Would a solution to the Gettier problem be..

I don't think that solves the Gettier problem. Take the sheep example - The evidence is the dog that the farmer thinks is a sheep. But that is not deceptive evidence by your definition (where x suggests that y is true but y is not true): x (the dog dressed as a sheep) suggests that y (that there is a sheep in the field) is true. And y is true (there is indeed a sheep in the field). Therefore the evidence is non-deceptive. So Gettier's problem still stands even by your modified definition of knowledge.

Sure but the belief in question here is "there is a sheep in the field" not "the thing I am looking at right now is a sheep". "There is a sheep in the field" is a justified true belief and since the evidence led you to that belief, that evidence is not deceptive. Whereas "The thing I am looking at right now is a sheep" is not a true belief therefore it doesn't really matter if we add your caveat about non-deceptive evidence to the standard definition of knowledge.

You could maybe combine the two. Something like "Knowledge is a true belief justified by other true beliefs". I think that would get round Gettier's sheep problem. But I don't think it works for something like the pyromaniac example. Here it is a true belief that these matches do generally light in normal circumstances. The belief was never that the matches were infallible but only that they are usually reliable.

If we try to fix this by limiting our justifications to infallible justifications then we eliminate any knowledge via induction. Suddenly you can't be said to have known the sun will rise today because there was a small chance it might not.

Yeah that's what I was getting at. I think that works for the sheep example but not for the pyromaniac one. The conclusion there is based on a true premise rather than a false premise ie "Sure-Fire matches usually light when the air is clear and the match is dry." While it is not true that the farmer is looking at a sheep, it is true that Sure-Fire matches are generally reliable. That this particular match was not (and only lit due to a fluke) does not make Sure-Fire matches as a whole unreliable. So the evidence is non-deceptive in that case.

It is true that in this instance the match would not have lit despite it being dry and clear, but it is also true that Sure-Fire matches usually do work when it is dry and clear. And that is what he justified his belief that they would work in this instance on.

Where we're disagreeing here is I think Conclusion 1 is actually true. Even though this particular match did not light by its own merit, the pyromaniac was still correct to think it should have done based on his past experience. His evidence (his past experience of matches) still justifies his conclusion about the general reliability of that brand of matches. So you have a justified true belief jystifying another justified true belief.

Well let's take a different approach to try and show where I'm coming from. Suppose you enter the lottery and I say you won't win based on my knowledge of how statistically unlikely it is to win the lottery. You end up winning. I was wrong, no arguments there. But would you say the statistics had "deceived" me?

Do you mean deceptive rather than non-deceptive?