{
  "arms": {
    "baseline": {
      "abstention_rate": 1.0,
      "adapt_selection_rate": 0.0,
      "ambiguous_write_count": 0,
      "arm": "baseline",
      "execution_error_rate": 0.525,
      "fallback_rate": 0.0,
      "mean_automated_score": -28.8,
      "mean_final_score": 4.3,
      "mean_score_auc": 9.55,
      "mean_steps": 4.0,
      "operational_failure_count": 0,
      "operational_failure_rate": 0.0,
      "pass_rate": 0.0,
      "pass_rate_ci95_high": 0.2775401687666166,
      "pass_rate_ci95_low": 0.0,
      "sample_count": 10,
      "success_count": 0,
      "tasks": [
        "iron_gear_wheel_throughput",
        "iron_ore_throughput"
      ],
      "total_adapt_latency_seconds": 0.0,
      "total_model_latency_seconds": 1757.8781490810006,
      "total_tokens": 114465
    },
    "domain_only": {
      "abstention_rate": 0.0,
      "adapt_selection_rate": 1.0,
      "ambiguous_write_count": 0,
      "arm": "domain_only",
      "execution_error_rate": 0.35,
      "fallback_rate": 0.0,
      "mean_automated_score": -29.1,
      "mean_final_score": 0.9,
      "mean_score_auc": 1.35,
      "mean_steps": 4.0,
      "operational_failure_count": 0,
      "operational_failure_rate": 0.0,
      "pass_rate": 0.0,
      "pass_rate_ci95_high": 0.2775401687666166,
      "pass_rate_ci95_low": 0.0,
      "sample_count": 10,
      "success_count": 0,
      "tasks": [
        "iron_gear_wheel_throughput",
        "iron_ore_throughput"
      ],
      "total_adapt_latency_seconds": 17.814461480995305,
      "total_model_latency_seconds": 1926.3853546639966,
      "total_tokens": 128995
    },
    "memory_only": {
      "abstention_rate": 1.0,
      "adapt_selection_rate": 0.0,
      "ambiguous_write_count": 0,
      "arm": "memory_only",
      "execution_error_rate": 0.55,
      "fallback_rate": 0.0,
      "mean_automated_score": -29.1,
      "mean_final_score": 0.9,
      "mean_score_auc": 1.95,
      "mean_steps": 4.0,
      "operational_failure_count": 0,
      "operational_failure_rate": 0.0,
      "pass_rate": 0.0,
      "pass_rate_ci95_high": 0.2775401687666166,
      "pass_rate_ci95_low": 0.0,
      "sample_count": 10,
      "success_count": 0,
      "tasks": [
        "iron_gear_wheel_throughput",
        "iron_ore_throughput"
      ],
      "total_adapt_latency_seconds": 6.9264798930034885,
      "total_model_latency_seconds": 1983.483683827999,
      "total_tokens": 131199
    },
    "warm_frozen": {
      "abstention_rate": 0.0,
      "adapt_selection_rate": 1.0,
      "ambiguous_write_count": 0,
      "arm": "warm_frozen",
      "execution_error_rate": 0.325,
      "fallback_rate": 0.0,
      "mean_automated_score": -30.0,
      "mean_final_score": 0.0,
      "mean_score_auc": 0.0,
      "mean_steps": 4.0,
      "operational_failure_count": 0,
      "operational_failure_rate": 0.0,
      "pass_rate": 0.0,
      "pass_rate_ci95_high": 0.2775401687666166,
      "pass_rate_ci95_low": 0.0,
      "sample_count": 10,
      "success_count": 0,
      "tasks": [
        "iron_gear_wheel_throughput",
        "iron_ore_throughput"
      ],
      "total_adapt_latency_seconds": 23.06145454299258,
      "total_model_latency_seconds": 1702.3130158450003,
      "total_tokens": 138262
    }
  },
  "comparison_fingerprint": "6d9ef73080714dfbb12d5b9346b07ca62c69e96b8fae43e7e47352acbb8ba058",
  "duplicate_cells": [],
  "experiment_complete": true,
  "experiment_id": "experiment-20260811T041432Z-3ff8ac0a",
  "generated_from": ".fle/adapt1/local-research-v1",
  "missing_cells": [],
  "runs": [
    {
      "abstention_count": 4,
      "adapt_latency_seconds": 0.0,
      "adapt_selection_count": 0,
      "ambiguous_write_count": 0,
      "completion_status": "trajectory_limit",
      "execution_error_count": 2,
      "fallback_count": 0,
      "final_automated_score": -30.0,
      "final_score": 0.0,
      "mode": "baseline",
      "model_latency_seconds": 184.104069635001,
      "normalized_reward_sum": -0.5,
      "operational_failure": false,
      "run_id": "eval-baseline-iron-gear-wheel-throughput-20260811T043805Z-9dbb4d13",
      "score_auc": 0.0,
      "steps": 4,
      "success": false,
      "token_count": 11495
    },
    {
      "abstention_count": 4,
      "adapt_latency_seconds": 0.0,
      "adapt_selection_count": 0,
      "ambiguous_write_count": 0,
      "completion_status": "trajectory_limit",
      "execution_error_count": 2,
      "fallback_count": 0,
      "final_automated_score": -30.0,
      "final_score": 0.0,
      "mode": "baseline",
