Finished up my repeatable squaring/positioning with series-wired motors:
I used some Avery labels to print 8" long adhesive rulers and little short segments with a Vernier-style gradations (similar to belt stretch approach before) so in theory I can eyeball position to less than a mm.
I also used my mirror “infinite hallway” approach to ensure that the zero position is parallel to the side rails to within one click of the motors.
I think this would be very slightly better than holding against endstops when energizing because there is no guarantee that the motor quantization will not be off by one cog between the two motors, depending what phase (in the cycle of four full steps) they happen to be in when energized. And for the same reason, position non-repeatability could be in the range of 0.4 mm.
From a more practical standpoint, holding against stops is inconvenient (just slightly) because I have to hold the center assembly with one hand and then select and run the ‘wake-up’ gcode file with the other (basically G92 X0 Y0 followed by G1 X0.1 Y0.1). So I find doing a proper G28 X Y and then a few seconds of forcing it square is nicer, nevermind the accuracy.