I like @mcneese.chad definitions as well; however, it all comes down to how AT is using the ratings. From what I can tell, they seem to have a significant impact. For example, I believe the issue in question was that I substituted a more difficult threshold session, completed it fine but it was tough. I rated as very hard based on how I was personally rating. This caused a slight downgrade, which seemed off since I completed the workout fine. The way we fixed it was simply changing the response.
Except we already know if you choose all out, you get a survey that indicates AT thinks the workout was too hard, and doesn’t give you PL credit. So if hypothetically I shifted everything two places: Easy → Hard, Moderate → Very Hard, etc. I would “break” AT. So there is some assumption with the way TR implemented the rating system and how it impacts AT.
This is the problem with ML: if the builders put in a bias due to the data set used for training, then the model is biased. I’m not saying TR purposely biased the model, but I’m pretty close to 100% sure that the model and training data is biased around how the ratings should be used. TR had use to use survey responses to tune the model, and these initial responses came from TR employees (hypothesis). I’m guessing there was some discussion initially, and the TR employee set isn’t a good proxy for the general set of TR users, so that dataset is skewed.
Yes, I’m stringing together a lot of assumptions to reach a conclusion, but I really think this is one place where TR could help people, and itself by providing more direction around the intent of the survey responses, and how to use it.
For me in relation to 1-10 (easier scale for me personally in terms of thinking about the effort):
Very Hard: 8-9
I know that is all subjective but this is my translation. And for example 4x20@92% usually falls into moderate/Hard, depending on a day, 4x15@97% is usually Hard. VO2 max is very hard and all-out is for testing. I treat all-out as “I have fallen off the bike and lying in the corner crying and cursing myself” - good example would be my “favourite” 5 min test. I rarely do all out workout - if I have not finish the workout there are usually some other factors.
I asked customer service this very question, and it is not based the perceived effort of the ride (i.e. RPE), so you can’t just mark rides easy if it felt easy and hard if they were hard. The survey is meant to be answered based on what you expected the ride to be (so endurance rides can be hard if exhausted) and threshold rides can be easy (if well rested, have increased fitness / FTP, etc.)
This thread, to me at least, highlights the need for TR support team to have a staff only thread or FAQ help page.
There’s a few repeated questions in the beta thread, missed information & here we have a few misunderstandings on personal feedback. The bug tracker page doesn’t really cover things like best practices or how results work mainly cos that’s not a bug just training.
This is a good thread. I was thinking of posting the same question myself. I have been answering the AT question with an interpretation exactly like that from @mcneese.chad , so it’s interesting to hear that AT might be working with a different definition. I’m sure it’s important for everyone to be using a consistent definition.
Yes I’m confused. I think I’ve been completing it in a different way to how TR is advising.
e.g. VO2 max felt hard to me (though I completed successfully) so I marked it as Hard.
Based on what I’m reading above, if I expect it to be hard, and it was hard, I should mark it as Moderate?
The rating doesn’t matter globally. A 2 for me will not be the same for anyone else and with machine learning it does not need to be. Machine learning is just looking at this like it’s an up or a down and as long as you are consistent it will score it relative to YOUR other scores, not anyone else’s. I think you should find a reasonable classification, such as Chad’s above to make it easier for you to be consistent with your choice. I like using Hard for most productive workouts because that gives me somewhere to go up if the workout is harder than normal, but then you still have one even higher when it’s way above your pay grade. I think having 2 notches below normal is also a smart choice for those workouts that you felt completely unchallenged or were pure recovery rides. A 5 star rating such as this seems like the easiest simple way for folks to gauge anything.
It should be this. An official communique saying “it doesn’t really matter as long as you are consistent with yourself”. Pretty sure that podcasts had that in one of them anyway.
What is the point of “how I expected to be”? If i expected to be hard (let’s say 4x15@97% FTP) and it felt moderate so easier than I expected how I should score it? If I score it moderate then it is based on RPE if I score it hard I am giving contradicting score.
I basically do not care as I do not use plans/plan builder etc but I am just curious.
I am attempting to pool some info here for reference, along with my thoughts on it that I hope TR can review and comment.
I have more comments to add and I would like to see what else we can learn. As of now, this is more confusing than helpful to me.
Yes, is it all based on some measure of perceived effort — it is just a question of what it is relative to. My understanding is that RPE is relative to just your perceived level of exertion (an example would be to rate how fast the Ford Mustang felt to drive), but TR is asking us to rate it based on the perceived level of exertion compared to what we are used to for that type of workout (how fast did the Mustang drive compared to a Toyota Camry). For your exact question, I would just rate it as something easier than hard depending on how much easier it felt.
FYI, here is the communication I had with customer service.
My question: Are the survey responses supposed to be similar to RPE (i.e. for example you should always expect an endurance workout to be less than threshold), or is it relative to how hard you expect an endurance or threshold work out to be?
Customer Service Answer: As for the survey responses, they should be answered relative to how you expect a ride in that zone to be, and not related to the RPE scale.
And that is why I have a problem with the TR comments. The one prior says 2-Moderate = as expected. But what if you are doing something like a challenging workout like Leconte (Painful Over-Unders)?
I have never been anywhere close to “Moderate” on that workout. I have been a 4 at best when everything was on point and I have had my fair share of what I would give a 5 too often with that beast. I guess we should “expect” it to be miserable and then rate it a 2 if I understand that logic?
Again, crazy to me if that is the intended use, with the fact that there are 3 ratings harder than that 2-Mod. Based on all I learned over the years, and in light of the new Workout Levels, I know that workout in particular is in Breakthrough/Not Recommend territory for me more often than not. So maybe that just points to that example being something that should not have been on my schedule.
All that is to say that I still don’t think I like or align with what we are learning from TR with respect to the “expected” use of surveys and the likely impact as a result. That is especially so since the initial and repeated comments since the start of the beta were “rate it how you feel” and then telling us the system would “learn from us”.
Maybe that is true, and my system not aligning initially with what we seem to be learning from TR are somehow meshing and becoming a proper rating system. But I have serious doubts based on their more recent comments, and looking back at the PL changes as a result of my ratings.
This is pretty frustrating to be told to just “go with it” in the beginning, only to learn that there may well be more specific definition they have on hand and use, without sharing until we pull harder to get the info.
Maybe they are still figuring out how to make AT work best for more users… and just go with it was all they had at the time, and later realized it led to inferior adaptations
Yeah, the two different comments seem to contradict. To me, the easiest way to have set this up would have been to simply use the traditional RPE scale, but based on the feedback I got that does not seem to be the case. We just need some clarification at this point, because we are getting different apparent answers from TR.
Yeah, sure could be the case. If so, that is great they are learning and working to improve.
But we need actual guidance from them if those prior instructions are not correct any longer. And that should be shared wide and far as opposed to the few people getting it via direct contact with TR Support. We only know what we know now because beta users are sharing (and I am thankful for that).
We need a clear and direct set of instructions (be the more clarity on the “old” system or whatever “new” system may be in place) via TR in these threads, the FAQ and such.
When I was trying to figure out which scenarios supported progression level updating, someone from TR was going to write up an FAQ. Did that happen?
I think I saw it posted on another thread that a 4 will slow and possibly even slightly reduce your progression rate.
Coming off of a rest week this morning I jumped backed into SSB-HV II with Antelope. Prior to riding AT adapted my plan to my SweetSpot PL (8.3). During the workout I needed a little extra rest between the last two sets so I rated it a 4. AT adapted the plan after the ride and made the subsequent workouts a little easier (PLs 6-7).