      "model_latency_seconds": 103.7113160079989,
      "normalized_reward_sum": -0.5,
      "operational_failure": false,
      "run_id": "eval-baseline-iron-gear-wheel-throughput-20260811T044153Z-9a6eb1a3",
      "score_auc": 0.0,
      "steps": 4,
      "success": false,
      "token_count": 11507
    },
    {
      "abstention_count": 4,
      "adapt_latency_seconds": 0.0,
      "adapt_selection_count": 0,
      "ambiguous_write_count": 0,
      "completion_status": "trajectory_limit",
      "execution_error_count": 2,
      "fallback_count": 0,
      "final_automated_score": -30.0,
      "final_score": 0.0,
      "mode": "baseline",
      "model_latency_seconds": 99.97653019400059,
      "normalized_reward_sum": -0.5,
      "operational_failure": false,
      "run_id": "eval-baseline-iron-gear-wheel-throughput-20260811T044421Z-11478b76",
      "score_auc": 0.0,
      "steps": 4,
      "success": false,
      "token_count": 11495
    },
    {
      "abstention_count": 4,
      "adapt_latency_seconds": 0.0,
      "adapt_selection_count": 0,
      "ambiguous_write_count": 0,
      "completion_status": "trajectory_limit",
      "execution_error_count": 2,
      "fallback_count": 0,
      "final_automated_score": -30.0,
      "final_score": 0.0,
      "mode": "baseline",
      "model_latency_seconds": 99.32107656200151,
      "normalized_reward_sum": -0.5,
      "operational_failure": false,
      "run_id": "eval-baseline-iron-gear-wheel-throughput-20260811T044644Z-f662d7b3",
      "score_auc": 0.0,
      "steps": 4,
      "success": false,
      "token_count": 11495
    },
    {
      "abstention_count": 4,
      "adapt_latency_seconds": 0.0,
      "adapt_selection_count": 0,
      "ambiguous_write_count": 0,
      "completion_status": "trajectory_limit",
      "execution_error_count": 2,
      "fallback_count": 0,
      "final_automated_score": -30.0,
      "final_score": 0.0,
      "mode": "baseline",
      "model_latency_seconds": 94.28871270400032,
      "normalized_reward_sum": -0.5,
      "operational_failure": false,
      "run_id": "eval-baseline-iron-gear-wheel-throughput-20260811T044908Z-68bb2c8b",
      "score_auc": 0.0,
      "steps": 4,
      "success": false,
      "token_count": 11495
    },
    {
      "abstention_count": 4,
      "adapt_latency_seconds": 0.0,
      "adapt_selection_count": 0,
      "ambiguous_write_count": 0,
      "completion_status": "trajectory_limit",
      "execution_error_count": 1,
      "fallback_count": 0,
      "final_automated_score": -30.0,
      "final_score": 0.0,
      "mode": "baseline",
      "model_latency_seconds": 212.3591295439992,
      "normalized_reward_sum": -0.25,
      "operational_failure": false,
      "run_id": "eval-baseline-iron-ore-throughput-20260811T041432Z-93ba39d0",
      "score_auc": 0.0,
      "steps": 4,
      "success": false,
      "token_count": 10784
    },
    {
      "abstention_count": 4,
      "adapt_latency_seconds": 0.0,
      "adapt_selection_count": 0,
      "ambiguous_write_count": 0,
      "completion_status": "trajectory_limit",
      "execution_error_count": 4,
      "fallback_count": 0,
      "final_automated_score": -27.0,
      "final_score": 3.0,
      "mode": "baseline",
      "model_latency_seconds": 369.8995767309998,
      "normalized_reward_sum": -0.8125,
      "operational_failure": false,
      "run_id": "eval-baseline-iron-ore-throughput-20260811T041847Z-3acc6551",
      "score_auc": 7.5,
      "steps": 4,
      "success": false,
      "token_count": 12233
    },
    {
      "abstention_count": 4,
      "adapt_latency_seconds": 0.0,
      "adapt_selection_count": 0,
      "ambiguous_write_count": 0,
      "completion_status": "trajectory_limit",
      "execution_error_count": 4,
      "fallback_count": 0,
      "final_automated_score": -30.0,
      "final_score": 31.0,
      "mode": "baseline",
      "model_latency_seconds": 280.55051645599997,
      "normalized_reward_sum": 0.25,
      "operational_failure": false,
      "run_id": "eval-baseline-iron-ore-throughput-20260811T042547Z-7062a984",
      "score_auc": 77.5,
      "steps": 4,
      "success": false,
      "token_count": 11337
    },
    {
      "abstention_count": 4,
      "adapt_latency_seconds": 0.0,
      "adapt_selection_count": 0,
      "ambiguous_write_count": 0,
      "completion_status": "trajectory_limit",
      "execution_error_count": 0,
      "fallback_count": 0,
      "final_automated_score": -24.0,
      "final_score": 6.0,
      "mode": "baseline",
      "model_latency_seconds": 197.4558083309994,
      "normalized_reward_sum": 0.375,
