Actually, the stat is not cumulative, it is a per-100 possessions stat. It doesn't appear to be anything too terribly complicated, it's using some of the player tracking data to determine who was the closest defender on any given defended shot and then comparing the shooting percentage on those shots relative to the shooting percentage on a wide open shot.
Silver points out that it's not a be-all, end-all stat. In fact, it does NOT incorporate important things like steals and rebounds (though I guess it would by nature incorporate blocks), but it certainly looks like a better way to gauge shot defense than any box score metric. They later combine it with a couple of box score metrics to make their updated "CARMELO" defensive rating, so in that it's only a part, not the whole thing.
What they don't actually show here is their updated CARMELO defensive rating, only the raw shooting stat (5 years and last year) and the *change* in the CARMELO rating. If you pick through their final table, you can see that their overall defensive rating of Giannis is +1.94 per 100 possessions, while Bjelica is +1.68 - both good defenders, Giannis better, but Bjelica getting a lot more boost from the stuff that their shot defense metric is measuring, which previously was really only an "eye-test" kind of thing, and then obviously not very reliable.
I agree that Bjelica was, by eye test, an underrated defender, so it's nice to see a metric that does seem to confirm his ability to force bad shots.