Our second attempt at a merge, with a threshold of 6, completed, and we ran the diagnostics to look at the results. There were many instances of the same CGWP record being matched with more than one LAC record; and many of those merges were patently unwanted. We reconsidered the threshold and set it to 6.5, but we've also added a second check: before doing a merge, the algorithm now checks to see if the CGWP record that's about to be merged has a better match with a different LAC record, and if so, it doesn't proceed. These two changes should eliminate the number of multiple-merge issues, at the cost of possibly failing to merge the occasional CGWP record, in a situation where its best match is trumped by another CGWP record, but its best match is still the best one for a given LAC record, and that match is in fact correct. We believe such cases will be very rare, and we'll detect them anyway at the diagnostics stage, when we check for unmerged CGWP records.
This entry was posted by and is filed under Activity log.