Right now, the scatterplot sets its own value accessor that looks at part of the key, presumably so that the coordinateGridMixin can use the value accessor to get Y values

And then in a couple of places it accesses d.value directly so that it knows whether a bubble exists or not.

Both of these things are somewhat wrong.

First, for a scatterplot, Y is really part of the key, not a value. If there are necessary assumptions in the coordinateGridMixin that value is Y, then the scatterplot is not really a coordinateGrid.

Second, we should never access d.value directly but should use accessors, so that people can reduce to whatever they want, not just single values. This part breaks reducing colors or sizes, for example. We should always support multivalue reductions, everywhere.

https://groups.google.com/forum/#!topic/dc-js-user-group/55nJcU0qDfg