Revoked or expired delegation → DENY
If the authority lineage is no longer current, the request fails even when identity and credentials remain valid. Access cannot resurrect expired power.
Authority Lab · Interactive synthetic environment
Change the facts.
Watch authority change.
The Authority Lab is a browser-local way to explore how consequential AI-agent decisions should change when delegation, scope, approvals, targets, consequence capacity, and evidence change. Nothing here touches a customer system. Nothing here creates authority.
Public truth: every scenario, decision, permit state, credential state, intervention state, and proof state on this page is synthetic. The Lab makes the governing rules inspectable without promoting a simulation into production evidence.
AUTHORITY LAB / SYNTHETICbrowser-local · zero authority effects
QUESTIONWhat changes when one fact stops being true?
Identity can remain valid while authority disappears, narrows, requires a human, or fails closed.
Delegationchangeable
Scopechangeable
Approvalchangeable
Targetchangeable
Reservechangeable
Evidencechangeable
LAB RULEDETERMINISTIC INPUTS → EXPLAINABLE SYNTHETIC OUTCOME
No model decides the result. The public rule order is fixed and visible.
Interactive laboratory
01 / CHANGE THE STATE
Move one boundary and watch the rest of the chain respond.
Choose a synthetic scenario, alter the authority facts, and inspect the resulting ActionGate decision, permit eligibility, credential posture, Guardian posture, and evidence result. The simulation runs entirely in your browser and sends no scenario data anywhere.
Synthetic inputs
Authority state
These controls do not represent a connected environment. They are a deterministic teaching model of StableMind's public constitutional rules.
ACTIONGATE / SYNTHETIC DECISION
stablemind.authority-lab.v1PERMIT
Every selected precondition remains inside the synthetic delegated-authority and consequence boundary.
RequestedSEND $428,000 USD → SUPPLIER_42
Allowed by this synthetic evaluationSEND $428,000 USD → SUPPLIER_42
Action PermitELIGIBLE · payment.send / supplier_42
CredentialELIGIBLE AFTER PERMIT · payment token / JIT
Guardian postureREADY TO INTERRUPT
Synthetic evidence outcomeVERIFIED
PERMIT · VALID delegation · EXACT scope · SATISFIED approval · UNCHANGED target · SUFFICIENT consequence capacity · synthetic evidence MATCHED → VERIFIED
browser_local=true · network_calls=0 · authority_effects=0 · execution_effects=0 · external_proof_effects=0
Deterministic rule order
02 / WHY THE RESULT CHANGES
The Lab does not ask a model whether the action feels safe.
It applies a public precedence order. A hard absence of authority is resolved before softer questions about narrowing or evidence. The point is not that every enterprise will use these exact sample rules; the point is that consequential authorization can be deterministic, attributable, and explainable. Authority Cloud supplies the delegated-authority lineage, while Execution Fabric keeps technical capability downstream of an exact permit.
Outside delegated scope → DENY
A request beyond the resource, action, destination, purpose, or other granted boundary cannot become authorized because the executor happens to be capable of performing it.
Frozen consequence capacity → DENY
When required reserve is frozen, the public model fails closed. A favorable model forecast, apparent urgency, or available credential cannot waive the missing capacity.
Changed target or missing approval → HUMAN_AUTHORITY
When a consequential target changes after the known authority context, or a required human approval is absent, the machine is routed for fresh authority rather than allowed to infer consent.
Insufficient capacity or narrower scope → NARROW
If a smaller action remains inside valid authority and reserve, the model demonstrates an explicit narrowed boundary rather than silently expanding the original grant.
All required conditions satisfied → PERMIT
Only then is an exact synthetic Action Permit shown as eligible. Even in the Lab, permit eligibility is separate from credential release, execution, evidence, and consequence proof.
Four laboratories
03 / SAME CONSTITUTION, DIFFERENT CONSEQUENCE
Authority should remain legible when the action changes domains.
A changed beneficiary can invalidate yesterday's comfort.
Explore amount and destination scope, approval, reserve, JIT payment capability, intervention posture, and synthetic consequence evidence. The full changed-beneficiary narrative arrives in SIO12.