      "operational_failure": false,
      "run_id": "eval-baseline-iron-ore-throughput-20260811T043113Z-a68f6de7",
      "score_auc": 6.0,
      "steps": 4,
      "success": false,
      "token_count": 12088
    },
    {
      "abstention_count": 4,
      "adapt_latency_seconds": 0.0,
      "adapt_selection_count": 0,
      "ambiguous_write_count": 0,
      "completion_status": "trajectory_limit",
      "execution_error_count": 2,
      "fallback_count": 0,
      "final_automated_score": -27.0,
      "final_score": 3.0,
      "mode": "baseline",
      "model_latency_seconds": 116.21141291599997,
      "normalized_reward_sum": -0.3125,
      "operational_failure": false,
      "run_id": "eval-baseline-iron-ore-throughput-20260811T043519Z-03ffbd8c",
      "score_auc": 4.5,
      "steps": 4,
      "success": false,
      "token_count": 10536
    },
    {
      "abstention_count": 0,
      "adapt_latency_seconds": 1.748698343999422,
      "adapt_selection_count": 4,
      "ambiguous_write_count": 0,
      "completion_status": "trajectory_limit",
      "execution_error_count": 2,
      "fallback_count": 0,
      "final_automated_score": -30.0,
      "final_score": 0.0,
      "mode": "frozen",
      "model_latency_seconds": 183.3295205059967,
      "normalized_reward_sum": -0.5,
      "operational_failure": false,
      "run_id": "eval-domain-only-iron-gear-wheel-throughput-20260811T055300Z-503192fe",
      "score_auc": 0.0,
      "steps": 4,
      "success": false,
      "token_count": 12762
    },
    {
      "abstention_count": 0,
      "adapt_latency_seconds": 1.725467223999658,
      "adapt_selection_count": 4,
      "ambiguous_write_count": 0,
      "completion_status": "trajectory_limit",
      "execution_error_count": 2,
      "fallback_count": 0,
      "final_automated_score": -30.0,
      "final_score": 0.0,
      "mode": "frozen",
      "model_latency_seconds": 104.19533724400026,
      "normalized_reward_sum": -0.5,
      "operational_failure": false,
      "run_id": "eval-domain-only-iron-gear-wheel-throughput-20260811T055650Z-56a74196",
      "score_auc": 0.0,
      "steps": 4,
      "success": false,
      "token_count": 12776
    },
    {
      "abstention_count": 0,
      "adapt_latency_seconds": 1.749128823999854,
      "adapt_selection_count": 4,
      "ambiguous_write_count": 0,
      "completion_status": "trajectory_limit",
      "execution_error_count": 2,
      "fallback_count": 0,
      "final_automated_score": -30.0,
      "final_score": 0.0,
      "mode": "frozen",
      "model_latency_seconds": 100.98425126600159,
      "normalized_reward_sum": -0.5,
      "operational_failure": false,
      "run_id": "eval-domain-only-iron-gear-wheel-throughput-20260811T055920Z-867a1d1a",
      "score_auc": 0.0,
      "steps": 4,
      "success": false,
      "token_count": 12776
    },
    {
      "abstention_count": 0,
      "adapt_latency_seconds": 1.8686770490003255,
      "adapt_selection_count": 4,
      "ambiguous_write_count": 0,
      "completion_status": "trajectory_limit",
      "execution_error_count": 2,
      "fallback_count": 0,
      "final_automated_score": -30.0,
      "final_score": 0.0,
      "mode": "frozen",
      "model_latency_seconds": 110.697940131,
      "normalized_reward_sum": -0.5,
      "operational_failure": false,
      "run_id": "eval-domain-only-iron-gear-wheel-throughput-20260811T060146Z-7eaebc2e",
      "score_auc": 0.0,
      "steps": 4,
      "success": false,
      "token_count": 12788
    },
    {
      "abstention_count": 0,
      "adapt_latency_seconds": 1.7458407780013658,
      "adapt_selection_count": 4,
      "ambiguous_write_count": 0,
      "completion_status": "trajectory_limit",
      "execution_error_count": 2,
      "fallback_count": 0,
      "final_automated_score": -30.0,
      "final_score": 0.0,
      "mode": "frozen",
      "model_latency_seconds": 111.96133488299893,
      "normalized_reward_sum": -0.5,
      "operational_failure": false,
      "run_id": "eval-domain-only-iron-gear-wheel-throughput-20260811T060423Z-5464cf81",
      "score_auc": 0.0,
      "steps": 4,
      "success": false,
      "token_count": 12788
    },
    {
      "abstention_count": 0,
      "adapt_latency_seconds": 1.7684618169987516,
      "adapt_selection_count": 4,
      "ambiguous_write_count": 0,
      "completion_status": "trajectory_limit",
      "execution_error_count": 0,
      "fallback_count": 0,
      "final_automated_score": -30.0,
      "final_score": 0.0,
      "mode": "frozen",
