I’m considering DeepSeek for assigning incoming support tickets to a fixed set of categories. The tricky cases aren’t the obvious ones; they’re short messages with missing context, two issues mixed together, or wording that doesn’t fit the taxonomy cleanly.

How well does it stick to an existing label set rather than inventing a new category or forcing an uncertain ticket into the wrong one? I’d also like to hear how people compare it with simple rules or a smaller classifier, and what they do with tickets that don’t have a clear match.