experiment.json

{
  "prompt": "Explain how an API should handle authentication errors.",
  "baseline": {
    "model": "Model A (example)",
    "reportedModel": null,
    "contextWindowTokens": null,
    "contextWindowSource": "unverified",
    "inputTokens": null,
    "outputTokens": null,
    "finishReason": null,
    "response": "When authentication fails, return an error message explaining what went wrong."
  },
  "candidate": {
    "model": "Model B (example)",
    "reportedModel": null,
    "contextWindowTokens": null,
    "contextWindowSource": "unverified",
    "inputTokens": null,
    "outputTokens": null,
    "finishReason": null,
    "response": "When authentication fails, return HTTP 401 for invalid credentials or HTTP 403 for insufficient permissions. Never expose tokens or internal implementation details."
  },
  "checks": {
    "requiredPhrases": ["401", "403"]
  }
}