Community-maintained patches for the BAFU life cycle assessment database (UVEK LCI data), distributed as EcoSpold 1 XML.
The pipeline reads the raw release, makes it schema-compliant, applies community patches, and writes EcoSpold 1 XML again, optionally with additional fields.
uv sync
uv run bafu-fixes list
uv run bafu-fixes run latest output/ --strict --report output/report.htmllatest downloads the latest BAFU release
(https://files.brightway.dev/bafu-2026-v1-ecospold.zip, about 37 MB). It's downloaded only
once: it's checked against a known SHA-256 checksum and kept in your user cache folder (on
macOS, ~/Library/Caches/bafu-community-fixes). You can also pass a local folder or file. A
full --strict run with a report takes about a minute.
Or use the terminal app, which needs no command-line knowledge. It uses the latest release by default, or a folder you pick in its file browser. Choose the output format and options, tick the patches to apply, and press Run. It shows each step with a progress bar, and opens the report in your web browser at the end. That last part can be switched off.
uv run bafu-fixes-tuiOptions for run:
--strict: also check every document against the official EcoSpold 1 XSD after import, and every changed dataset after each patch--only ID [ID ...]/--skip ID [ID ...]: choose which patches to apply--no-extensions: leave out our additional fields, giving plain EcoSpold 1--report FILE.html: write a run report (see below)--placeholder-year,--placeholder-number,--placeholder-country,--placeholder-text: values filled in where required data is missing (defaults1111,1111,AQ,UNKNOWN). Blank values are kept, since the schema allows them. Placeholders are validated before the run: the text is also used for company codes, so it can be at most 7 characters. The terminal app has the same settings.-v: log progress
- Parse the XML and strip the EcoSpold namespace (the raw release has none anyway).
- Import fixes (
import_fixes.py) make the raw release readable. They work on the raw lxml tree, always run, and run in the order they are defined. Runbafu-fixes listto see them. - Build models (
models.py): strict pydantic models that mirror the XSD. Anything that doesn't fit stops the run with a list of every problem: unknown attributes or elements, invalid values, broken references. With--strict, the official XSD is checked as well, as an independent second check. - Patches (
patches/) run one at a time, in order of their id. After each patch, every dataset is revalidated, and any error is reported against that patch. The pipeline records the patch id on every dataset it changed. - Write EcoSpold 1 XML, in schema element order, with canonical number and date formatting.
The models are strict. Values must already have the right Python type, unknown fields are rejected, and assignments are validated as they happen:
exchange.mean_value = "1.5" # ValidationError: not a float
exchange.name = "x" * 300 # ValidationError: longer than 255 characters
exchange.uncertainty_type = 1 # ValidationError: use UncertaintyType.LOGNORMAL
person.country_code = "XX" # ValidationError: not an ISO code in the XSD--report writes a single, self-contained HTML file, with the Sentier branding and nothing
external to load. It's meant to be shared with data users. It shows:
- which import fixes and patches ran, in order, and who wrote and authorised each one
- a prominent warning for every step that changed data without recorded authorisation
- per step: datasets and values changed, an example from the data showing the XML before and after the step, the most common before → after changes for each field, and every individual change for the first 10 datasets
- the schema check results, an input fingerprint (SHA-256), options, and software versions
If a run fails, the report is still written, with the error.
Our own data lives in <bcf:extensions> (namespace urn:bafu-community-fixes:extensions:1)
at the end of each dataset. The EcoSpold 1 schema allows foreign elements there, so the
output stays valid. Extensions record:
- which import fixes and patches changed each dataset
- the original BAFU number of each exchange that got a dummy number (
originalNumber)
BAFU exchange numbers are identifiers: the reference product has the dataset's own number, a technosphere input almost always (99%) has the number of the dataset that supplies it, and an elementary flow has a number for its substance. They aren't unique within a dataset, which EcoSpold 1 requires, mostly because one substance emitted to several subcompartments shares a number.
So numbers are kept wherever possible (about 73% of exchanges). Where several exchanges
share a number, the reference product keeps it, or else the first one. The others get a
dummy number, 1,000,000,000 + original number × 1,000 + occurrence. For example,
1000631002 is the second exchange numbered 631. Any number of 1,000,000,000 or more is
a dummy and links to nothing.
In Python they are dataset.ext and exchange.ext. --no-extensions leaves them out.
Add a module to src/bafu_community_fixes/patches/; it is discovered automatically.
from bafu_community_fixes import Database, data_file, patch
from bafu_community_fixes.models import UncertaintyType
@patch(
"0011-fix-something",
authored_by="Jane Doe", # who wrote this patch
authorised_by="Chris Mutel", # who approved it for the release; leave out until then
reported_by="John Roe, Example AG", # optional: who found the problem
confirmed_by="BAFU (email, 1 November 2026)", # optional: the data owner's confirmation
references=("https://github.com/sentier-dev/sentier-bafu/issues/1",),
)
def fix_something(db: Database) -> None:
"""One-line description, shown by `bafu-fixes list` and in the report.
The rest of the docstring explains why the change is needed. It appears in the
run report, so write it for data users.
"""
for dataset in db:
for exchange in dataset.exchanges:
if exchange.name == "Sulfur dioxide" and exchange.uncertainty_type is None:
exchange.uncertainty_type = UncertaintyType.LOGNORMAL- Field names are the EcoSpold names in snake_case (
meanValuebecomesmean_value).Datasetalso has shortcuts:name,location,reference_function,exchanges.db.exchanges()yields(dataset, exchange)pairs. - Put supporting data in
src/bafu_community_fixes/data/, prefixed with the patch id, and read it withdata_file(name). - Don't record your changes yourself. The pipeline compares every dataset before and after your patch and reports each changed value.
