One number out of 5, built from real public data only. No invented ratings, no estimated user counts, no paid reviews.
Each signal a listing has is ranked against the same signal of the other listings in its category (or the whole store when fewer than four in the category have it). The ranks are averaged with the weights above, over the signals that listing actually has: a missing signal is left out, never guessed. The average maps to a score from 1.0 to 5.0, with one decimal.
Every listing shows each input under its score: the source, the value it reported and the day we read it. Sources are read again every few hours.
Ratings from MLX users who have the app installed (checked by MLX Node on their own Mac) will blend in as they come, weighted by how many there are. They're shown separately too.
What we never do: show a rating count we didn't read, present an estimate as a fact, write reviews ourselves or with AI, or pay for reviews. There's no paid placement, so nothing can buy a higher score. See the rule book.