Lately I had been thinking the projection was useless but that is a good point. It will be interesting to play around with my weekend rides and see what happens
Maybe that would work. Just a question of how to rate the RPE to train the AI detection properly. I think previous recommendation has been to rate it as maximum effort if you lower the intensity. How does the new system handle things? Does it match the RPE with how the workout was prescribed or with how it was performed?
I’m maybe going to try this and see what happens once I’ve finished my current block in a couple of weeks. I’ll set the RPE to reflect the work I did. It’ll be interesting to see what happens.
So did another Zwift race today, were I held 344 for 5m and 321w for 15m. I averaged 309w and normalized 320 over 27 minutes:
on WKO5 my mFTP shot up to 307w and my zFTP went up to 311w from this race. TrainerRoad on the other hand predicts a FTP drop in 9 days:
I did start a Xert sub to run concurrent to my TR sub, and 297 is my Xert Threshold power. I did note that TR did predict an FTP bump to 320w if I followed a generic training plan this month though.
Based on 0 workouts in the next 9 days. Just put 2 workouts in there and look again.
TR predicts based on your rating of previous rides, power and HR as well as upcoming workouts. It currently sees you doing 0 workouts in the next 9 days since you have nothing scheduled, hence the drop in FTP.
The 297 might be a little pessimistic, but it wouldn’t be out of line based on those numbers IMO (at least for how I look at FTP). If you had a lot left in the tank, that’s another matter, it just appears you were close to a max 30’ effort if 175 is really your max HR. Assuming it was close to max, I’d probably estimate FTP ~305, but it’s really hard to know how much anaerobic contribution someone gets from a shorter effort like that (same reason a 20’ test tries to blow out the anaerobic contribution with a 5’ max effort first). If you feel your FTP is higher, just set it higher and see if you like how the workouts are feeling.
And all of this really goes back to what TR is doing with the FTP number. I haven’t really been following all the forum chatter about it recently, but I assume TR is still saying it’s primarily a number to base workouts on rather than a more traditional physiological definition (so all bets are off on what the real life performance proxy should be). Or maybe they have provided that, I just haven’t seen it.
Yea its exactly a copy of xert, and xert is able to arrive at the number less than half price. My max HR is 189 (best in the last 365 days on intervals.icu, using a polar h10), and my LTHR is 169 (garmin measured).
I still have this race in the same period, this is why every other platform is saying my ftp is higher.
Your current ftp in TR is 302w correct? WKO5 says 307 and xert says 297 so not every platform is saying you have a higher ftp than TR.
The prediction is that it will decrease to 297 because you have zero workouts scheduled between now and that detection date… I don’t see why having a race scheduled (but no workouts between now and then) would stop your FTP from dropping. If you follow that plan you will probably lose at least some fitness.
What happens when you apply a plan to your calendar or even just load in some workouts ad hoc?
I’m not going to pretend that only TR workouts count as training. Zwift, Intervals, WKO, and my race data all show my FTP going up. TR is the only platform that lowers FTP because I’m not doing their workouts, and I’m not rearranging my training just to make their prediction model happy.
It’s not lowering your (current) ftp.
It knows about the efforts you have done, but it has zero knowledge of your future plans, that’s where the ftp drop is coming from.
You could very well get to the race and have an ftp bump because of the work you have done and will do, but the prediction won’t work when you don’t tell it your plans.
I understand how TR’s prediction engine works. My point is that it’s not modeling physiology. It’s modeling whether I schedule TR workouts. Even if I did nothing for 9 days, my FTP wouldn’t drop to 297. Every other platform and my race data show the opposite trend.
How does exert predict an FTP? Have you put future workouts in there?
So why are you surprised it’s showing a drop if you don’t plan in any workouts?
Your current FTP is 302w - it will probably be higher when you do your next detection in a few days - I struggle to work out what your complaint is? 
EDIT: I think you might be ever so slightly over fixated on FTP? My first though was “these are all the same number” but then realised I’ve already said the same thing to one of your posts a few weeks ago 
Could you schedule just a simple AI workout in the window just to test if it could be a bug that no prediction was running. And then remove it again.
