They ride a track with a Powertap, possibly with an SRM as well (in most recent tests), at a moderate speed (150 watts, resulting speed 23-28.5 kph) then use AnalyticCycling to extract a Crr, which they temperature-adjust using results from a control tire. This isn't the most scientific method possible, as they don't say how they adjust for Crr.. personally I like the Chung method which is similar but slightly different. They inflated tires to the Berto recommended pressure, which targets 15% drop (ie loss of rolling radius due to load = 15% of the tire radius measured laterally, on the rim). But the results are sufficiently comprehensive that they are of interest. An in any case, if I want to criticize their methods, I should do my own freakin' test.
I'm glad they've switched to this kind of test, though of course I'm guessing that if the protocol is exactly as you describe their precision isn't very good. Here's how Heine described his Crr testing protocol last year
, which involved a hill coast down, two spotters spaced 184m apart, and an unspecified way to synchronize the timing devices.
My protocol is pretty different because it's not done at either constant speed or constant power.
Some readers might find a Crr of 0.34% more familiar as .0034, but it can be handy if you recognize that Crr scales exactly like a slope, so an additional 0.1% in Crr is just like a slope which is 0.1% steeper.