{
  "version": "0.1",
  "date": "2026-10-04",
  "trainingExamples": 48,
  "provenance": "Synthetic author-created held-out diagnostic set; not independently validated",
  "trainingSha256": "7d077677f0dd9a3aa3d118ccddbf3343da50523a7685db085fdc8135a25b494d",
  "testSha256": "1dc214974effa91e1dac8e4fbd0e4e72e2004235ced68e44fac25892610da5b2",
  "total": 24,
  "correct": 20,
  "accuracy": 0.8333333333333334,
  "abstentions": 6,
  "routed": 18,
  "routedCorrect": 16,
  "reviewExpected": 6,
  "reviewCorrect": 4,
  "perLabel": [
    {
      "label": "Decision",
      "total": 6,
      "correct": 6
    },
    {
      "label": "Evaluation",
      "total": 6,
      "correct": 5
    },
    {
      "label": "Research",
      "total": 6,
      "correct": 5
    },
    {
      "label": "Needs review",
      "total": 6,
      "correct": 4
    }
  ],
  "results": [
    {
      "text": "Find quotes about budget in the supplied report",
      "label": "Research",
      "predicted": "Research"
    },
    {
      "text": "Which documents contain evidence about implementation risks",
      "label": "Research",
      "predicted": "Research"
    },
    {
      "text": "Extract relevant source passages for this question",
      "label": "Research",
      "predicted": "Research"
    },
    {
      "text": "Prepare research findings with source references",
      "label": "Research",
      "predicted": "Research"
    },
    {
      "text": "Gather supporting citations from the project notes",
      "label": "Research",
      "predicted": "Research"
    },
    {
      "text": "Rank the alternatives by weighted quality and cost",
      "label": "Decision",
      "predicted": "Decision"
    },
    {
      "text": "Compare our options with a decision matrix",
      "label": "Decision",
      "predicted": "Decision"
    },
    {
      "text": "Test how changing weights affects the highest score",
      "label": "Decision",
      "predicted": "Decision"
    },
    {
      "text": "Score these alternatives using four criteria",
      "label": "Decision",
      "predicted": "Decision"
    },
    {
      "text": "Evaluate the tradeoffs between pilot options",
      "label": "Decision",
      "predicted": "Decision"
    },
    {
      "text": "Measure classification errors on labeled test examples",
      "label": "Evaluation",
      "predicted": "Evaluation"
    },
    {
      "text": "Retrain this routing classifier with my dataset",
      "label": "Evaluation",
      "predicted": "Evaluation"
    },
    {
      "text": "Inspect predicted labels and accuracy",
      "label": "Evaluation",
      "predicted": "Evaluation"
    },
    {
      "text": "Run the model evaluation on unseen cases",
      "label": "Evaluation",
      "predicted": "Evaluation"
    },
    {
      "text": "Show classifier performance and abstentions",
      "label": "Evaluation",
      "predicted": "Evaluation"
    },
    {
      "text": "Write a birthday poem for my friend",
      "label": "Needs review",
      "predicted": "Needs review"
    },
    {
      "text": "Translate this recipe into Japanese",
      "label": "Needs review",
      "predicted": "Needs review"
    },
    {
      "text": "",
      "label": "Needs review",
      "predicted": "Needs review"
    },
    {
      "text": "Look through the notes for unanswered questions",
      "label": "Research",
      "predicted": "Needs review"
    },
    {
      "text": "Which path offers the best balance of speed and expense",
      "label": "Decision",
      "predicted": "Decision"
    },
    {
      "text": "How often does the routing system get things wrong",
      "label": "Evaluation",
      "predicted": "Needs review"
    },
    {
      "text": "Write an advertisement about our research model",
      "label": "Needs review",
      "predicted": "Research"
    },
    {
      "text": "Explain why my car engine makes a clicking noise",
      "label": "Needs review",
      "predicted": "Needs review"
    },
    {
      "text": "Translate this research brief into Spanish",
      "label": "Needs review",
      "predicted": "Research"
    }
  ]
}