First I want to say that I totally agree with Jonathan's answer. I've been through the same ideas in my forays into genealogy database modelling, but have found that quantifying uncertainties, other than on a very rough scale, is simply too much work for too little benefit.
In my own research database I've got surety levels for the parent-child relationships, which in effect take the values Certain, Probable, Possible. I will only publish those that I've estimated as Certain. Quite a few of those actually lack direct evidence, but they at least have pretty strong circumstantial evidence.
I've got this profile of one of my g4-grandmothers, Marte Andersdatter, that I've been using as an example many times. I've never found a record which says that the Marte Andersdatter born at Veholt in 1736 is the same Marte Andersdatter who marries Abraham Jonsen in 1758. But it fits in all kinds of ways, and having worked with this parish for 23 years, I've never found any evidence that weakens or disproves it. But, as long as there's no direct evidence that proves it, it's still just a hypothesis. How do you put an actual number to that kind of uncertainty?
As Jonathan says, any day a hitherto unknown document might surface that either proves or disproves this relation. Then the probability goes to either 1 or 0. But until then, I won't assign any other probability value to it than "pretty convincing".