- Add a test.
tests/conftest.pyprovides an imported fixture database.
| Patch | What it does |
|---|---|
| 0001 | Corrects polyethylene disposal for three strawberry datasets (confirmed by BAFU) |
| 0002 | Uses the supplier's location for 25 exchanges (ENTSO → ENTSO-E, two CH; confirmed by BAFU) |
| 0003 | Flags the 1,289 xx/xxx datasets as not maintained (extension attribute) |
| 0004 | Adds missing CAS numbers to 28 elementary flows (Brightway Flows) |
| 0005 | Replaces CAS numbers naming a different substance, on 17 elementary flows (Brightway Flows) |
| 0006 | Removes CAS numbers naming a different substance, from 3 elementary flows (Brightway Flows) |
| 0007 | Removes units from 3 elementary flow names (Brightway Flows) |
| 0008 | Moves Carbon dioxide, in air to the in air resource compartment (Brightway Flows) |
| 0009 | Writes elementary flows' CAS numbers zero-padded, and on every exchange of a flow |
| 0010 | Links every elementary exchange to its Brightway Flows flow (extension attributes) |
| 0011 | Marks 40 aggregated datasets (eco-profiles, iridium, cabinet panel) as system terminated (type=2) |
Patches 0004–0008 change BAFU's own flow attributes and don't depend on 0010, which
only adds links. The Brightway Flows data in src/bafu_community_fixes/data/ is used
with its author's permission (see the README there); rebuild it from public sources
with uv run python scripts/build_brightway_flows_data.py.
uv run pytest
BAFU_ECOSPOLD_DIR=path/to/bafu-2026-v1 uv run pytest # also checks the full release
uv run ruff check && uv run ruff formatThis repository is also the public Sentier review workspace for BAFU/UVEK. It connects
partner evidence, Brightcon hackathon work, candidate improvements and the discussion
around the data, and returns actionable recommendations to BAFU. Its findings, records
and site came from sentier-dev/sentier-bafu; the pipeline above came from
sentier-dev/bafu-community-fixes. Both Git histories are kept.
The intended order of work is: the provider release, then this pipeline's import fixes and selected patches, then a Sentier import of the patched output. That is not yet how the import page reads: Import & use still describes the Sentier import of the raw release, the baseline, until Sentier inventory consumes the patched output (community-fixes #2). The findings and contributions are reconciled with the patches in the reconciliation receipt.
Import & use · Findings ledger · Reconciliation · Evidence & testing · Contribute · Consensus
This is a candidate integration (#19), to be confirmed with the community maintainers. Community-fixes is authoritative where the two projects conflict; its package, commands and XML extension namespace are unchanged. The file-disposition receipt accounts for every file of both sources. See the plan, the consolidation issues and the architecture decisions. The original workspace README can be read at a13bbbf; verbatim copies of both original READMEs are kept as history in docs/migration-history.
| Step | What belongs here |
|---|---|
| Canonical import | Source release, identifiers, hashes and reproducible import path |
| Partner contributions | Findings, proposed mappings, models and scenarios with evidence |
| Application & testing | Separate candidates, named exchange changes, coverage and score comparisons |
| Reviewed consensus | Actual reviewer decisions, objections and their disposition |
| Recommendations to BAFU | Agreed findings, delivery evidence and provider responses |
A finding starts with the source release, dataset code, what you measured and how. Read the ledger, then open an issue. Reproduce observations against the provider's source files before concluding that BAFU's data is wrong. An importer defect belongs to the component that introduced it.
A contribution layer gets a record, isolated candidate build and validation plan. A reconstruction of an aggregated inventory is a new inventory claim; representation parity, inventory correction and forecasting require different tests.
The contribution records and site have their own checks, independent of the package environment above. They run on Python 3.12 outside the package project:
uv run --no-project --python 3.12 python scripts/community.py check
uv run --no-project --python 3.12 python -m unittest discover -s community/tests
uv run --no-project --python 3.12 python scripts/build_checks.py community/templates/build-audit.json
uv run --no-project --python 3.12 --with pandas==2.2.3 --with pyarrow==19.0.1 python -m unittest discover -s registry-testsThe synthetic audit example is a format demonstration, not a measured BAFU build. The earlier Sentier import route, current work and measured audits are described in the original workspace README.
Open the BAFU / UVEK site. The
Documentation workflow checks pull requests and publishes
updates from main to GitHub Pages, separately from the package CI.
Build locally with uv run --no-project --python 3.12 --with markdown==3.7 scripts/build_site.py; see
site maintenance. Generated pages contain coordination material only,
with no inventory downloads.
Source datasets retain their own terms, attribution and modification requirements. Keep restricted exports and confidential partner evidence local. No new reviewer consensus, sent recommendation or provider endorsement is implied by the workspace. See the workflow, recommendation tracker and terms. Licences of the code and bundled files are listed below.
The code and original documentation are MIT (LICENSE). Bundled files keep their own terms:
- The EcoSpold 1 XSD files in
src/bafu_community_fixes/schemas/ecospold1/are under the EcoSpold Public License 1.0 (see their file headers and README). - The Brightway Flows data in
src/bafu_community_fixes/data/is under the Open Database License 1.0, not MIT (see its README). - The Rubik font embedded in run reports is under the SIL Open Font License 1.1 (Rubik-OFL.txt), as is the site's Nunito font (OFL.txt).
- The Départ de Sentier logo (
site/assets/dds-logo.svg) is by Sylvain Prevost, under CC BY 4.0 (see site maintenance).