NEWS
trialdiff 0.2.0 (2026-09-27)
- Added an analysis/output registry:
td_output() and output_registry()
declare analyses and outputs and their inputs, and add_outputs(),
add_edges() and remove_edges() graft, extend or override a lineage graph.
- Added
lineage_from_metadata(), which derives a lineage graph from a
\pkg{metacore} object (or equivalent metadata tables), with provenance and a
review list for ambiguous or unparsed references. aliases resolve tokens
that would otherwise be ambiguous or unknown, and overrides merge an
analysis/output registry into the generated graph.
- Added a
"diffdf" comparison backend to compare_cut(), which delegates
value comparison to the \pkg{diffdf} package and translates its output into a
tdiff object (schema comparison remains built in).
- Added a case-study vignette using public
pharmaverseadam data.
as_register() now coerces key columns to character, fixing a failure when
key columns were not character.
- New records in visit-based datasets are classified as
new_visit rather than
new_assessment.
- Added edge-case tests; package coverage now exceeds 90%.
- Bumped GitHub Actions and added Codecov configuration.
trialdiff 0.1.0
- First tagged release. Five composable layers: compare, classify, trace,
assess and report.
compare_cut() compares two data cuts of a clinical dataset and reports
added, removed and modified observations plus schema changes.
classify_changes() applies transparent, prioritised rules to produce a
clinical change taxonomy.
define_lineage() and trace_dependencies() build and query a lineage graph.
assess_impact() grades downstream impact as definitely/potentially/unlikely.
report_diff() produces HTML, Quarto, JSON and list reports.
- Synthetic
ADSL/ADLB data cuts and an example lineage are included.
trialdiff 0.0.0.9000
- Initial development version.