skills/ai-scientist/_scores.json

{
  "version": "1.0.0",
  "skillHash": "sha256:81bf01434f3b39504170aa0892b7f4731de910c0f5f03eb7e3fb5df420afca45",
  "scoredAt": "2026-05-13T13:44:41.991Z",
  "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": "The instructions lack coverage of error handling and edge cases, focusing only on happy paths."
      },
      {
        "severity": "MEDIUM",
        "category": "crossPlatform",
        "detail": "The examples are tailored to the Claude model, limiting use with other AI agents."
      }
    ]
  },
  "security": {
    "verdict": "SAFE",
    "issues": []
  },
  "impact": {
    "multiplier": 9,
    "baselineAvg": 0,
    "treatmentAvg": 90,
    "scenarios": [
      {
        "name": "full-research-pipeline",
        "baseline": 0,
        "treatment": 80,
        "rationale": "Response A covers most rubric items but omits explicitly stating the best hypothesis, while Response B provides no content."
      },
      {
        "name": "experiment-design",
        "baseline": 0,
        "treatment": 100,
        "rationale": "Response B satisfies all rubric items, including the required mention of ExperimentDesigner.design, while Response A does not mention ExperimentDesigner.design and thus fails that key requirement."
      }
    ]
  }
}