How we rank repos.

// Inputs · 24h window

• Star delta1.0×

  Stars gained in the last 24h. The primary signal — momentum on the canonical platform.

• Cross-source mentions0.6× / source

  Count of distinct discussions on Hacker News, Bluesky, Lobsters, dev.to, and ProductHunt. Multiple sources stack — three or more active platforms talking is the breakout signal.

• Engagement composite0.3×

  Combined score of comments, upvotes, and retweets per mention. A mention with 200 comments outranks a mention with 2.

• Momentum decayexponential

  A star burst 1h ago counts more than one 23h ago. The half-life is 6h — we are scoring acceleration, not history.

• Category baselinenormalized

  Each category (devtools, AI, frontend, infra…) has its own median. Raw scores are normalized against that median — otherwise everything in AI would look like it's trending.

// What goes into a breakout

A repo earns the BREAKOUT tag when its 24h score lands more than 3× over its 7-day rolling median and at least three distinct sources are talking about it inside the window. Both conditions matter:

• →The 3× threshold cuts noise. Normal day-to-day variance sits below 2×.
• →The 3-source rule kills single-platform amplification. One viral discussion isn't a trend.
• →Both gates close on bot spikes — our per-source anomaly detector drops them before they reach the score.

// What we exclude

• ×Star count alone — old, popular repos would dominate forever.
• ×Single-source spikes that look bot-like (per-source anomaly detection drops them before scoring).
• ×Repos archived or marked disabled on GitHub.
• ×Forks of currently-trending repos — the original gets the credit.
• ×Self-promotion loops where the maintainer is also the only mentioner.

// Refresh cadence

The trending pipeline runs every 20 minutes (UTC :07, :27, :47). Mention pipelines run on per-source schedules — each source has its own rate limits and we respect them:

// Inspect the data

Every repo's detail page (e.g. /repo/vercel/next.js) shows the raw signal breakdown — per-source mention counts, star deltas across windows, and the live trend score with every weighted input visible. The pipeline source is on GitHub:

github.com/0motionguy/starscreener →