Coach Jack plans, their description and Zwift

Training and Performance Fundamentals
1

This is not about syncing to Zwift, but more about using CJ plan/structured training in parallel with doing free rides on Zwift or any other platform or even outside - for that matter.

Also - I am not training for any competition, just to stay fit. I am getting closer to my 6x decade on this planet, picked up biking several years ago and go to about 40-50 percentile in capabilities for may age group… Maybe I can get a bit more - and that is why I need some training structure.

For quite sometime already I am, looking for a tool, which would let me do structured training, but then allow me to do my own riding 1-2 days a week and keep workout plan adjusted to accommodate the load from free rides - if more/less needed of particular workout types… So far I couldn’t find anything. I haven’t tried TrainerRoad (and don’t really want to), Xert can do this in theory, but it is not that good in practice (but their new “magic buckets” might be a step in right direction).

I would love TrainerDay to improve CJ to work in load incurred in free rides, but there is another way - maybe some workouts can be just replaced with specific Zwift rides - for example, I can do long endurance ride with an appropriate robopacer or do VO2max while climbing AdZ.

So I was very interested when I saw this while checking CJ recently:
image

My excitement was short lived though :slight_smile:

  • The frame to read CJ plan descriptions is too small - hard to read while scrolling - there is a lot of useful information
  • Would be really great if all these descriptions can be published as a web page/pdf/epub - I had to go extra mile, copy them all together and create the doc for myself…
  • This specific paragraph is out of place - I believe it belongs to the The Serious Italian section - and if all of these would have been published together, such error can be quickly recognized - @Alex you may want to correct this…
  • And the last one - there is no guidance on how to augment The Serious Italian plan with out of plan riding - it would be great if some guidance can be published on this.

I guess, my question is - are there any guidelines available for the last item above?

Would it be a good idea to map different workout types to different type of free rides in a table, then ride and keep a tab on the load/rest/recovery via intervals.icu (which I love as well and wish it would get some training recommendation engine :wink: ).

Any advice would be appreciated!

1 Like
2

Thanks, we recently did some translation and changed things here, it’s possible that caused issues… But we will look at this.

To answer your question. All of our blocks are on the premise of 3 key workouts, in the beginning these are not too hard, if you start at the lower intestines so you have more flexibility in what you add. In these 3 key workouts their is a main focus. The SFR intervals, threshold intervals and long ride. So basically the is a lot of flexibility, meaning you can combine any secondary work you want to this. Overall I would try to keep that mostly as lower stress activities. Z2/Z3… but a little intensity is not so bad either… Focus on keeping your volume progressions in check meaning don’t start the season at 15 hours a week and end it at 5 hours a week… Get some rest weeks.

As you get the harder phases of the block then be more careful not to do stuff that affects the quality of your recovery or the quality of the workouts. That’s personal.

So there is no black and white answers as to what is good and what is bad… Usually riding more is good as long as you can recover from it (good sleep) and not doing way too much for too long.

Not sure if that helped… I kind of would say just use common sense following common training advice. Do as much as you can adequately recover from :slight_smile:

1 Like
3

Thank you for your suggestions, Alex! very useful ideas…
I am at ~7-9 hours per week for a while and manage load and rest via intervals.icu, so I was planning to keep that volume about level and follow CJ suggestion - I see it can bump up 8hrs/week to 12hrs/week for 12-16 weeks plans…
As for doing Z1/Z2 in Zwift - it may be boring, I rode all flats already, most routes left are climbing - and doing Alpe Du Zwift in Z2 is not fun :grinning:
But I guess, you are right - without any specialized tool to balance weekly load based on history, common sense and intervals.icu would be the only sensible way to have fun and exercise…

4

The thing you have to remember is no numbers or tracking 100% align with the stress that we get at any point in time from any particular activity, it’s very broad brush. It just means if Intervals TSB (or whatever it is called there)… is severely negative pay close attention.

So many variables like sleep, outside stress and diet can severely affect how you recover. HRV can pick up on some of this but it’s not perfect in most cases either. But yes sounds like you have the right perspective :slight_smile:

What I do, is if I am increasing training stress a lot, I make 100% sure I am sleeping as much as possible. Last week I slept 9 hours a night 5 days a row according to Garmin (Garmin’s a bit of a liar but sounds good anyway…). I still felt very rested and it helped me recover faster from some of my latest increased training stress. For me my legs sure feel better with an hour in the hot tub but there does not seem to be a lot of science to validate this…

1 Like
5

Yes, these are excellent points! Since I use Polar pacer Pro to monitor sleep and use H10 HRM for workouts, Polar is able to collect good amount of information and indicate sleep recovery pretty well and even make available some HRV-derived stuff to intervals.icu (but that is from workouts only, as I understand).
So yes - good sleep, no alcohol help with recovery and allow easier ramp up of the load…
I like some ideas join.cc app implemented - they seem to rely on past load and RPE to recommend rest and next workout - very interesting!
But it mobile-only and does not have much of analytics/details (intervals.icu can do that), so I decided not to continue after a brief trial…

1 Like
6

HRV in workouts doesn’t work. You have the polar H 10 the app called Elite HRV is free. I would experiment with it and ideally take a one minute morning reading each day, especially if you are increasing your volume or intensity a lot.

