Yes, you may share the details here. Thanks for asking.
I may not be a spring chicken any more, but I’ve got many years of interval training in me (most of which with TR) to have a decent idea what good, consistent training, recovery and nutrition feels like for me. Since January I think I’ve been very good (not perfect!).
My experience is that the predicted FTP assumes a slope of FTP gain over the next 4 weeks based on your past data (+ the past data of users with your age/gender?) Then it relies heavily on your difficulty rating of each workout.
But, here is where it gets tricky. You would assume that if you rated a workout at the predicted difficulty (e.g. a VO2 workout as HARD or VERY HARD), you would stay on the curve it predicted for your FTP gain. This does not seem to be the case (for me, anyway). In order to continue on the trajectory, I have to rate everything at least one step lower than the difficulty prediction (EASY or MODERATE for sweet spot, MODERATE for VO2). You can change difficulty back and forth to test this, but of course it’s best to leave it at an honest difficulty rating when you’re done testing.
My conclusion based on this very unscientific analysis is that the predictor gives you a FTP but to stay on the trajectory to reach it, the workouts should feel slightly easier than it predicted. If they feel as hard as predicted, it pulls you off the trajectory, even to a decrease in FTP at the end.
to come back to this, idk if this helps; if you do the intervals workouts inside and only the endurance outside the effect on ftp isn’t as prominent and it usually stays the same. I’ve seen it maybe drop 1 watt in prediction doing that.
Alright, so here’s what I’m seeing. 
You’re on a plan that generates an initial fatigue warning. While you don’t too often go far above the prescribed amount of hours each week, you do regularly exceed the weekly TSS targets. You’re getting a fair amount of red and yellow days, but they don’t seem to be interfering with your training too much since you’ve built a schedule that caters to your two hard days a week well. Would you be better off with less volume? Maybe. Probably. Based on the data, we think so. 
Aside from that, ever since you’ve been riding outside again (around mid-March), your training compliance has changed a bit. You’re often doing one or two group/solo rides a week in place of one of your hard workouts. With Fridays being your sustained hard workout day, that’s a tough one to miss due to group rides.. It might help to switch your Monday and Friday workouts around every week or two if you’re going to do group rides on Fridays to ensure that you still get some threshold work in, and not just VO2.
Also, your easy days don’t follow the prescribed structure and are often way too hard. March 24th is a great example where you had Pioneer -3 scheduled, but rode outside without structure and ended with an IF of .91.
Losing half of your hard, productive work and then turning up your easy days is likely a recipe for your gains to slow down. I totally understand why what you’re doing right now is appealing, though.. I’ve waited 6 months to ride outside at this point, so I’d personally throw my structure out the window in favor of simply riding my bike for a while, but that won’t help my fitness too much. It’s better than nothing, but once you’re well-trained and near 4 w/kg, it, unfortunately, takes a fair bit of focus and discipline to keep your fitness up and make gains.
You also failed a couple of workouts in mid-march – one due to “sleep” and the other “injured.” Was there something going on then, or were those minor issues?
What are your goals? I see that you have an A race a few months out. You’ve got plenty of time to build into that, but if you need some time to just ride your bike, I’d say go for it. Just don’t expect crazy gains, and you may even lose some fitness in terms of FTP from TR’s perspective, but FTP isn’t everything, and that fitness can come back quickly with some focus.
Let me know your thoughts.
Hi @eddie ,
Thanks for the deep dive!
This is interesting @eddie . Back in January, around the launch of the Ai features, I did the thing that Nate and others were recommending all over the place - I played around with different plan options and volume and let the resulting Ai FTP prediction guide me. When I added a fifth day per week (Tuesdays) plus the dynamic endurance extended time on Saturdays my prediction went up. Similarly, when I’ve tried removing a day here and there since then, the prediction drops - it doesn’t increase. So where is the system telling me it thinks I should do less?
