Any downside to manual FTP adjustment with the new AI system?

Training
291

I have been thinking a lot about the recent discussions regarding AI FTP Detection and how the model handles timeframes.

Currently the 28 day detection window acts as a stability guardrail. For most users this makes perfect sense because fitness does not usually jump or crash overnight and preventing users from obsessing over daily fluctuations is good product design.

However for racers or athletes returning from interruptions such as illness, injury, or in my case a corrective surgery that removed a limiter, physiological reality can move much faster than the 28 day window allows the UI to update.

The Problem: The Weather Forecast Analogy

Right now the 28 day prediction feels like receiving a monthly weather forecast.

The first few days are highly accurate because the model knows my recent rides and fatigue. However days 14 through 28 are statistical guesswork. Just like predicting rain 3 weeks out the cone of uncertainty widens massively because life variables like sleep, stress, and missed workouts accumulate.

If I have a breakthrough performance on day 5 such as a race or a Zwift event that smashes my priors the monthly forecast becomes obsolete. But the UI is still locked for another 23 days which forces me to train at an intensity I have objectively proven is too low.

The Request: Two Potential Solutions

1. High Frequency Detection Setting Allow an optional setting perhaps hidden in Early Access to lower the detection window to 7 or 14 days. This would be for users who want to see the volatility and are willing to accept that their FTP might go up and down more frequently. We are providing high fidelity data so letting the model react to it faster would be beneficial.

2. Visualizing Confidence Intervals Instead of a single hard number locked for a month show us the cone of uncertainty. For example it could display Current Predicted FTP 302W plus or minus 5W. As we get further from a verification effort or deep into a block without max efforts the cone could widen. If a user pushes a 20 minute power that pierces the top of the cone it could auto trigger an unlocking of the detection.

Why this helps

It solves the friction between compliance to the plan and the physiology of what the body can actually do today. If the backend model already knows I am riding stronger than the plan predicts giving me a way to officialize that number sooner keeps my training zones relevant and prevents me from having to manually override the system.

Would love to hear thoughts on if a high volatility setting is feasible for the model!

292

For reference, the styling of your posts (various aspects including title phrasing and use of em dashes) has all the hallarks of being the output of an LLM chatbot…

My view is that where numerous people are going to great lengths to provide repeated detailed replies to you to help correct your misunderstanding, it’s pretty poor form to be churning out LLM replies to them. I barely even bother reading (skimming!) posts styled like that as it’s obvious the author couldn’t even be bothered to write it themselves. :person_shrugging:

293

Yeah, I’ve seen this repeated many times, but there must be a duration the ML knows users can hold for a certain length of time, or it wouldn’t be able to prescribe those threshold PL 3 workouts - and know it should be HARD, and not too hard (or too easy).

This really doesn’t make sense if there isn’t any threshold. Jonathan has ‘reminded us’ many times in the podcast for us to remember what the ‘F’ stands for in FTP. Not sure why he thinks people have forgotten what the F stands for. I want to know the ‘T’ means! You can’t just tell people what two of the letters represent and not the third letter….Like I said, maybe they have but I may have missed it.

295

I’m honestly not so certain there is (or needs to be) a set duration in time this new AIFTP relates to - I think it’s possible that it’s different for different people.

I think if TR released models for 20min and 60min powers then AIFTP will sit somewhere in-between them but will be biased one way or the other depending on the athlete and the training that the AI determines is most productive.

296

This is my selfish reason for wanting to know what AIFTP actually means for the threshold part. I don’t think I could hold my AIFTP for more than 10-12 minutes - so kind of want a sanity check on the number.

297

About as much of an explanation as we’ve gotten is that it calibrates you to a level 3 Threshold. I think I’ve seen it posted that you might expect to get a level 4 threshold or thereabouts prescribed, and rate it appropriately “hard”. This is why everyone was getting Cloudripper -3 prescribed - it’s a Threshold 3.6

So it may match up or be close for some people but it’s not tied to a ramp test, or a 20 minute test, or an estimate of your actual physiological FTP. It’s designed to give you productive workouts within the TR ecosystem.

I think where it is challenged is that people do have different natural TTE’s (Workout Level) for threshold, so that can throw it off as an actual estimate of FTP and in certain cases can result in workouts that are effectively pushed into a different zone (although, still achievable)

300

Follow up to my own post.

I am curious about the Level 3 Threshold Calibration though, anecdotally it feels more than that for me or based on my N=1

I haven’t done a Threshold workout since last spring, I’ve done a lot of block based training (I start a lot of Threshold in about 3 weeks). But, because of that my Threshold workout level is artificially low right now (3.3) but high Sweet Spot and VO2 Workout Levels because I have been working those (8.6, 6.6)

It has me slated to bump up ~11 watts tomorrow via prediction, despite already being a level 3 Threshold.

301

At one point, TR support said they were working on a statement/definition that would help us understand what AIFTP should equate to from a physiological/performance standpoint. If they have issued that, I haven’t seen it. And honestly, that statement might do more harm that good, so I can appreciate their apprehension about adding fuel to the FTP fire. It’s a bit of a no win situation trying to pin it down to something performance based. It’s squishy concept to begin with and pinning it down to something measurable would probably just expose more examples where it doesn’t match.

