Surely the high data in August should have been looked at again. The scalings clearly go wrong after the August breakdown until the next calibration in early September. This was reported as the final dataset.