Seems fine to me, though I assume some others have more opinions about tacking stuff onto the sys module. I think it's fine.

I think the biggest win here is really unifying the logic in one place.

Does not it need documentation?

Would not it be more useful to expose build_prefix (None or not set if is not Python ) rather of simple boolean _is_python_build?

Does not it need documentation?

As proposed, I see it as an internal implementation detail, so wouldn't be documented.

Would not it be more useful to expose build_prefix (None or not set if is not Python ) rather of simple boolean _is_python_build?

+1. is_python_build() could still return bool(sys._build_prefix).

Is sys._is_python_build always a correct indicator at runtime?

Nope, but we don't have a proper one. Any marker file could disappear or be copied, so we may as well look for the things we actually care about (stdlib modules) and guess, rather than look for something that's there solely to help us guess.

At least if we expose it properly, it'll be easier to change it to something more reliable if we come up with one.

Does not it need documentation?

It's missing a news entry, but other than that, no, it's an internal change.

Would not it be more useful to expose build_prefix (None or not set if is not Python ) rather of simple boolean _is_python_build?

We can do that. The reason I didn't do it here was because I didn't want to add any more public API, so I just used the already exported _is_build_prefix. I opened #99273 following this suggestion.