Personally, I’m quite happy pushing forward just manually setting my FTP in TR to match my own physiological/performance definition. I do the same in other platforms where my definition doesn’t line up with how they calculate it. And I can totally see how my approach would have caused problems for many athletes to get proper workouts, so I understand why the TR system would have worked better for those athletes using the AIFTP number.

And I’d argue that you don’t even really lose the FTP prediction feature when using a manually set FTP. At any time, you can accept the AIFTP value and see where the prediction stands (and how it’s trending since the last time you looked at it). And then, just delete that AIFTP record from your FTP history and it’s back to your manually set value. I have AIFTP scheduled for tomorrow and I was wondering what the system is going to give me, so I just enabled the AIFTP value and looked at prediction (which I assume is what I’ll see tomorrow). For what it’s worth, that prediction says I have 0% change (318 to 318). That’s still at least a handful of watts higher than my own definition for a physiological FTP, but it’s getting closer. A month ago, I would have struggled to hold that for 10 minutes, I could probably hold it for 20+ minutes today after a month of heavy sweet spot and o/u’s. I will be manually adjusting my FTP to 310 tomorrow and I think that number is going to work well and put me in a workout level range I prefer.

302

Maybe the system is also anticipating additional sweet spot progression and “making room” for you to keep growing there.

I’ve been doing a good bit of sweet spot during base and I can see where TR would have run into issues with prescribing workouts for me if I wasn’t able to significantly extend duration on my sweet spot workouts. My original plan started with only 90 minutes allocated to my interval sessions and the system almost immediately had to recommend longer duration workouts to make the workouts continue to progress. Last week (my last SS workout before this week’s FTP bump), my sweet spot workout was a level 11 and required 2.5 hours (almost 2 hours in zone). For me, that’s where I want to be before bumping up FTP (I prefer adding duration over intensity during base), but I can see how that approach wouldn’t work for many athletes.

303

I actually think it may be the VO2’s in my case. Pretty certain my Sweet Spot Level was already pretty high at my last AIFTP, but I hadn’t been doing VO2’s yet. Basically, my VO2 work and workout level is what’s changed since then.

(Edit - either that or it’s the combination of High Sweet Spot, Low Threshold, High VO2. The AI might be looking and my Threshold Workout Level and effectively not believing it…)

You and I are similar on the Sweet Spot Approach, my last workout was going to be a Level 11, but had to cut it and end the block a little early after tweaking my calf - so rested up and went straight to VO2’s.

304

I feel the same way. It may not sound like it, but I really like the new TR update. The AIFTP doesn’t work for me, but I use all the other features, and simply turn off the AI future prediction and use my own lower FTP as an anchor. Like you, I check it once in a while to see where it thinks I am.

It is selecting great workouts, giving me the appropriate volume/ tss each week, and is accurate at the fatigue detection and adapting when needed. No complaints here.

I do think they should at least give a range for the power you can theoretically hold for AIFTP. I’m not going to lose sleep if not, but I hope they at least consider clarifying this someway.

305

Agree and that’s really the important stuff IMO. With the new system, this is the first time I can recall TR being able to align pretty well with how I like to train as a higher volume athlete. I’ve always managed my overall progression and training stress on my own because TR never seemed “onboard” with pushing higher TSS. But if I look at my calendar today, I can see TR prescribing weeks that push into 900+ weekly TSS. And I assume that will go higher as much as I continue to ramp up and recovery well. As an athlete who focused on long events and has time to train, it’s a welcome change. I still like to have my hands in my training, but feel like I don’t need to work “around” the new system as much as I did with the old. Time will tell, I’m still several weeks away from starting my build phase.

306

Whoa, first of all that’s pretty darn high and second, that’s great it knows you can handle it.

I had around 650tss last week and it knows this is normal for me, so I’m also happy that it is prescribing higher volumes also and adapting the right level for my interval days. So far the manual FTP has been working perfectly for me.

307

What you might be missing here us that the WL only represent the work you actually and recently did in each zone. They don’t represent what you are capable in a zone.

However, the AI model used to choose the workouts and calculate the FTP likely “knows" that you are capable of more in the threshold zone than your WLs indicate, because it sees what you are able to do during your SS and VO2 workouts.

308

That’s what I’m saying and I think is the case. It’s just that the only information (that I’m aware of anyways) that TR has given us is that it’s calibrated to a threshold level 3.

309

I don’t think I could hold my AIFTP (197W) for 10 minutes. I did a VO2 Max 5x5 workout just over a week ago at 196, 198, 201, 198, 196 and it felt very close to my limit (marked it as Very Hard).

313

I had this same exact experience at the start of my build phase. 4% or so bump, then first threshold workout I completed but it was rough. I marked it very hard. It bumped me down 0.5 prog level for week 2. I could not complete the last interval without adjustment. I marked it max effort. It bumped me down 0.9 prog level for week 3. I could not complete the second interval without adjustment. All the threshold workouts were the same main set, just varying in the endurance block bookends and maybe the rest interval times.

So my fitness got worse during build, which sucks. I couldn’t complete threshold 2.7 in the last week. It bumped my ftp down 1%. I got recovery this week and will give it another go next week. I’m considering manual ftp change or decreasing workout intensity from the get go…

314

I’d consider the manual FTP adjustment. It worked very well for me, as I progressed quite quickly after doing it.

It’s also nice to not fear the workouts. That kind of max effort is demanding on the soul.