Brightcon 2026, hackathon and courses

Agribalyse is one of the most widely used public food LCA databases, but reproducing its SimaPro-based results in Brightway is not a simple CSV import. The source data arrive with SimaPro elementary-flow conventions, adapted EF 3.1 characterization factors, background datasets copied or renamed across ecoinvent versions, unit drift, and waste-treatment modelling conventions. This creates a practical collaboration challenge: the data and tools are open, but the translation layer that makes results comparable is often hidden in notebooks, local patches, or expert memory.

This presentation proposes a structured plan for making that translation layer explicit, reproducible, and reviewable:

1. Build a mapping registry. Every biosphere match, technosphere match, unit conversion, deletion, edge-label correction, and unmatchable decision is compiled into source-hashed parquet files with priority tiers and provenance. Inputs include randonneur packages, Sentier harmonised flows, curated workbooks, and reviewed residual matches.

2. Run the full Agribalyse to Brightway pipeline. The D-D-S adapter imports the ADEME Agribalyse 3.2 SimaPro export, links it against ecoinvent 3.9.1 cutoff, applies transformations for aggregated ecoinvent copies, edge labels, production exchanges, waste treatment records, and unresolved product links, then produces a square Brightway scoring matrix.

3. Register and augment EF v3.1 methods. For flows with clear ecoinvent equivalents, the pipeline uses the ecoinvent biosphere. For flows without such equivalents, it installs an EF-native biosphere subset so that Brightway can score Agribalyse-specific flows against JRC EF v3.1 characterization factors without collapsing them onto less precise substitutes.

4. Validate against SimaPro reference results. The current run links 100% of technosphere exchanges and 99.98% of biosphere exchanges before scoring, builds 19 headline EF v3.1 methods, and compares 2,451 mapped food products against ADEME's published product table.

5. Turn disagreement into diagnosis. Score differences are not treated only as failures. For most methods, median differences are within a few percent; remaining divergences reveal structural differences between SimaPro's adapted EF 3.1 method and the JRC/Brightway stack, including sub-compartment resolution, USEtox bridge gaps, biogenic methane naming, and water-use CF placement.

The session will combine architecture walkthrough and live demonstration: registry build, linking report, score calculation, and backtest dashboard. It will close with a reusable collaboration pattern for future database migrations: publish mapping datapackages, keep residuals reviewable, benchmark against known references, and make every difference between tools explainable instead of opaque.

Short Version

Agribalyse 3.2 is public, but reproducing its SimaPro-based LCIA scores in Brightway requires a visible translation plan: source parsing, mapping, transformation, LCIA registration, scoring, and validation.

We present the D-D-S Agribalyse to Brightway pipeline as a structured migration workflow. First, all mapping decisions are compiled into a source-hashed parquet registry. Second, the ADEME SimaPro export is imported, transformed, and linked against ecoinvent 3.9.1 cutoff. Third, EF v3.1 methods are registered across both ecoinvent biosphere flows and an EF-native biosphere subset. Fourth, Brightway scores are backtested against ADEME's published SimaPro product table.

The current run links 100% of technosphere exchanges and 99.98% of biosphere exchanges, builds 19 EF v3.1 methods, and compares 2,451 mapped food products. Remaining differences become diagnostics for sub-compartment handling, USEtox bridge gaps, biogenic methane naming, and water-use CF placement. The session combines architecture walkthrough, live demo, and a reusable plan for collaborative database migration.