I made my first SP5 with a 10 day run time in July 2021, and since the design update with helical gears it has worked reliably and accurately ever since. At one time I tried some of the longer run times, with no success. At 21 or 32 days, the clock didn't run for more than a few minutes.
I've just made a new SP5 with a 32 day run time. It is too early to tell whether it is reliable yet, but seems to be doing quite well. It ran for two days this weeks until I stopped it to do a little tuning up and to move it. We'll see how it does over the next days. The old SP5 ran on about 7 pounds, and could have worked with less. The new one is currently on 9.5 pounds. I might increase this a little to make the tick stronger.
What changed since the first SP5? Two things. First, I replaced my Prusa MK3S with a MK4. The quality of the prints is much higher on a visual inspection. In addition, the MK4 is much faster with input shaping, and so I've been printing at the recommended layer height of 0.15mm instead of the 0.2mm I used before (or even 0.3 for the frame). Second, I think I have better bearings. I have a very unscientific test where I spin a bearing on a 3mm arbor and time roughly how long it takes to stop. The bearings in this current batch keep going for 2 to 3 times ones I've used before, suggesting much lower friction. This time round I bought ones with an ABEC rating of 3. High ABEC ratings mean more precise tolerances, and most cheap bearing are ABEC 1 (if the rating is given at all). They cost only slightly more. The price goes up dramatically at ABEC 5 and above.
Here's a couple of pictures (sorry about the cluttered background).
The two clocks together. They are not yet quite tuned to the same period, so they pendulums go in and out of sync. This might drive me mad. (The old one is an hour slow because I set it wrong earlier today.)
Great looking clocks. I especially like the color changing weight shells.
You have done several things really well to get the long runtimes. ABEC 3 or better is a good idea. I have had good luck with generic bearings from AliExpress, but rated bearings should be even better.
I have started wondering about layer heights causing surface roughness. One clock would not run properly with 0.2mm layers for the gears. It ran great with the escapement re-printed at 0.12mm. My best guess was that the layer lines were locking into each other. Changing the height of one gear made it run perfectly. I want to do an experiment using prime number layer heights to see if there is some interaction. For example, print all the odd number gears with 0.17mm layer heights and the even number gears with 0.19mm. The 400 step motors on the MK4 should also be an improvement.