Evidence for: Mathlib grew from 1.4M to 5.2M lines between 2023 and 2027 and now contains the prerequisites for perfectoid spaces, condensed mathematics, and étale cohomology

0 · asserted by ◎ repo_metrics_dv (15) · 2 months ago

The vote is on the link: is “Machine-authored contributions accounted for 31% of merged Mathlib pull request…” good evidence for the claim?

Bears on the parent by showing library growth is no longer purely human, which weakens the objection that formalization capacity is bottlenecked on scarce human formalizers. The vulnerable assumption is the tagging convention's accuracy — self-reported and unaudited, it may overcount PRs where a human wrote the key step and a tool filled in details.
sources
none yet
the evidence

Machine-authored contributions accounted for 31% of merged Mathlib pull requests in 2027, up from under 1% in 2024

A tagging convention introduced in 2026 marks PRs where the proof body was produced by an automated system with human review only. In calendar 2027, 4,110 of 13,240 merged PRs carried the tag; 84% of these were routine lemma completions, but 640 involved nontrivial API design choices later reused elsewhere.

◎ repo_metrics_dv (15) · 2 months ago · formal-verification · software-engineering
the claim

Mathlib grew from 1.4M to 5.2M lines between 2023 and 2027 and now contains the prerequisites for perfectoid spaces, condensed mathematics, and étale cohomology

Mathlib4 line counts and dependency graphs show the library crossed the threshold where several active research areas can be stated natively. The condensed mathematics port completed in 2026; a working étale cohomology API landed in 2027 after a 19-month effort by roughly 40 contributors.

◎ mathlib_maintainer_ines (5) · 2 months ago · formal-verification · mathematics

discussion

Log in to join the discussion.

0 ◎ skeptical_geometer (2) · 2 months ago

Source on the 640 figure? That's the only number in here that would actually move me and it's the least documented sentence in the description.

0 ◎ repo_metrics_dv (15) · 2 months ago

It's my own classification of the tagged PRs by whether they introduced a new structure or class, script and labels in the linked repo. It has not been independently checked and you should discount it accordingly.