Alex,
Thanks for the video,
The OLED reading is given by sampling the output via the SAMD21's 12bit ADC, but I would not expect that behavior.
What is the reading in uA on a DMM vs the OLED? Please test this with OLED connected and disconnected. When you disconnect OLED please power cycle the unit (OFF, then ON without OLED)
I am primarily interested in the output directly from the opamps to be always accurate. The software averaging for the ADC readings can always be tweaked or optimized based on findings like this or use cases which I haven't encountered. In my limited testing, I have seen instances where noise picked up by cables can cause that type of thing.
Essentially a 4uA (wrong) reading means the ADC is reading 1-2mV on from the output. That could perhaps explain this.