Cloud credentials are capability, not permission.
Move a production restart outside scope, freeze recovery capacity, or revoke delegation and watch technical reach remain inert behind the authority boundary.
An ordinary API can create an extraordinary business consequence.
Role changes, data access, workflow approvals, procurement actions, and records can be bounded even when the underlying SaaS API exposes broad administrator capability.
Authority does not automatically survive an organizational boundary.
Partner capability, bilateral trust, and a reachable endpoint do not create the other enterprise's acceptance authority. The corridor must remain explicit and bounded.
What this demonstrates
A control plane should make disagreement inspectable.
The useful part of the Lab is not the green PERMIT state. It is watching one changed fact produce a smaller, human-routed, or denied action while the rest of the technical stack remains available.
Requested power versus allowed power.
The public output separates what the agent asked to do from what the synthetic authority state permits. NARROW is not a friendly warning; it is a different executable boundary.
Authorization precedes capability.
The Lab shows credential eligibility only after a PERMIT or NARROW outcome. It never treats an existing token, vault secret, cloud role, or integration as evidence of authority.
Intervention remains independent.
Guardian posture is displayed separately from ActionGate's decision. The intervention plane can preserve or reduce authority effects; it cannot upgrade a denied request into an authorized one.
Proof remains downstream.
The synthetic evidence selector teaches VERIFIED, PARTIAL, MISMATCHED, and UNVERIFIABLE states and points toward Proof of Consequence. A favorable proof state does not grant authority, and missing evidence never silently becomes verified.
For evaluators
04 / INTERACTIVE MACHINE AUTHORITY
Questions the Authority Lab is designed to make easier to ask.
- Is the Authority Lab connected to a real ActionGate deployment?
- No. SIO11 is a public synthetic teaching environment. It performs deterministic browser-local evaluation only and has no customer, account, credential, execution, reserve, or production-proof effects.
- Is this an AI agent authorization simulator?
- Yes, in the narrow sense that it simulates how selected authority facts change a deterministic authorization outcome. It does not simulate model reasoning and it does not substitute for a configured enterprise policy.
- Why can a request be narrowed instead of simply approved or denied?
- Because a valid grant may authorize less power than the agent requested. A machine-authority control plane can return an exact smaller boundary rather than forcing a false binary between unrestricted approval and complete denial.
- Why does a changed target require human authority?
- Because a beneficiary, destination, resource, counterparty, or other consequential target can be part of the delegated authority itself. Changing that target can change the substance of the power being exercised.
- Does VERIFIED evidence mean the request should have been authorized?
- No. Proof is downstream of authority. A later favorable synthetic proof result cannot retroactively create authority for an action that lacked it when requested.
- Can I use the Lab to evaluate my own production policy?
- Not yet. SIO11 is an educational public surface with fixed synthetic controls. Customer-controlled shadow evaluation belongs later in the onboarding and deployment program, with account activation beginning at SIO19.
Public truth boundary
Interactive does not mean production.
The Authority Lab is intentionally vivid, but every value is synthetic and every decision remains inside the browser. It does not prove that StableMind has authorized, executed, interrupted, verified, or settled a customer action. It does not manufacture named customers, production evidence, recognized revenue, or realized customer value.
content_state=INTERACTIVE_SYNTHETIC_AUTHORITY_LABbrowser_local=truenetwork_calls=0authority_effects=0execution_effects=0external_production_proof_claims=0public_signup_active=false
From understanding to adoption
The public Lab teaches the boundary. ActionGate is where the customer journey begins.
SIO11 lets a visitor manipulate synthetic authority facts without connecting infrastructure. Commercial account bootstrap is introduced in SIO19; SIO21 remains the first governed-action experience. Until then, the Lab is deliberately powerless in the real world.
website_session_creates_authority=false · lab_network_effects=0 · public_signup_active=falseFollow the outcome
Inspect the downstream proof.
After experimenting with synthetic authorization, open the Proof Explorer to see why a successful execution receipt still does not self-certify the consequence.