onelife1

If you actually read the CFAI text, it’s much clearer.
On p 236, the text says you can use the standard formula to calculate the exact stdev with the covariance table given.
Or you can take a short cut to approximate this stdev by a linear combination of 2 adjacent corner portfolios.
In the particular example given, the EF bows out to the left implying that a linear approximation will over-estimate the exact stdev which, in effect,  reflect a less-than-one positive correlation between 2 CPs.
You could draw a  diagram to visualise this. Think like convexity adjustment to bond duration.