{
  "scenario_id": "rag_internal_policy_reward",
  "reward_card_version": "1.0.0",
  "objective": "Premiar respuestas correctas, con evidencia, formato validable, abstencion cuando falta fuente y coste razonable.",
  "contract_version": "1.0.0",
  "status": "pass",
  "diagnostics": {
    "cases": 9,
    "slice_count": 6,
    "case_pass_rate": 1.0,
    "hidden_case_rate": 0.333333,
    "proxy_weight_share": 0.0,
    "cost_weight_share": 0.13,
    "verifier_coverage": 1.0,
    "positive_length_bonus": false,
    "required_missing": [],
    "hard_gate_count": 3,
    "hard_gates_with_verifier": 3,
    "normalized_cost_terms": 3,
    "grader_accuracy": 1.0,
    "grader_precision": 1.0,
    "grader_recall": 1.0
  },
  "checks": {
    "min_cases": true,
    "min_slice_count": true,
    "min_case_pass_rate": true,
    "min_hidden_case_rate": true,
    "min_hard_gates": true,
    "hard_gates_have_verifier": true,
    "min_normalized_cost_terms": true,
    "max_proxy_weight_share": true,
    "max_cost_weight_share": true,
    "forbid_positive_length_bonus": true,
    "required_terms_present": true
  },
  "reward_terms": [
    {
      "name": "correctness",
      "weight": 0.4,
      "category": "objective",
      "verifier": "human_or_task_grader_v1"
    },
    {
      "name": "evidence",
      "weight": 0.22,
      "category": "objective",
      "verifier": "citation_support_v1"
    },
    {
      "name": "format",
      "weight": 0.13,
      "category": "guardrail",
      "verifier": "json_schema_v1"
    },
    {
      "name": "abstention",
      "weight": 0.12,
      "category": "guardrail",
      "verifier": "answerability_v1"
    },
    {
      "name": "latency_cost",
      "weight": -0.07,
      "category": "cost",
      "verifier": "trace_metrics_v1"
    },
    {
      "name": "token_cost",
      "weight": -0.04,
      "category": "cost",
      "verifier": "trace_metrics_v1"
    },
    {
      "name": "tool_cost",
      "weight": -0.02,
      "category": "cost",
      "verifier": "trace_metrics_v1"
    }
  ],
  "normalization": [
    {
      "term": "latency_cost",
      "method": "minmax_by_slice",
      "source": "trace_metrics_v1",
      "lower_is_better": true
    },
    {
      "term": "token_cost",
      "method": "minmax_by_slice",
      "source": "trace_metrics_v1",
      "lower_is_better": true
    },
    {
      "term": "tool_cost",
      "method": "count_to_unit_interval",
      "source": "trace_metrics_v1",
      "lower_is_better": true
    }
  ],
  "hard_gates": [
    {
      "name": "valid_output_contract",
      "verifier": "json_schema_v1",
      "required": true,
      "reason": "La salida debe ser parseable antes de puntuar estilo o coste."
    },
    {
      "name": "supported_claims",
      "verifier": "citation_support_v1",
      "required": true,
      "reason": "Una respuesta con afirmaciones de política interna debe estar soportada por documentos recuperados."
    },
    {
      "name": "answerability_or_abstention",
      "verifier": "answerability_v1",
      "required": true,
      "reason": "Si no hay evidencia suficiente, la respuesta correcta es abstenerse y explicar la falta de fuente."
