Home · Methodology
What we measureThe rubric is fixed, versioned, and visible to your whole team — the same bar for everyone. Trivial changes are filtered out before they cost you anything. We never rank on lines of code.
Two things are scored on every meaningful change: the twelve dimensions below, and the business value the work delivered. Training and paste-ready prompts sit beside those scores — they teach the next change, they are not a third career metric.
Every raw model response is stored. The trail is developer → rollup → analysis row → quoted diff. Methodology changes are a new version of the rubric, not a silent re-grade of history you can no longer explain.
Lockfiles, typos, and three-line no-ops never hit the model. You pay for judgment, not noise.
The rubric measures engineering judgment — correctness, testing, error handling, design — in whatever language or framework the repo already uses.
Quality is not the same as impact. We score both, so a careful cleanup and a high-priority delivery are not forced onto one number.