I scrolled down the page and saw the first part of the message, you seriously got me. I was about to be like 
give me a break!
2 Likes
Some interesting results from the past few weeks of training
About a month ago, estimated FTP was 338, which seemed reasonable given the data available to it.
One month later of training, including two pretty notable signal workouts:
2x20 HIT workout of 30/30s and 15/15s at an average power of 350 and 340 (366 and 350NP)
1x60minute burst SST workout at an average of 322w
AI detection says FTP is 328w
Real FTP is probably ~345w with a TTE of 40-45mins given my previous state of training and how I feel right now compared to the past and the 2x20 HIT workout is pretty indicative of that since they were around threshold RPE and threshold HR.
I don’t do any off the shelf TR workouts so I’m guessing there is still some room on the unstructured ride classification. I replicated the workouts in workout builder and associated them, but the classifier for custom workouts still gives crazy results for the progression levels.
3 Likes
I’m just wondering about how AI FTP treats a recovery week. Does it matter if i have AI estimate my FTP after the last intense workout of my build phase BEFORE the final recovery week? Will my FTP rise in this recovery week because of the training adaptation that takes place or will it drop because of the drop in workload?
The best way to use AI FTP Detection is probably to wait till after recovery and have AI estimate the new training base number. But i’m still curious if the AI conciders “training adaptation” or is more driven by pure workload?
Conceptually, and at least from my own peeking, I think you want to do the AIFTPD after the recovery week. It is aimed to replicate the results of a Ramp Test, and unless you had a habit of doing a RT before the recovery week and using that, I think it’s a misuse (timing at least) of the tool.
I essentially tested this with AIFTPD peeks 3 weeks in a row:
- Before the final work week = 247w
- Before the recovery week = 248w
- After the recovery week = 246w
All follows what I would expect in a general sense (certainly could be ‘noise’ too), and means that the AIFTPD likely is best wherever you’d actually test.
1 Like
I’m not sure it is as explicitly intelligent as you are implying. At least I hope it’s not trying to be, since I don’t see much utility in that. Should be an objective number that gives best case scenario on what intensity should be, where what is actually prescribed is based on current PL and choosing appropriate workout level depending on load/de-load cycle.
I’ve used it midway through a recovery week, and then 2 weeks later in the middle of a block and got similar results in increases. Hard to say from that, but while it has technically stated it is meant to predict ramp, I think what is actually meant is that it will predict what an ideal ramp test would give you with current fitness without the need to be rested in it, etc.
those are my exact numbers too, still pissed it stole 2 W from me
edit same age too, though i’m 11kg heavier 
1 Like
This exactly answers what i was wondering about. So my thought about your FTP going down while in a recovery week is probably correct. But as @jjmc says, estimating FTP on the day you would normaly ramp test is still the best moment to establish the new baseline for the upcoming phase.
I added in a ramp test today just to see what it would come up with and it gave me a +8 points after 2 weeks of build on a nice round number of 260. I have a minor case of OCD on round numbers so I was hoping to get this number after recovery week, but concidering both your answers it’s probably going to be like 258 or 259 after next week. Not that it matters, a few watts won’t make a world of difference.
Thanks for the responses!
1 Like
I don’t do this, but just out of curiosity, is anyone here simply using FTP detection to update their FTP every two weeks? 
That’s not what I was saying. I don’t think it matters much at all what day you do it. There are reported fluctuations that are likely due to inherent noise in the ML model.
Not really any reason to do it more than 1x every 4 weeks, but not really any harm to doing it more or less frequently either. Shouldn’t matter if you do it at beginning or end of deload week (or any other time). Numbers will change if you check them more frequently, but that shouldn’t affect training in a significant way since PL adjusts for those variations.
That said, there is evidence of kinks in the system that challenge that thinking. Like what @MI-XC has reported, and may be related to a timing nuance of some sort. I don’t have any explanation of seeing a change like he did that seems as not a good estimate to accept.
I did it on release day and then 2 weeks later just to test a hypothesis that it doesn’t matter much. But I don’t plan to make this a regular thing. It’s not a useful thing to spend much mental energy on really.
1 Like
From your post in FTP detection results thread https://www.trainerroad.com/forum/t/your-ai-ftp-detection-results/70285/9 you said:
I did a ramp test and it was 1W lower than the AI FTP prediction.
