FTP prediction dropped from from +11 to no change the day after my last workout of the block

[jbeld]

Quite confused, I followed all of the workouts bar one with a stomach bug and added a couple easy endurance rides. Every workout I either met or exceeded the power targets and rpe stayed steady to where the prediction said it should be, yet after every workout the prediction dropped. Starting at 254 and steadily dropping to 248, honestly not a big deal.

My last endurance ride of the block, Sunday my wahoo died 2 minutes in so I completed the ride without data then marked it as complete without power data, ftp prediction did not change, stayed predicted at 248. The next day, Monday, after changing nothing on the calendar and no additional workouts completed or planned my prediction dropped to no change and down to 237. I thought it was a bug with that Sunday workout so I deleted it off the calendar and nothing changed but I cannot add it back in the past. For reference it was James, 90 minutes z2

Obviously this is frustrating, my most recent 2x20 SS workout I didn’t really pay attention to power and rode off rpe and did 244 for the second 20 minute interval marked it moderate yet detection says my ftp is still 237. My next block has VO2 intervals at 249w which seems far too low for VO2 work. Prediction for the next block is 264 from 237 which seems unrealistic.

Consider me confused, anyone else have similar issues? Nearly every workout was completed outdoors.

[chrism99]

I had exactly this with James at the weekend. End of block. Dropped from 271 to 262 overnight. I rode outside same IF, average power, np for James. I raised a ticket as it just looks weird. Hoping ftp detection is accurate tomorrow. If I delete the ride from the calendar no difference. If I unmatch it to James then no difference. In the outside ride I set 1 hr power record matching requested.

[Eushev24]

There is known bug that inflates outdoor riders predicted ftp, but actual detection should Be accurate.

[jbeld]

Actual detection gave me no change from the start of the block to the end

[nene]

My experience is the same since this FTP prediction feature was introduced. It always starts with some great prediction and then day after day it just keeps dropping until it reaches something very close to my previous FTP.

Like just a few days ago I remember it predicting something like 1.5% increase. And today it’s predicting 0.7% increase. And I only did a super easy recovery ride in-between.

What I do know is that I’m feeling much stronger. But I don’t really expect to see any massive FTP gains in such a short time anyway as TrainerRoad wants to predict. I don’t really want to care about this FTP number, but seeing it drop every day is sort of discouraging. I wish there was a way to just disable this feature.

[Pbase]

It’s easy to turn off. Go to your calendar and find your next detection. Click on it. Click on the “edit” pencil. Turn off.

[nene]

Great to learn that.

BTW, in the mean time I did one endurance ride (more than was prescribed to me, but really nothing crazy - yesterday I did less, today I did a bit more, so really it shouldn’t make like a massive difference is grand scheme), and the result of it was that my FTP prediction turned from 0.7% to -1%. And it’s predicting my FTP for tomorrow.

I turned off this feature.

[eddie]

I’d be happy to dig in here and let you know what I find. Can I share that stuff here?

[Chain_of_Fools]

I hate to say it but I’ve come to really dislike the AI FTP feature. It seems mostly like a random number generator. Mine dropped 7 watts just before its “detection” date, apparently because I did an outdoor ride instead of a planned workout. Sorry, but there no single workout that could cause a 7 watt change in FTP in either direction. I also did an actual TT effort that, when plugged into a standard 20 min FTP test protocol should’ve bumped my FTP by 15 watts or so. Nope, AI FTP doesn’t care.

I think the best option is to either ignore or turn off AI FTP. At this point it seems super unreliable and quasi-random, and I really don’t see what useful information it provides.

[jbeld]

Please do

[agk8624]

@Chain_of_Fools

Mine dropped 7 watts just before its “detection” date, apparently because I did an outdoor ride instead of a planned workout. Sorry, but there no single workout that could cause a 7 watt change in FTP in either direction. I also did an actual TT effort that, when plugged into a standard 20 min FTP test protocol should’ve bumped my FTP by 15 watts or so.

