How we see things differently
Much of computing rests on choices made decades ago that quietly became defaults. We go back to first principles and ask which of those defaults are actually required. These are the questions that shape our work.
First principles means first principles
Most work that calls itself first-principles keeps the familiar foundations and rethinks what sits on top of them. We do not.
Conventional computing has been built on the same architecture since the 1940s: the von Neumann model. We take none of it into the room with us. We make one concession, and it is practical rather than architectural: for now, our systems have to run on the hardware the world already has. They do, on any conventional machine.
Everything above that line is a clean slate. Every structure, every rule and every decision has to be earned from the ground up. If it cannot be derived, it does not go in, however familiar it is.
Start from what must be true
Conventional engineering starts from what usually works and refines it. We start from what must be true and build only what follows. It is slower at the beginning and far more stable at the end.
Order before clocks
Most systems treat time as a ticking counter. We treat order as the primary thing: what must come before what. Clocks stay useful for people. They are not the foundation.
Derived, never tuned
A result that needs a tuned parameter is a result we do not yet understand. Our standard is zero free parameters, in engineering and in theory.
A discrepancy is missing structure
When a prediction and a measurement disagree, we do not adjust the model until they match. We look for what is missing. That rule has cost us claims, and we keep it.
Nothing counts until it is witnessed
A report that work is finished is not evidence that it is. In our engineering process, a result counts only when there is captured evidence that it happened.
Prototypes are prototypes
We keep three things separate: what is demonstrated, what is encouraging, and what is still a question. Our own audits retire claims that fail.
Questions first, then systems
The same habits of thought run through our computing work and our research programme. If they match how you think, we should talk.