The description of P and Q values as mere 32-bit constants makes it sound like any value would be feasible and that these would be easily selected. They actual

The description of P and Q values as mere 32-bit constants makes it sound like any value would be feasible and that these would be easily selected. They actually hold the two chosen points on the elliptic curve. Also, since these are part of the public-domain knowledge of the system, it is reasonable to expect that the values might even be built into hardware in the field, etc. and not easily changed. It might be much more feasible to revert to an earlier value if the original value was already stored somewhere. The real problem with DualEC-PRNG is that if P and Q were somehow related, you wouldn't know this by looking at them and knowledge of the relationship would be restricted to the issuers (in the case of the "recommended" P and Q values, the NSA; in the case of Juniper, their own employees). The hidden relationship could be used to deduce the output of the generator and therefore deduce any keys based on the now-predictable values.