Two tracks in question, original album version:
A version that appears on a compilation:
The version that appears on a compilation is only slightly shorter due to a more aggressive fade-out… the recording is no different any other way.
So question, should this be merged into the album version or does this recording stand on it’s own?
Style / Recording - MusicBrainz already covers fades. They’re not considered separate recordings if they don’t “cause the structure of the song to change”
An approx rule is if you can play then at the same time and not spot a difference, then they are the same recording. Faded earlier is still the same recording, just less of it. Compilations often cheat some time off of tracks to gain some extra space on the disk
I find Audacity handy for a visual compare
You mean parallel? Then I would notice the earlier ending even it’s only about 2 seconds. 
If played one after the other, more difference would be required.
But how much less can it be before it becomes a separate recording?
In my example (my problem) it’s about considerably less Mark Knopfler guitar solo (though fading):
Early Dire Straits CD releases have many tracks shortened (for whatever reason) compared to the original vinyls. But on Making Movies it’s extreme. 5 of the 7 tracks are shortened, one by ~5 secs, three by 10 secs and one by 20 seconds. The 1996 remaster uses again the full length versions (as on vinyls).
In detail:
Tunnel of Love with 5 seconds is merged, okay.
Skateaway and Skateaway with 20 seconds difference are separate, also okay - you would probably notice the difference.
My problem is that one of the songs with 10 seconds difference is perfectly separated (Expresso Love and Expresso Love), the other two are merged. All three are actually shortened by almost exactly the same amount, almost exactly 10 seconds.
In short:
Can we agree on a number of seconds up to which an average listener will not notice a difference?
It would be easiest to merge, but I feel a little uncomfortable about it. So far I’ve set the limit at around 5 seconds. And it took some effort to keep these versions separated. 
I can see the logic to the current agreement that the recording is still the same recording, just shorter. A note in the annotation can cover that. Something common is the cutting off of crowd noise. Or fading out an extra repeat of a chorus. It makes sense keeping them together if a compilation trimmed it for space. The OP example of a compilation just chopping for time seems a good reason for the guideline.
I don’t think anyone would shoot you if you separated the Dire Straits tracks that are 10 and 20 seconds different and annotated the links between them. That has been done to make a very specific difference between LP and CD edits. Not just hacking a few seconds off for a compilation. Something worth noting in an annotation, but don’t be surprised if another editor then merged them back together.
Where this is a little inconsistent is this guideline only applies to someone cutting the end off of a recording. If someone cuts five seconds off of the front of a recording then it gets a separate Recording.
I have sorted something similar out with Chumbawamba’s Tubthumping. These are all exactly the same recording but all have been topped and tailed in different ways.
I split it four ways:
Technically the second and third versions there could be merged as the only difference is the ten second fade, but I left them separated.
This track appears on so many different compilations it is actually interesting to track which version appears where. Well, it was interesting to this music geek 
. And I feel other people will like to know which version they have.
The main reason I left the split recording is due to how many examples there are. If I had only found two or three places where it was faded early, then I would have merged it.
This also makes it easier for editors to find it from the Release\Recordings page when adding compilations. All four versions appear on compilations and three are audibly very different start and ends.
…but that’s about a few seconds. In case of Dire Straits I would have classified these significantly shortened versions as edits.
However, if I read the guidelines carefully, I would also tend to merge ← it does not “cause the structure of the song to change”.
I don’t think that’s what it says: “…different volume fades at either end of multiple tracks are not reasons to maintain separate recordings”
Thus the recording of “Skateaway” (20secs) should also be merged. 5s at the beginning (fade in) and 15s at the end are missing. But I think I will leave it separate and consider it an edit.
This must be guideline for works, it cannot be guideline for recordings. 
Here if a recording is missing a significant part of the outro solo, I would not merge.
This is from Style / Recording - MusicBrainz → Merging recordings / Recordings with different durations
My interpretation of the guidelines seems to be disputed:
These edits are about merging recordings of a “bootleg” compilation which vandalized some Dire Straits songs.
→ Edit #105629853 - MusicBrainz (fade in at 0:07, at least 15 seconds shortened at the end)
→ https://musicbrainz.org/edit/105663317 (at least 14 seconds shortened)
But they do not “cause the structure of the song to change”. They have still “different volume fades at either end” and are “considered mastering differences”.
In the first 10 seconds of Telegraph Road you can hear almost nothing at a moderate volume level - the sound increases over 18 seconds of which the first 7 have been removed.
I do not appreciate this and maybe they should be different recordings. But it seems to me like the guidelines say not to keep them separate. (→ Style / Recording - MusicBrainz)