I work for OpenSignal, and produced the data used in this report, so I'd like give some assurances around the coverage aspect. The principle is to calculate the percentage of time that each user was connected to LTE over the period in question, and then average over such users to arrive at an average figure per network. Naturally I'm understating some complexities around minor issues (mainly around cleaning and combining users' measurements), but it isn't like we've gone completely off-road and developed a black-box procedure here - it's pretty much what you would expect.

On the first point, yes, you're completely right that we are measuring the whole package, however we'd argue that this is what users really care about.