Layer Interfaces as Contracts
Each layer exposes an explicit, testable interface to its neighbors — the contracts that make the stack certifiable one layer at a time.
Interfaces, not entanglements
The eight layers hold together only because the boundaries between them are contracts, not conveniences. Each layer promises a defined interface to the layer above and below: defined data shapes, defined latency classes, and defined failure behavior. Because the contracts are explicit, a layer can be tested, replaced, or certified without opening the ones around it.
What a contract specifies
- Data shape — the schema of features, predictions, or commands crossing the boundary.
- Latency class — real-time, standard telemetry, or offline, per the line semantics.
- Failure behavior — what the interface does when the neighbor is late, absent, or wrong.
- Authority — whether the crossing carries data, a bounded request, or a command.
The most important contract
The L1|L3 contract is the one the safety case rests on. L3 may send L1 a bounded actuation trajectory; L1 promises to execute it deterministically or to reject it and fall back to safe control. Critically, L1's failsafe behavior is defined without reference to L3 at all — so the contract holds even when L3 is silent. This is how intelligence and determinism coexist.
Why contracts enable evolution
Because interfaces are fixed, the physics-specific parts of the stack can evolve independently. A better breeder equilibrium PINN or a new burner plug model can be swapped in at L3 without touching L1, L2, or the layers above — provided it honors the contract. This is what lets one architecture serve two machines and improve across reactor generations.
The contracts realize the eight-layer separation and the flows in how the layers connect.