They don't go by past data. If they did the prediction in March July and late September would have become more accurate with each wave with latest actuals between somewhere between the lower and upper quartile predictions.
Last September the lower quartile prediction was between two and three times HIGHER than the actual numbers.
As a statistician the SAGE models really, really bug me. They are shit and are are virtually useless.
They are so consistently bad that I will take their lower quartile prediction as the very worst case and scale it back from there.