If you have no theoretical model to support your analysis, reporting the saturated model without fit statstics seems the best option. Removing non-significant results will gives you artificially good fits statistics, just because you develop the model in a statistics fashion rather than a theoretical one.

If you have no theoretical model to support your analysis, reporting the saturated model without fit statistics seems the best option. Removing non-significant results will gives you artificially good fit statistics, just because you developed the model in a statistics fashion (removing non significant path) rather than a theoretical one (justified conceptually).