It’s predicting your FTP in 9 days. So if you don’t train for 9 days it’s modeling that you will detrain some. You’re getting angry just to get angry.
Every model that you posted is within the range of error for an FTP model. You are comparing three backwards looking models to a prediction of the future. You need to compare apples to apples.
But I honestly don’t get why you are so angry. FTP is primarily to set training. If you were mad about paying for TR and the training it served up that would be logical. Here you are making yourself angry about comparing three backwards looking and one forward looking prediction. 10 days off is leads to a 1 to 3% drop in FTP. That is what you are seeing. It’s saying in 10 days this is the training stimulus that would be appropriate.
Post it. Model the same TSS that you’ve been training at.
That’s completely incorrect in my experience. The only time of the year that I use TR workouts is during my race season (~Jan - July). But I ride and race year around and TR certainly looks at those activities in AIFTP estimation.
You can disagree with the number being generated (I don’t agree with mine either), but TR is absolutely using those non-TR workouts in whatever logic/AI is being applied to come up with the number.
And I can poke holes in most of the tools out there when comparing different FTP numbers. At the end of the day, my definition for FTP is just that - my definition. I’ve been training long enough that I have a good feel for “my” number within ~5 watts as long as I’ve been doing some efforts around threshold (over/unders always give me a great feel for where my FTP sits). TR’s calculation for FTP often differed from mine in the past and this latest release seems to have widened that gap. And that is fine and doesn’t keep me from getting value from the system.
It sounds like you have strong opinions about what your FTP is based on your own performance and what some other tools are telling you. Why not just manually set that number in TR? All you lose is the prediction functionality (which you aren’t going to believe anyway). I guess I just don’t understand the “anger” in an FTP estimate that you feel is off by 5-10 watts. I use Intervals.ICU for a bunch of stuff and love the tool. But intervals.icu estimates my FTP 35 watts lower than TR. And my estimate for FTP sits between those 2 numbers (and that’s the number I manually set in TR and intervals). It doesn’t make me hate intervals (or TR) because their FTP calculation doesn’t line up with my definition, they are still both awesome tools IMO.
If the AI requires me to manually calculate my own FTP using WKO5 to get a productive workout, the value proposition is broken.
Regarding the forward looking defense, a prediction model that ignores verifiably consistent, recent peak performance in favor of an empty calendar is a bad model.
Here is the hard data from the last two weeks showing why a 297W prediction is physiologically impossible:
1. The Hour Power Proof (Jan 27 - Ven-Top)
Duration: 1:31:47
Performance: I held 298W for 60 minutes and 289W for 90 minutes.
Physiology: Avg HR was 163bpm (well under my LTHR of 169).
The Verdict: I have already proven I can hold 298W for an hour at sub-threshold heart rates. Predicting my FTP will drop to 297W means the AI thinks my future Hour Power will be lower than what I just held for 60 minutes comfortably.
2. The Consistency Check (Feb 3 - London)
Duration: 1:23:38
Performance: NP 280W / Avg 275W
Physiology: Avg HR 158bpm (Endurance/Tempo zone).
The Verdict: Consistent aerobic engine maintenance. No signs of decay.
3. The Intensity Check (Feb 9 - Crit City)
Duration: 37:23
Performance: NP 305W / Avg 290W
Physiology: Avg HR 163bpm.
The Verdict: Able to punch well above the predicted FTP for nearly 40 minutes without cardiac drift.
4. The Current State (Feb 14 - Richmond)
Duration: 27:15
Performance: NP 319W / Avg 309W. (Included 5m at 344W and 15m at 321W)
Physiology: Avg HR 167bpm (Still under LTHR).
The Verdict: Yesterday, I normalized 319W for nearly 30 minutes.
Conclusion:
In the last 18 days, I have held roughly 298W for an hour and normalized roughly 320W for a half-hour race, all while keeping HR under threshold.
The AI predicting a drop to 297W is not forward looking analysis, it is a compliance algorithm punishing me for not scheduling TR training plan, completely ignoring that my actual engine is running 10 to 15 watts hotter than its estimate. I do not want safe numbers that ensure I pass a workout; I want accurate numbers that reflect the work I am actually putting out.