Beta Forecast AI - Super interested

170+ days from an event is likely too far into the future.
Try experimenting with a shorter time frame for an imaginary event.

I think they’ll need to add more pre-forecast guardrails that alert you to unlikely combos before you attempt to forecast a plan. For example, your event date is too far into the future, best to plan an interim goal. Or your work schedule prevents you from attaining the required TL (max hours/week) for the goal you set.
Some checks are working such as unreasonable increase in Focus Power or signature changes with too short of a target date.
With the right schedule and TL mix there is a good chance of raising some values but not others. For example, PP has a tighter limit before you hit your physiological wall.
What other conditions would likely result in an unworkable forecast?
Best to catch them before a forecast attempt.
OTOH Xert needs to know what combos we will be trying to forecast to see where the problems lie in the Beta. Keep on experimenting.

I have noticed an oddity with selecting Terrain for a road race where Hilly gets a warning but Flat or Mountainous does not. Duration does not seem to matter.

I see that the Planner XSS circles are color-coded when you complete a workout close to forecast spec. Got a “task completed” green circle today. :smiley:



Lots of questions will no doubt be answered as soon as some docs are published. :wink:

1 Like

You can run the forecast as far out as you want. Might take a bit longer to run depending on your computer. Interestingly, more than about 300 days out, you don’t need to train. :smile: That’s if you have periodization set to midway. If you slide it to the left (no periodization), you’ll get training presrcribed every week.

Anyone fathom why?

You’ve raised a similar question in the past, and I think I sort of understand. I think I know that training load has a decay of half life, so the work you do now can’t possible directly effect your fitness 300 days from now. If I’ve got that right, then what I still don’t understand is how starting from a TL of 50 now, vs 10, wouldn’t allow you to build up to a higher TL 300 days from now.

Would no periodization mean no variations in focus duration and intensities with a progression limited to ramp rate?

Are all the sliders active in the Beta?

I am testing the Forecast AI Beta now, and it looks like the Forecast items generated do not take any existing planned activities into account. This means that any activities you already have on your plan (e.g. group rides or events) will just come in addition to the plan generated by Forecast AI Beta, and you will end up with some days which are too hard.

Take a look at the two screenshots below, and you will see the effect of adding a 122 XSS activity on December 18. The forecasted load days are similar in both cases (updated after using Adapt Forecast).

Without any planned activities:

Activity added on December 18:

One question: Is the idea to create the forecast once or should one use the Adapt Forecast every day/after completing an activity? The latter will create an adaptive plan based on what you have done so far.

Before talking issues: This is truly impressing stuff. Given it will actually produce the projected results, thats light years away from anything else out there I have seen.

I think I see two bugs (or features I don’t yet understand?)

  1. Right after I calculate a new forecast or update an existing one, the training tab doesn’t show the new forecast diagram, but the “old” xata advicer

Also, in the planner, for future activities, the “set availability” button is not shown.


Only after I reload the xert page, the “set availability” button appears in the planner and the new forecast diagram apprears in the training tab:

  1. High training load doesn’t align between forecasted AI performance target, estimated tagret fitness signature nor AI workout effects. Let me show an example:

Notice the target training load is 46, 3, 0.3

Now the forecast run successfully and I save the result to the planner

Now I check the resulting estimated progression:

As you see, low (46) and peak load (0.3) matches the set target. But not high load (3.0 vs. 1.7)
This problem showed every time I tried so far.

Then I create the AI workout for todays training, which also shows a mis-alignment regarding high training load: Forecast AI prescribes 2.4, AI workout delivers only 1.8

All looks cool and happy to see the ‘teasers’ as I’m not part of testing.

May be better to set up a private channel for sorting bugs during beta testing though?

1 Like

I noticed the recommend activities doesn’t tally the forecast training. I would like to see correct recommend activities forecast training especially for high intensity session. Normally , i would do my high intensity indoor. Do the right one is pretty impt to me.

A more extrem example of high TL misalignement:

Well, in this example the AI generated workout missed all parameters (low, high, peak), not just the high TL.

I believe the target load is what you will end up at the goal date, which I believe you set to June 2024.

1 Like

Yes, but as you see, I would be way to little.

As I see it, given my last example:

  1. The forecast target adjust screen says: For what you tell me you want to achieve, you’d need 8.3 high training load.
  2. Then it generates a training plan and puts it to the calendar, which should lead to exactly that. Just that it doesn’t, as you can see at the goal date, where high TL will only end up being at 1.1
  3. The fitness signature graph within the progression page agrees with the calendar in that, it also estimates just 1.1 high training load at the target date.

I assume something is either going wrong during the plan generation, e.g., it prescribes less high intensity compared to what would be needed. Or, the estimated result is shown incorrectly in the calendar and the progression graph,
(Well, or I got something fundamentally wrong :slight_smile: )

1 Like

Thanks for posting about your experience.

As it turns out, your target power goal’s low/high/peak training loads are not reachable within the given timeframe but the target power is. Hence, you ended up with a different plan but your target power is reached. We’ll need to add a warning for this.

The workout generator appears to have a bug although seeing No Target Event Forecasted might mean the page is stale and the workout generator isn’t using the data from the forecasted activity. Reload the page and see if the problem goes away.

Check your filter.

Workout generator doesn’t handle very short high intensity workouts well at the moment. This is due to having a warmup and cooldown that take up most of the workout’s time.

Perhaps some type of ghost entry could be made when the forecast is overlayed on a calendar with existing entries. I.e. your existing entry is ghosted (ignored) to show what you planned to do that day but the forecast takes precedent.
If you complete something other than to forecast spec you would be alerted to recast.

If you want to plan an activity against the forecast, click the forecast activity title. If you want to plan an additional activity, click the +, choose it, then click Adapt to have Forecast AI include it in the training.

How is the planner supposed to handle if I am not going to follow the daily training suggestion?


The 88xss above are from my endurance ride today.
Means, I am not going to do the high intensity training the forecast layed out (card below with 69xss).
Can I somehow delete that one?
I have the feeling my freshness status currently assumes that I did both workouts (88 + 69 xss respectively), putting me into red.

Also, the training tab shows I did 157xss today (I only did the 88xss)


Bug has been reported.

1 Like

@xertedbrain I’d be interested in participating in closed beta!

Not sure how this is supposed to work, I set a goal a few months out and I also set my daily availability to about an hour each weekday. Yet AI generated a plan that has 1:45 and 2 hour sessions during week. Is this how this is supposed to work.

Also, then went to the table view to delete all those generated sessions and thinking that it was only selecting those shown on the page I hit delete and deleted all the activity in my account… is there an option to select only the page shown as opposed to the entire history?