They're looking at average speeds seen in the dataset of the users of their speed-test app. So the idea is instead of looking at peak speeds in good conditions,

They're looking at average speeds seen in the dataset of the users of their speed-test app. So the idea is instead of looking at peak speeds in good conditions, looking at actual speeds in real-life conditions, to see how well the modem deals with cell selection, poor reception, congested networks, etc. That said, the phone isn't out yet? Where is this data coming from? If it's all staff at Apple Campus that's not going to be a realistic comparison.