The overview had no mention of a lack of support for "not transitioning" it's certainly possible I'm missing it or it's in the full report (which I'll read when I get a few minutes).
One mention of the need for corresponding levels of support for de-transitioning and some mentions of increased support for other issues alongside the gender based ones.
It sounds like OP had a specific section/sections in mind, if this is indeed the report they were referencing I'd appreciate some indication to which part they were referencing specifically.
"The overview didn't mention it, but its somewhere in this 232 page report" isn't the most useful when trying to understand where someone is coming from.
Lots of practice.
For me it works like an elaborate pattern recognition tree.
e.g. This face in this context means x thing 75% of the time so far.
Then it's "strong opinions held weakly", you now have a working hypothesis but it's just that, a hypothesis.
Every facial/body/word/etc change could be a modifier to the previous assumption. You could also match some newly remembered memory to the situation that also changes the impression of what is going on.
It's exhausting, but it becomes easier with practice.
It's gets more refined the more you are around the same people, as you get a 'feel' for their patterns.
You also start to build up a library of 'shortcuts' that you can sometimes apply to unfamiliar situations/people.
At some point it starts to become 'muscle memory' and the energy required to do it is greatly reduced.
YMMV however, I've no idea if this will work for anyone else in the way I have described.
I'd also say to remember that everyone is guessing to some degree or another it's just that your guesses might require a bit more intention, whichever method (s) you settle on.
You do what you can with what you have, that is the best that can be reasonably expected of anyone.