The difference is that TR knows what training plans athletes are supposed to follow, whereas Strava and Garmin do not have that information. That is a key advantage and makes TR’s data pool unique in the business.
I’m working in data analysis and this kind of information is gold, worth much more than a bunch of ride files.
Could you explain how that elevates the accuracy of recovery suggestions above Garmin (or other platforms) that use far more data points? As that was the original point, and I’d be interested to understand how you feel TRs dataset with knowledge of the chosen training plan achieves that.
As I’ve said I think RLGL is a nice addition and should lead to some really cool additions in the future, but some of the suggestions in this thread about the quality of this function as it stands today are a little far fetched.
But RLGL does include unstructured rides? Unstructured rides are taken into account for everything bar PL’s at this stage.
Yes….i got a red light for Sunday after a very long unstructured Zwift ride and I got yellow lights after some stacked, long-ish outdoor rides a few weeks ago.
Its doing so for me, my weekend rides/ commutes (HR data only) its triggering yellow days following and after my commute it has adapted my indoor workouts down. As per the TR vid:
I didn’t say that it doesn’t?
Caveat: following is possible but I don’t actually know whether this is taken into account behind the curtains.
So, if AI knows that after back-to-back moderate/hard days follows rest day, it can be more lenient before triggering warning on 2nd hard day? This pattern describes long weekend + Monday’s rest layout in stock TR HV plans.
As opposed to Garmin that in perpetuity thinks that before my next workout I should take recovery for 24-72h.
Presumably, TR have more outcome data to feed into their ML; ramp test results, FTP changes, along with survey feedback, what plans, plan compliance, etc.
I would delineate between better detection and better suggestions. Of course, I don’t know how WLv2 and RL/GL work in detail, so I can’t say how it actually works under the hood.
One obvious difference is that RL/GL modifies your training schedule, i. e. it won’t just tell you to take it easy or take a rest day, but it will adapt your entire training schedule. Garmin and Strava only see what you do and can make suggestions in the moment, what you should do today or tomorrow. They can’t tell you what workout you should do next week.
You could see more features building on that, e. g. that Adaptive Training v2 will change training volume according to your fatigue levels.
I see also advantages with fatigue detection: if I’m flying too close to the sun, I often see a pattern where I manage two of the hard workouts, but struggle with the third. RL/GL has often added yellow days after the second hard workout.
Moreover, if you do TR workouts, you can compare power targets and expected heart rate with actual power and heart rate. I reckon unusual heart rates at given power levels or not hitting power targets is seen as an indicator of fatigue.
On a technical level, here is how you could do it: on the level of individual workouts, you could introduce fatigue/freshness as a parameter and then predict e. g. power and heart rate for a given workout type given this level of fatigue. Inverting this relationship would then give you the fatigue level. You could then use this to bootstrap to the next level and consider actual fatigue with predicted fatigue if you followed the next workout(s). Based on your past workout history, TR could compute individual threshold criteria for green, yellow and red. I would expect that the actual algorithms are more complicated than this.
Garmin cannot extrapolate training plans with good accuracy, the best they could do is roughly anticipate how many hours you might train in the near future. They could try to predict your training plan, but that’d be a giant effort and very error prone. At present, Garmin does use a few data points that TR does not at present include, most notably sleep data, resting heart rate and heart rate variability. But unlike Garmin with training plans, it seems more straightforward to me for TR to include those pieces of information.
Indeed I agree with the crux of that, but I am differentiating RLGL from adaptive training. In fact, I agree with you completely (as I stated before) that RLGL is foundational work for further potential functionality.
Also to repeat myself, I like RLGL. It’s a great addition. I just struggle with the suggestion that it gives a ‘better’ indication of when to provide a warning / suggestion to rest than other platforms. It does not look forwards as far as I can tell when determining whether to show yellow or red, only backwards. You can test this easily. Stick in a load of workouts and previous RLGL colours do not change.
It is probably better than TSB, but I don’t think many people use TSB as a black and white indicator of when to rest. RLGL seems pretty reasonable to me, but it’s just another data point. If I could only have RLGL or what I had before (Garmin training readiness, intervals.icu load and form, HRV, feel) then I clearly get a much better overall picture of when to decide to rest from the latter. Of course I don’t have to make that choice, so all is good.
It does integrate into AT by offering to adapt your training plan, so I don’t think you can separate the two. To be honest, I don’t know whether TR considers RLGL a feature of AT itself or an independent feature that integrates with AT. Probably semantics at the end of the day.
It only adapts workouts that are part of TR training plans. Note that AT (and thus, by extension RLGL) does not constantly adapt your entire training plan even though it could, only workouts that are in the near future (block-wise as far as I can tell).
Yes I know that. I think you have to differentiate RLGL from AT even though the maths behind RLGL informs AT, for the purposes of this debate, which is whether it does a better job of telling you when to rest than alternatives. We’re not debating whether it does a better job of adjusting your upcoming training plan, seeing as only TR offers that.
Correct, AT will only change workouts within the active/current training phase.
Where I’ve a hard workout scheduled on a yellow day, AT changes it to an easier endurance workout (I have had that happen to me, when I was moving workout days around to suit life). So I personally don’t think you can differentiate RLGL from AT. It’s all part of the TrainerRoad system to me, given that they impact each other.
Is that true? I feel like I recently read something saying that even if you have a custom workout scheduled, RLGL will suggest a more appropriate alternative (presumably a TR workout). I was surprised by that (in a good way), so it stuck in my head. I can’t remember where though, so I could be wrong.
They don’t impact each other.
RLGL obviously impacts adaptive training. It would be nonsensical if it didn’t. However your training plan doesn’t affect RLGL in any way. It works exactly the same with a calendar full of non TR activities.
That’s why it is fine to have a debate around how good RLGL is at suggesting when to rest without bringing AT in to it.
AT makes decisions for the future, RLGL is only interested in the past.
It is certainly using the past as it’s foundation, but the present (aka today) is the reason for that prior focus. RLGL can make changes (for those using TR plans) on top of the basic R/Y day color which is a basic recommendation for the current day (which can apply to any non-TR work as well). And keep in mind the whole reason RLGL does what it does is with an eye on the future, to mitigate or prevent overtraining. All that to say that RLGL is about more than “the past”.
I disagree. The colour of the light is not in any way impacted by your training plan for today or the future.
Adaptive Training takes all of that past information that is used for RLGL to suggest or adapt training.
That may be semantics, but it’s important when the suggestion that started this was that somehow RLGL was better at telling you when to rest than other platforms. (Not by TR staff, btw. By TR fans that got a bit carried away).
Yeah, I avoid “better/worse” types of comments, especially when there are unknowns like the back end of RLGL that we simply won’t get in enough detail to make a comparison to the other methods that seem mostly known. To be honest, I simply don’t care to compare the other options and evaluate RLGL with respect to my actual experience via notes and results.
They have for me in early access. I got a yellow day, which triggered an AT adaptation from VO2 Max workout to an endurance workout? I never had such adaptations before RLGL.
Saturday Long outside spin on a yellow day → Sunday Red Day (rest day scheduled) → Monday harder than expected return commute (thanks to wind) → Yellow Day, which AT suggested adaptation from the Planned VO2 Max to Endurance.
The only caveat is that I was picking more “on off” than “sustained” VO2 Max that the plan originally suggested, so I honestly can’t remember which one adapted but I definitely got an AT adaptation.