I put together a precision PWM DAC and automated a test with 0.5V steps from -9.5V to +9.5V.
DMM6500 set to 1NPLC, high impedance input, autozero enabled.
Per step; 80 seconds to allow the PWM DAC output to settle, approximately 1 minute of samples acquired and averaged to give a single value for the chart.
Linearity chart is attached.
I'm mostly certain that the majority of the error shown in the chart originates from the DMM6500 rather than the PWM DAC.
(The error in a PWM DAC, due to switch resistance mismatch [0.71 ohms in my case], is a smooth inverse x^2 curve with max error around the midpoint which I have corrected in the DAC firmware, not at all visible as even the slightest x^2 curve trend in the chart so I'm comfortable that this source of error is eliminated)
It's comparable to the typical factory test data spreadsheet given earlier in this thread and it's well within tolerance, which is pleasing.