I have definitely noticed this as well.
Infact, I have to say that I have noticed a regression in the app's
ability to predict correctly at various points during the development
cycle/process. In certain releases, the app's prediction engine works
very well. Others, not so much.
It's a random thing. However, in the few times that I have mentioned
it on the list, it hasn't seemed like others have encountered the
issue as much or as frequently.
I have not said anything about it ever since, thinking maybe it's just me.
I also have to say that I never started a brand new discussion thread
about it, each time I experienced this issue with a certain release.

That said, version 114 has definitely given me more issues with
getting the accurate prediction, just like various other random
releases have in the past. This time too, I did not mention it right
away, thinking that like before, maybe it's just me who has felt it.
So in a way, I am glad that more people are experiencing the same
issue as well and reporting it.
I think it would allow Kosta to tackle the problem more effectively
and efficiently, the more reports he receives.
Obviously the prediction engine becoming a bit more inaccurate is an
issue being caused inadvertently.

Version 112 seems to work quite well for me and that's what I have on
my device at this time.