Most A/B tests are experiments on features, not on their users.

There’s a difference in finding out what features users likes and let’s see if we can manipulate their feelings or get them depressed.

The one Facebook did is not really A/B feature testing.

Those who opt-in to tests can have biased results, it’s like asking those running Windows 11 if they like Windows 11 more than Windows 10.

Maybe their caching doesn’t work for private subreddits

A/B testing is cheaper than hiring real testers