Positive and Negative Feedback, or how to build a CO2-Bridge

What I see about feeback in climate change science is pretty mixed, some think that positive feebacks exceed all other set of feedbacks (which I think is very unlikely, given that this is a natural system), others tend to hope that at least some negative feedback through water vapor may be existent.

Let me explain what positive feedback is and why it is so dangerous in assuming one. Positive feedback is a terminology about the stability of a model or system. If the models response is positive feedback, it is like a ball lying on the tip of hill. If you push it, it will roll down the side and accelerate during its run. If the warming process due to CO2 were such a process, it would continue unlimited (or until CO2 saturation in the atmosphere is reached) and even gain momentum during its rise.
However, if some negative feedback in the model would pre-dominate, we would have something like a valley with a ball in it. If you try to push it up the sides of the surrounding mountains, the ball will come back.

Today, the major understanding is something like a combination of both feedbacks. They tend to view that there is an M-shaped feedback. First, we have negative feedback effects that always push down the warming, but if tried to hard, we can get the warming over the small hill and it will run down a even longer and steeper side (thus ending positively).
We don't know the exact heighth or steep of the mountain, but they assume there is one.
This always presupposes that the positive-feedback part of the system will be dominant in the long run!

The hill between the valley and the decend into global crisis may be called the negative-feedback buffer for my convenience now. So, if this is the state of global climate science, I could at least partly say that this might be a possibility, although still unlikely, because most systems are rather like a W, with a few positive-feedbacks, but the negative-feedback superposes all others by an order of magnitude.

Still, modelers seem to suggest that the M is more likely, although they have almost no reliable information about the buffer-zone. How can they start modelling then? Is guessing now science?

But if I am mistaken, and read something into the folks at realclimate, I am willing to amend. BUt if my fault is only to assume an M instead of a positive-feedback only effect, then I think they have missed some classes in model theory...