Yellow — AI-Workflow and Optimization

Yellow blocks are where learning and optimization live — training pipelines, surrogates, MPC, and the loops that make the plant intelligent.

What yellow means

Yellow marks the AI-workflow role: model training, surrogate inference, optimization, and model-predictive control. Yellow is the color most associated with the word intelligent, and it deliberately does not touch hardware directly — yellow proposes, and lower layers dispose.

What lives in yellow

• Offline batch retraining at L0 and the MLOps pipeline that promotes models.
• PINN and GNN inference at L3 — equilibria, imputation, stability surrogates.
• MPC agents computing safe actuation trajectories over the twin.
• Optimization loops — shape optimization for the breeder, plug tuning for the burner.

Surrogate speedups

Yellow's value is speed. A first-principles solve of a Grad-Shafranov equilibrium or a neutronics transport problem is far too slow for control. Trained surrogates reproduce those solutions many orders of magnitude faster, letting the MPC loop evaluate candidate actions inside a control cycle rather than offline.

Bounded by lower layers

A yellow MPC agent never actuates on its own. Its output passes through L4's rules engine and safety bounds, then to L1 for deterministic execution. If a yellow model is uncertain or drifts, the pink feedback path flags it and the plant falls back to conservative control — the intelligence is always subordinate to the safety layers.

Yellow's models are documented in the L3 Twin Modeling overview; its dependence on determinism is the theme of determinism versus intelligence.