We want to model things as they are, not as they should be.

Simplifying assumptions are ok, if applied judiciously.

Example: no need to model all 250 byte codes, as long as one chooses to model all the interesting ones:

e.g., Ignoring fadd  is fine, ignoring subroutines is not.

Accuracy is important: when in doubt, ask us.

Examples: