Currently, the qcJson script drops NaN values. To a consumer of the .json this looks like a missing value.
The script should report NaN values as is and report NA for values that it is missing. In no case should the script drop fields from its output, but it should always report the same set of fields.