ruler-and-vibes
A zero-infrastructure personal model-chooser kit — run models yourself, judge the outputs, keep your own results.
Every rubric combines objective checks (the ruler) with judged criteria (the vibes). The bank spans 40 rubric categories and 759 test forms, usually three parallel forms per facet, organized as a nested breadth ladder: Core (34) ⊂ Extended (123) ⊂ Full (759). Plain Markdown, JSON, JavaScript, and Node built-ins — no package manager, framework, CDN, API, hosted service, or build step.
Why it matters
Public leaderboards answer someone else's question with someone else's data. Ruler & Vibes is evaluation you run yourself: fresh-session runners follow RUN.md, a separate evaluator session follows JUDGE.md, and results stay yours to keep or publish. Core is the execution default; Full must always be named explicitly.
Repo facts
Public, MIT licensed. 40 categories, 759 test forms, three parallel forms per facet, eight-domain taxonomy. Zero-build static report bundle. Explicitly not an official public leaderboard — a personal model-chooser kit.