I took a break from training (moving, getting married, and a bad run of getting sick over and over).
I had an AI FTP detection right away, but it was super low (150W). I manually set it (170W and later 180W) and trained against how I felt. I’m currently feeling like 180W is actually about 10W low, but I want to be consistent for a while and have left it at 180W.
None of the AI prediction stuff is working currently when I use a manual value because it seems to want to force me to use its predicted value. When I accept the 150W number, it predicts I’ll gain 47W in 28 days!! lol. To me, this looks like the AI knows that 150W is a ridiculous number but some other condition in the code somewhere is overriding the number the AI is coming up with and using some baseline number.
I’m concerned because at my current workout levels, a 150W FTP (if I just keep it), will mean weeks of basically Z1 “sweet spot” and Z2 “threshold/vo2 max” workouts with no guarantee that the next FTP detect wont just sit on 150W again.
It seems to be ignoring all of my recent workouts, (done with a 180W FTP), successfully completed, rated RPE, and with HR data that fits that level.
Would a ramp test work for overriding the AI FTP Detect and hopefully set me where I should be again?
The AI FTP projection is an output of the AI model. I don’t think that over-riding the FTP will change the prescribed workouts. Similarly, doing a ramp test and accepting the new FTP in itself wouldn’t change the behavior. However, the workout data from the ramp test will be an input to the model and that could change things. Of course, an FTP test is not a prescribed workout so that has an impact too. The AI model is a huge amount of black box maths, and I’m not sure that anyone can tell you exactly what will happen until you try …
For myself, I took a ramp test and the number was 2% down from my AI FTP prediction from a couple of day’s prior. Accepting that reduced FTP number did not change the next AI FTP prediction, though the predicted percentage gain did go up because of coming from a lower base. This is during a base phase of my current training plan.
If I was in your shoes and unsure of my current FTP after a break, I would just do the ramp test. How else are you really going to know for sure how much you have improved?
Update: Not sure if somebody at TR did anything, but a new AI FTP detect suddenly became available today. I ran it and got a number that’s a lot more useful.
Guessing that rippled through workout levels, etc. The current workout plan in context of the new FTP makes a lot more sense now.
The AI prediction stuff is also working again and life is good.
Yeah, you must be right that workouts will change.
If you change your FTP then any workout will get scaled by your FTP and will become a different load. I understand that the FTP is not an input to AI TR and that it recommends a workout based on absolute Watts. I suppose it must map that recommendation into the closest workout given your FTP - this is where absolute Watts from the AI gets maps into FTP-relative workouts. So a workout using the old FTP gets changed to a new workout using the new FTP such that the absolute Watts are the same. Other details may change given the finite nature of the workout library.
The workout name changes but the workout Watts and workout intent should be similar. At least, that’s my guess on what would happen given the statement from TR that they don’t use the FTP number. Of course the AI model is a huge pile of maths and weights that nobody really understands so YMMV.
Same seemed to happen to me. I first my FTP was detected at least 10% low but I just bumped up the percent of work out and it was still easy. But after a 7 workout of doing that and reaching the new AI FTP detection it is up 17% by following the workouts, which are easier than my normal load. I am sure that is helping and being more structured but i was already ‘well trained’. I am sticking with my plan but it is scary to think if it doesn’t ‘work’ my A goal of Nationals is compromised. Fingers crossed for the TR method and Ai.