I am using a highly accurate, digitally programmable, calibrator with a resolution of 10uV.
I am also using a Fluke189 (5.5 digit) multimeter.
The output of the voltage calibrator has a very low impedance and therefore does not affect the Ad input even when the buffer is bypassed.
I am definitely sure of the voltage applied to the AD input.

I have set the AD in order to use either single-ended or differential input and tested both buffered and unbuffered input mode.

I made an Excel datasheet with all the results of all the tests I conducted, if you send me your mail address I will send it to you.

Besides the non linearity encountered in the single-ended mode (not in the differential), I noticed that the minimum step change is 10mV that equals to 0100H, it seens that the lower byte is not read.

Do you have any code example, maybe in 'C' to show me?

Thanks,

Diego