Pathways › Eaternity EOS · Overview · EDB Changelog · Regression · Benchmark

Rating Benchmarks

How A-E ratings are derived from real-world consumption data

Ratings reflect a recipe's environmental performance relative to average Swiss consumption. Climate and water each use an independent benchmark derived from sold-portion-weighted averages across a large restaurant panel. Rainforest and animal welfare are rated against fixed criteria instead of a benchmark (see below). The nutrition score (Vita-Score) is shown for information only; it is not part of the scheme. The benchmark is the consumption-weighted average and the upper limit of B: a rating of B or better is below average (a result exactly at the benchmark is a C). A represents a science-based target of 50% below the benchmark.

Computed on database release v1.12.5 (the previous benchmark of record, on v1.12.4, was 2,810.6 g CO2eq/DFU). Ratings use it from the switch after the validation; until then production rates against its previous benchmark. A result exactly on a threshold gets the worse letter.

CO2 Climate

2,650.3
g CO2eq / DFU
RatingThreshold
A< 1,325.15
B< 2,650.3
C< 5,300.6
D< 10,601.2
E≥ 10,601.2
Science-based target: A = 50% reduction

H2O Water Scarcity

2,967.05
L world-eq / DFU (AWARE 2.0)
RatingThreshold
A< 1,483.525
B< 2,967.05
C< 5,934.1
D< 11,868.2
E≥ 11,868.2
AWARE 2.0 (annual agricultural), on the validation environment since 3 November 2026; production still rates water on its previous basis

V Vita-Score (Nutrition) Not part of the scheme

336
points / DFU
Sample: 6,067 recipes (DFU-weighted)
RatingThreshold
A≤ 270
B≤ 340
C≤ 410
D≤ 450
E> 450
Lower is better (Vita-Score inverts Nutri-Score)
For information only. The Vita-Score is a nutrition rating, not an environmental claim; it is not covered by the scheme or by its assessment.
Criteria-based ratings

Rainforest and Animal Welfare

Part of the scheme as provider-declared impact categories (ESFC EmpCo Rulebook v1.1, Section 5.6). They are checked differently from climate and water: there is no benchmark. Each rating classifies the rated product's content by origin and certification evidence, in grams per 100 g of the whole product. Only A, B and E are used. A category enters certified scope only with a positive conclusion of the assessment body.

AW Animal Welfare

C = welfare-certified animal-derived content, U = uncertified animal-derived content (g per 100 g)
RatingRule
AU ≤ 1; any C, including none
BC > 1 and U > 1
EU > 1 and C ≤ 1
The rule classifies certification evidence; it does not measure each animal’s welfare directly.

RF Rainforest

R = rainforest-specific certified critical content (soy, palm), O = other-certified critical content, U = uncertified critical content (g per 100 g)
RatingRule
AU ≤ 1 and O ≤ 1; any R
B(O > 1 and U ≤ 1) or (R > 1 and U > 1)
EU > 1 and R ≤ 1
Critical content is modelled soy and palm, including feed assumptions; it can differ from ingredient mass.

Amounts of 1.00 g per 100 g or less count as absent. B is an intermediate class: it does not mean that most of the content is certified. The full rules, with worked examples, are in the pathway annex (PA-06.2, Animal Welfare and Rainforest decision rules).

Visual Reference

Climate Rating Scale

The rating scale for CO2, showing how thresholds divide the performance range. The benchmark (B/C boundary at 2,650.3 g/DFU) is the consumption-weighted mean of the kitchen benchmark on the Swiss inventory basis, 2026 DFU definition.

A
B
C
D
E
< 1,325.15 < 2,650.3 < 5,300.6 < 10,601.2 ≥ 10,601.2
Values in g CO2eq/DFU (2026 DFU definition). Logarithmic band geometry: B, C and D each span one doubling of the threshold (×0.5, ×1, ×2, ×4 of the benchmark); A and E are open-ended, drawn at equal width. The B/C boundary at 2,650.3 equals the consumption-weighted benchmark.
Data Foundation

Data Coverage

Benchmarks are derived from real contract-catering procurement data: complete delivery records of the sampled kitchens, not menu plans.

Restaurants
151
Export selection ("223 kitchens selected" cost-centre label); 108 kitchen IDs carry delivery data
Observation Period
3 years
2019–2021 (per-year kitchens 102 / 96 / 89; 83 in all three)
Delivery Rows
1,738,698
Per-product delivery records, each with its own nutrients and CO2
Weighting
Delivered quantities
All deliveries of the sampled kitchens (includes kitchen waste; no menu-plan estimates)
Region
Europe
European contract catering
Methodology

How Benchmarks Are Calculated

Daily Food Unit (DFU) is a standardized unit representing one person's daily food intake — the 2026 definition, whose five denominators all derive from one reference person (WHO/FAO 60 kg adult, 2,000 kcal/day). It is calculated as a 5-component average:

DFU = average( protein/50g, fat/66g, rest_energy/5076kJ, water/2500g, dry_weight/325g )
rest_energy = energy − 17×protein − 37×fat     dry_weight = mass − water − protein − fat

Each component normalizes a nutritional dimension to its daily reference value. Rest-energy and dry weight both strip protein and fat symmetrically, so those two macronutrients are counted exactly once; the 325 g dry-weight reference is the reference person's macro-stripped daily dry matter (carbohydrate 289 g + fibre 25 g + ash 11 g). The DFU ensures fair comparison across food types with different nutritional profiles.

Benchmark calculation: For each metric, the benchmark equals the consumption-weighted average — total impact divided by total DFU delivered — across the observation set. The benchmark of record is measured on real purchasing data: 108 kitchens, three years (2019–2021), 1,738,698 delivery rows on the Swiss (UVEK/BAFU) inventory basis of database release v1.12.5. Numerator and denominator come from the same export: the denominator is the engine's own Daily Food Unit column under the 2026 definition.

