Here is my take on the subject.
First of all there are 2 different things in this topic, one is the principle of loudness mastering ( LUFS ) against the Digital Full Scale Mastering approach ( dBfs ) & the second one is more about references & technical LUFS levels that you should aim for when putting your music online.
Like Phill exhaustively already explained, both stories are starting at the same and the most important point : what do you want to go for you music when mixing & mastering a track ? Do you want more dynamic in your track vs loudness & heavy level competing & matching commercial mixes or club ones. Therefore, you won't take care of a mix the same way and it's also obvious that depending of the genre of music, dynamic would be more or less present.
In general, EDM style and electronic music would have less dynamic than pop, rock or classical music, but again, like Phil already said, if your music is sounding good to you and if you take care about keeping dynamic range while maintaining some loudness in your mix, then your music will translate correctly in many situations. That is for offline & local mix production, the classic approach of mixing using dBfs and taking care of keeping Dynamic range and headroom around -6dB works in most case, but it's better to take a look at Lufs right from the start to save you from running into issue when you decide to put your music online.
Therefore, getting back to your question about Lufs references levels for uploading your music on digital platforms, you should look at it as an extra step ( and the last one ) of your mastering process in order to match a correct final output level after the AAC & MP3 digital audio encoding process by those online platforms.
Of course these normalisation algorithms & the encoding process will introduce changes in your music because they are going to find the peaking & weakest frequencies in your track and squash or raise the overall mix levels according to the algorithm founds.
That is why LUFS is the metering reference chosen to achieve this, because it is capable of extracting a full track dynamic as opposed to dBfs which is more for instant peaking metering. But those audio formats like AAC & MP3 are no lossless formats, therefore, if your mix is too far away from those targets recommendations, it could very well end up in sound quality degradation.
Audio files normalisation is basically aiming at reaching a kind of "consistant" level in order to provide a friendly listener experience while achieving file size reduction. Those companies know that people are using "playlist" features and they don't want them to have to reach for the level knob on each new playing track, and they also have to respect and conform to standards based on health care, potential ear damage issues in an era where digital music is mainly consumed on portable devices with an extensive use of headphones for hours in a day. But the purpose under the hood is also to reduce file size for server hosting on huge platforms with millions & more of audio files, as well as for individual portable devices storage capabilities.
And that's where it hurts, because of the non lossless audio file format that are used to achieve this. So that's why it's important to try to get as close as possible to the Lufs target recommendation of each platforms. As already said by Phil, the route to go for in this case is aiming for dynamic vs loudness. If you submit an already mastered audio file using the typical dBfs mixing approach and reaching - 3 or -4 dBfs on your regular meter without taking care of those Lufs recommendations, chances are that your track is going to be heavily squashed & that sound quality won't conform at all with what's coming out from your master channel. In the opposite you will have greater results by uploading a mix with more dynamics.
Now how to achieve this and what are those requirements you may ask.
Well, it's a common practice to aim at a general Lufs level right from the start at the mixing stage. Reaching for min 12 & max 14 Lufs is a good mid range start point to aim for. So that means mixing at low levels and following rules like the Bob Katz K-14 or even K-18, it's worth mentioning here that 0 dBfs in the digital world is the limit: there' nothing above this point, the audio information is simply lost if you are clipping above 0. Also it's worth mentioning that the digital 0 dBfs is matching -18 or -21 db on any analog desk !!
You get the picture... If you mix & master just taking care about the common -6 dBfs dynamic range safety net recommandation & aiming at - 3 or even above on your master channel, chances are that you've been fulled by your ears and perceptive loudness right from the start. You will then miss the real loudness in your track, as written in the first part, you'll manage when mixing for offline audio file rendering, but it's gonna cause issues for online uploads.
So again, it's good to keep in mind that dBfs is more of an instant measurement while Lufs metering is relevant on a full track length, that's why it's the metering reference when it comes to digital normalisation algorithms because as mentioned before it's a matter of consistant level matching across a track or even an album.
So if you take the habit to mix at low level and use the K-14 metering measurement you'll avoid perceptive loudness issue first of all.
Sending out your audio outputs to a pre-master bus in your DAW is also a good habit, it will give you the chance to adjust the overall Lufs level on your entire Mix and performing a correct gain staging per track or/and groups and taking care of both Dbfs & Lufs metering for those.
The goal is to reach your pre-master Bus with your mix hitting not more than -13 or -14 on the Lufs scale metering.
Keep your master channel vanilla and never mess-up with your master fader, just leave it at 0 dBfs.
Good levels have to be achieved during mixing & gain staging, not during mastering stage.
When you're happy with your mix on the pre-master, it's a matter of starting to get the desired Lufs level on your Master Channel : something like -16 for more dynamic and uploads and even - 8 for the loudest commercial releases ( that is if you don't plan to upload online ) . This is achieved mainly by using a good limiter at the very end of your Master Channel, in order to get from this -14 Lufs to -16 or more. If you started your mix using the k-14 scale, it's not rare to have to crank up your limiter with 8 or 12 dB !! That's why you need a good one, as much transparent as possible.
Once you reached this target according to the Digital Medium Lufs requirements for upload, you should be good to go and retain the best audio quality once your mix has ben converted to AAC or MP3.
Many tools can be used to help you achieving this, some of them will even provide you with target profiles to help you match those medium's references more easily.
To mentioned a few and to see some tutorials here on Sonic Academy :
Expose from Mastering the Mix :
and before reaching expose ( more a checking stage process ) you want to have a look at their Levels plugin which will help you with Lufs per target monitoring and Dynamic range monitoring as well during mixing. Those plugins are great tools and won't break the bank.
You'll also find some very handy and affordable tools by HoRNet plugins like their HoRNet LU Meter with an auto gain feature and there's a bunch of Lufs metering tools out there, but you need one.
In another price range, especially the Pro version, is the excellent MasterCheck by NuGen Audio which offers an extensive set of different metering & levelling target profiles and will let you hear and see the difference between the target and your mix in real time and therefore help you compensating it via your limiter. Though this tutorial here by Dom Kane is showing another great way to use the plugin, it will give you an overview of what this tool can do
Other tools like WaveLab9, Ozone from Izotope or The Fraunhofer Pro-Codec or their Codecs Tool Box from Sonnox and again MasterCheck Pro from Nugen Audio let you listen to your track as if they would have been encoded offline following standards like AAC or MP3 and some others codecs, this can also be handy.
Finally for macOSX users, it 's good to know that there is an Apple ‘RoundtripAAC’ that lets you preview how your track will sound in a lossy format, this is included in a set of free tools for iTunes Mastering, you'll get it here if under macOS:
Have a look at this topic on Mastering The Mix website, it will give you some Lufs Targets as well ( as per as 2017 ) but most of all a good picture of the principle.
Hope this could help,