I'm reverting back to using single digit 7 segment modules instead of dual digit ones. I simply can't find any .56" modules that match the look of the .3" module (for the 10th-sec digit) as well, and the ones that are closest don't cost less.

Meanwhile, I have one more experiment to try - I'm getting a version of the board made with a 3.2x2.5 mm crystal. If you do the math, the 180 µs PPS latency of the current design could be reduced to 120 µs by adding a 12 MHz crystal, so I want to see if that works. I'm not entirely convinced that I'm going to actually add the crystal, though. I'm not entirely sure that the resulting improvement is significant enough to bother with. If it were a 16 MHz crystal, that would get us under 100 µs, but when running the ATTiny841 at 3.3 volts the maximum safe speed is 12 MHz.

EDIT:

I got that board built, and it works, and it does indeed have 120 µs instead of 180 µs. But, again, I'm not going to go out of my way to rev the boards in the store for it.

I could try to go down the road of attempting to predict the PPS timing - that is, effectively using an FLL with the PPS interrupt as the timing source for the entire clock. That would likely replace the fixed latency with some indeterminate amount of jitter (but likely at least an order of magnitude down). But the trouble is that I think these things and then have to remember that this is a clock for humans with a granularity of 100 ms - being accurate 3 orders of magnitude down from that is surely sufficient.

All that said, I am working up to getting a batch of the clock boards manufactured - they're popular enough now to warrant that level of investment.