Every model's population-weighted heating and cooling degree days, on one table — CONUS, the census divisions CPC publishes on, the power ISOs and the gas regions. Degree days are taken per grid cell before any weighting (the max(0,·) never commutes with an average), on the same census-population cube the truth uses, so a model line and the verifying analysis are the same quantity from different sources. Blue dots are what actually happened (URMA, preliminary — nClimGrid settles the record).

The cluster engine cuts each ensemble pool into competing futures; here each scenario prices the CONUS degree days for its own accumulation window — Days 1–5 clusters price Days 1–5, and so on. Share × consequence is the book: the probability-weighted expectation, the spread the scenario split is worth, and each scenario’s share trajectory across recent runs.