FTP Detection Way Out - What Did I Do Wrong?

Hi Guys,

I think I may have broken the system…

Quick history:

FTP usually around 275-280.

I got ill.

Month ish off.

TR Ramp Test - 253 watts. I get this this is fine.

I stick to my program and get some good progression levels.

3 weeks later - FTP detection says it should now be 239 watts.

Day later, I PB at a time trial with NP of 278 watts which would give me an FTP of around 265 watts.

Now either I broke the system, or I did something wrong.

Would be good to try and workout why my predicted is so low.



what was the avg power and actual duration?

1 Like

100% this


Average power 278. NP 278.

23 minutes and a few seconds.

Flat ish course.

VI 1.0

IF 1.1

I’ve been training a fair few years and the 265 feels about right.

And 265 watts would align with the TR sessions I’ve been doing and the small increase from 253 watts 3 weeks ago.

  • Please contact support@trainerroad.com for issues like this. We have so little info compared to the mass of data they have. Not to mention that we can only guess as to the “secret sauce” that hides within the AIFTPD feature.

Thanks Chad - thought I’d start here in case it was more common or I had done something stupid, but I guess one of the clever techies needs a quick look at my data and see what’s what.


1 Like

Yup, as you’ve seen, even with what you provided, we are guessing at best compared to the background TR can access in seconds. Far better to get their feedback compared to our perspective which is barely us peaking through the keyhole of the door to your fitness. :smiley:

My guess is that AT FTP prediction is based on much more than just a recent PD curve. I have been seeing what seem to me to be reasonable results from the other direction - even having not done any maximal efforts for quite some time, it still has a reasonable estimate, whereas intervals.icu was way low due to a PD curve that doesn’t match my actual fitness. Some recent maximal efforts were in-line with TR’s estimate.
I would expect/hope that the TR estimate would ‘catch up’ if you continue doing work at a 265ish FTP. I guess they don’t have any simple checks on the estimates to make sure they are not way below what a PD and/or 20ish minute effort gives. I’d bring this up with TR support, as this does sound like a case they would be interested in looking at, as the estimate seems way off.

265 might be right but 95% of this 20ish min effort is not the same as 95% of a 20 min test with a 5 min clearance max effort prior.


Thanks for closing the other thread Chad.

Will update once I hear back from the TR guys.

I’m pretty well versed in the 20 minute test - I did a comparison of the 3 minute FTP test v’s the 20 minute test for my degree dissertation.

I’m also pretty in tune with 10 mile TT’s and how fast I go in relation to my approximate FTP.



How does your indoor environment compare to outdoors? If you are overheating indoors its maybe limiting your potential. If so with that potentially and +/- error, motivation, an anaerobic contribution (FTP should be with that exhausted), etc 9.8% isn’t really that far off.

Sure thing. Let us know what they say. Keep in mind this is still a “new and unreleased” feature that is like a beta test. So reporting anything odd or out of the expected range directly to support is worthwhile so they can look to improve this tool for you and anyone in a similar use case to yours.

Pretty close - I have a big fan behind me and built in fans on the Tacx bike.

Historically always tend to go better indoors!

239 to 265 is a big difference in training intensity going forwards.

For now, going to stick with 265 and see how I get on.


1 Like

Yip see how you get on. The majority of folk go better out than in but I don’t either; I think mainly because I can psychologically push right to edge (well less so now I’m older) inside when I’m not worrying about other people, bends etc. I’ve had similar one offs (if it was one) and it may have been too high with a lot of the other possibilities for discrepancy (+/- differences, motivation, anaerobic pollution etc) and whilst in the past that could have been potentially damaging to my training I think it is less so to you with Adaptive Training. Good luck :+1:

Hey @PTC

I looked at your account and based on the data you gave it, it did a good job. You told us 8 min intervals between 223-238 were “hard”, and really didn’t have any NP rides greater than 208 np.

Then you ran AI FTP the day before you did that 20 min effort. Then that effort was WAY outside what you’ve done recently and how you marked your RPE.

I don’t think any human looking at your workouts between Ramp Test and your 20 min effort would have guessed you’d do 278 np for 20 mins!

Like you said “8 minutes at 230 is hard!” then you ripped 278 for 20! I applaud the effort but please cut AI FTP Detection some slack. It seems magical but it’s not actually magic :smiley:.

We do need some sort of data (not capacitive efforts) but something a bit closer to your actual fitness level.

