figure_spec.json

{
  "question": "How often do observed learner states persist or change?",
  "data": "A supplied sequence of state labels, not inferred psychological states.",
  "method": "Count adjacent transitions; normalize outgoing counts; calculate entropy and self-transition frequency.",
  "formula": "P(j|i) = count(i→j)/sum_j count(i→j); persistence = self-transitions/(n-1).",
  "interpretation": "A terminal state with no outgoing observation has no estimated transition row. Fewer than two observations make persistence undefined. These are descriptive first-order frequencies, not a trained sequence predictor.",
  "status": "SYNTHETIC / RULE-BASED PROTOTYPE | no educational validity claim",
  "source": "results/review_examples.json",
  "rows": [
    {
      "label": "self-transition proportion",
      "path": [
        "outputs",
        "self-transition proportion"
      ],
      "unit": "unitless"
    },
    {
      "label": "work transition entropy (bits)",
      "path": [
        "outputs",
        "work transition entropy (bits)"
      ],
      "unit": "unitless"
    }
  ],
  "table_title": "Worked example — illustrative, not a measured research result",
  "evidence_note": "Illustrative inputs and calculations are in scripts/review_examples.py.",
  "finding": "This repository summarizes supplied learner-state sequences through transition probabilities, outgoing-state entropy, and persistence. Counts are normalized only where transitions are observed, and insufficient sequences return an undefined persistence value rather than a fabricated estimate. It is a descriptive baseline for sequence analysis; the labels must be justified separately and should not be treated as inferred mental states."
}