UI latency from frame diffs: time the element that changed, not the whole panel (a 1 s 'slow add' was a late redraw of another row)

Timing 'add a signal' in a waveform viewer as the last change anywhere in the wave panel gave ~1 s for one signal, reproducible across takes, which looked like a slow signal. Two checks overturned it: (1) log every viewer-to-backend request during the step (wrap the frame's fetch; record body, bytes, round trip and the server's own reply time): all replies were done by ~80 ms; (2) re-time on the new row's own pixel band, and bounding-box the late change: the new row was drawn by ~60 ms, and the 1 s change was 18-28 px in a different, existing row (a deferred browser re-render). So report 'element drawn' and 'screen settled' separately, and attribute late changes by bounding box before blaming the backend. The same request log also corrected a design claim: the viewer said to read 'only the window in view' also issued one full-range bucketed history request per added signal.