Define and ship the catalog data model

[larnsce]

Goal

Three tibbles, shipped as the internal object owd_catalog in R/sysdata.rda and exported as CSV for the website.

Rewritten 2026-09-04 per dev/decisions-2026-09.md.

owd_packages (all detected packages)

column	type	provenance
pkg_name	chr	DESCRIPTION Package
title	chr	DESCRIPTION Title
description	chr	DESCRIPTION Description, whitespace-collapsed
version	chr	DESCRIPTION Version, leading v stripped
published	lgl	Zenodo DOI (10.5281/zenodo.*) present in CITATION.cff
reviewed	lgl	NA until the pkgreview check is wired; then TRUE when the review issues are closed
maintainer	chr	Authors@R role cre, "Given Family"
maintainer_orcid	chr	Authors@R comment ORCID
authors	chr	all aut/cre collapsed with "; "
license	chr	DESCRIPTION License
date	Date	DESCRIPTION Date, NA-tolerant
keywords	chr	X-schema.org-keywords, GitHub topics as fallback
spatial_coverage	chr	X-schema.org-spatialCoverage, NA if absent
temporal_coverage	chr	X-schema.org-temporalCoverage, NA if absent
doi	chr	CITATION.cff doi, NA if absent
cff_version	chr	CITATION.cff version, drift diagnostic
url_github	chr	constructed, cross-checked with DESCRIPTION URL
url_docs	chr	https://openwashdata.github.io/[pkg]/
n_datasets	int	count of data/*.rda
latest_release	chr	GitHub releases/latest tag, NA if none
default_branch	chr	GitHub repos API
created_at, last_commit	Date	GitHub repos API
topics	chr	GitHub topics, collapsed
has_dictionary, has_extdata	lgl	contents listing

The columns the website renders today (name, location, status, description, published date, website link, maintainer, DOI, temporal coverage, keywords) all have a source here. spatial_coverage is the website's location column, published is its status column.

owd_datasets (published packages only)

pkg_name, dataset_name, title, description, n_rows, n_vars (from @Format), n_vars_dictionary (cross-check), csv_url, xlsx_url (raw.githubusercontent on the default branch, NA when the file does not exist).

owd_variables (published packages only)

pkg_name, dataset_name, variable_name, variable_type (first class only), description.

Outputs of a harvest

• R/sysdata.rda: owd_catalog, a list of meta (harvest_date, org, backend), the three tibbles, and the prebuilt search index
• inst/extdata/owd_packages.csv, owd_datasets.csv, owd_variables.csv, UTF-8, na = ""
• No xlsx, no catalog.json, no files table

Acceptance criteria

• test-schema.R asserts column names, types, non-empty tables, and the value gates from Build the harvest orchestrator with local and GitHub backends #4
• The three CSVs have the same rows and columns as the tibbles
• The column reference lives in the roxygen of the accessors (Implement the exported functions #6)

[larnsce]

Premortem findings for this issue (advisory, non-blocking)

From the 2026-07-18 premortem (report in docs/ on branch claude/implementation-premortem-jd5udn):

Shipping a weekly-changing catalog as installed package data freezes every user's answer to "what data exists" at their install date. The site stays fresh because it rebuilds from main, but owd_search() reads the installed snapshot, and nothing bumps the DESCRIPTION Version on harvest commits, so update.packages() sees nothing to update and two installs of the same version can carry different data (which also makes bug reports irreproducible). A later CRAN release would freeze one snapshot for months.

Suggestions, to decide here or in #6 rather than after launch:

• Either add an owd_refresh() that fetches the live catalog.json at runtime with the installed data as fallback, or bump Version on every harvest commit so updates propagate and installs are identifiable.
• Print a staleness warning on attach when the installed harvest date is more than about eight weeks old (the harvest date is already in the startup message; the comparison is the missing piece).
• Treat near-term CRAN submission as incompatible with weekly data rather than a goal.

---

Generated by Claude Code