I’m two days away from my next AIFTP detection. Realistically, what I do in the next two days should not have a significant effect on the results of that detection (and also not on my “physiological FTP” but I don’t want to get into that here…). My plan has Merced -2 listed for tomorrow (4/30), a 5.2 VO2Max workout that is 5 sets of 5 decreasing intensity 40/20s. I changed it to Bird, a 5.4 VO2Max workout which is 3 sets of 5 1-minute on-offs. That single change dropped the Friday 5/1 prediction an entire 1%. I thought, “ok, maybe AIFTP thinks Bird is too hard for me right now and would be unproductive.” Bird is estimated at Very Hard, but from experience, I know my physiology is predisposed to short VO2Max efforts like Bird, so I’m fairly confident it wouldn’t be Very Hard, probably just Hard at most. (I did Bashful last week, a workout almost identical to Bird, and rated it Moderate.) I swapped in Bird -2, a 4.6 level workout, and the prediction still went down 0.6%.
What’s going on here? Does AIFTP think that 40/20s are more productive for me than 1-minute on-offs at this point? Having a single workout swing the prediction this much when it’s only two days away seems like a great example of AIFTP being overly sensitive when almost all the work in the AIFTP training window has been completed.
Yes, TR staff can view my calendar and comment on it here.
It looks like you may be running into a known bug here. This isn’t a model issue but rather a bug in projections. The model is working as it should when it comes to detection day, but if the plan is set up as outside workouts (which looks to be the case on your last training block), this affects your predictions more than most. It starts higher and decreases over time… But as I mentioned, the detection itself is not affected.
More info from Jon here too:
Now, when I look at your training I would not be surprised if the detection stays the same or is a bit lower because 1) you’re in the Specialty Phase and we don’t see great increases on FTP here, 2) If we backtrack to your detection on March 6th (349), you’ve been consistently rating workouts as Very Hard, Maximum Effort, Did Not Pass. I feel this is likely the cause + on this last block there was a full week (April 13th) where you mostly just walked/ran which means your key workouts (intervals) that bring FTP gains, were missed.
Yeah, I wasn’t expecting a big bump during specialty, and I’m fully aware that the “true” results of my training has also been affected by some external challenges in March/April. I was more wondering about why different workouts within the same category (e.g. VO2Max) seemed to have different effects on the prediction. Would these different effects persist for indoor workouts, or is that related to the outdoor workout issue?
@Caro.Gomez-Villafane After reading Jonathan’s note on the outdoor workouts and seeing some other threads that are running into the outdoor workouts bug, what is going to be my best option for planning future outdoor workouts? I’m at the end of a plan and will very likely be setting up a new plan for summer/fall 2026. All of those workouts will be outside (barring wildfire smoke issues). Is AI FTP going to predict sufficiently difficult workouts for my fitness to increase if they’re 100% outside? It sounds like the prediction is the issue, not the workout modeling, but if the prediction results affects which workouts I choose and complete, how can I be confident that the workouts TR is suggesting will be of sufficient difficult to increase fitness vs maintaining or decreasing?
What workouts are being recommended because these are based on your power data + current FTP.
AI FTP Detection on detection day.
So it’s just the prediction aspect… It starts with a higher FTP prediction if your training is set to Outside and as you start completing workouts and feeding the system data it starts to adjust, hence the decreases. On detection day, the AI FTP Detection reads all the actual data and detects your FTP.
You can always turn AI FTP Prediction off for visual purposes, and AI FTP Detection will still pop up after 28 days.
@Caro.Gomez-Villafane It sure seems like something else is screwed up with outdoor workouts.
I just set up a new plan this week and the 28-day FTP prediction was +4.8%/+16 watts. Probably a slight overestimate, but clearly an increase.
All workouts were left scheduled as indoor workouts. I will be changing them to outdoor the day of the workout.
I completed English outside yesterday, hitting the workout targets well, and rating it Hard (TR predicted Hard or Very Hard).
FTP prediction dropped 22 watts to -1.8%.
Any workouts I didn’t manually change in the 28-day window also changed, with workout levels dropping about 2 levels per workout (from 4 or 5 to 2 or 3.
This doesn’t make any sense - I hit the targets, my experience matched the TR estimated difficulty, and TR still thinks I’m struggling?? Even disassociating the workout and changing the rating to Easy didn’t undo the 28-day changes, which is very frustrating. It makes sense that the FTP prediction would drop if the workout levels dropped, but I think TR is not “giving credit” for the completed workout correctly.
One thing I’m concerned about - does AI FTP look at NP or average for evaluating an outdoor workout’s success? For short intervals, e.g. 1-minute on-offs like English, NP will always be way lower than average power, inside or outside. I had a similar FTP drop after completing Merced -2 outside recently, and that similarly had short repeated intervals where NP is necessarily going to be way lower than average power. Here’s a ride I did in ERG mode last winter with 2-minute intervals - NP is still lower than average power, but TR considered this a full pass: TrainerRoad.
