Bind position is the vertex position relative to a specific bone at the time that the mesh is bound to the bone. The way linear blend skinning works is the bind pose vertex position is multiplied by the bone's world transform matrix, then weights are applied (see VertexAttachment source code).
Most people don't need to be bothered by how this works, since the Spine Runtimes handle this for you. Maybe it would help to describe your goals and work backward from there?
