Notes · results
What we measured before we drew a single mask.
A chip design should be a conclusion, not a premise. Here is the evidence ours is drawn from.
12 Aug 2026 · founding team
The usual order of operations for a new quantum hardware company is: draw the chip, raise on the drawing, then find out what the physics thinks. We spent our first months doing the opposite — running experiments on rented, commodity superconducting processors, as ordinary paying cloud customers, to find out exactly where today’s devices lose their qubits before committing a single design decision.
Four findings shaped the design brief, and all of them are on the public ledger with their caveats, dates, and job IDs attached.
Encoding works, and we can certify it. We prepared logical states of a small error-detecting code and measured them in all three bases — not just the easy one — and then held thirty-one such encoded pairs simultaneously, each one certified above the entanglement bound. The caveats matter: detection is not correction, and post-selection does not scale. Both statements are on the ledger in full.
Making a code bigger genuinely helps — until it doesn’t. Growing a repetition code’s distance cut failures measurably, then hit a floor. Finding the floor is the point: it tells you what limits today’s devices, and it is why we publish the step that didn’t improve alongside the one that did.
Connectivity is the wall. The full surface code — the code the field’s roadmaps run on — got worse with size on the degree-3 lattice we rented, while the same code and the same decoder improve with size on a degree-4 lattice in simulation. That gap is not a verdict on anyone’s machine; it is the measured price of a missing coupling degree. It is also, more than any other single number, the reason our chip looks the way it does.
The compiler is part of the physics. Rebuilding a scattered layout into local clusters removed most of a circuit’s two-qubit gates before the hardware ever saw it. On today’s devices, routing waste is an error source you can delete at a desk.
None of this is a physics record, and none of it required owning a quantum computer. It required care, pre-registration, and a willingness to publish the runs that failed. That discipline — not the numbers — is the asset we are building the company on.