Saying you did an outdoor ride isn’t specific. Was the planned ride and endurance ride, and instead you did a group ride which was basically a race? There’s a wide range of rides one can do outdoors that can negatively impact a training plan.

A TT effort at what distance? A 60 minute effort? That’s a lot of TSS, and while it’s good validation for an FTP value, that doesn’t mean it won’t impact your future training. And just because a more traditional FTP test would give you a higher FTP value, that doesn’t mean that higher FTP value will give you good workouts in the future.

Finally, there’s a known bug where planned outdoor rides inflate predicted FTP by a noticeable amount as soon as the ride is completed, the predicted FTP value drops to a “true” predicted FTP value.

[eddie]

I’ll definitely echo a bit of what was mentioned in your open support ticket, which is essentially what @Eushev24 touched on a couple of days ago:

You’re primarily being affected by a known bug with how the model handles outside workouts, where it’s very common to see the prediction drop a few watts after each one. It can also sometimes drop more dramatically as you get closer to the detection date, like you saw with that 11-watt drop.

I did notice that you had two 2x20 SS workouts within a week of each other that looked like they felt pretty different.

The first was on March 27th, where you did 230 watts at 164 bpm, and 233 watts at 168 bpm. You rated this workout as “Hard.”

Then on April 2nd, you did 230 watts at 163 bpm and 244 watts at 172 bpm and rated that workout as “Moderate.”

Are you rating the workouts based on how the work felt? I’m mostly just curious, but if you’re rating them in a different way, that could have an effect on your predictions, detections, and overall training.

I would have assumed, based on the power and HR data, that the second ride would have felt harder, but maybe there were other factors that made that effort easier than the first one..

It seems like you’re getting through all your workouts pretty well, though, so that’s a good sign..

Let me know your thoughts!

[eddie]

I could look at your historical prediction data to see if there’s anything strange (aside from the outside workout bug mentioned above) that’s causing issues for you.

Let me know if that would help and, if so, if I should share that data here.

[Chain_of_Fools]

Hey Eddie, I have no problem with you digging into my data and sharing it here. That said, I’m now in a part of the season where I’m racing or traveling nearly every weekend (and sometimes doing a crit mid-week), so I’m usually either recovering or tapering and am not really able to follow a plan effectively. My observations on AI FTP are based on my experience as a beta tester and earlier in the year when I was adhering to a plan more closely.

And, I’ll add that I still find TR effective, and the AI FTP number is good enough to get me workout intensities that work for me. Realistically +/- 10 watts is probably within the margin of error of my trainer’s resistance anyway.

To me, the issue here is one of UI and emphasis: TR users are hyper-focused on the FTP or predicted FTP number, and the interface puts it front-and-center on the landing page. That number then jumps around by up to dozens of watts for reasons that are hard to fathom, and surely can’t reflect actual fitness (e.g. changing one workout leading to a 5-10 watt drop in predicted FTP.) Given this it’s perhaps no surprise that a majority of forum posts now seem to be some version of “my AI FTP number is too high/low/changed and I’m confused about it.” Maybe this is a form of engagement that improves plan adherence, but I ultimately find predicted AI FTP to be a non-useful metric, so I’ve turned off the prediction.

TL;DR: You’re welcome to dig into my data, but that may be a waste of your time; I think AI FTP prediction is working “as designed,” I just don’t think its an accurate or useful piece of information for me.

[edgarminault]

TBH I have the same issue with ever.changing ai ftp values.

• Up until 22nd of March i was super cautious about following their plans; ai ftp prediction would be way up to my actual figures up until the last day prior the new ai estimate; even though i was following all trainings of my polarized block (vo2 + threshold to the T but happened that i split endurance rides into 2 because time crunched - always made sure i had very similar load & tss in endurance related to those split-into-2 workouts).
• Now that good weather has returned, i stopped being super cautious about training; Doing threshold intervals + endurance but way-less vo2 (just a few sprints here & there once a week);

The AI ftp is doing even worse in second use case.

