I am comparing designs centered around the MAX6126-25 and the TL431B (0.4% grade). I have an HP 3457 that is in cal from Agilent and a NI DAQ unit that is more useful at seeing the noise (unshielded) on the output (which is +/- 600 uVs or so) and logging the warm-up time to a (+/- 1 mV) stable output. I also have a multi-channel Keithley 740 thermometer scanner so I know the ambient at which readings are being made. What is considered the break-over point of precision with low uncertainty versus cost to a group like this? Is there a rule-of-thumb for the cost of each additional digit of precision after N digits? I mean these as simple, don't get into all of the uncertainty contributions and general traps of metrology, questions. If I sell someone a reference that I've ascertained is 2.50163v @70.3 F with a calculated uncertainty, is it valuable as a 0.1% reference even though the error may be much less, like +/- 0.08%? Thanks for your time and feedback. Russ