      "model_latency_seconds": 220.29862151100133,
      "normalized_reward_sum": 0.0,
      "operational_failure": false,
      "run_id": "eval-domain-only-iron-ore-throughput-20260811T052724Z-18fe1905",
      "score_auc": 0.0,
      "steps": 4,
      "success": false,
      "token_count": 13140
    },
    {
      "abstention_count": 0,
      "adapt_latency_seconds": 1.701113772996905,
      "adapt_selection_count": 4,
      "ambiguous_write_count": 0,
      "completion_status": "trajectory_limit",
      "execution_error_count": 0,
      "fallback_count": 0,
      "final_automated_score": -27.0,
      "final_score": 3.0,
      "mode": "frozen",
      "model_latency_seconds": 197.37766378699962,
      "normalized_reward_sum": 0.1875,
      "operational_failure": false,
      "run_id": "eval-domain-only-iron-ore-throughput-20260811T053143Z-82ee189d",
      "score_auc": 4.5,
      "steps": 4,
      "success": false,
      "token_count": 13063
    },
    {
      "abstention_count": 0,
      "adapt_latency_seconds": 1.8012566169982165,
      "adapt_selection_count": 4,
      "ambiguous_write_count": 0,
      "completion_status": "trajectory_limit",
      "execution_error_count": 0,
      "fallback_count": 0,
      "final_automated_score": -30.0,
      "final_score": 0.0,
      "mode": "frozen",
      "model_latency_seconds": 192.86376461700092,
      "normalized_reward_sum": 0.0,
      "operational_failure": false,
      "run_id": "eval-domain-only-iron-ore-throughput-20260811T053546Z-4f08aeaf",
      "score_auc": 0.0,
      "steps": 4,
      "success": false,
      "token_count": 13100
    },
    {
      "abstention_count": 0,
      "adapt_latency_seconds": 1.855793339000229,
      "adapt_selection_count": 4,
      "ambiguous_write_count": 0,
      "completion_status": "trajectory_limit",
      "execution_error_count": 0,
      "fallback_count": 0,
      "final_automated_score": -30.0,
      "final_score": 0.0,
      "mode": "frozen",
      "model_latency_seconds": 179.95120584100005,
      "normalized_reward_sum": 0.0,
      "operational_failure": false,
      "run_id": "eval-domain-only-iron-ore-throughput-20260811T053939Z-fc49e8bc",
      "score_auc": 0.0,
      "steps": 4,
      "success": false,
      "token_count": 12601
    },
    {
      "abstention_count": 0,
      "adapt_latency_seconds": 1.8500237160005781,
      "adapt_selection_count": 4,
      "ambiguous_write_count": 0,
      "completion_status": "trajectory_limit",
      "execution_error_count": 4,
      "fallback_count": 0,
      "final_automated_score": -24.0,
      "final_score": 6.0,
      "mode": "frozen",
      "model_latency_seconds": 524.7257148779972,
      "normalized_reward_sum": -0.625,
      "operational_failure": false,
      "run_id": "eval-domain-only-iron-ore-throughput-20260811T054318Z-4c5a8c82",
      "score_auc": 9.0,
      "steps": 4,
      "success": false,
      "token_count": 13201
    },
    {
      "abstention_count": 4,
      "adapt_latency_seconds": 0.6505647300018609,
      "adapt_selection_count": 0,
      "ambiguous_write_count": 0,
      "completion_status": "trajectory_limit",
      "execution_error_count": 3,
      "fallback_count": 0,
      "final_automated_score": -30.0,
      "final_score": 0.0,
      "mode": "frozen",
      "model_latency_seconds": 172.81314948999716,
      "normalized_reward_sum": -0.75,
      "operational_failure": false,
      "run_id": "eval-memory-only-iron-gear-wheel-throughput-20260811T063515Z-c7ad1d97",
      "score_auc": 0.0,
      "steps": 4,
      "success": false,
      "token_count": 11490
    },
    {
      "abstention_count": 4,
      "adapt_latency_seconds": 0.7088123499979702,
      "adapt_selection_count": 0,
      "ambiguous_write_count": 0,
      "completion_status": "trajectory_limit",
      "execution_error_count": 2,
      "fallback_count": 0,
      "final_automated_score": -30.0,
      "final_score": 0.0,
      "mode": "frozen",
      "model_latency_seconds": 88.88181130799967,
      "normalized_reward_sum": -0.5,
      "operational_failure": false,
      "run_id": "eval-memory-only-iron-gear-wheel-throughput-20260811T063853Z-f31f68aa",
      "score_auc": 0.0,
      "steps": 4,
      "success": false,
      "token_count": 11507
    },
    {
      "abstention_count": 4,
      "adapt_latency_seconds": 0.6020742529981362,
      "adapt_selection_count": 0,
      "ambiguous_write_count": 0,
      "completion_status": "trajectory_limit",
      "execution_error_count": 2,
      "fallback_count": 0,
      "final_automated_score": -30.0,
      "final_score": 0.0,
