Looking at your schematics V_OUT_N is bouncing up and down at 400 V pk-pk which is not at all suppressed by your unbalanced filter (VO_L- and V_OUT_N are the same net). That's a bad idea from an EMC perspective, but it also means that the voltage measurement circuit needs to have very good common mode rejection to measure it the differential voltage in the presence of that signal. Does the measurement circuit rely on a floating supply to work? Capacitance across the isolation barrier can upset that sort of thing.
I think I would start by fixing the filter, it needs to change to avoid producing awful noise, but that will probably also help your measurement.
Side note, it also makes a difference which switching strategy you use. One approach puts the PWM on all the transistors, and the other just PWMs one side whilst driving the other at a leisurely 50/60 Hz.