No, they are not. It’s been mentioned above, and they know that it is not “right”.
1 Like
Some one (not a Bellend 
) said in one of the other threads that as you probably expect TR is not calculating levels for custom workouts correctly 
1 Like
IS THIS A TRAP???
I think the team would be hesitant to provide a definitive timeline, as these features in beta can be really quick to roll out, or take a long time to be available to all depending upon those bugs and issues we identify and resolve. Some are quick fixes, and others more complex, so you can see how its really difficult to estimate what that process will look like. The last thing we want to do is provide a timeline that has to keep changing, or lets or athletes down when we don’t end up meeting a loosely estimated timeline.
9 Likes
is there a way to see the “sub” zone types on workouts? Now that I find myself manually adding workouts to get the intensity I feel I need, it would be helpful. For example, I know that Spencer +2 is Vo2 Max, but is it sub labeled as attacks, continuous, float sets, max efforts, mixed intervals, on-offs ramps, steps, sustained intervals, sustained power, or traditional? Searching for the right new workout would be a lot easier if I knew how other workouts that I am more familiar with were categorized.
1 Like
Again, you and others are assuming that TR somehow made up some changes to the plans based on some super small internal trials. This is a laughable assumption for anyone that has listened to the podcasts and their deep dive into minute details.
The TR team bases their training plans on lots of scientific studies plus experience of what it takes to compete at a very high level. Why would they suddenly throw that approach away?
Go read more about machine learning. To train the AI you have to start by giving it massive, massive amounts of well prepared PAST data. The PAST data gets used to train the ML model BEFORE you try to use it to predict anything.
TR have probably the world’s biggest dataset for people’s goals, planned workouts, actual compliance and performance improvement. It’s all the PAST data from you and everyone else’s training. They have something like 10k customers and probably 100k completed workouts in their data.
So the obvious explanation is that they have already trained the model on this past data and arrived at the improved plans from that data study, by comparing people that trained similar to the new plans to people following the old plans (many people train using custom plans or workouts). They can A vs B test so, so many things using this technique.
Most studies only have N= 5-30 participants and people jump to use the results for their training. TR has several orders of magnitude more participants. Give them so benefit of the doubt. Yes, ideally they would be open book and tell us all the nitty gritty details. But they won’t for competitive reasons. I bet they will slightly tip their cards, but that will probably come directly from Nate, in the podcast.
1 Like
There might be an easier way, but if i don’t know the sub category i search for a particular workout and check the sub categories one by one for the overall category and the search result will disappear from the results until i check the right sub category - Traditional in this case for Spencer +2
5 Likes
that’ll do in a pinch. Thanks!
1 Like
Dude! Great hack!!! Thanks for sharing 
2 Likes
I think that a lot of people underestimate how difficult it is to produce a useful and accurate ML/AI model. Even with massive amounts of data, there will always be subgroups of the data on which the model will underperform or make poor decisions.
This is not to say that TR has no idea what they’re doing—with the dataset they have and the talent they’ve employed, I think there is no company better equipped to tackle this challenge than TR.
However, it feels like every other response in this thread is something along the lines of “…but they used machine learning!” Okay, great, I know they did… but can they (generally speaking) show me how the model performed!?
For example: based on the ML models that TR has developed, what percentage of users are expected to benefit from the updated plans versus the older plans… is that number over 95% or closer to, say, 75%? Are there subgroups of the TR user-base that might actually be better served by sticking to the older, higher-load plans… maybe intermediate/expert riders in their 20s/30s? It is doubtful that every single TR user will benefit more from the updated plans than from their previous variants.
Yes—TR has updated their plans, presumably because they will provide increased training adaptation for the majority of their users.
But, here’s the thing—I don’t care if the plans will improve the performance of a majority of users… I care that the plans provide a performance improvement for ME. And to be confident that the new plans will benefit me, I think it’s important to see some statistics and data, rather than just parroting the same argument over and over again: “you should trust them because they used machine learning.”
8 Likes
If they released statistics, anyone who didn’t like the new plans would put themselves in the 15%, or whatever, of athletes who would do better on the old plans. Right now, most of us haven’t tried anything like the new plans, and we’ve seen improvement on the old plans. It makes sense to stick with what we know. But I don’t think that TR putting a bunch of stats out there is going to change many minds.
5 Likes
Well from my knowledge of AI you actually feed it a subset of data then present it with information about a known outcome then from there see what it predicts about a known result. Then you scale it up to start doing future prediction.
Trust me if they wanted to see this look at the completion rate of something like Elephant +1 from the old Oly Tri Plan where I found that if I did the Rx Tempo run the day before I was not going to be able to do it or was going to have to fight every minute.
I have listened to the podcast quite intently and don’t disagree with you that they do great thorough deep dives.
