HOW GITFIND SCORES

The Early Signal Score is a single 0–100 rating of how much a repo is moving right now — built so a builder can trust it without trusting us. Here is all of it: the signals, the weights, the tiers, and the rules we don’t break.

§ 1 — the signals

fig. 01

fig. 01 — seven weighted signals, example values

star growth25%████████████░░░░░░░░61

New stars per week, and how that compares to the repo’s own history.

active builders20%████████░░░░░░░░░░░░42

Contributors committing in the last 30 days. A team, not a solo push.

community buzz15%████████████░░░░░░░░58

Mentions across Hacker News, Reddit, and GitHub discussions.

commit pace10%██████████████░░░░░░70

Commits landing in the last 30 days. Alive repos breathe.

fork activity10%███████░░░░░░░░░░░░░35

New forks — people building on the work, not just watching it.

star momentum10%█████████░░░░░░░░░░░44

Whether this week’s star growth is accelerating or cooling off.

fork momentum10%██████░░░░░░░░░░░░░░30

Same idea, for forks. Acceleration beats raw totals.

raw signals are normalized against the repo’s own history and against every other repo we track, then blended into one score. a repo doesn’t compete with the biggest projects on GitHub — it competes with itself.

§ 2 — the tiers

fig. 02

fig. 02 — three tiers, no mystery

Breakout70 – 100Strong early signal — breaking out. The repo is compounding on every axis at once.

Hot40 – 69Gaining traction — heating up. Real momentum, still building.

Active0 – 39On the radar — early signal detected. Worth watching, not yet moving.

§ 3 — the manipulation filter

fig. 03

Star counts can be bought, and we assume some are. When a repo’s star pattern looks inorganic — for example, near-identical weekly gains two weeks running, a pattern genuine growth almost never produces — a manipulation penalty is deducted from the score and shown openly on the project page. When you see it, you’re watching the filter work. Stars alone are never the story: engagement, commits, and community conversation are far harder to fake.

§ 4 — data & cadence

fig. 04

sources

· GitHub API — stars, forks, contributors, commits, licenses, daily snapshots
· public GitHub event data (BigQuery) — ai-written commit volume
· Hacker News, Reddit, GitHub discussions — community mentions
· package registries — npm and PyPI weekly downloads, where linked

cadence

· pipeline runs nightly — new snapshots, new scores
· pages refresh hourly — nothing you read is more than a day old
· every score is stamped with its measurement date on the project page

§ 5 — the rules we don’t break

fig. 05

01every delta carries its timeframe — “+508%” means nothing without “this week”

02zero is a state, not a verdict — invisible to a metric is not the same as dead

03no fabricated precision — if a tool can’t be measured honestly, we say so on the page

04the score shows its work — every number links back here

form 27-b — subscription

THE TUESDAY BRIEFING

The repos that moved this week, why they matter, and what to watch next. One email. No noise.