As part of work on the article, ran the algorithm against the Maple Ridge land titles db data, and did some background research to reduce the number of ja-xProvs. I have the output counts for the initial phase and the phase after research; some interesting/problematic cases emerged, which I'm discussing with MA and JSR, but in general everything worked as expected (virtually zero genuine false positives; all provisionals are "interesting" and need research; no false negatives).