The polar nighttime reading might be fine as well. The Garmin nighttime HRV seems mostly reasonable. I can also see it fairly clearly from my lowest nighttime heart rate which seems to correlate very well with HRV. There can be false positives, but it’s still very interesting learning and as you triangulate your training, stress balance, TSB and your HRV as well as how you feel you can start to learn when you’re going over the edge or not getting enough Qualityrecovery.

1 Like
7

Thank you for the advice, I’ll check out Elite HRV.
I tried one of the apps a while ago and it didn’t work properly, maybe this one will :slight_smile:

Since Polar records sleep HR using optical sensor, it does not have HRV data, but I guess, they have pretty good algorithms to judge the recovery based on HR + accelerometer - whatever they show matches my feelings pretty well (and I use there watches for several years now).
But yes - correlating this with HRV was something I wanted to do.

1 Like
8

Yeah, the idea of collecting RPE like join is doing and collecting user feedback and adjusting the plan sound like a good idea to me, the problem is I’m not so sure it really works in most cases. I have tried building stuff like this or testing stuff like this and I just realize that when I’m giving feedback each day at least personally it’s not nuanced enough to make good decisions off of. One of the main benefits would be if somebody starts a plan and just hates it you could ask them you know you could make a major adjustment to intensity or you know collect some feedback but nuance day-to-day stuff especially in the coach Jack type of model just doesn’t make sense. I do know that if I was talking to a coach and describing something on a daily basis, I could give them good information but at the end of a ride and a question I’m just gonna click quickly. I’m not gonna go into a lot of details and so it’s hard to make plan adjustments to that.

I’m talking out loud because I would love to find this great idea of a way to actually build this and I’ve been trying for five years to do this, but I think others do it more as a good marketing concept rather than a great dynamic plan execution.

2 Likes
9

Interesting view on Join. I had my 9-day trial run up just now, I didn’t really use it, but selected a plan and was adding other workouts I was doing (currently finishing up a month testing Xert).
It seems that Join looks at the TSS/training load, not just RPE. From what I read, that RPE question doesn’t mean much - there is an algorithm, which monitors any training you do an modifies your plan on the fly. Same thing as TrainerRoad and Xert are doing to some extent, but without providing any detailed analysis to the user.

Interestingly, another issue I find is that most of the training plans start with relatively easy weeks, targeting people, who didn’t train much.
Your CJ can adjust the starting level, although I am not sure how good it works, but I plan test things out starting early next month :slight_smile:

1 Like
10

Yeah I am not speaking of join specific problems, I am just talking about the problem of user feedback and adjusting the plan. I fully agree for example monitoring TSS but not automatically adjusting but asking the user questions when AI notices something. Hey it looks like you did a lot this last week, you might want to take it easy for a few days “Ok?” Meaning the relationship of negative TSB (too much TSS too quickly) is very fluid and has a lot of “in depends” so you need to triangulate a bunch of things before making changes to the plan.

The importance of dynamics also is highly dependent on the nature of how you design plans. For example if you are xert or TR and you are really pushing people to the limit, well if you are killing them and they tell you, “Hey you are killing me.” Then sure AI should reduce the intensity or the volume. We don’t believe in this approach to training. So like in Coach Jack if you gradually increase from easy to hard, as long as people don’t take long periods off, it will be fine in 95% of the cases. The main problem is many people just feel Coach Jack is just too easy in the beginning. Or they crank it up until it is hard and then they don’t adapt fast enough and need to back off…

It’s just like going to the gym. If you previously were strong, and you come back after a break if you don’t take it easy then you feel the pain… So you learn to take it easy at first. People don’t see it so clearly with bike / endurance training but it works the same way… It should feel easy at first and slowly progress. It’s safer and produces better results long term. This is obviously our opinion and you as the cyclist need to chose what you want to believe and following your own beliefs. There are 1,000 different opinions out there and merit in many of them. Most people have to learn to curb their enthusiasm for SUSTAINED seasonal intensity the hard way (via suffering and failure)…

11

Yes! Such approach would be really good! It should be a guidance - here is the plan, here are suggested workouts, you can do your own rides and our system will give you corrective advice…

