yes that temp is huge. my original error takes several hours to get. so this test case is a much smaller pdb and mtz file.
The small reproducer doesn't help in this case, although it is a good idea in general. Simulated Annealing inherently carries the risk of "exploding". We use tricks to achieve stability in most cases. This would have to be revisited and we'd need the actual problem case to be sure the additional tricks help. However, I doubt it will make it to the top of the priority list anytime soon. Practically the best response is to use a lower start temperature or to apply some quick manual fixes before the SA run. Another thing you could try is interleaved_minimization.number_of_iterations=1 which will make the trajectories more stable. Ralf