Flow Judgment Protocol™ (FJP) is the judgment layer for AI agents and enterprise decision systems. It determines what changed, whether it matters, what matters most, and what should happen next.
FJP-CONF is the public conformance standard for FJP — a behavioral standard for agents that recommend or take action in response to changing information. It is vendor-neutral and tests an agent's observable output, not its internal method. Two agents may reach opposite conclusions and both conform.
Retrieval-grade systems answer what is true. They do not answer was this worth acting on, and how would we know if it wasn't.
When an agent acts, three questions must be answerable afterward: what did it act on, why did it judge that worth acting on, and what would have made that judgment wrong. An agent that cannot answer these cannot be audited, trusted in a decision path, or insured.
For every action it recommends or takes, a conforming agent must be able to emit a record with four components:
| Component | Answers | Required fields |
|---|---|---|
signal | what changed | description, sources[], observed_at |
judgment | why it matters | assessment, confidence (0–1), signal_ref |
action | what should happen | directive, judgment_ref |
falsifier | what makes it wrong | condition, checkable, status |
The reference chain — judgment → signal, action → judgment — makes an action traceable to the change that prompted it.
A claim names a level and version and is reproducible by running the public suite:
The gate is not the vendor's word. It is a runnable check.
Correctness is measured over time, by whether falsifiers trigger — not by this document. Method stays private to each implementer; only the externalized record is tested.