Metrics on new AI FTP Detection + TrainerRoad AI

Announcements
92

Another problem with calling it ‘FTP’, but instead being vaguely tied to ability to complete a specific workout level.
If someone is using ‘FTP’ detection and self selecting workouts, you’ve now muddied the waters about their ‘FTP’ and the workouts they can complete.

What if with their ‘real FTP’ they can compete the workouts they want, but now their ‘TRFTP’ pushes their “FTP” too high, or too low because they’re good at a specific type of workout?

Maybe this approach produces better results if you let AI take the wheel, but if you’re just using it as a guideline rather than doing ramp/ftp tests, its useless for that now.

93

If you wanna regularly do a ~60min TT to use to guide your FTP, more power to you.

The issue with that is the vast majority of riders are unwilling to do that. It’s like your doctor telling you that self-administered intra-ocular injections of meds are the best way to treat high blood pressure. :wink: Any method to predict training zones needs to both be precise, and also palatable to the average rider so that they actually do it

It also doesn’t work great for me personally. If I used 1h power for my FTP last year, then I had multiple weeks in a row where I did over 8 hours a week of TiZ above 0.85, which doesn’t make sense lol.

94

But you don’t have to. You can regularly do some max efforts then use your PD curve and smooth it with a math formula and look at the inflection point for your FTP. WKO, Intervals.icu, etc will do this for you.

95

I mean that’s not unusual. if your tte is shorter than 60m then attempting to use a 60m effort won’t be accurate, there is a reason it’s a range and not a set duration.

96

>I mean that’s not unusual. if your tte is shorter than 60m then attempting to use a 60m effort won’t be accurate, there is a reason it’s a range and not a set duration.

You and I know this. But it means we’re not just blindly following a number - we’re applying judgement we’ve learned from experience to tweak it. That’s how “FTP” is supposed to be used.

But it also means that “FTP” isn’t a consistent thing between people if we all need to tweak it differently to get it to work properly to guide our training.

I’d also suggest that means it’s an imperfect method alone to guide training. It’s part of the puzzle, but it’s hard because the other pieces are more subjective and require training experience.

Imagine you’re a cyclist who’s new to all this. You have no experience of what things are supposed to feel like to weigh this all against. People are using different tests, different definitions, arguing about abbreviations you’ve never heard of before. It can feel overwhelming. If there was an easier way for them to set the appropriate training zones, that would probably be really helpful.

97

I’ve tried numerous iterations of the plans now. Most recently, I got given 6 minute sweet spot intervals for my first workout. This is infuriating. Six minutes is not a sweet spot duration. This whole system is let down based on the fact that there are workouts in the catalogue that shouldn’t really be there.

I will be very clear. If you cannot do at least 3 x 10 min at Threshold, it is not your threshold. Calibrating other zones off this false assumption is broken.

98

I mean the issue here is that a “Level 3” threshold workout is unlikely to give you any threshold adaptations. If you can’t do a level 4, it’s straight up not your threshold.

Unless, of course, it’s overestimated your FTP to an extent that you’re actually doing over/unders with no under. Or even worse, long VO2 workouts.

99

If they can’t get past 15 mins of the first interval then it’s not their FTP. This is why longer form tests are a must. They allow you to calibrate your RPE.

100

Yeah. My concern is that every 4 weeks after AI FTP I will get set back to level 3 Threshold whereas maybe I would have wanted to progress further. But I’ll just see what happens and maybe I’ll just manually pick alternative workouts with longer intervals. And I think @Nate_Pearson mentioned that AI FTP prediction date will be movable…so I’ll just move it forward and keep progressing levels on the old AI FTP…guess that would be my ideal solution.

101

I’m concerned about that too. Looking forward to seeing what happens on Feb 6th.

102

You are describing my first workout today with my new inflated FTP (at least 15 watts over any reasonable physiological FTP). It was doable (but very hard) since it was only 3x9’ with the “overs” only being 1’, but the 2’ unders were not unders. Last week was proper over/unders and 4x12’, so much more time in zone, but at much lower wattages (but still very hard). I guess today’s workout was still productive, but I just started base and it’s just counter-intuitive to be doing shorter and more intense intervals pushing into vo2max. I didn’t really look for alternates, maybe I could have found something longer with lower wattage, but the over/under workouts are all kind of dependent on a somewhat accurate FTP sitting between the unders and the overs. For sweet spot, it’s not really an issue because I can find an appropriate duration and wattage that makes sense. I’m gonna keep playing with the system with my new FTP in place for a bit, but I’m struggling with it mentally. If there is some good news, my future FTP is predicted to remain unchanged, so maybe I’d grow into it. But the current value is higher than I’ve ever seen and I’m not getting any younger at 57.

