{
  "version": 1,
  "kind": "live_model",
  "status": "completed",
  "capsule": {
    "title": "Job Radar — frozen input-isolated trial",
    "checks": [
      {
        "id": "check-1",
        "name": "Check 1",
        "category": "intent",
        "basis": "Acceptance definition excluded from this export",
        "held_out": true
      },
      {
        "id": "check-2",
        "name": "Check 2",
        "category": "intent",
        "basis": "Acceptance definition excluded from this export",
        "held_out": true
      },
      {
        "id": "check-3",
        "name": "Check 3",
        "category": "intent",
        "basis": "Acceptance definition excluded from this export",
        "held_out": true
      },
      {
        "id": "check-4",
        "name": "Check 4",
        "category": "preservation",
        "basis": "Acceptance definition excluded from this export",
        "held_out": true
      },
      {
        "id": "check-5",
        "name": "Check 5",
        "category": "preservation",
        "basis": "Acceptance definition excluded from this export",
        "held_out": true
      },
      {
        "id": "check-6",
        "name": "Check 6",
        "category": "preservation",
        "basis": "Acceptance definition excluded from this export",
        "held_out": true
      },
      {
        "id": "check-7",
        "name": "Check 7",
        "category": "preservation",
        "basis": "Acceptance definition excluded from this export",
        "held_out": true
      },
      {
        "id": "check-8",
        "name": "Check 8",
        "category": "preservation",
        "basis": "Acceptance definition excluded from this export"
      },
      {
        "id": "check-9",
        "name": "Check 9",
        "category": "preservation",
        "basis": "Acceptance definition excluded from this export"
      }
    ],
    "intent": {
      "text": "Intent text was not included in this export."
    },
    "baseline": {
      "provenance": "Historical model authorship is unverified; source provenance excluded from this summary."
    }
  },
  "arms": {
    "A": {
      "label": "Saved artifact",
      "status": "completed",
      "models": [
        "unknown"
      ],
      "cost_usd": 0,
      "tokens": {
        "total": 0
      },
      "evaluation": {
        "status": "completed",
        "passed": 6,
        "total": 9,
        "checks": [
          {
            "id": "check-1",
            "passed": false,
            "error_kind": "behavior"
          },
          {
            "id": "check-2",
            "passed": false,
            "error_kind": "behavior"
          },
          {
            "id": "check-3",
            "passed": false,
            "error_kind": "behavior"
          },
          {
            "id": "check-4",
            "passed": true,
            "error_kind": "behavior"
          },
          {
            "id": "check-5",
            "passed": true,
            "error_kind": "behavior"
          },
          {
            "id": "check-6",
            "passed": true,
            "error_kind": "behavior"
          },
          {
            "id": "check-7",
            "passed": true,
            "error_kind": "behavior"
          },
          {
            "id": "check-8",
            "passed": true,
            "error_kind": "behavior"
          },
          {
            "id": "check-9",
            "passed": true,
            "error_kind": "behavior"
          }
        ],
        "screenshots": {}
      }
    },
    "B": {
      "label": "Ordinary improvement",
      "status": "completed",
      "models": [
        "claude-sonnet-5-5"
      ],
      "cost_usd": 0.0414374,
      "tokens": {
        "total": 8642
      },
      "budget_usd": 0.3333333333333333,
      "evaluation": {
        "status": "completed",
        "passed": 9,
        "total": 9,
        "checks": [
          {
            "id": "check-1",
            "passed": true,
            "error_kind": "behavior"
          },
          {
            "id": "check-2",
            "passed": true,
            "error_kind": "behavior"
          },
          {
            "id": "check-3",
            "passed": true,
            "error_kind": "behavior"
          },
          {
            "id": "check-4",
            "passed": true,
            "error_kind": "behavior"
          },
          {
            "id": "check-5",
            "passed": true,
            "error_kind": "behavior"
          },
          {
            "id": "check-6",
            "passed": true,
            "error_kind": "behavior"
          },
          {
            "id": "check-7",
            "passed": true,
            "error_kind": "behavior"
          },
          {
            "id": "check-8",
            "passed": true,
            "error_kind": "behavior"
