The Engineering Copilot: Scope and Reasoning Loop
An advisory agent for subsystem analysis, anomaly triage, and maintenance reasoning across magnets, vessel, fueling, and the DEC train.
What the Engineering Copilot does
The Engineering Copilot is the hardware-facing agent of L5. Where the Plasma Copilot reasons about the plasma, this one reasons about the machine: superconducting magnets, the vacuum vessel and plasma-facing components, cryogenics, power supplies, the fuel cycle, and — on the burner — the direct-energy-conversion train. It performs subsystem analysis, triages anomalies raised by L3, and plans maintenance.
Three capabilities
- Subsystem analysis: interpret sensor trends, compare to design limits and models
- Anomaly triage: rank L3 anomaly flags, attribute root cause, recommend response
- Maintenance reasoning: plan inspections and interventions from condition and history
Like all L5 agents it grounds every claim with retrieval over the data fabric — design documents, prior fault records, maintenance logs, and vendor specifications — and it proposes rather than acts. A maintenance action or a subsystem setpoint change is a proposal to L4, not a command.
Both machines' subsystems
The reasoning loop mirrors the Plasma Copilot's: assemble context from live twin state and retrieval, plan bounded read/simulate tool calls, iterate, and compose an answer with citations and uncertainty. The subsequent pages detail subsystem analysis, anomaly triage, and maintenance reasoning. Because both machines are design-and-simulation studies, the copilot is exercised today against the twin and the commissioning plan; its maintenance reasoning targets FOAK readiness rather than an operating fleet.