I’ve been following this thread fairly religiously and am really enjoying AT but I’m unclear on this as well. Specifically, for the post-ride survey, I’m trying to figure out how to respond on the spectrum of, let’s call it “objective difficulty vs. difficulty relative to expectations,” and how that influences the choices AT makes. This is most clear at the extreme end of the spectrum but actually gets a little more confusing somewhere in the middle.
Two examples:
Redondo +2: the only way this workout isn’t going to be brutal is if you’ve significantly underestimated your FTP. So assuming you complete the workout and are wrecked at the end how do you respond?
- Moderate - does this tell the machine that it’s what you expected, as in “moderate vs. expectations, not harder than what I expected” (so AT makes no changes; everything is on track) or would it mean “I expected a more difficult workout than it gave me, so it was only moderately difficult compared to what I was expecting?” (so upgrading the difficulty on some future workouts).
- Hard - this seems the most likely answer because it works on both fronts: you expected a hard workout and got one, and it also just kicked your ass. And I’m guessing this wouldn’t cause AT to make any adaptations, but just note that you worked hard and things seem on track.
- Very Hard - the flip side of Moderate - does this mean “I expected it to be hard but it was way too freaking hard” and cause AT to downgrade future works based on difficulty finishing the workout, or does it mean “I expected an ass-kicking and I got an ass-kicking” so everything is on track?
Lazy Mountain: this is a recovery ride, generally speaking, so do you mark it -
- Easy - You expected an easy workout and got one, so everything is on track and AT doesn’t make any changes. Or does it mean you expected something easy, but this way too easy, so future rest day recommendations will be scaled up a bit?
- Moderate - The opposite of Easy. “I expected a very easy workout, but this was moderately difficult so downgrade it next time.” or “I was expecting Easy, so this Moderate relative to my expectations. Everything is on track.”
So having written all that - can we get a little more clarity about what the survey responses actually mean and how they affect the choices AT makes? It might also be worthwhile to re-think the actual response options to make the impact of the choices a little more clear to the user. It could give AT users more of a feeling of agency and control, as opposed to plugging answers into a black box (which is pretty much what using Machine Learning is at this point, of course).