Ok so this one’s… a couple of things.
The first is that this is the same problem in new app as in old app. Samples at the edge of the visit that fall just outside the radius of the visit will get pushed back out again, when we attempt to assign them to the visit.
The change you’re noticing is that in early AT4 builds there was a separate bug that was slowing down that push-back-out behaviour. So it could look like your reassignment back into the visit was successful, but in reality it was still getting pushed back out, just very slowly, so you wouldn’t see it happen when checking the results immediately after your edit.
More recently I fixed that “edge stealing” slowness bug. It’s now done near instantly, as it’s meant to be. But that makes the other bug/weakness (that both old and new apps have) obvious again. Win some, lose some!
There’s two things I’ve got planned for fixing or improving this. One is… BIG-408, which is the quick one, that will improve the situation marginally, but not solve the problem properly. That one’s about merely tuning the “gravity” of visits, to make them more eager to pull in edge samples rather than push them out.
The other is BIG-174, which is for the more “proper” solution. The speculative idea there is to support “multi centre” places and visits, so that visits aren’t just a circle with a centre, but actually have proper shape, that better represents the nature of the data within them.
For places that would mean place models recognising (in a sense) that there’s multiple rooms within a building, for example. Though how BIG-174 will actually play out, I’m unsure yet. Once we get to it it might end up being smarter to approach it a different way.
Anyway, yep, this is definitely on my radar! As chance would have it I just now had to work around the same problem in my data. I ended up having to assign the uncooperative edge samples to a different place, because timeline processing wouldn’t let me win 
Hopefully the BIG-408 one will land in one of the updates very soon, and improve things a bit so that it’s at least not as annoying. Then in a larger update some point later, we’ll get BIG-174 in.