That fifth day also made sense against my experience most of last year of being relatively stagnant while following a “masters” 4/days per week plan. I felt like I needed more volume to make gains.
I only noticed the fatigue warning when I went back into Plan Builder in your account to review your schedule.
I think it’s tricky sometimes when athletes are able to put together a schedule that tells them it will bring the most gains, because there’s a lot that goes into each training plan. While one plan might claim to predict the highest gains, another tool in the machine might also notice that plan will bring on more fatigue than you’re used to.
It’s kind of like the training approach feature. The more aggressive training approaches state that there is some risk of overtraining compared to the others. For many athletes, it’s worth the risk.
I think this is something we can develop further to be clearer and more intuitive for our athletes. I’ll bring that feedback to the team.
Right. But what’s the system really looking at to adjust individual workouts and make FTP detections? The actual work done and qualitative responses (I hope)? Or expectations based on trends… and if so then likely also confirmation bias (I hope not)?
Also, this now leaves me confused over that guidance I’ve heard from Nate, Jonathan and others in podcasts and forum posts to do that very thing. Play around with plans and find out which is “best” based on the predicted FTP.
@eddie Would love your thoughts on my data if it helps refine the algorithm. Feel free to look at my rides.
Yesterday I rode Boarstone (2h zone 2) outdoors. Rated it easy and received a 5W drop in my predicted FTP about two weeks from today. I hit the prescribed average power on Boarstone but rode an extra 8 minutes at 25-50% FTP to get home and did a single 5 second sprint at the very end. One other variable I noticed: indoor Boarstone is a series of 15’ intervals in zone 2 (ranging from 165 to 194W). Outdoors the workout is a single 1:15 interval with a requested range of 165-194. I did 189NP 187average over this interval
My question is whether any of the things listed above accounted for the drop in predicted: NP? TSS from extra 8’ or the sprint? Or was it something else unrelated to Boarstone?
TIA
There is a known bug with doing rides outdoors causing issues with FTP prediction.
Yes, I’m aware - been reading these posts and commenting occasionally for at least a month. But the issue is inconsistent - some indoor workouts also cause a drop, whereas most outdoor VO2 and threshold workouts have, for me, not. hence, some insight from TR would be useful. 
This does sound like the bug that @Pbase is referencing.
The fact that your prediction dropped after completing some indoor workouts likely isn’t related, but I can’t say that for certain without knowing more of the details.
Could you send me the details of a workout where that happened, and I can take a look at what was happening on the backend?
Thanks for chatting off-forum about this @eddie
Similar experience today with Polar Bear -3. Same result. Completed outdoors, rated easy (honestly, if there was a “very easy” I would have picked it) and got a 4W drop in predicted along with a decrease in the IF/TSS for the next few scheduled interval workouts.
Does the outdoor workout bug have to do with going harder than the expected TSS/IF? Polar Bear -3 is supposed to be 2h, 102TSS 0.71 IF. I did 2:10 , 117TSS, 0.74 IF. It’s pretty difficult to nail TSS and IF perfectly outdoors. But, it would be trivial to set FTP manually so that the workout hits the expected numbers perfectly. Not a great solution, but a possible band-aid until this is sorted on the back end?
2 hours at .74 should not be “Very Easy” (assuming you did a very easy ride and didn’t bump the IF up with sprints or something). This kind of sounds like an FTP issue. Or maybe you and I just have different views on what “Very Easy” means?
Felt like the chain wasn’t connected. 
Definitely not .74 
Have you not had one of those days when the legs were weightless and everything was super easy? I’ve only had a few like that, including recent a VO2 Max workout. It was just one of those days that you wish you were racing or targeting KOM’s. In the same way that we can sometimes have “bad leg days” (such as my last race - argh!), it happens the other way too.
I don’t believe there is a good workaround at the moment. I think this is more about how we process outside ride data, not what the data shows specifically, if that makes sense.
We’re making progress on this one! Hang in there!