Understanding the dynamics of cascades in Twitter is an important modeling problem with multiple applications like viral marketing and the detection and forecasting of emerging events. Key hashtags rise in popularity to a peak and fall, with profiles characteristic to the specific topical area of the hashtag. Traditional text-based classification approaches are inadequate as new hashtags get created dynamically and because social media vocabulary evolves. We demonstrate a text-free approach SansText to classify emerging cascades by modeling the phenomenological patterns of rise and fall. We illustrate the utility of this approach over several specific event classes as well as more general topics in a collection of more than 2 million tweets from multiple countries of Latin America.