That could make sense if the data didn't correct itself randomly as well though right? It's not as if the Visitor never is classified, its that they are classified, unclassified, and reclassified within 5 minutes of each other.
I was pulling some Order metrics off a classification Dimension "Name (Product)", where we have a friendly name instead of a product ID. When I pulled the data for a specific product with a segment, it told me we had 424 Orders. Then when adding different segments as new columns to get different cut...