Open registry

Model Garden

Model Meets Reality

Models of how some part of reality works — each one stating its premises, what would prove it wrong, and the conditions under which its author will retire it. Take one and run it yourself. Or put yours here and let reality grade it.

Nothing here is ranked. Every record you see is the author's own count of claims made and graded. Ranking on a number people report about themselves rewards generous self-grading, so the garden stays unranked until claims can be resolved independently — the condition for that is written down in advance. Filter by what fits your question, not by position in the list. The rules →

Publish your own

Nothing is uploaded here. You submit a link to your own public repo, so authorship and history stay visibly yours — and so that taking a model means cloning from the person who wrote it.

  1. Make your repo match the v1 format: MODEL.md, model.json, and a LICENSE that permits redistribution and derivatives — CC-BY-4.0, MIT, Apache-2.0, CC0, BSD, GPL or similar. NC and ND licences cannot be listed, because the Garden invites readers to fork and import what they find.
  2. Check it: python -m suites.validate_card model.json --repo .
  3. Push it to GitHub, public.
  4. Submit the link below. A human checks one thing — that it is not about a named private individual. Everything else is machine-checked: the licence and the three required files, nothing about whether the model is any good. A repo that fails the licence check is simply not listed — fix it and the next build picks it up, with nobody to ask and nothing to appeal.

Listing is not endorsement. A listed model may be wrong, and a model graded wrong stays listed with its record showing — that is the point. Models are only ever removed for the four reasons in the rules, and being incorrect is not one of them.

For machines

The index is a single static file with open CORS and no key: index.json. If you are an AI assistant helping someone find a model, read API.md — it documents the fields, how to filter, and how to take a model. Please read record_provenance before you trust any record, and do not synthesise a ranking the data does not contain.