You both seem to be saying that any amount of timing difference - regardless of whether it is sub 1 millisecond, 1-2 milliseconds, 2-5 milliseconds, or above - can be refered to as 'latency'. Can you please clarify?
I can only say "yes, that's true". You may have a different idea of the meaning of the world. That's not just fine, no, I even absolutely get it. I can perfectly understand someone not willing to use the term "latency" when it comes to phase issues, comb filtering and what not - and if I was to describe those, I'd probably not use the term, either.
Still, what I'm saying is this (and it's in fact the same thing just looked at from another POV): One of the things you may run into when dealing with latency, is phase issues and the likes.
And what I'm additionally saying is this: When we're dealing with latencies (of whatever kind), we should rather be aware of all of these aspects, not just about "tactile latency".
See, my example with tracking a singer was 100% real life experience. Happened to me way back. I had very little clue about things, but I had that very powerful PC I've built myself (I think it was when CPUs topped out at around 500-800mHz and when they were single CPUs) with that RME Hammerfall PCI slapped in. Must've been the same time when the first at least sort of usuable guitar amp plugins came up (SimulAnalog Guitar Suite anyone? I think they're the heads behind Overloud these days...), so I was enjoying that an incredible lot, as I was able to run my system @ 64 samples and some RTL values around <5ms.
Then it came to tracking a singer. And we were running into the aforementioned issues. As said, I had no idea what it was - because I was able to track timing critical guitars with ease. So WTF was that stupid singer complaining about with his soft attack "instrument" and his way less than in-the-pocket overall time feel? "It feels so strange" and what not. F***** p***y. That's what I thought. I then checked his headphones singing myself for a bit and that was revealing what he meant - still couldn't explain it.
So we decided to ditch the neat monitoring mix of his voice and monitor through the RME hardware, using TotalMix (was it even already called TotalMix back then?).
Later on however, when we were already checking takes, he said he wanted to do another take or two and I forgot to recall the tracking patch in Total Mix and software monitoring was on for reasons. And all of a sudden he was "maaan, that is a glorious monitoring sound!"
WTF?!?
Ok, it's so long ago I don't exactly recall how I examined things after that event, so long story made short: It turned out that some plugin (I think a compressor) *raised* the monitoring latency - and because of that, the comb filtering phasey-ness went away in favour of what was likely sort of a nice, thickening doubler effect.
This very thing even came up again shortly after, on the LUG (Logic Users Group) when it was still a Yahoo Mailing List, after I started a little comparison thread about interface latencies. There's been another bloke *exactly* confirming my findings of that vocal tracking session.
Now, maybe this is common knowledge these days, back then it wasn't, at least not for me. Anyhow, my takeaway to this day is to be aware of any practical aspects of latency.
And as said, we haven't even touched latency compensation in this thread...