The developer should select reliability methods that match how the test is built. If the total score comes from a substantially multidimensional test, it should be treated as a composite, and the two halves used in a split-half analysis should be alike in content and statistical behaviour. For adaptive tests, simulation from the model and model-based conditional standard errors can be appropriate.
The graph holds this control, the 0 it maps to, and the evidence behind each claim, over MCP and REST.