      "mode": "frozen",
      "model_latency_seconds": 88.1305624029992,
      "normalized_reward_sum": -0.5,
      "operational_failure": false,
      "run_id": "eval-memory-only-iron-gear-wheel-throughput-20260811T064106Z-46af8192",
      "score_auc": 0.0,
      "steps": 4,
      "success": false,
      "token_count": 11507
    },
    {
      "abstention_count": 4,
      "adapt_latency_seconds": 0.6610415020004439,
      "adapt_selection_count": 0,
      "ambiguous_write_count": 0,
      "completion_status": "trajectory_limit",
      "execution_error_count": 2,
      "fallback_count": 0,
      "final_automated_score": -30.0,
      "final_score": 0.0,
      "mode": "frozen",
      "model_latency_seconds": 89.02740298599929,
      "normalized_reward_sum": -0.5,
      "operational_failure": false,
      "run_id": "eval-memory-only-iron-gear-wheel-throughput-20260811T064319Z-b8e542b1",
      "score_auc": 0.0,
      "steps": 4,
      "success": false,
      "token_count": 11504
    },
    {
      "abstention_count": 4,
      "adapt_latency_seconds": 0.6161615189994336,
      "adapt_selection_count": 0,
      "ambiguous_write_count": 0,
      "completion_status": "trajectory_limit",
      "execution_error_count": 2,
      "fallback_count": 0,
      "final_automated_score": -30.0,
      "final_score": 0.0,
      "mode": "frozen",
      "model_latency_seconds": 84.73087063599996,
      "normalized_reward_sum": -0.5,
      "operational_failure": false,
      "run_id": "eval-memory-only-iron-gear-wheel-throughput-20260811T064532Z-495221c9",
      "score_auc": 0.0,
      "steps": 4,
      "success": false,
      "token_count": 11495
    },
    {
      "abstention_count": 4,
      "adapt_latency_seconds": 0.7226878070014209,
      "adapt_selection_count": 0,
      "ambiguous_write_count": 0,
      "completion_status": "trajectory_limit",
      "execution_error_count": 1,
      "fallback_count": 0,
      "final_automated_score": -30.0,
      "final_score": 0.0,
      "mode": "frozen",
      "model_latency_seconds": 266.41544960800275,
      "normalized_reward_sum": -0.25,
      "operational_failure": false,
      "run_id": "eval-memory-only-iron-ore-throughput-20260811T060701Z-a7e9c777",
      "score_auc": 0.0,
      "steps": 4,
      "success": false,
      "token_count": 14027
    },
    {
      "abstention_count": 4,
      "adapt_latency_seconds": 0.746186601001682,
      "adapt_selection_count": 0,
      "ambiguous_write_count": 0,
      "completion_status": "trajectory_limit",
      "execution_error_count": 3,
      "fallback_count": 0,
      "final_automated_score": -27.0,
      "final_score": 3.0,
      "mode": "frozen",
      "model_latency_seconds": 282.62715138000203,
      "normalized_reward_sum": -0.5625,
      "operational_failure": false,
      "run_id": "eval-memory-only-iron-ore-throughput-20260811T061212Z-ed24ade5",
      "score_auc": 7.5,
      "steps": 4,
      "success": false,
      "token_count": 14937
    },
    {
      "abstention_count": 4,
      "adapt_latency_seconds": 0.7364082120020612,
      "adapt_selection_count": 0,
      "ambiguous_write_count": 0,
      "completion_status": "trajectory_limit",
      "execution_error_count": 4,
      "fallback_count": 0,
      "final_automated_score": -30.0,
      "final_score": 0.0,
      "mode": "frozen",
      "model_latency_seconds": 386.803294274001,
      "normalized_reward_sum": -1.0,
      "operational_failure": false,
      "run_id": "eval-memory-only-iron-ore-throughput-20260811T061738Z-15f0740f",
      "score_auc": 0.0,
      "steps": 4,
      "success": false,
      "token_count": 15455
    },
    {
      "abstention_count": 4,
      "adapt_latency_seconds": 0.763324677998753,
      "adapt_selection_count": 0,
      "ambiguous_write_count": 0,
      "completion_status": "trajectory_limit",
      "execution_error_count": 1,
      "fallback_count": 0,
      "final_automated_score": -30.0,
      "final_score": 0.0,
      "mode": "frozen",
      "model_latency_seconds": 216.2899796439997,
      "normalized_reward_sum": -0.25,
      "operational_failure": false,
      "run_id": "eval-memory-only-iron-ore-throughput-20260811T062450Z-43fffb4f",
      "score_auc": 0.0,
      "steps": 4,
      "success": false,
      "token_count": 14009
    },
    {
      "abstention_count": 4,
      "adapt_latency_seconds": 0.7192182410017267,
      "adapt_selection_count": 0,
      "ambiguous_write_count": 0,
      "completion_status": "trajectory_limit",
      "execution_error_count": 2,
      "fallback_count": 0,
