skills/geo-deep-learning/references/authoritative-sources.md

# Authoritative sources

- Last verified: 2026-07-19
- Review cadence: every 3 months
- Refresh triggers: TorchGeo or PyTorch major release; pretrained-weight or dataset deprecation

## Canonical sources

- [TorchGeo stable documentation](https://docs.torchgeo.org/en/stable/) — CRS-aware datasets, samplers, transforms, and models.
- [PyTorch reproducibility notes](https://docs.pytorch.org/docs/stable/notes/randomness.html) — deterministic behavior and reproducibility limits.
- [PyTorch model saving guidance](https://docs.pytorch.org/tutorials/beginner/saving_loading_models.html) — portable model artifact handling.

Capture library and weight versions, dataset revisions, seeds, split geometry, and preprocessing parameters. Revalidate recipes after framework or weight changes.