Comment by ssivark

As one example of such a more general framework -- (variants of) belief propagation might be a good answer if dataset constraints could be cleanly formulated as distributions to be reasoned with.