everyone.

Just to close the thread... unfortunately with no useful answer.

I aborted the simulation in progress after 16 hours and 21 minutes, absolutely fed up. It was on about 50% of the simulation (about 49000 out of 98000). Then, I added some tracking of the duration (coarse, counting seconds) of both code blocks (list generation from files, and CNN simulation), and re-run the same "49000" simulations as the aborted execution. Surprisingly, it took "only" 14 hours and 34 minutes, with regular durations of every code block. That is, all the list generations took about the same time, and so the CNN simulations. So, no apparent degradation showed.

Then, I added, at the end of the main loop, a "list".clear() of all lists generated, and repeated the "49000" simulations of the CNN. Again, the duration of both blocks was the same in all iterations, and the overall simulation time was 14 hours and 23 minutes, just a few shorter than without the list clearing.

So, I guess that there is no problem with my code after all. Probably, the slowdown that I experienced could be due to any kind of interference by the OS (Windows 11; perhaps any update or "internal operation"?) or the anti-virus. Well, I'll never know, because I'm not going to lose more time repeating such slow experiment. I'll just go on with my test campaign, trying not to desperate (Zzzzzz).

Anyway, I want to thank you all your interest and your comments. As I'm evolving to "pythonic", I'll try to incorporate your tricks. Thanks!