Forecast AI Beta: How is it working?

Why does the AI Forecaster schedule workouts that do not exist? How would I create them?

For example this Sunday the schedule calls for a 2hr 8 min Pure CG Specialist workout with an XSS of 157. There are Pure GC Specialist workouts with an XSS close to that, but they are around 1 hour.

Yes, it does. Both the Glycolytic system & aerobic system burn carbs. HIE relies more heavily on anaerobic pathways than aerobic.

The durations from XFAI are estimated durations based on the XSS. The existing calculations work well for endurance focus workouts, but not always with shorter/more intense focus types. We’re working on improvements to estimated duration based on the Focus & specificity of the training. Likely to be added soon.

I have a similar situation myself tomorrow - XFAI is calling for 135 XSS Mixed GC Specialist training (estimated for 2h6m), but the top recommended workout (SMART - Bangarang - Extended) accomplishes the desired XSS in just over 90 minutes. So I get my necessary training done and an extra 30 min for my morning coffee :smiley:

You might also consider auto-generating a workout for tomorrow if you don’t like any of the options from the main Xert library. HTH

1 Like

I guess the question is which is the less stressful workout in terms of fatigue.

I basically use Xert now to decide which workouts to do and how much.

I normally run TrainerRoad workouts based on the suggestions in Xert or export workouts from Xert to TrainerRoad. All workouts are run using the TrainerRoad app (or outside).

1 Like

Hello, I don’t really understand the difference between the “event readiness” setting and “program difficulty”. for me if you set more readiness the program will be harder so what is the difference please?

Can’t get the forecast to work for me. I’ve set a target date of 7th July, I run the forecast and get a plan. Then when I try to edit my availability as I can’t do Mondays or Fridays, it just gets stuck and I see the message ‘can’t forecast taper week’. If I only change Mondays it seems to work okay, but as soon as I try and change Fridays it gets stuck. I’d much rather be able to specify which days I am able to train and how long I can train for on those days and it build me the best plan possible (might not be the ‘perfect’ plan, but a good plan would be good enough for me) based on my availability, rather than going through the process of building a forecast only to have to change everything it’s built due to not being available on certain days of the week.

1 Like

That’s probably a bug in the new update. Send that to support please.

Sorry for the trouble.

1st 3 month block + race completed on 3/16.
At the start, it predicted I would get to 270 ftp. I’ve never achieved that in my life but hey, let’s give it a go. Lots of hard days right off the bat. I achieved my all time hi 261 Right at the start of February. But this wasn’t according to the plan schedule. I did Zwift races every Friday through the winter and was getting breakthroughs in the races. So… from early February to race in March, I was pretty much just maintaining although I did try to follow the schedule as much as I could. I did my race at 255 and had a great race. Unknown if I could have done much better.

So the last couple weeks I’ve just done fun rides and. mostly chilled.

No on to my 2nd block with next race scheduled for 6/8. I have another interim smaller race in May. I was able to put the interim race on the calendar and mark the day as unavailable for training. The adapt forecast did indeed remove training for that day without me having to run the whole AI forecast again. That’s nice. My new predicted target for on the training tab is only 256W and I’m sitting at 251W. IDK if that’s a good realistic goal or If I should change something in the settings to try to push harder.

Looking forward to the day when the Race option is ready for FAI.

I would love to have the ability to generate and play around with alternative AI forecasts/plans without messing up my current active AI training plan. It could even be implemented as a way to “mark” the current state and then “restore” it later.