ccrank — Claude Code leaderboard ranked by OUTPUT, not token spend

[flaskovystupel]

ccrank — the first output-based Claude Code leaderboard

Every existing Claude Code leaderboard ranks you by how many tokens you consumed. More spending = higher rank. That tells you nothing about actual developer productivity.

ccrank measures what you actually BUILD:

Dimension	Weight	Metric
Velocity	25%	Commits per active day
Efficiency	30%	Lines shipped / tokens consumed
Impact	25%	Weighted file + line changes
Consistency	20%	Active days + streak bonus

Efficiency is weighted highest — it rewards developers who get MORE done with LESS tokens.

Try it now

npx ccrank

One command. Scans your git repos + Claude Code usage data, calculates your score (0-1000), and ranks you globally. Tiers from Bronze to Legendary.

Links

• Web: https://ccrank-five.vercel.app
• GitHub: https://github.com/flaskovystupel/ccrank
• npm: https://www.npmjs.com/package/ccrank

Tech: Next.js 15, Prisma, PostgreSQL (Neon), zero-dependency CLI.

---

If you think output > spending, give it a try and share your score! Star the repo if you like it ⭐

Feedback on the scoring algorithm is welcome — what else should we measure?

[affaan-m]

Output-weighted measurement is the right instinct. Token spend by itself mostly measures burn rate, not productivity.

The tricky part is preventing the metric from rewarding churn. Lines shipped and commits per day are useful signals, but I would weight them with quality gates too: tests passing, CI status, review acceptance, issue closure, rollback rate, and maybe ratio of generated churn to retained code after a week.

The other important boundary is privacy. A global leaderboard that reads git repos plus Claude usage data should make collection, redaction, upload behavior, and opt-out very explicit.

As a local self-audit tool, this is very aligned with ECC. As a public leaderboard, the trust model matters as much as the score formula.