I don’t think the problem is the 5 point scale needs to be 10 or even 100 points, it’s trying to represent a ride with a single number. Since it will be the only feedback you are giving the model I understand how everyone overthinks it.
Rating a 90 minute VO2 session is pretty straightforward IMO, but then when rating a 4 hr outdoor endurance ride it becomes a lot less about the specific workout. Especially this time of year when sun & heat become a major factor by 10am. I get to answering the question and asking if the ride was “hard” because of workout intensity, training fatigue at this point in the block, life stress, or because I didn’t get out of the door until 9.
I would absolutely welcome a few more options in the RPE. Just this week my two interval sessions were very straddled. Both were between medium and hard. I ended up putting the very slightly easier one at medium and the slightly more difficult as hard. Both realistically landed pretty squarely between the two. Maybe my surveys balanced out? Not sure.
I actually wish the model relied less on survey responses and more on objective data (your actual power performance and HR). Rather than a more fine-grained scale, I would support a simple binary question - “did this workout feel appropriate for its type?” - anything else is noise and only a “no” answer should raise red flags. Perhaps a follow-up question if you do answer “no” to gauge the extent or cause, but otherwise no need to think about it any more than that.
To a certain extent I already do this. Feels between Medium & Hard but my heart rate is hovering in low Z3 gets a Medium. Hard & Very Hard but that last set my HR was pinned? Better call it Very Hard. I don’t know that I’m in tune enough with my body to pull out anything more subtle than that.
That is what my instinct tells me too but I think there is just too much of the equation tied to the riders “experience” and no amount of objective data can reach into that. Things might look completely normal in the data but your legs were just heavy or the data might’ve looked out of whack but you felt strong. There are some days when my HR is very low and either TR or Garmin interpret that as being excellent aerobic fitness. Other days it seems to interpret it as fatigue. Same goes for days when I have a slightly elevated heart rate.
I do agree with you, I have worried out loud here about the problem of “perceived” exertion given how differently we can perceive things day to day. I just wonder how any coach could ever really have a full picture without an IRL report from the rider. And given that, I think it’s good to have it as granular as it needs to be to best zero in on the target.
Perhaps, but I’m thinking in terms of only the post-workout/ride subjective evaluation, similar in purpose to the current survey questions, but more free-form.
Yeah it’s a tough one, and HR definitely comes with problems, but maybe a good ML model would be able to look through daily variances and only gauge actual fitness changes through broader trends.And while your actual daily HR level can vary for different reasons, the HR rise/recover pattern and timing might hold valuable insights too?
I don’t know. I just feel there is way too much riding on a simple survey response, which makes people overthink it, and despite all the tips from TR about it, I still find it hard to know if I’m answering “correctly”. To me, making it more fine-grained just makes this worse. Putting more work on users in the hope it will improve the AI model.
This has nailed it. It would be fine if the TR scale was a 1-5, but it’s not. The steps in the scale aren’t weighted the same at all. “Easy” is meant to occupy a huge chunk of rides, then the rest are slivers; with Hard/Very Hard dividing their own sliver.