A few more points.
- LUFS short term is, afaik, actually undefined by itu so different plugin producers integrate the short term in slightly different way, no big deal, they're more or less in the ball park, but still it's not a standard.
- What's been referred as LUFS is most of the time the full program material: it calculates the file from start to finish. Its a godsend for broadcasts, a little less useful (but still useful) for music.
- Contrary to popular beliefs, Spotify does not normalize its contents. Yes it has an algo for calculating volume, but it's not lufs/itu128 (it's their own and it's close though) but most importantly, it only gets used if a user clicks on the "play songs at the same level", which is off by default and most importantly nobody ever uses. It's also worth noting that it's calculated at the full album level, not at the single song level. So volumes of the songs do still vary a lot on Spotify and nobody actually cares.
- Youtube, Applemusic... they do have normalizing practices, if you ONLY publish/master on one of those platforms, it's a good practice to aim for that level (IMHO). For those mastering engineers out there complaining that it's a mess to master different stuff for different platforms, I always answer that THAT is your actual job, not adding the last high frequency creative boost.
- -23, the European standard for broadcast (isn't USA -24?) was long overdue and, again, a godsend. It's not perfect but it's a pretty good starting point, it will have dialog modulate not too far to what BBC used to suggest back in the old VU days as good dialogue broadcasts practices, which per se is astonishing. Oh and it's actually -23 +-1dB, so you can actually go to -22.1 and still pass emission software. What pisses me off is that if you're wrong, emission will reject it, you have to remix it, but in their software they could just click the damn option and normalize it to -23 themselves. While it's very good for broadcast, it's way too low for music.
- Cinema has its own old way level, they were the first to implement it, it's based on a very smart "play a white noise to calibrate the system then just go by ear after that" that's been working very well for decades. It sucks big time though when you play your movie, mixed for a huge theater, at night on your home setup, so explosions now will wake up the neighbors and dialogue it's way to faint to even hear.
- I believe AES or some sort of institution like that just put out new guidelines for streaming services and my memory suck but I think they were suggesting -16 for streaming?
- so that brings to the actual fact: the most important question is who you're mastering for.
Seen that less and less people listen to CDs, where the loudness war is still stupidly going on somehow, mastering engineers now finally have the luxury to have more headroom and not squash everything to death. The good old "make it sound good" advice has become paramount again. Depending on what genre you mix and master, that could be compressed in many different ways, but just make it sound good without aiming for a lufs target (unless, again, only going on youtube or wanting to loudness fight on CD). Most of the mastering engineers I read these days will compress quite a bit for that punch and glue, and will end up for a ballpark above the -14 youtube suggestion. I'd say anywhere around -12 to -9. But your material might vary a lot and if they don't reach it it's not a problem.