Could you please make it easier to compare the ride’s power curve with the workout’s power curve.
Great tip. I did it tonight and it stayed at the predicted number after the workout rating. Only a sample size of one, but so far so good. Thank you.
I was doing the method of “plan inside, then swap to outside the day of the session” but was still seeing a little bit of fluctuation. Again, unsure if that’s the bug or not.
As an experiment, I decided to manually create a planned TR workout using Intervals.icu and do that outside instead; basically circumventing the bug. Afterwards, on TR I matched the ride to the workout, and to my surprise the FTP prediction actually went up. I don’t know what this actually means; it could be the variable of the commute to and from where I performed the workout. Though prior to doing this I had never seen the prediction go up.
Honestly, doing this was a pain in the ass, but to me, the friction of recreating the workout on a different platform was less painful than all the uncertainty of the bug. Specifically the unknown of whether or not the algorithm is assigning workouts of inappropriate difficulty… and then having to manually selecting a harder one after that.
Wondering if any of you are reverting to traditional FTP testing.
My account should be public, so please feel free to take a look.
Based on my power curves, I deem the AI FTP pretty accurate; however, it has been stagnating for a while. Was it a case of being too high in the last detection? Maybe yes.
A new prediction in 10 days gives me a +5%, it will likely fall rapidly, as I do most of my workouts outside and am affected by the bug.
Was wondering what was the best testing protocol, if any of you is manually testing his data.
For Outside Workouts? I’m pushing for that! 
I did a 20min test a few weeks ago that confirmed my AIFTP was WAY off from what I traditionally considered FTP. I’ve since quit using the TR plan, but left AIFTP on just to watch it and see what happens over the next month.
I’ve checked in on ai detection with a ramp test for years, last time was January and it’s always bang on or within 1W. I have no desire to do that in a heat wave 
I often wonder how accurate and how persistent people think FTP should be.
I’ve been doing a traditional 20min outdoor tests every 8 weeks or so along side my TR training. On the same closed veloway I do all my outdoor workouts on. My tests always reflected higher numbers than TR which I was fine with since both metrics were trending upwards. With all the caveats that come with indoor/outdoor tests I didn’t give it too much thought. Since I don’t own an indoor trainer the outdoor tests are mostly used as a benchmark.
What testing protocol are you using? I am well aware that there isn’t one “best” protocol; it’s just a matter of keeping up with the same one for consistency.
Maybe I should take a Critical Power test next week. AI FTP is recommending right difficulty workouts indoors, slightly on the easier side outdoors. However, I don’t see myself matching my AIFTP in a true test anytime soon.
For me, I prefer the Kolie Moore test, but I did the TR 20 minute test just to see how it compared to my AIFTP. It was off so badly that I couldn’t even come close to holding it for the 20 min segment, not to mention something like 40-60 minutes.
I do agree with you that it’s not productive to worry about a couple of watts, and I know TR’s AIFTP isn’t meant to be “hour power, or whatever” but it was so far off from my Intervals.icu and Garmin estimated FTPs that I felt the need to know. Lots of folks say it’s not similar, but they’re being served appropriate workouts and estimated RPE so it doesn’t really matter to them, but this was not the case for me. I’m glad I did the test. I’m back to feeling positive about my training for the first time in months, even after the ego hit.
In just not sure how many I think is too many, or how much it varies day to day or intra-day.
I’m having shocking sleep in this heat can’t ride at all - have I lost ten or twenty watts until the situation sorts itself out?
In normal conditions is it +- 1% from morning to night or 5%?
Those sort of questions - I have no idea what the answer is really.
Actual FTP varies very little day by day, one’s ability to express it does however.
It’s why (imo) longer tests and a well calibrated RPE (from those longer tests) is so helpful, it lets you dial in the best effort on the day for your training.
I switched my planned inside ride to outside last night, did the ride and my prediction also went up
I am.
More details: I am just finishing a 2 week taper for my A event. The taper was from a LLM recommendation based on past 8 weeks of training. AI FTP (detected, not predicted) was +1% from the 4 week prior. Negligible but at least it’s positive.
After this race, I’m resuming TR circa 2020: FTP test every 4 weeks, and doing EXACTLY the same workouts and outdoor rides, rest days, for 3 months that lead to my last breakthrough. It’s plenty of 800+ TSS weeks, tons of days with an indoor interval session followed by a Z2 outdoor ride. Lots of long rides with a couple of PR setting efforts on climbs. No surprise, TRs AI is blowing a gasket 
, “not recommended” for a lot of rides, plenty of yellow and red. It’ll be an interesting experiment. Just because it worked before doesn’t mean it will again for hundreds of reasons. Only one way to find out!
Lost a point on tonight’s last minute switch which is weird bc it was a 30 minute easy zone 2 ride which felt easy. I’m wondering if we need a “very easy” rating option.
What is a reasonable FTP % increase? This 28-day block I was predicted for a 13% increase. I do my endurance and recovery rides outside, and after every one, my predicted FTP dropped. I’m now scheduled for a 4% increase.
What is actually a reasonable increase? I’m not sure if I should manually adjust my FTP up to where it was predicted to go, or somewhere in between the predicted and actual? Or is 4% a “reasonable” gain in a 28-day spell?
(TR profile is Evntprincess if any of the support techs want to take a look)
4% during a build phase 4wk block is reasonable. 13% maybe if you’re fresh off the couch and have never done any training but not if you’ve been at it for years. Why not do whatever testing you used to do (20’, ramp etc) before AI to see where you’re at.
How are your workouts feeling? That’s what’s most important.
If your workouts are going well, then things are probably in a good place.
If all your workouts are too easy, then you could probably bump your FTP up a bit. If they’re harder than they should be, then maybe you could bump it down.
I’d be happy to look a bit closer if you have other questions. Just say the word!
My FTP set in TR is 277w.
I did a vo2 workout on Tuesday which although it was more than recommended, it even said Not recommended, but I still did it, RPE Hard instead of Very Hard.
AI Prediction next week I think it shows 266-267w…
Wednesday (yesterday) I couldn’t go out because of the weather for a training of about 2h of z2.
Today AI prediction 270w. Ok something has changed for the better, but anyway irrelevant.
I do the recommended sweet spot training, the values on intervals were 7-10w more than what was recommended, but without forcing it. I answer the questionnaire with Moderate, because it was actually not more. Surprise AI Prediction 263w!!!
Funny how I have to improvise to have a training plan as close as possible to what I need!! But I’m slowly losing patience, especially since I had my hopes up that after 3 years I would return to TR, I think it’s already been 3 months since I returned and I’m still having problems.
This is the sort of volatility we are aiming to fix.
We talked about how volatile “FTP” likely is in real life on today’s podcast, but we also know that it can be disconcerting to see it fluctuate so commonly, and after something seemingly benign.
We have some solutions we are working on for it, but in the meantime, my personal approach with my own training is to not sweat these small/frequent fluctuations.
Again, we’re eager to get a solution in place that will take care of this for ya.
9 Likes