PHY.10:4.4 - Compare the responses that can actually be distinguished

For each candidate design, obtain the readouts allowed by each physical account, including consequential parameter freedom and uncertainty in preparation, conversion and recording. MMP.7 supplies a probability law when records are random; C.16.IR helps determine what an indication can resolve.

Disjoint bounded response ranges give a particularly simple separation under their premises. If the ranges overlap, some outcomes can still discriminate the accounts while others remain compatible with both. Equal expected values likewise need not mean equal distributions or equal time dependence.

When discrimination depends on statistical evidence, use the appropriate model-comparison method with its dependence, sampling and error assumptions. Repeated readings do not automatically supply independent trials. A mathematical design criterion is useful only after the physical design and the interpretation of its records are supplied.

Compare designs on the distinction needed by the work and their full resource demand. A more informative record about one parameter need not be more informative about the rival mechanisms. Use a cheap sufficient contrast before constructing a costly optimization. C.11 compares available options; C.11.DUA addresses the value of another observation or computation.

Before performing the comparison, make its outcome interpretation clear enough to resist changing the question after seeing the result. State what would remain compatible with each account and which apparatus or preparation failure could mimic the disputed effect. The existing explanation or protocol can carry this reasoning.