One important characteristic of any solution to temporal classification is that it should be robust to temporal variations of the kind that occur in temporal domains, which were previously discussed in Section 2.5. Careful consideration reveals that metafeatures implicitly capture the temporal variation as parameter values, and in so doing, allow temporal variation to be handled. For instance, if an Increasing event occurred slightly later than expected, then the difference is mapped into a difference of the midtime parameter; and the greater the difference in the midtime, the greater the difference in that parameter's value. Hence, the temporal variation between the two cases is mapped into proximal parameter values, where the usual distance metrics still apply in a sensible manner. In terms of the parameter space, this means that two instantiated features that occur at approximately the same time or last for approximately the same duration should be close. Hence the representation as midtime and duration would be more appropriate.