Thanks for the detailed feedback, and great question. You’re right to push back on this.
The “lack of sustained aerobic base” comment is misleading given your volume. 23,000km in a year is a massive aerobic foundation, and completing Mallorca 312 at 0.77 IF in 11.5 hours confirms that. What the AI is actually reacting to is the drop-off between your 20-minute and 60-minute power numbers. It interprets a steep curve there as a sign of poor endurance, but that’s not the right conclusion for someone with your profile. A more accurate read would be: your aerobic base is excellent, but there might be room to raise your sustainable threshold power through more structured work in that specific zone. That’s a different thing.
On efficiency: the Efficiency Factor on Intervals.icu (Normalized Power / Average HR) is a measure of aerobic decoupling and cardiac efficiency. The “metabolic efficiency” the AI mentions is a looser term referring to how well you sustain power over long durations relative to shorter efforts. So they’re related but not the same metric. Your EF of 1.4 is strong and consistent with someone who has a well-developed aerobic engine.
I’ll look at improving how the AI interprets power profiles for ultra-endurance athletes like yourself, where high volume and sustained output matter more than raw FTP. The analysis shouldn’t be telling someone with your background that they lack an aerobic base.
“undefined% FTP” on ramps — I thought I had fixed this a week ago, but clearly it’s still happening. I’ll take another look.
Cadence on recovery/cooldown — You’re right, having a target RPM on recovery intervals is annoying. That’s on my list to fix, along with your idea of ramping cadence during warmup instead of hitting 90rpm from the start. Good suggestion.
10min recovery before 12min cooldown — Yeah, that’s redundant. The AI sometimes adds a long recovery at the end of the last set when it should just flow into the cooldown. I’ll tighten up the workout structure prompts.
AI saying “completed” when you skipped a set — Right now the AI just checks that you did the workout, not how closely you followed it. I’ll check if Intervals.icu has a match percentage that I should be feeding into the analysis so it gives more accurate feedback.
Zwift repeat grouping — You’re right, this should be grouped as 3x. Sometimes the AI writes out each effort individually instead of using the repeat block syntax, which breaks the grouping in Zwift. I’ve added this to my list.





