The Eight-Layer Stack

L0 Foundation to L7 Ecosystem: a strict layered separation that isolates microsecond safety from petabyte-scale learning.

The eight layers

Kronos separates concerns into eight layers so that the fastest, most safety-critical functions never depend on the slowest, most compute-heavy ones. Each layer exposes a defined interface to its neighbors and can be reasoned about, tested, and certified independently.

• L0 Foundation — cloud HPC and bare-metal supercomputing for offline multi-physics Monte Carlo and batch retraining.
• L1 Control Plane — edge FPGAs enforcing microsecond determinism and the autonomous hardware failsafe.
• L2 Data Fabric — 60+ port telemetry ingest, validation, feature engineering, archive, feature store, vector DB.
• L3 Twin Modeling & AI — GNNs, PINNs, anomaly ensembles, MPC agents, and the KRONOS-CTRL digital twin.
• L4 Orchestration — event streaming, workflow engine, rules and safety bounds, human-in-the-loop routing.
• L5 Applications & Copilots — agentic Plasma, Engineering, and Operations copilots.
• L6 Experience — control-room 3D twin overlays, plant-floor SCADA, mobile engineering.
• L7 Ecosystem & Strategy — grid, supply-chain, maintenance, and regulatory integration via the unified API.

Why the numbering runs both ways

Data and telemetry flow up the stack, from sensors at L2 toward decisions at L3-L5 and human oversight at L6-L7. Authority and commands flow down, from strategy and copilots toward the microsecond actuation at L1. The stepped L7-to-L0 hierarchy describes that command descent explicitly.

Isolation as a safety property

The separation is deliberate: an L0 retraining job can run for hours across a GPU cluster without ever touching the sub-10-microsecond command boundary at L1. If L3's AI stalls, the L1 failsafe still dumps a magnet quench on its own. Layer isolation is what lets a fusion plant be both intelligent and provably safe.

Each layer has its own dedicated overview: L0, L1, L2, and L3.