tcaf-model

True Cost Accounting of Food (TCAF) build of the Transition Compass model.

It is a separate build of transition-compass-model, not a newer version of it. It keeps the food side (agriculture, crop, livestock, land use, dietary habits, alcoholic beverages, population, true cost) and drops the other sectors. Region: Switzerland.

Deployed at https://transition-compass-tcaf.epfl.ch/.

Branches

tcaf-import is the research work exactly as it was written in 2050Calculators/tcaf-calc, branch model, replayed commit by commit with the original authors and dates. That fork was taken from the web app repository before the app and the model were split, so the model code sat in backend/model/. Here it sits in tcaf_model/model/, the same place as in the main model repo, and nothing else changed.

tcaf adds what the code needs to be a real python package: the __init__.py files, pyproject.toml, the lint setup and the release workflow. The science is untouched.

The package is called tcaf_model, not transition_compass_model, on purpose. The web app installs both, and picks one with MODEL_PROFILE. Two packages with the same import name could not live in the same environment.

See issue #11 for the background.

Install

make install          # uv sync --all-groups + pre-commit

Use it

from tcaf_model.model.interactions import runner
from tcaf_model.model.common.auxiliary_functions import (
    filter_country_and_load_data_from_pickles,
)

DM_input = filter_country_and_load_data_from_pickles(
    country_list=["Switzerland"], modules_list=["population", "agriculture"]
)
output, kpi = runner(lever_setting, years_setting, DM_input, ["agriculture"], logger)

Each module also runs on its own, on the pickles in _database/:

python -m tcaf_model.model.crop_module

Release

The package is on PyPI as tcaf-model. A tcaf-v* tag on branch tcaf publishes it (tags are a human decision):

git tag tcaf-v1.0.0 && git push origin tcaf-v1.0.0

The workflow (.github/workflows/publish-tcaf-pypi.yml) lints, builds the wheel with the pickles inside and publishes it to PyPI. Then it opens a bump PR in leure-speed-to-zero that pins the new tcaf-model version.

Working here

• Data files (*.pickle, *.pdf) are in git LFS. Run git lfs pull if you get small text files instead of data.
• make format before committing, CI runs ruff and codespell.
• Keep the module names importable: no dash in a .py file name.