Keyboard shortcuts

Press ← or → to navigate between chapters

Press S or / to search in the book

Press ? to show this help

Press Esc to hide this help

World source manifests

World schema v18 replaces the old name/URL/license labels with a canonical, typed manifest for every distribution used by the offline compiler. Each entry has a stable ID, release status, canonical URL and optional DOI, typed licence, operational notices, access method, spatial and temporal coverage, preparation recipe, and content identity. Human-readable Markdown remains alongside these fields for the in-game audit trail.

The compiler sorts entries by stable ID and rejects duplicates, non-canonical order, empty or oversized fields, missing notices, malformed hashes, report claims without their source entry, and unknown serde fields. There is no Unknown identity. A source is either reproducible from a raw/prepared SHA-256 or curated revision digest, or explicitly records an unpinned or release-blocked state. The latter states never report themselves as reproducible.

Identity and cache boundary

The manifest digest is BLAKE3 over the world schema version, inference-rules version, world year, complete SpatialGridSpec, and sorted distribution manifests (including notices and preparation). Changing any field changes the digest. Source ordering, local paths, and file timestamps do not. The complete compiled artifact is still hashed separately; both IDs are retained by the SpacetimeDB import session. No separate world-data cache currently exists, so this digest defines the key that a future cache must use.

The import session also retains complete deterministic audit Markdown for every used distribution: ID, name, release, licence, content status and identity, URL, DOI, notices, access, spatial and temporal coverage, preparation, and notes. Each typed source is embedded as canonical compact JSON. The compiler rejects the import before calling a reducer if the fully JSON-quoted argument would exceed the 24,000-character Windows transport budget; legal text is never silently truncated.

The stored digest is compiler/operator-attested under the trusted first-caller import model. SpacetimeDB syntax-checks the digest and binds it to the import session, but it does not receive the typed manifest and therefore cannot recompute the digest. It is an audit/cache identity, not independent proof against a malicious or noncanonical client.

Operational notices and unresolved boundaries

  • Viabundus: retain attribution and CC BY-SA 4.0. The project applies a conservative BY-SA treatment to its generated map and terrain contributions, keeps those artifacts outside the software AGPL, and carries the complete source-specific terms in MAP_DATA_LICENSE.md. This policy is specific to those reviewed packs and does not authorize an unrelated combined output.
  • Copernicus DEM: retain the prescribed Copernicus/WorldDEM production credit and European Commission/ESA no-liability notice.
  • HYDE 3.5: retain HYDE attribution, the CC BY 3.0 licence link, and an indication that Adventure Simulator interpolates and classifies the source.
  • Copernicus forest and EU-Hydro: credit the European Union/Copernicus, identify project modifications, and do not imply endorsement.
  • Jung PNV, SoilGrids, and EGDI: retain CC BY 4.0 attribution and identify gameplay conversion changes. EGDI additionally retains the Maltese contribution and disclaimer.
  • EU-Trees4F: retain dataset and publication citation despite CC0.
  • IEG religion: source images remain rights-reserved and are not redistributed. Only the coarse curated intermediate is checked in.
  • NOAA OWDA: cite NCEI and Cook et al.; compiled output remains bounded per-settlement derived data, not the grid or annual series.

Current release blockers are explicit: GLO-30 tile selection, the manually acquired HYDE 3.5 files, EGDI, and EU-Hydro lack checked complete content inventories. The forest marker pins only the preparation format, so forest remains non-reproducible until every consumed raster is inventoried and hashed. SoilGrids is a rolling service whose strict prepared manifest supplies the actual retrieval timestamp and snapshot identity; that does not make raw latest reacquisition reproducible. These entries do not claim legal resolution or reproducibility that is not present.

scripts/world_source_init.py provides bounded plan/init/verify workflows for these accepted sources. Only the immutable EU-Trees4F archive is acquired; the authenticated or incompletely pinned sources fail closed until reviewed inventories are committed. The curated IEG CSV is validated in place and its rights-reserved reference images are never mirrored. Sidecars use sorted fixed metadata and actual sizes/hashes, reject unknown inventory fields, traversal, symlinks, redirects outside fixed hosts, oversized content, and partial or checksum-failing publication.

The EU-Trees4F initializer pins the exact JRC ENS_CLIM archive. It retains the EU-Trees4F v2 Figshare citation and CC0 notice, but does not claim that a Figshare-hosted archive is byte-identical; confirming that relationship remains an explicit provenance blocker.

Source-separated developer bundles

scripts/world_data_bundle.py provides a distinct distribution boundary for development inputs. Its ZIP is a manifest-and-notices collection: source components remain separately addressed and are installed into the compiler's existing source paths only after complete verification. It carries no compiled world artifact. The policy includes every active input but rejects LUH1, rights-reserved IEG map images, and raw OWDA grid/annual data; IEG's committed coarse CSV stays in the repository and OWDA may appear only as a bounded per-settlement derived profile. This engineering policy is fail-closed and does not itself resolve source-specific redistribution terms or the licence status of a future combined world-data release. See wiki/reference/world-data-bundles.md.