Consider also how knowledge interacts with conditional orders. Suppose a prison guard is ordered to shoot a prisoner if and only if they are trying to escape. If the guard knows someone is trying to escape and yet does not shoot he will be held accountable. Suppose meanwhile he does not know that someone is trying to escape but shoots them anyway, acting on a belief grounded in a baseless hunch that they were trying to escape. Here again the person will be faulted, even if the person is in fact trying to escape. Our common practice is to require knowledge of the antecedent of a conditional order in order to discharge it.The principle to take from this seems to be this:
KAct: If you oughtn't X unless C obtains, you oughtn't X unless you know C obtains.
Consider two claims about knowledge and warranted assertion:
KAN: You oughtn't assert what you don't know.
KAS: You may assert what you know.
P1. In C1, you know p but aren't in a position to know that you do [~KK].
P2. You may assert p in C1[P1, KAS].
P3. You shouldn't assert p in C1 unless you know p in C1 [P1, KAN].
P4. You shouldn't assert p in C1 unless you know that you know p in C1 [P3, Kact].
P5. You don't know that you know p in C1 [P1].
P6. You shouldn't assert p in C1 [P4, P5].
P6 is incompatible with P2, so something has to give. I think Kact has to be false, but I also think that principles in the neighborhood of Kact have to be more fundamental than principles that govern assertion. Since KAS seems much more plausible than KAN or Kact, I'd try to winnow the requirements on warrant, permission, etc. to something much weaker. I'd also worry about the coherence of principles in the neighborhood of Kact. I discuss this worry in the book and in my JPhil paper. In the literature, everyone seems to be fixated on Gettier cases and false belief cases, but these structural problems strike me as much more interesting.