sentier-brightway

flowchart LR
    inv[sentier-inventory] -->|fetch| sb[sentier-brightway]
    voc[sentier-vocab] -->|fetch| sb
    met[sentier-methods] -->|fetch| sb
    map[sentier-mappings] -->|fetch| sb
    sb -->|install| bw[Brightway project]
    sb -->|write| files[Plain files]

Loading

What it is

BAFU-2026 v1 life cycle inventory, the EF 3.1 biosphere and the 25 EF 3.1 impact methods, linked through the Sentier flow mappings. Ready for Brightway and Activity Browser. Data is fetched from pinned commits of the four Sentier data repos on first use. Nothing ships in the package. Install it in the Python environment where Brightway or Activity Browser lives.

flowchart LR
    f[fetch pinned data] --> b[relink BAFU flows onto EF 3.1] --> w[write project or files]

Loading

Install

uv pip install git+https://github.com/sentier-dev/sentier-brightway

Use

Command	What it does
uv run sentier-brightway db --project NAME	Installs the databases and methods into a Brightway project.
uv run sentier-brightway files --out DIR	Writes the same build as plain files (parquet + bw_processing datapackages). No bw2data needed.
uv run sentier-brightway coverage	Prints how many BAFU flows link to EF 3.1, without touching Brightway.
uv run sentier-brightway backtest --out DIR	Scores every process, compares with BAFU's own openLCA results, writes a dashboard.
Python API	import_db(project), import_files(out_dir) and backtest.run_backtest(files_dir, xlsx, out_dir) in sentier_brightway; import_bafu_db / import_bafu_files are the same functions under their original names.

Every command installs one inventory source: sentier-inventory keeps one folder per published release, data/<source>-<source_version>/ (bafu-2026-v1/ today), and tags each process row with source, source_version and sector. --source picks the folder (bafu-2026 by default), --source-version only matters when the data root serves two releases of the same source. The inventory and residual databases are named after the source, never the version, so a second source installs next to BAFU instead of over it, and a re-release replaces it in place. Process nodes carry the three tags; sector is a column, not a folder, so filter on it.

Brightway project

Object	Content
database <source> (bafu-2026)	the source's processes: 11,947 BAFU-2026 v1 processes, each carrying source, source_version and sector
database ef-3.1-biosphere	EF 3.1 elementary flows, with the source's emissions relinked onto them
database <source>-residual (bafu-2026-residual)	source flows with no EF 3.1 counterpart (kept, no factor)
methods ("sentier", "EF v3.1", <category>)	25 EF 3.1 impact categories

Link your own activities to bafu-2026 processes and ef-3.1-biosphere flows, then score with stock bw2calc. In Activity Browser, reload the project after the install; the methods sit under sentier > EF v3.1.

db works on both Brightway generations (bw2data 3.x with Activity Browser 2.x, bw2data 4.x with Activity Browser 3.x). One mix is refused before the project is opened: bw2data 4 next to bw2io < 0.9, which bw2data itself crashes on while opening a project (TypeError: '<' not supported between instances of 'str' and 'tuple'). The fix is pip install -U "bw2io>=0.9.3" in that environment.

Plain files

Path	Content
registry/	processes, biosphere, exchanges, methods, characterization factors (parquet)
mappings/	the BAFU -> EF 3.1 mapping files, verbatim
bw_package/	bw_processing datapackages: <source>/ for the inventory and methods/<slug>/ one per method
manifest.json	source and source_version, database names, data pins, citation, coverage, row counts

Every table shares an integer bw_id, which is also the matrix index of the datapackages. Stock bw2calc reads them: bw_processing.load_datapackage on bw_package/bafu-2026 plus one bw_package/methods/<slug> feeds bw2calc.LCA({bw_id: 1.0}, data_objs=[...]). sentier_brightway.datapackage.score(out, code, "ef-3.1:climate-change") does the same in one call (CH low-voltage electricity: 0.0320835 kg CO2 eq per kWh).

Backtest dashboard

uv run sentier-brightway backtest --out dashboard

Then serve the folder (for example uv run python -m http.server 8000 --directory dashboard) and open backtest_dashboard.html. --xlsx points at BAFU's results xlsx or zip; the fast extra adds the pypardiso solver.

The page opens on 25 box plots of %Δ vs BAFU's openLCA results, one per EF 3.1 category, on a symmetric-log axis: box quartiles, whiskers at 1.5 IQR, outliers as dots. A sector dropdown facets the boxes by BAFU's top-level category, and hovering a box or a dot shows a tooltip. Categories with a documented cause, such as the chromium counting difference behind human toxicity cancer, carry a ⓘ note.

Click a box (or its label) for the worst-200 list of that category, sortable and filterable by name, location and sector. Click a row in that list to open the per-process bar panel across all 25 categories; it loads vs_bafu.csv once, on demand, and caches it in memory.

backtest writes emissions.csv, vs_bafu.csv, boxes.json and worst/<short>.json alongside the page.

Options

Flag	Commands	Effect
--source ID	all	Inventory source to install (data/<source>-<version>/ in sentier-inventory); names the databases. Default bafu-2026.
--source-version V	all	Release of the source to read when the data root holds several (v1); the only release present otherwise.
--overwrite	db, files	Replace a previous install or export.
--data-root DIR	all	Read the Sentier data repos from local clones under DIR instead of downloading ($SENTIER_DATA_ROOT works too).
--skip-nomenclature	all	Leave BAFU flows whose EF counterpart carries no factor in the residual database.
--no-datapackages	files	Skip bw_package/.
--files DIR	backtest	Reuse an existing files export instead of writing one.

Data

Pinned data links 2,566 of 2,679 BAFU flows (95.8 %) and 96.8 % of biosphere exchanges to EF 3.1. The 113 unlinked flows stay in bafu-2026-residual with their exchanges intact. Pins live in src/sentier_brightway/sources.toml; uv run sentier-brightway coverage prints the full report.

Limitations (v0.1): global EF factors only, no regionalized factors such as country water use; only BAFU-2026 v1 and EF 3.1; static values, no uncertainty distributions; no flow drill-down in the backtest dashboard.

Contributing

• uv run --extra testing pytest runs the tests.
• uv run --extra dev pre-commit run --all-files runs the formatter and lint hooks.
• uv run python scripts/pin_sources.py re-pins sources.toml after the data repos change.

Licence and citation

Code: MIT. Data comes from the Sentier repositories under their own licences. BAFU data: Source: Life Cycle Inventory database of the Swiss Federal Administration, BAFU:2026. Keep this citation in any work derived from the installed data.