They are looking at average quality metrics. I think they should be looking at perc based quality metrics. That would have reflected a drop and the variability far earlier on. In this sense, it's akin to performance metrics — you want to find out what % of your users are experiencing what levels of service.