The reliable encoding mechanism is velocities. OpenUSD attributes may vary over time through time samples, but position interpolation assumes correspondence between sampled array elements. When topology changes over time, adjacent samples may not contain the same number of points, and even matching indices may no longer identify the same physical point. NVIDIA's Learn OpenUSD glossary defines variability as whether a property can change over time, with varying attributes supporting time samples and interpolation behavior. (docs.nvidia.com)

The OpenUSD geometry specification is explicit: ''Using velocities is the only reliable way of encoding the motion of primitives whose topology is varying over time,'' because neighboring sample indices may be unrelated or may not have the same element count. (openusd.org)

Option B is therefore correct. Positions alone are insufficient when topology changes, because linear interpolation between position arrays depends on stable point correspondence. Orientations describe rotational state, commonly relevant to transforms or instancing, but they do not solve topology-varying point motion. This maps to Data Modeling Time Samples, Attribute Variability, UsdGeom Point-Based Motion, Velocities, and Animated Geometry.