Perplexed by the latest iteration of Adaptive Training and "AI"

[eddie]

I haven’t looked at your career yet, so I don’t have any idea what to expect.

It’s worth mentioning that I genuinely have no personal interest in steering a narrative in any certain direction. If there is something we at TR can discover that can help us improve, that’s a win for us.

When our perspective of a situation differs from an athlete’s, things can be hard to decipher in a public setting without a full understanding of what’s actually happening behind the scenes.

We work for TR, and look at athlete accounts all day long. We know how the product works, and also the places where it doesn’t.

The goal here is to bring everything down to a level playing field so that we can understand how our athletes are using the product, where issues lie, and where things might be misunderstood, or misused. That way either we’re going to learn how to improve, or our athletes will learn more about how the product is intended to work..

Essentially, the feedback we get here is highly valued, but in order to have a productive discussion where we can bring our expertise into the discussion, we need to make sure our athletes are okay with it. If there’s every anythig you don’t want discussed in public, feel free to reach out to us directly through DM or support@trainerroad.com.

When there are discussions happening in public places where we feel like we can help, we want to get involved!

[andho]

Following up on my post above.

With 1 day to go, I did a60 mile road race that was a “B” event on my calendar since start of my block. No power PRs, but I felt better than I had ever felt on this race and made the selection for the final break of 5 of ~30 starters. The race was attrition in cross winds and relentless kickers. Best finish in 10 years in this particular race. Missed the podium by a bike length.

Today I did a 20 min effort up a local climb. Did not do the full 5 minute blowout effort but did 3 high VO2 1’ efforts before. This was nowhere near my PR on the segment but 95% of my 20’ power was 15W higher than my AI predicted FTP. Again, I did not do the full 5 min blowout effort and this was outdoors on a climb. So, I am not arguing that my 95% of 20’ power should be what AI predicts.

Having said that, I conclude that the training is working in terms of real world results. At the end of the day, that’s all I care about. I wonder if the AI predicted FTP indicates what I could hold for an hour on an indoor trainer, seated, with a steady cadence. Since the road and XC events I race do not involve sitting on a trainer for an hour of steady pedaling, I plan to continue with the plan and just ignore the AI FTP going forward.

I also will second other posts that have noticed you can ‘game’ the system by just rating everything ‘easy’ … if you want to see the number grow, that’s the secret sauce.

[AussieRider]

@eddie

I haven’t looked at your career yet, so I don’t have any idea what to expect.

Look forward to what you come up with!

[eddie]

Me too! Is this a good place to continue the discussion about what I find?

[AussieRider]

Yes, If anyone else can learn from the challenges I’m having (or had?), then that’s the best possible outcome.

[OreoCookie]

@AussieRider

Has anyone else experienced this? Am I misunderstanding how Adaptive Training interprets workout ratings or progression?

Ironically, despite my manual increases, FTP prediction has me on track to gain 2w this month.

Looking at your calendar and assuming it is representative of your recent training, I think it is pretty clear why TR AI does what it does. A few things I have noticed:

• The first week has workouts on 7/7 days, no rest day. Immediate giant red flag.
• Second week has five or six workouts (depending on whether you ahve done the strength workout). It includes a “recovery ride” that’s 2 hours long at an IF > 0.6, i. e. an endurance ride.
• On some days you have two workouts cycling workouts, on one you seem to do two VO2max workouts, one is 4:15 hours long!
• Your training schedule seems very inconsistent.

Here is what I would do:

• Decide on a training schedule and stick to it.
• Include rest days. I’d start with 2 rest days where you do nothing. No gym. No “2-hour recovery ride”. Just rest.
• Decide on the hard days. The entire rest of the week should be designed around these key sessions. For instance, you don’t ever want to double-up on VO2max workouts. You should do one VO2max workout that’s so hard that you are done for the day (on the bike). You shouldn’t even be able to do another workout. You shouldn’t tack on a very long endurance ride either.
• If life stresses allow, schedule your gym sessions on your hard days.

[AussieRider]

@OreoCookie

Looking at your calendar and assuming it is representative of your recent training, I think it is pretty clear why TR AI does what it does. A few things I have noticed:

• The first week has workouts on 7/7 days, no rest day. Immediate giant red flag.
• Second week has five or six workouts (depending on whether you ahve done the strength workout). It includes a “recovery ride” that’s 2 hours long at an IF > 0.6, i. e. an endurance ride.
• On some days you have two workouts cycling workouts, on one you seem to do two VO2max workouts, one is 4:15 hours long!
• Your training schedule seems very inconsistent.

