Well, here is my guess as to the issue. In short, don’t look at the Interval Summary data, because it misses the start and end values from lag.
TR aims to fix this a bit by sending the resistance change 1-2 seconds early, but some jumps are still missed. If you look at the actual time around the target power (even if it doesn’t align with the blue blocks) you are probably getting the full effect of the training stimulus.