I’m happy to test
Also sync is down for me today, have tried disconnecting and re-connecting - still won’t sync todays workout
Unfortunately the reason is that the app reached daily rate limit for Intervals.icu API…
Good news, it’s not a sync bug and it will start working normally tomorrow.
1 Like
Too popular - guess its a good probelm 
1 Like
I have yet another request, I just got a core sensor, all the data gets written to Intervals from both my headunit (outdoors) and TPV (indoors). It would be good to see that added to the charts/zones etc for time in HSI zones.
I’m sure the coach could then use as another metric, be a good differentiator for your platform from all the ai coaches.
I have just downloaded the app via play store,
Still trying to understand it, I think it will be good to add other watch brand too like Amazfit, Coros and the likes. Currently I see the only watch that can sync directly is Polar.
That’s great idea, and I’ve already added support for the core temperature streams. They are processed during the import, so you’ll see them only for the activities that were imported starting today.
1 Like
Hi,
thanks for trying out the app. For the apps/devices that MyTrainPal does not connect to directly we rely on Intervals.icu as a data source. The full list is displayed in the app’s connection settings.
Some apps/devices offer direct integration for more seamless experience, but adding support for more takes time and sometimes it’s not possible at all (like Garmin that has the developer program submissions closed currently).
If your experience with the particular app/device is significantly worse because of integration through Intervals.icu, please let me know.
1 Like
Hiya, still broken this morning.
I have a load of missing workouts for the next 2 weeks on intervals that are planned in app.
I manage to get Saturdays to push but it needed an amend for the zones range in the app, coach did that fine but now it can’t update on intervals.
I have disconnected and re connected and done manual syncs as suggested by the coach but everything is stuck again and says sync failed.
If I manually edit the workout I get th option to push to intervals in the 3 dots but that is failing to due to rate limit.
Hi,
As you probably noticed it’s a rate-limit again.
@david could you please help here, it wasn’t a problem until yesterday. Did anything change on your side? Is there anything I can do to resolve this issue? Thanks in advance for help.
Best Regards,
Krzysztof
1 Like
Did a workout today with core sensor on - when I try the “Heat and temperature” chart it just errors.
I can send you a .fit file with the data if it’ll help you troubleshoot it
So far new coach ai not so great - My workout tomorrow was missing, so I sent it straight from the workout, it sent an old version, I question that and the ai said delete from intervals and the right one will send. I did as told, it then vanished from ap due to sync, then it couldn’t recreate it, then it old me it had, but hadn’t - finally after a load of prompting it made it and it synced.
I’ll keep persevering for now
Thanks for sharing this feedback.
I have to admit that while new model was providing generally better, more interesting analyses, it can be less reliable for some administrative tasks.
You don’t need to persevere btw. 
just let me know and I’ll configure previous stable model for you.
Thank you, I managed to find some example form core sensor online. I pushed fix to the web version. I hope it will work with your data as well. If it won’t I’ll need you to assist me by sending the .fit file as you generously proposed.
It’s working now, can I request that all 3 tooltip values go to 1 decimal point, for HSI that’s quite important but with be good for temps too.
I had a chat with the new ai re the workout, initially it flagged cardiac drift, I explained it was heat stress and had a core sensor on, it came back with better analysis, but said i was dangerously hot. I told it go and read core’s website re how heat training and their metrics work, it came back with way better analysis, apologies and referenced all their zones.
I asked it to remember all what it had learnt in regards to my heat training and adaptions and it has committed it all to memory. So I think the new ai did a great job in that respect, hopefully it can improve on workout generation and amendments.
It should improve in that area as well. I’ve also pushed a small fix to improve the precision of the tooltips.
Thanks again for all the feedback you’ve shared so far. A lot of work has gone into the backend to make sure the app is more than just another generic AI-slop training tool, so I especially appreciate input from experienced users who understand their own training needs and are willing to share thoughtful, constructive feedback.
Thanks again for the kind words. I know you posted this more than two months ago, but while reading through some forum threads I came across this paper: [2503.14841] The three-dimensional impulse-response model: Modeling the training process in accordance with energy system-specific adaptation
I’m guessing the matching name is not a coincidence and that you’re one of the authors, if so, huge respect for the work.
The 3D impulse-response model immediately caught my attention because it seems like it could be particularly well suited to AI-based coaching systems such as MyTrainPal. Some of the concepts described in the paper are already reflected indirectly in the AI’s reasoning, but there is currently no explicit tooling or guidance telling the model to rely on this framework more heavily when the necessary inputs are available, such as power-meter data.
I’d be very interested to hear your perspective. Does my intuition make sense that this type of model could work especially well in combination with AI-driven training analysis? And have you experimented with using the 3D model alongside AI or LLM-based systems yourself? If so, I’d really appreciate hearing about your experience.
Cheers,
Krzysztof
2 more QOL requests please.
Add HSI zones to the time in zones charts
For Personal records, don’t use virtual rides as these are all unrealistic, I have some great IRL PRs from time-trial and races but they don’t show due to artificial virtual numbers. The 100km is from an IRL ride but all the others are from a virtual race.
Hi @Krzysztof ,
Yes, that is indeed our paper! Thanks for your interest. It is purely theoretical at this stage, and obviously a simple model (even if 3D instead of 1D) can’t fully capture the nuances of adaptation, but we think it’s a step in the right direction as it includes an aspect of specificity.
I’ve recently finished collecting data for a validation study where cyclists (half of them recruited from Intervals.icu) did weekly testing while all their training data were collected and strain scores calculated as described in that paper. I don’t want to speculate too much when it comes to the results and conclusions, at least publicly, as I only have the preliminary results so far (these data can be analyzed using many different approaches) and none of it has been submitted to peer review yet. But, there are certainly interesting trends and findings which likely will help the field to move on from the simple CTL/ATL model that in many scenarios is less useful. These ideas have long been implemented by Xert (https://www.xertonline.com) btw, although the approach is lacking scientific validation.
I’m an exercise physiologist and not a developer so I’ve only used AI for data processing and analysis, but the model could certainly be implemented in AI-based systems. Not sure how practical it is to individualize the 3D model parameters though as it requires lots of maximal test efforts (like in the validation study we have just finished). That would be the ideal scenario but without it you’d just overfit the sparse data.
It is still a relatively simple model with three fitness time constants (3 x tau1) and three fitness coefficients (3 x k1) at a minimum. If the fatigue component is added (this is sometimes a little tricky to model), it is twice as many parameters, although tau2 can be combined into one in my opinion as fatigue is not clearly system-specific. So it might become a 10-parameter model, which realistically you cannot use for most individuals in real life due to the lack of frequent maximal performance data points. But, perhaps you can fix the parameters to population averages or give them some room to move based on individual characteristics (because it appears that both k1 and tau1 have huge individual variation and they are affected by baseline fitness, training history, and training intensity distribution).
Me again, back on HRV, I know I mentioned that the way your iOS app only captures one figure at a random time each day is far from ideal and leads to massive swings and renders the readiness score useless.
You said you’d add it to the roadmap. Any idea when it will be implemented as it’s a bit of an issue for now and how the coach thinks you are capable of todays workout.
Here’s today figures from MTP, Bevel and Apple Health, you can see bevel thinks my normal range is 55-92ms, Apple has my average at 76ms so both agree, MTP has my average at 156ms which is just wrong, my HRV today is 55ms which is fair, yet MTP has 78ms which if it was correct is bang on my baseline, yet it thinks its suppressed.