The assumptions you’re making about fatigue and inconsistency don’t match the actual structure. The weeks in question had two intensity days, mostly easy volume, and built-in rest.

I’m adjusting sessions based on how I’m absorbing the work, not randomly adding load. If anything, the issue I’m seeing is under-prescription from TR, not overreach.

Mixed interval work isn’t unusual; it’s widely used when there’s a purpose behind it. The idea that VO2 work can’t be combined or extended within a session isn’t really accurate.

In this case, the added work didn’t compromise the quality of the session, and it wasn’t turning into a second maximal effort. It was still controlled. That’s the key constraint, not whether intervals are mixed.

[OreoCookie]

@AussieRider

The assumptions you’re making about fatigue and inconsistency don’t match the actual structure. The weeks in question had two intensity days, mostly easy volume, and built-in rest.

I did not make any assumptions beyond supposing that the two weeks you posted were representative. Durations and IFs were read off as they were. One week had literally zero days off, for instance.

We can quibble about all of this, but I still think a week where you don’t take a single day completely off is a mark of you doing too much. And 2-hour rides at > 0.6 IF are too hard to count as recovery rides (1-hour max, < 0.55 IF).

@AussieRider

I’m adjusting sessions based on how I’m absorbing the work, not randomly adding load. If anything, the issue I’m seeing is under-prescription from TR, not overreach.

You asked us for input and feedback. I for one am not surprised at all. That’s all.

Regarding under-prescription, are you sure it really is under-prescription? Or just a better plan than what you are doing now? In the short term, doing more can feel better, it feels like you are trying harder and all that. But when you are under-recovered to really benefit from hard sessions (≠ being able to do hard sessions), then you are stunting your growth.

I got to my second-best fitness this year with significantly less structured training. Although I do spend 2 x 25 minutes commuting in Z2, which add up to an additional 5–6 hours once I include grocery runs and errands.

@AussieRider

Mixed interval work isn’t unusual; it’s widely used when there’s a purpose behind it. The idea that VO2 work can’t be combined or extended within a session isn’t really accurate.

But is mixed interval work the right thing for you at this very moment? Mixed intervals are typically not about making you faster, but being able to perform intervals of a certain type (e. g. sweet spot) after riding for several hours. This is training for race winning moves.

@AussieRider

In this case, the added work didn’t compromise the quality of the session, and it wasn’t turning into a second maximal effort. It was still controlled. That’s the key constraint, not whether intervals are mixed.

How do you know that? Because you are able to finish the workouts? That’s not a good measure, especially if they are adapted for your state of fatigue.

[AussieRider]

You’re drawing some pretty strong conclusions from two weeks of training without any real context. I’ve seen your posts before and know you tend to provoke. I’m not interested in engaging here, so please take it to another thread.

[AussieRider]

That’s a lofty prediction. My guess is I will do the prescribed workouts and this will readjust the morning of to a 2w increase.

[eddie]

Since you’re doing your workouts exclusively outside, you’re running into the same bug as a bunch of other athletes, where your predictions are inflated. Your detections are still accurate, but you start with a high prediction, and it drops over the month to your detection (which is still a fitness bump in your case).

We’re working on that now. It’s proven to be a complex issue, so it’s taking longer than most bugs do, but I think we’ve recently made a bunch of progress recently, so hang in there.

In terms of your workouts being adapted down, as your calendar stands now, it looks like you’ve got a solid week on deck next week. The two hard workouts that you have scheduled are decent jumps from what you did this week, and based on your survey responses from those, I’d say things are working as intended there.

If anything changes between now and those hard workouts, let me know, and I’ll take another look.

---

Looking back, I wonder if we were trying to slow you down a bit because you often spontaneously extend the duration of your workouts quite a bit. If we expected your two hard workouts to be 90 minutes each, but you end up riding for two to three hours each day instead, that affects how we expect you to adapt and recover, which means that your upcoming workouts should change.

Over time, if you’re performing well consistently, how the system reacts in those scenarios will change, and maybe that’s what I’m seeing now ?

If next week’s workouts remain the same, and you get through this week without a load of yellow days, that could be a sign that the software is adapting to your training.

You did have a dip in TSS between late October and January, which could have caused things to be a bit more conservative than you were expecting, but your 6-week TSS is back up to where it was before that dip now.

Let me know if this makes sense and if you have any questions.

I’d like to stay in touch and see how things go for you wrapping up this week and into next week.

[AussieRider]

@eddie

Let me know if this makes sense and if you have any questions.

I’d like to stay in touch and see how things go for you wrapping up this week and into next week.

Sounds great. Thanks for the thorough analysis and thoughtful feedback.

