# Adding a tutorial

<!-- generated by docs/scripts/generate_tutorial_index.py; do not edit -->

Put the notebook under `tutorials/` and give it this notebook metadata, then run `python docs/scripts/generate_tutorial_index.py` (CI does):

```json
"zoomy": {"title": "...", "summary": "<one line>", "tags": ["type:tutorial", "dimension:1d", "physics:viscous"]}
```

Tags are `category:value`; an unknown tag or a missing `type` fails the build.  A notebook may carry several values of one category (except `type`).

- `type:` tutorial / use-case (required)
- `dimension:` 1d / 2d / 3d
- `physics:` viscous / inviscid / hydrostatic / non-hydrostatic / compressible / incompressible / free-surface
- `component:` model / system-model / numerical-system-model / solver / code-generation / mesh
- `method:` finite-volume / limiting / implicit-explicit / predictor-corrector
- `feature:` automatic-differentiation / parallel / gpu / coupling / function-hook
- `backend:` numpy / jax / openfoam / amrex

`requires:` cpu / gpu / mpi / container is no filter.  It tells the test harness what the notebook needs (default `requires:cpu`): `pytest tests/tutorials -m tutorial` executes every notebook that needs only the CPU and reports the others as skipped.

The book renders every notebook that has a card, from a copy in `docs/book/tutorials/ipynb/` (git-ignored); a notebook under `tutorials/pyodide/` is the content of the JupyterLite deployment and its card links there.