This would indicate the root cause of input acceleration is not corrected in the latest update yet, only some of the input acceleration at a level maybe closer to the input polling loop only, whereas the root cause is a global acceleration “state” which should be implemented as a per-input acceleration “state” instead. For optimization purposes it is possible the code is not maintaining and checking a per-input state, but a state for some inputs only, and they’d need therefore to manually add the ones they want to differentiate in a list of some sort. But this is just speculation at this stage.