Project status — food-processing gate-to-gate LCI replacements

Eaternity AG · version v1.2.2 · built from commit 096606b · downloads & quick start

ZHAW FCM (Claudio Beretta) and any LCA practitioner replacing an ecoinvent 3.3 / Agribalyse v1.2 food-processing background.

What this is

BAFU/UVEK Open Government Data (Swiss Federal Administration, BAFU 2026) + Eaternity tributary LCIs (food-processing, fruit, dairy, livestock, ...); NO ecoinvent flow data, NO biosphere3, NO EDB.
Method: EF 3.1 (BAFU) 16-category impact set; plus GLAM (UNEP Life Cycle Initiative Global LCIA Method) BAFU-companion methods, Ecological Scarcity 2021 UBP, and IPCC 2021 GWP100 — all BAFU-biosphere bound.
Functional unit: 1 kg processed product, factory gate (gate-to-gate processing).

Scope: Gate-to-gate PROCESSING ONLY. Upstream agriculture is a SEPARATE layer. Per-node scope_marker tells you which double-counting rule applies: 'gate_to_gate' = substrate left unbound (add your own farm-gate input); 'cradle_to_gate_proxy' = upstream baked in (do NOT add upstream again).

Coverage

Shipped by scope marker (the double-counting axis):

Shipped by inventory kind:

Why the 107 excluded entries cannot ship (by category, then by the non-OGD leaf database their closure reaches):

Computed processing climate (our values)

OUR EF 3.1 (BAFU) Climate Change per kg, computed live on every shipped node (no external reference value exists for the processing layer — Claudio computes against his own background).

Methods shipped

EF 3.1 (BAFU) categories: 16/16; total method objects: 55.

Quality & verification

Data-quality basis per node (a 5-axis pedigree is emitted ONLY where a real one exists; elsewhere the honest signal is the internal confidence — never a fabricated PEF/DQR class):

Verification: verify_self_contained.py: PASS — package imports into a FRESH Brightway project with NO ecoinvent / biosphere3 / EDB; layers not collapsed; every shipped node computes; 16/16 EF 3.1 categories register; the hero processing row lands in the plausible per-kg band.
Identity: Shipped nodes carry sediment.terms identity keyed by GADM location plus food-taxonomy terms (FoodOn, FoodEx2, AGROVOC, LangUaL, HESTIA). Coverage is per-namespace — see identity_coverage — NOT every-node-all-namespaces. This is the version-agnostic bridge to FCM's FoodOn-based food-impact layer (the standalone FoodOn crosswalk is published separately at zhaw-fcm-bridge).

Identity coverage (nodes carrying each sediment.terms namespace, of 514 shipped — per-namespace, not every-node-all-namespaces):

Provenance shape — what openLCA shows you, per layer

How each layer of the shipped package carries its provenance, and therefore what an openLCA reviewer sees in the Documentation tab. Both shapes are real provenance — the difference is whether it is machine-readable facets or the prose the tributary wrote. Every process in the package is documented; none is silent.

Where the remaining gap sits — named, not hidden:

This is a workspace-wide gap, not a food-processing one. Across 51 mapping artifacts, 15040 of 15088 mapping entries carry a sediment identity (99%) but only 15088 carry structured provenance (100%). The identity layer was built out everywhere; the provenance layer was not. Only STRUCTURED provenance can be projected into an openLCA Documentation tab, so a low figure here is exactly what makes a well-documented dataset look undocumented downstream. At 100%: lci-agri-integration, lci-algae, lci-aquaculture, lci-biodiversity, lci-cashew, lci-chemicals, lci-coffee, lci-cotton, lci-crops, lci-dairy, lci-electricity, lci-fermentation, lci-fertiliser, lci-fishing, lci-fishing-equipment, lci-fishing-gear, lci-food-loss, lci-food-processing, lci-fruit, lci-fruit-main, lci-fuel-combustion, lci-greenhouse, lci-icbm, lci-indigo-n, lci-irrigation, lci-lentil, lci-livestock, lci-luc, lci-market, lci-metals, lci-mushroom, lci-oilpalm, lci-origin, lci-packaging, lci-pesticide, lci-rainforest, lci-recipe, lci-rice-ch4, lci-rothc, lci-salca, lci-salt, lci-seed, lci-spices-additives, lci-storage, lci-textile-processing, lci-transport, lci-waste-treatment. Full table: provenance_coverage.json.

Nutrient bridge (lci-nutrients × esfc-glossary)

205 of 509 shipped inventories carry a per-100 g energy + macronutrient profile, joined on the sediment identity layer. 304 candidate matches were REJECTED and ship with their evidence but no numbers — a withheld value beats a wrong one.

Read the quality fields in this order: identity_verified (False is rare but ~100% precise — a genuinely wrong food), then match_tier (sediment_direct > glossary_name > embedding), then match_confidence (ADVISORY, ~56% precision against a 41% base rate). embedding_cosine/embedding_distance are recorded for ordering a review queue ONLY: upstream measurement over 109 labelled links found similarity indistinguishable between correct and wrong picks (0.857 vs 0.869). Do not threshold correctness on it.

Nutrients come from national food-composition tables via lci-nutrients, joined on the sediment identity layer (esfc-glossary bridges vocabulary mismatches). They are NOT derived from the LCI and add no impact. Values are a multi-country weighted-median blend; national tables differ in energy conversion factors and recipe procedure, so the spread in nutrient_uncertainty is irreducible, not noise to be averaged away.

Known limitations / open questions

• 107 of 621 mapping entries are NOT in the package because they are not OGD-closeable: their upstream substrate links to ecoinvent / EDB / EDB_remapped_final (e.g. at-farm herbs, market for soy sauce, cocoa beans, coconut milk). They are recorded in coverage.json::excluded_non_ogd with the exact leaf database and reason. We NEVER relabel a licensed/EDB node as OGD and NEVER ship a placeholder as a measured inventory — the honest deliverable is the 514 entries whose entire closure is BAFU/tributary OGD.
• VERSION ALIGNMENT (formerly a stated drift): the processing inventories + their technosphere closure live in the EDB_BAFU project against the `bafu` DB, which since the 2026-07-24 refresh is BAFU:2026 v1 (11,947 activities, code-identical to the bafu_2026 DB). The method CFs are captured from the bafu_2026 project — the SAME vintage. The former 2025-technosphere/2026-CF drift caveat is resolved; CFs bind by (bafu_biosphere, code).
• cradle_to_gate_proxy nodes bake the upstream into the processing node (no separate substrate exchange). A consumer who also adds a farm-gate input would double-count — the scope_marker field is the guard. gate_to_gate nodes leave the substrate unbound; their OWN EF 3.1 climate can read near-zero (processing energy only) until the consumer supplies the farm-gate input.
• 0 entries are Pattern-B direct-emission (processing-energy CO2 carried as bafu_biosphere flows because no BAFU technosphere equivalent exists for the process) — legitimate per the workspace LCI Inventory Policy exempt match types, shipped verbatim, never forced onto a fabricated technosphere anchor.
• GLAM damage/endpoint methods read near-zero on these BAFU-anchored food inventories (the agriculture-dominant flows are not yet fully bridged into the GLAM BAFU companion); GLAM midpoint Climate Change matches EF 3.1 / GWP100. Treat GLAM damage results as not-yet-reliable. EF 3.1 (BAFU) is the recommended method family.

Changelog

Generated from the live shipped artifacts (coverage.json, provenance.json, methods_cf.json, node_impacts.json); regenerated on every release so this page cannot drift from the dataset. Code Apache-2.0; data CC-BY-4.0; BAFU/UVEK OGD 2026; no ecoinvent.