The flaw in your post is that deterministic does not mean showing everyone the same thing. You can deterministically decide how much evidence to gather about underexposed articles as well, which is what I advocated

Though to be fair, a random number generator would be used to determine which users to expose the article too (just not how many of them).

Even if you fail to come up with a deterministic algorithm, you'll probably come up with a better random one simply by coming up with better probability distribution to draw from.