Yeah that board looks nice. Something like that is what I expect from a reference for sale.

To be clear, this is not a 10V reference, nor is it trimmed. It is however, well built.

He's using the REF102 untrimmed, and some OPA2141's for gaining the 10V output to +/-14 Volts

The OPA2141 has max Vos=3mV. This is used on the +14V output, and there is another one that simply inverts that for the -14V, which adds another 3mV offset. From the datasheet, OPA2141's are 0.14ppm/C @ 14V output, that's not bad.

The REF102 is 0.1%, which is 10mV. The first OPA2141 is gaining up the Ref voltage and the Ref error and the opamp Vos of 3 mv by 1.39. That's 1 + 0.39 because it's a non-inverting op-amp. The 0.39 part is due to the divider resistors, and if using 0.1% divider resistors on the opamp, then the gain error could be ((4k7*1.01)/12k) - (4k7/12k) = 0.39%. So the 0.39 part of the gain equation could be off by .39%.

So the max gain will be 1 + (0.39 * 1.0039) =1.392

(10 + 10mv + 3mv ) x 1.392 = 10.013 mV * 1.392 = 13.94V.

Op-amp input bias is in the pA range, so let's ignore it.

Finally, for the -14V output, add the 3 mV Vos of the second OPA2141 in inverting unity gain, for a maximum output of -13.97V.

final error is |(13.97 - 14)| / 14 = 0.21%

So there you have a +/- 14V DC reference, which you can only say is 0.21% absolute accuracy. Let's be conservative and say 0.25%

It's still not bad though, for a +/-14V reference. Too bad the way they designed it, the +14V does not track the -14V exactly.