(Srap Tasmaner) I think you are identifying the right problems, but I would suggest rather different solutions. I’m not happy with the causal theory of perception (though I’m not up to date with more recent ideas about it.) because what we see is so heavily dependent on interpretation, which doesn’t fit happily with causality.
Many Gettier problems depend on an inference typified by existential generalization in formal logic. We can infer from “Alice is in the field” to “There is a cow in the field” The catch is that if Alice is not in the field (or even if it isn’t Alice that the farmer saw), the inference collapses, and yet “Daisy is in the field” (if true) is a truth-condition for “There is a cow in the field.” The same applies if the farmer does not know which cow he saw or thought he saw. If the farmer saw a cow, there is a specific cow that he saw. If the cow that he doesn’t know about is the one that establishes the truth, then he didn’t know there was a cow in the field.
“A cow” is ambiguous between “a certain cow” and “a cow” as in “some cow or other”. If the farmer sees a cow, there is some specific cow that is seen, even if he doesn’t know which one it was; the scope of “a” is limited. However, suppose that the farmer has told one of his workers to put some cows in the field without specifying which ones or how many, and says to someone else “There is a (i.e. at least one) cow in the field.” That would be “cow” in the sense of “some cow or other”, which would be made true by any of the cows in the field, so it wouldn’t be a Gettier problem. The reason is that the scope of the justification matches the scope of the proposition. In Gettier cases, it doesn’t, and that’s the root of the problem. This may not apply to some cases proposed as Gettier cases, such as Russell’s clock. But those cases seem to me to have a different format.
I would like to pursue Gettier’s belief that it is possible to be justified in believing that p even when p is false. After all, this is where the door opens for Gettier cases.
Clearly, this falls away when justification is conclusive because falsity does not arise. (If one thinks one has a conclusive justification and it turns out that p is false, one needs to downgrade the justification to partial.)
Partial justification will undoubtedly always be more common than conclusive justifications, so it is worth considering in more detail than Gettier provides.
I can’t see that there is a problem with Gettier’s point when the falsity of p is merely a possibility. Even when p is false, but unknown to anyone, I can't see that it would affect anyone's belief or knowledge.
What matters is what happens when the falsity of p is known, and who knows it.
First, the clearest case. If S knows that p is false, S needs to consider this evidence in relation to the justification for believing that p. Since p is (by definition) conclusively false, the new evidence will outweigh any possible justification available to S, so S will cease to believe that p (or continue to believe that p on irrational grounds). In other words, S cannot believe that S is justified in believing that p and p is false; it is a variant of Moore’s paradox. Hence, of course, Gettier cases always specify that S does not know that p is false. (I have never seen this explained.)
Second, what happens when we know that p is false, but S doesn’t? Gettier cases never specify whether the falsity of p is known to anyone, but it has to be, because we could not appreciate the problem if we don’t. Can we, do we simply say that S is justified in believing p and p is false?
It seems pretty obvious that it is not entirely a matter for S to decide whether we accept his justification; if it were, then any old rubbish could be counted as a justification, and that’s precisely what the J clause was invented to exclude. So, if p is false, then either S’s evidence does not support p, or S’s evidence is false. So the fact that p is false does undermine S’s justification.