quanta
From-scratch quantum-circuit simulator: 20 crates, 7 backends, 824 tests. Beats a standard industry simulator 3.8–7.8× at high precision.
01
What I'm chasing
Thought experiment · in progress
The Gadsden Serpent — could 600 mph maglev actually be cheap?
Could the United States have interstate maglev at 600+ mph — the speed that turns coast-to-coast into an afternoon — without the price tag that has killed every serious high-speed proposal here? I worked it from both ends, the physics and the cost. I think there's a path that comes out cheaper, safer, and faster than both today's steel-wheel high-speed rail and the evacuated-tube bet. No vacuum tube.
The vehicle
A maglev "skytram" — elevated, magnetically levitated, zero rolling friction. SR-71-inspired low-drag body plus a flow strategy I call self-organized collapse: shape the wake to settle into a collapsed, low-drag state instead of fighting the air.
The propulsion
Fully electromagnetic and distributed — active coils on every car, no dead weight, so the whole vehicle drives itself. Energy lives onboard in novel sodium batteries, not in a continuous wayside power plant.
The guideway
Because propulsion rides the vehicle, the guideway stays cheap and passive — and powers its own signaling and failsafe loads off waste heat via thermoelectric harvesting (50–150 MW continuous network-wide).
The routes
Drawn in public by slime-mold simulation and quantum optimization — never a consultant's lines. Protected land is encoded as edges the simulation physically cannot cross. Every route traces to a published objective function.
State — honest: a feasibility thought experiment, not a built thing. Solid: the self-powering guideway and the sodium-cell chemistry (already verified in my physics sim). The open core is energy — holding 600 mph against open-air drag is a power problem, and that's exactly the work I'd want real resources behind. I'd rather show you the open problem than pretend it's closed.
02
Novel findings
Where something is proven, I say proven; where it's a hunch, I say hunch. I verify before I believe — usually to the last digit a computer can hold.
The LeGrandean theorems
quantum information · verified
A small family of exact rules in how structured quantum states come
apart. Three are identities. One — a doubling rule — I proved
directly and verified to ~10⁻¹⁵, machine precision.
Another result overturned an assumption I'd carried: the standard
entangler isn't optimal, and the advantage of the better one grows with
system size. I think the framework connects to training and running AI
models more cheaply, and I have one working data point.
TKFQC
topological QC · theory + sim
A quantum-computer architecture built on a foam lattice — the geometry
soap bubbles settle into — that appears to delete magic-state distillation, the single most expensive step in fault-tolerant quantum computing.
My simulator suggests a 10–80× reduction in spacetime
cost versus the standard approach. A physical device is gated on two
hard problems I've stated honestly rather than hand-waved.
Laedere
physics · early
A particle-collider simulator, fed by the same quantum engine. Early,
but already doing real physics — not stub code.
03
What I've built
moire
A language model built on quantum interference (the Born rule) instead of softmax. Runs on quanta.
vivarium + terrarium + aether
A runtime you can edit while it runs, plus a 3D engine and a verification-first physics sim built on top of it.
proteus
A Mac-native virtual-machine manager that boots Linux or Windows in ~2 seconds. Homebrew-distributed.
rustfusion
The Windows games API, rebuilt clean-room in safe Rust under a five-tier unsafe-code discipline.
rosetta-rust-lens + cargo-tier
A disciplined way to translate C/asm/shell into safe Rust — proven on 19 real ports, distilled into 511 rules, turned into a tool others can run.
tkfqc
A topological quantum-computer architecture on a foam lattice that appears to delete the most expensive step in fault tolerance (~10–80× cheaper).
THE FIFTH + two commercial sites
A 4-player co-op horror game (first phase shipped) and two live, revenue-bearing Cloudflare sites I build and maintain. I finish things.
04
How I work
One problem at a time, all the way to the bottom. Split my attention across five tasks and you'll get five mediocre ones. Give me a single hard problem that actually matters and room to disappear into it, and I'll come back with something nobody expected.
- I verify before I believe. I don't trust a result — mine or anyone else's — until I've checked it numerically. Most of what's here has a machine-checkable proof behind it.
- I document what fails. My research files hold the dead ends written up as carefully as the wins — that's how I keep from fooling myself.
- I'm AI-leveraged, and honest about it. I think, architect, verify, and direct AI to build. It's how one person produced the work above in four months — a preview of how minds like mine are meant to work, not something to apologize for.
- I don't need much. A problem worth my whole attention, the resources to chase it, a pointer here and there, and then room. Run me like a factory line and we'll both lose.
I'm not the right person for everything, and I'd tell you so. I'm the right person for the one thing nobody can quite crack and everybody has started walking past.
If you have a problem like that, I'd like to hear it.
I'm looking for the arrangement that gets the most out of a mind like mine: paid to do the work, pointed in a direction, given the resources — and then left to do what I do best.