{
  "schema_version": 2,
  "device": "cpu",
  "accelerator": "CPU",
  "seeds": [
    42,
    43,
    44
  ],
  "code_commit": "7b78ce1dbfc4006fbc81fdb181c3ead45aeb2457",
  "source_sha256": {
    "scripts/run_oct24_overlap_components.py": "ae6fbb18c6697e02f1e8c0f0d613828fa3006603c46e58cd12a6208f17ca143e",
    "scripts/oct18_frozen_sources.py": "da08355049c1f6b5184ee659643c9be9decedf2041ef6a87a3dce2e4582e385a",
    "src/auto_research/agent_research/physical_engineering_exam.py": "5c979d63b51b954ec11575efff76d4b7cbd64bb1f43cba6d27a61742d4b22132",
    "src/auto_research/foundation_models/ttr_compressor.py": "3d4b4dafc53fe0a5bc685a4fa236ae4d03b6449b969575f811fb79fbdf3cdd2d",
    "src/auto_research/post_training/userarena_optd.py": "fac048b0d687007becc4acf4a270f5287812f80d523052243f0fdd7e5c52fa01",
    "src/auto_research/post_training/moer.py": "d2d2dbfee1cc54207dc942b083eb1540c78fc735b3543446c1414e921fe09a1d"
  },
  "components": {
    "physical-engineering-exam": {
      "per_seed": [
        {
          "seed": 42,
          "checks": {
            "declared_deliverable_usable": true,
            "missing_task_not_usable": true,
            "manifest_reward_not_a_grade": true,
            "hard_constraint_gate_caps_score": true,
            "full_task_denominator_and_three_runs": true,
            "blob_provenance_consistent": true,
            "strict_reproduction_threshold": true
          },
          "metrics": {
            "contract_cases": 7.0,
            "model_calls": 0.0,
            "container_runs": 0.0,
            "external_commands": 0.0
          },
          "evaluation_tier": "l1_mechanism",
          "diagnostic_only": true,
          "promotion_eligible": false,
          "evolve_integrated": false,
          "requires_gpu_validation": false,
          "full_experiment_status": "not_reproduced_high_cost",
          "upstream_revision": "70ed96281d87380926a0d7c08e21799fd5edee63",
          "execution_device": "cpu",
          "execution_accelerator": "CPU"
        },
        {
          "seed": 43,
          "checks": {
            "declared_deliverable_usable": true,
            "missing_task_not_usable": true,
            "manifest_reward_not_a_grade": true,
            "hard_constraint_gate_caps_score": true,
            "full_task_denominator_and_three_runs": true,
            "blob_provenance_consistent": true,
            "strict_reproduction_threshold": true
          },
          "metrics": {
            "contract_cases": 7.0,
            "model_calls": 0.0,
            "container_runs": 0.0,
            "external_commands": 0.0
          },
          "evaluation_tier": "l1_mechanism",
          "diagnostic_only": true,
          "promotion_eligible": false,
          "evolve_integrated": false,
          "requires_gpu_validation": false,
          "full_experiment_status": "not_reproduced_high_cost",
          "upstream_revision": "70ed96281d87380926a0d7c08e21799fd5edee63",
          "execution_device": "cpu",
          "execution_accelerator": "CPU"
        },
        {
          "seed": 44,
          "checks": {
            "declared_deliverable_usable": true,
            "missing_task_not_usable": true,
            "manifest_reward_not_a_grade": true,
            "hard_constraint_gate_caps_score": true,
            "full_task_denominator_and_three_runs": true,
            "blob_provenance_consistent": true,
            "strict_reproduction_threshold": true
          },
          "metrics": {
            "contract_cases": 7.0,
            "model_calls": 0.0,
            "container_runs": 0.0,
            "external_commands": 0.0
          },
          "evaluation_tier": "l1_mechanism",
          "diagnostic_only": true,
          "promotion_eligible": false,
          "evolve_integrated": false,
          "requires_gpu_validation": false,
          "full_experiment_status": "not_reproduced_high_cost",
          "upstream_revision": "70ed96281d87380926a0d7c08e21799fd5edee63",
          "execution_device": "cpu",
          "execution_accelerator": "CPU"
        }
      ]
    },
    "ttr-compressor": {
