Hi TrainerRoad coaches—I’m a subscriber and would appreciate a review of my TrainerRoad training calendar/log. My TrainerRoad username is ssiegal. Since the AI FTP has been implemented I have noticed a consistent issue that I just don’t understand.
At the beginning of each block/month of a training plan, the AI FTP analyzes and predicts some sort of FTP increase (current build block started with an increase prediction of something like 8% or so). Then I complete each workout without failure, appropriately rate the RPE but after nearly each workout the predicted AI FTP goes down and down and down until by the end of the block the actual increase is nearly 0% (maybe 1-2 watt increase).
What I dont understand is the following:
Why does the predicted AI FTP go down after each workout despite finishing and having a properly rated RPE?
On the contrary, if the AI truly thinks the FTP should go down over the course of the block, why does the block being with such a high increase predicted? That would argue the algorithm does a poor job of predicting.
I just dont get how my FTP would reliably go up after training programs prior to AI, and now it just stalls out.
Just because you are rating a workout properly, it doesn’t mean it might not ding you.
For example, if you complete a vo2 max workout that AI estimated you should consider it “hard” but you find it to be “very hard” or more, it’s likely going to lower your predicted FTP over time. Same with endurance workouts if you rate them moderate or hard. I believe it also factors in your heart rate during the evaluation.
You should continue to rate them properly, but as an example you could play with a bunch of earlier workouts in a block and change them easier and I’d bet your predicted FTP would go up. I definitely wouldn’t suggest actually gaming the system like that, but just an example of how that may affect it.
Admittedly, this is also just all assumptions since you brought up RPE multiple times. Maybe you are rating it what it predicts (or easier).
Most likely reasons without being able to see your data (because your account is private) …
You’re rating the RPE appropriately, but rating workouts harder than the AI thought you would
The AI is dialling down the difficulty of your workouts across each 4 week cycle, partly because of Point 1, and/or because it’s (mis?)managing your fatigue.
I found success when I changed how I rate workouts. Not to game the system but realising that the workout RPE is not a linear scale (you see this in Garmin equivalent RPE rating).
The scale is 1-5 and the granularity between 3-5 is very important. You can think of it on a linear 1-10 scale being:
Easy becomes 1-4
Moderate becomes 5-6
Hard becomes 7-8
Very hard becomes 9
All out becomes 10
What tweaked for me was seeing that difference from Garmin RPE and TR RPE and also reading the guidance a lot and thinking about how I could or not complete parts of the workout.
A TR 5 all out should be when you basically fail the workout but just about scrape through
A TR 4 very hard means you could not, at all, under any circumstance squeeze out another interval (VO2 level or a short threshold over or under similar)
When you think about it, these are very hard to recover from and you shouldn’t be doing these efforts much. That’s why the AI drops your prediction.
Equally if you were able to do another interval and rated it as hard then that’s having some left in the tank ( but not too much) and that’s what the AI is optimising for
I agree and understand your point. However, I do not think the FTP adjusts appropriately. For example I may rate something 4-6 and it still may lower my FTP. I almost never rate something a 9-10 and it’ll still lower FTP. And some de load or base rides I’ll give it a 1-3 and it won’t increase. So I dont think RPE adjust appropriately either
I switched to public if you are willing to take a look. to point #1, I knew this argument and have previously tried to rate a slightly lower RPE and it neither increased FTP prediction nor seemed to make the adjusted future workout harder. To point 2, I dont get why it would dial down when I rate something in the 5-7 range. and Why would it BOTH dial down the difficulty and the FTP prediction? That just seems to point that this new AI FTP in TR isn’t really FTP at all and really just a training difficulty number
I don’t do any indoor workouts during the spring/summer/fall and I’ve seen the exact same as you with my FTP. Changing my plan to indoor then switching them to outdoor workouts might have helped a bit. I think making sure to only associate rides with the workout when I really can exactly mirror the workout has help too. Case in point, today I did an endurance ride that didn’t exactly follow the workout, so I kept it separate and rated it on its own.
I see from your calendar that you use an HR meter. I wonder if those those of us having trouble are getting HR readings that the AI doesn’t expect for some reason. Maybe @eddie could comment on that?
100% agree. In fact I have read the guidance from TR about their RPE scale and think about it every single time before I click the button to rate it. But I dont think it translates well at all for me. I wonder, however, I do my TR workouts in Zwift where their scale is 1-10 which I still try to be in the TR mindset when I click it in Zwift.
My initial observations looking at your calendar is that you might not be as consistent as you think and that you seem to have ended up in a situation where you fight the Ai.
Looking at your previous weeks there are a lot of weeks with a key workout missing or cut short. There has also been some travel that seldom helps.
Looking at your upcoming weeks your prompted to adapt to longer workouts since the Ai has trouble finding suitable workouts at one hour. This then leads to yellow days that in turn makes your weekend endurance rides easier. I don’t know it it’s the case for you, but in previous cases where the Ai has kept the FTP (too) low, and subsequently run out of suitable workouts, trying to game the system with the workout ratings was a big part of the explanation.
Right now, immediately, and without delay, take a screenshot of all the rides that are scheduled between now and Sunday 6th when your prediction ends and your next detection is due, making a note of the predicted RPE for each ride.
Follow the plan until Sunday 6th.
At that point you’ll be able to look back and say “Well, if I’d followed everything to the letter I should be at 270w. If I’m not, let’s compare what I actually ended up doing to what I actually did to try to explain where the watts went.”
The problem with having plans that are dynamic - as TR’s are - is that it makes variance analysis pretty much impossible unless you create your own baseline as I’ve suggested above.
