We Don't Know it All: on Understanding and Correcting Concurrent Computations with Inaccurate Observations