Called it. 
I want to give @eddie a hug
This was one of my final posts in the beta thread before it went live:
I’m what I consider a TR super user…based on the fact I’m here on the forum in the beta and listen to every single podcast is probably more than 95% of users do. If I’m confused by this, you think the average athlete jumping into TR won’t be?
I want this to succeed. I think TR support is going to be inundated with tickets because of this feature. Just my opinion.
This is the kind of stuff I’d be interested in hearing from @Nate_Pearson on. Like how accurate were the predictions in their testing, why would the model think 30% in a month is real, what influences the prediction model (what is the window), etc.
The workout delivery has been amazing but the AI Prediction feels really noisy.
Or gives a range that tightens. Problem is that range will often cross zero at the start. They should know the variability in it from their internal testing beforehand.
The workout picker is amazing. The FTP prediction is a little bit too much of a magic black box.
I’m having fun with the new TRAI and have enjoyed the improved Calander and am really happy with the workouts I’m being assigned.
I think the new predicted AIFTP is a good thing, mainly to stop me doing too much, or at least understanding the effects of this better.
I have tweaked the hours I’m doing down a little and am trying to keep a bit of a lid on it, during the weekly group ride.
I didn’t get much of a variation in my FTP/AIFTP, dropping from 277 to 274, which, as I took some time off over Xmas, was probably a little less of a drop than expected.
I’ve been enjoying the predicted AIFTP feature. From memory this started out at 284, but as I’ve got closer, is now 278.
For a long time, 1-month productive training has meant a 3 or 4 watt rise in my FTP, so it’s amusing to see after all the hullabaloo, in my case, nothing much had changed.
I was worried that seeing the expected effort level could influence how I rate my workouts, and this was true yesterday.
I did a threshold workout which was 54.5%/42.2% hard/very hard.
I marked it hard and everything on my calendar stayed the same, I’m one week from AIFTP detection and only have two more hard workouts to complete.
This morning, I decided to change this to very hard, as this was a truer reflection of how I felt.
This had the effect of dropping my AIFTP prediction down 8-watts to 270, so a month of training and a 4-watt loss, which is a little discouraging.
The difference to my schedule was my next Threshold workout drops from a 3.5 to a 2.2, which seemed a lot. I’m sure this is fatigue management kicking in.
Out of interest, I then marked yesterday’s workout as moderate. This gave me an AIFTP boost to 281 by next Thursday. Mainly by upping the AL on my upcoming Vo2 workout from 5.2 to 5.7 and my Threshold workout from 3.5 to 3.7.
I thought I’d mention it as it is a bit of a surprise to me that with just two hard workouts to go, how I rate my effort level for my last workout is having such a big effect on my predicted AIFTP.
This is influencing the rating I am giving yesterday’s ride, which I’m leaving as hard, as I can’t quite deal with having a 4 watt drop in my AIFTP!
Ultimately, my workouts are getting a little harder and I’m confident I’m getting a bit fitter, but it would be nice to see this being reflected in the predicted AIFTP and certainly don’t want to see a decrease.
Yeah I think you’re right
This is a common misunderstanding of math. Here, the “math” is just giving you a model of reality. A model implies a modeler, i.e.someone had to make decisions about how to encode reality - they had feelings (albeit hopefully informed feelings!) about what variables to capture and what the values of those variables might imply. Math isn’t perfect magic. There is a human (or machine modeling a human!) here, making imperfect choices.
I agree. And in this case, the algorithm is measuring work done against time. If you stop pedaling for 15 minutes, that’s 15 minutes of zeros. If you discount those zeros, you’re not going to get a real picture of the work done. It’s like people who set their bicycle computer to stop when they stop and then talk about average speed, discounting all of the zeros. This is how people were hacking Zwift races for a while too.
Very solid workouts so far for me. Goldilocks zone stuff. Been really difficult choosing between Moderate and Hard ratings as they feel right on the cusp.
I stopped 3 times during a workout
Lowering to 95% and stopped 10 mins early
Marked every workout very hard even when they were easy.
WHY is my predicted FTP lower ?!?!?
