$ paw-test judge --diff old.json new.json
exit code: 0
--- stdout ---
Flipped verdicts (1/2):

  Input: today
    Output: 2026-09-11
    anthropic/claude-haiku-4-5/temperature=0.0: True (correct)  ->  anthropic/claude-haiku-4-5/temperature=0.0: False
(flipped)

Flip rate: 1/2 (50.0%) -- all 2 scored
