catalogue.json lists the sample projects and their data files, descriptions,
byte sizes, and SHA-256 checksums. All catalogue data files are Parquet; collection
READMEs retain provenance and citation information.
- Queensland election: 2,380 tweets and 133 candidate records.
- Honi Soit: 100 UTF-8 articles with original filenames and document text.
- Reddit: the existing four Parquet tables.
Wordflow desktop creates remote DuckDB views over these files. Resolve main
to a commit SHA, then use that SHA in both catalogue and data URLs for a stable
snapshot. Queries require network access until views are materialized.
The historical demo_snapshots/ bundles are separate from the sample data catalogue
and retain their original formats.
With a checkout containing Git history and uv installed:
uv run --no-project --with pyarrow==23.0.1 python scripts/convert_to_parquet.pyThe script reads originals from revision 17fce77, validates all converted values
against their sources and Parquet round trips, and refreshes catalogue checksums.
No source download or Python runtime is needed by Wordflow to read the Parquet data.