Hello,
please help me understand, why the workout level assessment after my todays training is off.
TR_AI gave me Brush, a Threshold 3.9 Training:
I used Zwift, and hit all power targets in ERG mode… now, after the ride TR tells me I achieved 3.4 out of 3.9. So what went wrong? Has “Brush” a wrong WL, or did the assessment recognize something I missed?
Thank you
Your career is already at 3.9 and this workout was a 3.4 So as you have t worked out above career number it’ll stay at 3.4 HTH
My career is actually at 3.6. The workout is supposed to be a 3.9. However executing it “yielded” a 3.4. So given that I hit all power targets, I wonder why it didn’t gave me 3.9 and +0.3 on my career.
Correct it’s a 3.9 they gave you a 3.4.
As for why? Can’t say especially with the zwift data import and not seeing the targets easily for a visual comparison (how some of the model was allegedly trained). But it’s saying based on that new model that it was rated a 3.4 at the end. I don’t see hr blips that would indicate a pause and it should now note any significant pauses which it also isn’t doing so it isn’t that.
Remember if you’re having ai pic the workouts then this is mostly irrelevant at this point it’s going to look at this work hour heart rate the rating and use that to pick future workouts. Just because you did a 3.9 at 3.4 doesn’t mean that it won’t give you a 4.1 next or that a 3.8 is a lesser workout than the 3.9, just different.
Personally at this point I’ve mostly stopped looking at the levels post workout, haven’t looked at them on me career page in over a month. They don’t matter to me anymore how I’m using the system. I’ll glance to see if I made the machine happy or not but that’s about it.
If you’re picking on your own then sure grab a 4.1 next or whatever if you feel that’s appropriate. No problem with that.
Have you recently had an FTP increase? Did the workout in Zwift have you at wattages based on your new FTP?
There might be some variant of WLv2 going on 
I had yesterday Bounty assigned. Because recoveries were long and boring, I increased their intensity to 65% of FTP instead of typical 40%. After completion threshold PL increased to 4.1, not assigned 3.9.
No recent change, dropped from 314 to 311 about 3 weeks ago. Getting a new number next week. Zwift had the exact numbers is prescribed by TR.
Dammit. It was such a good theory!
Thanks for the reply. I can only say no breaks, no back-pedalling, just hitting the numbers during the entire workout.
I’m not particularly worried about not getting the career level. I use WL to get mentally prepared for what’s to come… Did I make it before? How much harder… that stuff.
But I confess, I’m trying to game the AI to receive the highest possible prediction - tweaking my schedule. Yesterday and the day before I just wanted to enjoy first spring weather and went out for some unstructured riding. Before I had Starlight -1 (Threshold 4 4.1) scheduled for today - after the outdoor rides I got Brush for today. I checked WL: 3.9 - works for me.
Apart from that, I also read AI is not caring about WL or FTP at all, just Watt. So my only explanation is that the displayed WL is wrong, it’s actually a 3.4 - assessed by AI I guess. I might have not accepted it as a 3.4 and played around. However, overall prediction is still ok. Been “stable” at around 321 W for 2 weeks now.
I had a similar thing happen to me a few weeks back. I did Cloudripper -2 which is. Threshold 4.0 workout. I did the whole thing in erg mode without any pauses/backpedaling/breaks/etc. I got to the end of the workout and was surprised to see that my Threshold WL only jumped to 3.6
I was wondering if AI is just really particular when it comes to over-unders and it sees the few seconds you might spend over or under the target (due to things like slowing or speeding up cadence or the adjustment when coming into an interval)
If we go by what Nate said on one of the podcasts about having humans review/rate workouts and then feed that into the ML so that it could learn what humans want to see. It apparently was way too sensitive to raw numbers at first so they had to add in some human nuance for it to learn.
Then I’m seeing quite a bit of green above the yellow in places.
Whether that’s a calibration error or power match error or who knows what, that workout doesn’t show the work completed as prescribed.
Basically what I’m saying is if the model is looking at the image like Nate said they trained it on, I can see too why this might be considered less than the 4.0.
Yeah, I think yo-yoing above and below the target is probably what my issue was. Looking ba the target for the 1-minute overs was 380 but my averages for the intervals were anywhere from 374 to 382. So despite being in erg mode you can still end up a little off target on these short intervals. It took me until the last set to get better at hitting the start just right so my average could be closer 380
In the end though I think it didn’t really matter in terms of getting what I need to the workout. I don’t think AI backed off my next threshold due to this one being 0.4 lower. But I wasn’t really paying attention to know for sure
I think one of the bigger things with shorter intervals (higher power as well) is that when it kicks in for a 45second interval, you might be well under the 400w target for the first 5-10seconds, then ERG settles in. But usually you make it up on the backend and pedal higher watts into the recovery. HOWEVER- if you make a conscious effort to ease off the gas when recovery/under interval happens you don’t get those extra seconds at 400w… you quickly dip down to the under recovery power.
So it is best before the hard short interval to slow cadence a little so you can just up cadence to hit the power target, always helps me get to the Over sooner. Also then when the rep is finished, carry on pushing and even up cadence a bit to maintain that high power until you spin out a bit. So AI will hopefully see the same shape it is looking for… just shifted/offset by ~8seconds.
