They seem to have repeatedly said that the model does not use FTP or workout levels, and the FTP and workout levels are really just the way the model converts back into something “understandable by humans.” If the concern is that the plan does not focus sufficiently on building out sustained power, then perhaps it makes sense to choose a different event type. For example i expect that selecting “time trial” or road race with long climbs as an event would likely have longer intervals at sweet spot and threshold.
I’d like to say a massive thanks for all the hard work the TR team has been putting in to this project. As with all tech/paradigm changes, there will be those that preferred the old model and that’s fine. There will always be some that dislike the “non-standard” version of FTP. Possibly due to the new number being lower (my new TR\AIftp was 25w lower than previously).
Despite this, the workouts I’ve been doing are far and away more productive, being at a higher avg wattage per session compared to previously, I’m actually more fatigued than before, and can’t wait for the rest week next week.
The “in your face” nature of the predicted change to my ftp does play havoc on my type A brain. I.e., I’d probably benefit from swerving Saturdays OU due to nagging illness, and fatigue, but downgrading or removing the workout results in lower number presented right there and then. I suspect many on here feel this way too.
Regardless, the ai adaptation is exceptional. From a pure technological perspective I don’t think people can appreciate its complexity, I certainly can’t. Chapeau. Keep up the good work.
I find it interesting to read that you also had a number of failed workouts after receiving the inflated new AI FTP. I wonder what the percentage is across the group of those of us that were at higher PLs before and hence received a substantial FTP increase. The data that @Nate_Pearson presented gives around 1.7 % of failed workouts over the entire beta user base, as far as I understand, but for me personally, the rate was as high as 50 % in the first two weeks; even though none of the were actually marked as “failed”, yet I had to cut intervals and reduce intensity. Maybe this polishes the statistics a bit…
Even 1.7% is too high IMO. I can’t even remember when I last failed a workout before that (OK, I just did, but just that one mixed interval a few months ago) and I thought the whole idea of TR was to aim for 0%. If the AI is now assigning workouts while considering 2% failure rate to be acceptable, I can see it being way more aggressive than before. I don’t mind that personally, as it’s new stimulus, but I do have this doubt in my mind if I can sustain it over a full season. Moreover, the failure rate was most probably spread unevenly among the population, and if AI wants to even that distribution out… ouch!
First of all: I had the pleasure to use this shiny new scale for a while in the Beta, and from my point of view it is a total gamechanger. Seeing in real time which effects which kind of training will have on your future progress (is it better to do a long work on Sunday or two shorter ones on Saturday and Sunday, which effect will commuting rides have and how hard should I do them etc.) is amazing. Thank you @Nate_Pearson and team for the great work and making this happen, and also for being so open and transparent in the communication!
I have one idea and one question with regards to the new scale, to stay in the analogy:
- idea: is there a possibility (maybe optional?) to have the new scale re-weigh you at the points previous historic AI FTP detections were done, so to effectively recalculate the FTP history? This would be great to see the historic development and comparing the current status with previous periods by actually comparing apples with apples.
- question: does the new scale also weigh the effects that non-cycling activities have on the predicted FTP (especially runs)?
For what it’s worth, I was a little concerned about my new ftp going up 10w with the new calculation especially as this was the first week I was back in full training action this week after taking a couple of weeks off earlier after stupid freak injury over xmas.
I almost said not to go with the new ftp but thought I’d trust the process, yesterday on new system was just 30m easy session after longer endurance the day before but that was actually a more interesting one than it had ever given me on the old system and the new zone 2 power was higher but my heart rate was still in zone 2.
Today was the one I was worried about 1.5h under / over threshold / sweet spot I hadn’t done before and with the new ftp just went for it and I’d say it nailed it. Literally got to the last 1m of the last threshold over wondering about my life choices but made it. Heart rate was high but in threshold didn’t blow me up was just hard, as it should have been, then had 12min sweet spot to finish with (thx for that 
) but again managed it fine.
Tomorrow I’ve got set up for the new dynamic solo easy ride outside and that’s nice to have that changing depending on workload.
So obviously early days but positive experience so far, also nice to be able to play around with workout types and actives and how they might impact future training as well.
Mark
I really like the concept of PL within workout types for all the reasons you describe. That is why I am uncertain about a system that tries to drag me back towards threshold PL 3 every month.
I’m curious why my FTP went down 7% after the update. From 300 to 279. That seems like a large drop. It’s not a big deal but I wonder what about my history would drop it that much. I’ve been hovering around 300 for a year or so I think.
