Vector Decisions.
Decision infrastructure for the moment between intelligence and action.
AI can generate an answer. An agent can propose an action. Vector Decisions asks the harder question: is this action authorized, supported, appropriate, and safe under these conditions?
The first primitive is deliberately small and deterministic. It evaluates identity, delegated scope, authority windows, evidence, policy, consequence and human-gating rules, then produces a decision trace.
From context to governed action.
Not every answer is “yes” or “no.”
A governed system can restrict scope, demand a human, escalate to another authority, route to a different capability, or substitute a safer action.
Run the first decision.
Should this agent be allowed to act?
Change the inputs. Vector Decisions evaluates authority, evidence, policy, consequence and context together.
The output is a governed decision, not a generated answer.
Run the example to produce a deterministic decision trace.
The system underneath.
GATRI and GIRL remain separate architectural directions. This page establishes the decision primitive first, so later assurance and routing systems have a stable vocabulary.
Not an “AI trust score.”
Not an LLM pretending to be a policy engine.
Not an autonomous execution gateway yet.
Not a replacement for security, legal, compliance, or human accountability.
This is the first public proof of the idea: authorization should be evaluated at the point of action, using explicit context, evidence, authority and policy.
GATRI turns decision into assurance.
The next stage is infrastructure that can attest that the authority used by a decision was valid, scoped, current and attributable — and eventually make that evidence enforceable at execution time.
GATRI architecture →