Manifesto · v1
AI made us faster.
Is it making us better?
A year into AI coding and nobody has a good answer to that question. So we built the number we wanted. Scoreflow is a mirror for craftsmen — not a productivity dashboard.
The acceptance-rate lie
Every AI-coding dashboard tracks the wrong thing: how much. How many tokens. How many lines AI wrote. How often you accepted the suggestion. Monotones — the more you do, the higher you score.
That is Goodhart's law waiting to happen. The moment your manager looks at a dashboard like that, the team learns to game it. More AI, not better AI.
Bands, not monotones
We score every signal as a band. There is a sweet spot, and drifting either way lowers your score.
Read/edit ratio too low? You are editing blind. Too high? Analysis paralysis. Plan-mode ratio at 0%? You never think. At 80%? You never ship. Bands make farming the metric mathematically worse than practising the craft.
Craftsmen measure what matters
Read before edit. Plan before prompt. Ship PRs that still stand 30 days later. Keep your rules file alive. Use more than one tool. Teach teammates your craftables.
Those are the things a craftsman does on autopilot and a vibecoder skips. Scoreflow measures the gap.
This is not surveillance
Team aggregates by default, never individuals without opt-in and audit. k=5 anonymity floor enforced at the query layer. No prompt content stored. EU-hosted. If any of that sounds annoying, it means the privacy is working.
Tiered like craftsmen
The level ladder references the people who built the field, not the platforms you pay:
Lv 1 Karpathy → Altman → Hassabis → LeCun → Hinton → Torvalds → Knuth → Hopper → Lovelace → Lv 100 Turing.
Grinding harder will not get you higher. Practising better will.
Want your number without signing up?