Weird handling of Metadata by RM Studio (Bug report?)

So I have this dataset where I want to map the last example of a given Binary attribute to a missing value. I do this by Mapping it to a unique string and then Declaring that unique string as missing value through the Set Data and Declare Missing operators respectively.

Later on in the process I need to predict on that attribute and I exclude the missing label I created earlier through a Filter Example operator. However, I cannot train it because it considers I have Three categories instead of Two. even after filtering it to include just the examples with the binary classes. On statistics I can see that it sees the old String I mapped as a third Category with 0 examples.

I worked around that by mapping that old string to one of the existing binary classes and since it has 0 examples, it does not alter my dataset.

But I think this is a bug.