Changes in version 0.1.1 (2026-10-07)                  

  - Hardened grid search against failed/nonfinite candidates and
    restricted EBIC recomputation to criterion = "ebic".
  - Fixed LOO handling of all-failed folds, required converged fold
    refits, and recomputed Adaptive Lasso pilots/weights within each
    training fold.
  - Preserved missing_x behavior in bootstrap refits and repaired
    complete-case group remapping and regression-imputation intercept
    handling.
  - Stored the full fixed-effect covariance matrix for valid
    Hessian-based fits and used (x^T V_beta x) for prediction standard
    errors.
  - Information criteria now use the number of informative response rows
    when completely missing responses are present.
  - Added explicit handling/documentation for transformed responses
    under censoring and for model-matrix missing-X imputation.
  - pecme_mi_combine(use_refit = TRUE) now requires converged refits
    with valid finite standard errors.