Ha, that’s a funny way to look at it! I also plan to creep mine up manually as I see my fitness improve.
I’m hoping my detection is on point so I can just let the system do its thing. I feel like, even though the manual correction was completely necessary, that I am only realizing some of the benefit of the AI tool. I want to see how it pushes me with a better understanding of my capabilities (I partly blame the new power view, which is cool).
If, after getting these additional weeks of data, it nails my next prediction, I’ll be super excited.
I’m a little scared of how much fun I’m having just nailing these workouts. I can’t neglect my road bike too much. Hah. But it’s fun to be dialed in.
Alright, I got home from work and decided I wanted to take a look at this, and re-accepted my AITFP of 258W (up from my manually set 245W).
Some context: I recently changed my current training plan from a Gran Fondo focused plan (all the way out in November) to a general FTP build. When that happened, my workouts changed from the 1 hour mark to mostly between 1:30 and 1:45.
Saturday’s threshold workout loses some time in zone (castle peak +1 to Brush). The unders are on the intimidating side (253W as opposed to the 233W scheduled in castle peak) while the overs are 276W, so more or less what I’ve done in recent weeks. I’m thinking this should be doable given the short duration of the intervals, but it will certainly be uncomfortable. One factor in my favor is I do better with an extended warm-up, and that’s baked into the workout already.
The nice thing about the increase duration of the workouts, compared to the previous plan, is that I don’t have to experience short recoveries with higher power figures than I’m used to. My next Sweet Spot workout, Glassy +1, is mostly a decreased on power versus earlier in this week with the smaller FTP value, with only about 10 minutes of threshold work thrown in across two different sets. Monitor +3 follows that up, with a small increase in power for each interval in the set, but with more recovery baked in. Same as before, it dips up into my self-chosen FTP range, but only briefly.
Starlight -1 is my final threshold workout before my rest week at 245W/271W. 271W for two minutes sounds scary, but it may be totally manageable after this next week of training.
My AIFTP prediction is 263W. We’ll see how that feels after this rest week.
Maybe my self-selected FTP workouts have made me overly confident, but I think I can probably do this, so I’m going to give it a try. Focusing harder on recovery and better nutrition, I may be better able to absorb these workouts than I expect.
Wish me luck!
Thanks for your insight! Yes, shortly after getting over a head cold, a bad cough infected our household. Back luck!
I’ve been having some issues whereby my predicted is starting off with some chunky gains at the end of a month (and I have finally started using the software ‘properly’ rather than sort of vibing it), but after completing, successfully, workouts outside, the predicted is dropping and it gives me much easier work following up from the workout I’ve just completed, rather than ‘sticking to the plan’. I have taken to adding in some high threshold workouts (3x18@99% with hard starts) to ‘prove’ to the computer that I can take it - and even then, the predicted is still 5w lower than it was before I did the workout this morning.
I thought this might be due to an issue I’m having - the laps aren’t matching up correctly owing to small pauses - not during intervals I should add, often just a few seconds at the end of each ‘open-ended’ before starting the next set. Sometimes I take a longer warmup to run an errand or go to the Engen to get some sweeties…
However, if I re-export from Strava rather than importing from Hammerhead directly, everything matches up again correctly, but there’s no improvement in terms of how ‘well’ it seems to think I’ve done.
Essentially, I’m confused as to why every time I successfully complete an outdoor workout - all my workouts are outside and are scheduled as such - the predicted drops, regardless of how well I perform.
Anyone have a similar thing happening? (I did submit a support request for this a week ago and haven’t heard back.)
I think I’m starting to get it, the pieces are falling into place. I think @Bones was the one that helped, though many others have through their posts here cutting through the glow of the gaslight.
I think it’s vitally important that people try the Kolie Moore FTP protocol, just try it if you haven’t. As much as people like to talk about ftp as this ‘arbitrary’ ‘always wrong for someone’ protocol there is still at its origin truth, and that truth is based upon a real thing. He who shall not be named on this forum used it as a fulcrum, or…maybe the ‘origin’ of the graph. There are good reasons to do so, real physiological reasons that have been borne out over time, which is why it is still the dominant artifact for training with power.
When you do the Kolie Moore test (if you follow the protocol correctly) you’ll get to this balance point, you’ll feel it, over which things get much harder. You could, if you are curious, even go over, and then under this point, feel it get harder, then recover. This is of course not a simple plot line, you will fatigue and that fatigue (which is the hamstring of the original FTP, and explained away by saying “well-trained”) will shift this balance point down the power scale. Something otherwise known as TTE. TTE is complex, (as all of this is) and can come down to muscle or even ligament resiliency, energy availability, hydration, heat, stress, whatever. But it doesn’t change the fact of the balance point. Some might call this MLSS, they are naming that as a reason the effort is suddenly more challenging. It’s called a ‘threshold’ for a reason, and this protocol teaches you to ride it by feel.
Now, in the AI FTP world IF the model is giving you a number significantly higher than that balance point…a few things happen. While the AI isn’t going to give you a workout it thinks you will fail, it is not prescribing what it says on the tin. If your ftp is 10% too high, then an over under (which has been said in this thread) is now a threshold → supra threshold workout. One you can achieve (possibly, more on that later) but nonetheless, it is not an over/under. That’s not moving my cheese, that’s calling velveeta cheese, and saying, hey it melts better and what is cheese anyway? 
This should set off alarm bells for anyone with a long training history. We can debate whether this model has the ‘right idea’ or not, I guess the proof is in the outcome of the training, but there is plenty of reason for the spidey sense to tingle. I was on Demanding, and I noticed that it said my ftp was going to go down, so I changed it to balanced and it said, hey 20w increase. I thought it was laughable, but now I get it. It was planning on scheduling me for supra-threshold efforts, and sweet spot that is actually threshold and ultimately vo2max that is anaerobic and it was rightly saying, if you try to do demanding you will perish. Good lookin’ out little AI buddy.
