A disclosure, computed

The standard fixes the arithmetic and publishes no coefficients. So this runs methodology 2.0.0 on numbers you supply, in your browser, and shows every step it took. Nothing is sent anywhere, and no value is filled in for you: until every field below carries a number, the panel says which ones it is still waiting on.

your inputs

inputTokens tokens Tokens sent to the model. outputTokens tokens Tokens the model returned. modelInputWhPer1k Wh per 1k tokens Your coefficient, with its evidence source. modelOutputWhPer1k Wh per 1k tokens Your coefficient, with its evidence source. providerPue ratio Your provider's published PUE. gridIntensityGPerKWh g CO2e per kWh Your region's intensity, with its source.

methodology 2.0.0, as published

energyWh = inputTokens / 1000 * modelInputWhPer1k
         + outputTokens / 1000 * modelOutputWhPer1k
energyWh = energyWh * providerPue
gCO2e    = energyWh / 1000 * gridIntensityGPerKWh

treeTimeMinutes = gCO2e / 21000 * 525600

your numbers, step by step

Waiting on all six values. Each one is yours to source and to evidence.

required with every rendered disclosure

Energy and carbon are modeled estimates.
Tree-Time and equivalents are educational comparisons.

Confidence for a figure computed this way is a token proxy, the weakest rung the standard names. Direct measurement is stronger than a hardware factor, which is stronger than this. Never upgrade it because the estimate looks plausible.

This computes Level 1 only, energy then carbon. A record also has to identify its id, subject, scope, time window, grid region, confidence, methodology version, generating source and generated-at timestamp. Those come from you, not from arithmetic. The field list is on get started, and the citable bytes are at /source/AIEDS.md, 8da923acee51.