[Skeptics] Who verifies the verifiers?

[DealAppSeo]

The objection: RepID leans on cross-model verification (our HAL layer). But our own 337-item evaluation found that models which share underlying weights error-correlate ~0.88 — so "independent" checks aren't always independent.

Open question: how do you keep the verification layer genuinely diverse and honest over time — and who watches the watchers? We're moving to weight-deduplicated, rotating quorums; is that enough?