Free technical books that anyone can check
AI drafts them, an automated pipeline keeps them current as the libraries they teach change, and anyone can check and correct them in the open. Every edition shows the date its examples last passed, how many pass and what changed.
always current every example runs: each column is a letter in binary, a = 1 to z = 26, top hole highest. Every cover carries a tape that spells its title.
What every edition shows
Nothing here is typed by hand. Each piece is generated when the edition is released, from the record the release writes.
The edition stamp
Every example runs
import pandas as pdfrom lib.store import DATA, load_labels, load_products, load_queriesproducts = load_products()queries = load_queries()labels = load_labels() # one row per judged (query, product) pairconflicts = pd.read_csv(DATA / "label_conflicts.csv") # pairs WANDS labelled two waysproducts: 12,000
queries: 80
train queries: 50
test queries: 30
…Python 3.12.15 · numpy 2.5.4 · pandas 3.0.6
The history tape
Anyone can check this book or correct it. Report an error · How to fix one
How a change shipsEvery cover, PDF and printout carries an edition code. Type it to see whether it's current.
No paywall and no ads. Text is CC BY 4.0 and code is MIT. Credit Rolling Edition, link back to the book and to the licence, and say what you changed. Assistants read the same editions as Markdown, with stable section links to cite.
How assistants read the booksThe update pipeline
Each book is rebuilt and rechecked whenever a library it pins releases a new version, and anyone can check the result or correct it.
A bot opens a pull request that moves the book's pinned version to the new release.
In the exact environment the book pins, with the network off. The output must match. The checks also confirm that every section keeps its permanent link.
Once every check passes, an agent approves the change and it merges. Anything with a serious risk, such as a security or copyright problem, waits for a Rolling Edition maintainer. If something is still wrong, anyone can correct it with another pull request.
A new edition number and code, a new tick on the cover and a new column on the history tape. The target is 4 days from a library's release, or 6 days for a breaking one. That includes the 3 days we wait before adopting any new release, in case it's withdrawn.
How search engines find and rank results, and how to measure whether a change helps: lexical, semantic and hybrid retrieval, reranking with learned models and LLMs, and judgments from people and LLMs, on a real product catalog.
Edition 2026.10.1 · Examples passed Oct 18, 2026 · 132 of 132 examples pass