Debug-Type-Coercion-V2

The model receives the prompt (and optional system message). The run uses scorer rubric_json_metrics with the JSON configuration below. Pass/fail and partial credit are determined entirely by that scorer against the model output; no human grading.

Return JSON only with keys diagnosis, fix, tests. A safeNumber parser accepts empty strings and whitespace as 0 because Number(value) coerces them. Diagnose and fix the coercion bug.

{
  "metrics": {
    "repro": {
      "checks": [
        {
          "contains": [
            "empty string"
          ]
        },
        {
          "contains": [
            "whitespace"
          ]
        },
        {
          "contains": [
            "0"
          ]
        },
        {
          "contains": [
            "coerce"
          ]
        }
      ]
    },
    "hidden": {
      "checks": [
        {
          "contains": [
            "trim"
          ]
        },
        {
          "contains": [
            "reject"
          ]
        },
        {
          "contains": [
            "Number.isFinite"
          ]
        }
      ]
    },
    "diagnose": {
      "checks": [
        {
          "contains": [
            "Number"
          ]
        },
        {
          "contains": [
            "type coercion"
          ]
        },
        {
          "contains": [
            "safeNumber"
          ]
        }
      ]
    }
  }
}