Skip to content
Technology How it works Breeder — Hyperion Burner — Aegis Burner — MetroVolt AI-Native Architecture Magnets Fuel cycle Safety Roadmap
Solutions AI & Data Centers Defense & Government Grid & Baseload Neutron Detection Quantum
Learn Technical Library
Proof Publications Whitepapers Technical Library Open Science & Reproducibility The Honest Gates
Company About / Mission Leadership Environment Health & Safety Investors Careers Press Contact
3D Model
AI Architecture › L6 · Experience
L6 · Experience

Alerting and Triage UX

Turning anomaly detections, precursors, and interlock events into a ranked, deduplicated, actionable stream an operator can work top-down under pressure.

THE STACK · click to jumpL7Ecosystem & StrategyL6Experience & VisualizationL5Applications & CopilotsL4OrchestrationL3Twin Modeling & AIL2Data FabricL1Control PlaneL0Foundation▲tlmctl▼L6 · EXPERIENCE & VISUALIZATIONHow people see, steer, and review the plant.1Control-Room 3D Twinlive overlays2Plant-Floor SCADAoperations HMI3Mobile Engineeringfield access4Alerting UXtriage & escalation5DashboardsKPIs & health6Replayincident reviewMACHINE TIESurfaces the L3 twin state and L5 copilots to human operators.KRONOS FUSION ENERGYAI-NATIVE S.M.A.R.T. GENERATOREXPERIENCE & VISUALIZATIONSHEET 08REV. 2026-08L6 · AI-NATIVE STACK
L6 · Experience & Visualization — its place in the stack (left, click any layer) and its internal components (right). Telemetry rises; control descends.

From detections to decisions

The lower layers emit many signals: L3 anomaly ensembles flag sub-threshold quench and disruption precursors, SCADA reports process excursions, L1 logs interlock events. Left raw, this is a flood. The alerting pipeline deduplicates, correlates, and ranks these into a single triage stream so an operator always knows the one thing that most needs attention now.

Ranking by expected consequence and lead time

Each alert carries a severity, a confidence from the detecting model, and — critically — a lead time: how long until the operator must act. A high-confidence disruption precursor with 40 ms of lead time outranks a low-confidence cryo drift with minutes of slack. Ranking blends these rather than sorting by raw severity, so a genuinely urgent, actionable item is never buried under chronic nuisances.

python
def triage_score(alert):
    # higher = surface sooner
    urgency   = 1.0 / max(alert.lead_time_s, 1e-3)   # less time -> more urgent
    stake     = alert.severity                        # potential consequence
    trust     = alert.model_confidence                # 0..1 from detector
    actionable= 1.0 if alert.has_recommended_action else 0.4
    return stake * urgency * trust * actionable

The triage card

Each alert renders as a card that answers four questions at a glance: what (the anomaly and where on the machine), how sure (model confidence and provenance), how long (lead time, counting down), and what now (the recommended response and who owns it). For the breeder a disruption-precursor card links straight to the disruption countdown; for the burner a plug-density excursion card links to the plug view.

Nuisance suppression is governed by the alarm-management rules so triage never degrades into an ignored red wall — see alarm flood suppression. Every alert and action is recorded for incident replay.

Content reviewed August 2026 · design-and-simulation stage