Trust · claims
Irreversible writes need two named people, not one operator and a model. The receipt is the same object both of them — and a stranger — can open.
A bind, payout, or withdrawal that one person can fire alone is not cleared. The second key is a named human on the receipt. A Slack reaction is not a second key.
You are not asked to trust a dashboard we control. You recompute a hash. If we changed the record, the hash fails. That is the pitch — not “our AI is safe.”