[cyphalle]

Review at_site() in src/nature.rs and the new fields in data/sites.json. Stacked on #6.

Closes #2.

Effect on the author's 30 days: CO₂e +14% (32.8 kg of 267.9 kg). The water term moves by less than 1%, because cooling is 3% of the litres.

Methodo & data decisions

• Hardware manufacturing: 40 gCO₂e per kWh, range 9-47. It is added to the climate pressure of every site. Sources: TPU LCA (Schneider 2025, 35-47), Gemini (Elsworth 2025, 42), Mistral/ADEME (37). Alternative: H100 PCF, 6-9. That covers only part of the server.
• Its water and minerals are not converted. Mineral scarcity is in the ReCiPe resources endpoint (USD), which does not add to species·yr.
• Water: terrestrial factor × CF_EQ of the site watershed / 0.3514 (US). Source: Pfister 2009 KMZ (ETH, Wayback). Alternative in Regionalise the water term at the basin, so basin stress moves the species·yr figure #2: WSI ratio. It is not used, because in Pfister 2009 WSI drives only human health. The US WSI is also not available.
• The multiplier applies only to cooling water. Power-plant water keeps the US factor, because it is drawn on the grid.
• The freshwater factor stays US: it is 1,000× smaller, and the Great Lakes and Columbia have no value.
• Limit, documented: CF_EQ does not see withdrawals. New Carlisle (Aqueduct High) gets the lowest multiplier, 0.31. The Aqueduct column stays.

Tests: 12 green, clippy clean.

🤖 Generated with Claude Code

https://claude.ai/code/session_01QUg7CcESDGqbwXB484UeRx