Changes in version 0.1.1                        

  - Correct per-child AIC improvement filtering and cache repeated model
    fits.
  - Validate inputs and exclude invalid fits; preserve safe predictor
    identities.
  - Correct bootstrap success accounting and report failed draws.
  - Handle null and empty model sets and undefined diagnostic metrics.
  - Repair dashboard data/split invalidation, confusion-matrix
    placement, ROC integration, test R², reproducibility, reports, and
    narrow-window layouts.
  - Remove test-data leakage from the diabetes example and align
    methodology documentation.
  - Add regression tests and automated R package checks.

Results may change because of the corrected search and bootstrap
accounting. stability()$B now reports successful draws; B_requested
records attempted draws. The core API requires finite numeric
predictors. Intercept-only models are exempt from variable-stability
filtering and have NA average stability.