Presuming it needs to not significantly degrade the stability of the reference, the approach I generally take is to have extra footprints for discrete resistors in parallel with each leg of the divider. At the ovenized reference level of stability, I would say that potentiometers are generally frowned upon, though if you are using one with a trim range of perhaps 100 ppm, you could probably make a reasonable case for using it. It seems like your circuit is probably part of an autocalibration block, so you could probably just deal with deviations from nominal in software. I think the conventional wisdom on this board would also be to use something like a NOMC 8-resistor array as the divider. I don't know offhand the best way to do 6:1 with that, but I think that for statistical averaging to be most effective, something other than the obvious 6 in series : 1 would be best. Perhaps three in series for the upper leg and two in parallel for the lower leg with the resistors interleaved would be good. You probably don't want to go higher than 500 uA divider current because the Joule heating tends to worsen the stability of the divider and make it more susceptible to thermal EMFs.
Many of the op amps one would normally use for buffering a small voltage with minimal error have current noise that could be problematic, especially if you have the inputs of multiple op amps connected to that node without a buffer. If you are using an autozero amplifier, you will probably want to bypass the lower leg of the divider with maybe 10 to 100 nF (C0G) as the switched inputs have lower current noise when there is some bypass capacitance. Depending on your target bandwidth, you could add some more bypass capacitance, but at <10 Hz you usually just deal with it.
Less obviously, you need a good way of measuring that voltage if you're going to trim it without needing to hold the test leads, preferably with things as close to fully assembled as possible. Through-hole test points where you can solder magnet wires (or similar) are good. Keep them next to each other to make sure the junctions experience similar thermal gradients or just use the footprint for a 2-pin 50 mil pitch header.