The power spike detection depends on FTP because it uses your assumed power curve to spot spikes. The FTP estimation does not but if the spike detection is “fixing” things it shouldn’t it won’t work well. If your power data doesn’t ever have spikes you can turn of detection globally. Intervals.icu uses max efforts of at least 3 minutes to put you on a pre-defined power curve (same as 75% of 5m power, or 95% of 20m power but generalised) and uses Morton’s 3P model parameters for that curve for FTP (using FTP = CP). You can see the model curves on the /power page.
For myself it gives numbers 1-2% lower that Xert and both are reasonable estimates to set zones for my training. When Xert says “breakthrough” Intervals.icu says “FTP up” so I think Xert is doing something very similar. If you add the W’bal series to an “FTP up” ride and enter W’ (get it from /power page) you can see W’bal going to zero or below at the “FTP up” moment. At this point you have exceeded the model prediction and so your FTP has likely gone up.
You can adjust the minimum duration that Intervals.icu uses for FTP estimation in /settings. Maybe 3 minutes is too short for you, 5m might be better.
How much too high is the intervals.icu estimate and how does it compare to a TR ramp test (if you have done one)?
The whole business of estimating FTP is very interesting and a lot fuzzier than what I thought before I started working on this (last December)!