That is crazy to me and I don’t believe it. Most everyone I ride with trains outside. I think obviously TR athletes skew inside, but I still wouldn’t believe that it is “rare”.
What I do believe - First is that TR is still a very good product for lots of athletes. That said, I have chosen not to use the product as they don’t fit how I train. One, I think they continue to cater their product towards the athlete that is going to train indoors, and they have made mistakes in addressing athletes that train outside, like this bug. I also believe that they cater to a lower volume athlete with not a long history of training. Bugs and distortions in training plans seem to pop disproportionately for higher level athletes (longer hours, higher FTPs, mature in their training history). I even wonder if their data that everyone talks so much about is overly skewed for these demographics, and they don’t have good data for the athletes I am talking about.
And that is ok, perhaps they should focus on those markets as that is their target market with the largest opportunity. But until they expand their focus more, it is not a tool that works for me.
Riding outside ≠ primarily training outside. I ride outdoors 6 days/week as I use my commutes for (Z2) training. But I do all intervals indoors. Where I live I cannot do long intervals or do hill climbs (we literally have no hills).
What percentage of TR athletes are primarily doing intervals outdoors. That’s something only TR can answer for sure, but if I had to guess, it is a small minority.
If you included people who do endurance rides outdoors or occasionally do intervals outdoors, the percentage might be higher.
When I read this bit, I felt as if you ascribe intention to that. It may very well be a self-reinforcing effect where because athletes who primarily train outdoors tend to be on other platforms, they make up a small share of the overall pool of athletes. Any issues that they run into will be harder to see in beta tests simply by statistics (≠ intention).
I don’t know. Higher-level athletes have more specialized needs that are harder to meet by a platform that is meant to cater from beginners to Olympians (and has).
The majority of their users will be, as you describe, beginners with a lower level of fitness and experience. But that’s just a fact that there are way more casual athletes out there than people who e. g. race.
I’m curious whether TR’s algorithms work better or worse for specific cohorts. It’s not a question only TR can answer. But I’d be cautious to extrapolate from what you see on these forums and think this is representative. It is whatever the reverse of survivorship bias is, you only get to see when people care strongly enough about giving feedback and know/care about online forums.
But at least for me (N = 1, >= 4.5 W/kg at age 45), TR AI has managed to give me my second-best overall peak. I think I was never better doing long sweet spot than last season. The algorithms and the new possibility have made TR a better product for me.
So you quoted me - which said “everyone I ride with TRAINS outside.” All interval sessions etc. When it is nice and warm outside, there is no way I am going to sweat away intervals in my basement, as is the case with everyone I ride with.
How representative are you and your riding buddies of TR athletes as a whole, though? That’s what matters.
Just to emphasize one point: being statistically an outlier doesn’t mean you’re doing something wrong or what you are doing is bad. It’s not a value judgement on my part. Just that there are fewer of you than those who primarily train indoors.
I do train /not just riding/ only outside all weather all seasons.. i missed workouts during snow storms and heavy rains / thats normal/. I assure you that in my area roughly 30k people and a lot of riders during the spring, summer, and autumn in the winder riders training outside are counting on one hand. There are always some commuters.
I do all of training indoors with an endurance ride outside on Sundays.
As of now for for me I can say the prediction has been on the money +/- about 1 watt when I haven’t made changes to the plan.
Even with the one or two Very Hard rating when it should have been Hard or Moderate when it should have been Easy. That one threw me for a loop.
This is riding five days about 6-7 hours a week.
I’ve noticed you’re using terminology like “a small percent predominantly do their workouts outside”. I think that’s true and it’s a great point, but it’s not clear to me if the bug only hits you if you do more than x% outside? Or, if you did one workout outside in February, will it still be impacting you in August? Does it only happen if you do outside workouts “predominantly”? Or, if you just go for a 2 hour zone 1/2 Sunday ride once a week, will this impact the prediction/detection, or only if you pair that to a workout? There are just so many possible scenarios, but it all gets thrown under the blanket “outside workout bug”.
This is why I keep trying to nail down what the true known issue and end state is. I know it comes off a little combative, but that’s honestly not my intent. I just want to clearly understand what the known issue is and what the “bug fix” will truly solve. I think a lot of people will find heavy frustration if, after what is now at least 7 full months of waiting, we are told “The Outside Bug” is fixed and it doesn’t solve the problem they’ve been patiently waiting to see remediated.
I think that the more of your training you do outside, the more significantly you’ll be impacted.
Also, I don’t think that doing a workout outside months ago will have lingering affects today. While I’d expect those symptoms to wash away before long after you went back to training inside, I’d assume that they’d completely disappear after your FTP cycle.
Again, I really have to be careful here and speak a bit generally, because the model is complex, and each situation is somewhat unique, but the goal of the work being done here is to eliminate unnecessary FTP volatility through predictions and detections. Predictions will be more accurate from the beginning, and they will accurately reflect the work done rather than bouncing around and dropping down when they shouldn’t.
I would describe “the outdoor bug” symptom as "predicted FTP and FTP detection reduces even if you perform workouts perfectly"
That is different to the (I believe) more common situation where the predicted AI FTP is not achieved because the athlete hasn’t done all the work or their body has not reacted to the training as predicted.
The key difference is that the bug causes an actual error in the FTP detection - and this is an outdoor only phenomenon as far as TR knows.
It’s worth noting that things like power data pulled together for marketing isn’t at all the same as what the product and engineering teams use.
There’s been a lot of recent play with pulling new types of data with new tools, but that’s not necessarily representative of how the rest of the company works.
@eddie if I ever meet you irl, I am absolutely buying you a drink for actively dealing with, and responding to, all of the frustration regarding this bug from others and myself. Hopefully it gets resolved soon, otherwise, it will be winter soon enough and indoor rides will be back with potentially less aiFTP anger
It’s unfortunate that outdoor workouts are impacted the most. In my experience here in Colorado, the athletes who are most serious about their training tend to train outside.
Prediction has always been higher than what I’ve actually achieved, but every drop in prediction corresponds to me either struggling with or failing a workout. One thing that I’ve said in other threads that bears repeating is that “struggling with” is fairly independent of my reported RPE. If I have a workout that is harder than it should be (usually due to poor decision making on my part), my elevated heart rate gives me away, regardless of how much I protest that the workout felt fine.
Thank you @eddie , it would be great to get some feedback on this one. Hopping off the bike in the recovery between intervals is what I tend to do. Just to stretch a bit and have a bit of a shake out. I can’t recall this happening previously.
I think it does go beyond a visual thing. I downloaded the FIT and imported into my own app. This is where I noticed the lap avg watts were off (some intervals weren’t picked up). So I guess what I saying its recorded into the FIT file. Oddly enough when I looked on Strava for the synced workout it showed my maximum speed and distance was 253kph / 253km! I can’t remember taking off on the workout
EDIT: Just remembered something else that caught me by surprise - I got asked why I found the workout difficult question. Assuming it thought I didn’t complete full interval at prescribed power (when I did).
I understand the thought experiment, but this isn’t how TR has ever worked for me.
Here’s a diagram of the TR-exported FIT file showing how the lap boundaries drifted relative to the paused workout timer.
I completed every interval, pausing when needed—the TR software displayed “Paused.” However, the exported FIT advanced the lap boundaries using wall-clock time rather than the paused workout timer.
This appears to have pinged my AI FTP prediction, and I guess future workouts may now be adjusted. No dramas, I just thought this could explain some of the issues others are seeing.
cc: @eddie may be helpful for finding out what happened
