Correctness of system compositions is automatically ensured by using formal behavioural models of services. However, such models are not always provided. We present a model inference technique for blackbox asynchronous services, that interleaves behavioural exploration and incremental model refinement. To save learning effort, only behaviour relevant to the desired system specification is explored. Compared to existing inference techniques that assume only controllable behaviour, our method addresses also uncontrollable events. Experimental results show that obtained models can be successfully used for a safe composition.

A lot of work has been done in the area of building componentbased systems with correct-by-construction adaptors. This is accomplished by using preexisting specifications of the component behaviour. But what happens when known components get to interact with incompletely specified, blackbox components, and errors occur? How can we avoid these errors without modifying existing/legacy components? We present a method to explore and control such systems. Our approach exploits information in correct and erroneous runs to build a controller that ensures our system will avoid observed errors. We consider the behavioural specifications for our known, legacy component as already documented and we infer partial behaviour information of the unknown component by studying its reactions to various interaction scenarios.