Changes in version 0.99.9                        

  - Correct development provenance to the maintainer-confirmed
    assistance scope.
  - Clarify strict material-reversal thresholds, equality boundaries and
    NULL summaries; add focused regression coverage.
  - Document table keys and distinguish audit issues from fitted event
    records.
  - Complete the source build, installation and evaluated-tutorial
    instructions.
  - Public APIs, numerical analysis, output schemas and example data are
    unchanged.

                       Changes in version 0.99.8                        

  - Document fixed backend settings, filtering defaults, design-audit
    components and the quantities shown by each plot.
  - Add a small single-cell aggregation example and clarify example-data
    metadata.
  - Centralize development provenance with a package-level source
    pointer.

                       Changes in version 0.99.7                        

  - Organize the README around installation, a minimal workflow and
    result interpretation; consolidate related-tool comparisons in the
    tutorial.
  - Place statistical and design limits alongside the results and inputs
    they explain, reducing repeated disclaimers while preserving their
    meaning.
  - Clarify package help and example-data documentation; public
    interfaces, executable analysis code and output schemas are
    unchanged.

                       Changes in version 0.99.6                        

  - Validate sparse and diagonal counts through public Matrix methods,
    retaining count error codes and sparse storage; add storage-class
    regression coverage.
  - Clarify two-condition design limits, cell aggregation and coverage
    semantics, and the distinction between current source and future
    Bioconductor installation.
  - Simplify development provenance and align source attribution
    pointers.
  - Make the output dictionary title version-independent and keep
    biocViews on one line; public APIs and output schemas are unchanged.

                       Changes in version 0.99.5                        

  - Reject logical and pattern Matrix counts with an INVALID_COUNTS
    error, before accessing numeric slots. Numeric sparse counts stay
    sparse.
  - Return the same 16 typed summary columns when no genes are retained.
  - Split design planning and fit execution into internal helpers while
    keeping the public interfaces, fixed filtering family and numerical
    backend intact.
  - Add package help, a complete output/status dictionary, synthetic
    data provenance, and a runnable mixed coverage/design-failure
    example.
  - Clarify the in-memory count requirement and preaggregated entry
    point, method references, functional positioning and AI-assisted
    development.

                       Changes in version 0.99.4                        

  - Revalidate prepared counts and gene IDs at both public model
    entries. Invalid edits fail before auditing or fitting; valid count
    edits, row subsets, and column reordering remain supported.
  - Use sample IDs for coverage plot positions, keeping donor/sample
    display labels separate so distinct samples cannot overlap due to
    label collisions.
  - Add dense and sparse prepared-object mutation regression tests.

                       Changes in version 0.99.3                        

  - Reject reserved covariate names before they can overwrite model
    columns.
  - Require syntactic covariate names and reject the special formula
    term ..
  - Use unique run IDs in effect plots, including for donors named
    baseline, and preserve positions for unavailable effects.
  - Add regression tests for both public model entries and plotting
    boundaries.

                       Changes in version 0.99.2                        

  - Added the maintainer ORCID and aligned the minimum testthat version
    with the stable mocking interface used by the tests.
  - Excluded data-raw from source builds while retaining its tracked
    files.
  - Added a clear vignette error when the Crowell19 selection rule
    yields fewer than two genes, and removed an unsupported fixed
    effect-direction claim.

                       Changes in version 0.99.1                        

  - Preserved backend warnings when the same donor refit later fails,
    and kept each deletion's original design diagnosis when the baseline
    is unavailable.
  - Added sample-aligned fit records and package/contrast provenance,
    plus consistency checks for pseudobulk objects whose columns or
    audit metadata were modified.
  - Clarified the output dictionary, related work, missing effects,
    observed ranges, and fixed-versus-re-estimated analysis choices in
    the vignette.

                       Changes in version 0.99.0                        

  - Fixed factor contrast coding so user R settings cannot reverse or
    rescale the named condition effect; added regression coverage for
    full and donor-deletion fits.
  - Expanded the Crowell19 vignette with a reproducible interpretation
    of per-donor effect changes and their limitations.
  - Added donor-aware pseudobulk preparation, design audits,
    leave-one-donor-out edgeR quasi-likelihood fits, summaries, and
    plots for two-condition studies.
  - Added explicit handling of incomplete pairs, missing cell types, low
    cell counts, unavailable fits, and genes that become all-zero after
    deletion.
  - Added simulated and Crowell19 Astrocyte pseudobulk walkthroughs,
    with runnable examples for the exported APIs.
  - Added Bioconductor installation guidance and a structured vignette.