_four",
        "model_hash": "6de37f0029ac",
        "fast_reward": 1.013364039643124,
        "detailed_reward": 1.0286095237393944,
        "fast_compliant": true,
        "detailed_compliant": true,
        "failed_checks": [],
        "impedance": {
          "fast_value": 6.267427838877267,
          "detailed_value": 5.8940719130995,
          "signed_relative_change": -0.05957083757100723,
          "lower_limit": 5.68674272883783,
          "upper_limit": 6.6089172254061275,
          "counterfactual_ok": true
        },
        "no_load_loss": {
          "fast_value": 2127.093775062052,
          "detailed_value": 2104.157176075426,
          "signed_relative_change": -0.010783069019115946,
          "upper_limit": 2320.401007298486,
          "counterfactual_ok": true
        },
        "load_loss": {
          "fast_value": 14926.69944321695,
          "detailed_value": 14596.5274626317,
          "signed_relative_change": -0.02211955709574416,
          "upper_limit": 19844.298659541804,
          "counterfactual_ok": true
        }
      }
    ]
  },
  "search": {
    "summary": {
      "task_count": 6,
      "rollouts": 24,
      "unique_positive_candidates": 24,
      "detailed_finalists": 18,
      "single_sample_detailed_pass_rate": 0.6666666666666666,
      "fast_best_detailed_pass_rate": 0.5,
      "detailed_search_pass_rate": 1.0,
      "detailed_search_gain_vs_single": 0.3333333333333333,
      "detailed_search_gain_vs_fast_best": 0.5,
      "tasks_selection_changed_from_fast_best": 3
    },
    "execution": {
      "final_attempt_remote_calls": 6,
      "final_attempt_wall_seconds": 122.87069012504071,
      "cache_reused_stage_count": 102,
      "cache_compatible": true,
      "solver_cache_fingerprint": "193aab215e0350c5e09e0206fdb68545ed793027a7a0b34485f30f3b1d689953"
    },
    "method": {
      "training_updates": 0,
      "shortlisting": "comparison anchors plus reward-margin Pareto and normalized farthest-first diversity",
      "ranking": "maximum original reward after converged detailed substitutions",
      "unanalyzed_thermal_policy": "retain fast thermal checks when whole-ONAN is skipped or unconverged"
    },
    "tasks": [
      {
        "task_id": "proc-design-20261003-9728-32371",
        "rollout_count": 4,
        "positive_rollout_count": 4,
        "unique_positive_candidate_count": 4,
        "single_sample_candidate_id": "proc-design-20261003-9728-32371::candidate::c3e9e927f8729905",
        "finalist_candidate_ids": [
          "proc-design-20261003-9728-32371::candidate::c3e9e927f8729905",
          "proc-design-20261003-9728-32371::candidate::2f37aacd5f03c7a3",
          "proc-design-20261003-9728-32371::candidate::d4f4dd38d4ee2983"
        ],
        "fast_best_candidate_id": "proc-design-20261003-9728-32371::candidate::2f37aacd5f03c7a3",
        "robust_best_candidate_id": "proc-design-20261003-9728-32371::candidate::c3e9e927f8729905",
        "pareto_front_size": 3,
        "single_sample_detailed_compliant": true,
        "fast_best_detailed_compliant": false,
        "selected_candidate_id": "proc-design-20261003-9728-32371::candidate::d4f4dd38d4ee2983",
        "selected_detailed_compliant": true,
        "selected_counterfactual_detailed_reward": 0.9970630782335249,
        "selected_fast_reward": 0.9963390224356657,
        "selection_changed_from_fast_best": true
      },
      {
        "task_id": "proc-design-20261003-5640-18639",
        "rollout_count": 4,
        "positive_rollout_count": 4,
        "unique_positive_candidate_count": 4,
        "single_sample_candidate_id": "proc-design-20261003-5640-18639::candidate::ac63f08159e5da42",
        "finalist_candidate_ids": [
          "proc-design-20261003-5640-18639::candidate::ac63f08159e5da42",
          "proc-design-20261003-5640-18639::candidate::018ae22c8a58f8bb",
          "proc-design-20261003-5640-18639::candidate::3a86932260b0794e"
        ],
        "fast_best_candidate_id": "proc-design-20261003-5640-18639::candidate::ac63f08159e5da42",
        "robust_best_candidate_id": "proc-design-20261003-5640-18639::candidate::ac63f08159e5da42",
        "pareto_front_size": 1,
        "single_sample_detailed_compliant": true,
        "fast_best_detailed_compliant": true,
        "selected_candidate_id": "proc-design-20261003-5640-18639::candidate::ac63f08159e5da42",
        "selected_detailed_compliant": true,
        "selected_counterfactual_detailed_reward": 1.2783350722228266,
