I haven’t. Zwift rides tend to sort of be sweet spotish so figured it’d work. I was initially planning on doing one 10 minute interval after but realized I was running behind time wise.
2 Likes
Then why would you see a 7 watt drop after completing a planned workout?
The model says based on X amount of workouts left until the end of the 28 days.
It’s very confusing. Other than this I like the new release but it’s weird.
1 Like
I’m pretty new to trainer road but i’m liking this feature.
I check it every day both before and after every workout.
293 → 313w Original prediction based on a FTP focused 4 weeks
Its currently reading 315w as i keep tweaking it slightly to see what it thinks of adding in some extra volume etc (such as an hour swim) to see if it trends me down.
I’m 2 weeks in and it seems to be holding steady for me. If it drops after a workout i assume i need to focus on fuelling or sleep and that i didn’t come into the workout optimal.
One thing I have noticed which is probably a massive flaw and trap is if you swap some of your rest week workouts to v02 intervals your get a higher number when your next assessment is at the end of that block 
. I think this is just a limitation of only doing the 28 day simulation as i imagine you would be at a lower point 3 months down the line. But its hard to judge how much rest you need in a rest week with the current simulation length.
I expect based on if your cadence has dropped through the intervals heart rate drift etc it can pick up hints of if you aren’t adapting as expected alongside the RPE feedback.
2 Likes
My next assessment is on the Wednesday of my recovery week. It still shows a rather unrealistic 6.8% increase. If I schedule a VO2Max workout for the preceding Monday, which should be a rest day, the prediction jumps to 10%.
Even better if you’re on a Masters plan … just add a bonus set of Anaerobic 30/30s in Week 1 and a VO2 Max workout in Week 2 for an extra 15 watt AIFTP bump at the end of Week 4. 
Just don’t be an idiot and fail the VO2 Max workout because you have a cold. Ask me how I know
The good news is at least now the AIFTP Prediction is a bit more realistic.
2 Likes
Feels like the number is a prediction of what training stimulus you can take. Which is a leading indicator of actual performance but not actual performance.
1 Like
Interesting. After a few more workouts, my prediction stayed at 2.6%. Taking an idea from a post in a different thread, I’ve just now retroactively changed my response for the Threshold workout on January 25 from “very hard” to “hard” (I could not decide on the day - I was somewhere between these two). This immediately changed the FTP prediction (now 15 days out) from 2.6% to 6% 
. That’s wild. I’m pretty rubbish in my RPE assessment (I never race and train for only a few months in winter), and this much sensitivity to RPE of a single workout is somewhat unexpected.
2 Likes
This doesn’t necessarily solve your problem, but at the least, using the “Chad Chart” will ensure your ratings are consistent.
Workout survey response seems to heavily play into FTP prediction. This is a bit of an issue since there isn’t much granularity in the survey. Would the model improve if they added:
“easy to moderate”
“moderate to hard”
And “hard to very hard” as additional survey choices.
It may take some time to sort out new response options with all of the old data, but we ultimately might not see such big changes based on a survey response that could easily go one way or the other.
3 Likes
I don’t think that’s true.
I’ve also noticed these wild swings but have also find that they are self correcting to a degree. I.e. you “fettle” your RPE answer response to get a higher prediction - but you also get harder workouts between now and the prediction date - which you will find a touch harder than expected - which brings the prediction back down.
Also you have to remember that the FTP value (both detected and predicted) isn’t a direct measure of fitness - it’s just one part of setting your workouts. You also need to look at the difficulty of the workouts following the prediction to get a better idea of your predicted fitness.
7 Likes
You could ONLY do ONE more interval or you could do another full set of intervals? Those don’t seem very close to me.
1 Like
Think it fair to say it deserves the current beta tag. Though from my seat, it is working really well and doing what I want TR to do, nudge me to make better decisions.
This and fatigue warnings have been assisting with adding workouts and re-arranging weeks around life. Lots of small nudges seem to be adding up to a good use of my time on the trainer and more suitable outdoor rides.
5 Likes
That’s what it does do - are we not just talking about the future prediction?
The prediction isn’t particularly complicated - it’s just a basic straight line assuming each workout follows the most likely RPE response.
The predicted RPE responses use AI and are brilliant - but most have at least a 1 in 3 chance of being something other than the most likely response - so it’s expected that when you stack a few of them together the accuracy will decrease.
You can only game it up until the point you can’t complete the workouts anymore 
12 Likes
Well, that depends. In my case, I knew I physically could do an interval (even though I would have to work hard) - that points to “hard” - but on that particular day, I was not sure I would be mentally able to force myself to do so (more in the “very hard” territory). The truth is, I would much prefer TR to get the information on how hard I had to work from the workout data instead of asking me 
. Especially if a single change can have such an outsized effect on the prediction. Which goes contra to what Jonathan and Nate said in the podcast - “do not think too much about it” (or something to that effect).
For the record, I do think the prediction has some merit to it, but we need more information (and/or it needs more finetuning) in order to be really useful to an average TR user.
And you aren’t gaming the test. Just setting yourself up to overachieve or fail a workout if you continue to try to game Chad (I think that is the name for the AI).
Chad doesn’t care if you are trying to trick him, he assumes you want what you are going to be given.
4 Likes
Your example to me is a clear Very Hard
3 Likes
If you consistently go with your first read, it’s probably more accurate than over thinking it.
5 Likes
I will admit that I find the future FTP prediction has a far greater steering effect on my survey responses than the AI RPE prediction in isolation.
1 Like
If you answer how you honestly feel, I think it is. Either you think you can do one more set, one more interval, or no more. I don’t think it’s gray if you just give your first answer. If it is, I would pick the one based on which workout I want next, a harder one or an easier one. Don’t overthink it. Go with your gut.
2 Likes