This concept might work if you only used your FTP to train within TrainerRoad, but many of us use FTP as a metric for other training, outdoors, or even with other software. Not being able to know your FTP as it is measured everywhere else is a backstep that I hope is resolved with an update.
This is my concern as well. Solely within TR, I couldn’t care less if it says my FTP is 200 or 2,000 watts, but we all use our FTPs to set our race/pacing plans for events. Do we now need to go back to ramp testing to do that? Please don’t make me do that!
My “FTP” went up 10% today, which feels great at first, but I don’t think that is accurate in the real world right now.
For ages now the AI FTP has been intentionally a somewhat artificial metric in the sense that you could have an FTP of 300W but low progression levels or an FTP of 300W with high progression levels and have quite different capabilities. This is at least partly down to the 28 day restriction and the fact that it’s not a ‘live’ reflection of current capabilities, and partly due to an understanding that a single parameter just can’t describe our fitness, so we need to let a gross parameter kind of lag while we work up to longer time in zone (etc.), before taking a step change in FTP setting.
The current (including new) construct allows users to manually shift the FTP value in order to manipulate their progression (/athlete) levels and make the workouts more appropriate for them. That’s perfectly valid, even if most people might be better off just letting the AI choose.
My preferred solution would be to decouple the ‘FTP estimate’ from this control lever, so that on any give day TR can show me its best estimate of ‘true’ FTP for some definition. I am quite happy for that to be decoupled from the AI’s workout suggestions. In order to show (and allow athlete control of) the orthogonal parameter akin to progression levels we could then have a secondary metric that could vary from e.g. 0-100 to effectively show where we are in the levels. So you can be sitting at 50 and fed roughly PL5 workouts, and if you really want to then you can bump it up to 80 and get PL8 workouts (which would automatically be of lower watts, because PL8 at (old FTP setting of) x watts is just as ‘easy’ as PL5 at (old FTP setting of) y watts. That ‘slider’ would probably need to be zone-specific, but that’s OK, no different really from the current PL construct.
FTP isn’t all that fuzzy if you validate it, and pace by it. A PDC like WKO would be nice, but that doesn’t mean that the number itself isn’t important.
I want to be able to go out and do 30+ minutes at 100%, or 60-120 minutes at 90%, and so on out on a training ride or in a race without blowing up.
And for me, AIFTP was real and was accurate and validated within a couple watts. And if the number they give doesn’t track with that, Id rather have them get rid of it as they’ve said they don’t need it.
It seems a bit of a headache for TR to find a way that allows the new ML model to fulfil its potential while still being constrained somewhat by the pre-existing product architecture that it has to operate within, and the training concepts and workout vocabulary that we’re accustomed to and often still expect to see…
If there was an easy/quick solution here, I feel it’d would probably already have been done. The “two different numbers” suggestion discussed probably raises as many (or more!) problems than it solves, while the “moving some things beneath the hood” suggestion likely means a massive engineering effort, not least because of the UI impact throughout the product… If TR have been working on the core technology here for 3 years, they understandably want to get this out the door now, as do we!
Without meaning any disrespect to them (just being practical), TR have had to “shoe-horn” this new underlying technology into the existing product somewhat, and it seems like they’ve done a great job of this, with the result looking very polished. Hopefully (“FTP number” debates aside…) the new ML training model proves its effectiveness in the months/years ahead. If TR really have nailed the core technology, it should allow focus to shift to leveraging it ever more effectively with new features, like the monte carlo stuff and novel plan structures that’ve been raised, plus a bunch of useful forecast performance metrics to suit different use cases. Could be a while though.
@Dark-Passenger If anything TR are getting closer to the meaning of FTP, as “descriptive not prescriptive.” amongst other things. I’m loathe to get drawn in to this kind of discussion again, but I understand your concerns. FTP has been a handy number for a long time.
My thoughts got to Luke, approaching the thermal exhaust port:
OBI-WAN: Use the AI, Luke. Let go, Luke.
DARTH VADER: The AI is strong with this one.
OBI-WAN: Luke, trust me.
REBEL: His FTP’s off. Luke, you switched off your FTP tracker! What’s wrong?
LUKE: Nothing. I’m all right.
So now you have decided that you do in fact trust it, but not the new model, and that doesn’t change the fact that both models are made up “AI” “FTP’s”, which is exactly what I said.
Except for a lot of us the “made up” AIFTP actually equals our real world FTP, which is why we trust it and are skeptical of something Nate says may not match up.
All the ftp numbers are estimates of a metabolic steady state, a steady state which fluctuates and is not fixed…what do you mean by “real world ftp”?
I’m getting sucked in.
My AI detected ftp and ramp tests usually align, my Garmin detected ftp is usually a bit different, and I haven’t done a 40k TT in a long time. My ironman power is usually approximately 66% of any of my ftp.
I’m extremely skeptical of AI FTP. Maybe just because it didn’t work well for me. So, maybe (almost certainly) that clouds my comments. I believe people who try to verify it just test to the number, which skews real world training. That’s why I love this new model. It doesn’t care what you think your FTP is on any given day, it just gives you the right workout for that moment based on multiple variables.
I get how that doesn’t work in the real world when not following a workout, but in that scenario, I would not have trusted the AI FTP and would have tested anyway. Again, I see where that’s skewed to my personal scenario. I just think some are missing where their own situation in believing the AI FTP is also skewed to their own personal scenario.
Seeing now what Nate has actually said about the new AIFTP, I’m a little more confused. The most important question for me is whether I can use this number for pacing outdoor real world rides and races. For me, my FTP of yesterday would give me a much different pacing plan for Leadville this summer than what I’m seeing today.
Hopefully someone from TR can give us some guidance here!
The amount of conclusions based on interpretation of Nate’s comments without trying it is concerning. I think pulling out of this conversation makes way more sense than jumping in
FTP was never designed to predict our performance like this but lots of us used it this way, for some it worked, for some it didn’t, for some it meant we went easier than we could have…there is no way to know for sure.
Also, very little science behind those predictive percentages.
I would still use it as a base prediction, but guided by your actual cycling performance in training.
My opinion: Real world FTP is the steady state power number that corresponds to MLSS, best estimated by long form tests like kolie moore tests, worst estimated by shorter duration tests with larger anaerobic contribution. You should be able to hold that steady state power for 40-80 minutes in the real world (sub 60 min for most)
And for me, AIFTP has been dead on with that number, which agrees with WKO when I feed it the right efforts. If you were to take my current AIFTP and increase by even 2-3% it would 100% put me over threshold at a 100% effort.
Hmm, except that we know that steady state fluctuates so if you tried to hold that same number today and succeeded, tomorrow you could fail, the day after you could go harder.
We need to accept a level of uncertainty and always have, and that we are unsure how much uncertainty there is.
