Why does dream coder use e-graphs vs fragment grammars?

I noticed that the original preprint uses fragment grammars but that was changed to use e-graphs in the PLDI accepted paper.

I was curious, why did that change happen? Was it that e-graph "worked better" or was it that the PL audience would be more welcoming of a PL/e-graph method for learning abstractions? 

cross: https://github.com/CatherineWong/dreamcoder/issues/1