@eddie

Since you’re doing your workouts exclusively outside, you’re running into the same bug as a bunch of other athletes, where your predictions are inflated. Your detections are still accurate, but you start with a high prediction, and it drops over the month to your detection (which is still a fitness bump in your case).

I assumed it might be a bug, so thanks for acknowledging that. It is still in beta, after all, so I wouldn’t expect it to behave like a fully finished product.

What’s interesting is that it’s currently predicting my FTP will stay the same after this four-week block. I wonder if that has something to do with being in the specialty phase leading into my A race at the end of the month.

For a seven-hour event, I’m a little surprised this TR iteration hasn’t included more threshold work. If you exclude Barker (March 26), which was more of a mixed Z4/Z5 session with 1- to 4-minute intervals, that leaves 6.5 weeks until race day without any sweet spot or threshold work.

That said, I’m excited to try something different. This plan has incorporated a lot more Z5/Z6 than I’m used to from TR- so it’ll be interesting to see how it translates on race day.

I just finished today’s workout and made a couple of adjustments, which I’ve pasted below. I significantly shortened the recovery periods (from 10 to 3-5 min) and increased the interval power slightly, but overall, the session felt about right.

I’d much rather finish with a little left in the tank than be right at my limit.

Thanks again. I’ll keep you posted.

[eddie]

The Rolling Road Race plan moves away from sustained work as you near your A race.

It’s designed to prepare you for shorter, harder efforts that you’ll likely experience during rolling hills and other types of surges.

[AussieRider]

@eddie

The Rolling Road Race plan moves away from sustained work as you near your A race.

Oh wow, this was definitely an oversight on my part.

Back in January, I must have selected the rolling road race option before I had a specific event in mind. Then in February, I committed to a 120-mile climbing road race, added the event, and adapted the plan from there.

Looking at the plan settings now, I can see there’s no mention of which plan type you’re on, or any obvious way to change it once it’s already started.

So this one’s on me. I’m really grateful for your response because I absolutely would have missed it otherwise.

I’ll add some threshold work back in over this last month.

[eddie]

You can change the discipline of your A race and we’ll ask if you want to update your training.

Gran Fondo is a great plan for sustained work and a balanced approach to riding a long course fast, but on your own terms.

Climbing Road Race is the best option for competing for the win on more sustained courses.

Let me know if you have any issues getting your plan changed over!

[AussieRider]

Hey @eddie,

I had a question out of genuine curiosity about how the system is adapting this season.

On March 10, I did two back-to-back VO₂ workouts with PLs of 6.4 and 5.3, both rated Hard. After a recovery week, a week with Zone 6 work, and another mixed VO₂ workout at PL 5.4 (also Hard), this week TR scheduled a PL 4.6 VO₂ workout.

I trusted the AI and did it as prescribed, but it ended up feeling too easy: 5 sets of 5 x 1-minute efforts at 341w with 6-minute recoveries. I averaged 362w on the work reps, cut the recoveries from 6 minutes to 3 minutes, and still rated it Hard.

Next week it has me on PL 5.6, which I’d expect to feel more like a 7/10 effort. My curiosity is why the progression seems to have stepped back despite several recent VO₂ sessions being completed successfully.

Could this be related to the newer FTP detection from outdoor rides, making VO₂ recommendations a bit conservative?

I’m also curious how TR treats workouts where I go well over target. If I’m consistently ~20w above prescribed power and shortening recovery, but still honestly marking the workout Hard, does the system account for those changes, or mostly just the survey response and planned workout structure?

I know TR usually figures it out, so this is really just me trying to better understand the logic, as this season the progression has felt more conservative than in the past.

[eddie]

Both of the hard workouts on your calendar next week appear to be productive.

I’m seeing a level 4.7 Anaerobic and a level 5.6 VO2 Max. Your current levels are 5.4 for VO2 and 4.6 for Anaerobic.

I do see what you mean with the power curve, though.

I’d try to avoid chopping up the workout structure too much if you can. Shortening rest periods might not be the best way to make a prescribed workout harder. If the intervals aren’t hard enough, I’d increase the intensity or use Workout Alternates to find something harder.

I’ve seen other athletes do this in the past, and at a certain point, the work you’re doing doesn’t resemble the prescription much, which I think becomes an issue.

I am noticing that most of your PRs for 5 minutes or shorter over the past 6 weeks have come from unstructured rides. My advice here would be to do what I suggested above if the prescribed workouts don’t seem hard enough, and then follow the structure of the workout closely and rate the work based on how it felt. That’s the best way to get TR to dish out the right workouts.

Let me know if this helps and how things go this week.