The addition of TR’s Ai training and subsequent FTP “calibration” has seemed to draw a lot of conversation from the community, positive and negative. FTP has been the metric we use to quantify our performance for a long time, but I believe we are entering a new era where this single dimension metric lacks enough information to satisfy our want for a performance comparison tool. If TR’s Ai does not use FTP to determine your training, why should we use it to determine our performance?
I do think FTP is a good metric, but it lacks specificity and range. Being a single number, it doesn’t show where your best performance is relative to FTP. For example, a time-trial specialist may find FTP to be a near perfect representation of their ability, but a sprinter or ultra cyclist could have a skewed perspective.
My idea for a tool that aligns with how TR Ai views performance (in my interpretation) is a theoretical power map. For a given duration, the power map would show how many watts you could sustain if you were to do an effort a given number of times. The visual of this would help us see where our strengths and weaknesses are, what area of performance our training is having the greatest impact on, and be the basis of some very in depth bench racing. This kind of tool would provide far more depth, perhaps to its detriment.
What are your thoughts on benchmark performance metrics? And how can they benefit your training?
I bet someone said this when FTP and PDC were first being introduced. “We have heart rate and RPE, to hell with power and lactate!” Hahaha
PDC is an excellent tool, but it is practical and does not take into account repeatability. “Durability” has become a buzz word and I think it boils down repeats. I think a 3d visual of theoretical power that included repeatability would be a very motivating to see. It would help curb the disappointment of a stagnant FTP if you could see improvements in other areas.
I worked with claude code to build it one in the past few minutes. I’ll put it up online so you can drop in a GPX file. I added options for fixed-length rest intervals, in addition to rest equal to work. The 1-2min rest produces the smoothest curves for me, 5 minutes makes some sense too.
My answer to your overall question is that if we learned to analyze these curves directly, we would understand that the AI’s FTP is generated by fitting a theoretical smooth shape on top of your personal curve here. As your progression levels increase, you’re flattening out the repeatability axis, and the theoretical curve shape (fit from all rider’s data) would show that your shape should actually be higher at the repeat=1 60min. Going from your actual surface to what the surface should look like if it’s fully explored (by testing).
I feel this too re the “FTP” value on tr. being a much lower number than the former AIftp. I struggle to contextualise what my performance is truly like in the really world as it’s often based on more nebulous things like repeatability and time to fatigue. seeing the number decrease or stagnate despite potential improvements in other areas is a huge hit to my confidence.