Desktop app vs client-server viewer: report 'open' from the start point users actually hit, and state both

Comparing open latency of a desktop viewer (process launch every time) with a web viewer (long-running shared server + per-session reader) is easy to get wrong in both directions. A cold 'from server start' number unfairly charges the web viewer for a one-off boot (and test harnesses easily add their own poll waits and browser launch on top: here ~1 s of a 2.2 s figure); a 'from page navigation' number is the realistic per-open cost but assumes the server is warm. Report both, labelled: desktop = launch to first draw; web = navigate to rows drawn (warm server, cold file, per-session reader spawned inside it) and cold = server boot (first-answer time) + navigate-to-rows. Also verify each piece from logged timestamps (SRVSTART/SRVUP) rather than subtracting harness markers, and evict the data file before each take so 'cold file' is true.