sentier-inventory

What it is

Standardized life cycle inventory (LCI) data for the Sentier platform. Processes and their exchanges, stored as parquet. A Data Layer repo: artifacts only.

• Fed by sentier-importers, which opens PRs with validated parquet.
• Read by sentier-brightway, which loads one source folder into one Brightway database.
• Cross-source flow ids resolve via sentier-mappings.
• No Python package, no loader. The only code is the CI validator.

Install

Nothing to install. Clone the repo and read the parquet files directly.

Use

Validate all source folders against the schema contracts. This is what CI runs on every PR.

uv run --with "jsonschema[format],pyyaml,pyarrow" python scripts/validate.py

Inspect a delivered parquet payload.

uv run --with pandas,pyarrow python -c "import pandas as pd; print(pd.read_parquet('data/bafu-2026-v1/processes.parquet'))"

Layout

schema/                  # the data contract, read by sentier-importers
data/<source>-<version>/ # one folder per published release of one datasource
scripts/validate.py      # CI validator, self-contained
.github/workflows/ci.yml # runs scripts/validate.py on every PR and branch push

Data

Inventory is organized by source. One folder per published release of one datasource, named <source>-<source_version> (bafu-2026-v1). A delivery replaces its source folder whole; sources never share a file. Every process row carries source, source_version and sector, so a consumer picks a folder by its metadata.json and filters rows by sector. See data/README.md for the id convention.

Each data/<source>-<source_version>/ folder holds:

file	rows
processes.parquet	one per process
exchanges.parquet	one per exchange (technosphere and biosphere edges)
metadata.json	source, source_version, title, publisher, license, citation, schema_version, sectors, row_counts

• Sector ids are lower-kebab enum values in schema/common.yaml: agriculture, electricity, chemicals, ...
• Background link targets such as ecoinvent are not sources here. They resolve through sentier-mappings.
• Parquet is committed directly. No git-LFS, no release artifacts.

Schema

Plain YAML. Each file lists the columns of one parquet table. LinkML is used only in sentier-vocab.

file	describes
schema/common.yaml	shared enums: flow_direction, flow_type, process_type, sector
schema/process.yaml	processes.parquet, primary key process_id
schema/exchange.yaml	exchanges.parquet, joins on process_id
schema/metadata.schema.json	JSON Schema for each source folder's metadata.json

CI checks columns, types, enums, row_counts, process_id uniqueness, the exchange to process foreign key, one source/source_version per folder matching its name, and sectors against the column.

Contributing

Data changes arrive as PRs from sentier-importers, not by hand. Schema changes go through a PR here. CI must pass.

License

MIT — open by default, client-loadable.