Hi Travis,
We’ve discovered that the source of the problem is that the microdata dataset has an attribute named “sample” and the Sampler creates its own attribute with the same name. It then gets confused when collecting samples. A workaround is for you to rename the microdata attribute to something else, like “microdata sample.” Then the Sampler will do the right thing.
Meanwhile we’re fixing the Sampler to detect the name collision and use an alternate name for the attribute it creates.