Changes in version 0.2.2 Bug fixes - syrona_connect() now defaults to read_only = FALSE. The previous default of TRUE always failed, because CDMConnector validates a CDM by writing a probe table. This also fixes extract_all() when db is a DuckDB file path. Documentation - README and vignettes install from CRAN; the GitHub install is listed as the development version. - README demo instructions use the bundled demo data instead of a repository clone. - Removed the read_only = TRUE examples and the claim that connections are read-only by default. Changes in version 0.2.1 (2026-09-04) CRAN review fixes - Removed single quotes around the OMOP CDM acronym in Title and Description. - Documented the return value of load_syrona_theme(), run_app() and syrona_disconnect(). - Replaced the \dontrun{} example on run_app() with an if (interactive()) block. - run_app() now restores the user's syrona.data_dir option on exit, and the examples reset any option they set. Changes in version 0.2.0 New features - Added a bundled demo dataset (inst/extdata/demo): a curated 32-concept subset (conditions, procedures, drugs) that runs the full comparison workflow with no database. See RUN_DEMO.html or ?compare_all for a runnable example. - compare_all() now ships a runnable \donttest example driven by the demo data. Improvements - Shipped standard-vocabulary lookups in inst/extdata/ so a fresh install has a working conditions chapter map and procedure device/site/morphology enrichment (icd10_lookup.csv and vocabulary/*.csv, resolved via syrona_extdata()). - Dashboard "About" panel now documents all four procedure attribute filters (site, method, device, morphology) and corrects the Sex/Age filter behavior (display-only; the meta-analysis is precomputed). - Population pyramids degrade gracefully with an explanatory message when a dataset has no demographics table (e.g. the demo). - DESCRIPTION metadata polished for CRAN (Paule-Mandel reference DOI, Depends: R (>= 4.1.0)). Bug fixes - Corrected the fold-threshold glossary wording (similarity boundary, not a clinical-relevance boundary).