But, maybe this is too much info for the more generic --diagnostics printout.

I might be wrong, but I think the only reason --diagnostics doesn't just do the same as --extendedDiagnostics is because we didn't want to break existing tools that parsed the output (and possibly that the latter measures things a little incorrectly?).

Yeah, we talked about this in the meeting. It's now a lot easier to backfill data so we can change things, but the main problem is that extendedDiagnostics requires native performance events which slows down the benchmarks and makes them less realistic (since as we add more trace points, the perf gets worse with the flag on, but not for any actual user).