16M Lines of Formalized Code: Axiom Moves an Entire Mathematical City into the Machine

Axiom posted a 16M-line formalization project on X, covering the entire Stacks Project. Turing Post reposted the announcement.

First, some context. The Stacks Project is a long-running collaborative reference text in algebraic geometry; researchers use it to look up definitions, theorems, and proofs. Formalization means writing this content as machine-checkable code. At 16M lines, this is beyond sample code: an entire body of text that needs to cross-reference itself has been moved into a codebase.

Axiom's judgment in the tweet is more direct: many people think large-scale vibecoding is slop, and this project is a counterexample. It says the code quality has been optimized, and the highly interconnected structure of the Stacks Project has also been preserved.

For formalization projects, structure is harder than line count. Definitions, theorems, and proofs must be able to reference one another; changing one place may affect a whole area. Axiom specifically highlights preserving the interconnected structure, which shows they know where the difficulty lies. Line count itself does not indicate quality, and 16M lines is not automatically a plus. Being checkable, locatable, and having clear dependencies is what determines whether this thing is usable. If it were only about line count, many repositories could win; the hard part is making every line hang in the place where it belongs.

The visuals compare the project to a mathematical city. General readers can understand it as: every building, every road, and every pipeline has a number, and the connections between them are also preserved. AI practitioners should look more at how this kind of project provides a stress-test scenario for code generation: large scale, dense relationships, and errors that cannot be glossed over by eye.

Source: Turing Post repost of Axiom's release