I have a new theory:
The encoders have a direction pin which can be used to set which direction of rotation is the positive direction. I left that pin unconnected because we didn’t need to change direction, but that was bad practice. I should have tied it to something to ensure that it couldn’t float around.
My current theory is that something in your environment is causing that pin to get pulled up sometimes causing the encoder to switch directions.
I’ve just sent you a second set of encoder boards which I’ve modified to make sure that pin is connected to GND. The weird thing is that I already sent you a first set of unmodified encoder boards yesterday (maybe two days ago), but I’m sending these modified ones express so they will get to you first. Hopefully it’s something weird in your environment and not a widespread issue 