I happened to see a link to this from Dave over at eevblog. TL;DR, it's a 5 ppm mechanical wristwatch for $800K.
That CNN says it keeps "perfect" time is laughable, but I'll give them that 5 ppm is certainly a high standard for a purely mechanical movement.
But it got me thinking... I've made all the crazy clocks I can think of... but just how accurate could a lavet stepper clock be?
Well, the going rate on eBay for a rubidium standard with 10 MHz output is around $180 or so. A simple firmware tweak should give us the same 10 Hz interrupt source: 10 MHz with a prescale of 16 is 625 kHz. 625 kHz divided by 1024 is 61 5/128, so 5 cycles of 62 and 123 cycles of 61 would be correct.
The result would be a lavet stepper clock that ticked with an accuracy of better than one part per billion. It would be no more than a tenth of a second off per century.
Of course, it wouldn't be synchronized to anything - it would depend on the user to actually point the hands to the right spot. But as long as it had power, you could count on it being a whole lot more accurate than a wristwatch 3.5 orders of magnitude more expensive.
You could do the same thing with GPS much cheaper, of course. You wouldn't even need a microcontroller. GPS modules have synchronized PPS outputs. You could use a flip-flop and two AND gates to turn the PPS output into alternating positive pulses on the two coil wires. If the PPS pulse isn't long enough, then you might need to add a pulse stretcher, but even then a microcontroller is still overkill.