It uses both the raw data as well as the survey to provide a more personalized progression. A good coach would also be asking about how you feel to make sure the workouts are correctly tailored to you, and AT is aiming to do just that. Below is the portion of the original podcast about how workouts are classified as super passes, passes, struggle passes, cut short and failure as well as the workout surveys and why they exist.
There’s a huge value to it! Imagine you finish a workout that should be relatively easy based on your FTP, but feels like an all-out effort. If your coach asks you how you felt, and you respond that the workout was extremely hard, they’re not going to give you another super hard workout the next day, and neither is Adaptive Training. Likewise, if workouts that should be super challenging consistently feel easier, your coach will identify that as a sign of progress and likely up the intensity of your future workouts a bit, and so will AT. It’s a natural way of tailoring your progression to what you’re ready for from day to day.
I can see this, but I don’t have any more faith in my ability to do this consistently than I do other “five star” systems.
Thing is that you can’t be wrong when answering it if you’re honest. All it cares about is recognizing patterns and adapting your training from them. Even if you are trying to break it, ML is quite good at identifying false patterns and disregarding them from what it’s actually trying to do. I would suspect the TR ML has ways for addressing this as well.
In short, I think you’ll be surprised how easy it is to answer each ride. If it’s difficult, you’re thinking about it too hard imo.
I refer you to this post:
I guess I dont see the value in using perceived output when you have the actual numbers.
To pick an analogy, do we trust the subjective feelings of consuming say Beet Juice(or whatever) or do we rely on seeing a scientific study to quantify what it does.
If we have the numbers and you continue to do workouts as per the TR set schedule…then the rate of increasing the difficulty of the workouts should increase. ie find out when I fail and tweak again.
If I think everything is easy and always say easy…AT will have to learn my easy could be someone elses it was fine, neither hard nor easy. It is this subjectivity that I think just creates noise in the data.
So, do you believe there is no value in a coach asking how a workout felt? If that were true there would be absolutely no value in a coach. Everyone could just follow the same random plan.
We don’t have sensors for perceived effort (not output), and power numbers don’t provide enough context for a model that’s supposed to give you “the right workout every time”.
Unfortunately, most of us don’t train in a lab, so we can’t measure the biological/neurological markers that determine how we feel during a workout.
There is no “set schedule” on AT. There’s a loose framework that AT fills in with sessions based on your performance and perceived effort.
AT tries to prevent you from failing workouts.
Someone else’s responses are not relevant to your progression. AT isn’t comparing your perceived effort to other perceived efforts and deciding on an objectively correct effort measurement. If you think everything is easy and always say easy, AT will give you progressively harder workouts.
Subjectivity is not noise. How you feel during a workout is relevant to your progression and your training goals.
Overreach is one thing (required) but taking that step further can lead to overtraining. So say you did Disaster successfully 3 days running you’re likely going to feel pretty bad, I’d imagine this system would actually never recommend that session in any circumstances.
It’s a way of preventing you from pushing yourself that step too far.
Edit - confirm that Disaster and Rockhouse don’t have AT ratings 
Thanks @SeanHurley , but what are the actual responses available then? That is what I wanted as I am NOT on AT and so cannot see them. And I want to back fill for my own records, consistent with what is coming.
(Bear in mind there is no manual or, as far as I know. blog post, that describes how it works and all the features, including the responses, apart from the long video and the enormous stream of comments and responses in this and the other forum posts, neither of which which does not constitute a practical, accessible, manual or guide 
)
They’ve already said user documentation is in progress.
That helps, thanks @ellotheth
I am sure they have said about documentation - somewhere in the various videos and 3000 forum posts! 
Here is my post on all survey responses, pulled from the video.
We aren’t talking about coaches. This is about data
But you have the data. You don’t need the subjectivity if you did the ride successfully. Atshould accumulate prior users abilities to do this and adjust. Why does it need subjectivity
You can follow a plan successfully but be in a hole by the end of it
I view it differently. If the ride is a success then data of success or failure should be sufficient. I don’t need my perception that work stress that day impacted how I felt riding.
But so far they indicated they are not factoring in telling you when to rest.
I think some of us are just particularly bad at RPE. I don’t think I’ve ever done more than a handful of “easy” rides because frankly, sitting on the trainer at any power level feels unpleasant to me. I did Petit yesterday and that felt almost hard because of the form sprints even though the power levels were trivial to hit.
I am envious of anyone able to quantify their discomfort levels to any meaningful degree of accuracy. The only one I really understand is “all out”, but that’s only by virtue of being on the cusp of either wanting to die, or having to work out ways to survive the intervals (or indeed, not getting through them properly).
There are situations where the data look perfectly fine but one would feel awful, so survey data just adds an additional measure that sometimes HR/pwr alone can’t measure.
Additionally, there is value in the likert scale stuff in health sciences. I don’t have a citation, just something I remember, but individuals reporting fair to poor on their health perception were more likely to have adverse outcomes compared to people self reporting better health. So there is value in the self reported stuff