Building it

Ask Physics is MIT licensed. Everything it learns from or ships is either written by the project or licensed so that it can be used, and credit travels with it.

The prose corpus

About 20 million words of English, built locally by scripts/build_corpus.py and never committed (ADR-017):

• OpenStax textbooks (33 repositories, 52 books: physics, biology, chemistry, history, economics, writing, and more), licensed under CC BY 4.0. OpenStax moved later versions to CC BY-NC-SA in March 2026. A CC BY license can't be taken back from a version already published under it, so each book is read at the last commit from before the change, and the build checks both the license file and the book's own metadata at that commit and stops if either isn't CC BY 4.0.
• Public-domain books from Project Gutenberg: only books where everyone credited died in or before 1955, sciences first, then other non-fiction, then literature. Project Gutenberg's header, footer, and trademark are removed.

The build keeps prose only (no figures, tables, exercises, or display math), drops repeated paragraphs and reference-list entries, and drops every paragraph with a slur or a swear word. The word list is stored only as hashes.

Each build writes ATTRIBUTION.md (every source and its license), the CC BY license text, and a lock file of exactly what went in. Any model trained on the corpus carries the same attribution on its model card.

Equations and constants

Every equation entry was written by the project, with a source cited for checking (mostly OpenStax, as a reference) and its own license field, almost always MIT. Constants cite CODATA and NIST. See Equations and Constants.

Everything else

Fonts, imagery, and software dependencies keep their own licenses, listed in THIRD_PARTY_LICENSES.md.