> What have I missed?

Automatically insuring a natural (assuring "blindly," irrespective of the count) makes no more sense than making any incorrect play that forfeits too much EV for the variance reduction it gains. The tradeoff is always between the EV sacrificed by making the apparently "wrong" play, and the commensurate reduction of variance that accompanies the "departure."

So, we examine the certainty equivalent to discover how much we can change an index, in the name of risk aversion. This, in turn, is a function of the spread we're using and the "volatility" of the play -- that is, how much the EV changes for each true count. That volatility is linear for insurance, and is rather large, so insuring at any index appreciably below the correct one just doesn't make sense.

Don