New method for evaluating out-and-back data

[HDPWR]

I am a new user of Aeropod. I bought the device to enhance my TT position and equipment. In the past days I tried to read all the manuals and instructions available and I have to say it is a lot of reading. Very impressive and huge thumbs up for the detailed and up-to-date documentation!

I would like to propose a new method for evaluating out-and-back data and really would like to hear your opinion about it.
In my opinion it would be more precise to cut off the turns at the end and beginning of an out-and-back lap. Let's say a 1 km long middle part could be selected based on GPS data, where the rider already got up to the target speed/power. The VA software could help in automatically dividing the record file into out legs and back legs, based on the segment selected by the user. These legs could be paired and then the "Analyze CdA" function could be called on selected pairs. To make the aero testing even more comfortable, the software could add one more column to show the average of selected analyzed CdA values, for example the average of 3 laps (6 legs in total) that was measured in the same position with same equipment.

Another, perhaps easier implementation would be to chop off the first 2 and last 2 seconds of each lap that was generated by Garmin. In this case the user should take a new lap on Garmin for each leg, but then the 2 seconds where the rider is in a transition to/from the racing position would be eliminated from the CdA calculations. In my opinion it is important for high-performance cyclists, for whom covering a 1000m leg is only 65-70 seconds and precision is key since their CdA is already in the 0.2-0.25 range.

With the current version of the software I could manually create laps for the legs, but then I would end up with an out leg lap, then a short turn lap and a back leg lap. Using the "Analyzed CdA" function on the out and back legs together is not possible, because they cannot be selected without the intermittent turn lap. Please, as a first update, make it possible to select laps in the VA software by CTRL+left click, not only with SHIFT+left click.
Another hack would be to divide the ride file manually and save only the meaningful out and back leg parts. Then merge these files in order to use the "Analyze CdA" function.

By the way, I am really graceful for your quick and thorugh replies to my previous emails before registering to the forum!

[Velocomp]

Our experience is that the "turn around" portions of test loops does not materially alter the overall CdA measurement.

Furthermore, as long as the test loops are consistent, any (minor) offset error introduced by turn arounds will be the same for the various measurements, so it has no real effect on measurements, particularly when you're evaluating alternative configurations.

If you select multiple laps in the column, the Analyzed CdA will be the averaged value of the multiple laps.

[HDPWR]

Thanks for the quick response!
Your statement surprises me quite a bit, especially because for the turn around the rider hast to change position on a TT bike to be able to brake. I will test the effect of turn arounds to see it myself.

Regarding the CTRL+left click lap selection in VA, could you maybe add this feature? It could be useful even in case of a screwed up lap somewhere in the middle of a long test.

[Velocomp]

Thanks for the quick response!
Your statement surprises me quite a bit, especially because for the turn around the rider hast to change position on a TT bike to be able to brake. I will test the effect of turn arounds to see it myself.

Regarding the CTRL+left click lap selection in VA, could you maybe add this feature? It could be useful even in case of a screwed up lap somewhere in the middle of a long test.

The amount of data collected during turnaround is small compared to data gathered otherwise. Also, due to slow speed of turnaround, data is even further insignificant.

You can add laps to any file as follows:

1) Click on the lap tab. The data will turn white

2) Select where you want to add a lap

3) Hold the Option key and click on the white data. You'll see something like

1.png (76.08 KiB) Viewed 351 times

this:

drjohnson@innovationsmedical.com

4) Select "Add lap..."to create a lap marker at the location you selected

[HDPWR]

So I did my first test yesterday and I am a little bit disappointed.
I used Profile 3 with no Garmin device to mark laps and the test route was a straight and flat approx. 1.5km long bike path segment. It is not too wide, in order to turn back I have to really slow down.

First I did 5 laps with a focused middle 1km part where I was fully in position with a target of 300W. For these laps I added approx 100m to slow down and turn around. This means usually to brake intensely from 48km/h to almost 0. As you can imagine, I completely changed position for the braking and also reduced the power output dramatically.