Another issue is picking up or changing plan (or app) in the middle of training. Say, you are doing your own training, somewhere in the middle of the next 6 week phase and want to switch to TrainerDay - if you just pick up a regular plan, it will be very easy, so the challenge is how to equalize the load to avoid steep change? Again, Xert tries to do so by analyzing your training history on Strava, but that can easily lead to immediate overtraining if math picks up wrong ramp rate… And let me runt about Xert a bit more…

Looks like Xert is too focused on the math, not physiology of training.
You can not apply the same exact formula to every person in the world. In order to advise on training load volume or rest, physiological measures have to be taken into account. I think, system like this should look at a combination of:

  • Obviously - the training plan
  • Age and training background - both long term and recent
  • Training load and rest over moving ~7-14 day window
  • Sleep, resting HR, HRV…
  • HR behavior over last several training sessions and how it changes compared to previous similar loads
  • Feedback - RPE and fatigue

Polar and Garmin are trying to get this covered to some degree, and I think Polar can manage running training program based on this data, but unfortunately, they are not interested in cycling.

Interesting that Xert/TR are so focused on suffering during workouts… Not sure about TR, but going through the weeks of Xert, I see that it is trying to over-train me. Why prescribe intense VO2 max workouts two or more days in a row? Why call “Pure Endurance” workout which is essentially Z3/Tempo? I setup Xert to just maintain my fitness, I know that I can do it with certain volume of rides since I been doing this for a while already, but Xert really wants to constantly increase load and hours - much more than I used to do for the last year… Their AI overestimates FTP, making their intense workouts even harder. And that’s not just my observation - I see similar complains in their forum all the time and the answer often is that user doesn’t understand how to use Xert😒

Their training workouts are often based on multiple repeats of micro intervals 180-40% or 120-80% for 30/30 and shorter variations - I guess, this works for some aspects of training, but should this be the primary workout type? Besides VO2 max, Xert is also pushing a lot of Sweet Spot workouts - it is possible to configure advice to your liking, but you may die on that hill while doing so :rofl:

I don’t want to make this post a complain about Xert - I like many of their ideas, but implementation might be problematic. If I play with settings, I can lower the load and increase recovery times, but these settings are not always documented well and may require an advanced knowledge on the way system operates. One example - why don’t they automatically shift recovery demand for a program based on age? You can’t really think that 25 and 60 year old athletes need same exact recovery time!
And - if someone learns and becomes proficient in Xert, I bet they can coach themselves without any need for their AI :grinning:

Anyway - I strongly believe that TrainerDay has a potential to get this kind of approach done well. I been (occasionally) using TD and monitoring/reading forums and changes. When I first tried TD more than two years ago, I was asking you to implement HR based workouts - and they are here and working well now.
So - how about getting some “dynamic” training advice system? :grinning:

1 Like
12

You understand the problems perfectly :slight_smile: Almost all platforms focused on the math and not the physiology. And from a marketing perspective intensity sells, and what I see at some level is once people invest in a platform they are more likely to stay there and make it work for them. I know many long term TR users that don’t follow their plans or advice for example but just stay for convenience, one is a really good friend and key forum moderator at TR :slight_smile: TR is a very solid product.

There is a ton of great sounding, un-validated scientific theory. TSS optimization for best performance is the most clear version of this and most platforms have invested too much in this thinking. Again, any of these can work well for the right person at the right time and some it can work for long term. So I am not trying to say it is all wrong, just based on the wrong premise. Xert has the same underlying problems but many people make it work for them. I like the ideas behind Xert but again it requires more investment or additional creativity/thought then they have been able to invest.

Regarding someone dropping in mid-stream plan/block, I would love to tackle this problem. I need someone like you passionate about it to help me :slight_smile: If I put my coach hat on I can solve this problem on a 1:1 basis but when I think about the amount of code required, the number of questions a person would have to answer and all the decision trees and thought behind all the variants it gets scary. It seems like something an LLM could handle but who really wants to put all their faith in an LLM for their training, or let’s say who should? With enough fine tuning and constraints it’s likely possible. It would be easy or at least much easier to “fake it” and make people believe it is a good idea with more un-validated science… I am not a believer that load = load… from a physiology stand point, so while that is one of the variables there is just a lot

13

This is a great discussion, thank you Alex!

It is kind of interesting that I decided not to test TrainerRoad - I think, over the last three years I tested almost all available applications for virtual cycling (at least to some degree). TR was front and center in my list, but even then it was the most expensive one. The price is not an issue, it is value per price - if I see TrainerDay or Xert with ton of great functionality, why would I even try TR? :grinning:
Maybe I’ll test them one day, but that day have yet to come.

I would rather spend time and help TrainerDay - if you think I can be of any help, of course.

