NEWS

CircularRegression 0.5.0

Add circular_regression() as the main fixed-effect modeling interface.
Add angular_two_step() as an explicit consensus-then-homogeneous workflow.
Add S3 methods for coefficients, fitted values, residuals, predictions, summaries, plots, log-likelihoods and information criteria.
Improve consensus numerical stability for large Bessel-function arguments.
Add validation for finite angles, non-negative modifiers, weights, controls and initial values.
Align logLik.consensus(), AIC.consensus() and BIC.consensus() with the full von Mises log-likelihood by including the normalizing constant. This changes absolute consensus likelihood and information-criterion values but does not change the fitted estimates.
Add summary.angular_re() and print.summary.angular_re().
Preserve model-frame na.action information in angular() objects.
Replace the draft overview vignette with two reproducible HTML vignettes.
Add pkgdown configuration and a workflow diagnostics article.
Add a package-data workflow vignette for R Journal reviewer support.
Add a minimal GitHub Actions R CMD check workflow.
Clarify random-effects and special-wrapper documentation.
Clarify documented provenance for noshiro and the remaining provenance limitations for multiplebison.
Add a related scientific reference for the multiplebison study context.
Expand tests for simulation recovery, predictions, NA handling, weights, modulo invariance and small-sample fits.

See CHANGELOG_0.4.0.md in the development repository for development notes. That file is not included in the CRAN build.

Improve stability of angular and consensus estimators with QR-based updates and better handling of reference-only models.
Implement observation weights in consensus() and remove the deprecated model argument across the API.
Make angular_re() usable without workarounds, add Hessian conditioning checks, and provide more robust predictions.
Refresh documentation and vignette examples, and add an automated test suite covering key model features.