Alain,

truncation is a possibility that came also to my mind. I tried adding "define adcbits 10" to the software yesterday, at that time it started to work, but it could be also that re-soldering one lead of the thermocouple at the same time was the real reason. I tried to remove the define, even erase the chip and reprogram in case that the define did a change in the configuration that was not reversable by removing it, but could not return to the erraneous state of affairs.

I wonder if the reason could be in the input. I checked the output from the op. amp. with oscilloscope, and it seems to be stable within around 5%, nothing like 4-fold variation in the input values.