How we count avoided emissions, including where the counting is weak.

A country's average carbon intensity describes the mix as a whole. Moving one load does not change that mix; it changes which generator responds at that moment. Crediting a shifted load with the difference between two national averages therefore counts something that did not happen.

The right question is which plant would have produced the marginal kilowatt-hour, and which one produces it after the shift. That difference, the marginal operating emissions rate, is what a demand-shifting product can honestly claim.

Marginal accounting usually yields lower savings than average accounting, sometimes much lower, and occasionally the opposite sign. We would rather publish the smaller defensible number than the larger one that a reviewer can dismantle.

Measured: the energy an appliance actually drew, and the hours it drew it in. Published by third parties: the intensity figures for each hour and zone. Estimated: the counterfactual, meaning what would have happened without us. That is a model, not an observation, and it is the largest source of uncertainty in any avoided-emissions claim, ours included.

The demonstration on the home page currently shows average intensity, because that is what the public source publishes. Production reporting will use marginal rates.

Marginal rates are themselves modelled, and different providers disagree. Widespread load shifting changes the marginal unit, so today's rates stop describing tomorrow's grid. The effect erodes as adoption grows, which is a good problem and an honest one to state. And a shift that reduces emissions can increase cost, or the reverse; the two objectives are correlated but not identical, which is why the product exposes the trade-off rather than hiding it.

Once pilots are running, the intention is third-party review of the method and publication of the raw per-load records behind any aggregate figure. Until that exists, no impact figure appears on this site.