I have been very interested in CHR (constraint handling rules) for this kind of activity. It was developed (I think) to provide a systematic basis for the various constraint libraries to 'speak the same language' instead of being implemented without knowledge of each other.

I don't see much literature out there on it (after the mid-2000s) but it always seemed a conceptually elegant way of doing forward-chaining (and inference) in the way of most rules engines. It only has a few simple rules: simplification (remove data/rules), simpagation and propagation (new data).

The best part is that it embeds very easily in Prolog.