AI FTP prediction working well for me (outdoor rides notwithstanding) but it’s human to be frustrated when a predicted FTP number is initially suggested and as you get closer the number shifts down. Im interpreting it as a median value shifting down as the CIs tighten and an overly optimistic initial range is revised. Would it feel better to instead predict the range then let us see that range slowly tighten overtime as it zeros in on the final value ? Less fixation on a singular number then. Additionally , showing some secondary analysis post workout explaining changing range would help interpretability (- heart rate drift or + perceived easier etc)
I’d definitely agree with this - a great parallel is the way MacroFactor shows calorie expenditure. The shaded band gets tighter the more consistent your logging and the better the model is working:
Obviously the presentation wouldn’t work for FTP but having some kind of confidence interval would help with a lot of the frustration!
Right now the prediction is best case scenario and we all know how often that works out. A range has been suggested several times, but so far 
The prediction is optimistic, but not always the best case (during 2 of my earlier blocks, I had the prediction drift upwards by ~5W).
An range (or even better, a CDF) would be welcome.
I usually don’t agree with the feature requests here but this one I support 100%. Any credible prediction model should include an error or uncertainty measure.
I like it. I just posted in another thread about revising PLs as well.
When this new AIFTP was released, there was a lot of talk from the TR team about not showing “FTP” at all, and focusing on PLs and then potentially showing estimates for 20m and 60m power. I think TR should lean into some version of PLs, but base them on absolute power/duration so they can be compared across history and athletes.
Hey @eoinos 
I will take this feedback to the team! Thank you!