New TR feature - FTP estimate - thoughts

This is how it works right now. But if you add it that day (like you add a ramp test to check), that’s when it will take a bit of time.

This same piece of ML is used to predict FTP based on what’s on your calendar. Then it should be able to optimize training plans based on different scenarios for individuals. At least that’s what the vision is.

We’re just detecting what the FTP is today vs predicting what it is in the future.

So this is one “UI” stop of adding value in the roadmap to some crazy stuff.


And you can come in with some fatigue and not worry.

Like I never like to do a hard workout(s) the weekend before a ramp test.

Plus we’ll be able to detect when an FTP could change in the middle of a block. That’s not much of an issue for all the forum users who are experienced athletes but for a new rider that can be killer. They can get some huge off the couch improvements in the first 6 weeks of a plan.


Exactly. And you get no training interruption.


This is another good point. I would like to make it where we don’t schedule FTP tests on your calendar and just detect when it needs to be raised and let the athlete know.

This would be similar to a coach watching your workouts. They keep making them harder until you get to a point where you need to raise the wattage on all of them to keep progressing.

The rate of that increase depends soooo much on training history. Consider the following scenarios.

  1. Off the couch athlete who has never done a workout
  2. Athlete with 5.0 w/kg with 20 years of consistent training
  3. Athlete who peaked at 5.0 w/kg when they were 30 but hasn’t done any training in 2 years.
  4. Athlete in their first year of training and at 3.0 w/kg

It’s hard to have a best testing frequency for all of those athletes. And It’s hard to say that they will all execute a test with the same precision or have the desire to go all out in a test.


A post was merged into an existing topic: Feature request/idea: Add intervals during workout / switching to +1 versions

Well I’m not a fan of the ramp test - one coming up Tuesday. It will leave me a gasping heaving mess running outside in kit in -15C bc I can’t suck in enough wind. And I’ll likely be disappointed in the result

However, I like the idea over time that I will get some kind of idea of progress. Maybe that’s optimistic. I’d like to see that the work = fitness progress. My objective here isn’t to race but to be a stronger more competent road cyclist in my terrain which include some intermediate length/grade climbs. I’m spending the time and the money and effort - I would like an idea of progress. I’m not at a point yet where I am content puttering or just riding my bike. Hopefully not for a while.

Yes I understand that “the number” is subject to so many variables and I’m trying not to work to a performance result. But again if the result helps to formulate a training program that helps me improve, not nuke, I’m in

So yes I will probably test ongoing. What will be interesting is if estimated FTP is even close to test results - and if not, then more importantly why not?

Until it comes out, then it’s bad.


Speaking about FTP prediction:
Is this tied to TR plans or more flexible?

I mean what if you looked at someone’s past training phases…would you see that someone might have had better results on individual trainings vs TR plans.

Would FTP prediction consider that and prescribe/adjust training outside the bounds of a TR plan framework?

Or maybe the better question to ask is:
Could FTP prediction go back at my past training and then tell me what it thinks worked best for me (by projecting forward from different points in the past)?

The amount of people who will get disappointed that their rides just aren’t that much work for progression levels is going to be somewhat hilarious. No, doing a couple of 3min efforts isn’t going to get your vo2 level high lol


It’s also just not FTP. It’s FTP + progression levels after the fact. So for some people we might start them at 6 vo2 and 4 threshold post ramp.

Someone else we might do 3 Vo2 and 3 Threshold.

The important part is to get you to the right workouts after the fact.


It’s tied to whatever is on your calendar.

We haven’t made it yet so it runs different scenarios and finds out the best. I’d like to do it but it’s still a ways off. Not as easy as it is to type it.


I don’t know about disappointed, I’d settle for them just being included. I’m not expecting gains, except in endurance and tempo.

Racing maybe different - I’m hoping it’s live before that, even for the FTP estimation given I can definitely go deeper in a race scenario (outside or on zwift!).

If a cyclist has been training long enough and has these easy to predict and proably small FTP changes the AI should be able to much easier predict their FTP. The riders who can’t really predict their FTP well are also the ones who the AI will be less acurate for in predicting their FTP and will need to test. Also proably the people who need to test more often as their FTP is changing more (just started training, recovery from tie off, etc)

I’m wondering if there are good measurements that correlate with how hard someone is going. As in since the computer can’t tell directly how hard you are pushing what data correlates with how hard you are pushing. HRV measurements like Alpha 1? Using HRV to compute respiration rate and calculate VT1/VT2? Does pedal smoothness and torque effectiveness change the harder you go vs your threshold? Use muscle oxygen at slope of zero Evaluating the Relationships between Muscle Oxygenation, Ventilation, and Lactate Thresholds – Spare Cycles
(ok, TR has to add support for a new sensor) or when muscle oxygen drops low enough Predicting Cycling Time To Exhaustion from Paraspinals Muscle Oxygenation – Spare Cycles

1 Like

Its not my fault my asthma makes me end the ramp early cause I’m not getting enough o2…


When does the FTP estimate feature go live?

1 Like

They have til the end of today because my ramp test is tomorrow , thanks :joy:


I have not read much of this thread and I’ve very new to AT, but my general sentiment is that TR is really just formalizing something (automatic FTP estimates) that is already going to be happening naturally behind the scenes as a byproduct of AT.

The old FTP number was really just a basis to put some structure around interval intensity levels for different power zones. It’s served that purpose reasonably well since it hit mainstream over a decade ago. But using a single number with fixed percentages only works for folks that fall right in the bell curve. Even with those people, it’s still flawed because the percentages don’t account for how trained you are in a specific zones at different stages of your training.

It seems that AT really addresses those issue by tracking progression levels within each energy system. It looks like they plan to keep FTP around (probably because it’s such a ubiquitous thing and they already have a system structured around it), but you could argue that it’s no longer relevant and the progression levels are just a mechanism to make the FTP reference point correct by giving you workouts that happen at a lower/higher percentage within the zone. It’s kind of like they are setting a different FTP “base” for each kind of workout, but keeping the visible number the same. Again, it’s been around for a while, so I get it, but I’m not sure centralizing around FTP would make sense if starting from scratch today.

Anyways, I don’t really care if they keep FTP as a key metric or not. You could hide it everywhere in TR tomorrow and the workouts would still ask for the same number of watts for the same amount of time and AT should be able to adjust those up/down as you progress. At least that’s my understanding of where it’s going.

For what it’s worth, I don’t do many ramp tests. Most of my adjustments are manual to make workouts productive. I actually think the ramp test is a great objective fitness test combining multiple energy systems, it’s just not optimal for setting all zones and it certainly isn’t the best test to determine real functional threshold power. I’ll take maybe 2-3 ramp tests per year as a general benchmark. I do other benchmark tests also. I think everyone likes to do tests that objectively measures past vs. current fitness to gauge progress, so that makes these tests valuable and helps validate training. The output of the ramp test just isn’t that valuable to setting training zones (in my opinion). Maybe for setting v02max training since the ramp test is really a MAP test, but I think AT should be able to set the other zones better than a ramp test.


We are working on a few issues that I added to reduce confusion. We did want it to go live today but I slowed it down a bit.

The big ones are:

  • Trigger FTP prediction in the background when you add a ramp test to your calendar
  • Give some type of feedback that AI FTP Detection is happening (It can take 20-30 seconds)
  • Do a load test on the server to see what happens if we have a LOT of estimations run through at once. Can it scale without crashing or slowing down a bunch?

Your thinking is exactly inline with our thinking.

Noice :sunglasses:.