Best

Default of SLOT_EXPLORE. It is the lower bound of the Wilson confidence interval.

Formula

p     = ups / n,   n = ups + downs
lower = (p + z²/2n − z·√(p(1−p)/n + z²/4n²)) / (1 + z²/n)

with ups = weightedLikes / 100, downs = weightedDislikes / 100, and z = 1.281551565545 — that is 80% confidence, the exact number Reddit uses, not 95% as many write-ups still copy.

Reading the formula

It asks "how sure are we", not "what's the average".

• 2 likes, 0 dislikes looks perfect but is nearly meaningless — the sample is too small.
• 900 likes, 50 dislikes has a worse ratio but is far better evidence.

A plain average would rank the first above; Wilson does not. That is how a new post with one upvote does not jump to the top just because no one has objected yet.

The implementation here is verified to match Reddit's results within 0.5% error.