When you’re in erg mode and complete a workout, how does TR AI determine how well you’ve done in order to adjust your next workouts? Does it rely solely on your RPE feedback?
There are quite a few factors that feed the model, but power data and RPE are big ones. 
But if you’re in erg mode, the power is set by the workout and the smart trainer isn’t it? As long as keep pedalling at whatever cadence suits you the power output is consistent, isn’t it?
Yeah, AI will likely score the power portion as spot on (or pretty close to) target. It will rely on that and RPE and heart rate to determine your future workouts (and also possibly some other metrics that TR hasn’t advertised to heavily).
We still look at the power data objectively, regardless of what trainer mode you use.
Power can fluctuate in ERG mode if you’re struggling, and there can be pauses, changes in intensity, etc., that we look at when analyzing the work done.
ERG is always adjusting based on the torque and cadence you’re producing, so the data is by no means perfectly smooth, but yes, it should usually be easier to hit the power targets consistely in ERG than in other trainer modes.
If you compare a couple of rides in erg mode from easy to very hard you will notice (at least it holds for me) that for easy rides your heart rate increases and plateaus and your rpm is more or less constant, for very hard rides my cadence is high in hard blocks until the last one where the power remains, but my cadence drops from 95 to 70. In very hard to failed rides the same happens earlier until I can’t hold the power anymore, the trainer goes into a dead spiral and I have to stop. These effects are so obvious that I’m quite certain there are many smaller ones I don’t even notice.
Thank you guys.
Both Ed and Ben have mentioned “other metrics”. I am curious about what these might be. As far as I am aware the only metrics that I deliver during a session are power, heart rate and cadence.
I think that I am right in saying that heart rate is not a perfect measurement due to the other factors that can affect it, other than the effort of the session ( which is why power is a more accurate measurement)
That leaves cadence which, as you will see from my two sessions attached, fluctuates a lot, particularly in the one where RPE is Hard. In this case, I know that this fluctuation is due to me struggling, but in other sessions I will purposely vary my cadence for variety and different objectives. Indeed some rides that have instructions ask you to do this. So if TR looks at Cadence in its assessment, I would like to know how it judges the fluctuations please.
Thank you for your help
What are you using for a heart rate monitor? The heart rate trace on the second workout you have posted is all over the place.
That would have been a Polar, which I have now replaced for the reason that you have spotted
One thing they seemed to be alluding to was the micro-variations in power that still develop in erg mode as the trainer adapts to your changes in cadence or any faltering. The power graphs may or may not show it depending on smoothing options, but your ability to hold the power through every microsecond is certainly measurable by the computer.
Is TR AI now using HR? In the past TR did not look at HR. If it is now using it, that’s new.
They are using HR now. They talked about it when the new AI rolled out a few months ago.
They have been using HR for a while now.
They said during the new model roll out that they were using HR for some time prior to that as well.
Thanks. If that is the case, is it reasonable to be measured based on a mechanical aspect outside of your control, I wonder
This would be huge if true. When I inquired specifically 3 or 4 months ago, the answer was no. And, it was related to asking why there would be any adaptations if a plan’s ERG mode workout was executed as scripted, meaning no deviations from the power or times. At the time it was stated that the AI would somehow magically know, which I consider to be basically BS.
Thankfully it is true, from what TR has said the ai weights power and RPE higher but does also looks at HR.
The system checks how closely you matched the target wattage intervals during the actual ride block. I did a sweet spot session last month where my smart trainer kept dropping connection, and the algorithm flagged it as a fail even though my manual survey said it felt easy. The actual compliance file overrides the questionnaire metrics.
Is there any official TR doc or communication that describes their new use of HR data? Where can I find it?
We don’t have any official documentation of how we use HR data in the model.
All I can say is that we do process it with everything else.