Data Availability
A hash is not useful if the underlying data cannot be retrieved.
Provenance
Historical Spawn correctly separated sourced data from engineering/model choices. v2 should preserve:
- dataset name/origin;
- license;
- hash/manifest;
- citations;
- sourced-vs-designed table;
- code/model license;
- explicit “what this is not” claims.
This makes claims attributable and stable. It does not prove that a paper supports the creator's interpretation or that the model is scientifically accurate.
Large datasets
A World can reference terabytes of source data, but the raw bytes should remain off-chain.
Suggested pipeline:
raw archive → preprocessing → derived runtime dataset → active working set → simulation.
The launch record commits to the exact versions/manifests used.
Availability
A verifier must be able to retrieve all data necessary to replay the disputed horizon. Therefore data required for verification must remain available through at least:
Season duration + challenge window + settlement finality.
Hashes without retrievable bytes are inadequate.
For a public replay claim, the manifest must resolve to retrievable content through a declared repository, content-addressed network, archive provider, or equivalent availability system. The location, retention period, access requirements, and redundancy policy must be disclosed. A hash by itself is only an identifier.
Cost
Storage, transfer, preprocessing, active memory, compute, and verification are separate resource categories and should be metered separately.