And AI - I think, it has too much marketing around it. But do you reallyt need it? Why not to use some tried and true machine learning techniques and analyze training data for the last month - either uploaded or pulled from Strava. Since you have your own workouts CJ uses to build plans - maybe you can create some kind of training load “signature” for them and see if you can find similar work in historical data. This will allow your app to decide at what level new user was training and ask what way he wants to go - maybe he’ll be fine with lower load to start, giving some time to recover from previous plan or user might choose to start in the middle - if new plan is more or less compatible with what was done previously.

You are right - one training load may not be the same as another even if TSS matches. You can’t compare long endurance ride with much shorter VO2 max or that to HIIT! That’s where the pattern recognition is needed - how the distribution of power over time compares between workouts and rides?
I think, if this can be solved, it would be a great start to analyze training more efficiently, including ability to manage weekly mix of workouts and free rides.

14

TR is going to push you. The base plans are typically going to start with over/unders and sweet spot adding VO2 type intervals before base is even over. Some people thrive on the intensity and some will burn out.

Their calendar functionality is very good. The app works well. I think it is down to your perspective on the intensity and whether you want an adaptive plan or not.

Dave

15

This is exactly right. And really pushing you is not always bad unless it’s always pushing you :slight_smile:

1 Like
16

Yes, I am on vacation for the next 2 weeks… I still work, just minimized… but I would be happy to discuss ideas. I am always looking for optimum ideas that could pull this forward.

So TrainerRoad used machine learning in their solution but think about this. What does ML give you?

  1. Marketing to say your have an AI solution
  2. Either a lot of things we already know or hypothesis on new things that need to be tested/validated.

Remember if you tell ML that you eat ice cream every and your FTP went up by 30% in 6 months, it’s going to say you need to eat more ice cream.

The reason I say all this is because the data that we have in fitness is rather low quality and very sparse compared to everything going on. Very few people have very high quality diet, sleep, hrv and testing protocols in their data to know what is going on. So if you put bad or extremely limited data in you are not going to get much value out. Ride more get faster, sleep more and recover better get faster, don’t injure yourself or get sick…

I am a strong believer that LLM’s are a huge amount of hype.

As much as people believe a system should be able to match training to their physiology it’s not very true unless you can have deep conversations to understand the human as much as possible. What people really want is something to match their expectations not their physiology because if it does not match their expectations or beliefs they won’t follow it anyway… This is where humans coaches and humans in general come in, they can build trust, so when they say, start out easy, even though you really want to start hard, you might give it a chance… So LLMs have the potential for reasonable quality conversation and collecting the data that could get a better match or even convince the person why something is right for them.

Again this is not easy, just sharing my thoughts. I am looking for a break though idea as I am stuck in my own head on this stuff… There is some basic simple stuff we could do that would add value and improve the matching process and make it more dynamic. We started something like this, but even that got complicated fast. I would say most “AI” systems now are just using random math ideas to generate plausible ideas that don’t match human physiology… That does not fit my interest or what I feel is in the best interest of our cyclists even if I could make a compelling story around it.

17

TR has big money… I just saw this… it highlights my point exactly (regarding matching expectations)

18

Well, that’s my experience with Xert as well - Their endurance is actually Tempo and it has no issue giving two VO2 max or HIIT workouts in a row. I can sustain it, but after 3 weeks I have to ask myself - do I have to or do I even want to? I am able to keep my training level and improve slowly just riding on my own in Zwift and outside. I have no issue pushing myself - when I feel that I can sustain good load. Or take it easy and enjoy the ride when I need a break and I monitor my fitness via Intervals.icu. But now I want to do a good long route in Zwift with some climbs and after Xert been pushing me every day - I feel that I am not recovering enough next day to take the load I want.

Thank you for the feedback, @dthrog00, very useful!

19

@Alex please enjoy your vacation! We have plenty of time to discuss this topic without any rush :slight_smile:

Well, do not feed this data to ML! You need to understand what signals correlate first before applying ML to them. If you have a training load signature of a good set of workouts, trying to find same patterns in other workouts and free rides can produce valuable information. But this is very specific use case, applicable to structured workouts and rides with power recorded. You may be able to add HR to that, but this will vary more.

Other signals - diet, HRV, sleep - yes, these are much lower quality and often just missing at all.

LLMs in general - yes, hype. But pattern matching in specific datasets is doable. Not sure if it would be easy to train LLM, but use some trending/Fourier analysis - may work.

20

I’ll see if I can find time to read that TR thread. I like their forums - plenty of useful info, but also a lot of noise :slight_smile:

But I immediately see the same thing I saw repeated over and over - when there is a complain like this from new user of the platform, the recommendation is often for a user to adapt to the program - in my view, this very backwards - it is program which has to adapt to specific user needs!
It is hard to implement and all that marketing about “AI” (which, I believe, TR started first is sports training) will get us nowhere by itself.