Analyzation: the great thing is that the calibration seems to be perfect and data is good. However, I can see a huge moving average with a window of approx. 2 mins. My laps look like these:

laps.jpg (120.45 KiB) Viewed 330 times

A significant rise in CdA at the turning point, but clearly averaged over a wider span. This makes my data unusable because as soon as my CdA would settle, there is again a turning point and the moving average gets into action. I also tried cutting out the 1km-long legs and put them together, but then I could clearly see that the CdA data was already averaged by the Aeropod and not by the VA software.

After the first 5 laps I did 6 other with extending the length of the braking phase to 250m. So basically 250m accel - 1km focused leg - 250m decel. This was way better, I had more time to brake, turn, accelerate and get into position. As you can see, the situation with the moving average was the same here, too.

6laps.jpg (122.82 KiB) Viewed 330 times

I attach the laps table and the advanced data of the first 5 laps.

5-laps-advancedmerged.png (254.05 KiB) Viewed 330 times

One thing I noted: I used dynamic power smoothing on the Aeropod, now I set it to 0s. Does it influence the CdA calculation or just the transferred power data that gets to my bike computer or both? Could it help me in this case?

In Profile 4, is there a similar moving average function or does that report really raw values without much filtering? Could I get useful data for my usecase with Profile 4? If yes, then I will buy a Garmin.

I am already afraid that Profile 4 also applies some additional filtering, at least I am guessing it because you told me that the lap button should be pressed on the Garmin to let the AP make some compensation calculations. I think for the normal out-and-back usecase it will not suit. I think most people cannot find a 5-7km long segment without cars to eliminate the effect of the turning. Even then it would be annoying to know that you must be pretty consistent position-wise during the turnings to get precise data.

I attach the cut-out of my first lap, the out and back 1km legs combined. Here you can see again that the CdA would settle in the middle. Side notes: my position was not fully fixed, hence the big difference in the min CdA out and back.

I really hope that you can give me a good solution for my problem.

[HDPWR]

For some reason the forum doesn't let me add more than 3 pics.

I attach the cut-out of my first lap, the out and back 1km legs combined. Here you can see again that the CdA would settle in the middle. Side notes: my position was not fully fixed, hence the big difference in the min CdA out and back.

first-lap.png (149.12 KiB) Viewed 328 times

[Velocomp]

With respect, I think the problem is that the data is not being interpreted correctly.

The whole purpose of the out/back test is to provide enough data to average out second-to-second changes as the route progresses. You seemingly are expecting the data to be exactly the same, with no CdA change during the test. If that were really the case then there would be no need for extended tests or repeated tests.

More data is better--that is the central principle guiding CdA analysis.

So, let's take a look at your data from this perspective. You have 8 laps (good!) where you repeated the same configuration. Here they are (lap number and CdA for the lap):

3 0.208
4 0.214
5 0.212
6 0.214
8 0.219
9 0.211
10 0.209
11 0.207

Lap 8 is a bit of an outlier (that can happen); discard it and the average CdA of these other 7 laps is 0.2107. Call that 0.211. The standard deviation of these 7 laps is 0.003, so you can say with 95% probability that your CdA is 0.211 +/- .003. This result is with profile 3, which does not have quite the precision of profile 4. Profile 4 can reduce the standard deviation...

Outdoor CdA measurement is really difficult because of natural variations in wind, slope and road surfaces. This is why velodromes are used (they mostly eliminate these variables). If you go to other alternatives such as wind tunnels or even other aero sensors in velodromes all of them will give you a number, but none of them will offer this kind of detailed analysis or ability to review and understand. And frankly I'm not sure any of them would deliver you a meaningfully more useful CdA measurement (though they might claim it).

I think you've done a good job, and you did it without the extra precision offered in profile 4. Well done!

[HDPWR]

Yes, I expect only a very minor change in CdA during the 1km leg, where position, power, conditions, etc are all constant. Let's say between 0.2 and 0.205 for example with 3s moving average, but of course the longer the leg, the better it is for calculating an average. However, when I calculate the average then I want to take into consideration the segment only where the conditions were constant, in my case it is the 1km leg and not the entire lap, because I cannot guarantee and control all the factors for the braking and turning around phases. Plus, after more than 30 mins at race pace, it is really hard to fully focus on keeping the same position and timing for the braking and turning.

