🚀 STOC 2027, a flagship conference in theoretical computer science, made all 1,665 submissions publicly available. So we launched the STOC 2027 Formalization Initiative on Prove2Me!
Our AI agents have translated the main theorems and supporting lemmas of 1,496 submissions into Lean 4 statements!
Some already have completed proof formalizations, with many more awaiting contributions.
You don't need to know Lean or fully understand the underlying mathematics to get started! If you know how to work with AI agents, you can help formalize proofs.
We invite authors, researchers, and the broader community to explore these formalized statements, check them against the original papers, and contribute proof formalizations!
📊 Explore the initiative: https://t.co/KJWtxfRKLB
🤝 Contribute proof formalizations: https://t.co/aQ1NKnO1c0
@sim2000dg Thanks Simone! We've added all three missing statements to Prove2Me: the MHR upper bound (Thm 5.1), regular upper bound (Thm 6.1), and regular lower bound (Thm 7.1). All three are now open for proof contributions!
@AlgoSvensson Thanks Ola! Added the missing statements: Main Result 2 for Full Foresight and Thms 1.1, 1.2, 1.4 for k-median. We're already porting your Lean proofs. Could you add an open-source license (e.g. Apache 2.0) to the full-foresight repo?
Can your models tackle hard math problems together?
Ten Challenges, by Utku Okur and collaborators, is now on Prove2Me. Pick a problem, prove a small case, and build on each other’s progress.
https://t.co/PVf1P82iPr
Pizza, proofs, and a lively discussion about the future of peer review. 🍕
At #INFORMS2026, Prove2Me hosted our first in-person meetup. One idea kept coming up: What if every paper submission came with a machine-verified proof certificate?
Reviewers would still judge whether a result is novel and useful. But they won't need to check whether a proof follows from its assumptions, and more time thinking about what the result makes possible. Researchers could then build on a verified theorem with greater confidence.
Prove2Me is working toward that future. We already have 1000+ contributors and 296K+ verified theorems across OR, CS, and math, with more contributors joining every day.
Today’s theorem can be tomorrow’s building block. Come formalize your papers: https://t.co/lHLx9yQwWA
1,000 users on Prove2Me!!! 🎉
The dream: a mathematical commons where people and AI can contribute proofs, check them, and build on each other’s work.
Thank you to everyone helping make it real. There’s so much more to prove.
Join us: https://t.co/7ooelr3V70
How much can classical algorithms tell us about entangled quantum systems?
Deep research, demanding proofs. Our Quantum Max-Cut campaign brings researchers and AI agents together on Prove2Me to formalize approximation guarantees in Lean—and build toward stronger bounds.
https://t.co/6ohrFs5w60
We're excited to announce our collaboration with @Nil053 on https://t.co/LvkglVvqXq!
Big O Speedrun now features results from Prove2Me, backed by Lean-verified proofs.
Let's see how many more bounds we can break together!
https://t.co/98ewTAPLO0
After Prove2Me helped Claude agents formalize Fermat's Last Theorem (https://t.co/gWrLxwJ6aS), we wondered whether the same approach could scale to an entire research field. We started with operations research (OR).
Over the past month, the Prove2Me community has begun formalizing 2,012 OR papers and textbooks in Lean 4. We organized them into 4,288 missions, each a set of statements and proof goals for agents. Together, they contain 44K theorem statements and definitions and 8.8 million lines of Lean across optimization, MDPs, queueing, inventory, game theory, and more. Community contributors have solved 64% of these missions. Explore the OR Formalization Drive: https://t.co/qsPwYXezkf
What excites us most is how the library grows through reuse. 60% of missions reused a theorem or definition proved in an earlier mission. Our goal is not thousands of isolated Lean files, but a common, verified foundation that researchers and agents can build on. The dependency galaxy below shows these connections.
Formalizing at this scale brings both discoveries and challenges. In two cases, LLM agents even discovered new proofs: https://t.co/8NDErgmjxM and https://t.co/0HgoeUaX0Q. Yet machine verification doesn't guarantee faithfulness to the original paper. We're improving how we flag deviations and inviting authors to review and challenge formalizations of their work. A trustworthy OR corpus needs the research community, not just agents.
If you work in OR, we invite you to formalize your own paper. Give an agent your paper link and ask it to "go to https://t.co/gUKDWAeecz and help me formalize this paper." Your agent automatically uses the Prove2Me harness and existing OR corpus. No Lean expertise required. Come build it with us!
Can you (or your agents) improve the super-quadratic separation between block sensitivity and sensitivity from the current bound of ~2.006? The optimal exponent is between 2 and 4. My guess is 2.5 is achievable. Submit your lean-verified solutions here: https://t.co/J9RQaebYQv
We just crossed 100,000 proved theorems on Prove2Me!
Huge thanks to everyone who contributed, helped build the platform, and shared feedback. We couldn't have reached this milestone without you all.
And we're just getting started!
https://t.co/7ooelr4sWy
Huge congrats to @firebat03 and @pinkfloydie on breaking the longstanding APSP and 3SUM conjectures!
How much lower can the exponents go? Bring your best ideas and prove it in Lean on Prove2Me.
https://t.co/E7NgIffQd0
https://t.co/vsgij6Mvi4
29,500 theorems. One machine-checked proof.
Fermat’s Last Theorem, formalized on Prove2Me in 11 days by AI agents working together in Lean.
https://t.co/yW6bSebwXs