103

It will be interesting to see what happens with that. I also wonder whether changing the Training Approach for Threshold would impact how it will proceed. (ie: to Demanding or above)

104

Just out of curiosity…how exactly is someone “put around level 3”? My speculation is that FTP is adjusted up / down until some threshold level 3 reference workouts result in a difficulty prediction of 50% hard / 50% very hard. But my thinking could be completely wrong, just some speculation :wink:

105

I have just finished a really good training block where I was loosely following the “old” TR 4 day a week plan, but adding extra endurance and/or changing some workouts to be ones I like better (not changing the intensity). I also did a 20 minute FTP test at the start of this block.

The new AI bumped my FTP by 4.5%, which seems reasonable to me as I was not finding the hard workouts hard enough (RPE slightly challenging). The block itself was hard overall.

So far liking what I see from the new AI Plans.

106

I think there is going to be a big decoupling between the label (SS, VO2, Threshold, etc) and the stimuli the workout elicits. I think that is because the threshold is being normed to PL 3 workouts, which for my money are not all that hard by a conventional understanding of threshold work.

I worry the system will be confusing a workout’s feasibility for me and it appropriateness for my season long plans. As @BCM said, this reminds of the pile on the intensity days of yore.

I know all the workout descriptions are going to read a touch off moving forward.

107

It might totally be the fact that the range is throwing things off. As Nate said it was seeing those as failures because of the power reductions.

As for using both a human coach and AI simultaneously being a disaster, I agree. Which is why I only use TR to track my stuff with the yellow zone/red zone fatigue and for the indoor workouts. I don’t use it for coaching me.

108

The issue is that TR has always had a problem with what good FTP/TTE workouts should look like, in a bid for:

  1. Variation. No one wants to do Starr King and Gray every week.
  2. Giving people the progression they crave. From prioritising the increase in the FTP number with little regard for the equally important TTE value, they are consistently pushing people into ridiculous 6 x 8min “threshold” workouts, which if not applied correctly will either give you no adaptations or burn you out.

If you try and apply a conventional understanding of physiology to these new “levels” it’ll do your head in.

Everyone, and I mean everyone, should listen to Empirical Cycling on the matter, specifically this podcast. Then come back and see how close your AI prescribed block adheres to the science.

109

@Nate_Pearson As a 10 year TR user, I am normally very comfortable trusting the process & getting on with the work.
However, I’m concerned about my very large (-28W) FTP drop upon entering the new AI program, which has taken me from 284W down to 256W. This is by far the lowest FTP I’ve seen for almost 5 years. I normally sit between 275-285W, all previously given to me by TR.

After a 3 month training block which had taken me from 276W to 284W (self prescribed workouts as I wanted a mental break from training plans), I had 2 week holiday, during which I didn’t ride but continued to run. I would have expected a 5W decrease, maybe a little more, but to get a 28W decrease feels highly unusual.

I’m concerned about the size of the decrease but also the implications for getting the right training stimulus when riding outdoors based on time in zone, and how to calculate the correct zones based on the FTP I enter into my Garmin.

As an example, after most group rides, I ride for an extra 60-90 mins at Zone 2. At 284W that works out to be approx 156W-213W, but with my 256W FTP it’s approx 140W-192W. Am I really going to get enough training stimulus at this reduced wattage, to make for effective and time-efficient adaptations?

The same would be true for when I add Threshold blocks when riding outside. Normally I’d sit around threshold for 6-7 minutes sections until I run out of road and hit a junction. If I rode today at 284W, or slightly lower due to 2 weeks off, I’d now firmly be sitting in VO2Max of my new FTP for the whole duration. Or if I ride at my new 256W FTP, I’d be only be riding at what 2 weeks ago was considered Sweetspot. Is this really going to provide enough time at Threshold in either scenario?

Having been 100% comfortable with TR plans, a massive fan of the workouts and the benefits I’ve seen over the years, this change has really thrown a lot of doubt onto how I’m going to apply my new FTP in practice on the road.

