[cyphalle]

@AntoineDarwin ratify or correct the six methodology choices below. Each one is already in the code and public.

The report now gives one nature footprint per candidate data centre, in species·yr (ReCiPe 2016, ecosystem quality), split by IPBES pressure. On 30 days of real use: climate is 86-90% of the total at every site, pollution 4-10%, water 2-3%, land 2-4%.

Methodo & data decisions

1. ReCiPe 2016 Hierarchist, US country factors where the report gives them (water, acidification, ozone). Rejected: world averages, as in Darwin's reference table.
2. Grid per site: eGRID2023 subregion (CO₂e, NOx, SO₂). Rejected: one 0.287 kg/kWh for all sites.
3. Land: Lovering 2022 medians on the eGRID mix. Wind and gas use the footprint (spacing rejected). Biomass uses residue (dedicated rejected).
4. All land as "artificial areas" occupation (0.73). Transformation not counted.
5. Water endpoint stays a country average. Basin stress (Aqueduct) is a separate column, not a factor.
6. Readable unit: species·yr / 1.48e-8 species/m² = m² that lose all their species for a year.

Published before ratification, by decision of @cyphalle on 2026-10-16. A correction goes in a follow-up PR.

---

Context for agents

• Every factor, with its table number in RIVM Report 2016-0104a, is in METHODOLOGY.md § Nature footprint.
• Data: data/nature.json (ReCiPe, Lovering), data/sites.json (Aqueduct, eGRID). Regeneration: scripts/*.sql.
• Not counted: hardware manufacture and mining, data centre land, Memphis on-site turbines, invasive species, human health.
• Next step for water: Pfister WSI per basin, the method behind the ReCiPe country factor.
• Tests: cargo test (11), clippy and fmt clean.

🤖 Generated with Claude Code

https://claude.ai/code/session_01QUg7CcESDGqbwXB484UeRx