I think the best solution is to have two detections:
The first would be as far out as the system feels is sensible (at the moment I guess this would be four weeks) and that moves on a daily basis so it’s always as far out as the AI simulation window allows.
The second would be one that the user can drag and drop, anywhere between today and whatever date the first detection defaults to, but not beyond.
I would suggest watching the Q&A - https://www.youtube.com/watch?v=X0_ViMBmHGM
I can’t remember who or where, but I feel like one of the TR folks said they’re going to make it movable.
I have. They are none of the obvious reasons.
You can easily get 0% by giving too easy workouts. Listen to the podcast where they discuss the new model. You need to balance giving too hard vs too easy. It’s like with government programs the optimal amount of fraud isn’t zero because of the cost to prevent it exceeds the cost of the fraud.
That’s my point! No FTP test is a one-size fits all thing for all users. Some formats of the test will be more accurate for setting training zones in some riders, other formats of the test will be more accurate in other riders.
You always need to calibrate after the fact. It’s just some number to guide your first guess of where about to roughly start your power targets.
And since the test is not the answer in isolation without combining it with additional real world data, the exact test you do is less important than the real world calibration part.
When people get a big discrepancy between their FTP test results and what they can actually do, we tell them to base it on what they can actually do. Because the real-world calibration step is more important for making sure training zones are setup properly than the FTP number.
So yes, it would be really nice (especially for novice riders who don’t know what things should feel like) to find a better way to set training zones.
And if TR thinks they’ve found a more accurate way to set training zones, one that also doesn’t need to involve regular formal testing (which decreases rider compliance), that sounds pretty sweet.
Here is something I dont seem to understand….
You’re basing ftp on level 3 threshold. Statically defining a time and duration for everyones ftp.
A well trained person could eventually do a workout like helvet, and it would be quite doable for them. That doesn’t mean their ftp went up. In this case, where would you set this persons ftp? Some translation that converts this absolute workout to a relative workout at lvl 3.0 with the same absolute performance level? That doesn’t seem the make sense since we know threshold training doesnt make threshold go up after being trained
I’m not sure if this is a bug or a hack….or the third option meaningless (obviously this one).
Anyway, I thought this would be interesting to share about the AI behavior:
I had Lick Rock on the agenda yesterday: It is an easy 1.5 hour sustained power ride with an IF of .67.
At about 1 hour in, my son called me for a pickup because his ski boot binding broke, so I ended the workout at 1 hour and marked ‘easy’ but noted the workout wasn’t finished due to time. After the survey my first thought was, I wonder how the AI will take this info and if any changes will be made. For context I had an over/under workout the next day (cloudripper -2) at a level 4.0.
The AI changed my estimated increase in 14 days from 296 - 304, down to 302w, and changed my over/under workout to an easier 3.4 level version. I thought that was odd that it would change my workout as theoretically, I should be more rested, so why would it downgrade?
As an experiment when I got home, I added an easy ride to the calendar of 30 min. to make up for the time I cut my workout (same day as the shortened workout), and the AI went back to my original 304w estimate in 14 days and put the harder cloudripper back for the next day. OK, kind of makes sense I guess, but then I deleted it, and it stayed the same! So now I’m back to where I thought I should have been and got my theoretical 2w back and my harder level 4 workout.
So now I’m rollin’ down Rodeo…
(Sorry, I couldn’t help myself. I needed to RATM for a second. Carry on.)
Great example of how confusing this can all be. I’m really curious how accurately the FTP prediction tool will play out over the long term as people (including me) continue to manipulate the system day to day to raise it a watt or two, and if people will scream when it’s off by a watt on redemption day.
I would recommend just try the new system for 4 weeks and stop worrying.
If anything would be estimated wrong it will readjust quickly.
I‘m a new TR user started beginning of January with the Beta.
It‘s just brilliant and I already see progress which I haven’t seen in a long time.
The training is not how I would have done it intuitively in the past but I don‘t care. It already moved the needle (lower heartrate at same power) and that’s the only important point.
Stop worrying about the „FTP“ and just let the AI do the work. I absolutely love it!
Ha, totally. I’ve entered the ‘gamification’ mode with the predicted FTP! I know it is silly, but I can’t help myself. My original was 296-308w, but I’ve been slowly losing a watt here or there for seemingly inconsequential reasons, including the example I used - I had to fight back lol!