By the way you just nicely avoided my question regarding Profile 4. Previously you mentioned that in Profile 4 there is access to raw data from FW. Could you say something about that and my assumption about the window of moving averaging in Profile 3? Perhaps in a private message or email?

With respect, I think the problem is that you don't try to understand my case. I am a fairly good amateur racer and would like to fine tune my position on the bike. Just like any professional, I am not expecting huge gains in terms of CdA. Let's say my initial CdA is 0.2, if I can make it to 0.19 then I did an amazing job. At 300W input, that would save almost 15W. So during my test runs, I am not expecting more than a 0.005 difference between the average of two sets of laps with different positions. During the braking and turning, I guess I have a CdA around 0.3, huge difference and it influences the average pretty badly. In my opinion it would be absurdity to expect one to lock in and out from the TT position with utmost precision in order to have the exact same timing and position for braking. Maybe with a 5km long leg it would be acceptable, but there is no such option around my location plus in that case a maximum of 3 laps (3x10km) could be performed on the same day at race pace and well-maintained race position.

Regarding my test, I didn't attach the picture of all laps, here it is:

all-laps.png (36 KiB) Viewed 290 times

As I wrote, I first did 5 laps - lap 2 to lap 6. First 2 with position A, last 3 with a different hand and head position (position B). Averages are 0.2105 vs 0.2137.
And then some rest in lap 7, then 6 another laps - lap 8 to 13. First 3 laps again with position A. Last 3 laps with position B. Averages are 0.213 vs 0.21.
Can you tell me, which position is faster? Based on this data, it would be just guessing, but I am pretty sure there is difference, it is just blurred by the braking the turning.

[HDPWR]

I have just checked the testing process of a competitor, they use exactly the method I proposed:

The Aero Test is recorded on an out-and-back ride, and it is essential to stop recording the test where you began it. Our algorithm will automatically remove the acceleration from the start, the u-turn, and the deceleration at the end. That way, we can analyse identical seperate sections to calculate the calibration factor and CdA.

I still think this is the best way to measure small changes in an out-and-back ride. Please consider implementing something like that.

[Velocomp]

I have just checked the testing process of a competitor, they use exactly the method I proposed:

The Aero Test is recorded on an out-and-back ride, and it is essential to stop recording the test where you began it. Our algorithm will automatically remove the acceleration from the start, the u-turn, and the deceleration at the end. That way, we can analyse identical seperate sections to calculate the calibration factor and CdA.

I still think this is the best way to measure small changes in an out-and-back ride. Please consider implementing something like that.

I am aware of this method. Fine; maybe flaws in their measurement algorithms require this in order for them to get decent results.

We have tested and verified AP methods with over 5 years of development, and we have been making aero sensors since 2008. Firmware changes cost money to implement and more money to test and verify. Changing things requires a justification other than "I think this is the best way".

The most important question is this: what will you do with your CdA tests? The most meaningful way to use them is comparison: ride position, equipment, bike fitting alternatives. If a test method is consistent then it is the DELTA in CdA that is of value, not the "absolute value" of CdA (whatever that means...).

I showed you how to interpret the data from your testing; you got very consistent results. Now you can start comparing things to really improve.

CdA is not like power, where there is an objective physics reference point. CdA is extremely dependent on measurement method and course conditions--these can never be the same except when the same rider is testing on the same course under the same conditions. A CdA of .211 measured in a wind tunnel is not necessarily "better" than a CdA of .215 measured on an outdoor course. In fact, just the opposite may be true...

If you don't like how we do things with AP you are free to spend nearly twice as much (or more) on an alternative. One thing I am certain of is that you won't get usefully better results (or better support ).