      "final_automated_score": -24.0,
      "final_score": 6.0,
      "mode": "frozen",
      "model_latency_seconds": 307.76401209899814,
      "normalized_reward_sum": -0.125,
      "operational_failure": false,
      "run_id": "eval-memory-only-iron-ore-throughput-20260811T062911Z-1796daff",
      "score_auc": 12.0,
      "steps": 4,
      "success": false,
      "token_count": 15268
    },
    {
      "abstention_count": 0,
      "adapt_latency_seconds": 2.3891100529999676,
      "adapt_selection_count": 4,
      "ambiguous_write_count": 0,
      "completion_status": "trajectory_limit",
      "execution_error_count": 2,
      "fallback_count": 0,
      "final_automated_score": -30.0,
      "final_score": 0.0,
      "mode": "frozen",
      "model_latency_seconds": 206.36011696399873,
      "normalized_reward_sum": -0.5,
      "operational_failure": false,
      "run_id": "eval-warm-frozen-iron-gear-wheel-throughput-20260811T051308Z-3f6827ca",
      "score_auc": 0.0,
      "steps": 4,
      "success": false,
      "token_count": 12745
    },
    {
      "abstention_count": 0,
      "adapt_latency_seconds": 2.2224771580004017,
      "adapt_selection_count": 4,
      "ambiguous_write_count": 0,
      "completion_status": "trajectory_limit",
      "execution_error_count": 2,
      "fallback_count": 0,
      "final_automated_score": -30.0,
      "final_score": 0.0,
      "mode": "frozen",
      "model_latency_seconds": 106.6218733909991,
      "normalized_reward_sum": -0.5,
      "operational_failure": false,
      "run_id": "eval-warm-frozen-iron-gear-wheel-throughput-20260811T051721Z-604a43f4",
      "score_auc": 0.0,
      "steps": 4,
      "success": false,
      "token_count": 12776
    },
    {
      "abstention_count": 0,
      "adapt_latency_seconds": 2.2383537629957573,
      "adapt_selection_count": 4,
      "ambiguous_write_count": 0,
      "completion_status": "trajectory_limit",
      "execution_error_count": 2,
      "fallback_count": 0,
      "final_automated_score": -30.0,
      "final_score": 0.0,
      "mode": "frozen",
      "model_latency_seconds": 99.06111720000081,
      "normalized_reward_sum": -0.5,
      "operational_failure": false,
      "run_id": "eval-warm-frozen-iron-gear-wheel-throughput-20260811T051954Z-8aa7609f",
      "score_auc": 0.0,
      "steps": 4,
      "success": false,
      "token_count": 12762
    },
    {
      "abstention_count": 0,
      "adapt_latency_seconds": 2.16969051599699,
      "adapt_selection_count": 4,
      "ambiguous_write_count": 0,
      "completion_status": "trajectory_limit",
      "execution_error_count": 2,
      "fallback_count": 0,
      "final_automated_score": -30.0,
      "final_score": 0.0,
      "mode": "frozen",
      "model_latency_seconds": 100.98124689200085,
      "normalized_reward_sum": -0.5,
      "operational_failure": false,
      "run_id": "eval-warm-frozen-iron-gear-wheel-throughput-20260811T052219Z-d99c55e8",
      "score_auc": 0.0,
      "steps": 4,
      "success": false,
      "token_count": 12762
    },
    {
      "abstention_count": 0,
      "adapt_latency_seconds": 2.2118679709983553,
      "adapt_selection_count": 4,
      "ambiguous_write_count": 0,
      "completion_status": "trajectory_limit",
      "execution_error_count": 0,
      "fallback_count": 0,
      "final_automated_score": -30.0,
      "final_score": 0.0,
      "mode": "frozen",
      "model_latency_seconds": 119.06327509500079,
      "normalized_reward_sum": 0.0,
      "operational_failure": false,
      "run_id": "eval-warm-frozen-iron-gear-wheel-throughput-20260811T052446Z-e8a54453",
      "score_auc": 0.0,
      "steps": 4,
      "success": false,
      "token_count": 11866
    },
    {
      "abstention_count": 0,
      "adapt_latency_seconds": 2.4001992119983697,
      "adapt_selection_count": 4,
      "ambiguous_write_count": 0,
      "completion_status": "trajectory_limit",
      "execution_error_count": 1,
      "fallback_count": 0,
      "final_automated_score": -30.0,
      "final_score": 0.0,
      "mode": "frozen",
      "model_latency_seconds": 265.57117886000015,
      "normalized_reward_sum": -0.25,
      "operational_failure": false,
      "run_id": "eval-warm-frozen-iron-ore-throughput-20260811T045126Z-3aa0b172",
      "score_auc": 0.0,
      "steps": 4,
      "success": false,
      "token_count": 15031
    },
    {
      "abstention_count": 0,
      "adapt_latency_seconds": 2.282597641999928,
      "adapt_selection_count": 4,
      "ambiguous_write_count": 0,
      "completion_status": "trajectory_limit",
