A close up of the fracture.

Hah, yes, I'd hate to have to try to cut some of those gears. The gear ratios are actually slightly changed from Strutt's design to replace some 6-tooth pinions (he used lanterns) with 8 or 10 tooth ones. There is also a clever trick called Ferguson's Mechanical Paradox which means there is no direct gearing down from the minutes to the hours.

One other change I forgot to mention. The clutch slipped at first so the minute hand didn't go round. I made the small spacers thicker with a slight improvement. The change which really helped was to roughen up the gear 5 spacer and the surfaces it comes into contact with using coarse sandpaper, to add a bit more friction.

For design from scratch, I use Fusion 360. For modifying STLs, I find it is often easier in Blender. In Blender I mostly work by moving vertices around in Edit Mode, or by using Boolean modifiers in Object Mode. Both take a long time to learn but there are plenty of resources. I've also heard recommendations for MeshMixer for modifying STLs.

Thanks for the reply! I have been using the gear generator in Fusion 360, which allows you to specify the backlash as a parameter, and based on observation of the fit, I was thinking that adding an extra gullet might be a good idea. Your test jig looks like a good idea.

I switched to the Revo 6 a few months back and it is sooooo easy to change the nozzle. I think I had only done it twice in all the year prior to that as it was such a pain with having to hold the heater block in place. Now it's just unload the filament, unscrew the old one, screw in the new one, ready to go.

There is an interesting followup to this from Thomas Sanladerer, where he suggests that the Arachne algorithm enables you to get the same precision with a 0.6mm nozzle that you would previously have needed a 0.4mm nozzle for, with a reduction in print times of typically 30%. I've done a few tests (I have a Revo 6), and my experience so far matches his.

I am trying this fix now. I built the 10-day version last August and get a consistent pattern of about 4-5 weeks of reliable operation before I saw the pendulum swing getting less (on average, as there is always some variation), before it stalled. Restarting might get a few more hours or a couple of days further running. Each time I disassembled the clock and fiddled with it in various ways, only to have the pattern repeat after 4-5 weeks. In February I switched the escapement to ceramic bearings, and it ran until June before stalling. After a restart it ran fine until the start of July since when I have had two stalls. I've now replaced the two gears with the new variants and we'll see how it does.

Yes, that looks very clever. (And Clayton is a super nice guy, who would probably chip in with advice if asked.)

Thanks, that's good to know! I did some test prints today (with a lot of switching between the CAD program and the slicer, just as you say), and the results look OK, but I have yet to try them in a train. I'll try modifying them the deepen the dedendums if they don't work well.

Have you seen this: https://www.thingiverse.com/thing:4828789?