      "per_seed": [
        {
          "seed": 42,
          "metrics": {
            "primal_dual_max_abs_error": 9.019388675923778e-07,
            "a_gradient_max_abs_error": 1.2564587983376896e-06,
            "rho_gradient_max_abs_error": 9.700646141919123e-07,
            "permutation_max_abs_error": 4.76837158203125e-07,
            "global_scale_max_abs_error": 0.0,
            "normalized_target_negative_control_gap": 3.1190908473268077,
            "repeated_evidence_max_abs_error": 2.9802322387695312e-08,
            "token_gradient_norm": 16.298301696777344,
            "outer_fixture_loss_before": 2.291396141052246,
            "outer_fixture_loss_after": 2.288318634033203,
            "groups": 4,
            "tokens_per_group": 7,
            "dimension": 12,
            "outer_steps": 1
          },
          "checks": {
            "dual_matches_independent_primal": true,
            "outer_a_gradient_eq9": true,
            "outer_rho_gradient_eq9": true,
            "token_gradient_finite_nonzero": true,
            "base_token_permutation_invariance": true,
            "unnormalized_target_scale_response": true,
            "normalized_target_negative_control_differs": true,
            "repeated_evidence_closed_form": true,
            "actual_outer_step_updates_both_parameters": true,
            "outer_step_finite_nonincreasing_fixture_loss": true
          },
          "execution_device": "cpu",
          "execution_accelerator": "CPU"
        },
        {
          "seed": 43,
          "metrics": {
            "primal_dual_max_abs_error": 7.3019780044703e-07,
            "a_gradient_max_abs_error": 1.8868610700550903e-06,
            "rho_gradient_max_abs_error": 1.8978941094260904e-06,
            "permutation_max_abs_error": 8.344650268554688e-07,
            "global_scale_max_abs_error": 0.0,
            "normalized_target_negative_control_gap": 6.583786463409712,
            "repeated_evidence_max_abs_error": 2.9802322387695312e-08,
            "token_gradient_norm": 18.85676383972168,
            "outer_fixture_loss_before": 6.530484676361084,
            "outer_fixture_loss_after": 6.496818542480469,
            "groups": 4,
            "tokens_per_group": 7,
            "dimension": 12,
            "outer_steps": 1
          },
          "checks": {
            "dual_matches_independent_primal": true,
            "outer_a_gradient_eq9": true,
            "outer_rho_gradient_eq9": true,
            "token_gradient_finite_nonzero": true,
            "base_token_permutation_invariance": true,
            "unnormalized_target_scale_response": true,
            "normalized_target_negative_control_differs": true,
            "repeated_evidence_closed_form": true,
            "actual_outer_step_updates_both_parameters": true,
            "outer_step_finite_nonincreasing_fixture_loss": true
          },
          "execution_device": "cpu",
          "execution_accelerator": "CPU"
        },
        {
          "seed": 44,
          "metrics": {
            "primal_dual_max_abs_error": 1.4137845232298218e-06,
            "a_gradient_max_abs_error": 2.232731237228336e-06,
            "rho_gradient_max_abs_error": 1.548664263495425e-06,
            "permutation_max_abs_error": 9.5367431640625e-07,
            "global_scale_max_abs_error": 0.0,
            "normalized_target_negative_control_gap": 6.999853148019153,
            "repeated_evidence_max_abs_error": 2.9802322387695312e-08,
            "token_gradient_norm": 14.562341690063477,
            "outer_fixture_loss_before": 6.388666152954102,
            "outer_fixture_loss_after": 6.368872165679932,
            "groups": 4,
            "tokens_per_group": 7,
            "dimension": 12,
            "outer_steps": 1
          },
          "checks": {
            "dual_matches_independent_primal": true,
            "outer_a_gradient_eq9": true,
            "outer_rho_gradient_eq9": true,
            "token_gradient_finite_nonzero": true,
            "base_token_permutation_invariance": true,
            "unnormalized_target_scale_response": true,
