Comparing a desktop waveform viewer with a web one: count only what lands on the shared host

When the question is 'how much shared compute does each viewer cost', the deployment model decides what to count. A desktop viewer (GTKWave) run on a remote build host renders there: its own process does layout and rasterization, so its RSS/CPU is host cost; the X server that shows the pixels sits on the user's machine under ssh -X, but the ssh hop encrypting every frame is host CPU too (measure it separately), and a VNC-style setup puts the display server and encoder on the host as well. A web viewer's host cost is just its server plus data reader; the browser's memory and CPU are on the client and should not be in the host column at all. In a single-machine test harness (Xvfb + headless browser on the same box), record everything but report by deployment, and write a one-line method note saying which side renders where. Also state that VmRSS sums over a process tree overcount shared pages (upper bound).