Scores can differ from rfair 0.1.0: several fixes below change what counts as
evidence, and new metadata sources find evidence that was missed before. Rerun
assessments before comparing them with 0.1.0 results. Agreement with the
reference F-UJI 4.0.0 service (metrics v0.8, five fixture DOIs, 85 metric
comparisons) rose from 91.8% to 97.6%; see tests/conformance/README.md.
id_parse() no longer treats a plain URL with a numeric path segment (for
example https://figshare.com/articles/dataset/foo/12345/1) as a Handle. Such
URLs were marked persistent and resolved through hdl.handle.net instead of
their own page. The Handle pattern is now anchored at the start, as in
F-UJI's verify_handle().resolved_url is NA, the new resolution element records the
attempt (URL, HTTP status, error), and print() shows it as unresolved.
FsF-A1-02MD-1 (metadata retrievable) now requires a harvested metadata
record, as F-UJI's testMetadataRetrievable does; a page that resolves but
offers no extractable metadata no longer earns it. A nonexistent DOI dropped
from 17.3% to 13.5%; the remaining points (identifier scheme, HTTP protocol)
are the ones F-UJI also awards.isAccessibleForFree is read as an access statement, as F-UJI
does (FsF-A1-01M, classify_access()).harvest_errors element.
API rate limits (GitHub, GitLab, Codeberg, repository APIs) are also raised
as an rfair_rate_limit warning and stop further calls to that API for the
assessment, instead of silently lowering software scores.license_reuse() recognizes the OGL, Etalab, CDLA, DL-DE, and NLOD open data
licenses, and classes MPL as copyleft.reference_data.https://doi.org/ra/), and the DataCite requests are skipped for DOIs
registered elsewhere. CSL metadata does not satisfy FsF-F4-01M-2, which is
specific to DataCite.GITHUB_PAT (then GITHUB_TOKEN), GITLAB_PAT, and
CODEBERG_TOKEN, and sent as a Bearer Authorization header, which libcurl
does not forward when a redirect leaves the host.IsSupplementTo link to GitHub) is bridged to it: the
repository supplies the software signals and the DOI counts as the registry
DOI. rfair's own Zenodo concept DOI rose from 2.2% to 100%.DESCRIPTION, PyPI from pyproject.toml or
setup.cfg), not only against text mentions.a$software.evidence_type = "heuristic", and print() says
so: the scores come from repository signals that have not been validated
against expert judgement. frsm_agreement() compares them with expert
ratings (percent agreement and Cohen's kappa per test, plus agreement
between raters), and inst/extdata/frsm_validation_template.csv is a
rating sheet for all 45 FRSM tests. The validation study itself is still to
be done.fair_recommendations() lists every failed test with one concrete action,
largest score gain first (the gain allows for each metric's cap). The Shiny
app shows it in a "How to improve" tab.fair_compare() reports per-metric or per-test changes between two
assessments.as_rdf() adds one DQV quality measurement per metric and one FAIR Test
Result per metric test in the OSTrails FAIR Testing Resource vocabulary
(https://w3id.org/ftr/, version 1.3.0): pass or fail, completion, the
evidence as a log, and for failed tests the fair_recommendations() action
as a suggestion. Turtle output now checks for the jsonld package it needs.assess_fair() gains max_time, a time budget for the whole assessment.assess_fair_batch() and assess_data_code() gain workers (parallel
assessment by forking; serial on Windows), keep (the full assessments in
the "assessments" attribute), and previous (resume an interrupted run),
plus resolved and http_status columns.options(rfair.rate_per_host)), and an optional HTTP
cache (options(rfair.cache_dir)). Bodies are decoded from their declared
charset to UTF-8.options(rfair.block_private_hosts = TRUE) refuses non-http(s) URLs and
hosts that resolve to loopback, private, link-local, or cloud metadata
addresses. The request connects to the address that was checked (libcurl's
CURLOPT_RESOLVE), so DNS rebinding cannot swap it, and every redirect hop is
checked, with credentials dropped when a redirect changes scheme, host, or
port. The bundled Plumber API and Shiny app turn it on, since both fetch
visitor-supplied URLs. Headless rendering (use_headless) runs in a browser
outside this guard.RFAIR_API_ALLOW_HEADLESS=true.CITATION.cff, codemeta.json, .zenodo.json, and
ro-crate-metadata.json are generated from DESCRIPTION by
data-raw/07-build-metadata.R, so title, authors, contributors, and
dependencies agree. citation("rfair") reads the title and version from the
package metadata. The CRAN DOI is recorded.httptest2, jqr, wand, and
covr. The documentation no longer mentions libmagic file sniffing, which
was never implemented.httr2::local_mocked_responses().First release. rfair is a native R implementation of the F-UJI / FAIRsFAIR
research data object assessment metrics and the FRSM (FAIR for Research Software)
metrics. It performs the entire assessment in R, with no external server.
assess_fair() resolves a DOI, persistent identifier, URL, or code
repository, harvests its metadata, and scores it against the FAIR metrics,
returning a fair_assessment object.rfair_metric_versions(): the current F-UJI
data metrics (v0.8) by default, several legacy and domain-specific versions
(0.2-0.8, plus social-science and environmental variants), and the FRSM
research-software metrics (0.7).metric_version = "0.7_software"
to score it against the FRSM metrics from its repository signals (license,
README, citation/codemeta, tests, CI, dependencies, coverage, releases,
contributors). The FRSM metrics operationalize the FAIR Principles for
Research Software (FAIR4RS; Chue Hong et al. 2022, doi:10.15497/RDA00068).id_parse() recognizes DOI, Handle, ARK, URN, UUID, identifiers.org / w3id,
and compact prefix:accession identifiers.fair_assessment object has print(), summary(), as.data.frame(),
and plot() methods. plot() draws a category scorecard, a per-metric
breakdown, or a concentric FAIR "sunburst".as_fuji_json() exports the assessment in the F-UJI FAIRResults JSON schema;
as_rdf() exports W3C DQV quality measurements plus a schema.org Rating
(JSON-LD or, with rdflib, Turtle).fair_example, is provided for offline use.license_reuse() judges whether a license actually permits reuse, using the
(Re)usable Data Project taxonomy; reusabledata_rating() looks up curated
repository ratings.classify_access() flags controlled-access and sensitive data (which are not
FAIR failures).identifier_hygiene() checks identifiers for layered or non-persistent forms.fair_tlc() reports the FAIR-TLC (Traceable, Licensed, Connected) indicators.fair_principles() and principle_definition() provide the canonical FAIR
principle definitions; fair4rs_principles() provides the FAIR4RS principles
for research software, and principle_definition() resolves FRSM software
metrics to their FAIR4RS statement.assess_fair_batch() scores a vector of identifiers into one tidy row each.assess_data_code() ingests the data and code identifiers that the
rtransparent package extracts from articles (its open_data_links and
open_code_links) and scores each (FsF for data, FRSM for code).split_identifiers() parses the " ; "-joined identifier strings.launch_rfair() opens a bslib Shiny app for interactive assessment.system.file("plumber", package = "rfair") and
system.file("openapi", package = "rfair").