Yeah I don’t know what to tell you. Looking at the last 2 years, with just 2 samples I can predict just as well as the AI FTP of TR what my ftp will be based upon my ctl alone. Any two samples. Maybe I’m just one of those squarely living in the bell curve of mediocrity.
I’m downloading my strava data now to try that intervals thing. looks pretty interesting. 
I uploaded my data and intervals gives me the basically the same result. eFTP at a given training load is pretty consistent. The model is ‘periodic’ at predicting my ftp because I didn’t have enough hard efforts I guess for it to make an estimate. Where it does move the eftp needle it correlates strongly to training load.
In fact when I dropped TR because my ftp was flat and declining and I felt horribly burned out, my training load was consistent. Intervals shows a much lower eftp than TR was suggesting I had. IF you map that load to my eftp it is exactly what I would predict as well based upon training load. During ‘specialty’, intervals.icu shows my eftp declining btw.
So for me anyway it’s a clear match. Obviously, your mileage has varied.
Training load and eFTP have nothing to do with each other. Your eFTP will be as accurate as your settings and the data you give it. It’s just based on a single effort.
The eftp and the ftp I set are the same at times where eftp shifts (it’s bumping up from a static state, I’m assuming it made some calculation to do so) most likely this is a big effort I made, or an ftp test. I’m telling you that my ftp and my training load are highly correlated. Because this is the case eftp and training load are highly correlated at times when eftp is being set/shifted(*see above). It’s another independent FTP predictor, that like the AI ftp predictor, and the Xert predictor, and the outcome of ftp tests which correlate to each other and to my training load. Btw, I rolled it back to 2017, and the data is almost exactly consistent with today’s data. I checked my overall stats and in 2017 I was pyramidal, in 2016 threshold, in 2020 base. Yet ALL the data, all the ftps from the different sources and the training load correlate exactly across all the different years and training types. I don’t know what to tell you.
I don’t know what to tell you either, other than correlation is not causation.
I’m not sure why you’re so beholden to the belief that you need a X CTL to get a Y FTP. That’s great that it correlates to you and your history, but other than you I’ve not heard it elsewhere. Maybe @The_Cog has an opinion regarding this.
Love it. Your post highlights why TSS/CTL/ATL model isn’t a settled science and doesn’t describe your training at all, which is only part of the reason why it can’t be used for predicting any certain gain. So while you may have increased your TSS by 40, your FTP didn’t go up, and that’s a function of training and your physiological response to the training. Would a different training plan that only raised your TSS by 25 have given you more FTP and repeatability? It’s a very real possibility…
PPP: It’s called training stress score and not training adaptation score for a reason.
CTL being an exponentially-weighted moving average of TSS, I therefore wouldn’t necessarily expect a strong, linear relationship between CTL and FTP.
On the other hand, though, it wouldn’t surprise me if there was some relationship, at least within a given individual and over a limited range of CTL.
I don’t know what to tell you, it is not that way for me at all.
Okay so let’s be clear since you called ‘dad’ to adjudicate.
I’ll even go you one further, let’s get into tin foil hat territory. Why doesn’t trainerroad have future predictive AI? Surely, SURELY they have the capability. I’ll tell you why, because people care about FTP, rightly or wrongly. Can you imagine looking at the prediction and seeing if you do specialty your ftp will drop? You mean I have to do all these v02 repeats and die a thousand deaths and my ftp will go down? What’s the likely outcome? And again, if you are trying to get faster, and specifically trying to win races, that might be exactly what you need, repeatability. The rule of specificity! Unless your goal is to win an FTP contest then you need specificity.
I’m going to bring it on back to the hero of the hour, our 200watt wunderkind. He’s the roughly the same height as I am, he’s roughly the same weight, he’s roughly the same age, had roughly the same ctl as me, had a roughly similar ftp at that ctl and was targeting an FTP that I currently have. Per “KNAPKIN MATH” I said, I think he can do it, and I would predict that mid 80s ctl would get him there. Take a deep breath. It’s true I made a prediction based upon what I considered to be a ‘roughly (knapkin) similar’ fit to my own data. What I didn’t ask was, how’s your expression though? I didn’t ask, what about your repeatability? I didn’t wonder about his cornering, or his hydration scheme. He’s targeting an FTP, a number provided by TR via “AI” or a ramp test.
Will I be proven wrong? Who cares? Will he reach his goal? I certainly hope so, and if in doing so he invalidates my hastily drawn knapkin art, I’ll die a slow death of great ignominy. 
In reality I’ll be super stoked for him to make that kind of comeback. Farmer Strong! (that is the wild card, I’m not strong at all)
So let’s put it to bed now. You disagree with me, your data don’t work that way. Great. Good on you. My data do work that way, and I can not only correlate I can calculate my outcomes based upon my ctl.
Great summary! I just about spit out my milk on that first sentence! 
Incredibly my motto is FArmer STrong (FAST)!
Thanks for the continued encouragement!
CTL is an exponentially-weighted moving average of TSS, with a default time constant of 42 d. That makes the half-life about a month, which means that CTL mostly reflects what you have done for the past 3 months.
Although WKO (I don’t know about other programs) allows you to adjust the time constant, as with Bannister’s original impulse response model, doing so has little effect.
ETA: I’d rather be called “Dad” than “Chad”.
Don’t want to distract from this thread, but I’ve learned a lot from observing. So “thank you,” I guess.
So if you’re someone who wants to increase their FTP by X amount, is there a good way to calculate what your CTL x time needs to be in order to hit that new FTP target by X date? A good calculator, or reference framework, by chance?
Here’s one way.
There might be, but it’s a bad idea because it causes an athlete to chase TSS, as that’s what increases CTL.
You don’t want to chase TSS, because that’s not the most efficient way to train. It might mean you forego interval training in place of more z2 or tempo, push too hard on recovery days so you pad your TSS for the week a little more, not take full recovery weeks so your CTL doesn’t drop too much, tire yourself out to the point that your hard days aren’t as hard as they could be…etc.
Do your 2-3 hard TR workouts a week, fill the rest of the time with as much z2 as you have time for…so long as it doesn’t impact the quality of your interval work, and your CTL will likely rise to a sustainable level, and your FTP will keep going up too.
