It’s nothing to do with your encoders (dual-potentiometers in fact) - it’s the bi-product of how a binary system can be divided, there are always even value ranges, but most parameters are bipolar and include a zero midpoint which means the range is asymmetric
Roland address this by limiting one range (dropping a potential value) so you get e.g. +/-63
How it’s implemented in this 14bit (High resolution) case is annoying - the highest potential value should not be handled as 127 and 127/128ths but should be rounded up on a one off basis for any similar value internally (i.e. treated as if 128)
so if the Modulation offers the accurate 99.99% of teh range, the final binary value should be assumed to equal 100% internally, even though the binary value is not that
Anyway . there is a bugs topic already, this isn’t a bug, so really this should be renamed and marked as resolved which i will do shortly