Today I ran into an interesting problem using NumericField.

The issue is that when a document is retrieved, fields encoded using NumericField come back as plain old strings. This isn’t a problem by itself, but if you retrieve a document, modify some fields and leave others alone, any unmodified field that was previously a NumericField reverts to a basic string value.

The issue is demonstrated by this sample NUnit test fixture:

Long story short, if your documents are going to take round trips and get re-indexed, you must re-encode any NumericFields before adding the document back to the index, even if they did not change.