Changes in version 0.99.20 - The core carries its own trigamma(), so the solver no longer depends on R supplying one. Agrees with R's to 1.8e-15 across the range it is used over. Changes in version 0.99.19 - Declares its dependency on S4Vectors, which stashPACE() uses. Changes in version 0.99.18 - Condition signals build their messages without paste(), and the ridge check handles empty input directly rather than suppressing the warning it would raise. Changes in version 0.99.17 - New compactFit() drops the per-cell matrices before saving a fit, which the vignette previously did by assigning into a slot. - The vignettes read fits through accessors rather than @, which clears BiocCheck's note about slot access in examples and vignettes. Changes in version 0.99.16 - New stashPACE() copies a fit's readouts into metadata(spe), so results can travel with the data they came from. The solver state is not copied. Changes in version 0.99.15 - Records the Australian Government Research Training Program as funder in Authors@R. Changes in version 0.99.14 - PACEFit gains a validity method. Replacing a slot wholesale used to pass silently and break every readout downstream; fit@fit <- list() is now refused. - New accessors cellTypes(), paceParams() and fitState(), and validated replacement methods for neighbourSlopes(), varianceDecomposition() and topDrivers(). Changes in version 0.99.13 - verbose = FALSE is now silent. The shrinkage step reported its progress regardless, because it never received the flag, and mashr printed its own on top of that. - Installation instructions use BiocManager::install(). - The vignette evaluates every chunk. - Minimum R version is 4.6.0. Changes in version 0.99.12 - plotDecomposition() draws the zoom panel's percentage labels upright, so that neighbouring bars of similar height no longer overlap each other's text. Changes in version 0.99.11 - Added a canonical multivariate adaptive shrinkage implementation to the compiled core, alongside the existing mashr path. It reproduces mashr's calls exactly on both reference cohorts. Changes in version 0.99.10 - The driver score columns are named for what the method calls them: driver_score and driver_score_spec4, previously MCSD and MCSD4. plotDrivers() takes panels = "score" in place of "mcsd". Changes in version 0.99.9 - topDrivers() documents which model term each column holds. On a fit with a condition the table carries two different slopes, and they are not a raw and shrunken pair of the same quantity. - The condition vignette shows the responder decomposition and driver figures, and prints the driver table. Changes in version 0.99.6 - The per-pair gene score is called the driver score throughout the documentation and plot labels, matching the method description. Changes in version 0.99.4 - Fixed portability defects found by the Bioconductor build matrix: bit identity on arm64, a crash on Windows from static thread_local destructors, an unsupported system2(env=) on Windows, and missing -pthread on compile flags for Wasm. Changes in version 0.99.0 - Initial Bioconductor submission.