            "normalized_target_negative_control_differs": true,
            "repeated_evidence_closed_form": true,
            "actual_outer_step_updates_both_parameters": true,
            "outer_step_finite_nonincreasing_fixture_loss": true
          },
          "execution_device": "cpu",
          "execution_accelerator": "CPU"
        }
      ]
    },
    "userarena-optd": {
      "per_seed": [
        {
          "seed": 42,
          "diagnostic_only": true,
          "component": "userarena_optd_equations_5_to_10",
          "metrics": {
            "loss_before": 0.3610756059799968,
            "loss_after": 0.36105673461107346,
            "formula_reference_error": 0.0,
            "parameter_delta": 0.0006064822147348367,
            "max_teacher_kl": 0.01997002332031962,
            "max_lambda": 0.25390625,
            "distillation": 0.10932683832207112,
            "grounding": 2.517487676579257,
            "optimizer_updates": 1,
            "model_calls": 0
          },
          "checks": {
            "teacher_detached": true,
            "formula_reference_matches": true,
            "per_token_trust_budget": true,
            "zero_budget_recovers_unconditioned": true,
            "bounded_jsd": true,
            "finite_gradients": true,
            "real_parameter_update": true,
            "separate_grounding_branch_updated": true,
            "padding_gradient_zero": true,
            "fixed_objective_decreases": true
          },
          "execution_device": "cpu",
          "execution_accelerator": "CPU"
        },
        {
          "seed": 43,
          "diagnostic_only": true,
          "component": "userarena_optd_equations_5_to_10",
          "metrics": {
            "loss_before": 0.327604761776583,
            "loss_after": 0.32758552576957184,
            "formula_reference_error": 5.551115123125783e-17,
            "parameter_delta": 0.000619657599621257,
            "max_teacher_kl": 0.01999728273275002,
            "max_lambda": 0.296875,
            "distillation": 0.12105869802170939,
            "grounding": 2.065460637548736,
            "optimizer_updates": 1,
            "model_calls": 0
          },
          "checks": {
            "teacher_detached": true,
            "formula_reference_matches": true,
            "per_token_trust_budget": true,
            "zero_budget_recovers_unconditioned": true,
            "bounded_jsd": true,
            "finite_gradients": true,
            "real_parameter_update": true,
            "separate_grounding_branch_updated": true,
            "padding_gradient_zero": true,
            "fixed_objective_decreases": true
          },
          "execution_device": "cpu",
          "execution_accelerator": "CPU"
        },
        {
          "seed": 44,
          "diagnostic_only": true,
          "component": "userarena_optd_equations_5_to_10",
          "metrics": {
            "loss_before": 0.35333988231310076,
            "loss_after": 0.35331923892490147,
            "formula_reference_error": 5.551115123125783e-17,
            "parameter_delta": 0.0006424830055630282,
            "max_teacher_kl": 0.019997852524657227,
            "max_lambda": 0.3359375,
            "distillation": 0.1421531533879938,
            "grounding": 2.1118672892510695,
            "optimizer_updates": 1,
            "model_calls": 0
          },
          "checks": {
            "teacher_detached": true,
            "formula_reference_matches": true,
            "per_token_trust_budget": true,
            "zero_budget_recovers_unconditioned": true,
            "bounded_jsd": true,
            "finite_gradients": true,
            "real_parameter_update": true,
            "separate_grounding_branch_updated": true,
            "padding_gradient_zero": true,
            "fixed_objective_decreases": true
          },
          "execution_device": "cpu",
          "execution_accelerator": "CPU"
        }
      ]
    },
    "moer": {
      "per_seed": [
        {
          "seed": 42,
          "diagnostic_only": true,
          "component": "paper_eq_1_3_6_single_candidate",
