But if that trend continues the ftp prediction will be about -2.5% by the end of the 28 day block 
2 Likes
As you mentioned AIFTP is based on a ML-model and is a prediction based on a simulation. A simulation has some boundaries, means, when everything is running well. The recovery, no stress and you are a average responder of the stimulus, than you can achieve the predicted FTP-value.
But if you could not recover as it is defined in the model or if you are a slow responder then the prediction must be adjusted based on the done workouts. The simulation uses this workout’s to tune the values in the correct way.
But as you have although said, no training according the plan has led to a decrease…. this looks suspicious.
I am not a sport scientist, but I recall there is stated in the sport literature an increase of FTP is in reality between 3-4% within 8 weeks, if you are not a cycling beginner? Maybe I am wrong.
I have started than as a beta tester and my prediction was +9% and having followed the plan strictly, now 1 week prior the 28 days is +3.4%. If this comes true and it happens every 28 days over the year, I am happy.
Regards, Markus
2 Likes
Well, 12/20 - 05/21 (my first indoor season) I went from 240 to 290 using the much criticised 10-12wk Zwift FTP Builder plan. My current AI FTP is 265, so I kind of expect this sort of progression 
Last season (12/24 - 03/25) from 265 to 281 using TR for the first time, which was less than I expected.
BTW I would like to find a TR plan which would be more efficient than the Zwift one (which is 5-7 hrs a week, whereas TR suggests significantly less than that for me).
I had the exact same issue. I was predicted a higher number on 2/6 but after todays easy ride (1.5hr z2 easy spin at 60%ftp), due to the fact that I had to take a phone call half way pausing 8 min (I thought about disregard and start over but it was already 40min in so I didn’t), my threshold workout tomorrow got downgraded from WL4.0 to WL3.4, and the prediction went lower. And if I manually add back the exact WL4.0 workout, it becomes “not recommended” and would actually yield an even lower FTP prediction. I am so frustrated as I am very confident I can complete the WL4.0 workout perfectly but don’t want to make the prediction go lower.
It’s a prediction. Who cares. The sensitivity of a pause in the workouts seems to have a greater impact then previously. We can always use our knowledge of what happened. Do the workout that was originally scheduled is what I would do.
2 Likes
I had the same issue with the workout McGergor +1 where on the day of the workout I couldn’t do it mentally because of the beating I got at work that day. Moved some workouts around that week so that I can do McGergor +1 the next day but TR was saying Not Recommended all of a sudden, at that point I wasn’t going to let the AI drive me mad, did McGergor +1 anyway and gave the system some good data.
Another vote for just do the workout. FTP will be what it is.
6 Likes
If you’re confident, do it. Show that AI who is the boss and see what it does.
4 Likes
People should care when the changes don’t make sense or have no apparent logic behind them.
Pausing for a few minutes during an easy endurance ride should not knock your threshold session way down.
Sure the user can adjust it back but that’s not the promise of the system.
1 Like
I try not to let it get to me but I can see how the FTP prediction can play head games. Just trying something’s I was changing workout RPE from Hard to Very Hard just to see how much of a hit the Prediction would take. The drop seemed excessive especially for a workout that I completed with normal power and heart rate readings.
It’s getting to the point where I’m starting to treat the Prediction let the HRM on my watch, look at and just think “oh that’s nice”.
I think once the option comes to disable it 100% with no reminders on the app Home Screen, I may opt in.
1 Like
Updates - after playing with the system (trying balanced vs demanding and various workout durations), the AI is willing to give me Sonoran at WL3.9 without the “not recommended” tag, and the prediction is only 1watt lower than before. However, I dived deeper and found out that if I were to manually change Sonoran to another workout “Bounty” which shares the exact same WL, the prediction would go done by another 5watts. There is definitely something behind the AI that I can’t understand. But since it’s a new feature I have been sticking to for 3wks, I am gonna give the AI a chance and stick to its calculations and see what’s gonna happen on the 2/6 ftp detection day.
