jeremygracey-ai/provenance-corpus

Provenance Corpus — OpenStax Anatomy & Physiology 2e (page images)

Page-image renders of the OpenStax Anatomy & Physiology 2e textbook (1,347 pages), used as the retrieval corpus for the Provenance demo — a citation-grounded RAG system that verifies every claim against the exact page image that proves it.

Live demo: provenance-web-self.vercel.app

What's here

pages/anatomy-physiology-2e_p{N}.png   # one PNG per PDF page, N = 1..1347, rendered at 150 DPI

Each page is referenced by id anatomy-physiology-2e#p{N}, where {N} is the 1-based PDF page index. The resolve URL for a page is:

https://huggingface.co/datasets/jeremygracey-ai/provenance-corpus/resolve/main/pages/anatomy-physiology-2e_p{N}.png

How it's used

In Provenance these images are embedded with Cohere Embed v4 (a multimodal model — query text and page images share one vector space) and retrieved by cosine similarity. The retrieved page images are then passed to Claude for answering and per-claim verification. The page numbers a Provenance answer cites index directly into this dataset.

The vector index itself is not stored here — it is committed to the Provenance repo (data/corpus/index.npy + manifest.json) and bundled into the serverless function. This dataset holds only the page images, served as public files so the cited pages load in the browser.

Source & license

• Source: OpenStax, Anatomy & Physiology 2e, © Rice University. https://openstax.org/details/books/anatomy-and-physiology-2e
• License: CC BY 4.0. These page images are a faithful reproduction of CC BY 4.0 content and are redistributed under the same license, with attribution to OpenStax / Rice University.