It makes sense.

Ride a course once, and the elevation data is pretty bad.

Ride the same course 100 times, or 1000 times, and take averages, and the data becomes more representative.

Perhaps the biggest flaw will be short dips in the road that could get lost in averages.

Or, the places where Strava has troubles resolving which road a person is on.
Here is a local segment where most of the leaderboard is filled with people going under the overpass rather than over it. So, will Strava dock that extra 50 feet of climbing?
https://www.strava.com/segments/1062935

And this one, someone else posted, with a switchback, and the leaderboard being filled by people descending rather than climbing. So, will Strava resolve the data properly?
https://www.strava.com/segments/3267169

It all depends on how much care went into their programming. But, with the errors that seem to have been made, one has to wonder.