LCIA methods (EF 3.x, IPCC, ReCiPe, …) for the Sentier platform — characterization-factor tables keyed by elementary flow, with normalization and weighting sets. A Data Layer repo: loadable artifacts only, no fetch/parse/calculate code.
- Fed by:
sentier-importers(PR / Push). - Read by:
sentier-platform(GET LCIA matrices). - Flow ↔ CF links resolve via
sentier-mappings.
schema/ # YAML column definitions + JSON Schema for metadata
data/ # raw parquet, one subfolder per datasource (ranked)
Each data/<NN>-<datasource>/ folder holds:
| file | rows |
|---|---|
methods.parquet |
one per (method, impact category) |
characterization-factors.parquet |
one per CF |
normalization-weighting.parquet |
optional normalization & weighting sets |
metadata.json |
datasource provenance |
The NN rank prefix orders datasources and sets resolution precedence (lower
wins). Parquet payloads are pushed by the Application Layer — the skeleton ships
only folder structure, schemas, and metadata.json stubs.
Plain-YAML table definitions (schema/method.yaml,
schema/characterization-factor.yaml, schema/normalization-weighting.yaml).
LinkML is used only in sentier-vocab; here schemas describe parquet columns
directly.
These files are the contract, not code. sentier-importers reads them to
validate and cast incoming data before delivering parquet here. This repo ships
no Python package, loader, or tests — all ingestion logic lives in the importer,
the read side in sentier-platform.
MIT — open by default, client-loadable.