Decision Graph Protocol

A decision graph protocol is a shared standard for expressing, executing, and verifying decisions as connected graphs — so that any system, human auditor, or AI agent can see exactly which inputs, rules, and authorities produced a given outcome.

What is a decision graph?

A decision graph is an explicit, executable representation of how an organization decides. Nodes represent the elements of a decision — observed inputs, policy rules, thresholds, required authorities, possible outcomes — and edges represent the logic connecting them: "if contract value exceeds this threshold, route to this approver; if the counterparty is on this list, halt."

Unlike a flowchart or a policy document, a decision graph is not a description of decision-making — it is the machinery of decision-making. Software traverses the graph to reach outcomes, and every traversal leaves a record. That makes decision graphs the natural substrate for governed AI: when an agent acts, its action can be constrained by, and recorded against, the same graph a human decision-maker would follow.

What does the "protocol" part add?

A protocol is what turns one company's internal decision graph into infrastructure the whole ecosystem can rely on. Without a shared standard, every vendor encodes decision logic in a proprietary format: the insurer's underwriting rules cannot be read by the reinsurer's systems, the bank's credit policy cannot be verified by the regulator's tooling, and migrating platforms means re-authoring every rule by hand.

A decision graph protocol defines the common grammar: how nodes and edges are serialized, how a graph declares the authorities it requires, how an execution engine proves it followed the graph faithfully, and how traces are signed and exchanged. HTTP did this for documents and TCP/IP for packets; a decision graph protocol aims to do it for consequential choices.

Why do governed decisions matter now?

Because AI agents have crossed the line from recommending decisions to making them. Agents now approve refunds, triage claims, screen vendors, and adjust pricing. Each of those actions used to pass through a human who could be asked "why?" — and who operated inside an implicit web of policy. When the actor is a model, both the "why" and the policy boundary vanish unless they are engineered explicitly.

Regulators have noticed. The EU AI Act, sector guidance in financial services and healthcare, and ordinary procurement questionnaires increasingly ask for evidence: who authorized this action, under which rule, with which data. An organization that executes decisions through a graph answers those questions by querying a trace. An organization that lets agents act directly answers them with a postmortem.

How does a decision graph protocol compare to existing tooling?

Each row is useful; none is a substitute. Workflow engines move work, rules engines evaluate predicates, logs record activity. A decision graph protocol is the layer that makes the decision itself — not just the motion around it — a first-class, verifiable object.

What decisiongraphprotocol.com is building

decisiongraphprotocol.com is an early-stage property in build by ROIZILLA, LLC. The planned work is a published protocol specification: node and edge types, authority declarations, execution semantics, and trace formats, released openly for implementers. None of that is live today — this page is an educational reference on the concept, and no products or services are offered yet.

Frequently asked questions

Is a decision graph just a flowchart?

No. A flowchart documents how a process is supposed to work; a decision graph executes it. Software traverses the graph to make real decisions, and every traversal produces a verifiable record.

Does the protocol lock you into one vendor?

The opposite. A protocol exists so that a graph authored in one system can be executed and audited in another. Proprietary decision tooling without a shared protocol is the lock-in.

How does it relate to AI agents?

Agents propose; the graph disposes. The agent gathers context and recommends, but the graph defines required authorities, escalation thresholds, and the evidence that must be recorded.

What is a decision trace?

The record of one execution: inputs observed, rules evaluated, authority invoked, outcome chosen. It turns "why did the system do that?" from an investigation into a lookup.

Is this the same as model explainability?

No. Explainability interprets a neural network's internals. A decision graph protocol governs the explicit logic around the model — where most auditable decision-making actually lives.

Who operates decisiongraphprotocol.com?

decisiongraphprotocol.com is owned and operated by ROIZILLA, LLC. It is a property in build: the protocol specification is planned, and this page currently serves as an educational reference.