In recent years, Optical Music Recognition (OMR) technologies have experienced a notable boost thanks mainly to the use of new pipelines based on machine learning, especially on deep neural networks. These methods are usually studied just from the point of view of the accuracy of the output of the networks. However, from a practical perspective in a real-world context, this is not enough. In this paper, we present a design of a tool devised for allowing the scientific study of the complete OMR workflow in different scenarios and notations, including both the possibility of analyzing the real impact of improvements in automatic recognition models and how they are integrated for practical purposes in the work of the transcriber.