{
  "epochs": [
    {
      "epoch": 1,
      "eval_metrics": {
        "avg_binary_accuracy": 0.795,
        "confirmation_required_accuracy": 0.8,
        "count": 25,
        "eval_batches": 25,
        "eval_loss": 7.704372234344483,
        "mode_accuracy": 0.44,
        "permission_bit_accuracy": 0.850909090909091,
        "precondition_ok_accuracy": 0.76,
        "quality_score": 0.21877532557574186,
        "rollback_supported_accuracy": 0.72,
        "schema_ok_accuracy": 0.76,
        "session_required_accuracy": 1.0,
        "tool_accuracy": 0.24,
        "trigger_accuracy": 0.76,
        "validator_required_accuracy": 0.8,
        "worktree_required_accuracy": 0.76
      },
      "improved": true,
      "quality_score": 0.21877532557574186,
      "train_metrics": {
        "loss": 0.5388983534915107,
        "micro_batches": 70,
        "optimizer_steps": 5
      }
    },
    {
      "epoch": 2,
      "eval_metrics": {
        "avg_binary_accuracy": 0.81,
        "confirmation_required_accuracy": 0.8,
        "count": 25,
        "eval_batches": 25,
        "eval_loss": 7.129253711700439,
        "mode_accuracy": 0.48,
        "permission_bit_accuracy": 0.8545454545454545,
        "precondition_ok_accuracy": 0.76,
        "quality_score": 0.4400157397790389,
        "rollback_supported_accuracy": 0.8,
        "schema_ok_accuracy": 0.76,
        "session_required_accuracy": 1.0,
        "tool_accuracy": 0.44,
        "trigger_accuracy": 0.76,
        "validator_required_accuracy": 0.8,
        "worktree_required_accuracy": 0.8
      },
      "improved": true,
      "quality_score": 0.4400157397790389,
      "train_metrics": {
        "loss": 0.4655745736190251,
        "micro_batches": 70,
        "optimizer_steps": 5
      }
    },
    {
      "epoch": 3,
      "eval_metrics": {
        "avg_binary_accuracy": 0.81,
        "confirmation_required_accuracy": 0.8,
        "count": 25,
        "eval_batches": 25,
        "eval_loss": 6.935494079589843,
        "mode_accuracy": 0.48,
        "permission_bit_accuracy": 0.8545454545454545,
        "precondition_ok_accuracy": 0.76,
        "quality_score": 0.4229221342606978,
        "rollback_supported_accuracy": 0.8,
        "schema_ok_accuracy": 0.76,
        "session_required_accuracy": 1.0,
        "tool_accuracy": 0.4,
        "trigger_accuracy": 0.76,
        "validator_required_accuracy": 0.8,
        "worktree_required_accuracy": 0.8
      },
      "improved": false,
      "quality_score": 0.4229221342606978,
      "train_metrics": {
        "loss": 0.43205853487764084,
        "micro_batches": 70,
        "optimizer_steps": 5
      }
    },
    {
      "epoch": 4,
      "eval_metrics": {
        "avg_binary_accuracy": 0.81,
        "confirmation_required_accuracy": 0.8,
        "count": 25,
        "eval_batches": 25,
        "eval_loss": 6.897225952148437,
        "mode_accuracy": 0.56,
        "permission_bit_accuracy": 0.8545454545454545,
        "precondition_ok_accuracy": 0.76,
        "quality_score": 0.46349615617231893,
        "rollback_supported_accuracy": 0.8,
        "schema_ok_accuracy": 0.76,
        "session_required_accuracy": 1.0,
        "tool_accuracy": 0.44,
        "trigger_accuracy": 0.76,
        "validator_required_accuracy": 0.8,
        "worktree_required_accuracy": 0.8
      },
      "improved": true,
      "quality_score": 0.46349615617231893,
      "train_metrics": {
        "loss": 0.42175399448190415,
        "micro_batches": 70,
        "optimizer_steps": 5
      }
    },
    {
      "epoch": 5,
      "eval_metrics": {
        "avg_binary_accuracy": 0.81,
        "confirmation_required_accuracy": 0.8,
        "count": 25,
        "eval_batches": 25,
        "eval_loss": 6.89590030670166,
        "mode_accuracy": 0.56,
        "permission_bit_accuracy": 0.8545454545454545,
        "precondition_ok_accuracy": 0.76,
        "quality_score": 0.4635160408540206,
        "rollback_supported_accuracy": 0.8,
        "schema_ok_accuracy": 0.76,
        "session_required_accuracy": 1.0,
        "tool_accuracy": 0.44,
        "trigger_accuracy": 0.76,
        "validator_required_accuracy": 0.8,
        "worktree_required_accuracy": 0.8
      },
      "improved": false,
      "quality_score": 0.4635160408540206,
      "train_metrics": {
        "loss": 0.4198248211826597,
        "micro_batches": 70,
        "optimizer_steps": 5
      }
    }
  ]
}