im going cheap, as usual ...
a while ago i did 16 plastic LT1021, paralleling together over some balancing resistors, with adjustment and inside a thermostat.
As the thermostat itself are good, all this construction became a bit better than disaster, an output swing at non predicted level, randomly going up and down.
Im goin to the next loop of cheap
so i pick random used LT1021-10 PDIP8 , add basic china multiturn resistor, 1N4148 as compensation, from +20C to +40C , i got negative swing of 4 microvolts. basically lucky. tossed in plastic container, add 2 5W resistors, container filled by oil , everything in a thermos. basic 12V regulator
it run it some period on March, (let say 500 hours) ; and on april I sporadically did a measurement , just to see a tendency .. ; measurement device 34401a , with positive tempco, and room temp usually +18, and was +20 during mid April heatwave.
im guesin - 4 (or 6) such modules , adjusted to min tempco , a bit running a couple month ,then adjusted to the same output , (avoiding balancing resistors) probably make sense.
attached file of sporadic measurement, and guessing Temp inside of oil bottle as thermistor not calibrated. 34401a I think +0.8PPM