[noelEOS]

Rewrites the README's opening around what has landed since 0.1.0, and corrects its DPT evidence statements.

What changes

• Opening. A "From satellite imagery to flood predictions" section: the Prithvi Sen1Floods11 model run from its released checkpoint, with the one-crop real-imagery result, the 2 × 2 region result, and the ViT-B/16 + DPT conversion. The radar water-mapping example follows it.
• Get started. A link to the Prithvi workflow, alongside the start guide added in Add a plain-language start guide and three Colab notebooks #45.
• What you can run. Shorter Prithvi and DPT rows; a sentence on the recently added raster blocks.
• Staged inference. Prithvi's cut lists and the 33 to 16 export measurement.
• Evidence and limits. A DPT bullet stating the current reference record: offline, the full-width head and logits are within 2.06e-5 of float32 PyTorch with zero argmax flips; two intermediates exceed 1e-4 against float32 (largest 1.88e-4) and are checked against float64; live checks cover reduced twins and individual blocks only. The README on main still says full width misses the 1e-4 target, which stopped being true when the DPT references were rebuilt in test(dpt): use reproducible native-scale hash references #44.
• Licence. A note that the Prithvi checkpoint is a separate third-party release under its own terms.

Every figure is quoted from docs/api.md, docs/vit-b16-dpt224.md and docs/converter.md at the base revision. No new evidence is added.

Base revision

84fd9ed (main).

Changed files

• Documentation: README.md.

Checks at 1987118

Check	Result
G1 (tools/graph_identity.py --check, with resources)	45 of 45 graphs equal: 36 fixture graphs and 9 trained graphs
G1-ext (--check-ext)	19 of 19 graphs equal
tools/block_catalogue.py --check	in sync
pytest -q tests/test_docs.py tests/test_history_citations.py tests/test_notebooks.py	36 passed
pytest -q (the offline suite)	3,738 passed, 5 skipped, 190 subtests passed
G3-offline, live plan	not applicable: no block, fixture or graph changes; no Earth Engine calls

Limitations

• Documentation only. The Prithvi result covers one labelled crop and is stated as an example, not an accuracy score.
• The DPT statements describe offline, random-weight evidence at full width; full-width live execution remains unvalidated.