Rating thresholds for CO2 and water are derived from each benchmark using the same fixed multipliers: A below 0.5×, B below 1.0× (benchmark), C below 2.0×, D below 4.0×, E from 4.0×; a result exactly on a threshold gets the worse letter. Vita-Score thresholds (for information; not part of the scheme) are derived from DFU-weighted percentiles: A = benchmark × 0.80 (~P10), B = benchmark (~P50), C = P80 (worst 20%), D = P90 (worst 10%).

Derivation record: the full measurement behind the benchmark — every product family, every tonne, traceable to source rows — is maintained as an internal review record and made available to the CAB during the engagement; a client-facing change communication (what moved against the previous ecoinvent basis, and why) is prepared separately.

EmpCo Scheme Document

Plausibility

Cross-Check Against Top-Down Studies

This benchmark is built bottom-up: real purchasing rows × life-cycle inventories. Global studies estimate food-system emissions top-down from emission inventories. The two must agree within their boundary differences; they agree in order of magnitude, and the open points are listed below. The chain below makes the comparison explicit; every step is arithmetic on published figures, not a dietary statement.

StepValueWhat it does
Benchmark, per delivered DFU2,650.3 g CO2eqConsumption-weighted mean of the kitchen sample (food as purchased, kitchen and plate waste inside)
× 365 days0.967 t / yearOne reference-person year at exactly 2,000 kcal/day of delivered food
× 2,985 / 2,000 kcal1.44 t / yearScaled to the actual global per-capita food supply (FAO Food Balance Sheets, 2022). The supply figure already contains consumer-stage food waste — UNEP measures 19% of consumer-available food wasted (households 60% of it) — so waste must not be added again on top of intake scaling
× 3,471 / 2,000 kcal1.68 t / yearSame scaling at the European supply level — the highest regional DES in 2022 (FAO: 3,471 kcal/cap/day), closer to this benchmark's European catering basket
Top-down reference2.2 t / yearEDGAR-FOOD (EU JRC, 2024 update, data year 2023): 18 Gt CO2eq — one third of global GHG emissions — over the world population

The residual (about 34 % on the world basis) is in the direction the boundary differences predict, but most of it is not yet attributed. EDGAR-FOOD counts stages this delivered-food benchmark does not carry in full: end-of-life waste management (6% of the food-system total), domestic food preparation (~2%), the full breadth of land-use-change accounting, and refrigerant F-gases (4%) — and it averages the world's diets, while this benchmark measures a European contract-catering basket. A further 13% of food is lost between harvest and retail (FAO); those upstream losses are largely embedded in the life-cycle inventories themselves. An independent bottom-up measurement landing below the top-down world average is expected from these differences; how much of the gap they explain is not yet measured (U1). (UNEP attributes 8–10% of global GHG emissions to food loss and waste alone — consistent with the waste-related wedge of this residual.) The named wedges quantify to at most ~0.27 t of the 0.76 t world-basis residual. What remains open is stated here rather than absorbed into the narrative:

  • U1 — Unattributed share: the named wedges (end-of-life 0.134 t, domestic preparation 0.051 t, F-gases 0.088 t) cover at most ~0.27 t of the 0.76 t residual; at least ~0.5 t is currently unattributed.
  • U2 — Boundary double-claim (the annex statement yielded, 28 September 2026): the pathway annex used to declare cradle-to-grave coverage including end-of-life. Pathway annex PA-07.4 now declares the operated boundary: Eaternity Gastro ratings run from agricultural production to delivery at the kitchen; end-of-life and the consumer phase are excluded, and kitchen preparation is not systematically modelled (open point: tag-triggered heat treatment, to be confirmed). The end-of-life and domestic-preparation wedges can therefore explain part of the residual, subject to that open point and to an inventory-level check of whether matched inventories embed such flows.
  • U3 — Basket direction: Europe's more animal-heavy supply (77.7 vs 44.5 kg meat/cap/yr, FAO) should raise per-kcal intensity above the world average, yet the European-scaled figure (1.68 t) sits below the world top-down (2.2 t); the compensating forces are plausible but unquantified.
  • U4 — GWP metric vintage: EDGAR-FOOD's GWP100 basis is unverified against our EF 3.1 / AR6 characterisation (35% CH4 share ⇒ several-percent sensitivity).
  • U5 — Missing like-for-like comparator: a European per-capita top-down figure from EDGAR-FOOD's country data would turn U1/U3 into arithmetic — the cheapest item to close.
  • U6 — kcal-only scaling: the supply scaling treats one DFU as 2,000 kcal; the DFU also normalises protein, fat, water and dry matter — a sensitivity check on pure kcal is defined.

None of these endangers the order-of-magnitude conclusion; each carries its resolution path in the evidence dossier, and the list shrinks — never silently — as items close.

Sources: Crippa et al. 2021, Nature Food (food systems = one third of global GHG emissions); EDGAR-FOOD, EU Joint Research Centre (2024 update: 18 Gt in 2023; 2.2 t CO2eq per person); FAO Food Balance Sheets 2010–2022 (global dietary energy supply 2,985 kcal/cap/day, 2022); UNEP Food Waste Index Report 2024 (19% of consumer-available food wasted; FAO: 13% lost post-harvest to pre-retail); Poore & Nemecek 2018, Science (independent bottom-up global estimate: 13.7 Gt CO2eq, 26% share). The comparison is a reference-frame reconciliation between system boundaries; none of these figures is a statement about what any person eats or should eat. Every claim above is documented with its verbatim source quote, access date and archive hash in the evidence dossier.