TR could somehow account for that if the power source vendor+model was an input to their ML models (IIRC I asked that before and never got an answer).
@dhengen Best option would be to offset one power source to match the other source in the power meter configuration (if available as for most power meters nowadays).
That would assume all of one model is out by the same amount though.
I am on my second set of Garmin Rally’s and these ones are 1% under my KICKR, where as my last set were 3% over.
Don’t know if it is possible, but obviously the answer is use the same power meter indoors and out.
Otherwise - your question really is, how does the model deal with me giving it bad (or at least inconsistent) data when it doesn’t know I am giving it bad data? Seems very hard to solve on TR’s end.
@AussieRider, the not recommended warning for the Threshold workout is because the chance of failure is high enough that the system does not recommend it. But this does not mean you can’t complete it as you did 
And there are plenty of power meters that read a greater % off at higher watts than they do at lower watts.
They’d have to develop a tool that created a PM profile for every single device and equalize them somehow, which sounds like an insanely huge lift to me.
A bad night’s sleep and one failed workout meant my predicted FTP went from +1% to -1.5%. I feel like it’s playing head games with me 
Just had first workout since I came back to platform. App has me at 286 ftp and I did Baird -1. Wanted me to do 343w for the one minute efforts I turned up difficulty and hit over 370 for each rep. Im Still so confused with these settings, this was honestly as hard ride in the post ride survey nothing more. If I did 340 I can guarantee not effective vo2 max training. I will turn all vo2 rides up this can’t be correct. What am I missing.
Ps I have been training last few weeks and can back up after a ride like this so the arguments of each ride shouldn’t wreck you etc are not relevant as I say this was a hard only despite being a lot over target
Are you on a TR plan? Did you accept the AI FTP Detection? Do you have recent rides similar to Baird-1 at higher watts that would show the AI what you’re capable of? If so, is your VO2 set to Aggressive or Demanding?
Also, if you share your calendar/training history/future plan with us, we can see if we see anything that might help.
I have a theory as to why TR’s FTP prediction might be a little over-optimistic and why users may not ultimately to their number on detection day.
Correct me if I’m wrong @SeanHurley & @Nate_Pearson
The vast majority of my base phase intensity rides are expected to be “hard.” Those workouts have about an 80% to be hard and a 20% chance to be “very hard.” If, for example, there are 10 intensity days in the simulation, from what I understand, TR is taking the most likely outcome for each workout (in this case hard) and developing a progression and prediction based on that response.
But…. although on a per workout basis “hard” is the probable answer, on an aggregate basis, we’re statistically more likely to rate two of the ten very hard - a result which would drag down the progression/prediction. The odds that we’d rate every workout as hard in this example are quite low, ~11%.
So, unless TR is changing the rating on the back end for some workouts or using some sort of rating weighting that only applies in the simulation, it should tend to overestimate FTP prediction for many of us.
Just because you can do something doesn’t mean that you should.
Most of my interval days have been expected medium. I’d imagine this is individualized, though I suppose its probably in a bell curve.
I have an FTP increase coming up next week, after week intervals are expected to be medium, hard, and hard, followed by recovery. Just working through winter base phase right now.
In general, fitness has been coming back fast during base, things are feeling easy right now.
Useful thanks. I understand training zones and if I train vo2 max and top out below threshold HR it’s not effective vo2 training. Your message is patronizing in tone btw
It does mean that vo2 max training might be ineffective though.
If the power is not high enough to elicit vo2 max then that is a problem.
This has become a very long thread, so this question may be answered somewhere in here already. Is the AI Detection tool still using rides that happen outside of TR? Specifically, what about Zwift rides (since TR and Zwift sync now, which I like a lot)?
A week before the new AI rollout–on Tuesday, January 13–the old Zwift AI Detection Tool–which I had always found to be accurate!–said that my FTP was 302w. This aligned with what I was seeing. That day, I had done a Zwift Racing League Race where my normalized power was 296w for 62:00. Needless to say, I took it pretty easy on my TrainerRoad workout the next day, January 14, turning down the power targets on the Wednesday group workout that I always host through TR.
The following Tuesday–January 20–I had another long Zwift race, this time doing 75:00 with a normalized power of 276w. And again, I took my Wednesday group workout on January 21 easy. Then, on Thursday, January 22, I did a TTT on Zwift where my normalized power for about 40:00 was 292w. Later that day, TR rolled out its new update, and the new FTP detection tool said that my FTP was 268w.
Since then, I’ve done a few more Zwift races with actually better numbers–an NP of 309 for 55:00 on February 3, and an NP of 308 for 45:00 just yesterday–but my FTP is stuck at 268 on TR. I have manually changed it, but I can’t access any of the features of the new AI system unless I accept this obviously-too-low FTP from January 22nd.
I can see all of the Zwift races and rides on my calendar, but is the new AI not taking them into account? Is it only looking at the TR workouts where I turned down the power because I was recovering while my friends did the group workout? Thanks for your help.
Ok, I’m out. Hope you get better advice from others.
Well that’s his assertion, which may or may not be true. The system is designed to give the most effective workout to elicit training adaptations so I’d err on TRs AI advice in the first instance. Just because an athlete can go harder, and thinks they know better, doesn’t make it true - of itself. Perhaps the calendar they share will support one or the other, but this particular athlete doesn’t like my post and/or principle so I’ll duck out in the name of being excellent 
Huh?
It’s very possible to not elicit vo2 max if the work is not hard enough, that’s all my post was saying.
If the power is not sufficiently high to elicit vo2 max then that is a problem when you are trying to do a vo2 max workout no?
I’m trying to duck out here! Ha ha.
The athlete may think that say 120% ftp is what he needs to do for best training.
The AI may have the data to say he actually needs to do x Watts for optimal training.
But there must some probability that you’d rate it hard? I rate my intensity rides hard or very hard. Some people primarily use moderate and hard, but the same principle would apply.
That’s still missing the point I was making.
If the power prescribed is too low to elicit vo2 max then it’s not a vo2 max workout. If their HR is still below threshold it’s almost definitely too easy and is not giving the adaptation that the workout is claiming to give.