Look into capturing marginal intensities rather than means

[fuglede]

When deciding to pulling energy from the grid for some operation, besides knowing the average intensity of all such operations (which is what we capture right now), it would also be nice to know the what energy sources one ends up drawing on (i.e. a forecast version of #4), and the intensities of those; it's not a priori that useful for decision making as it's not like people queue up to use power, but it has an impact on the uncertainty of the estimates, cf. e.g. this blog post.