Make student simulation content-aware and hybrid

[Copilot]

The simulator’s statistical model did not reliably incorporate the current workshop pages. This change combines Monte Carlo modelling with evidence-backed agentic assessment of page state.

Changes

• Content-aware ranking

  • Rank high-risk steps using dropout rates, curriculum metrics, and semantic page review.
  • Score state readiness, path clarity, and recovery support across all mapped pages.

• Hybrid probability model

  • Convert weighted semantic scores into bounded probability adjustments.
  • Apply those adjustments to the existing learner and environment model.

• Input integrity

  • Bind insights to page content hashes and reject stale assessments.
  • Reject incomplete scores, unknown keys, and out-of-range adjustments.

• Regression coverage

  • Cover semantic-score effects, score normalization, stale insights, and adjustment bounds.