It's not 15% of the ballots randomly distributed across the state.

It's typically that some places have reported all or nearly all of their votes, some haven't reported anything yet. We know how those places have voted in past elections and how their demographics are (or aren't) similar to other places in the state.

Have enough of those, and you have a pretty good picture of how the state is voting. And if they're running ahead of the numbers they were expected to need to win in those various places....odds are that's going to happen in the unknowns as well.

Example:

Imagine Portsmouth has voted D+20 as an average of the past couple elections, and Dover has voted D+15.

If you have 100% of the vote in for Portsmouth, and it's only D+10, Dover is probably D+5, and it more generally suggests Dems are doing far worse than usual. If the overall race was supposed to be D+2, statewide you might be looking at R+3.

(in reality, you would want a bunch of different kinds of places to make these sorts of inferences for the state overall).

This is also how you can get calls while the actual counted votes are the still opposite of who they're calling it for. - because they're way underperforming where they need to rack up votes, even if they're ahead in the total count at that time.

It's pretty rare for major media outlets to miscall races. Not non-existent, but uncommon.