NEWS.md
The example datasets data_benchmarkExample and data_test_confusion_matrix_scores now ship with prolfquabenchmark (they were removed from prolfqua 1.9.0); load them with prolfqua::prolfqua_data(<name>, package = "prolfquabenchmark").
SqueezeVar_comparison no longer lists lm_missing among the prolfqua facades that feed variance moderation.
The vignettes Benchmark_prolfqua, Benchmark_pipeline_demo and Benchmark_Model_IonStar_With2Factors take ContrastsMissing and ContrastsLMMissingFacade from the new package prolfquatakeoutmissing (Suggests), since prolfqua 1.9.0 no longer has them.
New vignette SqueezeVar_comparison comparing variance moderation across limma::squeezeVar(), msqrob2, and prolfqua::squeezeVarRob(). It shows msqrob2 delegates to limma, and that prolfqua::squeezeVarRob() is limma’s classic fitFDist plus a min_df filter — explaining why old MSqRob forked it (classic limma collapses to df.prior = Inf on fractional/low-df features). It documents that prolfqua reaches this regime through its rlm / rfit, lmer_nested, and imputation facades (fractional / low residual df; firth is not a consumer), and includes a benchmark (synthetic + real IonStar spike-in data) showing the former fork and modern limma::squeezeVar() give near-identical results — the evidence behind prolfqua migrating its moderation to limma::squeezeVar().
Began tracking user-visible changes in NEWS.md. For changes before this version, see the git history.