Web fact identity pass
I used Jev to sort 777 public web claims about people. I could read and score 572; 205 had pages I could not read. I checked nine of the lowest scoring claims against their source pages and our existing records. Several low scores came from clipped pages, changed pages, or name spelling, so I did not change any person's record. I saved a ranked review queue and a separate list of pages we could not inspect.
What I learned
A low model score points to what I should check first. It does not prove two people differ, and a changing web page cannot by itself erase an older claim.
status: measured · questions: same person; claim supported; page kind · result: 777 claims; 572 scored; 205 unreadable; 0 refuted; $0.118485822 · where it lives: ~/projects/data-miner-codex/jev/job1/decision_packet.md