Not how much you use AI. How well.
The Gradient Score measures the quality of how you work with AI — judged across five dimensions, not counted in tokens. Quality over volume, and your data never leaves your machine.
Not affiliated with or endorsed by Anthropic.
sample score · The Ascent
From session to score
How you get your score, step by step.
No setup headaches, no surprises. Here is exactly what happens — and what is shared (almost nothing).
Already happening
Claude Code is already reviewing your sessions — privately.
On your own machine, Claude Code already keeps a private review of your sessions (via /insights). It contains NO prompts and NO code — only judgements like “Was the goal reached?”, “Was Claude helpful?”, and “What went wrong?”.
You choose
Open “Get your score” and pick how to start.
Either try it instantly with demo data, or drag in your own files from ~/.claude/usage-data. Your call — both work right away.
In your browser
Everything is analysed in your browser.
Nothing is uploaded behind your back. You see the exact little summary that would be shared (no text, no code) — and you confirm it before anything leaves.
Your result
Your Gradient Score appears.
Your number, your tier and your skill radar — plus concrete tips on how to get better.
On the board
Your result is placed on the leaderboard.
You see your rank right next to everyone else — pride without exposure.
The thesis
Burning tokens isn’t a skill.
Every AI leaderboard today ranks the same thing: consumption. More tokens, bigger bills, higher rank. That measures your spend — not your craft.
Viberank, CCgather and the rest rank you by tokens burned and six-figure bills.
Even their own makers call it gameable — a vanity metric you can inflate by simply spending more. It rewards the wrong instinct: throw more compute at every problem.
- Pay more, climb higher
- One number, no nuance
- Trivially gamed
Gradient measures how well you work — not how much you spend.
Precision, delegation, verification, toolcraft, efficiency. Five dimensions, judged from the shape of your sessions — not their size. A clean, one-shot result beats a thousand wasted tokens.
- Quality over volume
- Multi-dimensional, hard to fake
- Spending more buys you nothing
How it works
Three steps. Your data never leaves home.
- 01
Connect your usage
The open-source connector Gradient Pulse (`pulse`) reads your local usage data right on your machine. No connector? Upload your /insights export, or try it instantly with demo data.
Open source · runs locally · nothing uploaded yet
- 02
Everything stays on your device
Your sessions are scored locally. Only your score and a handful of aggregates ever leave — and you see an exact transparency diff of the JSON before anything is sent.
You approve the upload · field-by-field diff
- 03
Get your Gradient Score
Your score, a five-dimension skill radar, your tier on the Ascent, and the marks you’ve earned. Share the card without ever revealing what you actually worked on.
Score · radar · tier · marks
$ pulse scan
→ Reads your local sessions. Nothing leaves yet.
$ pulse score
Reads everything locally. Sends only your score.The measure
Five dimensions, not one vanity number.
One metric is trivial to game. Five independent signals, read from the shape of how you actually work, are not. This is craft made legible.
Precision
How clearly you brief a task — few clarifying loops, few misreads.
Delegation
How well you break work down and hand it to subagents and tasks.
Verification
How consistently you review and catch errors before they cost.
Toolcraft
How broadly you wield the tooling — skills, MCP, agents, search.
Efficiency
How much you achieve per iteration — little rework, no token grind.
Your shape, not your spend. The radar shows where you’re sharp — and where there’s room to climb.
Privacy
Built so you can verify it yourself.
A score about how you work is only worth having if it doesn’t cost you your privacy. Every part of Gradient is built so you can check that for yourself, instead of taking our word for it.
Local-first by default
Your sessions are read and scored on your own machine. The raw transcripts, prompts and project paths never leave it.
Open-source connector
Gradient Pulse is open source. You can read exactly what it does before you run it — no black box between you and your score.
See the JSON before upload
Before anything is sent, you review the exact payload, field by field, with a transparency diff showing what stays local and what is shared.
Only the essentials are public
The leaderboard shows a pseudonym, your score and your percentile — nothing about what you actually worked on. Pride without exposure.
measured locally · only the score is shared
Ready when you are
See where you stand on The Ascent.
It takes less than five minutes — and nothing but your score ever leaves your machine.