We present a number of, somewhat unusual, ways of describing what Craig's interpolation theorem achieves, and use them to identify some open problems and further directions. Keywords Interpolation, definability, cross-model relations, first-order fragments, meta-theory.