The read mapper aligns the reads according to a user-specified scoring scheme, that expresses your expectation of what the data looks like and guides the mapper in deciding whether a particular alignment is good or bad.
As a rule of thumb, you want to choose a scoring scheme that matches the expected characteristics of the data. For example, if you are using a sequencing technology that has a tendency to produce errors in the form of gaps, you may want to lower the gap cost slightly to indicate, that gaps are expected to occur frequently.
Keep in mind, that the mapper deals with reads individually and so cannot distinguish between biological events and technical errors. Since technical errors are (typically) much more likely to occur than true genetic variation, you should set your costs according to what you expect to see in the raw reads, rather than which biological events you expect to be present in the sample itself.