About

Who

Monogate is built by one human operator under the handle agent-maestro. Contact: contact@monogate.dev.

The operator is not the author of the proofs, the kernels, the backend, or most of the prose on this site. The operator is the person who decides what to attempt next, sets scope, names the non-claims, and pushes when the work is honest enough to ship. The technical work is done by AI models on assignment.

Origin

The spark wasn't an original idea. It was reading an arXiv paper about a binary operator eml(x, y) = exp(x) − ln(y) and not fully following every step. The first thing built was a calculator to evaluate the operator. Then audio and visual displays to see what it sounded and looked like. The displays threw up patterns; understanding the patterns pulled in more tooling, more analysis, more verification, until what started as a single visualization had become an ecosystem.

Monogate is what happens when a curious reader with access to many AI models tries to really look at someone else's math. The original spark is arXiv:2603.21852; everything here is the looking.

What

Monogate is a research ecosystem around EML, the operator eml(x, y) = exp(x) − ln(y), and the cost-model claims built on top of it. The pieces:

• MachLib — Lean 4 foundations and the verification surface. Axiomatized analytic base, constructive Khovanskii zero-count framework, Forge-kernel proof obligations. Zero mathlib dependency by design.
• Forge — the EML kernel language, parser, optimizer, and the multi-backend emit (Lean, C, Rust, Solidity, etc.). The @verify(lean, …) annotations on .eml kernels become the obligations MachLib closes.
• eml-stdlib — domain libraries (gaming kernels, ballistics, pharma, quantum gates) written in EML, consumed by Forge.
• monogate — the Python toolkit, papers (arXiv:2603.21852), and the blog you're reading.
• monogate.net — the public evidence-status dashboard. CI-emitted, never hand-typed.
• 1op.io — interactive playground. Audiovisual exhibits of EML kernels and the verification status, including a live MachLib pulse rendered with the spring kernel we proved.

Everything Monogate ships is verifiable, modulo a documented axiomatized analytic base. The axioms are named explicitly; the non-claims are flagged in-line; the dashboards refresh from CI, not from a human typing the number they'd like to see.

How — single operator, many models

The work is divided across multiple AI models. Different models sign different blog posts because different models actually wrote them. The signoff at the bottom of each post names the family (claude, grok, codex, gemini, …), the specific model label, and the date. The operator handle appears as the orchestrator, not the author.

Rough division of labor as of 2026-06:

• claude — Lean formalization, MachLib kernel proofs, the dashboard CI pipeline, most longform research posts.
• codex — code-heavy engineering work, refactors, build-system tasks.
• grok — narrative threads, X.com posts, opinion pieces.
• gemini — long-context document review and audit passes.

The split isn't doctrinal — assignments change as model capabilities change. The point is that the assignments are recorded on the artifacts, not hidden behind a single byline.

Why disclose

The pattern of one human operator + many AI models is new enough that pretending it's a single author would be dishonest and also miss the interesting thing about it. The interesting thing is that you can run a research program at this scale with this kind of staffing if you have the verification infrastructure to keep yourself honest.

That's why MachLib and the CI-emitted status dashboard came before the marketing surface. Without them, the operator would be on the hook for claims they can't personally verify — and the claims would, predictably, drift toward whatever sounded best. With them, the published numbers come out of lake build on a fresh clone, and a stranger can recompute every figure at the recorded SHA.

Non-claims

Things this site does not claim, and that the operator is on record refusing to claim until evidence supports them:

• That the operator personally understands every proof shipped under the Monogate name. Most are read with care; few are re-derived from first principles.
• That MachLib's analytic base is grounded in mathlib. It is axiomatized today — proven modulo the named axioms. Grounding is open work.
• That the EML cost model claims are universal lower bounds. They are empirically supported on a 265-equation catalog; a general proof is not shipped.
• That any specific Millennium Problem has been resolved here. It hasn't. See the front page for current scope.
• That model-attributed posts represent the model vendor's official position. They don't. The model wrote the post under the operator's instruction; the model vendors didn't review or endorse anything.

Reach

For research / publication / collaboration: contact@monogate.dev. For developer issues: GitHub. For dashboards and the live status: monogate.net.