The rabbit hole is in front of me…..
You've already slipped in.
That's really the truth of it.
When you've invested some £700+ in a 6.5 digit bench meter simply to overcome the tempco issue of a couple of cheap 9999 counts Mestek DM91As you'd previously thought of as being very stable considering the 41 quid total invested, the best you can say is that you're now clinging onto the edge of that particular rabbit hole by your fingernails.
I've had my 3065X permanently powered up ever since I took delivery of it a week into October and similarly the cheap AD584 voltage reference I'd bought about ten days later.
The 3065X seems to suffer very little tempco since it tracked the slow downward decline of a μ7805 mounted on a temperature controlled baseplate (Rubidium oscillator reference project) without any indication of a tempco. The AD548, otoh, has only been consistent in that it's variation with temperature very closely tracks room temperature with no sign of any ageing related drift (at least over the last month of its 10 weeks run time to date).
As has been previously mentioned, absolute accuracy is less important than consistency in the face of ambient temperature variations and ageing effects, particularly true with self contained hobbyist projects. Absolute accuracy is a bonus as far as hobby activities are concerned and only becomes important when comparing results between different lab setups in the wider world.
The Chinese seller I'd bought the AD548 from was honest enough not to insult my intelligence by making any attempt at supplying a voltage calibration chart of any sort, not even the more likely accurate hand written type which I'd have still ignored anyway.
At best, it could only be as good as the calibration of the 7 1/2 digit DMM used and the care taken to create such a chart - too many unknowns to invest much faith in afaiac. Far safer simply to assume that Siglent had been able to calibrate the 3065X to within its stated specifications and allow for the tolerances built into its specified calibration accuracy.
After monitoring that very lightly loaded (20mA) temperature stabilised (36*C) μ7805 for nearly two months, I'm quite confident that the 3065X is well within its stated tempco specification. With just over 2000 hours run time clocked up (and some 1800 hours for the AD548), the worst of whatever ageing drift it had is now well and truly behind it.
Absolute accuracy in a home lab is always going to have a question mark hanging over it but the stability seems to be as good as I could possibly have hoped for out of my SDM3065X investment. Stability trumps absolute accuracy every time simply because without it, trying to recalibrate to within a tight tolerance of absolute accuracy will be a waste of both time and money.