What it is not

Naming these is not a hedge: a factor handed over without them is what fails assurance.

Lifecycle, not generation-only, and CO₂e. The Scope 2 methods use factors assessed at the point of generation. What happens upstream of the plant (extracting and moving the fuel, methane leaks, building the plant) is reported in Scope 3, category 3. This series includes it: it is Ember's estimate, each fuel's generation multiplied by a lifecycle factor, in CO₂ equivalents. So it reads higher than the factor the standard describes, and a company that also reports category 3 would count the upstream part twice. The clean end shows it: the 7 grids above 10 TWh that are at least 99.5% low-carbon carry 23.1 to 27.8 g per kWh, where a generation-only factor would be at or near zero.

One open series does count generation only, for one region: the European Environment Agency's for the EU. For 2023 this series reads higher than it in 23 of the 27 members, by a median of 34 g per kWh, and 14% higher for the 27 together. It reads lower in four, Estonia most of all, because the two differ in method as well as in boundary: the EEA divides the emissions a country reports by what it generates, and Ember multiplies generation by a factor for each fuel.

Location-based only. This is the grid average where a site sits. A market-based factor reflects the contracts a company actually holds, such as RECs, Guarantees of Origin, PPAs and supplier-specific residual mixes, and no public dataset carries those. A company reporting both bases will find the two lines differ substantially, and only this one can be built from open data.

An annual average, not hourly matching. A site drawing power overnight on a wind-heavy grid, or at a summer peak met by gas, is not on the annual mean. 24/7 carbon-free-energy accounting needs hourly generation data; a yearly grain structurally cannot express it.

Production-based, not consumption-based. OWID's series is the carbon intensity of electricity generated in a country. It ignores trade, so a country that imports much of its power is assigned only what it generates itself: one importing a neighbour's hydro looks dirtier than the mix it consumes, and one importing coal power looks cleaner.

One more, from the warehouse rather than the standard: five territories in OWID's energy data (Guadeloupe, Martinique, Réunion, French Guiana and the Falklands) are not countries in this warehouse's country list, so they carry no factor here. That list is what counts as a country throughout, and it is also what keeps World Bank groupings like "World" and "European Union" out of every total. The coverage page is where absences are rows.

The factor is OWID's carbon_intensity_elec, which OWID takes from Ember's Yearly Electricity Data and titles "Lifecycle carbon intensity of electricity generation", modelled as marts.dim_grid_emission_factors and snapshotted as history.snap_grid_emission_factors (SCD2, check strategy, 2015 onwards).