Changes in version 2.5.0 (2026-05-25)                  

Added

  - Prediction for all interaction methods: predict() and predict_pls()
    now support product_indicator and orthogonal interaction models, in
    addition to two_stage. Previously, only two_stage interactions could
    generate out-of-sample predictions; the other methods threw an
    error. All three methods now fully support single predictions
    (predict()), k-fold cross-validation, and LOOCV via predict_pls().
  - Quadratic term prediction: quadratic_term() models (using any
    interaction method) can now generate predictions.
  - Parallel k-fold cross-validation: predict_pls() now supports
    parallel execution for k-fold CV when cores is specified (e.g.,
    predict_pls(model, noFolds = 50, cores = 4)). Previously,
    parallelization was only available for LOOCV.
  - Interaction method detection: New internal
    detect_interaction_method() function provides clean dispatch based
    on interaction class attributes.
  - Custom confidence levels in plots: plot() accepts a user-specified
    confidence level for bootstrapped models, allowing displays at any
    alpha (e.g., 90%, 99%) instead of the fixed 95% default (#407).
  - Public accessor API for constructs and measurement-model elements: A
    set of helper functions is now exported and documented for use by
    downstream packages and user scripts. Container-first argument order
    (model or measurement-model first): construct_items(x,
    construct_name) (S3 generic), construct_names(x) (S3 generic),
    construct_name(construct), construct_mode(mmMatrix, construct),
    construct_type(model, construct), all_factors(model),
    all_composites(model), all_non_interactions(measurement_model).
    These replace and consolidate a set of non-exported internal
    helpers; downstream code should migrate off seminr::: triple-colon
    access and use these exported functions instead.

Changed

  - predict.seminr_model() dispatch refactored: uses switch() on
    detected interaction method instead of pattern-matching on
    measurement model names.
  - Interaction estimation now stores prediction-relevant parameters on
    the model object (model$interaction_params), including
    orthogonalization regression coefficients needed for out-of-sample
    prediction of orthogonal models.
  - Mixed interaction methods (e.g., one two_stage and one
    product_indicator in the same model) produce an informative error at
    prediction time.

Fixed

  - Plot significance stars now use bootstrap p-values for consistency
    with reported significance (#412).
  - construct_items() and all_LOC_items() return a character vector
    instead of a single-column matrix, restoring expected downstream
    behavior (#364).
  - Construct/item name collision check now correctly detects name
    conflicts that were previously missed (#402).
  - CBSEM summary path significance now displays in the conventional IV
    → DV direction (#404).
  - Quadratic interaction terms with a single indicator no longer fail
    due to matrix-to-vector coercion (#327).

                 Changes in version 2.4.2 (2026-02-18)                  

Fixed

  - PLSpredict now works correctly with non-standard (character)
    rownames (#390)
  - Plot symbols use BMP-compatible Greek letters for cross-platform
    rendering (#226)
  - Plot displays capital R² for coefficient of determination (#389)
  - Summary reports now work correctly for PLS-SEM models with
    higher-order constructs (#369)
  - vif_items() always returns a named list structure (#377)

Changed

  - Modernized GitHub Actions CI workflow for Ubuntu 24.04

                 Changes in version 2.4.0 (2026-01-27)                  

  - (See previous CRAN release notes)