It's R419, and its purpose is quite obvious.. it draws -5.6mA out of GND, exactly compensating all currents into GND from the reference circuit.
OK, you got me! Thanks for the correction-- yes, when I zoom in on my very low resolution copy of the schematic, it does say R419, and not R418. Oops!
Well, this is one of the theories that I have heard as to it's purpose, but I have never seen any experimental results proving such a claim. If this is true (and R419 is indeed necessary), then the value would have to be adjusted if we choose a better op-amp that might have a different quiescent current than the LT1013.
Experimental result?
Instead of measuring the currents, you can do a simple calculation beforehand, using typical or even maximum current values:
Zener current ~ 0.40V (@90°C) /111Ohm = 3.6mA
collector currents ~ 2x(7.2V (typ.) -0.4V) / 74.25k = 183µA
divider current ~ 7.2V /16k =450µA
LT1013 supply = 2 x 0.5mA (max. value) = 1mA
Sum~5.2mA versus 15V/2.67kOhm = 5.6mA
Evidently, these currents nearly cancel, leaving the analogue ground supply 'A', pin J400 (4) with some 100µA, depending on sample variation. I can't overlook, where on the main analogue board, or on the A/D board the Ref GND is married with analogue GND, but very probably, it's not buffered, in contrast to the +7V Ref signal.
On the A/D board, at least there's R168 (?), a 10k2 resistor, sinking -1.2mA out of Ref GND (it's labelled ZR_LO2 there).
And on the analogue board, the potential of Ref_GND can be measured by the DCV input multiplexer also..
In the resistor reference current circuit, you'll also find a similar compensation resistor, that's R311 (9k09) which sinks 2mA from analogue ground, balancing the reference currents through resistors RP300A .. C..
Back to the LTZ circuit... a current of 5.6mA over GND would create a potential difference of about 100µV over only 20mOhm, or on the order of 15ppm.. So if the connector would change its resistance by that amount, that would create a well measurable error.
I bet, that removing this 2k67 resistor would shift the calibration remarkably. (Would be the experimental proof, that I won't do for obvious reasons)
Fluke uses active current cancellation circuits in their 732B, and as far as I remember, in their reference circuits for their calibrators also. The purpose is the same, very obviously.
By the way, my 3458A showed a clearly visible -0.5ppm shift, after returning it from a volt-nuts meeting two years ago, that's been 2x200km ride by car, with some vibration, although I got a very comfortable French car, with smooth suspension.
I directly thougt that the connector of the reference had changed its contact resistance towards smaller values.. and I also thought, that sending these instruments to hp for calibration, that it would always give some shift by the transport itself.
Frank