Markov Balance Satisfaction Improves Performance in Strictly Batch Offline Imitation Learning | Synapse