# Licensing of generated assets

## Recommendation

For the project's generated assets — SVGs, rendered images, annotations,
critique records and dataset records — we recommend:

> **CC0 1.0 Universal (Public Domain Dedication)**

Rationale:

- The project's purpose is a public, reproducible benchmark: maximal
  reuse with zero friction.
- Attribution is recorded in provenance metadata and on the site, but CC0
  makes downstream use (research, derivatives, redistribution) unambiguous.
- It matches the permissive spirit of the MIT license used for code.

### Alternative

If attribution is preferred, **CC BY 4.0** is the recommended alternative.
It is one click harder for reusers, which slightly reduces the benchmark's
usability.

## Important caveats

- **Legal uncertainty is real.** Copyright status of AI-generated content
  varies by jurisdiction (e.g. US Copyright Office guidance vs. UK/EU
  positions), and our outputs may incorporate model-training influences we
  cannot fully trace. We make no warranty of enforceability, and we claim no
  rights in any pre-existing artwork, font, logo or recognizable design that
  may appear in an output.
- **Do not include third-party assets.** Never commit images, fonts, logos or
  vector art copied from other sources into the dataset unless their license
  explicitly permits it and it is recorded.
- **Reference images** generated with third-party image services are treated
  separately: their publication and reuse terms are checked and documented
  before any publication, and they may remain internal-only.

## Application

- Code and documentation in this repository: MIT.
- Generated assets: CC0 1.0 (recommended), recorded per-asset in metadata.
- Third-party quotes (e.g. the benchmark prompt itself): attributed, not
  re-licensed.