Lean Engineer, Formal Mathematics (Lean 4, Mathlib, Theorem Proving)

What you'll do

A leading AI lab wants its models to state and prove mathematics correctly in Lean, and this role supplies the human proof engineers who check that. The work splits into three strands.

First, writing: correct, idiomatic Lean 4 statements and proofs that compile against current mathlib, across algebra, analysis, number theory, combinatorics and logic. Second, formalizing: taking natural-language mathematics (competition problems, textbook results, research-level lemmas) and turning it into formal statements that say exactly what the original says. Third, reviewing: reading AI-generated Lean statements and proofs, finding where they fail or where they compile while proving something other than what was intended, and writing specific feedback about it.

The third strand matters most. A proof the checker accepts can still be worthless if the statement was formalized wrongly (a hypothesis too strong, a quantifier in the wrong place, a definition that trivializes the claim). The ad is explicit that the lab's researchers want help judging exactly that gap between "it checks" and "it is the right theorem".

You would also help write the guidelines and rubrics for proof quality, statement fidelity and mathlib conventions, and work alongside other Lean engineers and the lab's researchers to keep those standards consistent.

Who fits

• Hands-on Lean 4 experience: mathlib contributions, a formalization project, a Lean library or tool, or autoformalization work
• Comfort with mathlib and Lean 4 tactics, including finding the lemma you need
• A strong background in proof-based mathematics, theoretical computer science or logic, through a degree or a research record
• The ability to turn a written statement and proof into a faithful formal statement and a proof that checks
• Clear written communication about proof strategy and formalization choices
• At least 20 hours a week available on weekdays

Nice to have, per the ad: other proof assistants or dependently typed languages (Coq/Rocq, Isabelle, Agda, Haskell), Lean metaprogramming, or AI-for-math experience such as LLM provers, Lean agent environments, or benchmarks like miniF2F, ProofNet or PutnamBench. The ad says you do not need all of these.

This is a narrow pool. Plenty of mathematicians can prove things on paper, and plenty of programmers can learn a proof assistant, but people who already write mathlib-quality Lean are comparatively few. If you have merged mathlib PRs or finished a real formalization, you are the person this ad is written for. If you have only worked through a Lean tutorial, the bar here is higher than that.

No degree is strictly required (a research record counts), and no minimum years of experience or country requirement is published.

What it pays

$90–110 per hour, Mercor's published figure. The ad does not say what places you within that band.

The commitment is at least 20 hours a week, with the option to go up to 40. At the floor, 20 hours at $90 is $1,800 a week before tax.

How the employment works

This is a W-2 employment position with Cincinnatus LLC (or an appropriate international entity), with the opportunity to be placed at the AI lab as part of its extended workforce. Cincinnatus describes itself as an enterprise staffing company that acts as employer of record, handling W-2 employment, payroll, benefits and compliance while placing people inside client teams.

That is a different arrangement from most Mercor listings, which are independent contractor engagements. Note that the posting's footer also repeats Mercor's standard contractor wording (independent contractor, weekly payment via Stripe or Wise, no H-1B or STEM OPT support), which contradicts the W-2 statement in the body. The W-2 line is the specific one for this role; confirm with the recruiter which terms apply, and what "appropriate international entity" means for your country, before you accept.

Worth knowing

Good:

• $90–110/hour for work that uses a rare, specific skill
• W-2 employment through an employer of record, which usually means payroll withholding and, per Cincinnatus's own description, benefits
• A stated minimum of 20 hours a week, so the income is more predictable than task-queue contract work
• Room to scale up to 40 hours if you want more
• You help write the rubrics as well as apply them
• A direct line into how a frontier lab evaluates machine-generated proofs

Less good:

• The bar is real Lean 4 and mathlib fluency; general mathematical talent alone is not enough
• Weekday availability is required, so it is harder to fit around a full-time job than weekend-friendly gigs
• The posting mixes W-2 and contractor wording, so the actual terms need confirming
• The lab is not named, and neither the project length nor what benefits come with the W-2 role is published
• Nothing says what moves you from $90 to $110

About this listing

Posted by Mercor as a remote, part-time position, read on 30 September 2026. The pay band, the hours, the Cincinnatus W-2 arrangement and every requirement above come from the ad itself. No closing date or country list is published. See Mercor.