How scores are calculated

The five signals

For each approved listing in a category, we gather five signals, normalize them, weight them, and sum them into a single score on a 0–10 scale.

1. 1. Community votes

   rolling 30 days

   Count of votes cast for this listing in its category over the last 30 days. One vote per person per day (deduped by a daily voter hash). Votes from authenticated GitHub users and anonymous visitors both count.

2. 2. GitHub stars

   refreshed hourly

   Pulled directly from the GitHub API for any listing that has a GitHub repo linked. Sites without a repo are excluded from the stars calculation entirely — they're not penalized for missing it, and they don't drag the normalization floor down for repos that do have one.

3. 3. Traffic

   rolling 30 days

   Real outbound clicks from this site to the listing's URL, summed over the last 30 days. No external traffic-estimation API is involved — it's measured from our own audience, which is exactly the audience the ranking is for.

4. 4. Recency

   half-life 60 days

   Exponential decay from the last GitHub release (falling back to the last push to the default branch). A fresh release scores 1.0, a 60-day-old release scores 0.5, a 4-month-old release scores 0.25, a year-old release is effectively zero. Only applies to listings with a GitHub repo.

5. 5. Editorial score

   0–10 manual

   A capped 0–10 rating set by our editors to reflect quality signals the other four metrics can't capture — documentation quality, maintainership, real-world reliability. Default is 5/10 for every new listing; we only move it when there's a clear reason to.

How they combine

Inside each category, the five signals are min-max normalized to a 0–1 range, so the formula compares listings against their peers rather than across unrelated categories. Each category has its own weights — stars matter more for dev tools, votes matter more for vibe-coded projects.

score =
    w_votes     · normalize(votes_30d)
  + w_stars     · normalize(github_stars)   // github sites only
  + w_traffic   · normalize(outbound_30d)
  + w_recency   · 0.5 ^ (days_since_release / 60)  // github sites only
  + w_editorial · normalize(editorial_0_10)

// rescaled by the sum of weights that applied,
// then multiplied by 10 for display

The final score is rescaled by the sum of weights that actually applied to that listing, so a listing with no GitHub repo competes on three signals instead of five without being silently demoted. The result is always on a consistent 0–10 scale.

What doesn't move the ranking

• — Paid sponsor slots. Sponsored listings are surfaced above the ranking in a separate, clearly labeled section; they don't alter any listing's score.
• — Submission recency. A brand-new submission starts with the same neutral editorial score as everything else; it wins or loses on the other four signals like any listing.
• — Click-throughs from paid ads, referrals from spammers. If a domain is spiking inorganically, we review it manually.

Refresh cadence

Signals pull hourly. Rankings recompute every hour and on every new vote. All data is rolling — a listing that stops shipping, losing traffic, or collecting votes will slide down; a newcomer that does the opposite will climb.