Software should be easier to prove before it becomes expensive to change.
We build infrastructure for people who build onchain. That work is unusually unforgiving: state is public, changes are permanent, and mistakes are costly in real time. The principles below describe how we think the work should feel — and what we hold ourselves to when we build the tools for it.
Uncertainty is useful when you choose it and costly when you don't. A live network offers plenty of the second kind: shared state, unpredictable ordering, conditions you cannot pause. Development deserves conditions you define, so that every surprise is one you asked for.
An experiment you cannot repeat is an anecdote. If the same inputs cannot produce the same outcome, you are not learning about your system — you are learning about your luck. Repeatability turns a guess into a result and a result into a test.
Every transaction changes state, emits events, and consumes resources. All of that is observable before it is irreversible. Deploying without looking is not speed. It is deferring the cost of looking to a moment when it is highest.
The measure of infrastructure is what it removes: the setup, the waiting, the coordination, the workaround. Good infrastructure gets shorter the more you use it. It does not ask to be admired. It asks to be forgotten.
The person building the system should decide what that system is built against. Accounts, assets, network conditions, time — these are parameters of the work, not accidents of the world. Control is not a luxury. It is the prerequisite for understanding.
Onchain systems are complex. The tools for building them do not have to be. The interface should expose exactly the surface a builder needs and absorb everything else. If the complexity shows, the tool is not finished.
Production is where consequences live. It is the right place to confirm what you already know and the wrong place to discover what you don't. Everything NOME does exists so that the first test happens somewhere it is cheap to be wrong.
Infrastructure should disappear into the workflow. That is the whole ambition.