        "selected_fast_reward": 1.3047975281678932,
        "selection_changed_from_fast_best": false
      },
      {
        "task_id": "proc-design-20261003-37998-128369",
        "rollout_count": 4,
        "positive_rollout_count": 4,
        "unique_positive_candidate_count": 4,
        "single_sample_candidate_id": "proc-design-20261003-37998-128369::candidate::b09e2dc06ac24a2f",
        "finalist_candidate_ids": [
          "proc-design-20261003-37998-128369::candidate::b09e2dc06ac24a2f",
          "proc-design-20261003-37998-128369::candidate::ecfe44e12c520707",
          "proc-design-20261003-37998-128369::candidate::8c0066361683be93"
        ],
        "fast_best_candidate_id": "proc-design-20261003-37998-128369::candidate::ecfe44e12c520707",
        "robust_best_candidate_id": "proc-design-20261003-37998-128369::candidate::8c0066361683be93",
        "pareto_front_size": 2,
        "single_sample_detailed_compliant": true,
        "fast_best_detailed_compliant": false,
        "selected_candidate_id": "proc-design-20261003-37998-128369::candidate::b09e2dc06ac24a2f",
        "selected_detailed_compliant": true,
        "selected_counterfactual_detailed_reward": 1.0385474692555032,
        "selected_fast_reward": 1.0404419658933297,
        "selection_changed_from_fast_best": true
      },
      {
        "task_id": "proc-design-20261003-5923-19584",
        "rollout_count": 4,
        "positive_rollout_count": 4,
        "unique_positive_candidate_count": 4,
        "single_sample_candidate_id": "proc-design-20261003-5923-19584::candidate::4462fa8990f1837b",
        "finalist_candidate_ids": [
          "proc-design-20261003-5923-19584::candidate::4462fa8990f1837b",
          "proc-design-20261003-5923-19584::candidate::d9bc9f5da66b1f78",
          "proc-design-20261003-5923-19584::candidate::e7d941266087d90a"
        ],
        "fast_best_candidate_id": "proc-design-20261003-5923-19584::candidate::d9bc9f5da66b1f78",
        "robust_best_candidate_id": "proc-design-20261003-5923-19584::candidate::d9bc9f5da66b1f78",
        "pareto_front_size": 1,
        "single_sample_detailed_compliant": true,
        "fast_best_detailed_compliant": true,
        "selected_candidate_id": "proc-design-20261003-5923-19584::candidate::d9bc9f5da66b1f78",
        "selected_detailed_compliant": true,
        "selected_counterfactual_detailed_reward": 1.0975497941283754,
        "selected_fast_reward": 1.1154896121653888,
        "selection_changed_from_fast_best": false
      },
      {
        "task_id": "proc-design-20261003-1248-4005",
        "rollout_count": 4,
        "positive_rollout_count": 4,
        "unique_positive_candidate_count": 4,
        "single_sample_candidate_id": "proc-design-20261003-1248-4005::candidate::3a1b514305795fae",
        "finalist_candidate_ids": [
          "proc-design-20261003-1248-4005::candidate::3a1b514305795fae",
          "proc-design-20261003-1248-4005::candidate::f63688398c93fdf3",
          "proc-design-20261003-1248-4005::candidate::b2540671dfb625f5"
        ],
        "fast_best_candidate_id": "proc-design-20261003-1248-4005::candidate::f63688398c93fdf3",
        "robust_best_candidate_id": "proc-design-20261003-1248-4005::candidate::b2540671dfb625f5",
        "pareto_front_size": 2,
        "single_sample_detailed_compliant": false,
        "fast_best_detailed_compliant": true,
        "selected_candidate_id": "proc-design-20261003-1248-4005::candidate::f63688398c93fdf3",
        "selected_detailed_compliant": true,
        "selected_counterfactual_detailed_reward": 1.3157080637858312,
        "selected_fast_reward": 1.334509212924861,
        "selection_changed_from_fast_best": false
      },
      {
        "task_id": "proc-design-20261003-49806-167270",
        "rollout_count": 4,
        "positive_rollout_count": 4,
        "unique_positive_candidate_count": 4,
        "single_sample_candidate_id": "proc-design-20261003-49806-167270::candidate::b3fa631b46a2897d",
        "finalist_candidate_ids": [
          "proc-design-20261003-49806-167270::candidate::b3fa631b46a2897d",
          "proc-design-20261003-49806-167270::candidate::e67305f58554c655",
          "proc-design-20261003-49806-167270::candidate::54dd258397480aba"
        ],
        "fast_best_candidate_id": "proc-design-20261003-49806-167270::candidate::b3fa631b46a2897d",
        "robust_best_candidate_id": "proc-design-20261003-49806-167270::candidate::e67305f58554c655",
        "pareto_front_size": 3,
