Today I had a 4x5 VO2Max workout. For the first interval I was bang on target, the second I was 3 watts above, then 8 above and finally 2 above for the last.
My prediction, which is tomorrow, dropped by 9 watts.
Bug fix: bring back Chad?
The plans have moved on. No need to bring him back. Both TR and Chad have been clear on that era being over.
It’d be nice to have Chad back. His knowledge and approach was invaluable.
I am not a fan of FTP prediction. Seems like there are more complaints and questions than anything. Maybe TR needs to rework that idea a bit.
For what it’s worth I also see this inside when I use my assioma duos in erg mode as the trainer power source and there’s slight variation in output throughout the course of an interval. Am I better off using my trainer power for TR workouts?
I think the main problem is with outdoor training, when I was doing my plan indoors it stayed consistent with the FTP prediction, but now that weather is better and days longer I’ve been doing my training outside and my FTP is dropping, not improving.
I’m seriously considering canceling my subscription and moving to something else and resubscribe in winter when I’ll go back to do my training indoors, because between the FTP prediction and that the workouts that TR is giving me don’t work too well outdoor.
People really rate the Prediction huh, it wasn’t a feature 5 months ago but fair enough if you move to 100% outdoor rides, though I do rate RLGL and the calendar highly, regardless of structure.
I kind of rank the main features as..
- AI workout selection
- RLGL
- Plans
- FTP detection
- Calendar
- FTP prediction
If a platform highlights this prominently a feature of course people will take it in consideration when using said platform, why should I ignore the prediction when it’s at the center of the interface?
think the main problem is with outdoor training, when I was doing my plan indoors it stayed consistent with the FTP prediction, but now that weather is better and days longer I’ve been doing my training outside and my FTP is dropping, not improving.
This is exactly what happens with the bug we’ve been discussing. A fix is on the way!
I don’t know if I’m affected by the same issue or not, but my ftp is due to update tomorrow. Currently 194 and all block it’s been predicting between 198 (at the start) making its way down to 194, then 193 briefly. It was 194 before I did my workout today (vo2 max level 4.8 - not the heights I’ve done in this block).
I did a cross race (so outside) 2 days ago (southern hemisphere) and unfortunately only recorded the 2nd half of it. My lap times were 2-3 seconds lower each lap, so I don’t think the numbers would have been too different except a few pinches it was definitely dropping off on (haven’t trained that repeated short high power at all and felt it!).
I did my openers the day before outside too.
But today I was back inside. My Bluetooth dropped out 20 minutes in and I had to restart to reconnect, so then skipped the warm up. I have the full set of intervals in the session that uploaded. I marked it as hard and selected equipment issues when it asked why I struggled.
Ftp prediction drops to 190. I changed the rating back to hard and it dropped even further to 187!
To be honest, training to a slightly lower ftp might get me better results in the short term. In the past I’ve tested using ramp tests and always topped out at 199 and really struggled to complete workouts. So spending more time a bit below might be the space my body needs to build more aerobic capacity.
So I’m not super bothered by the number. It’s not like it will feel that different at 190 vs 194. But why? I followed the programme almost to the letter. Moved days around a bit to better fit my schedule, and with the exception of a stage race a couple of months ago have been careful with red/amber days (recovery is really critical for me).
Overall I’m totally happy - 166 when I started at the end of Jan to 194 (until tomorrow!) is pretty much exactly where I was hoping to get. I just would have taken longer on my own. And it’s been nice not having to second guess my training.
Just the idea that it might have dropped back to 187 in a day, even though the session was completed as prescribed (glitch notwithstanding) is frustrating. It almost makes me want to do a ramp test (almost…).
If anyone wants to look at my calendar behind the scenes feel free. I’d quite like to know what’s going on. Is it the outdoor rides? Did the ‘incomplete’ workout upset it? Is it something else?
I don’t mean to be overly critical, but looking at it with a different lens, you only recorded half of your race, you had a ride where you skipped part of it, and you moved the days around in the calendar. This is real life and things happen, but I wouldn’t call that “following the program almost to the letter”. It sucks, but remember that every little deviation is going to cause the AI to react (and in my personal experience, often OVERreact).
Unfortunately, I’m running into this issue again in this training block as well.
I do all my training outdoors, and during this last 3-week block I completed every single workout as prescribed:
-
no failed intervals
-
no underpowered workouts
-
nothing marked as “Very Hard” or above
Actually, because this already happened in my previous block too, I even started riding slightly above the prescribed targets during intervals (+5–7 W).
I’m currently preparing for a TT in two weeks, so most of this block was threshold-focused work.
Then after a simple 1-hour endurance ride this Sunday, my AI FTP Detection dropped my FTP estimate from 260 → 239 W (current FTP 241 W).
That honestly makes no sense to me.
After 3 weeks of consistent training, solid compliance, and good sensations on the bike, the model suddenly thinks I got significantly worse?
I started using TrainerRoad after winter, and initially AI FTP Detection worked really well. But lately — maybe after some model update — especially with outdoor workouts, the estimates feel completely disconnected from reality.
Another thing I don’t understand:
I can see that next week is supposed to contain 3 threshold workouts, but after completing only 1–2 workouts, one of the harder sessions suddenly changes into an easy endurance ride because the AI FTP changes again.
At this point I constantly feel like I have to manually select harder workouts just to keep the progression realistic and avoid losing FTP in the prediction.
So I have a few questions:
-
How should I pace my upcoming TT if AI FTP Detection doesn’t seem reliable?
-
Is this issue with outdoor workouts known, and will it be fixed?
-
Should we still be doing manual FTP tests after each block to recalibrate the model?
-
Have there been any recent changes to the AI FTP Detection model?
I thought the whole point of AI FTP Detection was to reduce the need for constant testing, but in my case the estimates seem to drift lower despite good training consistency and performance.
Also interesting:
During my first block after winter (march), FTP progression looked reasonable. But in this block(may) despite better fitness and harder work, the gains are minimal — or even negative.
And it seems like this mainly happens with outdoor workouts.
I’d really appreciate some clarification, because right now it’s honestly starting to reduce my motivation and confidence in the platform.
Any updates to this thread and the outside riding prediction bug?
However, doesn’t the prediction become the detected AI FTP as time draws nearer? I assume it’s based on the same algorithm.
In my experience yes the detected FTP has always matched the predicted FTP (despite being hint at by TR it does’t)
I don’t have an update at the moment, but I’ll be sure to follow up if I hear anything tomorrow when the rest of the team is online. 
What we’ve always said is that the prediction started too high and dropped over time before converging with the detection.
Eddie, I interprited your quote below differently
Carl