In a previous comment on karma systems I suggested a scheme where there was a voting 'diamond' up, down, left, right, and center. Where up was "good comment/submission" and down was "not-good comment/submission", left was "less like this", and right was "more like this" and center was perfect.

That allows you to establish both what you like the site to have in it, and how you much you liked this instance of it. Perfect would mean simply that "spot on" or "this."

The second part of the system would then adjust the presentation of your karma (to you and others) unscaled vector along the line to the 'perfect' site in the context of the viewer.

To use a simple example, lets say HN was over run with cat pictures, and I thought they were the best so I vote like this and "+1" (up and to the right), You the reader hate cat pictures so you always vote down/less like this.

So we've created virtual root nodes at +loves cat pictures and one at +hates cat pictures. Now I can plot your karma as negative with respect to mine if we assume that the center point between your root nexus and mine is 0. Now do this with enough topics and you get multiple consituencies all in the same discussion space where your view of their karma will inform you how likely it is you'll like to read what they wrote :-)