        "single_sample_detailed_compliant": false,
        "fast_best_detailed_compliant": false,
        "selected_candidate_id": "proc-design-20261003-49806-167270::candidate::54dd258397480aba",
        "selected_detailed_compliant": true,
        "selected_counterfactual_detailed_reward": 0.9827777032389041,
        "selected_fast_reward": 0.987995157683699,
        "selection_changed_from_fast_best": true
      }
    ]
  },
  "exploratory": {
    "status": "nominal_fast_verifier_only",
    "experiment": {
      "policy_rollouts": 4,
      "maximum_turns": 17,
      "maximum_verifier_calls": 16,
      "completed_compliant_rollouts": 3,
      "selection_rule": "highest nominal reward among four frozen rollouts",
      "elapsed_seconds": 110.56022816710174
    },
    "spec": {
      "spec_id": "exploratory-20mva-gsu-34p5-13p8",
      "kva": 20000.0,
      "hv_kv_ll": 34.5,
      "lv_kv_ll": 13.8,
      "hv_bil_kv": 150.0,
      "lv_bil_kv": 95.0,
      "z_pct_target": 8.0,
      "max_no_load_loss_w": 25000.0,
      "max_load_loss_w": 140000.0,
      "phases": 3,
      "freq_hz": 60.0,
      "connection": "YNd",
      "z_pct_tol_of_value": 7.5,
      "ratio_tol_pct": 0.5,
      "max_avg_winding_rise_k": 65.0,
      "max_top_oil_rise_k": 65.0,
      "a_usd_per_w_no_load": 5.0,
      "b_usd_per_w_load": 1.5
    },
    "policy_output": {
      "schema_version": "ironbench.compiler-parameters.v1",
      "grade": "M5",
      "b_target_t": 1.6,
      "vpt_factor": 0.48,
      "j_target_a_mm2": 2.2,
      "lv_layers": 3,
      "hv_layers": 22,
      "extra_gap_mm": 3.0,
      "ducts": 2,
      "margin_profile": "standard"
    },
    "result": {
      "design_sha256": "bf817b566085617f4240c011c7bc51486fb9da5aff5b1bee1fdf5fbb2b24be3a",
      "nominal_compliant": true,
      "nominal_reward": 0.8685939252218119,
      "toc_usd": 459696.26721537893,
      "baseline_toc_usd": 399289.3851504209,
      "cost_ratio": 1.1512859703049743,
      "metrics": {
        "lv_current_density_a_mm2": 2.1566597653554176,
        "hv_current_density_a_mm2": 2.179009168137174,
        "flux_density_t": 1.6,
        "no_load_loss_w": 17282.373811991543,
        "load_loss_w": 62480.085046446955,
        "reactance_pct": 7.525323656002343,
        "resistance_pct": 0.31240042523223477,
        "impedance_pct": 7.531805238671122,
        "mean_oil_rise_k": 42.44752563698517,
        "top_oil_rise_k": 50.9370307643822,
        "lv_average_winding_rise_k": 63.944097356983576,
        "hv_average_winding_rise_k": 58.94219135520819,
        "core_mass_kg": 10788.058493036711,
        "copper_mass_kg": 4985.863158734885,
        "tank_mass_kg": 7164.598431809937,
        "oil_mass_kg": 24698.48310025298,
        "capex_usd": 279564.27058575075,
        "efficiency_50pct": 0.9967205506663994,
        "efficiency_100pct": 0.9960277190020618
      }
    },
    "scope_warning": "The policy generated and passed this candidate through the nominal fast verifier. A 20 MVA YNd transformer is outside the frozen IronBench-Sim-v1 75--2500 kVA Dyn domain, so no authoritative or detailed-solver compliance claim is made.",
    "audit": {
      "solver_domain": "IronBench-Utility-Exploratory-v0",
      "run": {
        "modal_run_url": "https://modal.com/apps/generaltrajectory/main/ap-lL1ff2U7mZqJjECH5nCH6G",
        "raw_results_path": "runs/tinker-20mva-gsu-detailed-results-v3.json",
        "raw_results_sha256": "ddf9bf89b3c33152f2bda05898f8d26e125992a116e52aa9fcce342e63a503ad",
        "solver_cache_fingerprint": "a5d33f6d2041a39418cf2a8151c466e1c0e7227d98f33e2a6acae8a838e9ed48",
        "suite_wall_seconds": 353.2365273339674,
        "all_stages_completed": true,
        "all_model_hashes_match": true,
        "all_numerical_gates_passed": false
      },
      "conclusion": {
        "status": "numerically_incomplete",
        "detailed_audit_passed": false,
        "failed_checks": [
          "top_oil_rise",
          "lv_avg_winding_rise",
          "hv_avg_winding_rise"
        ],
        "unconverged_checks": [
          "openfoam_onan"
        ],
        "scope_warning": "Six of seven detailed stages converged, but whole-transformer OpenFOAM ONAN did not satisfy iteration or heat-balance convergence. The stricter reduced-order ONAN check independently exceeds all 65 K thermal-rise limits. This exploratory utility-scale design is therefore not detailed-verified and remains outside the calibrated IronBench-Sim-v1 reward domain."