[Velocomp]

Yes, I expect only a very minor change in CdA during the 1km leg, where position, power, conditions, etc are all constant. Let's say between 0.2 and 0.205 for example with 3s moving average, but of course the longer the leg, the better it is for calculating an average. However, when I calculate the average then I want to take into consideration the segment only where the conditions were constant, in my case it is the 1km leg and not the entire lap, because I cannot guarantee and control all the factors for the braking and turning around phases. Plus, after more than 30 mins at race pace, it is really hard to fully focus on keeping the same position and timing for the braking and turning.

By the way you just nicely avoided my question regarding Profile 4. Previously you mentioned that in Profile 4 there is access to raw data from FW. Could you say something about that and my assumption about the window of moving averaging in Profile 3? Perhaps in a private message or email?

With respect, I think the problem is that you don't try to understand my case. I am a fairly good amateur racer and would like to fine tune my position on the bike. Just like any professional, I am not expecting huge gains in terms of CdA. Let's say my initial CdA is 0.2, if I can make it to 0.19 then I did an amazing job. At 300W input, that would save almost 15W. So during my test runs, I am not expecting more than a 0.005 difference between the average of two sets of laps with different positions. During the braking and turning, I guess I have a CdA around 0.3, huge difference and it influences the average pretty badly. In my opinion it would be absurdity to expect one to lock in and out from the TT position with utmost precision in order to have the exact same timing and position for braking. Maybe with a 5km long leg it would be acceptable, but there is no such option around my location plus in that case a maximum of 3 laps (3x10km) could be performed on the same day at race pace and well-maintained race position.

Regarding my test, I didn't attach the picture of all laps, here it is:
all-laps.png

As I wrote, I first did 5 laps - lap 2 to lap 6. First 2 with position A, last 3 with a different hand and head position (position B). Averages are 0.2105 vs 0.2137.
And then some rest in lap 7, then 6 another laps - lap 8 to 13. First 3 laps again with position A. Last 3 laps with position B. Averages are 0.213 vs 0.21.
Can you tell me, which position is faster? Based on this data, it would be just guessing, but I am pretty sure there is difference, it is just blurred by the braking the turning.

You are wanting AP to do something it was not designed to do (read the instructions). You are doing lap testing using profile 3 (training/racing) because you don't have a Garmin. If you want to do highest-accuracy CdA testing then you will need to get a Garmin bike computer and use AP in profile 4, with the lap feature. What we do in profile 4 is proprietary and is not something we share.

While it's possible get decent accuracy in profile 3 you will need to do at least 5 laps in succession for each ride configuration.

[HDPWR]

I really appreciate your quick answers!

I totally agree with your sentences but I still have the feeling you could not precisely measure the delta CdA between my position A and B, because the braking and turning phase will severely influence the average.

Anyway, we are speaking about measuring the aero drag of a body (the rider), and you would do it while the form of the body is changing? (Position changes due to braking and turning).

I could totally live with the raw CdA data for out leg and back leg, for example 0.21 and 0.195 for position A. Then maybe for position B it would be 0.208 and 0.193. Of course these values should be averages of multiple repeats.

Btw just one more thing came to my mind: testing helmets. In my test setup with the 1km leg, the braking part is too long and significant, but it could be that helmet A has better CdA during braking while helmet B is better in locked in race position. The current measurement method would make it impossible to give a valid difference.

[Velocomp]

I really appreciate your quick answers!

I totally agree with your sentences but I still have the feeling you could not precisely measure the delta CdA between my position A and B, because the braking and turning phase will severely influence the average.

Anyway, we are speaking about measuring the aero drag of a body (the rider), and you would do it while the form of the body is changing? (Position changes due to braking and turning).

I could totally live with the raw CdA data for out leg and back leg, for example 0.21 and 0.195 for position A. Then maybe for position B it would be 0.208 and 0.193. Of course these values should be averages of multiple repeats.

Btw just one more thing came to my mind: testing helmets. In my test setup with the 1km leg, the braking part is too long and significant, but it could be that helmet A has better CdA during braking while helmet B is better in locked in race position. The current measurement method would make it impossible to give a valid difference.

To repeat, you are asking to do something in profile 3 that AP is not designed to do. Profile 3 is intended for regular training/racing. For what you want to measure you need to make use of the additional procedures accessible in profile 4--which requires a Garmin. By the way: for out/back measurement procedures our competitors also require a Garmin...

