{
  "schema_version": 1,
  "created_at_utc": "2026-09-22T09:16:40.648130+00:00",
  "configuration_sha256": "c3ef9d5fbce230d4b87a109e53ccffade53f6dc01f4008f5a213c7c11364c0e5",
  "experiment": "Separately frozen same-gain PD-only versus FF+PD ablation after initial benchmark",
  "episodes": 240,
  "seeds": [
    200,
    201,
    202,
    203,
    204,
    205,
    206,
    207,
    208,
    209,
    210,
    211,
    212,
    213,
    214,
    215,
    216,
    217,
    218,
    219
  ],
  "conditions": [
    {
      "condition": "nominal",
      "policies": {
        "ff_only": {
          "episodes": 20,
          "rmse_deg_mean": 69.60432037251967,
          "rmse_deg_sd": 15.1493520229937,
          "completion_count": 0,
          "completion_fraction": 0.0,
          "stage_completion_fraction": 0.2800000000000001,
          "peak_error_deg_max": 241.47203776100332,
          "peak_joint_limit_violation_deg_max": 4.489075421650467,
          "mean_saturation_fraction": 0.003570918440222825,
          "recovered_count": null,
          "recovery_time_mean_s": null,
          "mean_absolute_mechanical_work_j": 92.17355472755797
        },
        "pd_only": {
          "episodes": 20,
          "rmse_deg_mean": 3.1099252128328914,
          "rmse_deg_sd": 2.3010411745503546e-08,
          "completion_count": 20,
          "completion_fraction": 1.0,
          "stage_completion_fraction": 1.0,
          "peak_error_deg_max": 23.631739367563313,
          "peak_joint_limit_violation_deg_max": 0.0,
          "mean_saturation_fraction": 0.006677617483216683,
          "recovered_count": null,
          "recovery_time_mean_s": null,
          "mean_absolute_mechanical_work_j": 75.77502539672025
        },
        "ff_pd": {
          "episodes": 20,
          "rmse_deg_mean": 0.6598702757614209,
          "rmse_deg_sd": 2.4006559010903236e-11,
          "completion_count": 20,
          "completion_fraction": 1.0,
          "stage_completion_fraction": 1.0,
          "peak_error_deg_max": 12.100295138166034,
          "peak_joint_limit_violation_deg_max": 0.0,
          "mean_saturation_fraction": 0.006034852163976575,
          "recovered_count": null,
          "recovery_time_mean_s": null,
          "mean_absolute_mechanical_work_j": 69.79537629767638
        }
      },
      "feedforward_rmse_improvement_deg": {
        "mean": 2.45005493707147,
        "ci95_percentile": [
          2.4500549270978547,
          2.4500549465263712
        ],
        "method": "Paired-seed nonparametric bootstrap,10000 resamples; internal simulator replicates, not real-world confidence."
      },
      "feedforward_completion_difference": {
        "mean": 0.0,
        "ci95_percentile": [
          0.0,
          0.0
        ],
        "method": "Paired-seed nonparametric bootstrap,10000 resamples; internal simulator replicates, not real-world confidence."
      },
      "feedforward_rmse_better_seed_count": 20
    },
    {
      "condition": "impulse",
      "policies": {
        "ff_only": {
          "episodes": 20,
          "rmse_deg_mean": 65.7574246503428,
          "rmse_deg_sd": 14.475472330691655,
          "completion_count": 0,
          "completion_fraction": 0.0,
          "stage_completion_fraction": 0.21000000000000005,
          "peak_error_deg_max": 241.47203775135085,
          "peak_joint_limit_violation_deg_max": 6.5052329473887225,
          "mean_saturation_fraction": 0.003570918440222825,
          "recovered_count": 20,
          "recovery_time_mean_s": 6.85248361017648,
          "mean_absolute_mechanical_work_j": 97.66167411297454
        },
        "pd_only": {
          "episodes": 20,
          "rmse_deg_mean": 3.1308670034426545,
          "rmse_deg_sd": 0.020901549953098727,
          "completion_count": 20,
          "completion_fraction": 1.0,
          "stage_completion_fraction": 1.0,
          "peak_error_deg_max": 23.631739367577012,
          "peak_joint_limit_violation_deg_max": 0.0,
          "mean_saturation_fraction": 0.006677617483216683,
          "recovered_count": 20,
          "recovery_time_mean_s": 0.0008836101764794879,
          "mean_absolute_mechanical_work_j": 76.27555703543351
        },
        "ff_pd": {
          "episodes": 20,
          "rmse_deg_mean": 0.7623685821885438,
          "rmse_deg_sd": 0.01769818308828679,
          "completion_count": 20,
          "completion_fraction": 1.0,
          "stage_completion_fraction": 1.0,
          "peak_error_deg_max": 12.100295138172994,
          "peak_joint_limit_violation_deg_max": 0.0,
          "mean_saturation_fraction": 0.006034852163976575,
          "recovered_count": 20,
          "recovery_time_mean_s": 0.0008836101764794879,
          "mean_absolute_mechanical_work_j": 70.29277396510174
        }
      },
      "feedforward_rmse_improvement_deg": {
        "mean": 2.368498421254111,
        "ci95_percentile": [
          2.356476135053861,
          2.3800028968812126
        ],
        "method": "Paired-seed nonparametric bootstrap,10000 resamples; internal simulator replicates, not real-world confidence."
      },
      "feedforward_completion_difference": {
        "mean": 0.0,
        "ci95_percentile": [
          0.0,
          0.0
        ],
        "method": "Paired-seed nonparametric bootstrap,10000 resamples; internal simulator replicates, not real-world confidence."
