Check out the link Chad posted here:
If you listen for 5-10 mins, Nate talks about where they are
Our devs are still hard at work on this and I promise it’s still our top priority to ship this out!
14 Likes
It’s a good think they are taking their time. It’s a hard problem to solve when you don’t have a model.
Hopefully they are looking to the future and really leveraging the data they have INDEPENDENT of the workout library.
A good lead data scientist these days is $750k a year and up. Hopefully, TR can use some of the funds freed from Chad’s departure.
1 Like
This. I think it would have been easy for TR to release something that sorta worked. I’d like them to ship only when they got it right.
Meanwhile, they have been releasing meaningful updates: updated plans, including updated polarized plans (e. g. rest weeks on my non-polarized plan are a lot easier than they used to be), they shipped AI FTP and allow users now to opt into what @Nate_Pearson outlined when he first presented AT to the public, that nobody needs to take an FTP test ever anymore (unless they want to).
In Europe they are cheaper, so maybe they should hunt for talent there. Although everybody and their dog wants a data scientist these days, so this is a very competitive market (for companies).
5 Likes
Another mention in today’s podcast:
I personally have mixed emotions with these mentions. I like hearing about the product direction, but we are years past the initial announcement (and officially into “outdoor riding season” yet again) without even limited beta access to a core group of non-TR employees or extremely private users.
I get the general “we want to get it right” aspect that seems to be driving these delays, but it sort of flies in the face of the “minimum viable product” aspect that has also been applied for numerous new TR features. Maybe what we will get eventually will still be that MVP, but as much as I wave the TR flag generally speaking, this delay is into a completely different territory for time & frustration waiting here.
#hopefulbutfrustrated
23 Likes
I mean, they could start that limited release tomorrow for all I know. They have hinted at that more than once this year… but always seem to hit some new roadblock. I suspect they are frustrated with the delays too, but I understand those that feel this is/was some sort of carrot (even though I don’t think that was their intention).
I do think that some or all of this has turned into one of those things that’s far more complicated to nail down than anyone would have expected. Plenty of stuff like that in life, and considering the potential complexity in analyzing a wide cross-section if indoor & outdoor files with a goal of leveraging them for PL’s, it’s understandable that they’d have some difficulty along the way.
Just a shame to get those teasers only to hear nothing at all or another “we got stuck again” statement that shows we will still be waiting.
4 Likes
I think it might almost be time for them to pull the pin on mentioning it and maybe put some devs on other projects.
There’s been plenty of other feature requests/ideas mentioned/stuff that could do with an update (having time limits for workouts when making plans, a modern calendar, better analysis, and a million other ideas).
Maybe outside v2 is just too hard to quantify so maybe put it on the back burner and get some other features mentioned rolled out. Then come back to it with fresh eyes.
6 Likes
Thanks, I listened to that and couldn’t help but walk away thinking “why must everything be reduced to PLs?” and would anyone outside TR understand what was being discussed? Wasn’t fully paying attention but near the end it was something like we looked at an outside workout with 2.something vo2 PL, and something something about impact and ROI from progressing 5.something vo2 PLs. Is this even rooted in underlying physiology and adaptation? It just seems to be all about progressing workout difficulty. Why have I been able to push my fitness higher and higher, while barely progressing workout difficulty? My coach is focused on adaptations, and from listening to the recent Evoq pod with Landry Bobo it sure seems like they focus on that too.
@firemunki back in the late 80s I had a really smart roommate, at 21 years old he had masters in both mechanical and electrical engineering (Cornell and Stanford). Smart guy, semi-retired at 39 yrs old, he took his Silicon Valley #gainz and bought a bunch of land in the Sac metro area and developed a gated community. Anyways, he used to say “if x isn’t working, then 1/x” - in other words, if it becomes clear you are going down the wrong road, turn around and find another road. Its not our place to pass that type of judgement on a system we know nothing about, but if I can get faster following different principles than what TR is doing, maybe what TR is doing only really works well in the domain of indoor training. 
Truly have no clue, just talking out loud. Best wishes they figure it out.
7 Likes
The issue could be more fundamental: what does “PL” mean practically for a ride that is an unstructured mix of everything (see the below example)? How is Threshold PL impacted if prior to a threshold level effort I do a bunch of VO2 Max or Anaerobic efforts, but then have some recovery time? What if the threshold is at the end of a 3 hour ride instead of at the start?
Example outside ride distribution
2 Likes
What does PL really mean anyway? Perhaps a PL only has meaning as one of several tools to progressively overload training. Its not the only tool. What about volume and kJ / capacity to do work? What happens if you use ML to attempt and optimize for those measures of overload? What if relatively small bumps in volume, week after week, over a month or two, sometimes at the expense of structured workouts, delivered $$$ money adaptations? Is the Adaptive Training system so rigid its bound to PLs?
I don’t know. But your questions / points point to a bigger question: is Adaptive Training optimizing for the next workout (a local optimum) or optimizing for performance (can of worms to define that) at a specific point in time?
Since Adaptive Training has no input for a specific future goal, it can’t be optimizing for future performance except to the extent that future performance is correlated to training consistency…
Right now the only lever Adaptive Training has to pull is workout “difficulty”. Not workout type, not work to recovery ratio, not volume
2 Likes
It does I believe, and why it’s recommended you use plan builder which selects phases from TR plans.
Some recent thread on polarized plans was showing differences in volume, between different users, depending on their starting PLs in zones.
- Definitely the latter, with that specific point in time as the end of the phase, but also A-event or final phase in a non-event PB use case.
