When using PivotGrid with a flat data source the handles for dragging the dimensions are labelled with a different string to the dimension headers (screenshot attached).
The dimension headers are labelled with the dataSource.schema.cube.dimensions.[name].caption.
The dimension drag handles are labelled with the dataSource.rows.name (or dataSource.columns.name).
Now, the dataSource.rows.name must be the key in the dataSource.schema.cube.dimensions of the dimension and this must also be a key in each item in the data array -- the field that stores the value of the dimension -- and therefore is in general not suitable for showing to users.
In any case, it seems to me to be a bug that the drag handle is labelled with a different string to the dimension caption.