MERICANII

Qwen 3.8 27B at 38 t/s and Qwen 3.6 35B-A3B at 56 t/s on a used RTX 3060 12GB, quality held to within one HumanEval task.

Re-verification of the Four Color Theorem's computer proofs, plus a new theorem.

A sorry-free Lean 4 formalization of the 2026 two-thirds-of-zeta-zeros theorem.

Validation-driven dictionary training that beats zstd's own trainer by 3–32%.

An RL environment where agents assemble non-halting proofs for Turing machines.

Whether sorting 13 inputs needs 44 comparators or 45 is still open.