wp.org search vs our merit ranking: “large language models (llm)”

Each plugin's merit score multiplies these inputs together: how recently it was updated, how current its "tested up to" WordPress version is, its install base (log scale, capped at 10 million), its support thread resolution rate, and a rating score. The rating score weights the star average by review volume against a sitewide prior, then scales it by the lower bound of the share of 4 and 5 star reviews. A plugin with 3.7 stars from tens of thousands of reviews can still outscore a 4.9 with a few dozen, because the small sample carries less certainty.

A health factor between 0.4 and 1.15 then adjusts that score for what wp.org's search can't see: PHP fatals or failed activations on a test install, PHP and WordPress compatibility, memory and page-speed footprint, suspicious review patterns, closure risk, and a large install base with very few reviews. Clean, adopted plugins earn a small lift.

A keyword ranking starts from wp.org's search order for the term, weighting each position by 1 divided by its place, then applies the merit correction and health factor above. Relevance to the keyword stays in the result, so the order can differ from the all-categories list on /best, which scores plugins with no search term.

Hover a row to see the main reason for its position.