Skip to content
pubscout

How scoring works

The health score (0 to 100) is a weighted mix of five parts. Every package page shows each part and the numbers behind it, so you can disagree with the score and still use the data.

Caps

Some problems outweigh good numbers. A discontinued package scores at most 20, an archived repository at most 30, and a package with no release in two years at most 50.

Missing data

When a part can't be measured, for example a package without a public GitHub repository, it's left out and the other parts are re-weighted. The breakdown says “no data” for it.

Categories and summaries

Categories come from three layers: rules on the package's metadata, an AI pass over its README, and human review, which always wins. Summaries are written by AI from the package's own README, changelog and pubspec, are labeled as AI-written on every page, and are refreshed when a package makes a breaking release.

Alternatives and “works with”

Alternatives are packages that do the same job and could replace each other, like dio and http. “Works with” lists packages used together: ones a package depends on, ones built on top of it, and packages from the same family, like firebase_core and firebase_auth. A package never appears in both lists.

Which packages are included

pub.dev's 2,000 most-used packages, plus a hand-picked set of essentials (theming, maps, code generation and more) that stay whatever their rank, plus any package someone requests. Everything is refreshed every night. Packages that drop out of the top list for a month are removed so their numbers don't go stale.

Explore packages