{
  "status": "RUN_ON_CPU",
  "seeds": {
    "7": {
      "bc_greedy": {
        "success": 0.9722222222222222,
        "cost": 5.041666666666667,
        "steps": 4.722222222222222,
        "tool_errors": 0.6388888888888888,
        "recovery_rate_given_error": 0.6111111111111112,
        "invalid_actions": 0.1388888888888889,
        "shortcut_count": 0.08333333333333333,
        "p95_steps": 6.0,
        "p99_cost": 8.299999999999997,
        "failure_taxonomy": {
          "truncated": 0,
          "premature_submit": 0,
          "permission_or_injection": 0,
          "tool_error_unrecovered": 1
        }
      },
      "terminal": {
        "success": 1.0,
        "cost": 4.930555555555555,
        "steps": 4.666666666666667,
        "tool_errors": 0.5277777777777778,
        "recovery_rate_given_error": 0.6875,
        "invalid_actions": 0.16666666666666666,
        "shortcut_count": 0.16666666666666666,
        "p95_steps": 6.0,
        "p99_cost": 8.299999999999997,
        "failure_taxonomy": {
          "truncated": 0,
          "premature_submit": 0,
          "permission_or_injection": 0,
          "tool_error_unrecovered": 0
        }
      },
      "shaped": {
        "success": 0.6666666666666666,
        "cost": 6.652777777777778,
        "steps": 6.444444444444445,
        "tool_errors": 0.4166666666666667,
        "recovery_rate_given_error": 0.21428571428571427,
        "invalid_actions": 0.1111111111111111,
        "shortcut_count": 0.9166666666666666,
        "p95_steps": 8.0,
        "p99_cost": 8.825,
        "failure_taxonomy": {
          "truncated": 6,
          "premature_submit": 0,
          "permission_or_injection": 1,
          "tool_error_unrecovered": 5
        }
      }
    },
    "17": {
      "bc_greedy": {
        "success": 0.8888888888888888,
        "cost": 5.347222222222222,
        "steps": 4.944444444444445,
        "tool_errors": 0.8055555555555556,
        "recovery_rate_given_error": 0.6086956521739131,
        "invalid_actions": 0.25,
        "shortcut_count": 0.16666666666666666,
        "p95_steps": 7.25,
        "p99_cost": 8.825,
        "failure_taxonomy": {
          "truncated": 0,
          "premature_submit": 3,
          "permission_or_injection": 0,
          "tool_error_unrecovered": 1
        }
      },
      "terminal": {
        "success": 0.9722222222222222,
        "cost": 5.319444444444445,
        "steps": 4.944444444444445,
        "tool_errors": 0.75,
        "recovery_rate_given_error": 0.6521739130434783,
        "invalid_actions": 0.16666666666666666,
        "shortcut_count": 0.1111111111111111,
        "p95_steps": 6.25,
        "p99_cost": 8.149999999999999,
        "failure_taxonomy": {
          "truncated": 0,
          "premature_submit": 0,
          "permission_or_injection": 0,
          "tool_error_unrecovered": 1
        }
      },
      "shaped": {
        "success": 0.3055555555555556,
        "cost": 7.430555555555555,
        "steps": 7.111111111111111,
        "tool_errors": 0.6388888888888888,
        "recovery_rate_given_error": 0.25,
        "invalid_actions": 0.1388888888888889,
        "shortcut_count": 1.5,
        "p95_steps": 8.0,
        "p99_cost": 9.0,
        "failure_taxonomy": {
          "truncated": 10,
          "premature_submit": 1,
          "permission_or_injection": 5,
          "tool_error_unrecovered": 9
        }
      }
    },
    "29": {
      "bc_greedy": {
        "success": 0.8611111111111112,
        "cost": 5.208333333333333,
        "steps": 4.888888888888889,
        "tool_errors": 0.6388888888888888,
        "recovery_rate_given_error": 0.47368421052631576,
        "invalid_actions": 0.2777777777777778,
        "shortcut_count": 0.1388888888888889,
        "p95_steps": 8.0,
        "p99_cost": 8.825,
        "failure_taxonomy": {
          "truncated": 0,
          "premature_submit": 2,
          "permission_or_injection": 2,
          "tool_error_unrecovered": 1
        }
      },
      "terminal": {
        "success": 0.9444444444444444,
        "cost": 5.291666666666667,
        "steps": 4.861111111111111,
        "tool_errors": 0.8611111111111112,
        "recovery_rate_given_error": 0.42857142857142855,
        "invalid_actions": 0.19444444444444445,
        "shortcut_count": 0.08333333333333333,
        "p95_steps": 7.0,
        "p99_cost": 9.149999999999999,
        "failure_taxonomy": {
          "truncated": 0,
          "premature_submit": 2,
          "permission_or_injection": 0,
          "tool_error_unrecovered": 0
        }
      },
      "shaped": {
        "success": 0.2222222222222222,
        "cost": 7.458333333333333,
        "steps": 7.388888888888889,
        "tool_errors": 0.1388888888888889,
        "recovery_rate_given_error": 0.4,
        "invalid_actions": 0.08333333333333333,
        "shortcut_count": 5.361111111111111,
        "p95_steps": 8.0,
        "p99_cost": 8.5,
        "failure_taxonomy": {
          "truncated": 26,
          "premature_submit": 0,
          "permission_or_injection": 2,
          "tool_error_unrecovered": 0
        }
      }
    }
  },
  "aggregate": {
    "bc_greedy": {
      "success_ci95": [
        0.9074074074074074,
        0.8518518518518519,
        0.9537037037037037
      ],
      "seed_means": [
        0.9722222222222222,
        0.8888888888888888,
        0.8611111111111112
      ]
    },
    "terminal": {
      "success_ci95": [
        0.9722222222222222,
        0.9351851851851852,
        1.0
      ],
      "seed_means": [
        1.0,
        0.9722222222222222,
        0.9444444444444444
      ],
      "paired_delta_vs_bc_ci95": [
        0.06481481481481481,
        0.009259259259259259,
        0.12962962962962962
      ]
    },
    "shaped": {
      "success_ci95": [
        0.39814814814814814,
        0.3055555555555556,
        0.49074074074074076
      ],
      "seed_means": [
        0.6666666666666666,
        0.3055555555555556,
        0.2222222222222222
      ],
      "paired_delta_vs_bc_ci95": [
        -0.5092592592592593,
        -0.6203703703703703,
        -0.40717592592592716
      ]
    }
  },
  "security_scope": {
    "red_team": "toy prompt injection + file permission denial",
    "production_safety_claim": false,
    "note": "No sandbox, auth service, network, or adversarial coverage."
  },
  "negative_result": "Process reward can favor repeated search/injection bait, raising return/steps without success.",
  "replay_check": "failure and recovery demo trajectories reproduced exactly"
}