When using a multi-column Lucene index which is updated with values for its corresponding properties, the Lucene-backed document has to be deleted and recreated with the updated values (Lucene doesn't support updates)

However, certain field information (e.g. whether a field was numeric or not) is not available when reading the existing document from the indexes. This means that when the update/merge logic is performed, certain field information may be lost which can affect indexing.