Measure before changing anything

Touch a known point and record the interval between the physical event and the visual response. Eyeballing “feels slow” leads to tuning the wrong stage.

StageTypical contribution
Sensor scan and packet emissionTens of milliseconds at 100 Hz+
Network transport1–5 ms on a quiet segment
Host application processingVaries wildly — usually the culprit
Rendering / display pipeline1–3 frames, i.e. 16–50 ms

Where the time actually goes

SymptomLikely stage
Delay grows over hoursHost application leaking frames or buffering
Delay constant but highDisplay pipeline, or vsync on a slow refresh
Delay only with many targetsHost-side tracking complexity, not the sensor
Delay only on one machineNetwork adapter offload, or an antivirus inspecting UDP

Cheap wins

ChangeEffect
Disable frame buffering in the rendererRemoves 1–3 frames
Process packets on arrival, not on a timerRemoves up to one polling interval
Isolate the sensor VLANRemoves retransmission delays from congestion
Lower the sensor frame rateSometimes reduces perceived lag if the host was dropping packets

When it really is the sensor

Only if the emitted packet rate is below the configured frame rate. Check that on the wire before blaming the hardware:

  tcpdump -i en0 -n udp port 10940 -c 200
  

Next

Acceptance criteria for latency: Interaction test.