Ok, a bit more details. And I think I see the problem.
Seiberg and Nelson claim that U(1)R -> broken SUSY, except in some special cases. This proof hinges on a redefinition of fields so that you can factor the superpotential into a part that has R charge and one that doesn't:
W = T f(a,b,c,...)
where T has r charge 2, and all of the other fields have r charge 0. What I didn't realize an hour ago is that this redefinition probably means that your fields aren't canonically normalized---i.e. at least ONE of them looks like A/X, where A and X are two chiral superfields.
You run into problems when T = 0...specifically, when T = 0, it could be that some other field, say a, can be zero, eg a = 0, and you can satisfy the F=0 constraints with a good U(1)R, so you might be fooled and think that you have a SUSY ground state. In the specific example I'm looking at (and probably more generally), a = A/x and T = X (this is a simple ORaifeartaigh model):
[tex]\mathcal{W} =\frac{1}{2} h X A^2 + mAB + gX[/tex].
In terms of the new fields:
[tex]\mathcal{W} = T \{\frac{1}{2}h a^2 + mab + g \},[/tex]
where g is the dimensionful constant, not a field!
Anyway, this means that when you now compute the scalar potential, you have to write
[tex]V \sim F_i \bar{F}_{\bar{j}} K^{i\bar{j}}[/tex]
i.e. the Kahler metric is no longer flat. In the case I am describing, when T = 0, the new Kahler metric (in terms of non-canonical fields) is now singular.
So it seems I have a bit of a hole: I know that F = 0, but K = infinity. Obviously something weird is happening. In terms of the original OR model, SUSY is clearly broken (SUSY is broken everywhere). But how do I see that SUSY breaking in terms of the new fields?
I suspect that there's something to do with a singular Kahler matrix, but I'd like a reference to a paper, or a gentle pat on the head with a "Good boy", or something.