StableMindMachine AuthorityLive Authority Graph Explorer · Synthetic and browser-local
Machine authority should not exist as a vague property attached to an agent. It should have an attributable source, a bounded path, conserved limits, a current validity state, and an inspectable relationship to every decision, permit, consequence reserve, execution, intervention, and proof artifact that follows.
Synthetic graph: every principal, agent, delegation, request, reserve, permit, execution, intervention, and evidence object below is fictional. The explorer runs entirely in your browser and creates zero machine authority.
Authority is not inferred from identity, credentials, organizational familiarity, or technical reach. The path must still be valid now.
One graph · Four deterministic states
The explorer uses a fixed synthetic lineage. An accountable principal delegates payment authority to a Treasury Agent, which creates a narrower child delegation for a Payment Agent. The request then crosses ActionGate, consequence-capacity checks, permit issuance, bounded execution, Guardian posture, and Evidence. Change the request or lineage state and downstream eligibility changes with it.
The accountable synthetic principal is the origin of this lineage. Its presence does not authorize an action by itself; the authority must travel through explicit current delegation.
What the graph proves
StableMind does not treat an authority graph as a picture of who is connected to whom. Each edge must describe a governed relationship that can be evaluated at decision time. A reachable node is not necessarily an authorized node, and a historical delegation is not necessarily a current delegation.
Machine authority begins with an accountable grantor or governed institutional source. The graph preserves who delegated power, to whom, for what purpose, over which resources, destinations, limits, validity period, and inherited conditions.
A child grant may preserve or narrow parent authority. It cannot expand action scope, amount, destination, time, consequence capacity, recursive depth, or inherited denial conditions. Shared budgets are conserved rather than duplicated across descendants.
Revocation, expiry, quarantine, signature failure, lifecycle supersession, or a missing ancestor can invalidate descendants. StableMind therefore resolves the current chain, not merely the existence of a historical edge in a database.
A valid graph is still only an input to ActionGate. It does not issue credentials, execute actions, reserve consequence capacity, or create Proof of Consequence. Exact authorization happens against the semantic request.
Four graph states
These synthetic states deliberately mirror the four ActionGate outcomes. They are teaching states, not claims about a customer deployment.
The request is $188,000 to the beneficiary ending 7744. The child grant allows up to $250,000 to that exact destination, the ancestor is current, and sufficient synthetic consequence capacity is available. A bounded permit can become eligible.
The request rises to $428,000 while the current child delegation remains capped at $250,000. The parent may have more capacity, but the child cannot silently borrow it. ActionGate narrows the allowed action to the current child boundary.
The amount returns to $188,000, but the beneficiary changes from ending 7744 to ending 9138. The agent, credential, and supplier label may remain familiar; the current lineage no longer proves authority for that destination, so fresh human authority is required.
The CFO Office revokes DLG-001. DLG-002 cannot remain independently alive beneath a dead ancestor. The descendant lineage becomes invalid, no Action Permit is eligible, and downstream technical capability stays inert.
The objects inside machine authority
AUTHORITY IS A LINEAGE, NOT A BOOLEAN
Human principals, governed organizations, trusteeship structures, or other authorized sources anchor why a machine possesses bounded power in the first place.
Agent identity lets the graph attribute an actor. It does not grant permission. The same authenticated agent may have different authority under different current delegations.
Delegation objects carry the machine-evaluable boundaries that ActionGate can later compare with an exact semantic request.
Where required, the graph can show bounded consequence capacity associated with a lineage while keeping reserve coverage constitutionally distinct from permission.
Action Permits bind a decision to a narrow action and time window. Execution Fabric then materializes only the capability that exact permit requires.
Revocation must propagate through dependent authority, and every revocation dependency should remain inspectable. Evidence preserves what was requested, decided, permitted, executed, interrupted, and independently observed afterward.
Public Truth boundary
The explorer intentionally resembles a product control plane because machine authority is easier to understand when relationships can be inspected. But the data is embedded synthetic teaching data. Selecting a scenario or node changes only local presentation state.
Authority graph FAQ
INSPECT THE RELATIONSHIP, NOT JUST THE NODE
No. Identity graphs answer valuable questions about principals, workloads, groups, roles, credentials, and relationships. An authority graph adds a different question: which consequential power was explicitly delegated, through which current lineage, for which purpose and scope, under which constraints, and with what revocation dependencies? StableMind uses identity as evidence about the actor while keeping permission to create consequence separately governed.
No. A current delegation path establishes authority context that ActionGate can evaluate against an exact semantic request. The request may still exceed amount, destination, resource, time, approval, consequence-capacity, or other boundaries. That is why the graph continues through the decision rather than ending at the agent node. A valid lineage can coexist with NARROW, HUMAN_AUTHORITY, or DENY for a particular request.
Because downstream artifacts should remain attributable to the authority that justified them without being confused with that authority. An Action Permit should point back to the request, decision, and current delegation context that made it eligible. Execution and evidence can then preserve a lineage forward. This lets reviewers inspect an end-to-end chain without pretending that a permit, credential, receipt, or successful execution retroactively creates permission.
Revocation is not simply a red label on one node. If a descendant depends on a revoked ancestor, the dependent path must become invalid according to explicit inheritance rules. Technical capability that still exists downstream should not keep the authority alive. Guardian may help contain residual capability, while ActionGate and Execution Fabric continue to fail closed on invalid current lineage.
StableMind does not assume that two narrow grants automatically combine into one broad grant. Composition can create surprising authority, especially when one grant supplies destination scope and another supplies economic capacity. Unless a governed composition rule explicitly permits that union, independent grants remain independent. The graph therefore needs enough semantics to show why a path exists, not merely that two nodes are connected.
It proves only that StableMind.io can explain the architecture through a deterministic browser-local teaching model. It does not prove a named customer deployment, live authority lineage, production revocation, governed transaction, external Proof of Consequence, realized savings, revenue, or customer outcome. StableMind's Public Truth contract deliberately prevents an interactive interface from promoting synthetic demonstration into external production evidence.
For architects, security teams, and investors
Every consequential machine action should be traceable to a current, attributable source of authority rather than a role name or standing credential.
If recursive delegation exists, attenuation must be enforced across every dimension that matters, including shared budgets and inherited conditions.
Revocation and expiry should invalidate dependent authority quickly and predictably, without trusting descendant agents to voluntarily surrender technical capability.
Independent grants should not be unioned into accidental super-authority unless a governed composition rule explicitly permits it.
A vault token, API reach, administrator role, service account, or successful connector test is technical capability. None of those facts should create machine authority.
Decision and evidence systems should preserve enough attributable lineage to reconstruct why an action was permitted, narrowed, human-routed, denied, executed, interrupted, or left unexecuted.
Continue the lineage
The Proof Explorer continues the inspectable lineage into request hashes, Action Permit, execution receipt, independent verifier observation, and final Proof of Consequence.