Updated September 30, 2026. The catalog is an ordinary Camino entity with search and readBook remote operations. It has an independent Studio presentation and an embeddable GutenbergPicker. Reader's document-author adapter imports the chosen excerpt through the existing durable publication connection; completion navigates the bookshelf to the new document. Manual text authoring keeps its own independent state.

Catalog and downloads

catalog.csv.gz contains the reviewed September 30, 2026 snapshot of Gutenberg's machine-readable CSV. Its SHA-256 is 62a53fdce11252795b8f247d4cdb51d100a2cf4856a716ff59d53959ad1c19c3. The build parses text editions and bundles the index into the service. Fuzzy search over titles/authors reads that index without contacting Gutenberg. Replace the snapshot deliberately and repin workspace resources when refreshing it. The daily feed is not a build-time dependency: older source revisions must remain buildable even after it changes. The index builder checks coverage and representative titles before accepting a snapshot.

Book text is requested when a user opens a result, from the listed generated-files mirror gutenberg.pglaf.org, following the robot access policy. The downloader reads only far enough to collect about 1,000 words, ending at the next complete paragraph, then cancels the remaining response body. Gutenberg header/footer markers are removed. Short books end at EOF. Exceptionally long paragraphs are bounded at 56,000 characters, preferring a word boundary and preserving Unicode surrogate pairs.

readBook({ebook}) returns {book, url, text, words} in one response. The picker shows a preview and an Import excerpt button; the standalone catalog offers Copy excerpt. There is no paragraph pagination or selection step.

Downloads are serialized, start at least two seconds apart, and coalesce requests for the same book. The process caches eight excerpts, bounded by 32 MiB. The UI also keeps eight excerpts while mounted, so going back to a book needs no new remote call. The first request still needs the mirror to respond. Downloads have a 20-second timeout and 12 MiB maximum; missing editions and failures can be retried. Search never downloads books.

Verification

node --test examples/local-first/packages/gutenberg-runtime/test/catalog.test.mjs
nix run .#camino-presentation-qualification

Unit tests cover paragraph/Unicode boundaries, truncated streams, stopping and cancelling a download, coalescing, caching, serialization and retries. Browser checks search the actual pinned index and stub only the external book response.