How to read it. Each row is one model run, oldest at the top; each column is six hours of the storm. A cell is how much that run put in that window, on the map's own colour scale. Read down a column to watch one night's snow appear, move and settle as the runs approach it; read along a row for one run's storm. Hatched cells had already happened when the run was made. The last row is what fell, in the only cells the analysis has — whole climate days — and the right-hand column is every run's total, with the truth ruled down it.

How to read it. Each row is a threshold somebody acts on. Every dot is one model run, placed at its lead time, filled if that run expected the threshold to be reached. A red ring means that run got the answer wrong. The green band and its flag mark the lock — the earliest run after which no later run changed its mind, and the mind it had was the right one. That is the honest answer to “when could we have acted”, and it is a different question from “what was the error”: a forecast that is right from four days out and one that is right only in the last twelve hours score the same on error and differently here.

A grid cell is not a gauge. Both the forecast and the analysis are 2.5 km averages sampled at the city's coordinates, so a valley town and the ridge above it can share a number that suits neither exactly. These are not station observations and should not be read as one.

The range stops at f054. Past two and a quarter days the blend publishes one snowfall number and nothing about its spread. Where a run shows no band, that is the product ending, not the uncertainty ending.

A negative lead is a nowcast. Runs initialised after the window opened are still shown, because watching a forecast settle once the storm is underway is part of the story — but they are not a forecast of the whole event and their totals cover only what was still to come.