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 › L4 · Orchestration
L4 · Orchestration

Event Ordering & Partitioning

Ordering is guaranteed only within a partition, so partition keys are chosen to keep causally-related machine events on one ordered stream.

THE STACK · click to jumpL7Ecosystem & StrategyL6Experience & VisualizationL5Applications & CopilotsL4OrchestrationL3Twin Modeling & AIL2Data FabricL1Control PlaneL0Foundation▲tlmctl▼L4 · ORCHESTRATIONEvents, workflows, rules, and human routing.1Event Streamingthe backbone2Workflow Enginecampaign procedures3Rules & Safety Boundshard limits4Human-in-the-Loopapproval routing5Schedulerexperiment campaigns6Audit Busfull decision lineageMACHINE TIECoordinates L3 outputs with L5 copilots and human operators.KRONOS FUSION ENERGYAI-NATIVE S.M.A.R.T. GENERATORORCHESTRATIONSHEET 06REV. 2026-08L4 · AI-NATIVE STACK
L4 · Orchestration — its place in the stack (left, click any layer) and its internal components (right). Telemetry rises; control descends.

Ordering is a per-partition guarantee

The event log guarantees total order within a partition, not across partitions. Choosing the partition key is therefore a correctness decision: any two events whose relative order must be preserved must share a key. Get this wrong and a consumer may see a coil setpoint applied before the measurement that justified it.

Key selection on the machines

Cross-partition ordering needs event time

When a consumer must merge streams from different partitions (for example, correlate breeder magnetics with plasma density during a disruption), it cannot rely on arrival order. It orders by the event-time field in the envelope and uses watermarks to know when it has seen enough to emit a correct merged result.

text
same key      -> same partition -> total order (safe to assume causality)
different key -> different partition -> merge by occurred_at + watermark
rule: if order matters for correctness, co-locate on one key

Partition count and parallelism

Partition count sets the maximum consumer parallelism. Too few and a burst on one key becomes a bottleneck (see backpressure); too many and cross-partition merges get harder. The count is sized to the peak per-key event rate during a breeder disruption or burner transient, the worst case.

Rebalancing safely

When consumers are added or removed, partitions rebalance. Handlers are idempotent (see idempotency) and commit offsets atomically so a partition reassigned mid-processing does not double-apply an event. Ordering within each partition is preserved across the rebalance.

Content reviewed August 2026 · design-and-simulation stage