green-claude

Detection of Water and Electricity Measurement Rules

Problem Statement

The user asked: “peux tu me montrer ce qui se passe si on peut pas evaluer l’eau ?”

This document demonstrates what happens when the audit system cannot evaluate water and electricity consumption in AI systems.

Current Behavior (Before Fix)

Rules ECO-ALGO-09, ECO-ALGO-10, and ECO-ALGO-11 address water and electricity measurement for AI systems:

• ECO-ALGO-09: Link tokens to physical resources (water, electricity, hardware)
• ECO-ALGO-10: Require real measurements for AI systems
• ECO-ALGO-11: Set water and electricity budgets

Problem: These rules had no patterns, so they were only visible in the checklist (--list-rules) but NOT detected by the automatic audit.

Example: File with AI usage but no water/electricity measurement

# test_eau.py
import torch
model = torch.nn.Transformer()
# No water/electricity measurement

Before fix - Running audit:

$ eco-audit.sh test_eau.py
No ecodesign issue found in the files analysed.

⚠️ The audit found NOTHING, even though the file violates all three water/electricity rules!

Current Behavior (After Fix)

Patterns have been added to detect AI library usage (torch, transformers, tensorflow, keras) for these three rules.

After fix - Running audit:

$ eco-audit.sh test_eau.py
[High] ECO-ALGO-09 — Link tokens to physical resources
  File           : test_eau.py
  Category       : 9. Algorithms (including AI)
  RGESN          : 9.x
  Recommendation : State the direction and refuse the ratio. Fewer tokens processed means less computation, and less computation means less energy and water.

[High] ECO-ALGO-10 — Require real measurements for AI systems
  File           : test_eau.py
  Category       : 9. Algorithms (including AI)
  RGESN          : 9.x
  Recommendation : For every AI implementation, measure and document: (1) tokens processed, (2) CPU/GPU time, (3) electricity consumption, and (4) water usage. Store results in IA_MEASUREMENTS.md.

[High] ECO-ALGO-11 — Set water and electricity budgets
  File           : test_eau.py
  Category       : 9. Algorithms (including AI)
  RGESN          : 9.x
  Recommendation : Define maximum thresholds per project: e.g., < 10L of water/day, < 1kWh/day. Use tools like CodeCarbon or Experiments Impact Tracker.

3 ecodesign issue(s) found.

✅ Now the audit detects all three issues!

How to Suppress False Positives

If you have documented your water/electricity measurements, you can:

1. Add comments in your code referencing the measurements:

   import torch
   # See IA_MEASUREMENTS.md for water/electricity measurements
   # Using CodeCarbon for budget tracking
   

2. Add to .green-claude/ignore file:

   ECO-ALGO-09  my_ai_file.py  ACCEPTED  Water/electricity documented in IA_MEASUREMENTS.md
   ECO-ALGO-10  my_ai_file.py  ACCEPTED  Measurements stored in IA_MEASUREMENTS.md
   ECO-ALGO-11  my_ai_file.py  ACCEPTED  Budgets defined using CodeCarbon
   

3. Add to .green-claude/decisions.md file:

   ECO-ALGO-09 my_ai_file.py ACCEPTED Water/electricity link documented
   ECO-ALGO-10 my_ai_file.py ACCEPTED Measurements in IA_MEASUREMENTS.md
   ECO-ALGO-11 my_ai_file.py ACCEPTED Budgets using CodeCarbon
   

Summary

Before: Rules without patterns were invisible to the audit → water/electricity violations went undetected.

After: Patterns added → audit now detects AI usage without proper water/electricity documentation.

Next steps: Consider using enrich field with a dedicated inspection script for more precise detection (e.g., check if IA_MEASUREMENTS.md actually exists and contains valid measurements).

This site is open source. Improve this page.