SOLVED | Why would increasing Dynamic Endurance duration hurt predicted FTP?

[WillS]

I’ve been playing with my training plan, and I get the best expected FTP increase by going with the Balanced training approach and leaving everything else alone. FTP prediction expects me to go form 367 to 375 (both the latest version of AI FTP) in my build phase over the next 4 weeks. This is when I leave the max duration for Dynamic Endurance rides at 90 minutes.

If I change the max duration, I get longer rides, but it expects my FTP to stay the same. I thought the idea with the dymanic endurance rides was to give the system the maximum time you have to ride and let it pick the best workout without going over. So, why is it trying to fill my available time when it expects a worse result?

I know there’s more to a “result” than FTP, but I always thought that was more during the specialty phase, and the build was when you worked on raising threshold. This build block also looks weirdly easy. Here are some screenshots and my calendar if anyone’s interested. Log In to TrainerRoad

[Torneng]

Both approaches have you doing two hour threshold on Saturdays, instead of the 1:30 you have been doing. You’re also going from SS to Anaerobic on Thursdays, with most of them expected to be very hard. So not quite sure why you feel the block looks easy.

The reason that the predicted FTP drops seems to be that the AI thinks it will have to give you lower and lower threshold workouts during the block. If you find you can handle them then probably the workouts will adapt and maybe your FTP prediction will increase.

Either way, it seems apparent that the AI scales down the rest of your week to fit in longer endurance. Why it does that when you have it set to dynamic endurance is a good question that maybe even TR doesn’t know the answer to.

[huges84]

Is often helpful to think of things in the extreme, absurd case when trying to answer ‘why does this complex thing do X?’

If you do hard workouts seven days a week will your FTP be higher? No, because you’ll crash and burn. You can’t handle that much fatigue and those intense workouts require you to not be too fatigued when you do them.

If you train 100 hours of endurance you will have the same result.

So there’s a right amount of endurance to do that doesn’t create too much fatigue, allowing you to nail those hard workouts. There’s a range of endurance volume that isn’t too fatiguing that will have about the same training impact for the next 4 weeks.

However, extra volume does pay off in the long term (12+ weeks), provided it’s not too much of an increase from what you’ve been doing recently. The AI is trying to manage the fatigue by not ramping you too fast.

[WillS]

I don’t love the 2 hour threshold days with such short intervals, but I’m trying to just follow the recommended plan. Maybe that’s why the block looks easy to me. I also feel like you can always squeak out another anaerobic interval if you have to. With anaerobic workouts usually being fewer kJs than SS I also feel like I recover from them pretty well.

I know Nate mentioned in one of the podcasts that seeing workout levels drop like that week-over-week means the system thinks you’ll be fatigued, but I can’t see how to correct. It picked everything for me, and even has me doing 4 weeks on before my next recovery week.

My last base block under the new update felt great, so maybe I should just trust it. In the back of my head, I’m wondering if the base phase is the most dialed since most of the beta testing was done over winter in the northern hemisphere.

[wik04]

The other thing to remember is that the four week block it throws on your calendar isn’t the four week block you will be doing EXACTLY. It will change as you go and that is only the first draft of that four weeks.

[Torneng]

@WillS

I don’t love the 2 hour threshold days with such short intervals, but I’m trying to just follow the recommended plan.

It’s still perfectly fine to use workout alternates. Does the AI let you switch tomorrows workout to something like Icefall +2, or is that not recommended?

[WillS]

My original question must have gotten buried in my attempt to provide context. Obviously going hard every day or resting every day won’t work. I understand that there is an optimal volume that you can recover from.

My understanding is that with dynamic endurance rides I can tell the AI how much time I have, and it will optimize volume within that. So, if I take a plan that predicts an 8w FTP increase over 4 weeks and the only change I make is telling it that I have more time on my dynamic endurance days it should only give the system more room to optimize, and I’d expect the result to be the same or better. Instead, it appears to be trying to fill my available time and telling me I’ll be less fit for it.

[LacticacidMCB]

@WillS

Instead, it appears to be trying to fill my available time and telling me I’ll be less fit for it.

