Energy measurement report checklist for AI systems
Everything a reader needs to check an energy figure, in the order a report should give it. Built from the Joule Ledger, JL-26, and usable with any method.
1. Identify the system and the method
- The system, hardware, firmware and software versions
- The method and its edition; for JL-26, the edition and revision, as listed in the changelog
- Who measured it, and the audit grade: Declared, Witnessed or Replicated
2. State the boundary and the period
- The meter point for every figure: M1 device, M2 board, M3 system or M4 facility
- The start and end of the measured period, and the task set
- For annual facility ratios, at least twelve months of integrated energy
3. Report every account
| Account | What to report |
|---|---|
| Intake | All energy entering the system, integrated at the meter point |
| ΔReserve | The change in stored energy over the period |
| Work | Energy that produced completed useful tasks |
| Overhead | Cooling, idle draw and conversion losses, sub-metered where possible |
| Waste | Energy that left as heat without doing work or overhead |
See the five accounts.
4. Show closure and uncertainty
- The closure residual, r = (Intake − sum of accounts) ÷ Intake, and whether the ledger closes; see the closure rule
- The combined expanded uncertainty, at coverage factor k = 2
- At least one account other than Intake measured independently
- A ledger that does not close, reported as "not closed" with its residual, never adjusted to fit
5. Label every number
- Measured or Estimated for each value; see labels and grades
- The instrument for measured values, and the assumptions for estimated ones
- Ratios with their meter points, such as OR(M4/M3) or DY(M3)
6. Leave out what the figures cannot support
- Adjectives such as "green" or "clean" without numbers
- Comparisons across different meter points
- Claims beyond the boundary; see substantiating AI energy claims
Questions
Can I use this checklist without JL-26?
Yes. Any method that names its boundary, period, accounts, uncertainty and labels gives readers what they need to check a figure.
Where can I see a complete example?
The worked examples show one ledger that closes and one that does not, with the diagnosis.
Keep reading
Worked examplesTwo illustrative Joule Ledgers for an AI inference rack: one that closes within uncertainty and one that fails, with the residual diagnosed step by step.The closure ruleThe closure rule: Intake = ΔReserve + Work + Overhead + Waste, the residual, the uncertainty test, and why one account must be measured independently.Ledger closure checkerFree closure checker: enter Intake, ΔReserve, Work, Overhead and Waste to see if your energy ledger closes, plus Overhead Ratio.Substantiating AI energy claimsEvidence and wording for AI energy claims as of November 2026: EU Directive 2024/825, the FTC Green Guides, boundaries, and measured vs modeled.
Take the next step.
What a credible energy report for an AI system must contain: boundary, period, accounts, closure, uncertainty, labels and grades. A free checklist.