Observed arrival · 2026-10-28
TheoremOS: a Lean workbench for mathematical agents
An online research workbench for connecting models to Lean and Mathlib proof tasks, then checking submitted proofs with an independent Lean process.
- For
- Mathematical-agent researchers and Lean or Mathlib users
- Worth noticing
- The archive lists six proof packages with official pass records, while attributing the work to DeepScientist and subsequent efforts.
Field notes
The workflow lets researchers hold the task, environment, model, and budget fixed while comparing five components, including proof construction, feedback, search, context selection, and collaboration. Runs have configurable round and time limits, with a stated 15-minute maximum. The archive lists six proof packages with official pass links, but attributes them to DeepScientist and later work; the displayed walkthrough is explicitly preset and not a live model run.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
One card from the complete issue