This property states that a good PRNG should be able to eventually recover from compromise even if the entropy is injected into the system at a very slow pace,

This property states that a good PRNG should be able to eventually recover from compromise even if the entropy is injected into the system at a very slow pace, and expresses the real-life expected behavior of existing PRNG designs. If someone figures out the internal state of a PRNG that doesn't take input entropy, they can predict all future outputs. If someone figures out the internal state of a PRNG that does take input entropy, they can mostly predict near-future output but shouldn't be able to predict what the output will be after enough new input entropy has been provided. I think this paper is saying that that "shouldn't" either isn't true of Linux, or takes more input entropy to happen than it ought to?