AUTHORITY OBSERVATORY
Parent communication, approved
The simplest version of the pattern, and the one a non-technical audience reads instantly. A draft becomes a sent message only when a human approval is redeemed at the boundary. Without the approval, the send does not happen — and the agent cannot step outside what it was delegated to reach the roster.
RECOMMENDATION → APPROVAL → ACTION (INVARIANT 5)
Parent Communication
Why did parents receive this message?
A schedule change triggers a Communications agent to draft a parent message. With director approval, a decision token is redeemed at the boundary and the message sends. Without approval, the send is blocked. And when the agent reaches for the roster — something it was never delegated — that is blocked too.
Watch for: The approved send redeems a token and executes. The unapproved send is blocked (SWARM_TOKEN_REQUIRED). The roster reach is blocked (SWARM_AMPLIFICATION). The comms agent literally cannot change the roster.
Replay proof — rewind & verify
The same run as an authority graph. Click a side effect to rewind its lineage to the human authority; hit Verify to re-derive every edge from the signed receipts. Delegation edges are drawn at a width proportional to their signed budget; blocked branches sit red, off to the side — governance working, not a failure.
Loading graph…
Real boundary, ephemeral key (never the production key), no database, feature dormant. Re-derive any executed action offline: npx @strixgov/verifier swarm swarm_academy_comms
See all three scenarios together in the Authority Observatory.