      },
      "feedforward_rmse_better_seed_count": 20
    },
    {
      "condition": "actuator_gain",
      "policies": {
        "ff_only": {
          "episodes": 20,
          "rmse_deg_mean": 56.25450459046465,
          "rmse_deg_sd": 0.6801647787886537,
          "completion_count": 0,
          "completion_fraction": 0.0,
          "stage_completion_fraction": 0.20000000000000004,
          "peak_error_deg_max": 182.24963181896973,
          "peak_joint_limit_violation_deg_max": 0.0,
          "mean_saturation_fraction": 0.003570918440222825,
          "recovered_count": null,
          "recovery_time_mean_s": null,
          "mean_absolute_mechanical_work_j": 69.17097943529384
        },
        "pd_only": {
          "episodes": 20,
          "rmse_deg_mean": 4.7485747761171755,
          "rmse_deg_sd": 0.526685240151829,
          "completion_count": 20,
          "completion_fraction": 1.0,
          "stage_completion_fraction": 1.0,
          "peak_error_deg_max": 50.5534081501461,
          "peak_joint_limit_violation_deg_max": 0.0,
          "mean_saturation_fraction": 0.014978217397514638,
          "recovered_count": null,
          "recovery_time_mean_s": null,
          "mean_absolute_mechanical_work_j": 79.62417736901489
        },
        "ff_pd": {
          "episodes": 20,
          "rmse_deg_mean": 2.3061715184689597,
          "rmse_deg_sd": 0.5754022925867488,
          "completion_count": 20,
          "completion_fraction": 1.0,
          "stage_completion_fraction": 1.0,
          "peak_error_deg_max": 40.42528231309074,
          "peak_joint_limit_violation_deg_max": 0.0,
          "mean_saturation_fraction": 0.014287244679331526,
          "recovered_count": null,
          "recovery_time_mean_s": null,
          "mean_absolute_mechanical_work_j": 73.04423465839892
        }
      },
      "feedforward_rmse_improvement_deg": {
        "mean": 2.442403257648216,
        "ci95_percentile": [
          2.40507124649886,
          2.476979818933939
        ],
        "method": "Paired-seed nonparametric bootstrap,10000 resamples; internal simulator replicates, not real-world confidence."
      },
      "feedforward_completion_difference": {
        "mean": 0.0,
        "ci95_percentile": [
          0.0,
          0.0
        ],
        "method": "Paired-seed nonparametric bootstrap,10000 resamples; internal simulator replicates, not real-world confidence."
      },
      "feedforward_rmse_better_seed_count": 20
    },
    {
      "condition": "combined",
      "policies": {
        "ff_only": {
          "episodes": 20,
          "rmse_deg_mean": 60.50264124532157,
          "rmse_deg_sd": 0.8601566553431147,
          "completion_count": 0,
          "completion_fraction": 0.0,
          "stage_completion_fraction": 0.04,
          "peak_error_deg_max": 227.64246000837167,
          "peak_joint_limit_violation_deg_max": 2.20198097720803,
          "mean_saturation_fraction": 0.003570918440222825,
          "recovered_count": 16,
          "recovery_time_mean_s": 6.589925727949327,
          "mean_absolute_mechanical_work_j": 67.53212537580605
        },
        "pd_only": {
          "episodes": 20,
          "rmse_deg_mean": 11.65608975338323,
          "rmse_deg_sd": 5.7814858028810665,
          "completion_count": 13,
          "completion_fraction": 0.65,
          "stage_completion_fraction": 0.8899999999999999,
          "peak_error_deg_max": 123.53159221447237,
          "peak_joint_limit_violation_deg_max": 0.0,
          "mean_saturation_fraction": 0.05988430224253679,
          "recovered_count": 20,
          "recovery_time_mean_s": 0.5684836101764795,
          "mean_absolute_mechanical_work_j": 106.80367361170777
        },
        "ff_pd": {
          "episodes": 20,
          "rmse_deg_mean": 10.02366790720219,
          "rmse_deg_sd": 6.498091646053419,
          "completion_count": 14,
          "completion_fraction": 0.7,
          "stage_completion_fraction": 0.9,
          "peak_error_deg_max": 126.37324194799434,
          "peak_joint_limit_violation_deg_max": 0.0,
          "mean_saturation_fraction": 0.06200364233680903,
          "recovered_count": 20,
          "recovery_time_mean_s": 0.5234836101764795,
          "mean_absolute_mechanical_work_j": 100.46933250227487
        }
      },
      "feedforward_rmse_improvement_deg": {
        "mean": 1.6324218461810382,
        "ci95_percentile": [
          1.2990074013186705,
          1.9637206713671016
        ],
        "method": "Paired-seed nonparametric bootstrap,10000 resamples; internal simulator replicates, not real-world confidence."
      },
      "feedforward_completion_difference": {
        "mean": 0.05,
        "ci95_percentile": [
          0.0,
          0.15
        ],
        "method": "Paired-seed nonparametric bootstrap,10000 resamples; internal simulator replicates, not real-world confidence."
      },
      "feedforward_rmse_better_seed_count": 20
    }
  ],
  "compute_wall_s": 27.799148299964145,
  "limitations": [
    "Designed after original evaluation; fresh internal simulator seeds, not external validation.",
    "Same gains across PD-only and FF+PD, not separately optimized policies.",
    "One synthetic video-derived reference and four planar arm DOF only.",
    "No change to main benchmark controller gains, torque limits, metrics or stage windows."
  ]
}