What i recommend is for you to do once every 2 weeks an all-out effort between 10’ to 30’ and check the ftp detected by intervals.icu. Much more precise + does an extrapolation of similar riders.

What i want to say to TR team:

• I do like the incremental side of trainerroad workouts aiming at pushing your np on specific time intervals overtime. I’ve seen significant increases.
• This ai ftp will not make me churn from the app but i do hope you get your act together to incorporate not-workout-related all-out efforts. Currently ai ftp is confusing at best, useless at worst

Love the product + polarized plans but dislike this useless feature (in its current form)

[eddie]

If you decide to turn AI FTP Predictions back on, let me know. I’d like to see how it’s working for you.

To me, the issue here is one of UI and emphasis: TR users are hyper-focused on the FTP or predicted FTP number, and the interface puts it front-and-center on the landing page. That number then jumps around by up to dozens of watts for reasons that are hard to fathom, and surely can’t reflect actual fitness (e.g. changing one workout leading to a 5-10 watt drop in predicted FTP.)

This is a big thing for us right now. There is a ton of noise on the forum, and it’s been hard to track each individual’s historical situation with predictions, but I think that it’s really easy for cases to blend together and create confusion.

Most often when athletes report multi-watt changes in predictions after a single workout, they come to the conclusion that we’ve changed their fitness based on that one singular performance, when in reality, it’s almost always (probably something like 99% of the time) from us adapting their upcoming workouts in a similar, though maybe less obvious way that we used to.

If an athlete is three weeks out from their detection and we change all of their sweet spot workouts between now and then, it will likely have an impact on their fitness by that detection date.

Again, there are many different scenarios that could, and do play out, and we’re looking into as many as we can. I know that this has created a lot of confusion with our athletes, and having so many cases shared on the forum definitely makes it harder for us and our athletes to read through the noise.

If the prediction isn’t useful for you, that’s totally understandable (I don’t really look at it myself), but we’re hoping that this is something our athletes can use, enjoy, and benefit from in the future more than they are right now.

[jbeld]

That makes sense, the workout marked moderate was initially marked hard and I changed it to see what it would do with the prediction out of curiosity and forgot to change it back…oops

The workouts seem to be great, however SS and VO2 the power targets seem low. Could be that I am suffering from a case of too big an ego, but would I be best suited riding to the target directly or aiming more for RPE and using the target power more as a guideline and trying to be “close enough”

I do really like the plan and it motivates me to stay the course, any idea when the outdoor ride bug might be fixed. Obviously I know nothing about how that would be fixed or how much work goes into something like that, just curious if there is a tentative timeline?

[andho]

The forum is full of nearly identical posts. With nearly identical answers.

The frustrating part is that TR sells the message “if you follow our plan you will get faster”

Then AI FTP prediction takes that to the next level: “Here is your plan. Follow it exactly and here is how much faster you will be at its end.”

So you follow the plan and the prediction gradually drops to (or below) where you started. That would be fine if it gave feedback regarding what you did wrong (whether or not that was under your control). This is frustrating.

So turn it off -or-

For those that want to see the number stay high, just rate everything ‘easy’. The prediction won’t be accurate but you will eventually reach the VO2 of a hummingbird

[eddie]

Here’s another one of the same answers.

So you follow the plan and the prediction gradually drops to (or below) where you started. That would be fine if it gave feedback regarding what you did wrong (whether or not that was under your control). This is frustrating.

As far as we know, this only happens with Outside Workouts.

Follow the plan with inside workouts, and you should reach your prediction.

That Outside Workout bug is our #1 priority right now. @jbeld, I have no idea on a timeline, but I just left a meeting where this was being discussed, and there is another one going on right now. We’re doing all we can!

[andho]

Reach the prediction if you rate as expected or rate below expected?