Frames captured: 0
Frames processed: 0
Frames dropped: 0
Every real-time camera feature on a phone — object detection, AR filters, document scanning — has to reconcile a fixed camera capture rate with a model inference time that rarely divides evenly into it. This simulator runs an actual discrete-event scheduler: frames spawn at the chosen camera rate, an inference "gate" holds each one for the chosen latency, and you pick between the three scheduling policies real mobile ML pipelines use — queue everything, drop frames while busy, or submit only every Nth frame. Watch the conveyor of frames turn orange while being processed, green once committed, or red when dropped; below it, a logic-analyzer strip traces the inference engine's busy/idle state and two live graphs plot effective FPS and staleness over time, exactly as an app developer would read them off a profiler while tuning against a battery and latency budget.