Payload v1 — Gutenberg plaintext corpus

The full corpus as byte-addressable release assets.

Individually gzipped books inside uncompressed tars: the container stays uncompressed so byte offsets remain exact, while each book is compressed separately (~2.6x measured). One HTTP Range request retrieves exactly one book — fetch ~170KB instead of ~450KB, with no dependency on anything else in the asset.

Every book carries a SHA-256 of its raw text in locator.sqlite, so upstream re-transcription is detectable rather than silently inherited. This is a pinned edition, not a mirror.

Offsets and integrity hashes: see locator.sqlite.gz.

Index v1 — 79,071 works, 36,456 people

Catalog and people-graph indexes, built from the official Project Gutenberg bulk catalog (pg_catalog.csv, 2026-08-03).

books.sqlite.gz (48 MB → 182 MB) — 79,071 works, every upstream column preserved verbatim plus the whole raw row as JSON. 184,624 subject edges, 206,269 shelf edges, 82,405 LoCC classifications covering 99.98% of texts.

people_books.sqlite.gz (7 MB → 23 MB) — 36,456 people, 107,987 person-work edges, 27,771 with life dates. Roles kept as distinct edges. BCE years negative.

Verified lossless: all nine upstream columns present, sampled rows round-trip exactly against source.

gh release download index-v1 -R sisodias/siso-book-library
gunzip *.gz