Interestring thread. I just thought I'd add my 2 cents.
I"m new to home grown retrofitting and manually programmed CNC, but I have many years experience as an engineer and have done a lot of machining for prototypes of all sorts. My comment is that I have to agree with the folks who opined that this is really an academic discussion as the minute error at issue here is insignificant for almost any real world application.
The 'accuracy' of a gear from a 'timing' perspective is related only to the number of teeth. Any inaccuracy in individual tooth spacing would only effect that specific tiny portion of the trace and be completely gone by the time the gear moved to the next tooth. i.e. 32 teeth is 32 teeth and that is how many teeth will pass per revolution, no matter if a tooth or two are slightly off of their true theoretical azimuth or slightly over or under thickness.
Manufactured gears have a tolerance. Many are broached in a single pass. Like bearings, there are many different accuracies available with corresponding cost. Gear tooth contour accuracy effects mostly the amount of sound (harmonics) produced and wear characteristics. No cut gears are really "accurate". The only applications that I know of requiring super accurate gears call for hardened and ground teeth and are for super precision, super smooth (quiet) or very high rpm service, none of which would apply to a clock, I would imagine.
Even in the most accurately produced gears, ultimately the 'final machining' is done by the gear mesh itself during break-in.
So, as has already been stated, use absolute angular positioning to eliminate a tolerance stack, rely on the accuracy of your setup (which from where I'm sitting looks more than adequate), machine and enjoy!