The UDA thesis

The Unified Design Architecture (UDA) is the claim that the layers we usually buy from different vendors — the chip, the OS, the language, the UI — are most powerful when they are one designed thing. Historically only a few systems achieved this: System/360, the Lisp Machines, Smalltalk-80, and — in our era — Apple’s M-series. Each was a vertical artifact where the abstractions matched the metal.

Why now

For most of computing history, building a full vertical stack was economically impossible for anyone but a giant. That constraint just broke. AI models and code-generation agents collapse the unit cost of producing a coherent stack — which means the co-designed system is buildable again, by a small team, on commodity silicon.

JouleOS takes the UDA seriously but refuses the one luxury the historical systems had: we do not own the silicon. So instead of assuming the hardware, the runtime measures it and lowers against a control vector:

• Address-space ownership — do we own the page tables, or are we a guest?
• Capability enforcement — hardware-assisted, or software-checked?
• Persistence directness — is memory durable, or must we stage to a device?
• Dispatch reach — what can we call without leaving our trust domain?

The same intent IR is specialized at boot against these axes. The result is one program with many honest lowerings — not a lowest-common-denominator portability layer.

The consequence

When the stack is co-designed, the abstractions stop lying. A pointer can be durable. A capability can be unforgeable. Energy can be measured rather than estimated. The rest of the docs are the working-out of that single move.