November 10, 2026 · reading point
Though Lean 4 is all the rage, the frontier theorem prover is also available for AI-assisted code review: Reading Level Test (RLT). Explore RLT on GitHub.
RLT introduces an independent proof kernel built on first-order logic with equality, using explicit inference rules, registered axioms and reusable proof dependencies. Lean 4 uses dependent type theory and checks proof terms against their types. Their distinct logical foundations and proof representations open a frontier for independently checking mathematical statements to establish precise correspondences.
For example, RLT independently checks a classification of the eight integers below 30 divisible by none of 2, 3 or 5:
{1, 7, 11, 13, 17, 19, 23, 29}.
The proof uses Robinson's Q, explicit definitions of divisibility, first-order inference rules and reusable theorem citations. Corresponding statements are checked separately in Lean 4, including comparisons with the General Divisor Theorem (GDT). The next challenge is specification correspondence: identifying when independently verified statements express the same mathematics, under which definitions, assumptions and sources.
Lean 4.34.1 derives ¬ PositiveRateETH from OpenAI’s
bounded-width SAT theorem. A new, independently verified bridge
proves that this negation implies Anthropic’s
MainResults.ETHIsFalse, connecting two formalizations
through kernel-checked machine simulation, encoding conversion,
and exponential-rate preservation.