Palomar: A registry of Lean verified mathematics
terrytao.wordpress.com
August 19, 2026
3 min read
55/100
Summary
Palomar, a registry for Lean-verified mathematics incubated by the Lean FRO and ICARM, has opened for submissions. It records fixed GitHub repository snapshots containing formal Lean proofs of old or new mathematical results, including work produced by humans, AI systems, or both. Terence Tao serves on its scientific advisory board alongside Jeremy Avigad, Matthew Ballard, Jaume de Dios, Nestor Guillen, Bryna Kra, Kim Morrison, Ravi Vakil, and Akshay Venkatesh. Each submission must include a challenge file that states the claimed results in human-readable Lean, a solution module containing proofs, and a formalization.yaml metadata file with an informal description and disclosures. Palomar mechanically uses Lean Comparator to verify that the solution typechecks and proves the challenge file’s statements. A large language model separately assesses whether the informal description appears to match those formal statements, while the registry also checks minimal repository standards. Registration does not constitute peer review for novelty, interest, or mathematical accuracy. Tao reported successfully submitting his recent Lean formalization of the proof of Sendov’s conjecture as a test and plans to submit older formalizations. AI agents can assist with submission mechanics, though Palomar recommends human review.
Key Takeaways
- Palomar registers specific GitHub commit snapshots of repositories containing Lean formalizations of mathematical results.
- Submissions require a Lean challenge file, a proof-bearing solution module, and a formalization.yaml file with informal descriptions and metadata.
- Lean Comparator mechanically checks that submitted proofs typecheck and establish the stated formal claims, while a large language model checks whether the informal and formal descriptions appear aligned.
- Palomar accepts human-generated, AI-generated, and mixed-origin formalizations, but it is not a peer-reviewed journal.
What the discussion said
Commenters treated Palomar less as a novelty registry and more as missing infrastructure for machine-checkable mathematics, with clear implications for AI-assisted research. The strongest enthusiasm centered on the prospect of a connected corpus of Lean proofs: it could make results independently checkable, expose flawed claims buried in the literature, and eventually let researchers or AI systems trace dependencies and discover reusable abstractions. Several readers saw the project’s practical submission path as encouraging, with at least one considering formalizing and submitting an upcoming paper. Tao’s involvement was also read as a meaningful endorsement of formal methods as a companion to AI in mathematics. The thread was not naïve about what gets certified. Readers stressed that Lean can guarantee a formal statement once accepted, but cannot guarantee that the encoded statement faithfully captures the informal theorem a mathematician intended. That translation remains the human bottleneck, so a registry is not an automatic truth machine. Skepticism also focused on whether Palomar genuinely advances the ecosystem beyond longstanding archives for Isabelle and Metamath, and whether Lean’s growing dominance is deserved. A mention of an LLM-based screen for vacuous formalizations drew interest because it addresses a real failure mode: mechanically valid proofs can still formalize something trivial or beside the claimed result. Overall, readers saw substantial promise, tempered by interoperability and semantic-validation limits.
Where opinion split
The sharpest dispute is whether a Lean registry meaningfully improves formal mathematics or merely rebuilds facilities that older proof-assistant communities already provide. Supporters see a curated, challenge-backed corpus with tooling to catch empty formalizations as infrastructure suited to AI-era mathematical work; critics argue Lean is duplicating established archives while concentrating attention on one assistant.
Community Sentiment
Positives
- A searchable network of verified proofs could give AI-assisted mathematics a reliable substrate for reusing lemmas, tracking dependencies, and finding structural patterns across fields.
- Formalizing the existing literature promises more than archival polish: machine checking may flush out long-surviving errors and force vague claims into precise statements.
- The registry’s approachable submission process makes contributors consider turning current research papers into reusable Lean artifacts rather than leaving proofs locked in prose.
- An LLM-assisted check for vacuous formalizations targets a crucial loophole, since proof assistants can validate code that proves a technically true but mathematically irrelevant claim.
Concerns
- Lean acceptance certifies only the encoded theorem; deciding whether that encoding preserves the intended mathematical idea remains an irreducibly difficult human judgment.
- Skeptics see Palomar as Lean recreating proof archives that Isabelle and Metamath have maintained for years, without a compelling reason to privilege one ecosystem.
- The project risks reinforcing Lean’s dominance even though mathematicians may need proof-assistant diversity and cross-system access for a genuinely durable formal corpus.