      "execution_error_count": 1,
      "fallback_count": 0,
      "final_automated_score": -30.0,
      "final_score": 0.0,
      "mode": "frozen",
      "model_latency_seconds": 232.29290490099993,
      "normalized_reward_sum": -0.25,
      "operational_failure": false,
      "run_id": "eval-warm-frozen-iron-ore-throughput-20260811T045638Z-343e61b7",
      "score_auc": 0.0,
      "steps": 4,
      "success": false,
      "token_count": 15059
    },
    {
      "abstention_count": 0,
      "adapt_latency_seconds": 2.4824934670014045,
      "adapt_selection_count": 4,
      "ambiguous_write_count": 0,
      "completion_status": "trajectory_limit",
      "execution_error_count": 1,
      "fallback_count": 0,
      "final_automated_score": -30.0,
      "final_score": 0.0,
      "mode": "frozen",
      "model_latency_seconds": 239.77889078299995,
      "normalized_reward_sum": -0.25,
      "operational_failure": false,
      "run_id": "eval-warm-frozen-iron-ore-throughput-20260811T050117Z-08f0016c",
      "score_auc": 0.0,
      "steps": 4,
      "success": false,
      "token_count": 15115
    },
    {
      "abstention_count": 0,
      "adapt_latency_seconds": 2.2888481259997207,
      "adapt_selection_count": 4,
      "ambiguous_write_count": 0,
      "completion_status": "trajectory_limit",
      "execution_error_count": 1,
      "fallback_count": 0,
      "final_automated_score": -30.0,
      "final_score": 0.0,
      "mode": "frozen",
      "model_latency_seconds": 93.67054061199997,
      "normalized_reward_sum": -0.25,
      "operational_failure": false,
      "run_id": "eval-warm-frozen-iron-ore-throughput-20260811T050603Z-a8c129ac",
      "score_auc": 0.0,
      "steps": 4,
      "success": false,
      "token_count": 15059
    },
    {
      "abstention_count": 0,
      "adapt_latency_seconds": 2.3758166350016836,
      "adapt_selection_count": 4,
      "ambiguous_write_count": 0,
      "completion_status": "trajectory_limit",
      "execution_error_count": 1,
      "fallback_count": 0,
      "final_automated_score": -30.0,
      "final_score": 0.0,
      "mode": "frozen",
      "model_latency_seconds": 238.91187114700006,
      "normalized_reward_sum": -0.25,
      "operational_failure": false,
      "run_id": "eval-warm-frozen-iron-ore-throughput-20260811T050823Z-6345d0a9",
      "score_auc": 0.0,
      "steps": 4,
      "success": false,
      "token_count": 15087
    }
  ],
  "task_arms": {
    "iron_gear_wheel_throughput": {
      "baseline": {
        "abstention_rate": 1.0,
        "adapt_selection_rate": 0.0,
        "ambiguous_write_count": 0,
        "arm": "baseline",
        "execution_error_rate": 0.5,
        "fallback_rate": 0.0,
        "mean_automated_score": -30.0,
        "mean_final_score": 0.0,
        "mean_score_auc": 0.0,
        "mean_steps": 4.0,
        "operational_failure_count": 0,
        "operational_failure_rate": 0.0,
        "pass_rate": 0.0,
        "pass_rate_ci95_high": 0.43449149475208104,
        "pass_rate_ci95_low": 0.0,
        "sample_count": 5,
        "success_count": 0,
        "tasks": [
          "iron_gear_wheel_throughput"
        ],
        "total_adapt_latency_seconds": 0.0,
        "total_model_latency_seconds": 581.4017051030023,
        "total_tokens": 57487
      },
      "domain_only": {
        "abstention_rate": 0.0,
        "adapt_selection_rate": 1.0,
        "ambiguous_write_count": 0,
        "arm": "domain_only",
        "execution_error_rate": 0.5,
        "fallback_rate": 0.0,
        "mean_automated_score": -30.0,
        "mean_final_score": 0.0,
        "mean_score_auc": 0.0,
        "mean_steps": 4.0,
        "operational_failure_count": 0,
        "operational_failure_rate": 0.0,
        "pass_rate": 0.0,
        "pass_rate_ci95_high": 0.43449149475208104,
        "pass_rate_ci95_low": 0.0,
        "sample_count": 5,
        "success_count": 0,
        "tasks": [
          "iron_gear_wheel_throughput"
        ],
        "total_adapt_latency_seconds": 8.837812219000625,
        "total_model_latency_seconds": 611.1683840299975,
        "total_tokens": 63890
      },
      "memory_only": {
        "abstention_rate": 1.0,
        "adapt_selection_rate": 0.0,
        "ambiguous_write_count": 0,
        "arm": "memory_only",
        "execution_error_rate": 0.55,
        "fallback_rate": 0.0,
        "mean_automated_score": -30.0,
        "mean_final_score": 0.0,
        "mean_score_auc": 0.0,
        "mean_steps": 4.0,
        "operational_failure_count": 0,
        "operational_failure_rate": 0.0,
