Drat, I must have misunderstood the paper, then, because it seemed (to me) to require keeping the old cyphertext in order to determine the GT relation.

Any chance you have a more-plain-English (e.g. like in a NIST pub) description of the paper's proposed system.