Claude comments often contain the whole iterative chain of decisions that led to the current state.

Useful for the LLM to know the "why", but not something a human would do, unless it's a very critical and confusing part of the code.