What Counts as a Deal-Breaker?

As I quoted below Paul Griffiths says:

[I]n the case of voting (which is also a deal: I vote, as I hope you do, in response to what a candidate advocates and has done), there are also deal breakers, which is to say actions done or positions advocated sufficient to make voting for someone improper, no matter what other good policies that person may advocate and no matter what other good things he or she may have done. …

…You might object to this line of argument by saying that there are no deal breakers in politics, that we have to look at the whole picture and make prudential, calculative judgments about what will, on the whole, be best for the country. On good days, I find this line half persuasive. But it really won’t do: if the two candidates were Hitler and Stalin, would you feel that you had to vote for one of them?

The argument here is clear enough: there are some actions that a candidate has done or proposes to do which make it categorically immoral to vote for him. In the present election, the deal-breakers Griffiths identifies are Kerry’s support of unrestricted abortion on demand and Bush’s launching of a pre-emptive war on grounds that have turned out to be false.

Now, we can dispute whether those are accurate descriptions of the candidates’ positions or actions as well as whether they are bad, or as bad as Griffiths claims. But what I’m interested in at the moment is the concept of a “deal-breaker” itself. How do we decide what policies constitute deal-breakers?

Griffiths’ remark about Hitler and Stalin should not, I think, be taken to imply that Bush and Kerry are comparable to Hitler and Stalin (though there are some who might say so!). Rather his point is to show that there are at least some cases where the candidates’ stances are such clear-cut instances of evil that we intuitively think it would be wrong to vote for either one. Thus, in principle, the notion of a deal-breaker makes sense.

What isn’t so clear, though, is precisely which cases count as deal-breakers. How would we distinguish between those policies which suffice “to make voting for someone improper” and policies which are merely imprudent or ill-advised?