Can you provide any insights into my dilemma please. Many thanks, Andy

110

My personal experience with the new AIFTP has been quite shaky. At first it gave me a 10% bump after the holidays. No way that’s real, but it’s beta, so I gave it a shot. I figured, if I ever did a workout at that intensity, it’d be a 30 minute anaerobic or something. AI gave me Berge 2.8: 2×90s 2×60s 2×30s, quick and dirty. Forced me into 42 all time PRs and I failed the 1-minute efforts at the 40s mark, both of them. Marked max effort. Next it gave me Carson 3.6. Threshold work masked as SweetSpot. At the end of the day, it’s a sub 60 minute workout and I’m in decent shape, I completed it, but it felt way harder than sweetspot should and I couldn’t recover. Marked Very hard (historically I mark SS moderate or hard). The AI saw that and gave me a yellow day which I ignored (that’s on me), because I had scheduled two days in a row and there was no way around it. Still, it knew in advance, it shouldn’t have burned me out on the first day of a back-to-back. But it did and I failed my following workout, a 90-minute SS lv 3.2. After a decent amount of complaining I got my FTP back to where it used to be and was prescribed a 90 min lvl 3 workout with some timewasting in it (endurance). Good as I needed to recover from the previous three. FFWD a week of productive training and I raise my FTP manually by 2.3%, because I start catching up on the gains I had right before the holidays (I was due a detection and expecting an increase). Next day the new AI goes live, triggering a new detection. Took it as FTP prediction would be otherwise disabled and I wanted to test out the feature (big thumbs down for TR here for releasing it less than 28 days after it was in beta, so essentially without any confirmation of how reliable it is at scale!!) Again, a bump, but more moderate at 3.7% (on top of the 2.3% I had given myself manually). Still enough to turn the high end of my Sweetspot into threshold though, and it did. My SS workout yesterday gave me a 20 minute low threshold interval (masked as sweetspot). Of course, I could complete it, and it’s not even very hard, because 20 minutes at threshold should be within a cyclist’s capabilities, but it’s not sweetspot. I had 60 all time PRs - again the indicator that AI is pushing the intensity.

That’s all less than a month within the new season. My A race is in June, I’m not sure I want to do threshold a few times a week starting now. At least the AI gets it. It doesn’t know about the FTP reported to the user and considers watts. It knows that something’s off and tries to cover up for it, systematically avoiding SS workouts at higher % of FTP if I go to alternates or try selecting something manually, also giving me my next threshold workout entirely below my FTP :smiley: I mean, that’s not what the number is, I get it. The AI gets it. My OUs are prescribed at 95/99%. 97% of my AIFTP, which somehow equals exactly my manually entered value before AIFTP detection, sits right in the middle. So as it stands, I’m better at detecting my FTP than the AIFTP detection, which means it’s a non-feature, yet TR promotes it. In fact, it was part of the reasoning behind last year’s price adjustments, so it is indeed considered a valuable feature.

My point is, we should just call it (AIFTP) something different already. TR does not exist in isolation and multiple people in the beta mentioned the ripple effects on IF and so on. Quote from the TR career page: “ © 2011-2026 TrainerRoad, LLC. Ver. 30846 — NP, IF and TSS are trademarks of Peaksware, LLC and are used with permission. “ So yes, the platform knows about the outside world. It uses these metrics, officially so, and has articles about them and how they’re related to FTP, like this one: Functional Threshold Power: What FTP Means to Cyclists - TrainerRoad . Note that pacing is also mentioned here, so people have full right to pose the question “How to pace from now on?” when their AIFTP suddenly becomes unreliable outside a very specific context.

111

“Levels’“ were introduced a while back as a relative measure of difficulty within a particular “zone” or workout type. In principle i think it acknowledges that FTP is a “point” estimate and two people with the same FTP might have quite different true zonings relative to FTP and different “density”. For example athlete A and B might both have a FTP of 300 watts based on some kind of FTP test, but one athlete is able to do 6 x 3 min V02 max at 125%, whereas another might struggle to do 115%. Or a different example, one athlete might be able to do a threshold workout of 3 x 20 min at 95% of FTP, whereas another might only be able to do 3 x 10 FTP comfortably.

Workout levels was a scale that tried to allow for the issue that FTP is fairly one dimnsional, but in reality each athlete has a multi-dimensional profile

It was used for picking workouts previously. Now it seems it is only used as a “translation” layer of what the AI model- which is actually looking at relationships betwen underlying power, duration, HR etc.