53x12 - if you're cycling the bike the inherent difficulties are (1) ensuring accurate measurement of all the different forces acting on the system (rider plus bike) and (2) repeating the experimental conditions, not least the rider movement or line along the road.
The delta we're attempting to isolate is very small.
I don't see that anything you've posted overcomes these difficulties and would allow you to effectively measure real world differences between frames accurately.
BTW, one small point to raise about the tests is that no-one can maintain absolutely constant power output. It is probably impossible for a human being to do. And the measurement from powermeters (and any other equipment) used to measure the various forces isn't accurate enough to allow compensation for this fact.
I could go on - but you get the point I hope.
Well, you're right about many of your points but not quite about the entirety of the message.
1. I completely agree that no one can maintain absolutely constant power output, and that it's probably impossible for a human being to do. However, this protocol doesn't require that you maintain constant power (or constant speed). In fact, this protocol works best if you have a wide range of power and speed.
2. You're absolutely right that the instantaneous accuracy of power meters can be larger than the difference in drag between, say, two frames. However, this protocol works not across instantaneous data samples but across aggregated segments or intervals of data samples. In that case the instantaneous accuracy of the power meter matters much (much) less. There's also a way to test for accuracy drift across tests, and it's good field practice to do that if you're trying to isolate small differences.
3. You're completely right that real-world conditions can affect the estimates. However, one of the nicest things about this protocol is that you get a diagnostic that tells you when the conditions interfere with a good estimate (like, when a car passes, or a momentary use of the brakes to avoid a hazard, or when the wind gusts). So while it would be great if we could eliminate all those problems the next best thing is to be able to tell when that happens so we can drop the "contaminated results" and re-test.
4. You're absolutely right that there's a detection limit beyond which field tests can't detect small changes in frames but that's also true about all measurement methods. Nonetheless, the detection limit is actually pretty small. The best I've ever seen in an outdoor field test was in a case where the weather conditions were nearly ideal, there was no other passing traffic, the rider was experienced in the method, and the venue was well-suited to getting a wide range of test speeds. In that case the SE on the estimated CdA was around .0006 m^2, or about 0.3% of total drag area. But, admittedly, that was a pretty clean case. The method is also being used in more controlled indoor (velodrome) settings, and SE's of that magnitude are more common. There are frames on the market that absolutely are more than .001 m^2 different in CdA (which would be twice the size of the SE).