Forecast AI Beta: How is it working?

@xertedbrain We are all trying out the new Forecast AI Beta now. Before we actually change our training, it would be nice to understand a bit more what is taken into account in the model what is not.

Rest days
What causes Xert to schedule rest days? An example I see from my training schedule is workouts in the following order: Low Intensity, Rest day, Low intensity. I would normally replace the 244 XSS day with two 122 XSS days. In this case the training status was red at midnight, but changed to blue at around 4:00 am.

High intensity vs low intensity days
What causes Xert to choose low intensity days vs high intensity days? I understand you need high intensity days to produce hight and peak XSS – or a focus different from 3 or 24 hours. But why do I end up with some weeks with no high intensity sessions and some with three? I understand/assume that the optimizer doesn’t really care about weeks, only different days.

Rest vs Recovery
What are the differences between those days?


I’m also curious how/why rest days are scheduled. I seem to get 2 or even 3 day rest day blocks.

1 Like


The main difference between a Rest Day and a Recovery Day is that a Recovery Day is needed when your status is red - unable to do low or high intensity due to being too tired. A Rest Day is just a day that you don’t need to train in order to reach your goal. Not everyone wants to train 7 days a week so you have Rest Days when you don’t need the XSS to reach your target goal/event.


Thanks. But is there some logic that cases Xert to prefer one longer (high XSS) workouts combined with a rest day to having the same XSS divided between the two days? I understand there can be a slight difference in training status/readiness because you will do some of the XSS one day later.

At least I prefer to have one workout each day unless I change my availability. Of course if you train less than say 7 hours a week you will probably prefer to have some rest days scheduled.

1 Like

The system looks at both 1 day train - 1 day rest and 2 days train into consideration and will depend on how the training fits into the overall plan and what your individual days looks like (sometimes one day always has a longer ride like a weekend day). Generally speaking though, you can get fitter with bigger/few rides than smaller/more frequent rides. This isn’t something we put into the system. It’s what it will see as more effective.

1 Like


Since you set a target fitness signature (or at least N min focus power) I assume there are multiple solutions (training plans) what will be feasible – or there might not be a feasible plan at all (too big increase in too short time so you will not be able to recover fast enough). Which parameter is used to choose the one you prefer/select? Are you trying to generate a plan that will require the least number of training hours, the least number of XSS, or …?

1 Like

Not sure if this has been asked in a different thread, so excuse me if I’m repeating a question.
I am showing a 266XSS day on Saturday. I am pretty sure I won’t be hitting that on the trainer.
Will the schedule automatically adapt the training or will I just run a deficit into the next session ?

1 Like

As far as I know you will have to manually adapt the plan. I think the Adapt Forecast button is supposed to do that. Not sure if it does, as the tooltip says it will adapt changes in the preferences (and does not mention completed training).

1 Like

There are a few ways to tackle this depending on where you are in your overall plan and how aggressive/inflexible it is (i.e. can you afford to miss any training). If it’s inflexible, you’ll need to change your availability and then click Adapt or do what you can and then click Adapt. Either way will make changes into the future to keep you on track (unless you have 0 flexiblity).

If you have flexibility in your plan and you are still early in it, having to Adapt is optional if you miss training infrequently.


Not sure if it fits here, but have recent changes removed recovery / easier weeks?

When I first generated a forecast I was getting slightly easier weeks every 3 to 5 weeks… after latest reforecast (or regeneration from scratch) I now have a steady increase in hours / TL for 4 months with no easier weeks.

1 Like

XFAI does things a bit differently than what traditional programs would have to (sort of blindly) account for, especially when it comes to overreaching.

XFAI avoids high amounts of overreaching. Overreaching means to have an athlete intentionally train when they are in need of recovery. This is a touchy thing to do and it can result in improved outcomes (see research on Block Periodization) but the challenge being that you need to allot and schedule recovery periods otherwise athletes get overtrained. We want to avoid the possibility of overtraining as much as possible so we avoid overreaching where the recovery demands are unknown.

