I don't know the answer to this, but am am going to make a logical assumption, so this is more of a question..... Its been said by many on here the the models are only as accurate as the data which is fed in. Fair enough, an obvious statement. But surely irrespective of the accuracy of the input data, the models accuracy is dependant on the quality of the program which is processing the data. Now in the case of our usual UK winter weather - atlantic dominated, there must have been a significant amount of learning put into the programming code to enable more accurate data processing. Hence we have more accurate mid range forcasting for this type of event. But in the case we have in hand at the moment - a rare easterly, the models do not have same quality of processing to provide accuracy over the mid range because there has been so little learning. This is why there has been such disagreement between them and also why we see constant upgrades and downgrades.