What the score is made of
The score has two halves. Quality asks whether this is real, maintained work by more than one person over time, read from the GitHub repository: a year of pushes, other humans committing, issues being filed. Popularity asks whether the people who find it actually use it, read from the owner's Google Analytics. The verdict is two words, one per axis: below the cut on quality is Slop, above is Craft; below the cut on popularity is Fringe, above is Popular. Nothing connected is Unknown, not zero. Connected and empty is zero.
This page is generated from the code that computes the score, so it is always current. Every rule is a plain function anyone can read in api/src/heuristics. No LLMs, no paid data, no stars.
Quality
heuristics v0.6.1Quality: is this real, maintained work by more than one person over time?
Public GitHub repository signals that are expensive to fake. Stars and forks are not used. Sites are scored on the repo they link; no repo is zero.
historyA year of pushes. The share of the last 52 weeks with at least one commit, times the age of the repo: nothing under one month, 40% at six months, full credit only at twelve months of history.+60contributorsOther humans have put real work in: people (not bots) with at least two commits, not counting the author. One person is zero.+25issuesIssues exist at all: the total number of issues ever filed, pull requests excluded. 3 is a start, 100 is full credit. Whether they were closed or who filed them is not quality.+15
Three things, all of which take time; weights sum to 100. Nothing that can be scaffolded in an afternoon counts: no README, license, CI, tests, stars or forks. A repo born today with one author scores 0, and full credit on history needs twelve months of pushes. No repo linked is unknown, not zero. Archived repos are capped at 20. Read with the owner's own GitHub login, so private repos work and nothing is shared.
Popularity
v0.5.0Popularity: do the people who find it actually use it?
Read from the owner's Google Analytics 4 property, connected with Google sign-in, for the last 30 days: active users, and how many of them Google classes as returning.
popularity = reach × (0.5 + 0.5 × returning). Reach gates everything: nobody there means zero, however loyal.
| 0.0% | 0 pts |
| 10.0% | 30 pts |
| 25.0% | 60 pts |
| 40.0% | 85 pts |
| 60.0% | 100 pts |
| 100 visitors | 0 pts |
| 1,000 visitors | 30 pts |
| 10,000 visitors | 60 pts |
| 100,000 visitors | 85 pts |
| 1,000,000 visitors | 100 pts |
No analytics connected is unknown, not zero. Connected with no visitors is zero: a thing nobody uses is slop, however well built. The Google connection is read-only and stored encrypted. Other analytics products are an open contribution.
Verdict
Two words, one per axis. Quality below the cut is Slop, at or above is Craft. Popularity below the cut is Fringe, at or above is Popular. Nothing connected is Unknown, not zero. Connected and empty (no history, no visitors) is zero.
Craft at 50 and above, Popular at 50 and above.
The overall score is Popularity and Quality averaged, 0 to 100, higher is better. It is unknown if either half is unknown.
- Slop quality below the cut
- Craft quality at or above
- Fringe popularity below the cut
- Popular popularity at or above
- Unknown nothing connected
Grades do not go stale: popularity is re-read every day and quality on the first of every month, from the owner's connected accounts. The history on each entry page shows the drift.
Terms version in force: 2026-10-22. Grades keep the heuristics version they were computed with; re-grading uses the current rules.