Let's talk over more pressing matters, software is like.. it works as is, and quick enough? Good, moving on.
I wonder if the analog parts of MicroChip aren't inherently 'bad'. As you mentoined, ADC's and DAC's in micro's are not as good as an external one. I don't know on what design they are making the external converters, but if it's the same as the micro's, you're still back to where we were.
They also make like 22-bit ADC's with a claim of 18+ ENOB. I have no idea whether they will meet this specification. I have some of these lying around (MCP3551), but I really wonder whether these are spot on as well. Maybe they have a terrible offset of +100 counts if both inputs tied to GND hard.. not sure.
I'd resort to another manufacturer, probably, for analog parts from seeing this. Well, they meet their specifications, but +/- 12 LSB is quite horrible (0.3% right there). Am I right in saying that the worst case situation, the output voltages could be off by 2-3% (voltage reference +/- 0.25%, FSR output +/-1%, linearity +/-0.3%, unwanted offsets)? Kinda a pity of all the precision resistors and current sense amplifiers, etc. I would probably pick a more precise DAC from AD or TI.
Ah well, the given resolution was way better than required anyway
Furthermore, if you really want a 'DC accurate' measurement, I think calibration is inevitable. A better resolution is ofcourse nicer because you can more precisely vary steps (only for that reason I'd smack in a overkill ADC or DAC for my hobby power supply which is probably a one or twice off).