I understand your concern about changing body position differences during turnaround and its effect on CdA. As long as the out/back course is significantly longer than the turnaround length (it had better be...), and as long as your test procedures doe not vary between tests, then any CdA difference introduced during turnaround will be relatively minor in comparison to the overall measurement, and any difference, whatever it is, will be the same for each test. The difference is an offset number, which has no net impact when comparing different tests (the difference subtracts out).

[HDPWR]

No, I am asking to do the thing in profile 4. But then you always repeat profile 4 needs a Garmin and marking laps. That means the AP will again do some calculation based on the entire lap with braking and turning making the data unusable for measuring CdA differences between two helmets on a short leg.

As long as the out/back course is significantly longer than the turnaround length (it had better be...), and as long as your test procedures doe not vary between tests, then any CdA difference introduced during turnaround will be relatively minor in comparison to the overall measurement, and any difference, whatever it is, will be the same for each test.

In my opinion it is still not possible. I will again sum up my reasons:

• With a 1km leg the travel time on the leg is 70-90s. If you change to braking position 1s sooner then that will increase the overall CdA by as much as 20s riding in a aerodynamically less efficient helmet. On the other hand, on this length one can maintain the race power for 10-12 total laps that makes it possible to compare 2 or 3 different setups.

• With a 5km leg the braking and turning would have less effect but the rider is not able to do more than 3 total laps in race pace, not enough data for averaging. Plus, ultra hard to find a 5km long straight, flat, good quality bike path or road without traffic.

• Still cannot maintain same braking form and pace, especially when getting tired.

By the way: for out/back measurement procedures our competitors also require a Garmin...

This is what I initally thought but turned out to be not true at least for one competitor:

Yes. xxxxxx can be used without a Garmin device. Recording can then be controlled directly from the xxxxxx using the side button (double click to start / double click to stop recording your ride) or from the xxxxxx App through an iOS device.
An iOS device such as an iPhone or iPad is mandatory to pair your xxxxxx with the ANT+ sensors and access your data stored in the xxxxxx.

[Velocomp]

No, I am asking to do the thing in profile 4. But then you always repeat profile 4 needs a Garmin and marking laps. That means the AP will again do some calculation based on the entire lap with braking and turning making the data unusable for measuring CdA differences between two helmets on a short leg.

As long as the out/back course is significantly longer than the turnaround length (it had better be...), and as long as your test procedures doe not vary between tests, then any CdA difference introduced during turnaround will be relatively minor in comparison to the overall measurement, and any difference, whatever it is, will be the same for each test.

In my opinion it is still not possible. I will again sum up my reasons:

• With a 1km leg the travel time on the leg is 70-90s. If you change to braking position 1s sooner then that will increase the overall CdA by as much as 20s riding in a aerodynamically less efficient helmet. On the other hand, on this length one can maintain the race power for 10-12 total laps that makes it possible to compare 2 or 3 different setups.

• With a 5km leg the braking and turning would have less effect but the rider is not able to do more than 3 total laps in race pace, not enough data for averaging. Plus, ultra hard to find a 5km long straight, flat, good quality bike path or road without traffic.

• Still cannot maintain same braking form and pace, especially when getting tired.

By the way: for out/back measurement procedures our competitors also require a Garmin...

This is what I initally thought but turned out to be not true at least for one competitor:

Yes. xxxxxx can be used without a Garmin device. Recording can then be controlled directly from the Notio using the side button (double click to start / double click to stop recording your ride) or from the Notio App through an iOS device.
An iOS device such as an iPhone or iPad is mandatory to pair your xxxxxx with the ANT+ sensors and access your data stored in the xxxxxx.

It would seem that you think you understand how our system works (or should work) better than we do. I can only suggest that you read our instructions again, which not only describe our CdA measurement procedures but the underlying dynamics of CdA measurement.

Notio is either out of business or going out of business...they have not updated their blog since 2021. And by the way: Notio was a licensee of our technology.

