When you set the attribute "City" in the Key Attributes parameter, the first feature passing through the DuplicateFilter with a certain city name in attribute "City" will exit through the 'Unique' port, all consequent features with the exact same city name in attribute "City" will exit through the 'Duplicate' port.
The duplicateFilter has two ports.
The first time a value is encountered it is output through the Unique port. Any subsequent features with that value are output through the Duplicate port.
So if you had 3 features with Vancouver, the first would come out the unique port along with the rest of the cities, while the remaining 2 Vancouver features would be in the duplicate port.
You may be more interested in the Matcher transformer, which will send all 3 Vancouvers to the Matched port .
I was also confused about the UniqueFilter, since features in the Unique port don't have to be unique values.
You can also use StatisticsCalculator with "Group Processing" and "Total Count", which is faster than the Matcher transformer.