Build a library

The uv-managed corpus/ workspace acquires pinned source material, normalizes text and attribution, exports reviewed records, builds a SQLite FTS5 database, and packages it into a .blobs archive. Android installs that archive through the same verified path as other libraries.

The 0.5.0 release ran this workspace's test suite before publication. The commands below reproduce that run.

Work from the existing pipelineCopy Anchor Link

From corpus/:

uv sync --locked
uv run --locked python -m unittest discover -s tests

Follow the corpus preparation guide for source acquisition, extracting a pinned archive or web revision, fetching reviewed records, and building an installable library. It contains the exact commands and revision identities for current editions.

Preserve the source contractCopy Anchor Link

Keep article identity, source URL, section, license, and attribution with each record. Source locks pin revisions and hashes; changed source material must become a new edition rather than silently replacing the published text.

Regional subsets reuse reviewed articles. The Bali subset selects 30 articles from the English travel edition; its selection lock preserves their identities and provenance.

Verify rights before publishingCopy Anchor Link

Confirm redistribution rights for each included work, including article-specific notices. A site-wide footer can differ from an article’s own license. Keep contributor-history links and adaptation notices, and omit media that hasn’t been cleared for redistribution.

Model weights and third-party datasets retain their own licenses. Original source-code licensing doesn’t grant rights to those assets or to Blobs artwork. See the data notices.

Publish an immutable editionCopy Anchor Link

Read trust and publishing for the publisher command, signing, and revision rules. A corrected payload needs a new package edition. Models, generated packages, caches, and signing keys stay out of Git; reviewed record locations and hashes are in the corpus guide.