Problem on DSGE modeling

Dear Professor Pfeifer,

I begin with a simple DSGE model and solve it using Dynare. This time I could get acceptable result of model fit (comparison between model implied moments and empirical moments ).
Then I want to add some little feature as the innovation into the DSGE model, and it could be successfully solved. However, the model fit is always getting bad results after the bayesian estimation. I have no idea why this always happened as I think the economic logic behind this feature could make sense.
Does this mean that this innovation is actually meaningless (garbage in and garbage out) ? Now I feel very confused on how to specify an ideal model with the feature I want. Is there any technical details that I overlooked ?

Thank you for your time.
Best regards

That is impossible to tell without knowing the model. Maybe your prior for the new feature is problematic so that you force the model to have it in there strongly instead of the data being able to tell you how strong it should be.