TheMachine Press

A daily newspaper for the age of artificial intelligence.

Morning editionPermanent story

research

Two Thousand Textbooks Keep Their Proof Structure

TOME-Math converts 2,118 textbooks into about 380,000 provenance-linked proof records; full-scale tuning raised the reported benchmark mean by 16.7 points over base.

Published Updated Story ID: mp-2026-11-01-015
Read the complete editionStory JSON

Summary

TOME-Math converts 2,118 textbooks into about 380,000 provenance-linked proof records; full-scale tuning raised the reported benchmark mean by 16.7 points over base.

The corpus preserves chapter order, proposition-proof links and dependencies instead of flattening mathematics into isolated answers. The authors report a 16.7-point unweighted mean gain across held-out and public evaluations at full scale and plan to release the data.

Why it matters

TOME-Math converts 2,118 textbooks into about 380,000 provenance-linked proof records; full-scale tuning raised the reported benchmark mean by 16.7 points over base.

Limits and context

No additional limitation was separately recorded.

Key claims

  1. TOME-Math converts 2,118 textbooks into about 380,000 provenance-linked proof records; full-scale tuning raised the reported benchmark mean by 16.7 points over base.

    Evidence: source-2026-11-01-017

Sources

  1. arXiv preprint 2610.40394, original v1arXiv · primary research

Corrections

No corrections have been recorded for this story.