Forecast AI and AI Generated Workout discrepancy

The AI Forecast had planned a high intensity 121 XSS workout today.

I generated and AI workout but after completing it showed a yellow circle instead of red.

The generated AI XSS was less than what the AI forecast called for.


The strange thing seems to be, the text description of the workout doesn’t match what is actually in it.
The description calls for 3 sets of 7 sprints at 535 watts.


In actuality, only the first set was 535 watts. The second set was 515 watts and the third set was 490 watts.

Hard to tell how accurately you executed the workout form what you posted. If you recorded with Xert you should be able to overlay the targets in the chart

Also, depending on how the workout was defined, it may have been a smart workout which dynamically adapted to your actual efforts… if your early reps were too hard, or recovery was at too high a wattage, it may adapt future rep intensity down… would need to check the workout details to confirm

It is normal for an Autogen HIT workout (watt target based) to state “3 sets of X at Y watts” but the sets will taper.
Perhaps the text should say “3 sets of X starting at Y watts and tapering as difficulty rises”.

Enable Target Power on your chart to gauge how close you were on interval targets across all 3 sets.
What trainer?

1 Like

Post the xss values of the workout created if you still have it.

Thanks for responses. I use a Elite Direto X along with a Garmin Vector 3S.
I completed the workout targets 100%.

I was still able to view the AI generated workout, it’s a little small but that’s exaclty how the workout was built, with decreased targets as it progressed. I guess the “starting with” was doing some heavy lifting in the description.

So the real question is why did the AI workout generator make a workout that didn’t meet the demands of the AI Forecast?

We could do this. Originally autogenerated workouts would generate multiple sets at the same intensity. However, we noticed that tapering our workouts towards the end (like the standard library) increases compliance to the workouts. We could update the text to reflect that.

For OP, I’m guessing it’s just a (unavoidable) difference/lag in the trainer getting to the high-intensity efforts intervals (most trainers will probably struggle with a jump from ~85 W to ~530 W.

To me, it looks like the workout, even with the tapering, still should have achieved the recommendation - the XSS will be different now since OP’s signature has changed & all auto-get workouts are in absolute Watts & not defined by their fitness signature:

Because workout autogeneration is (expletive) hard to do when you have 3 dimensions to match of nearly every combination and need to control difficulty and duration for every possible fitness signature combination.

It actually came pretty close to the forecast requirements. Closer than you did. :wink:

Being off by a bit isn’t the end of the world though. You still did well and had a great workout with the right intent. That’s far ahead of the half-random, no-clue-what-this-really-does-for-me workouts you’d do elsewhere.

Try enabling Target Power on the activity details chart to see how much different target watts and actuals were.
I circled some more obvious entries below but you’ll be able to zoom in on your chart and see how much the square top targets differ from the spikes you produced and their duration at target watts.

As a result of your trainer response, the High strain portion of the workout was short of target.
One thing you might consider next time is riding an Autogen workout in Slope mode and manually hit those targets (exceeding a bit if necessary) similar to how you’d ride the workout outdoors.
Any trainer response issue will be eliminated and you can watch the 10-minute power chart on the Session Player to monitor compliance and control the results.
There is some leeway built into the XSS target ratio goal to achieve all green checkmarks.

Thanks everyone for helping me put the pieces together.

The workout generated was pretty close the requirements but following it in real life was a bit harder.

Here’s a zoomed in peak with the target power, with the delay of the trainer catching up. I hoped the tail off on the back end of the interval would balance out the delay but it’s not so. Hitting about 95% of each interval times 21 adds up.

Overall I’ve been very happy with the AI forecasting and the style of the AI workouts.

The next time I have a high insenity workout I’ll try setting it to 103% to see if that accounts for lag in the trainer.


I also have a Direto and with ERG workouts (at least on Zwift) it really helps me hit targets if I purposefully spin up my cadence a few seconds before the high watt intervals hit. I’m new to Xert but it seemed to help with the workout I did Wednesday.