Signal pipeline

Two emission patterns feed one bus. Continuous-stream sources (accelerometer, microphone, gyroscope, ambient light) run their own detection pipeline over a high-frequency sample stream and emit one Reaction when the pipeline crosses threshold. Discrete state-transition sources (USB, power, audio, Bluetooth, Thunderbolt, display, sleep/wake, lid angle, thermal) observe an OS notification or KVO surface and emit one Reaction per user-meaningful state change. Same bus, same envelope, same fan-out.

Two categories of source publish onto the ReactionBus. Impact sources run three parallel sensor streams through ImpactFusion, which applies consensus and rearm gating before publishing Reaction.impact. Infrastructure event sources (USB, power, Bluetooth, etc.) publish discrete Reaction cases directly. The bus runs a pre-fan-out enricher that resolves audio clip URL, clip duration, and face indices exactly once per reaction. Four independent outputs each subscribe to the bus via bus.subscribe(), receiving an AsyncStream<FiredReaction>, and pattern-match the reaction kinds their per-output toggle matrix enables.

      End-to-end signal flow

The bus enricher runs once per reaction before fan-out. All subscribers receive a FiredReaction with identical, pre-resolved values: the same soundURL, the same clipDuration, the same faceIndices array, and the same publishedAt timestamp. No output re-selects a clip or picks a face. they just consume what the enricher resolved.