                        Changes in version 0.2.0                        

  - Reads the easyQDA-CSV-Collection (contract easyqda-collection), the
    lossless relational CSV serialisation of a project beside the
    REFI-QDA .qdpx: qda_read_collection() returns a named list of data
    frames, one per thematic table, each checked against the shipped
    collection-v1.json (stamp, columns, version);
    qda_read_collection_table() reads a single table. The contract and
    reference samples are byte-identical with zotQDA.
  - The exchange contract moved to v2 under the easyQDA brand: the stamp
    column is easyqdaFormat, the tables are easyqda-*, and format ids
    are X/2.

                 Changes in version 0.1.0 (2026-09-05)                  

First release.

  - Reads all nine formats of the zotQDA exchange contract v1,
    validating the format stamp, the version and the promised columns;
    refuses an export from a newer version rather than guessing at it.
    Reference files are installed with the package, so every example,
    test and vignette chunk runs without a Zotero installation and
    without network access.
  - Projects from other QDA software can be read through the REFI-QDA
    interchange standard (qda_read_qdpx()); the subset a .qdpx file
    supports is reported on import.
  - Reproduces the plugins' graphics with ggplot2 -- code frequencies,
    the code-document matrix, saturation, timeline, MDS -- and can
    additionally re-render the original Vega-Lite chart specifications
    unchanged via the suggested vegawidget package.
  - Intercoder reliability reimplemented independently of the plugins:
    percentage agreement, Cohen's and Fleiss' kappa, Brennan and
    Prediger's coefficient, Krippendorff's alpha and Gwet's AC1, each
    with bootstrap or Wilson confidence intervals, plus level-wise
    agreement over a hierarchical code system, the confusion table and
    the annotation paradox diagnostics. Checked against frozen results
    of the plugins' JavaScript implementation.
  - The reliability of the segmentation itself, before categories are
    even compared: Krippendorff's unitizing alpha, Mathet's gamma with
    its best alignment, WindowDiff and Pk.
  - Progress and reporting: saturation measures with a stopping
    criterion, code drift across coding sessions, and generated COREQ
    and SRQR checklists.
  - Inferential statistics the plugins deliberately leave out:
    chi-squared tests of code-by-group tables with Cramér's V and an
    exact or Monte Carlo fallback, Jaccard code distances, classical
    multidimensional scaling, hierarchical clustering with the
    cophenetic correlation, and correspondence analysis with the share
    of inertia shown.
  - Licensed AGPL-3 with one additional permission on top: what you
    produce with the package -- results, figures, reports -- is yours,
    unconditionally.