I have nothing further to contribute to this particular thread, so I am going to move on.

[HDPWR]

I have no doubts about your tech, I think it is powerful and accurate for the purpose. My issue is with the firmware and software solutions. My impression is that the Aeropod was tailored for casual cyclists and not for time trialists.

Just to prove my right I show a picture from the competitor with the F1 engineers. I'd like to highlight the Valid speed min, the Valid brake deceleration and the Brake time offset parameters.

Protocol.png (266.15 KiB) Viewed 174 times

In my opinion this protocol with a fixed min speed and required hard braking is not better than what Notio applied. Either pressing the Garmin and cutting off 2-3s after the press of the button or simply select the laps later by GPS in an automatic way seems to be better to me.

[Velocomp]

Out of respect for the sincerity of your postings and your obvious desire to make your points clear, I will try, one more time, to do the same.

First, let me state that I understand your concern about "turnaround" errors. There is no doubt but that CdA measurements can change whenever riding conditions change, such as happens when making the turn at the end of the "out" portion of a ride.

Here are the points I would like to make:

1) Comparative, highest-accuracy CdA testing of different configurations (helmet, ride position etc), should be performed in profile 4 ("CdA testing", as described in our instructions). You are not doing profile 4 testing because you don't have the required Garmin computer, which is needed to mark the start and stop points of each test lap.

For the objective you have, the instructions tell you to use profile 4. Since you do not have the necessary equipment to use profile 4 you are in no position to judge the accuracy of profile 4 measurements, nor the methods used in it. What we do in profile 4 is proprietary, but how it functions is considerably different from profile 3, and the turnaround concerns you have do not apply. To be 100% clear: for the objective you have (highest accuracy comparative CdA) you are using AP in a manner contrary to our instructions. This is a user issue, not a product issue.

2) You are using profile 3 measurements (CdA training/racing) for your testing, because you don't have the Garmin needed to do testing in profile 4. Profile 3 is not our recommended way to do CdA testing with the accuracy you desire, but as your own results have shown you can get decent results even in profile 3, as long as you do a lot of successive runs for the each test configuration.

3) Your concern is that CdA errors in profile 3 might be introduced at the turnaround point. I get it. But the key point is this: any error introduced during turnaround is an "offset" value that adjusts the theoretical CdA measurement up or down by some fixed amount. The offset error, whatever it might be, is not important, because when comparing test configurations, (helmets, ride position, whatever), the thing to compare is the DIFFERENCE between your test configuration results. And when comparing measurement differences, offset errors cancel out.

4) For example, suppose you get a measurement result of 0.212 for helmet "A", and 0.216 for helmet "B". The relevant comparison is that helmet A is better because it has a lower measured CdA value of .004. The offset error, if any, introduced during A/B testing subtracts out when comparing test results.

5) As an example, let's suppose there is a CdA "offset error" due to the turnaround of .005. If you perform the A/B test sequences in the same manner (route, speed, power level), the turnaround offset error of .005 will be the SAME for each test sequence, so when you compare A/B results, any offset error will subtract out!

-------------------------

You have made multiple references to the Aerosensor, alleging its superior test method. It's a nice protocol, but as it so happens I own an Aerosensor and I have used it. There are things Aerosensor does in hardware that necessarily compromises its performance--compromises made because otherwise it would infringe on our patents. We can argue all night about which test protocol is better, but in reality I found nothing about its actual performance that made me envious.

In a previous posting I said I would not be commenting further on this thread. I hope everything I have explained here is clear and helpful. With this post I really do have nothing further to add to the topic.

[HDPWR]

Although we don't agree on the topic but I also truly respect that we can discuss it so openly here on the forum!

1) Short reaction to your arguments: My problem with the turnaround is that I cannot guarantee that I can reproduce them in the same way. In my opinion biggest problem is the timing: If in one lap I change the position 1s later then I spend 1s more in a 0.21 CdA position and 1s less in a 0.3 CdA position. Same for acceleration, if I start accelerating in the 0.3 CdA position instead of the other. And yes I understand if I would do 100 laps then the offset caused by these uncertainties would average up nicely, but if I do only 3-5 laps it will not be enough for averaging.