          },
          {
            "id": "check-9",
            "passed": true,
            "error_kind": "behavior"
          }
        ],
        "screenshots": {}
      }
    },
    "C": {
      "label": "Intent-first reframe",
      "status": "completed",
      "models": [
        "claude-sonnet-5-5"
      ],
      "cost_usd": 0.07513720000000002,
      "tokens": {
        "total": 14298
      },
      "budget_usd": 0.3333333333333333,
      "evaluation": {
        "status": "completed",
        "passed": 9,
        "total": 9,
        "checks": [
          {
            "id": "check-1",
            "passed": true,
            "error_kind": "behavior"
          },
          {
            "id": "check-2",
            "passed": true,
            "error_kind": "behavior"
          },
          {
            "id": "check-3",
            "passed": true,
            "error_kind": "behavior"
          },
          {
            "id": "check-4",
            "passed": true,
            "error_kind": "behavior"
          },
          {
            "id": "check-5",
            "passed": true,
            "error_kind": "behavior"
          },
          {
            "id": "check-6",
            "passed": true,
            "error_kind": "behavior"
          },
          {
            "id": "check-7",
            "passed": true,
            "error_kind": "behavior"
          },
          {
            "id": "check-8",
            "passed": true,
            "error_kind": "behavior"
          },
          {
            "id": "check-9",
            "passed": true,
            "error_kind": "behavior"
          }
        ],
        "screenshots": {}
      }
    },
    "D": {
      "label": "Clean-slate rebuild",
      "status": "completed",
      "models": [
        "claude-sonnet-5-5"
      ],
      "cost_usd": 0.034864,
      "tokens": {
        "total": 5274
      },
      "budget_usd": 0.3333333333333333,
      "evaluation": {
        "status": "completed",
        "passed": 9,
        "total": 9,
        "checks": [
          {
            "id": "check-1",
            "passed": true,
            "error_kind": "behavior"
          },
          {
            "id": "check-2",
            "passed": true,
            "error_kind": "behavior"
          },
          {
            "id": "check-3",
            "passed": true,
            "error_kind": "behavior"
          },
          {
            "id": "check-4",
            "passed": true,
            "error_kind": "behavior"
          },
          {
            "id": "check-5",
            "passed": true,
            "error_kind": "behavior"
          },
          {
            "id": "check-6",
            "passed": true,
            "error_kind": "behavior"
          },
          {
            "id": "check-7",
            "passed": true,
            "error_kind": "behavior"
          },
          {
            "id": "check-8",
            "passed": true,
            "error_kind": "behavior"
          },
          {
            "id": "check-9",
            "passed": true,
            "error_kind": "behavior"
          }
        ],
        "screenshots": {}
      }
    }
  },
  "comparisons": {
    "A_B": {
      "verdict": "improved",
      "improvements": [
        "check-1",
        "check-2",
        "check-3"
      ],
      "regressions": []
    },
    "A_C": {
      "verdict": "improved",
      "improvements": [
        "check-1",
        "check-2",
        "check-3"
      ],
      "regressions": []
    },
    "A_D": {
      "verdict": "improved",
      "improvements": [
        "check-1",
        "check-2",
        "check-3"
      ],
      "regressions": []
    },
    "B_C": {
      "verdict": "no_measured_gain",
      "improvements": [],
      "regressions": []
    },
    "B_D": {
      "verdict": "no_measured_gain",
      "improvements": [],
      "regressions": []
    },
    "C_D": {
      "verdict": "no_measured_gain",
      "improvements": [],
      "regressions": []
    }
  },
  "manifest": {
    "input_result_sha256": "f582d48796b2f91a4a8ee74144b7f51d89df3c2eae7d85467c5d6d0e0db30680",
    "source": "1f634cfabb3add1def3795369f041c622c091974e9cd5c0eb7c53526ce4995b1",
    "checks": "673403f661e5caba6125eb17029e6561ad86df1a623ce34e235514d334c0e23f"
  },
  "model_calls": 4,
  "sharing": {
    "metrics_only": true,
    "intent_included": false,
    "screenshots_included": false,
    "note": "Source, raw prompts/receipts, check definitions, private labels and paths are excluded. Metrics are reported from the input result; this export is not independent verification."
  },
  "source_unchanged": true,
  "cost": {
    "total_usd": 0.1514386,
    "limit_usd": 1.0,
    "known_spend_usd": 0.1514386,
    "basis": "Claude CLI reported API-equivalent USD; not a subscription invoice"
  }
}