The complement of event A is every outcome in the same sample space that is not in A. The two events cannot occur together and together cover the entire sample space, so P(A)+P(not A)=1.
The complement rule does not require equally likely outcomes. If a supermarket promotion gives a particular novelty toy with probability 0.18, then the probability of receiving any other toy is 1−0.18=0.82.
Always define the sample space before naming the complement. “Not late” is the complement of “late” only when those two categories cover every possibility being considered. In binary artificial-intelligence classification such as spam/not spam or fraud/not fraud, the model outputs may be treated as complementary categories only if the classification system is genuinely binary.