> how would the model know what this unexpected message means? It would probably just ignore it and continue playing as normal.

Then that’s a poor model then. Significant anomalies should be flagged for manual review, otherwise you corrupt datasets unintentionally.