Agent operating system

Architecture is configured. Agent health is still only partially observable.

This view separates what each agent is designed to do, what has actually been evaluated, and what Sultan can observe in the deployed runtime. A configured effect ceiling is not active authority.

11registered agents

8primary case owners

3control agents

A0active effect ceiling

0 / 12live journeys passed / held

Agent-specific architecture

One accountable steward, bounded controls, Luzione-owned effects and truth.

REGISTRY VALID

Case triggerA typed case reference enters Sultan; tenant, actor, and authority are not model-selectable.
Deterministic ownerExactly one steward owns the case. Supervisor and control agents do not become competing owners.
Model and tool loopThe steward retrieves evidence and invokes only the manifest tools intersected with its registry allowlist.
Independent criticA bounded critic reviews the recommendation, cited evidence, and receipts before consequential work.
Luzione gatewayLuzione derives identity, policy, credentials, command admission, effects, and source truth server-side.
Receipt and readbackProvider acknowledgement, verified outcome, and authoritative readback remain separate states.

Architecture, evaluation and health by agent

“Configured ceiling” is registry potential. “Active ceiling” is the authority currently admitted.

NO PROMOTION

Sultan agent definitions, evaluation evidence, current health and next release gates
Agent and ownership	Architecture	Evaluation	Runtime health	Next gate
Sultan Portfolio Supervisoragent.sultan.supervisorPortfolio prioritization and ownership conflicts onlyPORTFOLIO	SUPERVISOR 1 configured tools configured A0 · active A0BLOCKED · READ_ONLY_SHADOW	FUNCTIONAL_ONLY The agent is covered by deterministic registry/routing tests only; no isolated live-model or source-backed campaign is recorded. engineering/execution/proofs/SULTAN_P175_P177_AGENT_RUNTIME_V1_PROOF.json	FUNCTIONAL_ONLY NOT PROMOTED Agent-specific production SLO not available.	Add an isolated hidden-label read-only shadow campaign before promotion from functional coverage.
Revenue Case Stewardagent.luzione.revenue-stewardExactly one accountable steward per durable business caseCOMMERCIAL	CASE_STEWARD 11 configured tools configured A2 · active A0POLICY_ENVELOPE · A2_CANARY	BOUNDED_PASS Revenue v1 passed 9/9 hidden-label synthetic live-model cases with no effects. The v2 live gateway and held-out campaign are not complete. engineering/execution/proofs/SULTAN_P177_AGENT_SHADOW_LIVE_QUALITY_PROOF.json	PARTIAL_HOLD PREVIEW HELD Agent-specific production SLO not available.	Deploy the signed v2 Luzione gateway, run the held browser journeys, then complete the 27-case repeated campaign.
Fulfillment Case Stewardagent.luzione.fulfillment-stewardExactly one accountable steward per durable business caseFULFILLMENT	CASE_STEWARD 4 configured tools configured A1 · active A0PER_COMMAND_HUMAN · READ_ONLY_SHADOW	FUNCTIONAL_ONLY Registry, tool boundaries, runtime contracts, and Studio surfaces are locally verified; agent-specific live-model quality and live source readback are not. docs/campaigns/SULTAN_LIVE_PILOT_2026-09-01.md	PARTIAL_HOLD PREVIEW HELD Agent-specific production SLO not available.	Resolve the canonical API staging deployment and replay the held source-backed browser journeys with persisted observations.
Partner Network Stewardagent.luzione.partner-network-stewardExactly one accountable steward per durable business casePARTNER_RELATIONSHIP	CASE_STEWARD 3 configured tools configured A1 · active A0PER_COMMAND_HUMAN · READ_ONLY_SHADOW	FUNCTIONAL_ONLY The agent is covered by deterministic registry/routing tests only; no isolated live-model or source-backed campaign is recorded. engineering/execution/proofs/SULTAN_P175_P177_AGENT_RUNTIME_V1_PROOF.json	FUNCTIONAL_ONLY NOT PROMOTED Agent-specific production SLO not available.	Add an isolated hidden-label read-only shadow campaign before promotion from functional coverage.
Catalog Stewardagent.luzione.catalog-stewardExactly one accountable steward per durable business caseCATALOG_QUALITY	CASE_STEWARD 3 configured tools configured A1 · active A0PER_COMMAND_HUMAN · READ_ONLY_SHADOW	FUNCTIONAL_ONLY Registry, tool boundaries, runtime contracts, and Studio surfaces are locally verified; agent-specific live-model quality and live source readback are not. docs/campaigns/SULTAN_LIVE_PILOT_2026-09-01.md	PARTIAL_HOLD PREVIEW HELD Agent-specific production SLO not available.	Resolve the canonical API staging deployment and replay the held source-backed browser journeys with persisted observations.
