Ill take that as a no, you have not used aiftp and have not been a paying member since it was released.
The workarounds (ftp test) are obvious, people here are not asking for the workarounds, but sharing an odd experience and trying to understand the algoritm.
Its a shame the observations cant be discussed due to the disruptive actions of a select few.
I never said it was fundamentally broken, nor have i called them disingenous. They are the opposite actually, and i also believe in AIFTP, and i have clearly stated that here. Im a big belidve of the platform. Why dont you show the quotes where i said that, making stuff up is not cool.
As for your data argument, you are making claims, minimizing peoples observations and trying to marginalize it. These people show THEiR data, which is not enough for you. The large data is guarded by TR and you know this, so you ask for it to prove a point.
Anecdotal data can be very powerful. Smoking and cancer was anecdotal data for a long time, until people started connecting the dots. Yes 1 case doesnt pove smoking causes cancer. Many cases warrant an investigation, and we all knew how that turned out.
Now you will probably read this again as that im tryi g to say there is a problem with AIFTP, like smoking… there may not be… however completing a build phase to the T wtih a 2.5% decline in ftp, is certainly worth a look in the algorithm. “Just do a ramp test” or “show me more data you have platued mr pogachar” are not the right responses.
Right, that is called an anecdote, not data. I have never once dismissed anyone’s individual issues with TR. as I have clearly stated multiple times, there will be issues with any product. No product will be 100% successful for all people. That does not mean there is an issue with product itself.
You said there was a “clear trend”, not me. But when asked to substantiate that claim, you continue to deflect and try to shift the burden to me. Asking for others to substantiate their claims is not making a claim, it is exposing the weakness in their thesis.
Moving on now.
The clear data is in this thread. Different users reporting similar problems, and given not anyone is on this forum, this could be the tip of the iceberg.
I can agree to moving on and agreeing to disagree. Although i still like to see the quotes for the false words put in my mouth.
There is no algorithm. If there was it would just be straight forward decades old style coding. It’s machine learning which has just seen lots of examples and makes a statistical inference.
Machine learning is an algorithm, can it can be tweaked and various parameters can be emphasized and fudged to achieve certain outcomes.
Nope it’s really not.
It really is. Its an alorithm, but not a simple linear equation.
Let me understand, you’re saying that TR had a previous version of AIFTP that worked really well, but recently, something might’ve changed, and this change is causing issues for some users, is that the discussion?
There have been increasing reports of people that complete their structured training programs to the T and that than to a surprise see a (very minor) decline in their AIftp. Sometimes these people have been on the platform for a while and historically seen a small bump while doing this kind of work, so people have been sharing those experiences. Then there are other confounding discussions within this thread how it shouldnt matter, that you should ramptest or that fitness is improving and it may not be reflected in ftp, and people sticking to their guns. That last bucket is less helpful as it is near impossible now to get a clear picture of actual “issue” with all the other noise
Just for clarity, most of us recommending people do a real ftp test if they don’t trust AIFTP are not recommending a Ramp Test. Ramp Tests have their own set of issues/biases. If you’re going to do a test, do a long form test, like the Kolie Moore one mentioned above.
I’m not arguing with your personal experience btw. Personally, I gave up on AIFTP a long time ago because it did not work well for me. I THINK you’re saying it DID work well for you in the past, but something has changed and you feel that change may have created a new issue for you and a few others who’ve posted about it. The suggestion from people saying to do a long form test isn’t meant to argue that your personal current issue is invalid. It’s just saying that we see a few people posting about issues with AI generated FTP estimates, and the simple solution is to do a long form test. Sure, we’d all like to know what TR has changed (if anything), but the easiest solution is to just do a real test. That’s all. It’s not meant to invalidate your personal experience. It’s just trying to give you a simple solution to move forward with your training progression if you’re concerned with the AIFTP estimate.
I hope you realize this is a public forum and not limited to only currently subscribed TR users, and almost definitely better off for it ;).
My suggestion was never meant as a permanent workaround but as a way to verify the AI result and have some actual data on its accuracy for you.
I was more than happy to not post in this thread anymore as I had already given my input but you couldn’t resist mentioning my username while for some reason also not tagging me.
So I will re iterate, I think you are better off working directly with TR if you are not doing so already in this case. The user base does not know if there was an update to how the AI determines FTP or not and if so what changed. TR does and ideally should be able to help yourself and any other affected users get a result that they are happy with.
I see and respect your opinion.
If the point is that "I’m completing my workouts but my FPT went down: (the topic), we have to understand that FTP is a rough number. There’s a lot of variation due to the human nature, environment and technical issues (1-3% margin of error of a PM).
Then, the way FTP is measured is another huge discussion, even the creators of it don’t fully state that testing 20min 95% minus 5min all out to deplete your anaerobic system will give you the gold standard. WKO5 uses mFTP, which for me is a fancy Critical Power model, and if you take the time to watch one of Kusik’s webinars, you’ll see that you have to test every other week - 5sec, then 1 min, 12 min, and so on - in order to have the best ESTIMATE of your FTP (mFTP?)
TR then offers AIFTP, which works reasonably well for most people, especially if you provide good data. With that in mind, we know that FTP isn’t accurate on the dot by any means.
Completing workouts won’t increase your fitness - fitness, not FTP - alone. A few points to understand: Is the person (people?) complaining eating well? Sleeping well? Doing the basic housekeeping? Stress level? It’s hard to tell, but it’s easy to blame the software/product/coach.
I had a coach last year and completed most of my workouts. My FTP went up. However, I burned out so badly that I had to pull the plug on my “A” race. That meant my FTP was roughly corrected assed, but my performance was not even close.
Was it his fault? I don’t think so. I realized later that I was underfueled.
To wrap this up, stating that “a lot of people” are reporting this issue of completing workouts and getting lower FTP numbers is a bit intangible. We don’t know the numbers of TR users to statistically define a sample and then test this sample to check how many people, statistically, aren’t getting the results they expect.
It may have been mentioned here, but you could have increased your TTE (time to exhaustion) based on the workouts without necessarily any change in FTP. A tiny decrease (or increase) in FTP is not really that big of deal since FTP isn’t an exact number and can vary day to day based on sleep, stress, fueling, etc. The AI is likely no exception as it uses your performance on the bike to estimate your FTP, which is likely to be impacted by the factors listed. What TrainerRoad does not provide is an estimated TTE number to accompany the estimated FTP, which would be beneficial if they added that feature. If your FTP remains the same over a month-long training block, but you improve your TTE from, say, 35 minutes to 50 minutes, then you are much more fit than you were. Resilience is another measurement that is highly relevant to race results, but it is not measured on TrainerRoad either.
Just got a reply from support. There must be a bug in aiFTP detection. They got back to me and increased my FTP because the system “ran into a snag”. Glad that they addressed the issue for me and will be interesting to see what happens next month with aiFTP.
My riding buddy was stuck at 245 for a solid year until we "gamed’ AI FTP by doing some way harder than prescribed rides. It seemingly woke things up and now he’s up 10 watts in the last 2 months and his workouts are challenging again.
*this is not advice.
Good to hear! A simple question with a straight forward answer. Sometimes intiutions are correct.
Not either
I am with you… just test. But alternatively they can always just add 10 or 20 or whatever watts they “think” their FTP is and then see how it goes. AIFTP may not be perfect, but probably more accurate than their “think” though.
How about “TR^2” for “TrainerRoad Training Number” 
It is an algorithm, and trainerroad calls it that way.
Workout Levels – TrainerRoad
