Different data streams like video, maps, and telemetry can present contradictory information, creating a dangerously incoherent view of reality for operators.
Network buffering and asynchronous processing can cause data to arrive out of order, making the common "last message wins" state update pattern unreliable.
Instead of directly updating state, treat incoming data as evidence to be validated and assembled into a single, confidence-aware operational snapshot.
Handle time mismatches between streams like video and telemetry by defining an explicit coherence window to decide when data can be presented as a single aligned view.
Prevent the UI main thread from blocking by moving the high-frequency data ingestion and processing into a dedicated web worker with separate delivery lanes.
Avoid creating "phantom vehicles" by ensuring that UI animations and motion predictions expire when the underlying data evidence becomes stale or untrustworthy.
Abstract away transport-level complexity by having React components subscribe to a clean, interpreted operational snapshot from a worker instead of raw data streams.
The architecture is guided by four rules: events are observations, model time and confidence, preserve critical transitions, and expire certainty with evidence.
Jobs that call for the skills explored in this talk.