Seems you have started two threads with similar questions. I believe Mark is discussing this with you in another thread. I must admit, it is very difficult to understand what you are studying, what questions are you trying to answer, etc.? What does rep mean? There are two ways that come to mind: If you have setup the treatment combination and gotten multiple data points without any changes to the treatments, this is repeats (or nested) and does not increase the degrees of freedom. These data points can be useful for estimating means and variation within treatment. This strategy is often used for purposes of understanding measurement error, within sample variation, sample-to-sample variation within treatment, etc. These are not experimental units.
If you are talking replicates, then this can also be treated two different ways:
1. Randomized replicates: Often used to get "unbiased" estimates of experimental error to be used in statistical tests such as F-test.
2. Blocks (RCBD or BIB): In this case, how the blocks are treated depends on how the blocks are created. They can be fixed effects and assigned as such in the model where block-by-factor interactions can be estimated (very useful for estimating robustness particularly when the noise has been identified and "managed" during the experiment) or random effects where only the main effect of block is assigned as a random effect in the model increasing inference space.
What do you mean when you refer to a treatment? Typically this is a set of factors at specified levels.
Fit Model and Fit Y by X are both used for regression modeling. The fit model platform is extremely flexible. It is very useful for understanding multiple x's in a model and the models can be linear and non-linear. I use the Fit Y by X when looking at just one X.