When calculating an R-value between two data sets, or producing a map or coefficients, you want the best scale factors so that you can get the best agreement between the two data sets. Therefore, you would allow Rfactor to choose all scale factors.
During refinement you want the individual temperature factors of the model to absorb the signal which causes the parameter B to be non-zero. The model cannot adjust to this signal if the signal has already been eliminated. B must be fixed with a value of zero in all refinement, and Rfactor does this automatically.
Additional restrictions must be placed on the scale factors if the data sets do not cover large areas of reciprocal space. These matters are discussed in the ``Setting Up a Project" chapter of the TNT Users' Guide.