Does your FTP prediction change after a workout then change again after the overnight refresh? Yesterday I got a predicted 3w increase … then it reverted to a 2w decrease.
Yep, that started happening pretty regularly. Usually not changing overnight by much, but a few watts up or down seems common.
I’ve never noticed an overnight change, any change I can recall happened immediately after a workout or if I modified a planned workout. I have heard of it happening for others though, and it’s possible I missed it.
Turning off the daily reminder of my ai ftp prediction is the best thing I have done for my training.
I don’t think it’s true that FTP is used to set training zones with the new AI. It’s focused on progressive overload, not training zones. So it’s targeting specific wattage and durations for intervals and using the FTP to pick whatever workout from the catalog has the correct percentages of your set FTP to accomplish that wattage.
I only notice it because when I open the app on my android phone it goes straight to the career page and you can see the FTP box refreshing.
I’m looking into this now.
I’ve been seeing similar issues on my end for the last month, though things have finally stabilized slightly…
- Mixed Intervals sessions wildly throwing off the model (I think they aren’t correctly calibrated) — I’ve removed these entirely from my plan
- Prediction dropping over night with no change in training — no idea what’s happen here
- Not agreeing 100% with prediction difficulty (I.e. Hard vs Moderate, which is pretty subjective and the podcasts said not to worry about edge cases like that) causing extreme FTP prediction swings — when I’m on the edge between two survey responses, I just agree with the prediction now
Since the update I’ve also seen wild swings in the suggested volume for the plan builder / check volume button and am largely not using this feature anymore.
I think the suggested workouts seem pretty good, and better with the new model, but the surrounding software seems to still have pretty large bugs.
Why care? It’s not changing what you are or what you will be, it’s just changing a prediction of what you may be, but might not be.
A few reasons to care from my point of view:
-
It’s interesting.
-
If the prediction is accurate, it can be used to fine to tune training to improve your FTP more.
-
If it’s accurate, it can be very motivating.
-
It can be helpful in realistic goal setting.
So how about waiting until AFTER the modified ride and the rating of the ride for RPE before jumping to modifying future workouts and boning the FTP prediction. This happens to me as well and it’s a PIA.
The point of forecasting future outcomes is so you can change your behavior and avoid those outcomes if they’re undesirable. And if the forecast is irrelevant or untrustworthy, then why would include it.
I think the answer is that the machine learning/artificial intelligence might be wrong (blasphemy, I know.) If I have more time today and I’ve got a 75 minute endurance workout and I’m feeling good, I don’t think that changing that workout to 120mins of similar power output, executing the workout and rating it ‘easy’ is really going to affect my FTP in a negative direction. Nor in a positive direction, for that matter.
That said, I think there’s probably too much ‘faith’ in that FTP number on a day-to-day basis. I think TRs model is more about predicting how we rate a given workout (easy, moderate, hard) at a given FTP setting than about predicting what our FTP actuallly IS on a day-to-day basis. I think the AI FTP detection that you can run every 4 weeks or whatever it is, uses machine learning to more accurately detect (or calculate) than the prediction engine.
Bottom line is that the AI is a black box, not infallible and maybe don’t worry about the predictions on a day to day basis because it might be off (wrong?) by a few points.