Stream-First Data

ScienceClaw favors stream-first environmental workflows:

Mode	Meaning	When to use
stream	Read chunks, windows, or metadata directly from a remote source	Default for STAC, COG, Zarr, and Parquet
cache	Copy only the needed subset into /workspace/cache	Useful when repeated analysis needs the same small subset
download	Explicitly copy a dataset into /workspace or /external_storage	Use only after human review and provenance notes

Avoid downloading giant source datasets into the container. Prefer STAC search, COG window reads, Zarr chunk access, Parquet filtering, and derived outputs.

Good public dashboard patterns:

1. Commit small public data under docs/assets/data/.
2. Reference public external data by URL.
3. Stream public STAC, COG, Zarr, or Parquet endpoints.
4. Publish static HTML, PNG, CSV, and lightweight JSON outputs.
5. Summarize large private data with metadata, plots, and links rather than committing it.