No, because schools open at the same time all year round.

That is typical with machine learning. The more complex the model is, the more accurate it will be on average, but the less explicable and extreme will be the mistakes. A simple function will be very smooth and won't have extreme behaviors, but the incredibly complex functions that these models generate can have absurd peaks all over the input space.

Apart from having to manually set the old clocks, and having to write special tests in my code for those dates, I never really understood where the problem of the time change is. I would have much more problems with the same time all year round because it would either means waking up in the middle of the dark on winters, or having the sun shining at 4am on summers.