Folks, take that ramp test on the same day for a proper comparison.
I’m squarely in the camp that a ramp test result matching the FTPD should not be used as a validation. Nor should any other FTP estimatation method or test. Doesn’t mean these won’t often be close.
FTPD is all about trying to best set training intensity. Hard Stop.
My concern is that if we take the mindset that test results like you indicate are the standard of validation - rather than tracking what really matters, long term performance gains - we will distrust the approach just based on if it doesn’t match. And there are many reasons why that might occur that have nothing to do with its utility.
6 Likes
@jjmc Thanks - makes sense! Reason I asked was because I was curious today and used FTPD and it gave me a 10 watt increase on my last result, two weeks ago. Seems a big increase for two weeks’ work so I’m sticking with my old FTP for the rest of my training plan…
@DewiJ for some reason there is a bug where often when I quote it immediately edits and deletes the quote. Anyways…
Yes, I got the same thing as you - a large bump after two weeks that I think was a stretch at best. And accepted it just because I continually argue that PLs correct for large variations in FTP input.
Results are early, but thus far this has held true.
So statements like “I accepted and my workouts are spot on” are less a validation of FTPD and more about AT setting PL extremely well over a wide margin on FTP input.
The goodness of the FTPD will take much longer, and be much harder, to flesh out. That can only show up through tracking progress through different ways of setting FTP, or more generally, if there are ideal PL ranges in different zones for different goals or individuals. I don’t think that is generally true, but there are likely special circumstances where it is for sure.
+1 to this.
FTP (whether from ramp test or AI detected) is just putting me in the right ball park, AT will then adjust things so I’m playing with the right team. I don’t think you can really use one without the other.
Interestingly Intervals.icu currently has its guess at my FTP at 274W as opposed to TR’s 297W, most likely because I haven’t done any of the 3 minute plus hard efforts for a while that it bases its calculations on.
My first workout post AI determined FTP was Keith-1, basically four mini ramp tests with surges! That was hard and I needed a back pedal in the final two blocks but I’d have struggled at my previous FTP of 288W as well, maybe not quite as much. I do seem to struggle with that style of workout - continuous ramping up of effort with no breaks, I struggled with the Basin series a while back as well.
3 Likes
Extremely well stated! And 
FTP input is meaningless in AT without also comparing against PL.
Regarding workouts, the levels are not as accurate as one would believe based on the granularity in the scoring, and my hunch is that it comes down to accuracy in the FTP input. I get a workouts scheduled I scratch my head at and/or the workout level doesn’t jive. But by and large I just ignore them. Part of it may be that volume and intensity/recovery is not being handled explicitly by AT yet. So it is up to the user to either do the workouts and struggle and system to correct, or user to proactively correct through selecting alternates. Additionally, it could just be my own personal strengths and weaknesses with certain workout structures. Likely a combination of the two.
That’s interesting because I find workout levels to be quite accurate and one of the best additions of all the recent updates. Within the same training zone the workout PL has greatly improved my training. Not to mention, has helped me understand why I struggled with some workouts in the past prior to PLs.
Workouts’ PLs are such a better why to judge difficulty over IF, TSS, duration and/or % of FTP.
7 Likes
I agree on all that. Accurate is a funny word without qualification. I just mean they aren’t perfect for a few different reasons. But they are a huge huge improvement that is my favorite part of AT by far.
1 Like
So everyone was doing meaningless training before AT? What about all that stuff that Coach Chad said and wrote about the importance of having an accurate FTP? It’s all meaningless huh? I don’t think you realize the implications of your statement.
Some interesting points of view in that series of posts.
5 Likes
Yeah, agree on the oddity hiding in that statement.
AT (and the related PLs) are just new tools that largely replace the old advice and recommendations of watching our workouts and making tweaks to suit our needs in each moment.
AIFTPD is just an easier tool to getting a Ramp Test value. It didn’t rewrite what FTP is or how it is leveraged in TR.
The mix of the two may allow for more margin of error in ways, while still getting the rider to move in a positive direction, which is beneficial. Notably it is essentially automatic so it removes most of the need for the rider to do the research (via TR resources as in the past) and makes it darn simple.
I love the combo and it’s working well for me, but it’s not like these changes moved us from the stone age to the bronze. It is still leveraging the same basic principles that TR used for years in the old days.
7 Likes
Yup.