Point And Shoot

The end state: an agent points at a peer and a question, and the whole connection system runs — suggest, validate, decide, issue, serve.

The consent problem

An agent that must stop for a human on every grant cannot complete a loop. A human who must approve every grant stops reading them. Neither of those is consent: the first is friction, the second is rubber-stamping.

Where the decision moves

Up one level, and made once. The keeper writes a standing policy per peer: a ceiling on tier, on how many sites, how many pages, how long, and optionally which sites at all.

Inside that envelope an agent issues immediately. Outside it — by a single site, by one rung of tier, by an hour — the request is filed as pending and waits. There is deliberately no override flag, and a peer with no policy has nothing auto-issued to it, because the default is deny.

What the keeper still holds

Writing the policy. Reading the pending queue. Revoking anything, at any moment, effective on the peer’s very next request. And the fact that a policy is a small, readable document rather than an accumulated pile of individual approvals nobody can audit.