io-notebook

A marimo notebook that teaches environmentally extended input-output analysis on a synthetic economy of 9 sectors: the Leontief quantity model, the carbon footprint of final demand and a scenario you change with a slider. It ends with self-checks modelled on the checks a scenario explorer runs against RaMa-Scene. It needs only numpy, so it also runs in the browser through Pyodide.

The notation follows Miller and Blair (2009), Input-Output Analysis: Foundations and Extensions, second edition: $\mathbf{A}$ for the technical coefficients, $\mathbf{L} = (\mathbf{I} - \mathbf{A})^{-1}$ for the Leontief inverse, and $q = \mathbf{e}'\mathbf{L}\mathbf{y}$ for the footprint.

Run

uv run marimo edit io_leontief.py   # edit and run
uv run marimo run io_leontief.py    # app view

Export to a static page (--mode edit lets readers change the code). The page loads Pyodide and packages from cdn.jsdelivr.net, so readers need internet access. dist/ is not committed.

uv run marimo export html-wasm io_leontief.py -o dist --mode run
python3 -m http.server --bind 127.0.0.1 8000 --directory dist

Checks

uv run pytest                     # pure functions, harness checks, and a headless run of every cell
uv run ruff check . && uv run ruff format --check .
uv run marimo check io_leontief.py
# browser smoke test against a served dist/ (install Chromium once first):
uv run --with playwright playwright install chromium
uv run --with playwright python scripts/check_wasm.py http://127.0.0.1:8000/ dist/shot.png

Deploy

The page is live at https://simonvanlierde.com/io-notebook/. Every push to main that passes the checks exports the page, smoke-tests it in Chromium and deploys it to GitHub Pages.

Each pull request gets a preview at a pr-<number> address on Cloudflare Workers, linked from the PR as its preview deployment (wrangler.jsonc). The job needs CLOUDFLARE_API_TOKEN and CLOUDFLARE_ACCOUNT_ID in the preview environment's secrets, and skips without them. Drafts get no preview. When a PR closes or turns draft, preview-retire.yml points its address at a short "preview closed" page, since wrangler cannot delete an alias.

Data and licence

The notebook uses synthetic data only. I took its recipe, a seeded numpy.random.default_rng and a dense, non-symmetric matrix, from a RaMa-Scene regression harness, which is public on my fork of RaMa-Scene. The same harness is an open pull request against the upstream repository (#72).

The code is released under the MIT licence.