Science tracks

One engine accepts TrackConfig; no food, water, or third-track pipeline is copied. The default food and water configurations request the same synthetic canopy-height and NIR-reflectance observations. Food uses a scalar encoder; water uses scalar and square. This demonstrates configuration and output independence, not a scientific food- or water-security indicator.

Each track declares question, variable requirements, modalities, spatial/temporal requirements, encoder, and scientific assumption. A track recipe references its core recipe; track observations retain parent canonical identifiers and parent provenance. Track-specific artifacts carry track identity and never overwrite core observations or another track's outputs.

Source records and shared canonical cache entries are immutable and shareable. A shared source does not imply shared interpretation. Future food rules might involve traits/productivity, and water rules might involve fluxes/soil/environment; no such scientific model is encoded until supplied and reviewed.

Add a track

1. Copy tracks/third_track/config.json to a new track folder or another JSON path.
2. Supply a scientific question, required variables/modalities, support, and explicit assumptions.
3. Define and review any nonconservative recipes and domain validation.
4. Enable the track and run prototype run --config path/to/config.json --output run/new.
5. Add configuration, scientific validation, provenance, and independence tests.

The third-track template is disabled because its scientific scope is undecided. An integration test enables a synthetic copy to demonstrate configuration-only extensibility without inventing a third scientific application. Custom domain recipes can use the independent harmonization API; the default orchestrator currently constructs only curated identity/selection recipes from declared metadata.