I don’t think -3.2 is an actual return. It is rather an expected return conditional on the realized information coefficient. And Q2 says, “given the realized IC…” which indicates that you should compare to the -3.2.
For Q3, check out the paragraph right above the blue box, that explains it. I will try to summarize it:
Once you observe the realized return, you can decompose the return variance in two parts:
a) IC variation (TC2)
b) constraint induced noise variation (1-TC2)
If TC is high, this means more of the variation in success is attributed to forecasting success (since loosely speaking, TC is one of the measures of forecasting ability of active management).