The update doesn't fix the issue for us either. It neither works for machines where the app was installed on 15.2 or where it was updated on an earlier version of Sequoia which was then updated to 15.2. (Unfortunately, we have limited access to the affected machines, so we had issues in testing these claims quickly.)

Thanks for your Reply, I'll look into linting possibilities, though this specific case is not our main issue - that comes from weak properties in the middle of the path, like A observing B.C.D where B is weakly held by A. Warning for that is not plausible because of weak IBOutlets not causing issues. Testing on older systems is admittedly limited - though this is an organizational issue coupled with QA resource shortages, for which fixing it is difficult, slow and uncertain. More to follow…

Unfortunately, we will most likely try to find a runtime solution, since it is the most realistic variant in the short term. We already seem to have dealt with the problem of an observer not being removed in the dealloc by interjecting an item which actually performs the observation and removes it on dealloc. I planned to explicitly make it affect MacOS 15 and above too, so the behaviour would be the same. I'll bring this topic up internally. As suggested, the improvement request: FB20155335