Data we store

Rubricary's storage splits along a simple line: prompt content lives in git, everything else in Postgres.

• Prompt content. prompt.md, evals.yaml and the license file for every prompt live in a real git repository with full commit history — one repo per prompt. Locally that's the filesystem; in production it's object storage (Cloudflare R2).
• Everything else — prompts' metadata, versions, scores, organizations, and accounts — lives in Postgres.
• Account data is name, email address, and which OAuth provider you signed in with. Nothing else is collected unless you type it into your profile yourself, and it stays editable or removable from Settings.
• The validator's submission log keeps the score, which model and provider produced it, token count, cost, whether it was answered from cache, and — if you were signed in — your account id. It does not keep the prompt text; see What the validator does with your text.
• Eval runs of published prompts store each case's response, the verdict, the judge's probability, the judge version and the run's cost. Producing them sends the prompt and case inputs to OpenAI, and the inputs and responses through OpenRouter to TypeSafe — see Methodology.

Full detail on retention and your options — export, deletion, what a public prompt makes visible — is in the Privacy Policy; the usage limits are also in the Terms of Use.