We haven’t built this level of overreaching into XFAI mostly because it’s really only useful to those with wide-open schedules that are trying to eke out every last watt from their training (often a coached athlete). Most of our users are going to run into some time limitation far before they need to worry about block periodization and recovery weeks. We simply plan recovery all the way along and avoid/minimize having you training high intensity when your status is yellow, for example. Or any training when your status is red. This is why there is no specific need for recovery weeks. You train so that you improve slowly enough that you can recover all the way along.

In terms of seeing “easier” weeks, these were likely due to how we were counting weeks. We used to count weekly hours from Sun-Sat and these could create higher variations than Mon-Sun. I’ll leave that for you to think about why this would make a difference. :smile:

As a note, we are looking to allow users to add in overreaching periods (such as a training camp) into your plan and manage training before and after towards a goal. XFAI sort of does this ok atm but I think there are some improvements we can make.


Is there a way to set a preference to do high intensity on the day before low intensity when I have back to back days e.g. weekends? I assume mathematically it’s somehow ‘better’ to do low intensity Saturday and high intensity Sunday, but I’d rather come into a high intensity day with fresh legs (not just blue status, but also not carrying fatigue from the day before)

1 Like

What is the best way to do this currently? I’ve played a bit and it seems if I enter a block of eg 4 back to back days I’m prescribed less training before (OK, to be fresh) but then quite a few days off afterwards (kind of OK, but it seems that then brings me back to the TL I would have had without the block I.e. there is no ‘benefit’). Ideally an ‘additional’ block should continue to have some benefit, and potentially even lead to a higher target at the end of the plan?

(Also interested in the other question I had re how to plan high intensity the day before low intensity)

1 Like

You can’t have it to give you high/low intensity on a given day. You can plan your own on a given day only atm. Use the planner to plan the ride you want to do.

It you go deep into the red during a block/training camp and don’t want to wait until you’re fresh again before getting prescribed training, you’ll need to manage your own training at the moment. XFAI won’t offer you training when you’re red. That needs to be overridden with your own planned rides. You can try reducing the Recovery Demands but you’ll need to be aware of what that will mean overall.

Once this overreach option becomes available, it will allow you to train up to the training camp block, but XFAI will still wait until you’re fresh again before prescribing any training after it. It may seem like you’ll lose fitness but if you’ve spiked your TLs and RLs, you’ll recover far faster than you’ll lose fitness.

1 Like

I tried the event → road race option and all I manage to get is " A training plan that reaches your target could not be found. Your target is not achievable.".


was excited to try this and was putting in planed events in the planner so that it would take into account those events. it scraps everything i put in and goes on its own every time. frustrating as it seems like a good idea, IF, it will adjust based on what you know you will be doing instead of what it wants you to do. a little more work on the programing side would help, but it seems like i will be on the self coaching side again unless this gets worked out. intelligence it is not, yet.


Forecast your plan.
If you have a weekly schedule you want to follow define Availability for a week.
Add your events, select clock icon on those dates and change to Unavailable.
Recast your plan.
This assumes your planned events are actual activities with strain scores if you want FAI to factor that in.


Hi Jim.

Thanks for taking the time to post.

When you say “scraps everything”, what do you mean? If you plan an activity in the planner, and then create or update a forecast, it includes it in your training towards your target.

1 Like

So far it’s an interesting addition. I’m enjoying, we’ll see how I feel in 3 months. :smiley:

anyway…lame jokes aside I do have two questions currently.

  1. Does forecast use your ‘decay’ model as part of its estimation? For example, if I am on no decay (say a base period) and then forecast, then switch to optimal during the “build” period of the forecast will it need to recalibrate?
  2. Given specific xss targets I noticed when I “autogen” the workout (at least in today’s case) it gives me an “optimal” workout at 24xss less than what was prescribed. (it’s yellow as well).

My question: is this because I’m a bit over my endurance xss target for the week and it’s pulling me to plan? Well rather than leading the witness as they say. Why is it autogenerating a yellow xss workout and calling it optimal?

Thanks, having fun so far! looking forward to seeing how it goes.

1 Like

Hey Eric.

  1. Your signature decay is only applied to activities recorded. XFAI doesn’t use the decay setting.
  2. The autogen workout looks broken. Send that into support so we can have another look at it. (It’s been getting confused on the simple endurance workouts lately). Generally autogen should match your requirement fairly closely in all L/H/P combinations.