I am conducting a structural equation model with three latent variables containing three indicators each. The latent variable CFA's are significant with constraints on 2/3 indicators, but the SEM is significant with only 1/3 constraints on the indicators for each of the three latent variables. Is it okay to use latent variables in the SEM model that have different constraints, or do the latent variables have to fit in the SEM in the same form that they were significant, as per CFA?