3 Likes
You are absolutely right - has this happened for anyone?
I though it would effect workout scoring but nothing else if you rate the work you did honestly
Apologies if there’s s an example in this thread I haven’t seen
1 Like
The system for me doesnt work as described. Doesn’t matter what RPE or what workout I pick my sweet spot workouts are at a threshold level even if I drop from Balanced to conservative. The FTP prediction at first had me increasing 16 watts. No way could I do the majority of those. After 3 workouts it has a drop of 4 watts. The workouts still not the best.
If we have an FTP prediction that designs a workout program that is dropping my expected FTP, the question is then why? In my view the inputs are wrong. My FTP is too high.
So I have ignored FTP prediction. I manually enter my FTP. FTP prediction now is off and the workouts are back like they were before the new system.
I have used TR for 10 years. They at time get things wrong but they will adapt and adjust. It is still early. At some point it may go back on for me.
As to why does a break result in a drop. The most likely reason is work produced during time for workout. So if we do 500 kilojoules of work in 1 hour vs 500 kilojoules in 62 minutes where is more work done. A break will reduce work done per minute. This is only my guess as to why a pause or decline has an impact. I dont need this level of detail.
I enjoy my workouts and when a system doesnt work then I adapt to what will work for me while they figure out the problem. The problem doesnt occur for all of us. Many are fine with it. So yes I dont care at this point about FTP projection. I have talked to support so this is where I ended up.
2 Likes
They aren’t punishing you, it sounds like you got full credit for the threshold aspect of it. It’s just rating it per its algorithm. If someone did all of them at 304 would you want them to get a higher score?
2 Likes
Sounds like they evened out. So it’s just reallocating from Sweet Spot to Threshold. Rather than docking you. If you just skipped the unders and did all at the over power would you want it to give you sweet spot credit?
I like the concept of this new feature, but agree that it probably has a good bit of room for improvement. My initial impression is that it’s not very smart about how much to ding for a drop/pause and it’s not very smart about how much credit to give for doing extra work. But I consider it a 1.0 release of this new feature and it’s been a long time coming. This feature has the potential to be one of my favorite new features. It basically allows you to adjust your workout on the fly if you are feeling good/bad and still get appropriate “credit” for it. And while I don’t think the adjusted numbers look very accurate (and I assume those numbers play into future workout recommendations), I’m mostly just happy the new system is actually looking at the work performed rather than incorrectly assuming I did the workout exactly as planned. A huge step in the right direction IMO (in some ways, it’s a big step towards WLV2).
Bottom line, I’d much rather TR release a feature once it has the basic/usable stuff ironed out. In my opinion, the best/fastest way to mature a feature is to release a basic version and iterate on it, not monkey with it in the back room until someone thinks it’s “done”. No matter how much you develop and test a feature prior to release, it’s almost never done and perfect until users get their hands on it and the real-life use cases all play out.
3 Likes
It happened to me dropping my threshold workout the next day.
One fun thing you can do with it is complete a random zwift ride/race and then score it if you load a workout. I did the zwift triple loop ride as a hard ride but not race then loaded up my planned workout. Did it as my cool down. Then it scores it.
Thank you so much for this explanation which, I intuitively anticipated before reading this post but just wanted to be more sure as to the “why” I would do a threshold workout one day and see my Predictive FTP number be larger than it was after the endurance workout done two days before, then to see it drop again two days after that threshold workout after another endurance level workout. Sort of amusing if you ask me.
1 Like
Thanks for sharing, this was on my list of things to try to see how close this functionality is getting to WLV2. Just curious, have you tried taking the same ride and assigning to a dramatically different TR workouts to see if the scoring is biased at all by the assigned TR workout structure? If the ride is being evaluated completely independent of the TR workout (which would be WLV2 in my mind), I don’t understand why they aren’t currently scoring the rides without a TR workout associated.