It is telling you that you will be training at a lower wattage. You might be more or less fit for the race you are doing. Optimizing for 28 days vs the season.

Not sure if TR has the data but maybe @Jonathan could do a podcast where they talk about expected gains during certain periods. They could look at historical data, the best ramp rate of ftp for new/intermediate/veteran rider is X. If you go much hotter than you risk plateau at a lower level. Or maybe the data shows it doesn’t matter, if you ramp faster you reach the peak earlier but you can still hold it. Or if you ramp faster you keep going up like a rocket.

Either way it would be an informative podcast for how to use the new tools and not go crazy if one is prone to it.

[WillS]

@LacticacidMCB

It is telling you that you will be training at a lower wattage. You might be more or less fit for the race you are doing. Optimizing for 28 days vs the season.

Fair enough. I do tend to win races by training at a higher wattage than everyone else so that I can show up and ride at a higher wattage than everyone else, so it’s hard to accept lower being better. I just want to know if there’s some reason the AI thinks doing more just because I have the time is better wen the result appears to be lower power numbers in my next block.

[Jonathan]

@WillS

Fair enough. I do tend to win races by training at a higher wattage than everyone else so that I can show up and ride at a higher wattage than everyone else, so it’s hard to accept lower being better. I just want to know if there’s some reason the AI thinks doing more just because I have the time is better wen the result appears to be lower power numbers in my next block.

In your case, when you increase your Dynamic Endurance duration, it is going to increase training volume on those days, so there will be a short-term negative impact on your body’s ability to perform in the other workouts. Long term, you’ll likely be better off.

I think it’s key to note that TR AI isn’t operating with a true north of “increase this athlete’s FTP before their next prediction”. I know you didn’t state that, but its easy to assume that, and I’ve fallen into that assumption before as well, hehe.

It’s also crucial to remember that the AI FTP Prediction is a calculation that, with a context window like yours with training scheduled 18/22 days between now and then, has SO many possible outcomes. The end result is likely to vary from what you see so far out with so many possible outcomes.

We want to ship an update to this that will increase the likelihood of increased precision over long time periods, but no ETA on that.

All of that said, I think you’re spot on with what you mentioned about going off your great training experience with TrainerRoad AI over the last few weeks. Let that number be whatever it is–TrainerRoad AI will give you the workouts you need to get fitter.

[LacticacidMCB]

@Jonathan

We want to ship an update to this that will increase the likelihood of increased precision over long time periods, but no ETA on that.

That would be pretty cool.

The most recent Watts Doc:

They reference a study from a gym where they had historical results from 15,000 members where they did a very rigid lifting regiment. They were able to calculate the various slopes of improvement for different members. Separated beginner gains vs experienced. Also about how the season to season gains and the within season gains for return to fitness.

I wonder with the massive data set you guys have you could do the same thing. Maybe that is what you are doing with the predictions and it’s nothing new. I thought the idea of a personal curve that projects based on others who had similar curves.

ChatGPT describes my idea better:

TrainerRoad could use a “nearest-neighbor” approach: for any athlete, build a feature profile (e.g., current FTP/W/kg, training age, recent volume/intensity distribution, consistency/compliance, plan phase, and maybe HR/power response patterns) and then find a large set of similar athletes in their historical database. Instead of relying only on a single-person model, TR would “borrow strength” from what happened to those neighbors and produce a forecast like: “people like you doing a block like this typically change by X watts over 4 weeks,” ideally with uncertainty bands (10th–90th percentile), not just a single point estimate.

This is conceptually the same move as the Dutch fit20 lifting dataset: they followed a huge number of people over long periods and showed a clear population-level pattern (big early gains, then diminishing returns), which lets you set expectations for individuals even though any single person varies. TrainerRoad could do that in cycling-specific form—conditioned on training style and phase—so a rider isn’t comparing themselves to a generic average, but to a cohort that matches their starting point and stimulus.

The payoff is better projections (especially for newer or noisier data situations) and, importantly, calibrated realism: instead of interpreting day-to-day changes in a point forecast as meaningful, athletes would see that the forecast is a distribution with a normal amount of variability, and that “meaningful change” depends on horizon and how their recent outcomes compare to what similar riders typically experience.