Giving AI and agentic systems a way to call on human judgment and authority when needed.
View Technical Specification
We created PAVE to give AI and agentic systems a way to call on human judgment and authority when needed. It represents human judgment as a protocol so it can be requested and applied within machine workflows while preserving its connection to the person it came from, the authority associated with it, and the conditions under which it can be used. Learn more about Callable Human Judgment
What PAVE does
Makes human judgment callable.
Represents human judgment so AI and agentic systems can request and apply it within a defined scope.
Works with decision agents.
Gives systems that make decisions a way to call on relevant human judgment and authority as part of the decision process.
Adds a layer before human-in-the-loop.
Lets systems call on represented human judgment first, while preserving escalation to a person when direct human participation or authority is required.
Keeps judgment connected to its source.
Preserves who the judgment came from, the authority associated with it and the provenance needed to understand its origin.
Carries permission and scope with the judgment.
Separates access to someone's judgment from permission to act, and defines the conditions under which represented judgment can be used.
Makes human contribution accountable.
Creates a record of how judgment is used, providing a basis for attribution, valuation and eventually compensation.
PAVE, Explained
Four short films exploring human judgment in AI systems, how PAVE works, and what becomes possible.
From Answers to Actions
AvailableDemonstrations
Recording to come
A manufacturer is considering a $480,000 supplier commitment that would reduce costs by 9% but concentrate a critical component with a single production site. The supplier has a recovery plan, but it has not yet been tested.
The company already uses OpenAI’s Decisions API to classify and route agent decisions. Its procurement policy requires human judgment when a sourcing exception introduces material continuity risk.
PAVE discovers the relevant procurement judgment capability through the Human Judgment MCP, invokes it, and returns a structured assessment to the workflow. The judgment identifies the missing continuity evidence and recommends that the system request additional validation before the supplier commitment proceeds.
The machine decision and the human judgment remain distinct — and the workflow can use both.
Callable Human Judgment & Human-in-the-Loop
Two ways of bringing human judgment into machine workflows.
| Aspect | Callable Human Judgment | Human-in-the-Loop |
|---|---|---|
| When | Callable Human JudgmentJudgment can be called during machine execution without requiring the person to be present. | Human-in-the-LoopA person enters the workflow at a defined point to review, decide, approve or intervene. |
| What the system calls | Callable Human JudgmentA represented contribution derived from a person's judgment. | Human-in-the-LoopThe person directly. |
| Scale | Callable Human JudgmentThe same represented judgment can be available across permitted machine workflows. | Human-in-the-LoopHuman participation is required each time the workflow reaches the intervention point. |
| Authority | Callable Human JudgmentAccess to represented judgment and permission to act remain separate. | Human-in-the-LoopThe person can exercise authority directly when they enter the workflow. |
| Escalation | Callable Human JudgmentCan precede human-in-the-loop and escalate when direct participation or authority is required. | Human-in-the-LoopProvides the direct human intervention when the workflow requires it. |
How PAVE Works
PAVE sits between a system's request and the human judgment relevant to it. It determines what represented judgment is available, whether it can be used in that context, and returns its application with the provenance needed to understand where it came from.
Permission to use someone's judgment and authority to act remain separate.
1. Request
A system encounters a decision where human judgment may be relevant.2. Discover
PAVE identifies represented judgment relevant to the request.3. Check permission and scope
PAVE determines whether that judgment can be used for this request and under what conditions.4. Apply judgment
The permitted judgment is applied within the defined scope.5. Return result and provenance
The calling system receives the result together with the information needed to trace the contribution to its human source.6. Decide or escalate
The system can use the result within its own authority, or escalate to a person when direct human participation or authority is required.Where PAVE can be used
PAVE becomes useful when a machine can make a decision, but policy or automation alone does not capture the judgment the situation requires.
Procurement
Supplier decisions, exceptions and commitments where organizational judgment or authority matters.
Credit
Risk, exceptions and lending decisions where scoring or policy alone may not resolve the decision.
Insurance and claims
Claims, eligibility and exceptions where contextual judgment may be required beyond established rules.
Healthcare
Clinical and operational decisions where machine recommendations intersect with professional judgment or authority.
Workforce
Hiring, evaluation and employment decisions where human judgment, escalation or authority needs to remain explicit.
Compliance and risk
Decisions where something may satisfy formal policy but still requires contextual judgment about risk, exceptions or appropriate action.
Interfaces & Integrations
Where Agents Work
| Integration | Type | Status | Note |
|---|---|---|---|
| MCP | Protocol | Available | Discovery and inspection of synthetic judgment resources demonstrated in a bounded sandbox. Live judgment execution and public access are not established. |
| OpenAI | Agent platform | Available | Private ChatGPT discovery and inspection of synthetic resources demonstrated when Human Judgment is selected. Live execution is not established; Decisions integration remains planned. |
| Agentforce | Enterprise platform | Planned | Salesforce Agentforce adapter planned. Platform integration and conformance testing have not started. |
| NVIDIA | Ecosystem / framework | Planned | NeMo ecosystem adapter planned. Component and version selection, integration and conformance testing have not started. |
Technical Documentation
Bring PAVE into your own environment.
PAVE is being developed for organizations building AI and agentic systems where human judgment, authority and provenance need to remain part of how decisions are made.
Questions that follow
Is PAVE the same as human-in-the-loop?
No. Human-in-the-loop brings a person directly into a workflow. PAVE allows represented human judgment to be called without requiring the person to be present for every use. The two can work together, with PAVE preceding escalation to direct human participation when needed.
Does PAVE give an AI system authority to act for a person?
No. Access to someone's represented judgment and authority to act are separate. PAVE is designed to preserve that distinction.
Whose judgment is being represented?
Judgment remains connected to the person it came from, along with the provenance, permission and scope under which it can be used.
Can the same judgment be used by more than one system?
Potentially, where the permissions and scope attached to that represented judgment allow it. This is part of what makes represented judgment different from requiring the person to participate in every workflow directly.
How does PAVE work with decision agents?
A decision agent can call on relevant represented human judgment as part of its decision process. PAVE can return that judgment and its provenance without collapsing the agent's decision or authority into the human contribution.
Is PAVE open source?
No. PAVE is proprietary technology developed by FRUIT. Selected technical materials are publicly available so its architecture, boundaries and development can be understood and referenced.
Can organizations integrate or license PAVE?
FRUIT is working with organizations exploring how human judgment and authority can be incorporated into AI and agentic systems.