I have looked back at what has caused certain situations with my training. I am brutally blunt with myself and can usually see that I have pushed myself too hard during runs on days where I maybe shouldn’t have.
I have faith that I will still see improvements from following the new plans the problem I have is that the solution to plans being potentially suboptimal for many users is a feature that < 1% have access to.
I would hope that the logic could be explained. For example Salado replaced Givens which is a drastic change for the late phase of Build
I agree that ML is very difficult and success is not guaranteed. Also, any training plan is unlikely to be optimal for everyone. However, that has always been true.
Also, I keep seeing people claim that these plans somehow are untested or based on a super small TR internal test group. That’s obviously false. They can do comparisons to see what has worked best for the most people.
1 Like
I think people are more open-minded and willing to change than you give them credit for!
If TR said that they expect 97% of users to perform 2-3% better on the updated plans than the old ones, most people would probably make the swap with no further questions asked (myself included).
But if TR said that they expect 51% of users to see a 2-3% improvement on the new plans? Well, then I’d start to wonder if I might be in the 49% who would be better served sticking with the old plans.
From TR’s perspective, both 51% and 97% are enough of a majority to implement a change “for the greater good”. But for me as an individual athlete, I want to be more than 51% confident that making the switch will benefit me.
3 Likes
Makes sense. I was thinking of a different set of stats.
I’m really happy with the new plans; I’m confident that I’m part of the curve that will benefit from the new plans as they are. I understand some of the concern in this thread, though. I think it will all be worked out when AT comes out of beta. But it’s not out of beta yet and we don’t know when that will come.
3 Likes
100%, couldn’t agree more! Thanks for clarifying your argument!
You’re right—no training plan is perfect for everyone. Adaptive Training will resolve that by individualizing each user’s own progression. But, unfortunately, AT isn’t here yet. 
So, until it is here, I have to make a decision between the old plans and the new plans. That decision isn’t immediately obvious to me (without more data/information from TR!).
But yes, anyone who assumes that the new plans were somehow conjured up based on a sample size of n=4 is also way off the mark. 
1 Like
I can’t speak for him but I subscribe for myriad reasons, and the plans aren’t one of them and haven’t been for years
3 Likes
I think the concept of A and B races you are looking at is one of motivation and intent. The two races, 4 weeks apart from each other, were just as important to you and you wanted to excel at both. And this is consistent and achievable.
The TR A/B concept looks more at physiology and the ability to cope with high load/specificity for your event. Unfortunately this is not really possible to train for. You need to recover from an event like an A half IM and you also need to taper for your A full IM, leaving only 2ish weeks to specifically train for the IM. You’re not likely to see any improvement from those 2 weeks of training.
If you are looking at having 2 (similar style) A events 4 weeks apart, I’d probably aim to peak for the first one and simply try to hold onto that same level of form until the second event (that’s just like my opinion, not backed up by any science though). You should perform really well in both
I totally get where you are coming from though, when I was younger my fitness peaks/valleys over the course of a season felt much more distinct, maybe 40-50w FTP difference. These days I might get an extra 10w if I really focus on an event/period, likewise if I take 6-7 weeks off the bike I’ll probably only lose a handful of W from my FTP. It does make it harder for me to stay motivated for those A events because the effort is high for a relatively low gain and when I do get there I want to hold onto it for as long as possible!
4 Likes
It depends on your individualized progression levels. They’ll be different for every body, and you don’t have access to them unless you’re in the closed beta or you reverse engineer them by looking at your recent workout levels and calculating where you think your levels fall for each system.
Based on some quick searches of the workout catalogue and comparing them to my levels, I surmise the following for how they’re determined:
- Recovery - Only 1.0 Endurance workouts exist for this difficulty.
- Achievable - Equal to or lower than your current progression level. It may also account for some workouts slightly over the current level, but likely no greater than 0.3 over your current level.
- Productive - Any workout equal to or greater than your current level by no more than 1.0.
- Stretch - Any workout between 1.0 and 2.5 greater than your current level.
- Breakthrough - Any workout 2.5 or greater than your current level.
Note: When I use the term “current level”, I mean the individualized level for each zone; those being Endurance, Tempo, Sweet Spot, Threshold, VO2 Max, Anaerobic and Sprint. Also, when trying to determine the above, I only referenced my Threshold and Anaerobic progression levels. It is quite likely that each zone has different parameters for determining workout difficulty classes. That said, I can say from my experience when a workout is 1.0 above your current PL, you’ll know, and if it’s 2.0 or greater, you’ll definitely know. That experience has been universal for my PLs so far in the beta.
3 Likes
Well… This is a whole different problem. Because runs nor swims are accounted for. So they have no way to know that you failed because you did a hard ass run workout the day before…
1 Like
Thanks for this!! It’s really going to help me piece together a good progression within the new plans. I just have to figure out how much I have to roll back the workout levels assuming I have ftp increases.