        "pass_rate": 0.0,
        "pass_rate_ci95_high": 0.43449149475208104,
        "pass_rate_ci95_low": 0.0,
        "sample_count": 5,
        "success_count": 0,
        "tasks": [
          "iron_gear_wheel_throughput"
        ],
        "total_adapt_latency_seconds": 3.2386543539978447,
        "total_model_latency_seconds": 523.5837968229953,
        "total_tokens": 57503
      },
      "warm_frozen": {
        "abstention_rate": 0.0,
        "adapt_selection_rate": 1.0,
        "ambiguous_write_count": 0,
        "arm": "warm_frozen",
        "execution_error_rate": 0.4,
        "fallback_rate": 0.0,
        "mean_automated_score": -30.0,
        "mean_final_score": 0.0,
        "mean_score_auc": 0.0,
        "mean_steps": 4.0,
        "operational_failure_count": 0,
        "operational_failure_rate": 0.0,
        "pass_rate": 0.0,
        "pass_rate_ci95_high": 0.43449149475208104,
        "pass_rate_ci95_low": 0.0,
        "sample_count": 5,
        "success_count": 0,
        "tasks": [
          "iron_gear_wheel_throughput"
        ],
        "total_adapt_latency_seconds": 11.231499460991472,
        "total_model_latency_seconds": 632.0876295420003,
        "total_tokens": 62911
      }
    },
    "iron_ore_throughput": {
      "baseline": {
        "abstention_rate": 1.0,
        "adapt_selection_rate": 0.0,
        "ambiguous_write_count": 0,
        "arm": "baseline",
        "execution_error_rate": 0.55,
        "fallback_rate": 0.0,
        "mean_automated_score": -27.6,
        "mean_final_score": 8.6,
        "mean_score_auc": 19.1,
        "mean_steps": 4.0,
        "operational_failure_count": 0,
        "operational_failure_rate": 0.0,
        "pass_rate": 0.0,
        "pass_rate_ci95_high": 0.43449149475208104,
        "pass_rate_ci95_low": 0.0,
        "sample_count": 5,
        "success_count": 0,
        "tasks": [
          "iron_ore_throughput"
        ],
        "total_adapt_latency_seconds": 0.0,
        "total_model_latency_seconds": 1176.4764439779983,
        "total_tokens": 56978
      },
      "domain_only": {
        "abstention_rate": 0.0,
        "adapt_selection_rate": 1.0,
        "ambiguous_write_count": 0,
        "arm": "domain_only",
        "execution_error_rate": 0.2,
        "fallback_rate": 0.0,
        "mean_automated_score": -28.2,
        "mean_final_score": 1.8,
        "mean_score_auc": 2.7,
        "mean_steps": 4.0,
        "operational_failure_count": 0,
        "operational_failure_rate": 0.0,
        "pass_rate": 0.0,
        "pass_rate_ci95_high": 0.43449149475208104,
        "pass_rate_ci95_low": 0.0,
        "sample_count": 5,
        "success_count": 0,
        "tasks": [
          "iron_ore_throughput"
        ],
        "total_adapt_latency_seconds": 8.97664926199468,
        "total_model_latency_seconds": 1315.2169706339992,
        "total_tokens": 65105
      },
      "memory_only": {
        "abstention_rate": 1.0,
        "adapt_selection_rate": 0.0,
        "ambiguous_write_count": 0,
        "arm": "memory_only",
        "execution_error_rate": 0.55,
        "fallback_rate": 0.0,
        "mean_automated_score": -28.2,
        "mean_final_score": 1.8,
        "mean_score_auc": 3.9,
        "mean_steps": 4.0,
        "operational_failure_count": 0,
        "operational_failure_rate": 0.0,
        "pass_rate": 0.0,
        "pass_rate_ci95_high": 0.43449149475208104,
        "pass_rate_ci95_low": 0.0,
        "sample_count": 5,
        "success_count": 0,
        "tasks": [
          "iron_ore_throughput"
        ],
        "total_adapt_latency_seconds": 3.687825539005644,
        "total_model_latency_seconds": 1459.8998870050036,
        "total_tokens": 73696
      },
      "warm_frozen": {
        "abstention_rate": 0.0,
        "adapt_selection_rate": 1.0,
        "ambiguous_write_count": 0,
        "arm": "warm_frozen",
        "execution_error_rate": 0.25,
        "fallback_rate": 0.0,
        "mean_automated_score": -30.0,
        "mean_final_score": 0.0,
        "mean_score_auc": 0.0,
        "mean_steps": 4.0,
        "operational_failure_count": 0,
        "operational_failure_rate": 0.0,
        "pass_rate": 0.0,
        "pass_rate_ci95_high": 0.43449149475208104,
        "pass_rate_ci95_low": 0.0,
        "sample_count": 5,
        "success_count": 0,
        "tasks": [
          "iron_ore_throughput"
        ],
        "total_adapt_latency_seconds": 11.829955082001106,
        "total_model_latency_seconds": 1070.225386303,
        "total_tokens": 75351
      }
    }
  },
  "total_runs": 40
}
