I am encountering this same issue with scans of people - This is very misleading/frustrating, as we got a full license based off the point cloud data that the demo version provides and are now extremely disappointed. We are reverting back to another software as it constructs these meshes without this additional geometry.
This is occurring with photo-based reconstructions, which go a full 360 around the subject. Our other software interprets these photos correctly and we get a full human figure. Reality Capture is only able to construct half of the subject, with the back side being a box of geometry. What is going on?? There is clearly something wrong with the algorithm, as it is not even using half of the photoset.