The current implementation relies on valid cross-reference offsets in the PDF documents so that PDF objects are easily found and parsed. However, we may implement a mechanism for importing documents with invalid offsets for the objects inside the stream.