Check every dataset's citation, license and DOI, structurally and against its authority

One row per (dataset, finding); a clean dataset has a single ok row. The structural half always runs: citation_main must be non-empty (missing_citation), carry a 4-digit year (no_year) and a locator — a DOI, a URL in the string, or a link_data_source (no_locator); license must be an active id in metadata/license.csv (missing_license when empty or unknown, license_unregistered otherwise), and custom needs a license_url (license_custom_no_url); doi must be bare (10.…/…).

Usage

check_dataset_citation(
  ingest_yaml,
  network = TRUE,
  cache_dir = NULL,
  license_csv = NULL,
  refresh = FALSE,
  fetch = NULL,
  timeout = 30
)

Arguments

ingest_yaml

named list from read_ingest_yaml()

network

fetch from the authorities (default TRUE); FALSE runs the structural half and compares against whatever is already cached

cache_dir

the metadata/ root: holds license.csv, each dataset's questions.csv and its citation_authority.json cache. Defaults to metadata/ beside the first ingest's .qmd.

license_csv

path to the license registry (default {cache_dir}/license.csv)

refresh

ignore the cache and fetch again (default FALSE)

fetch

the HTTP function, function(url, accept = NULL, method = "GET", …) returning list(status, content, url); the default uses curl. The tests inject one that serves saved responses.

timeout

seconds per request

Value

A tibble: dataset_key, finding, detail, authority, authority_citation, checked, plus level (ok/error/warn, see citation_findings()), exempt and question.

Details

The network half (network = TRUE, i.e. not CALCOFI_SKIP_LINK_CHECK) asks the source's own authority, chosen by link_data_source: EDI's cite service (packageid=<scope>.<id>.<rev>, or the newest revision when only scope/identifier are given), an NCEI landing page ("Cite as"), an ERDDAP .das (globals), and DataCite for any DOI (its rightsList SPDX id, and doi.org content negotiation for the formatted citation). A declared doi must answer 200/30x at doi.org (doi_unresolved). Every fetch is cached in {cache_dir}/{provider}/{dataset}/citation_authority.json (authority, url, citation, license, creator, title, checked, …) so a re-run costs nothing; pass refresh = TRUE to fetch again. A cached authority is compared even when network = FALSE. A fetched citation (or SPDX license) that differs from the declared one after normalize_citation() is authority_drift, with both strings in detail; a resolver that cannot be reached is authority_unavailable. Nothing is ever written into a notebook's YAML — the author's string is the record, the authority is a proposal.

A finding is exempt when the dataset's questions.csv ({cache_dir}/{provider}/{dataset}/questions.csv) holds an open or proposed row with related_table = dataset whose related_field is empty or names the field the finding is about (citation_main, license, doi); question carries the row's label. assert_dataset_citation() stops on any non-exempt error-level row.

See also

read_license_registry(), assert_dataset_citation(), release_citation()