Conversation
The dataset shipped with data_repo_url and data_repo NA for all 8 papers: the Crossref relation metadata in the raw harvest was empty and the fallback planned for issue #27's download step never ran, so the platform mention scan (#26) had nothing to search. Resolve every paper against Crossref relations, DataCite resource types, and the articles' own availability sections (MDPI pages via the Wayback Machine). Record the links in data-raw/datapapers_repo_fixes.csv, one row per DOI with its source and check date, applied in 03_datapapers_process.R through the existing coalesce helper so harvest metadata still wins if it ever appears. parse_repo_name() learns GBIF and IEEE DataPort. Seven papers deposit in general repositories (GBIF twice, IEEE DataPort, Figshare, Dryad, NCBI BioProject, Zenodo); one shares its data only in the article tables; none uses a WASH sector platform. The rebuilt dataset, refreshed mention scan, vignette sentence, and re-knitted README carry that finding. Refs #26, #27, #28 Assisted-by: Claude claude-fable-5
Personal lesson notes kept in the repo root, not part of the package. Assisted-by: Claude claude-fable-5
The stock claim "all relevant data are included in the paper or its supplementary information" is the modal DAS across the corpus and nearly uninformative on its own. Tier 1 of #47: score the claim jointly with the supplement fields already recorded instead of taking it at face value. New exported function das_in_paper_support() detects the claim in both forms (normalized "in paper" in washdev, the raw stock sentence in the IWA snapshots) and classifies each claim by the strongest attachment that could carry the data: no supplement, unstructured (pdf/images), structured (docx/xlsx), or open format (csv, json). data-raw/das_in_paper_support.R reproduces the summary in das-in-paper-support-summary.csv. Across 2,599 claims in washdev, jwh, aqua, and ws: 71.5% have no supplement at all, 26.5% a structured one, 2.1% unstructured, none an open format. Spread per journal: Water Supply 80.7% bare claims, AQUA 77.6%, washdev 58.4%, JWH 55.4%. The IWA snapshots serialize an empty supp_url as "[]", which defeats plain non-emptiness checks; the function handles it. Normalize this to NA when the snapshots become package datasets, or score_fair()'s Accessible dimension will overscore them. NEWS.md also catches up the entry for the datapapers repository-link backfill committed in 33f1224. Refs #47 Assisted-by: Claude claude-fable-5
Tier 2 first heuristic: filter the no-supplement in-paper claims by article type, since editorials, errata, and reviews make the claim trivially true. Types come from OpenAlex (batch-fetched, DOI-to-type lookup committed so reruns cost no requests) with title patterns as a front-matter fallback. Negative result: of 1,857 bare claims, 1,832 are typed article, 9 review, 16 unmatched; only 10 claims (0.5%) are trivially true. The filter does not shrink the population, most likely because front matter in these journals never carries a DAS and so never enters the claim set. The 71.5% no-supplement share needs no discounting; 1,847 substantive bare claims remain for the table-shape and supplement audit tiers. Refs #47 Assisted-by: Claude claude-fable-5
Tier 3: download the IWA supplementary files and classify what each one actually shares, instead of trusting the file extension. The pre-signed CDN links from the July 2026 scrape expire between 2026-08-18 and 2026-08-30, so the download ran just in time: 741 of the 745 still-valid files were fetched (661 MB); 572 links were already dead. The files themselves are a gitignored cache; the committed record is data-raw/suppfiles/manifest.csv with signature-stripped URLs (the signed queries trip secret scanners, see #10), HTTP status, size, and sha256 per file. suppfiles_parse.R classifies each file with transparent structural heuristics: docx via pandoc-to-markdown pipe-table metrics (rows, numeric share, summary-statistic terms in headers), xlsx/xls via readxl sheet metrics plus openxlsx merge counts. Classes: observation tables, summary tables, figures, prose only. Per-table metrics are kept so thresholds can be audited. Results: docx supplements are 57% prose only, 25% summary tables, 14% observation tables; xlsx nearly always holds data (25 of 29). Joined to the in-paper claims (das_in_paper_suppfile_audit.R, 290 structured-supplement claims covered, 42% of that population): 54% prose only, 21% summary tables, 20% observation tables. Roughly 5% of all in-paper claims are visibly backed by data-shaped files. The heuristics await validation: validation_sample.csv holds a seeded, stratified 30-file sample for human rating before these numbers are used in the paper. Refs #47 Assisted-by: Claude claude-fable-5
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
datapapersdataset via a scripted fix table (Evaluate how and what data is accessible through WASH data platforms #26, Categorise shared data and identify PII and other sensitive content #27), regenerating the exported .rda, .csv, and .xlsx files.Test plan
Rscript -e 'devtools::load_all()'succeedsdata-raw/suppfiles/validation_sample.csvcontains 30 sample rows for manual ratinginst/extdata/datapapers.csvhave non-empty repository links matchingdata-raw/datapapers_repo_fixes.csvdata-raw/das-in-paper-support-summary.csvlooks plausible