saw in the manual that the threshold setting improves the detection so im assuming it does. it seems to me likely that people will not understand the subtlety of a control like that and tend to set it too low. i know i was setting it low before reading that
in my experience the threshold for something like that should be mostly automatic/ gain adjusted for the general signal level and shouldn’t require much user input. maybe just a good bit of extra gain to push the threshold up higher than it would normally need in case of a noisy or complex signal but certainly not allow setting it lower. the setting is very coarse anyway so it seems unlikely that the average setting on it would end up being at the optimal level for a given signal of arbitrary level. really for best results the threshold should be be very tightly coupled to the input signal ideally in multiple dimensions.
like for instance spiking up during transients so their energy doesn’t get into the detection and ruin it. also the ratio of number of spectral peaks to not peaks vs the ratios of their accumulated respective gains can inform the interplay between the partials and the remnant, their gain and spread, at a given signal moment
anyway, it sounds to me like too many amplitudes are getting counted into detection and adding or fouling notes, even when i had it set high enough that at other times it was not properly triggering chords. frequency domain pitch detection is a bit of a pet passion of mine i know all about how not trivial it is to get it working reliably for all input conditions