washr

• update_metadata() is rewritten as the one FAIR step (lifecycle: experimental). It derives a schema.org Dataset description from DESCRIPTION, data-raw/dictionary.csv, CITATION.cff and the files in inst/extdata, writes it as JSON-LD into pkgdown/templates/in-header.html so every page of the site carries it, and ends by listing the fields it could not fill. Spatial and temporal coverage and keywords live in DESCRIPTION as X-schema.org-spatialCoverage, X-schema.org-temporalCoverage and X-schema.org-keywords. It no longer calls the dataspice helpers, creates no data/metadata folder, and writes no inst/extdata/metadata.json; existing copies of both are ignored. generate_jsonld() is no longer exported (it is the internal builder), and lubridate leaves Imports (#68, #70, #67).
• New use_brand() installs the openwashdata brand (_brand.yml and the logo files it references) from the central openwashdata/brand repository into the active package, refreshes an existing copy idempotently, and wires an existing _pkgdown.yml to the brand through bslib so the package site renders with the brand fonts and colors (#109).
• setup_readme() gains has_example, the argument the guide documents. With has_example = TRUE the README carries an Example section with a commented ggplot2 scaffold for a first plot, pairing with the argument of the same name on setup_website(). The function now stops with a clear message when data/ holds no data object instead of writing a README with NA in it, and says which data object the template documents when there are several (#74, supersedes #24).
• update_citation() gains build. With build = FALSE it regenerates the citation files and adds the badge without rebuilding README.md and the site, for scripts and tests (#75).

Changed

• Idempotency and ergonomics sweep of the core (#73). Every function that rewrites a file follows read-merge-write and is safe to re-run:
• setup_website() keeps an existing _pkgdown.yml as it is and only rebuilds the site, so the guide's "answer No when prompted" step goes away; it no longer crashes without a .gitignore; and it leaves docs ignored when a pkgdown workflow deploys the site, or when the new track_docs = FALSE argument says so (#104). The example article is created once.
• The _pkgdown.yml template carries the Pages URL as the site URL, the explanatory comments, and a reference index with one entry per data object, matching the openwashdata review standard's template.
• update_citation() re-run without a doi keeps the DOI already on file instead of dropping it, moves keywords typed into CITATION.cff by hand to X-schema.org-keywords in DESCRIPTION (their canonical home, from where cffr carries them forward), and rebuilds the README only when the badge changes.
• setup_roxygen() re-run regenerates only the @format block and keeps everything below it (@source, @examples, and other text), writes nothing when the result is unchanged, and errors clearly when a file has no @format line instead of crashing.
• setup_dictionary() records the first class of multi-class columns (POSIXct, ordered, Date) instead of a deparsed vector.
• Loading a .rda file with several objects, or a file that is not an .rda, now errors with the file name and the reason instead of documenting the first object or crashing.

Fixed

• setup_readme() no longer writes a dead license link. The README template carried the package name placeholder in URL encoded form, so whisker never substituted it and every generated README linked to .../%7B%7B%7Bpackagename%7D%7D%7D/blob/main/LICENSE.md (#101).
• update_citation() adds CITATION.cff to .Rbuildignore, so R CMD check no longer reports a non-standard file at the top level of the data package. cffr only adds the entry itself when handed a file path, and washr hands it a cff object (#102).

Removed

• update_gsheet_metadata() is removed. It appended a row to a private openwashdata Google Sheet, needed interactive Google authentication, and never shipped on CRAN. googlesheets4 leaves Imports with it. The catalogue update becomes org-internal tooling outside the package (#69).
• The dataspice helpers are removed: add_metadata(), add_creator(), update_access(), update_attributes() and update_biblio(). None of them shipped on CRAN. update_metadata() replaced their output, and a package that still carries data/metadata/ keeps it; washr ignores the folder. fill_dictionary() and generate_roxygen_docs() are no longer exported; setup_dictionary() and setup_roxygen() call them. The export surface is now nine functions. dataspice, dplyr, readr, stringr and tibble leave Imports with the removed code (#71, #100, #72).

Dependencies

• devtools moves from Imports to Suggests. update_citation() still rebuilds README.md through devtools::build_readme(), because the README loads the data package, and asks to install devtools when it is missing; the site rebuild calls pkgdown::build_site() directly. setup_website() no longer renders the example article on its own, since the site build renders it. The version constraint on utils is dropped; utils is a base package and the constraint silently required R 4.3.3. With the seven packages removed by the FAIR layer work, Imports go from 16 to 10 (#72).

Documentation and tests

• Test bar (#75). Every export has a behavioral test that asserts on file content or output; every expect_error() names its message; the setup_rawdata() tests check the rendered template. A test coverage workflow measures coverage on every push and pull request and writes it to the job summary.
• Documentation (#76). The Get started vignette walks through the workflow in the order of the publishing guide, one function per step; the README gives the toolkit overview with the stable core and the experimental FAIR layer, and carries the CRAN status badge; every export names its neighbours in the workflow under See also; the reference index is grouped by workflow stage; and the pkgdown site URL points at the openwashdata organisation (it pointed at the dev fork, which broke canonical links).