I have not been able to find a way to directly use a column as reconciliation results. In the case of Wikidata, reconciling a column of item IDs works (I assume it’s just due to the fact that wbsearchentities supports searching for the item ID), but it takes an unnecessarily long time.

This would be very useful for a kind of workflow where I don’t want to work with the dataset in OpenRefine all the time as one monolithic project, but rather prefer to use plain old text files and associated Unix tools (cut, join, awk, …) as much as possible, and exchange parts of the data with OpenRefine as needed – see https://github.com/lucaswerkmeister/bundestag/ for example (most of the details are in the commit messages).