Maybe not. 
First, as you say, if this happened, it would have happened to someone else long ago.
I wasn't calling setNumber() at all, but I was pounding on refreshDisplay(), expecting that it would either not matter, or be handled by a timer throttle of some kind in the routine itself, too soon to step the multiplexing along kinda thing.
Placing a healthy delay() on the loop() that does nothing but scan lets you see the multiplexing happen, one step per call, so either all segments of one digit at a time or one segment of all digits at a time. Calling refresh very often shouldn't matter too much, I think, until one was doing ridiculously too fast, faster than the loop can go?
But. I was using the wokwi simulator. It has rarely been the problem, but here I believe it is.
The wokwi's strong point is not hardware. A LED connected between ground and an input pin with its internal pull-up enabled will glow as brightly as the same LED with no current limiting series resistor on an output pin set to HIGH…
So, without wasting more of anyone's time, I will guess at some point in the scan of the segments the lines are set such that a tiny amount of current will flow in segment 'A' of the first (or is it the last) digit, for the full interval between on call to refresh and the next.
@DianeBMaker had the wrong resistor placement (or setting) or was also using the simulator.
The defect disappeared in the simulator by switching the scan to digit by digit.
It occurs to just now that with a longer delay, I can pause the wokwi at the point where the 'A' is glowing and see what all the lines are doing. I will have to do that. Not just now.
THX for the help, this was some kinda fun.
a7