GitScore methodology

GitScore is a single 0–100 number that ranks every gitlawb repo by six independent signals. Built so a distorted dimension (fake stars, batched commits) can't dominate the result.

Why six components

Star counts alone are easy to fake. Commit counts alone reward churn over substance. A score is only useful if it's hard to game with a single dimension. GitScore combines:

Component	Weight	Signal
Activity	30%	Commits in the last 90 days (log, saturates ~30)
Freshness	20%	Days since last push (linear decay, 0 at 90d)
Longevity	15%	Repo age (log, saturates ~2 years)
Contributors	15%	Distinct commit authors (log, saturates ~10)
Popularity	10%	Stars (log, saturates ~100)
Originality	10%	1.0 if original, 0.3 if fork

Final score = weighted average × 100, rounded. Missing components are skipped and the remaining weights are renormalized — a repo with no commit data can still get a Lite score from longevity + popularity + originality + freshness alone.

Formulas

Each component returns a value in [0, 1]. The math is pure — no DB lookups, no external state, no randomness.

// Repo age, log-saturating at ~2 years (730d).
longevity = clamp01(log1p(ageDays) / log1p(730))

// Stars on a log scale, saturating at ~100.
popularity = clamp01(log1p(stars) / log1p(100))

// Originals score full; forks discounted.
originality = isFork ? 0.3 : 1.0

// Linear decay from 1 (today) to 0 (90 days since last push).
freshness = clamp01(1 - daysSincePush / 90)

// Recent commits, log-saturating at ~30 per 90d window.
activity = clamp01(log1p(commitsLast90d) / log1p(30))

// Distinct authors, log-saturating at ~10.
contributors = clamp01(log1p(distinctAuthors) / log1p(10))

Tier colors

The web app and CLI both color the score by tier:

Score	Color	Tier
≥80	emerald	high
60–79	lime	active
40–59	amber	medium
`<40`	zinc	low / new

Lite vs Full

What's open-sourced in the @gitdex-labs/gitscore CLI is the Lite version: commits-only activity, no pull-count blending, no ownerTrust, no daily snapshot history. The full version that powers gitdex.app adds:

Pull-count blending in the activity score (mixes commits and merged PRs for a less commit-spammable signal)
Daily ScoreSnapshot so we can compute "Rising 7d" — repos whose GitScore climbed the most in the last week
ownerTrust placeholder (weight 0, reserved for future agent-trust integration)

The formulas above describe the Lite version. The Full version uses the same six components with the same weights, just enriched.

How to raise your repo's GitScore

Push more often in the last 90 days (raises activity and freshness)
Welcome contributors — distinct commit-author count drives the contributors component
Don't fork unless you mean it — originality is the cheapest way to gain 10 points
Time helps — longevity ramps over the first 2 years and then plateaus

What doesn't move the needle:

Star-farming — saturates at 100 stars; beyond that, popularity adds 0
Single-day commit dumps — distinct authors and a 90-day window dilute one-time bursts
Renaming or forking — fork status is detected from the repo metadata

Where the data comes from

All inputs come from the gitlawb network API: /api/v1/repos for metadata, /api/v1/repos/{owner}/{name}/commits for commit history. The CLI talks to node.gitlawb.com directly; the web app indexes the same data daily into Postgres and snapshots scores for trend analysis.