skills/apache-spark/_scores.json

{
  "version": "1.0.0",
  "skillHash": "sha256:d9d071dfd0ca3d9a72896f8a2c12bee81154f0b2eeeb1012a4f1d8a363b7af0f",
  "scoredAt": "2026-05-13T14:01:04.206Z",
  "backend": "ollama",
  "model": "gpt-oss:20b",
  "quality": {
    "score": 87,
    "dimensions": {
      "clarity": "PASS",
      "completeness": "WEAK",
      "conciseness": "PASS",
      "actionability": "PASS",
      "crossPlatform": "WEAK",
      "examples": "PASS"
    },
    "issues": [
      {
        "severity": "MEDIUM",
        "category": "completeness",
        "detail": "Instructions lack error handling and edge case coverage."
      },
      {
        "severity": "MEDIUM",
        "category": "crossPlatform",
        "detail": "Instructions only provide Python (PySpark) examples, not for Scala, Java, or R."
      }
    ]
  },
  "security": {
    "verdict": "SAFE",
    "issues": []
  },
  "impact": {
    "multiplier": 2.5,
    "baselineAvg": 40,
    "treatmentAvg": 100,
    "scenarios": [
      {
        "name": "etl-pipeline-parquet",
        "baseline": 0,
        "treatment": 100,
        "rationale": "Response A meets every rubric requirement, while Response B fails to use parquet, correct revenue calculation, and full aggregation."
      },
      {
        "name": "streaming-kafka-window-count",
        "baseline": 80,
        "treatment": 100,
        "rationale": "Response A satisfies every rubric requirement, while Response B uses 'complete' mode instead of the required 'update' mode."
      }
    ]
  }
}