1) Models are hard to get right. You need to find the right balance between simplifications and reality. The results indicate something is not calibrated or functioning the same as our current state.
2) Models are intended to be simplifications. We draw meaningful results from very simple models (e.g. supply and demand) but without modeling the complex interactions it's hard to say what impact could a certain policy have on the system. Agent-based models are a way to do this and the computing power available is allowing us to experiment with this.
Re: #1. We may find that modeling and seeing outcomes similar to our own is very hard to get to. This may be due to model specification, or it could be that given the rules that govern our own interactions and intentions, our current state may be the one we just chanced upon.
This is a hard challenge to tackle but I'm glad that the authors have published this and shared their code. Don't expect to have prescriptive solutions from it anytime soon though.