Here’s what I see:

Ramp test with 253 FTP Result
5/8 - Jordan - SweetSpot 2.0 (Moderate)
5/10 - Zwift 55 min - 208 np
5/11 - Ericsson - SweetSpot 2.8 - 8 min intervals between 223-238 - np 200 (Hard)
5/14 - Holt Hill -2 - Tempo 2.8 - 177 np (easy)
5/16 - Hogeback - Tempo 3.9 - 191 np (Moderate)
5/17 - Zwift - 1 hour - 164 np
5/18 - Carter - Endurance 3.3 - 170 np (easy)
5/22 - Beech - Endurance 4.1 - 171 np (easy)
5/23 - Zwift - 10min - 194 np

5/23 Run AI FTP Detection - 239 ftp

5/25 - Zwift - 30min - 260 np <-----WAY OUTSIDE THE CURVE!

I think that 5/25 looks like a different athlete.

You also haven’t given us much historical data, this makes it harder for AT to judge your “rebound” from sickness that I suspect you’re having.

With this new data, I re-ran AI FTP Detection for you today and it’s got you at 249.

Can you do me a favor and run Marion at the FTP you think you are at (265)? Log In to TrainerRoad

That’s a 4.2 threshold at 3x12 @ 98%. If you can do it, then you’ll give AT more data to lock you in. If you have problems, then I think AI FTP Detection might be a bit more accurate. I suspect you’re somewhere in the middle of those two…but I’d like to see!


Thanks Nate!

Is one effort in 14 days enough? Subjectively saying Ericsson is hard, a 3 out of 5, and then nailing a couple Tempo workouts, seems like AI FTP should tell AT and the user to get another SS or Threshold effort before using AI FTP. Because giving a 5.5% reduction after two weeks of training seems a bit odd, don’t you think? Maybe you have data that taking a break and then training for two weeks leads to a loss of fitness, but that just doesn’t seem right.

That last Zwift ride did push the FTP prediction up a ton.

One workout with four (hard) 8 min efforts around 230 aren’t enough to push the model to a 265 FTP.

I mean, if I told you 4x8 mins @230 watts were “hard” for me, would you guess my FTP was 265?

ML is basically statistics, and what we generate are likelihoods of FTPs at different watts. The more data you give us, the tighter the curve is.

But with AT, we can zero in on where on that curve you are after a couple of workouts if you don’t give us much data.

With this new harder workout, the model raised their FTP to about 250. And it still gives about a 13% chance to be 265. So it’s entirely possible that is their FTP.

But the beauty of this + AT, is that if they do a threshold workout at 250 and it’s not so bad, it will raise their PLs and zero in on the appropriate training level for that athlete. :tada:


Thanks again, and that last Zwift ride is an outlier for sure. Appreciate the info in your two posts, good stuff. My observation was that in this particular case, it feels like AI FTP is missing a guardrail.

Don’t ask me to guess because I’m terrible at rating perceived exertion!!! Almost two years of writing post-ride notes and rating RPE on every workout. #StruggleFestWithSubjectiveRatings Majority of workouts see a 3 out of 10, or 5 out of 10 on TrainingPeaks. Recently did 5x6-min at 100% and wrote ‘easy peasy’ rated it 3-of-10, then 3x8-min at 106-108% and wrote ‘these felt righteous’ and rated it a 5-of-10. :man_shrugging: Sandwiched in there was 23-min at threshold to calibrate my brain and power targets. Subjective rating of feelings is so, umm, subjective :rofl: (sorry couldn’t help it)

So if you told me 4x8 mins at SS (from a recent ramp test) were ‘hard’ I really wouldn’t know what to say!!!

In other words, I wouldn’t want to be that algorithm! :joy: What is hard? If hard was barely hanging on to power, can’t catch my breath, feeling like a fish out of water, I’d probably think your ramp FTP was too high and those weren’t sweet spot.

Different story when it comes to my power:HR, and Garmin 530’s ML on my HR, HRV, and power data.

Keep up the good work!

1 Like

Hi Guys.

Thanks for all the input.

So I did Marion today (3x12mins @98% of 265 watts) and finished it.

I reported the effort as hard, but I could have probably done another 12 minute interval.

It was similar “hardness” to the 8 minute sweet spot intervals I said were hard a few weeks ago.

I do find “hard” very subjective though and can be affected by so many factors such as how my bum felt on the saddle.

Does replying “hard” soften future sessions and should it only be based on effort level in future?

I like to work hard (bury myself) on the trainer as that is what seems to bring out the best in me and translates well on race day.

I did find the last few weeks easier overall than when I’ve done SSB on its own before AT came about.

And damn @Nate_Pearson, I thought it really was all magic!

I’m not sure what the take away here is - maybe when I think a session is hard (effort wise), I should mark it down as moderate?

Thanks for all you guys do,


1 Like