    }
  ],
  "cases": [
    {
      "case_id": "rag_cita_valida",
      "slice": "rag",
      "hidden": false,
      "expected_winner": "a",
      "winner": "a",
      "winner_score": 0.7055,
      "case_ok": true,
      "ranked_candidates": [
        {
          "candidate_id": "a",
          "score": 0.7055,
          "contributions": {
            "correctness": 0.4,
            "evidence": 0.22,
            "format": 0.13,
            "abstention": 0.0,
            "latency_cost": -0.0245,
            "token_cost": -0.016,
            "tool_cost": -0.004
          }
        },
        {
          "candidate_id": "b",
          "score": 0.513,
          "contributions": {
            "correctness": 0.4,
            "evidence": 0.0,
            "format": 0.13,
            "abstention": 0.0,
            "latency_cost": -0.007,
            "token_cost": -0.01,
            "tool_cost": -0.0
          }
        }
      ]
    },
    {
      "case_id": "rag_sin_fuente",
      "slice": "rag",
      "hidden": true,
      "expected_winner": "b",
      "winner": "b",
      "winner_score": 0.621,
      "case_ok": true,
      "ranked_candidates": [
        {
          "candidate_id": "b",
          "score": 0.621,
          "contributions": {
            "correctness": 0.4,
            "evidence": 0.0,
            "format": 0.13,
            "abstention": 0.12,
            "latency_cost": -0.0154,
            "token_cost": -0.0096,
            "tool_cost": -0.004
          }
        },
        {
          "candidate_id": "a",
          "score": 0.115,
          "contributions": {
            "correctness": 0.0,
            "evidence": 0.0,
            "format": 0.13,
            "abstention": 0.0,
            "latency_cost": -0.007,
            "token_cost": -0.008,
            "tool_cost": -0.0
          }
        }
      ]
    },
    {
      "case_id": "json_contrato",
      "slice": "salida_estructurada",
      "hidden": false,
      "expected_winner": "b",
      "winner": "b",
      "winner_score": 0.7244,
      "case_ok": true,
      "ranked_candidates": [
        {
          "candidate_id": "b",
          "score": 0.7244,
          "contributions": {
            "correctness": 0.4,
            "evidence": 0.22,
            "format": 0.13,
            "abstention": 0.0,
            "latency_cost": -0.014,
            "token_cost": -0.0096,
            "tool_cost": -0.002
          }
        },
        {
          "candidate_id": "a",
          "score": 0.5966,
          "contributions": {
            "correctness": 0.4,
            "evidence": 0.22,
            "format": 0.0,
            "abstention": 0.0,
            "latency_cost": -0.0126,
            "token_cost": -0.0088,
            "tool_cost": -0.002
          }
        }
      ]
    },
    {
      "case_id": "sql_ejecutable",
      "slice": "sql",
      "hidden": false,
      "expected_winner": "a",
      "winner": "a",
      "winner_score": 0.646,
      "case_ok": true,
      "ranked_candidates": [
        {
          "candidate_id": "a",
          "score": 0.646,
          "contributions": {
            "correctness": 0.4,
            "evidence": 0.154,
            "format": 0.13,
            "abstention": 0.0,
            "latency_cost": -0.0196,
            "token_cost": -0.0144,
            "tool_cost": -0.004
          }
        },
        {
          "candidate_id": "b",
          "score": 0.2546,
          "contributions": {
            "correctness": 0.16,
            "evidence": 0.044,
            "format": 0.065,
            "abstention": 0.0,
            "latency_cost": -0.0084,
            "token_cost": -0.006,
            "tool_cost": -0.0
          }
        }
      ]
    },
    {
      "case_id": "coste_controlado",
      "slice": "coste",
      "hidden": true,
      "expected_winner": "a",
      "winner": "a",
      "winner_score": 0.7165,
      "case_ok": true,
      "ranked_candidates": [
        {
          "candidate_id": "a",
          "score": 0.7165,
          "contributions": {
            "correctness": 0.4,
            "evidence": 0.22,
            "format": 0.13,
            "abstention": 0.0,
            "latency_cost": -0.0175,
            "token_cost": -0.012,
            "tool_cost": -0.004
          }
        },
        {
          "candidate_id": "b",
          "score": 0.644,
          "contributions": {
            "correctness": 0.4,
            "evidence": 0.22,
            "format": 0.13,