          "metrics": {
            "loss_before": 0.10613693748543677,
            "loss_after": 0.10609197674354218,
            "gradient_norm": 0.06741184452650403,
            "parameter_delta": 0.0006741184452649588,
            "formula_reference_error": 1.3877787807814457e-17,
            "gradient_reference_error": 3.469446951953614e-18,
            "valid_tokens": 17,
            "optimizer_updates": 1,
            "model_calls": 0
          },
          "checks": {
            "formula_reference_matches": true,
            "gradient_reference_matches": true,
            "old_policy_detached": true,
            "rewards_detached": true,
            "finite_gradient": true,
            "real_parameter_update": true,
            "padding_zero_gradient": true,
            "union_group_mean_no_std": true,
            "global_token_normalization": true,
            "guided_completion_is_off_policy": true,
            "trust_region_gradient_not_cancelled": true,
            "rollback_replaces_only_failed_candidate": true,
            "cosine_schedule_endpoints": true
          },
          "execution_device": "cpu",
          "execution_accelerator": "CPU"
        },
        {
          "seed": 43,
          "diagnostic_only": true,
          "component": "paper_eq_1_3_6_single_candidate",
          "metrics": {
            "loss_before": 0.10196078431372546,
            "loss_after": 0.10192486720816524,
            "gradient_norm": 0.0611095287423114,
            "parameter_delta": 0.0006110952874231185,
            "formula_reference_error": 1.3877787807814457e-17,
            "gradient_reference_error": 6.938893903907228e-18,
            "valid_tokens": 17,
            "optimizer_updates": 1,
            "model_calls": 0
          },
          "checks": {
            "formula_reference_matches": true,
            "gradient_reference_matches": true,
            "old_policy_detached": true,
            "rewards_detached": true,
            "finite_gradient": true,
            "real_parameter_update": true,
            "padding_zero_gradient": true,
            "union_group_mean_no_std": true,
            "global_token_normalization": true,
            "guided_completion_is_off_policy": true,
            "trust_region_gradient_not_cancelled": true,
            "rollback_replaces_only_failed_candidate": true,
            "cosine_schedule_endpoints": true
          },
          "execution_device": "cpu",
          "execution_accelerator": "CPU"
        },
        {
          "seed": 44,
          "diagnostic_only": true,
          "component": "paper_eq_1_3_6_single_candidate",
          "metrics": {
            "loss_before": 0.10196078431372546,
            "loss_after": 0.10192073453989252,
            "gradient_norm": 0.06445100494022657,
            "parameter_delta": 0.0006445100494022486,
            "formula_reference_error": 1.3877787807814457e-17,
            "gradient_reference_error": 6.938893903907228e-18,
            "valid_tokens": 17,
            "optimizer_updates": 1,
            "model_calls": 0
          },
          "checks": {
            "formula_reference_matches": true,
            "gradient_reference_matches": true,
            "old_policy_detached": true,
            "rewards_detached": true,
            "finite_gradient": true,
            "real_parameter_update": true,
            "padding_zero_gradient": true,
            "union_group_mean_no_std": true,
            "global_token_normalization": true,
            "guided_completion_is_off_policy": true,
            "trust_region_gradient_not_cancelled": true,
            "rollback_replaces_only_failed_candidate": true,
            "cosine_schedule_endpoints": true
          },
          "execution_device": "cpu",
          "execution_accelerator": "CPU"
        }
      ]
    }
  },
  "elapsed_seconds": 4.907266708090901,
  "evaluation_tier": "l1_mechanism",
  "diagnostic_only": true,
  "formal_comparison": false,
  "promotion_eligible": false,
  "evolve_integrated": false,
  "pretrained_model_calls": 0,
  "full_experiments_executed": false
}