-
This statement is wrong because AT IS aiming for a “future goal” based precisely on the training phase you are currently within. This can come through Plan Builder (with or without an A event at the end) or individual phases added ad hoc to your calendar.
-
Any of these are the foundation that contain things like a ramp rate, specific training zones and associated workout ranges (PL’s as the guide) to “progress” a rider from an initial state to a higher one.
-
Again, not quite right. With respect to Difficulty, that is the simplified proxy for a more complex selection process. A rider PL along with the phase goals are used to pick workouts aligned with the goal as covered above. We can summarize this all as a “workout picker” to a degree, but there’s more to it than that.
-
As mentioned, and covered by that other discussion, volume is actually a variably and not fixed in all phase or methodologies. From memory, POL & TB sure see duration stretch as PL in Endurance increases. SSB is more fixed, but it’s not right to say volume isn’t a factor across the board.
3 Likes
From what I understand and what I can observe, these factors are definitely included when computing PLs. However, whenever you are reducing a lot of different dimensions into one, you loose data.
I completely agree with you that TR should be more disciplined in their communication or say honestly that “We don’t know when it will be released.” My OCD-riddled brain would like proper scoring of outdoor rides, but I’d likely benefit more from improvements in calendaring and more flexibility when it comes to scheduling of workouts with the help of Plan Builder (time limits on particular days).
Ironically, when @Nate_Pearson presented AT to the public, he envisioned a future where you’d never ever have to do an FTP test again. And it sounded as if that was farther off in the future. On the other hand, scoring of outdoor rides at the time was supposed to be a feature of AT v1.0. Funnily enough, we are in the future where you no longer need to do FTP tests (unless you want to), but we don’t have scoring of outdoor rides yet.
I suspect that the latter is just super hard to do as the impact of unstructured rides can vary a lot, but I reckon is likely very small (i. e. few fitness gains). Whenever you are in a situation where you have lots of noise and a small effect on average, you’ll have trouble to extract anything meaningful.
You mention a lot of the reasons why (I think) scoring outdoor rides is a very hard problem. Like you wrote, even those averages don’t tell the full story. On a mountain bike ride, your power profile might be very spiky.
Yeah, that’s such a good point and likely what I’d do in their situation.
Maybe scoring outdoor rides isn’t actually needed, but you might want to quantify fatigue incurred by an outdoor ride to adapt next week’s training. E. g. I would like AT to adapt my workouts next week taking into account that I did this long mountain bike ride in the Austrian alps last weekend.
5 Likes
I’ve worked as a software engineer in companies of various sizes for more than 25 years, and one of the canonical struggles is sales and marketing selling features that aren’t done. When everything comes out on the planned schedule, great; but in software, the name of the game is unforeseen challenges popping up during all phases. I imagine throwing ML in the mix makes it even worse. Obviously, TR has chosen this path very intentionally, but continuing to push marketing for an inaccessible feature seems risky at best.
4 Likes
Agreed, but I also think there is a bigger issue. Whether they release ‘V2’ today, or work on it another year, in 3 years time (or sooner), they will need to release ‘V3’. I don’t think that this is the type of thing (ie modeling human physiology in new ways at scale) that will be ‘done’ anytime soon, as after they have 3 years of data and feedback on V2, they will certainly identify areas that need improvement. I hope they move to a more incremental rollout of AT features/improvements after this big drop.
3 Likes
V1 wasn’t modeling human physiology “at scale” and there are existing models they could easily pick up and use. Heck, have you seen the really well done sustained interval and climbing detection in the free as in price and speech GoldenCheetah? The extended CP model in GoldenCheetah also matches up nicely with WKO5. There really isn’t a need to invent a new model
5 Likes
Except for many TR athletes, there is not much time spent inside during the season, so for AT to really function properly, it needs to take outdoor rides into account.
Let’s say I basically only ride out doors from Apr - Oct, but occasionally ride inside due to weather, family, whatever. Any adaptations AI would recommend would be based on a mess of seemingly missed workouts (maybe some are tagged as outside workouts, maybe not) and very few actual workouts. So AI is going to recommend a sub-optimal workout.
Now, I would also probably argue that it is really irrelevant in the scope of things and PL’s really don’t matter. As noted above, they are somewhat made-up metrics within the TR ecosphere.
6 Likes
I understand that. This is why I wrote that they should maybe start with computing fatigue incurred by outdoor rides and adjust training accordingly. It may not give you as much information about what improvements you have made when riding outdoors, but it may still improve your training compared to what we have now.
PS I fully understand that having PLs is the superior solution, but it might be out of reach for now. Deducing fatigue levels from unstructured outdoor rides might still substantially improve your experience with TR, and might be easier to implement.
9 Likes
I might miss something here, but all this effort and resources dedicated to essentially extending the current PLs outdoor?
Is this PL really useful anyway?
I would understand the effort if it actually changed the plan’s structure, I.e moving hard days if the user says he’s too fatigued or it’d actually create tailored plans based on someone’s time available.
At the current state of art though it merely suggests a different workout.
I mean, it might not even be the optimal workout anyway.
Say one starts with a PL of 0 or thereabouts (e.g starting vo2 after endurance block), the initial progression might still be too easy.
I really struggle to see people’s expectations with this PL2.
Most likely many more will have an endurance pl of 10 as people did not sit 4 hours on a turbo and long outdoor rides will count, but then coming indoors again what would the suggested workout be?
Also forgot to add, the issue really is with kitchen sink workouts.
If one follows a tr workout, he/she could already match the outdoor workout with one existing in the library.
3 Likes