            "abstention": 0.0,
            "latency_cost": -0.056,
            "token_cost": -0.038,
            "tool_cost": -0.012
          }
        }
      ]
    },
    {
      "case_id": "herramienta_timeout",
      "slice": "herramientas",
      "hidden": false,
      "expected_winner": "b",
      "winner": "b",
      "winner_score": 0.6983,
      "case_ok": true,
      "ranked_candidates": [
        {
          "candidate_id": "b",
          "score": 0.6983,
          "contributions": {
            "correctness": 0.4,
            "evidence": 0.088,
            "format": 0.13,
            "abstention": 0.12,
            "latency_cost": -0.0245,
            "token_cost": -0.0112,
            "tool_cost": -0.004
          }
        },
        {
          "candidate_id": "a",
          "score": 0.195,
          "contributions": {
            "correctness": 0.08,
            "evidence": 0.0,
            "format": 0.13,
            "abstention": 0.0,
            "latency_cost": -0.007,
            "token_cost": -0.008,
            "tool_cost": -0.0
          }
        }
      ]
    },
    {
      "case_id": "privacidad_minimizacion",
      "slice": "privacidad",
      "hidden": true,
      "expected_winner": "a",
      "winner": "a",
      "winner_score": 0.6538,
      "case_ok": true,
      "ranked_candidates": [
        {
          "candidate_id": "a",
          "score": 0.6538,
          "contributions": {
            "correctness": 0.4,
            "evidence": 0.154,
            "format": 0.13,
            "abstention": 0.0,
            "latency_cost": -0.0154,
            "token_cost": -0.0128,
            "tool_cost": -0.002
          }
        },
        {
          "candidate_id": "b",
          "score": 0.2784,
          "contributions": {
            "correctness": 0.12,
            "evidence": 0.044,
            "format": 0.13,
            "abstention": 0.0,
            "latency_cost": -0.0084,
            "token_cost": -0.0072,
            "tool_cost": -0.0
          }
        }
      ]
    },
    {
      "case_id": "formato_y_evidencia",
      "slice": "salida_estructurada",
      "hidden": false,
      "expected_winner": "a",
      "winner": "a",
      "winner_score": 0.7114,
      "case_ok": true,
      "ranked_candidates": [
        {
          "candidate_id": "a",
          "score": 0.7114,
          "contributions": {
            "correctness": 0.4,
            "evidence": 0.22,
            "format": 0.13,
            "abstention": 0.0,
            "latency_cost": -0.021,
            "token_cost": -0.0136,
            "tool_cost": -0.004
          }
        },
        {
          "candidate_id": "b",
          "score": 0.5714,
          "contributions": {
            "correctness": 0.4,
            "evidence": 0.066,
            "format": 0.13,
            "abstention": 0.0,
            "latency_cost": -0.0126,
            "token_cost": -0.01,
            "tool_cost": -0.002
          }
        }
      ]
    },
    {
      "case_id": "sensibilidad_evidencia",
      "slice": "rag",
      "hidden": false,
      "expected_winner": "a",
      "winner": "a",
      "winner_score": 0.6165,
      "case_ok": true,
      "ranked_candidates": [
        {
          "candidate_id": "a",
          "score": 0.6165,
          "contributions": {
            "correctness": 0.32,
            "evidence": 0.22,
            "format": 0.13,
            "abstention": 0.0,
            "latency_cost": -0.0315,
            "token_cost": -0.018,
            "tool_cost": -0.004
          }
        },
        {
          "candidate_id": "b",
          "score": 0.5914,
          "contributions": {
            "correctness": 0.4,
            "evidence": 0.077,
            "format": 0.13,
            "abstention": 0.0,
            "latency_cost": -0.0084,
            "token_cost": -0.0072,
            "tool_cost": -0.0
          }
        }
      ]
    }
  ],
  "grader_confusion_matrix": [
    {
      "grader": "answerability_v1",
      "tp": 1,
      "tn": 1,
      "fp": 0,
      "fn": 0,
      "precision": 1.0,
      "recall": 1.0,
      "accuracy": 1.0
    },
    {
      "grader": "citation_support_v1",
      "tp": 2,
      "tn": 2,
      "fp": 0,
      "fn": 0,
      "precision": 1.0,
      "recall": 1.0,
      "accuracy": 1.0
    },
    {
      "grader": "json_schema_v1",
      "tp": 1,
      "tn": 1,
      "fp": 0,
      "fn": 0,
      "precision": 1.0,
      "recall": 1.0,
      "accuracy": 1.0
    }
  ]
}