2) Your confidence makes me doubt myself. Can you tell me what I will get if I use profile 4 with a Garmin but create laps for the 1km legs? So a lap would not start and end at the same location. I assume I would get a bit different CdA values for the out and back legs, for example 0.19 out and 0.22 for the back. That would be fine for me, I can manually calculate the average of them and compare. I already checked profile 4 and noticed that it gives back a cumulative CdA value, as described in the manuals.

[lorduintah]

If the turn around portion of any "lap" is a very small percentage of the remainder - any error introduced by that portion (from variability) should ultimately contribute a very small deviation to the lap average. The same could be said for the time taken to attain the riding position (or equipment) one is attempting to replicate or differentiate from lap to lap. A baseline lap, repeated several time might allow you to estimate the variance from the turn around portion of your trials. (See my comments related to DOE, below.)

Even doing a normal ride, one cannot expect to attain the exact reproduction of all but one element (I think the same can be said if Profile #4 is the test mode.) So it would seem that you might attempt to extend the evaluation portion of any lap such that the prep portion of any lap is markedly small = even if that portion is not reproduced exactly, That small deviation from ride to ride (in terms of percent deviation from the overall test) will probably average out to an offset that ideally will be consistent over all your test rides and effectively cancel out for your comparisons of each test variation.

If you want to consider that the preliminary setup portion of each test is another variable, you could design that as another variable in your design and consider several levels of conditions for that as part of the testing.

Are you familiar with the process of Design of Experiment? I seems to me that you are trying to invoke portions of that in your experimenting. I also appreciate that by suggesting DOE, the whole sequence likely becomes more complex.

[Velocomp]

Although we don't agree on the topic but I also truly respect that we can discuss it so openly here on the forum!

1) Short reaction to your arguments: My problem with the turnaround is that I cannot guarantee that I can reproduce them in the same way. In my opinion biggest problem is the timing: If in one lap I change the position 1s later then I spend 1s more in a 0.21 CdA position and 1s less in a 0.3 CdA position. Same for acceleration, if I start accelerating in the 0.3 CdA position instead of the other. And yes I understand if I would do 100 laps then the offset caused by these uncertainties would average up nicely, but if I do only 3-5 laps it will not be enough for averaging.

2) Your confidence makes me doubt myself. Can you tell me what I will get if I use profile 4 with a Garmin but create laps for the 1km legs? So a lap would not start and end at the same location. I assume I would get a bit different CdA values for the out and back legs, for example 0.19 out and 0.22 for the back. That would be fine for me, I can manually calculate the average of them and compare. I already checked profile 4 and noticed that it gives back a cumulative CdA value, as described in the manuals.

CdA testing requires the completion of an ENTIRE out/back closed loop; interim measurements (CdA out and CdA back) are not meaningful individually. Profile 4 is the best way to do closed loop testing; you can potentially get good results from profile 3 if you do multiple laps, but profile 4 will always be better.

I repeat, I repeat: you are using profile 3 in a manner for which it is not designed. Why do you continue to make theoretical objections about it when, for the things you are trying to accomplish, you are using AP improperly?? I have zero interest or patience in responding to your theories, because they have no relevance in profile 4 measurements.

Slight differences in speed and acceleration during turnaround are insignificant; if an out and back course is 3KM total (the short end of what we recommend), at 30 km/h that is about a 3600 second test time. A difference of one second during turnaround times is a potential measurement error of only 1/3600 =0.0278%! The same observation applies to lap location; it makes no practical difference if the start/stop points are different by a few meters. And if you read the instructions carefully, you'll find that you are supposed to avoid accelerating during testing.

AP doesn't require you to do manual averaging of out and back CdA (unlike some competitors); the profile 4 CdA number you get at the end of each profile 4 closed loop is the result for the entire test.

It would be really, really good if you re-read the instructions and did the testing just as it is described. It would be even better if you borrowed a Garmin, so you could use profile 4 lap testing. You are way, way overthinking this and that is not a good use of anyone's time.