data is the root of trust
A model is only as trustworthy as its data. Seal indexes datasets, fingerprints their content, and records where every split came from — so a model's lineage can be traced to actual, verifiable sources.
what a dataset seal records
- content hash + per-split hashes
- source URLs and licenses
- contributors and their $SEAL rewards
- any derived subsets (fineweb-edu → your filtered corpus)
submit a dataset and earn $SEAL when it is sealed and used by downstream models.