Dylan Johnson's "The Problem with TrainerRoad Training Plans": it's gonna be a busy day around here

I’m going from memory here, but I remember TR staff mentioning the following metrics:

  • Progression Levels for different power zones
  • Scoring of unstructured rides (this is an internal alpha)
  • Failure rates of workouts (e. g. with and without AT)
  • Pauses during workouts (to my knowledge, this is not publicly accessible)
  • An internal fitness metric that subsumes FTP, all Progression Levels, etc. into a single number.
  • Benchmarking of modified versions of AT against old plans and older version of AT
  • Ramp rates

@Nate_Pearson was a bit vague on what makes AT tick, but he mentioned that they have tested plenty of models, including some with more and others with less parameters. For obvious reasons he didn’t want to go into a lot of details, but if you know anything about Machine Learning and the fact that TR has been working on this for at least 4 years, then it is clear that they have consider many more factors than that.

Again, I’m not saying it is perfect or that TR is better than coaches, I’m just saying that they have probably thought of any metric we can come up with and more.

Yes, what does lactate testing have to do with anything? You now mention FTP tests. All FTP tests are based on prior research, and that includes the ramp test. What TR (and all other platforms I am aware of) do that is problematic is (1) rely on research whose test subjects were likely fitter than the average TR member and (2) they use a fixed percentage to compute FTP from tested power (e. g. 95 % of FTP20 power or 75 % of MAP). With the data set TR has, they can use other techniques to infer FTP. Large-scale lactate testing could conceivable only address point (1), but not point (2).

So yeah, I think lactate testing has nothing to do with it. My recommendation is that you simply don’t hand off your brains when you train, no matter if it is a coach, a piece of training software or a book: validate things yourself. Is the new FTP you got (using whatever approach there is) accurate? Well, try a threshold workout with longer stretches at 100 %, a VO2max workout and, if you can, an outdoor ride where you ride stretches at these power levels. You’ll learn what riding at FTP feels like. Use the knowledge and adjust your FTP if necessary.

Stupid question since this is an opt-in beta feature: have you heard of AI FTP detection? TR has done exactly what you think they did not do: they validated their algorithms against the data that they had.

Now I don’t claim AI FTP detection is perfect, I actually prefer testing (because I like it, really), but clearly they have been working on this for quite a while and at least have attempted exactly what you want them to do.

If you think that, I recommend you read some of my other posts, e. g. when it comes to the state of TR’s apps, lack of Health integration, lack of Apple Watch integration and lack of metrics. I just happen to think that the vision they have articulated for TR is a good one.

How do you know that averaging 60 % FTP on your endurance ride will have the same or better outcomes? Is that just your opinion?

I don’t think such a blanket statement is correct in either direction is correct. If possible, longer rides at lower intensity are equal or better than shorter, slightly higher intensity rides (all within Z2). But that is usually a function of how much time you can (or allow the algorithm to) allot to your training on a given day.

(Personally, I prefer to do longer, mellower endurance rides outdoors. But I usually can’t spend more than 3 hours on a ride, I’m a family man.)

6 Likes