Recently, I faced with one issue where two seemingly similar solutions performed differently. I want to know why it is happening?

Code 1: Used an XOR operation (num ^ some_randint) to preprocess numbers before counting their frequencies

Code 2: Counted numbers directly without any transformation.

Both codec followed identical logic after this step, but code 1 passed the test cases efficiently, while code 2 gotaTLE.

WHY? I would be happy if someone could explain.