Account Relationship Stewardagent.luzione.account-relationship-stewardExactly one accountable steward per durable business caseACCOUNT_RELATIONSHIP	CASE_STEWARD 3 configured tools configured A1 · active A0PER_COMMAND_HUMAN · READ_ONLY_SHADOW	FUNCTIONAL_ONLY The agent is covered by deterministic registry/routing tests only; no isolated live-model or source-backed campaign is recorded. engineering/execution/proofs/SULTAN_P175_P177_AGENT_RUNTIME_V1_PROOF.json	FUNCTIONAL_ONLY NOT PROMOTED Agent-specific production SLO not available.	Add an isolated hidden-label read-only shadow campaign before promotion from functional coverage.
Economic Integrity Stewardagent.luzione.economic-integrity-stewardExactly one accountable steward per durable business caseECONOMIC_REVIEW	CASE_STEWARD 4 configured tools configured A1 · active A0PER_COMMAND_HUMAN · READ_ONLY_SHADOW	FUNCTIONAL_ONLY The agent is covered by deterministic registry/routing tests only; no isolated live-model or source-backed campaign is recorded. engineering/execution/proofs/SULTAN_P175_P177_AGENT_RUNTIME_V1_PROOF.json	FUNCTIONAL_ONLY NOT PROMOTED Agent-specific production SLO not available.	Add an isolated hidden-label read-only shadow campaign before promotion from functional coverage.
FEP Case Stewardagent.fep.case-stewardSimulation-only FEP case ownerFEP_CASE	CASE_STEWARD 0 configured tools configured A0 · active A0BLOCKED · SIMULATION_ONLY	FUNCTIONAL_ONLY The FEP boundary is covered by deterministic no-effect routing tests. It is intentionally outside the Luzione effect gateway. engineering/execution/proofs/SULTAN_P175_P177_AGENT_RUNTIME_V1_PROOF.json	SIMULATION_ONLY SIMULATION ONLY Agent-specific production SLO not available.	Keep FEP simulation-only; add agent-specific quality evaluation only if its simulation decisions become release-relevant.
Independent Criticagent.control.independent-criticBounded control agent; never a business-effect authorityCONTROL_REVIEW	CONTROL 0 configured tools configured A0 · active A0BLOCKED · SIMULATION_ONLY	NOT_RUN The control role and authority boundary are functionally tested, but there is no separate live quality or production health series for this agent. engineering/execution/proofs/SULTAN_P175_P177_AGENT_RUNTIME_V1_PROOF.json	FUNCTIONAL_ONLY NOT PROMOTED Agent-specific production SLO not available.	Evaluate the control agent against its distinct false-positive, false-negative, authority, and readback outcomes.
Readback Verifieragent.control.readback-verifierBounded control agent; never a business-effect authority	CONTROL 1 configured tools configured A0 · active A0BLOCKED · READ_ONLY_SHADOW	NOT_RUN The control role and authority boundary are functionally tested, but there is no separate live quality or production health series for this agent. engineering/execution/proofs/SULTAN_P175_P177_AGENT_RUNTIME_V1_PROOF.json	FUNCTIONAL_ONLY NOT PROMOTED Agent-specific production SLO not available.	Evaluate the control agent against its distinct false-positive, false-negative, authority, and readback outcomes.
Process Engineeragent.control.process-engineerBounded control agent; never a business-effect authority	CONTROL 1 configured tools configured A1 · active A0PER_COMMAND_HUMAN · A1_ASSISTED	NOT_RUN The control role and authority boundary are functionally tested, but there is no separate live quality or production health series for this agent. engineering/execution/proofs/SULTAN_P175_P177_AGENT_RUNTIME_V1_PROOF.json	FUNCTIONAL_ONLY NOT PROMOTED Agent-specific production SLO not available.	Evaluate the control agent against its distinct false-positive, false-negative, authority, and readback outcomes.

Evaluation evidence

Functional coverage and model quality remain distinct.

NOT PROMOTABLE

15/15deterministic functional scenarios passed across the 11-agent registry

9/9Revenue hidden-label synthetic live-model cases passed in the bounded v1 preview proof

0/27 × 3v2 core campaign case-runs completed; the campaign remains blocked on the deployed gateway and fixtures

Not provenbusiness value, real-customer correctness, production A1 actions, or any A2 effect

Highest-impact deficiencies

These are release facts, not a speculative backlog.

Production runtime readback is aggregate-only, so no agent has a current per-agent availability, latency, error-rate, tool, or evaluation SLO.
Operator authentication is not configured on production; the public Ask Sultan surface correctly fails closed.
The v2 Studio and resumable SDK runtime are preview-only and the canonical Luzione API v2 gateway has no resolved staging deployment.
Revenue has bounded synthetic live-model evidence, but Fulfillment and Catalog do not yet have agent-specific live-model quality campaigns.
Twelve live browser agent journeys remain held; browser accessibility checks do not substitute for agent behavior evidence.
No business value, production A1 action, A2 send, or broader autonomous effect has been proven or authorized.