Câmon @Nate_Pearson , you guys will have no problem to integrate RHR/HRV (rMSSD/SDNN) plus a few subjective scores (mood, stress, soreness, motivation) in the algorithm to get a better readiness score to train. Specially targeting against the post workout evaluation. We canât have just load to do this
Anyhow, thx for the improvements.
@Nate_Pearson Will this update help users understand the long term differences and adaptations between a regular training plan and a master plan?
Unless the data isnât useful, then there is a problem.
Listen to recent podcast where Jonathan discusses how all his devices disagree with each other and how often times users notes about hrv and other metrics donât line up to their workout results.
Long time and very regular user of TrainerRoad. I like the new interface. Itâs logical and intuitive. Seamless to select alternative workouts (easier, shorter, harder, longer). The predictive difficulty is a nice touch. I think TR did a good job with these improvements.
I donât see the ftp prediction feature
My only gripe and this is a legacy gripe, the post workout survey should be a scale of 1-5 âhow hard was thisâ. Nothing is ever truly easy and I donât think Iâve ever rated a workout âvery hardâ.
Please tell me youâre going to name the LLM feature something like âCoaching Heuristics & Adaptive Dialogueâ or the like.
I love the way he thinks, that we think, that weâre chatting to humans in these forumsâŠ
Of course. Low HRV/readiness doesnât mean you canât complete the workout; it means your body wonât absorb the training load effectively. Itâs 2025, and Jonathan is still posting videos claiming that Zone 2 training isnât meant to fatigue you so you can train harder. I understand how this aligns with their business model, but at this point, TrainerRoad has effectively become, for me, a fancy workout player.
As for their metrics, what we actually need is an individual ML model that factors in all of our data, links it to training outcomes, and predicts a truly meaningful readiness score. This is a relatively straightforward problem, and they already have massive amounts of dataâat least on the workout sideâto make it accurate enough.
My issue is that you can throw in whatever fancy AI they want to claim, but if it isnât using physiological and subjective data (e.g., how you slept, work stress, family issues, or whether yesterdayâs run left you sore), it will never be good enough. Ever.
I agree, there is evidence that shows how useful some of these metrics can be. In my personal experience HRV etc. were not useful at all for my training/early detection of illness or whatever.
Studies show also how useful surveys in the morning about feeling, stress, etc can be. I could not find a platform that was able to make any meaningful use of it.
What works shockingly well for me is ChatGPT. Over the last twenty years I was lucky to learn a lot about training principles, so I do not desperately need a training plan.
Right now I am experimenting with Chat GPT as a Triathlon Coach that doesnât prescribe me workouts, but works as a interlocutor. I tell it, what I do, how my sleep was and how my body feels, and it helps me to not overshoot and take breaks, when I would just dig me in a hole.
In my experience this works a lot better than metrics that try to know my body better than me.
The strength here is in my opinion, that I can give a verbal feedback to my sleep, energylevel etc. and how that changed after a workout/sleep, etc. That is far more nuanced than a slider from 1-10. Since in triathlon you do often more than one workout a day, Chatgpt is with this constant feedback very close to the state of my body and mind and adjusts.
When its loadmanagement seems strange I can discuss with it, when I would decide differently.
Long story short: If Trainerroads integrates a LLM (that would probably far more sports science specific than ChatGPT is and would have direct access to my real workout data), I imagine this has the potential to be far more usefull than metrics, if I could talk with it about feeling/sleep/etc. and it makes recommendations how to manage load and not get carried away by my overambitions.
It would be amazing to have one more tab within the TR app, just to take your morning HRV measurement (1-2-3-5, whatever you prefer). They could even include an orthostatic test, finishing with a simple subjective evaluation, and thatâs it.
Itâs good enough for me! 
Is that based on personal opinion? Or the TR guidelines?
just noticed it appears the TR guidelines disappeared on TR app o ln the iPhone.
Someone shared this before and I find it helpful.
Thatâs helpful. Thank you
Thatâs not solidly proven though and that is part of the issue .
Nate has said that they want to get this data from more wearables and see how it looks in the data. They may then be one of the best sources for that with all the ride data they have. But no oneâs proven this is as viable a thing yet.
Nate also mentioned that the pre workout subject survey they have had for years canât be tracked to anything reliably which is why they donât use it in their modeling.
Again listen to the podcast. Jonathan has two sleep trackers and they can vary by hours. One doesnât pick up when he gets out of bed in the middle of the night. Detecting sleep quality varies between them.
Yes if they can apply it that would be great but the data doesnât seem to be great.
Wearables sux. You need a reliable chest strap and do this first thing in the morning. Yes, thatâs a barrier.
Cool so itâs not happening then.
So I currently donât follow a TR plan or workouts as I have a triathlon coach who sets the sessions for me and I then do the bike sessions in the TR platform.
I use the FTP auto detection but can I now really rely on this? As I understand it the new FTP AI model sets the FTP to better select the workouts you do rather than being a reasonably accurate estimate of your actual FTP. Do I have this understanding correct?
If I do then it seems I am having to revert to old fashioned testing when a big selling point of TR was taking that necessity away?
If I canât rely on FTP AI detection then I may have go consider dropping TR and just use my Training Peaks account and their virtual system and do tests and rides on that platform.
I would say that this is being completely blown out of proportion and that for the vast majority these numbers will be within a rounding error of each other.
The new model doesnât even use ftp for selecting working itâs.
Letâs see when I get my next FTP number I will test it against a ramp test result.
