One other way of thinking about it, which may be entirely wrong, but is my first impression as I'm reading it on the first pass, is that we think of assigning/clearing terms in circuit satisfaction as forming and breaking clusters of 'ordered structure', and that he's then using statistical physics to show that there isn't enough speed in the polynomial operations to perform those operations in a particular "phase space" of operations, because these "clusters" end up being too far apart.
One other way of thinking about it, which may be entirely wrong, but is my first impression as I'm reading it on the first pass, is that we think of assigning/clearing terms in circuit satisfaction as forming and breaking clusters of 'ordered structure', and that he's then using statistical physics to show that there isn't enough speed in the polynomial operations to perform those operations in a particular "phase space" of operations, because these "clusters" end up being too far apart.