As for whether you’ll succeed at its prediction…well that really depends. For reasons that can only be described as goofy self experimentation I have not long ago pushed my body into an uncommon power curve. You could find these curves probably on a lot of ultra-endurance, some ironman distance triathletes possibly, anyone with a lot of low and slow, or even a ton of sweet spot and no threshold or above work. I have at a number of times in my training had very stable and reasonably large threshold values, and would have no qualms riding 60 minutes at threshold. And often my vo2max was poor at best. That sort of profile is a disaster in this new model. a 20 minute threshold effort wouldn’t have been enough to get my HR above endurance values, you can imagine what my ftp would’ve been. It’s a double whammy, a weak vo2max and a vastly over estimated threshold.
When I recently joined back up to try all this fancy AI stuff it vastly, over estimated my ftp. probably near to 10% too high. I even did a ramp test to convince it it was wrong (I didn’t know) and then just set it. And the whole time it was trying to bump up my ftp. Even at my adjusted ftp it was programming over unders that really pushed the boundaries, I had to mark one as very hard and it finally relented a bit and lowered my predicted ftp 9w. But the threshold workouts come much closer to what I would call a vo2max effort breath wise, happily shorter durations.
Luckily for me I was pretty detrained before re-onboarding, so I can be entrained by a new model and see how it goes. Good to mix it up. So I’m not dunking on the AI here, I’m just sayin, I see you…
Well thanks for coming to my TED talk Final thoughts: I guess the danger in this AI is similar to the danger in a lot of AI, it will atrophy your expertise if you have any, and it will rob you of learning if you don’t. I’m not down on TR, or AI (I use AI for work and pleasure every day), but I’m well aware of the weight on the other side of the scale. I tend to prefer transparency over expediency, even at the cost of complexity. But that’s just me.
I rather enjoyed your ‘ted talk’ - it’s refreshing to read something you know wasn’t AI generated!
I share a lot of what you are reflecting on. It’s a big advantage to have years of training experience and knowledge, but also a curse as you run the risk of over analyzing things. Guilty on the over analyzing.
I’ve debated on the fact TR may have a sort of ‘secret sauce’ behind this AIFTP, but I can’t accept it as I’d have to deviate from core training principles around energy system development and TiZ to address strengths and weaknesses.
I believe the AIFTP as a functional anchor is simplifying things a little too much. Do I believe it works, yes. Do I believe it is optimized for me, no. TR must know this also, but I can’t blame them for rolling this out in this fashion - it will make you faster with the right level of adherence.
As you mentioned, over / unders have to be performed where your under is actually under. Without this, things fall a part and TiZ will be greatly compromised. SS intervals need not be too hard as you need to progressively advance these. The way I’ve gotten around this is to set my FTP at the appropriate level for these workouts, where I have room to grow them before changing my functional wattage. I mentioned this in another thread - my predicted AIFTP is now higher than it gave me using TR’s FTP and my workouts feel correct (I know super subjective, but experience has to count for something).
Edit to add: TR must have a roadmap that will further optimize to individual athletes, so it’s silly to treat this as a final product. This release is a huge step in that direction IMO.
Following up here, they got in touch and basically let me know that the reason my predicted was dropping was because I was putting myself into too many yellow days. I’m not entirely satisfied as the TSS I performed was literally 2 higher than what was in the original prediction, which doesn’t seem to equate to a -16w/5% difference in FTP, but hey. I’m just going with the flow now and trying to make it as sustainably difficult as possible. In the process, I feel like I’m getting better at 1. trainerroad workouts and 2. cycling, so that’s a win.
Jonathan has a great instagram post out about preparing for Leadville.
It appears he is targeting 4.80 w/kg for a sub 7 hour finish, with the watt portion estimated by tr AI FTP detection.
I take this to mean that TR AI FTP estimation does in fact try to estimate maximum sustainable power at LT2 - as this is my understanding of what is generally understood about w/kg goals for performance on long events like Leadville. I absolutely may be wrong. This would be a departure from TR AI FTP detection simply normalizing around threshold level 3 productive workouts
Who knows? TR has been pretty coy on how they actually define their version of “FTP”. I don’t think we are getting an answer at this point.
Is TR “FTP’“ like Strava’s fitness score? Just a number that represents how your fitness is for you today compared to your personal history, and not how you compare to others?
No. Strava fitness score is just CTL.
The reason I ask is I am seeing another 20% FTP gain since the last update. That seems to be unreal. In years of training, I never seen gains of 20% let alone two back to back.
I can’t speak to how tr determines their version of “FTP”.
From what we have seen it does not necessarily correlate with what is the common definition of FTP used elsewhere.
But if it is going up that probably means your fitness is also increasing.
You could look at your pdc for each block to see improvements or go do your own testing up a local climb as a benchmark every few months.
Is that 20% gain a prediction?
First one was 20%, and at the beginning was predicted to be a 20% increase. The second one is predicted to be, so with what happened last time I am expecting the prediction to be close.
@Nate_Pearson is there a way to see if changing from a demanding to a balanced approach would yield better results without overrriding the current plan. My current plan is working well and i seem to be making much better progress since i moved to a master’s plan instead of the 3 hard days a week but i’m always curious if i can do better. Just. don’t want to override my current plan if not necessary.
Just go in and play with the sliders. If you change them back everything else will go back too.
Hard to say with seeing what you’ve done in the last 4 weeks and what you have planned for the next 4 weeks. 
For all we know you may have a week of holiday next week.
Generally you were doing too much hard work.
Have you tried using a custom plan yet?
Have you unpinned the threshold workouts?