Smile 
It’s a happy day
Jokes are ok
Seriously sorry if I triggered you
It feels like this forum is very sensitive recently 
Really happy with the evolution of TR. I understand there will be some teething challenges especially while in beta.
I do 2 structured TR workouts per week + 3 solo/group rides. Since starting up the structured rides AI is predicting my ftp will go down (even though I’m more focused) and continues to give rides with lower PL’s.
I assume it thinks I’m over training but if anything im doing less work than I typically do and the training so far feels like top of moderate bottom of hard effort level.
Not overly concerned. Just a little disheartening when your screen is filled with a lower FTP predicted. I’m sure it’ll sort itself out
Keep up the good work.
Red days should trigger a block on forum posting.
I’m not an expert, but my impression is with ML stuff that you don’t actually know “why” the model is predicting what it is. You train it, give it weights and inputs, validate it against data, but the underlying why is complex prediction based on all the weights and averages.
I guess it could be return to Fitness gains. Time off. It lowers the FTP. Time back you ramp up quickly.
I can understand, especially if it’s anything like exercise-induced asthma, which I have experienced, & I was very nervous about work over threshold for a while thereafter.
This is really important information in the context of getting help from the community. The reality is that some sort of FTP test is going to be necessary to reset the system & allow it to find your level, & the ramp test is being suggested, probably because it carries the least fatigue moving forward. Your mention of medical reasons makes sense & is to be respected, but you didn’t mention that until later & it could’ve saved a lot of unnecessary discussion.
Perhaps the 20’ FTP test might be more to your liking. It’s hard for much longer, you need to be really good at pacing, but is not as violent as the end of the ramp test. Yes you’ll be breathing hard (working at 105.3% of threshold by definition), but it’ll be more that your legs will want to fall off. This is actually my preferred test protocol & I feel it makes for a good threshold workout anyway.
I haven’t been able to find a true “hour power” test in the TR workout catalogue. There are workouts with efforts of that duration, but I don’t think they can set FTP.
I’m confused, and with apologies, I got 40 posts into this thread and I’m not reading the whole thing. I will happily switch to a support query if that makes more sense. If bafflement is part of the spirit of curiosity, I’m all in… 
I wrote a much longer version, but I’m going to try to distill to bullet points that address the intro post…
- I was okay with having my nominal FTP dropped from 247 to 242, because “new scale” metaphor (even if I was a little bummed to lose my proximity to the 3W/kg threshold).
- I got an AI FTP estimate of 252W. I could be happy with that.
- Estimate dropped after every workout, going from +10 to +5, and today, to -3W. Which is a really rough way to end a 75-minute Threshold workout that went well.
- After reading this pinned post, I twiddled “approach” and got just back to just barely positive by switching to “Moderate” after trying a few (which seems odd; I’m not new to structured training, but I am 54).
- I re-read the “AI can’t suggest volume.” I used plan builder, but after reading that I hit “Check Volume,” after which my predicted FTP went negative again after it appeared to suggest that I should switch my Wednesday endurance ride to Dynamic instead of Fixed. Or maybe it meant I should have kept it fixed, but I hadn’t been cognizant of this setting, and after hitting the button it showed Wednesday as Fixed and Saturday as Dynamic. They were previously 60 and 45 min respectively, and after making them both Dynamic and hitting save, it immediately made them both 90 minutes excepting rest weeks, and returned my FTP prediction to a negative gain.
I’m flummoxed. I’ve been seeing slow, steady gains, which runs counter to predicting none. I know there’s more to fitness than FTP, but it’s awfully strange to see a downward trend in the one metric TR can give me, and during the build phase to boot. It’s also odd that dynamic volume wanted to bump me right up but that also re-negated my gains.
I feel adrift in terms of knowing how to twiddle the adjustments at my disposal, and while I was prepared to have everything move around a bit, I’m finding it confusing and demotivating to see the numbers fall without any clear understanding of even “this is why they’re doing that right now but at Phase X they’ll come back up” or something like that. I’m a hamster on a treadmill: “FTP-ish number go up” is all I can chase outside of a handful of races that are still months off, and which hinge on staying motivated enough to do these workouts.
Feels bulky to post screenshots; calendar is here: https://www.trainerroad.com/app/calendar/jesseransom