Outdoor AI FTP Detection model update timeline? My FTP predictions are swinging despite successful training

Training
1

Hi TR team,

I wanted to ask if there is any update on the outdoor-optimized AI FTP Detection model that Nate mentioned in Ask a Cycling Coach Episode 590 around the 5–10 minute mark.

In that episode, Nate discussed some of the issues athletes have been seeing with outdoor rides affecting AI FTP Detection and mentioned that a new FTP model is planned to be released into Early Access. I have been checking the Early Access settings regularly since then, but I have not seen anything appear yet.

I completely understand that ML model updates take time, and I have been very sympathetic to this rollout. I have generally been very happy with TrainerRoad and AI FTP, and I understand that a new system will have edge cases. However, my current training experience feels like I am getting misleading signals from the model.

Some context:

My FTP detections this year have been:

  • 261 W

  • 295 W

  • 307 W

  • 296 W

The frustrating part was the most recent decrease. I finished my previous training block feeling very strong, and the AI FTP Prediction was showing an improvement. During my deload week, I got sick, missed some recovery workouts, and when I returned to training I received an 11 W FTP decrease.

I understand that fitness can change, and I do not expect FTP to only go up. However, missing a few recovery rides during a deload did not seem consistent with an 11 W decrease, especially given that the previous block felt successful.

I have also noticed a pattern where AI FTP Prediction starts a block with a very optimistic prediction (~340–350 W), then gradually falls throughout the block. Usually, I can complete the prescribed workouts and often feel like they were successful, but the prediction still drops afterward.

The problem is that these signals influence training decisions and confidence. Seeing the prediction repeatedly decrease despite completing workouts successfully makes it feel like the system is telling me my training is going worse than my actual experience suggests.

Most of my training is outdoors, so I suspect I am experiencing exactly the type of issue Nate discussed in the podcast. I do my best to execute workouts outside, but outdoor rides obviously do not have the same perfect execution as ERG mode.

My questions:

  1. Is there an estimated timeline for the outdoor-optimized AI FTP model?

  2. Will this be available through Early Access before a wider release?

  3. Is there any way athletes experiencing this issue can participate in testing before public release?

At this point, I would be happy to try an experimental model. My current experience already feels like the model is giving me unreliable feedback, so I do not feel I have much to lose by helping test an improved version.

Thanks for all the work the team is putting into this. I appreciate that this is a difficult ML problem, and I am excited to see where TrainerRoad AI goes. I would just love to get back to trusting the feedback from the system.