AUTHORITY OBSERVATORY · GOVERNED OPERATIONS PILOT

One pattern, everywhere

The point of this pilot is not that an organization is “run by AI.” It is that AI can operate inside a governed authority system — with real people, real approvals, real money, and a real audit trail. Below are three scenarios from one real operating environment. Each is a different governance shape, and each runs through the same execution boundary that governs an agent swarm. Switch between them and watch the same underlying structure appear every time.

AuthorityEvidenceDecisionActionProof

ADMISSIBILITY RE-EVALUATED AT EXECUTION

Player Eligibility

Why was this player allowed to play?

A tournament director’s approved run authorizes an Eligibility Office, which delegates bounded check authority to age, grade, and waiver verifiers plus a roster agent. Each check runs under governance; the eligible player is approved onto the roster. An ineligible player’s approval is denied at the boundary — admissibility is re-checked at the moment of execution, not assumed from the submission.

Watch for: Three checks and one roster approval execute. The ineligible approval is refused with a real SWARM_POLICY_DENIED verdict — the roster is never modified.

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_eligibility

Eligibility, refunds, communications, healthcare, agent swarms — the same shape

Strip away the domain and each of these is one structure: an authority delegates, evidence accrues, a decision is made, an action is taken, and a proof is produced that an auditor can re-derive. That is the category — not “sports software” and not “agent orchestration.”

DomainAuthorityDecisionAction
Player eligibilityTournament rulebookEligible / ineligibleRoster approved
Tournament refundsDirector → OperationsRefund within budget?Refund issued
Parent communicationsOperations + approvalApproved to send?Message sent
Healthcare prior-authPayer policyMedically necessary?Service authorized
Agent swarmsHuman → delegated agentsAdmissible at execution?Side effect runs

The healthcare prior-authorization solution is the same engine in a regulated domain — see it here.

What makes this honest

Every node and edge changes because the boundary returned a real, Ed25519-signed verdict — the chain is re-verified at point-of-use, attenuation and budget are re-checked, and evidence is written before any side effect. These scenarios sign with an ephemeral key (never the production key) and touch no database. Deep-link a single scenario: eligibility, refunds, communications. Sibling demos: the deal-pipeline demo room and generic budget controls.