Creativity in narrative requires careful management of knowledge but story generation systems focusing on creativity have typically circumvented this level of detail by using high level descriptions of events and relations. While this has proven effective for plot generation, narrative generation can be drastically enriched with a grounded representation of actions based on low level simulation. This level of detail and robust knowledge representation can form the basis for a conceptual space exploration driven by narrative knowledge, namely by guiding non-deterministic generation of successive simulation states composing a story. This paper presents and updated version of the story generation system STellA that implements this hybrid model, along with results and discussion on the relative benefits of the described approach.