It would be nice to be able to allow the same data point to be labelled multiple times when using superintendent in "distributed" mode. It would be even nicer to incorporate the labeller's ID in model that can determine reliability of labellers.
The resolution of conflicting labels should probably be optional and configurable.