Monogate Research

Frontier map

The current frontier is no longer "EML wins everywhere." It is a repeatable review loop: EML expressions become packets, packets become guard decisions, guard decisions become lowering plans or proof obligations, and monogate.org keeps the public claim boundary honest.

Current center

EML Advantage Lab

The Advantage Lab asks the corrected research question: where does EML help, where does standard/protected math win, and which claims remain blocked? The first scoreboard is deliberately mixed: EML is strongest as a generator, proof/search grammar, and teaching lens; standard or protected math currently wins many runtime and stability cases.

Active instrument

Proof-carrying rescue suite

The rescue suite tracks four optimizer boundary events through named rescue operators and replayable witness packets. The generated Explorer fixture is derived from the Forge manifest, replay JSON, obligation registry, and approval gate.

compiler boundary

Guarded lowering

A9 through A11.3 convert EML advantage evidence into conservative compiler policy: preserve EML for proof/search shape, lower near-zero `exp(x)-1` to protected `expm1`, lower softplus/log-sum-exp to protected forms, require positive-domain guards, and block unstable deep trees.

Open packet builder

symbolic search

PySR reality check

The first real private PySR run did not show a robust EML grammar advantage on the prime-residual fixture. That is useful evidence: it narrows the claim to better feature design, sharper targets, and pre-registered controls instead of treating EML universality as a performance claim.

Open template search

machlib

Small checked witnesses

MachLib has small witnesses for softplus positivity, sigmoid denominator nonzero, eml(x, 1) = exp(x), and the Atlas-facing subtraction boundary eml(log(v), exp(u)) = v - u under 0 < v. These are scoped footholds, not full EML language semantics.

See status table

claim boundary

What is claimed

We claim generated analysis artifacts, simulated traces, replay checks, and explicit proof-obligation names. We do not yet claim production optimizer rewrites, broad EML superiority, compiler correctness, hardware observations, zeta-zero discovery, or fully discharged formal proofs for the whole stack.

Surface split

Research record, workbench, visualization

monogate.org is the research record. monogate.dev is the developer playground where people inspect packets and run tools. 1op.io is the games and visualization surface. This split matters: the dev site should be usable, while the research site carries the claims, caveats, and public status.