> My high credibility makes someone else have bad credibility, which changes the rankings again. How many iterations do we run before we consider the results stable?
Tangential to your point but this is actually a solved problem and you only have to run once. You might even recognize the solution [1] [2].
Why only once? When we did pagerank in class we chose a precision factor and then iterated pagerank until the delta values for the matrices were less than the precision factor
More generally factor graph optimization / msg passing, if you don't need the constraint that it must be an eigenvector/eigenvalue operation. The number of iterations is bounded by graph width but is locally constant, if memory serves.