David Benson's article ( https://webdocs.cs.ualberta.ca/~games/go/seminar/2002/020717/benson.pdf) is cited as the benchmark for eye definitions, but I've found two problems with it, one of which seems crucial. I claim Rule 2 is incomplete and Theorem 2 is incorrect. The former is easy to fix: the rule doesn't prohibit either player from placing an isolated stone where it will be suffocated. Here's a concrete example where white has just placed a stone illegally, but without violating Rule 2:
A correct statement of the no-suicide rule would be this: it is illegal to place a stone on the board that would not capture any enemy stones and whose block would have no liberties.
A more serious problem is with Theorem 2, which is outright false. Here's my counter-example:
Clearly there is only one black block b. Also, b is safe, because if black passes every single turn, then white will be unable to fill either of the regions f and g. So the set of all safe black blocks is X = {b}. However, contrary to what the theorem states, X is not unconditionally alive. By definition, in order to be unconditionally alive, X would need to have two distinct vital regions. Then, by definition of vitality, b would have two distinct healthy regions. The only possible healthy regions would be f, g and h, since no other region is surrounded by black stones. But neither g nor h is healthy, because they don't satisfy condition (3) of the definition of small black-enclosed region: they contain empty intersections in their interior! So b doesn't have two distinct healthy regions, which is a contradiction, thus proving that X is not unconditionally alive.
My analysis above suggests that condition (3) is the problem. In fact, intuitively, I have no idea where it comes from. Why do you need the interiors of the regions to be white, anyway? Also, the health condition that "every intersection in region R is a liberty of block b" implies condition (3), so it is also nonsense. Why would need every intersection in the region to touch the block? Intuitively, all you need is for each block to be touched by at least one region, so that it becomes protected.
Small correction: at the end I said "Intuitively, all you need is for each block to be touched by at least one region, so that it becomes protected." when I meant to say "Intuitively, all you need is for each block to be touched by at least two regions (satisfying certain properties), so that it becomes protected."
Ahh, I was wrong in my analysis! White can just put black's block in atari without filling region g, so the block is not safe! Here's how white would do it:
Also, my mistake is very instructive: the reason the regions have to be small (no empty intersections in the interior) or healthy (every empty intersection is a liberty of the block) is because otherwise white could put black's block in atari without filling the whole region.