But how can that be when "... the engineering lead owning conhost.exe: I can't find anything in our code that would trace user input, or data effectively derived from user input, to even a local event stream."
I mean, I suppose that the key sequences around copy/paste are technically user input. Those events get processed into telemetry as to whether they were done in processed or raw input mode, and whether Quick Edit was engaged at the time.
Additional telemetry points include whether the "Find" dialog is being used (explicitly, whether "find next" has been clicked, and how long the search string is -- no other user-generated search content is included in event logs), when a process detaches from the console, and when the window is resized.
I think this disassembly supports their assertion. The number of times a user copies and pastes isn't in itself input to the console, just a metric of some out of band events.
Additional telemetry points include whether the "Find" dialog is being used (explicitly, whether "find next" has been clicked, and how long the search string is -- no other user-generated search content is included in event logs), when a process detaches from the console, and when the window is resized.