This paper describes an adaptable framework that facilitates exploratory analysis, interpretation and classification of beat-to-beat data extracted from the electrocardiogram (ECG). The system supports a variety of user-defined annotations and allows the definition of analysis programs. Special care is taken on the correct treatment of corrupted and missing data, ubiquitously found in real world problems. Besides the computation, the performance of single features can be inspected using different kinds of diagrams provided by the system. Combinations of features can be evaluated using a polynomial classifier. Both the computation and combination of features are defined as tasks that can be dispatched by a server to various clients. The framework is easily adaptable to different problem structures and has been used successfully in three studies.