Most code health scores are vibes. A number goes up, a number goes down, and nobody checks whether the files it flags are the files that actually break later. I wanted to know if the score I built does better than that, so I ran it against a defect corpus and put it head to head with the leading commercial code-health tool.
On the same 2,770 files across 9 languages, scored at the same leakage-free commit against the same defect labels, the score surfaces 2.3x the defects under a fixed review budget.
This post is how that works, and the four other layers sitting next to it in repowise.
What the score is
Every file gets a 1 to 10 score from 25 deterministic biomarkers. McCabe complexity, deep nesting, brain methods, class cohesion (LCOM4), god classes, native Rabin-Karp clone detection, untested hotspots, function-level churn, code-age volatility, ownership dispersion, change entropy, co-change scatter, prior-defect history, test-quality smells, and more.
