Abstract--In numeric optimization algorithms errors at application level considerably affect the performance of their execution on distributed infrastructures. Hours of execution can be lost only due to bad parameter configurations. Though current grid workflow systems have facilitated the deployment of complex scientific applications on distributed environments, the error handling mechanisms remain mostly those provided by the middleware. In this paper, we propose a collaborative platform for the execution of scientific experiments in which we integrate a new approach for treating application errors, using the dynamicity and exception handling mechanisms of the YAWL workflow management system. Thus, application errors are correctly detected and appropriate handling procedures are triggered in order to save as much as possible of the work already executed.