      },
      "fast_thermal_cross_check": {
        "model": "ironbench.onan_thermal_network",
        "top_oil_rise_k": 76.2578628153465,
        "lv_average_winding_rise_k": 85.00903505198802,
        "hv_average_winding_rise_k": 80.65834651758769,
        "limit_k": 65.0,
        "compliant": false
      },
      "stages": {
        "getdp_leakage": {
          "ok": true,
          "numerically_converged": true,
          "elapsed_seconds": 20.988598762,
          "authoritative": false
        },
        "getdp_dielectric": {
          "ok": true,
          "numerically_converged": true,
          "elapsed_seconds": 5.486648864999999,
          "authoritative": false
        },
        "elmer_magnetic": {
          "ok": true,
          "numerically_converged": true,
          "elapsed_seconds": 10.352011044000001,
          "authoritative": false
        },
        "elmer_core_loss": {
          "ok": true,
          "numerically_converged": true,
          "elapsed_seconds": 17.212301122,
          "authoritative": false
        },
        "elmer_harmonic_loss": {
          "ok": true,
          "numerically_converged": true,
          "elapsed_seconds": 14.870068325999998,
          "authoritative": false
        },
        "elmer_short_circuit_thermal": {
          "ok": true,
          "numerically_converged": true,
          "elapsed_seconds": 19.348662781999998,
          "authoritative": false
        },
        "openfoam_onan": {
          "ok": false,
          "numerically_converged": false,
          "elapsed_seconds": 310.407560547,
          "authoritative": false
        }
      }
    }
  },
  "inference": {
    "baseline": {
      "provider": "Tinker",
      "completion_tokens": 30720,
      "elapsed_seconds": 869.085884809494,
      "generated_tokens_per_second": 35.34748468125668,
      "scope": "one retained IronBench reasoning trajectory; not aggregate service capacity"
    },
    "runtime": {
      "model": "nvidia/Kimi-K2.6-NVFP4",
      "model_revision": "2fd3a800dedd098b8327eb49e93ebc75f85da19f",
      "draft_model": "lightseekorg/kimi-k2.6-eagle3.1-mla",
      "draft_model_revision": "35194ee8feb2826812f716eb42a924f99a5404f3",
      "engine_image": "vllm/vllm-openai:v0.26.0-ubuntu2404",
      "hardware": "4x NVIDIA B200 183359 MiB",
      "modal_gpu_hourly_usd": 25.0,
      "kv_cache_dtype": "fp8",
      "output_tokens_per_request": 1024
    },
    "target_only": {
      "startup_seconds": 1633.5132261059998,
      "aggregate_generated_tokens_per_second": {
        "1": 145.32431981549075,
        "4": 508.86587592410655,
        "8": 1009.1534646609872
      }
    },
    "eagle_3_1": {
      "startup_seconds": 1773.448387219,
      "aggregate_generated_tokens_per_second": {
        "1": 265.3104539319237,
        "4": 143.8200909078386,
        "8": 124.53765196985312
      },
      "speedup_over_target_only": {
        "1": 1.8256438720564587,
        "4": 0.28262868019330156,
        "8": 0.12340804082924099
      }
    }
  }
}
