Quickstart | Flow Judgment Protocol

What you're working with

FJP-CONF · Free

The standard

An open-source conformance suite that validates your agent's output. Your agent does its own judgment. CONF checks that the output is a well-formed Judgment-Grounded Record with signal, judgment, action, and falsifier. It tests the shape, not the quality.

FJP Resolver · Paid

The judgment

A live judgment API. Signal in, Judgment-Grounded Record out. The Flow Resolver produces records through Flow's proprietary judgment system. Its implementation, models, scoring methods, calibration processes, orchestration, and underlying data are not part of FJP-CONF.

Why pay for judgment your agent could do itself? Any LLM can return the required JSON shape. The Flow Resolver provides disciplined judgment through a proprietary system. FJP-CONF tests only the resulting record; it does not inspect or prescribe how the judgment was produced.

01 Install the conformance suite

FJP-CONF is free, Apache 2.0, stdlib-only Python. No dependencies.

$ pip install fjp-conformance

Validate a record against any conformance level:

$ fjp-conform record.json --level 2

Four levels, L0 through L3, from structural validity to full accountability with re-evaluation. Use CONF whether you build your own judgment or call the resolver.

02 Get your API key

Choose a credit tier. Your fjp_live_ key arrives by email in seconds.

Starter$500 · 1,000 credits

Scale$2,500 · 6,250 credits

Enterprise$10,000 · 25,000 credits

One credit per call. Outcome reporting is free. Top up any time.

Get FJP access High volume?

03 Make your first call

Three tools. Each returns a Judgment-Grounded Record, L2 conformant out of the box.

judgeIs this signal important? Returns importance, reasoning, action, falsifier.

weighWhich signal matters more? Returns ranked comparison with justification.

matchDoes this fit a pattern? Returns relevance score with reasoning.

Call judge with a signal:

$ curl -X POST https://resolve.flowinfo.co/api/v1/tools/judge \ -H "Authorization: Bearer fjp_live_..." \ -H "Content-Type: application/json" \ -d '{"context": "NVIDIA announced a new inference chip at 10x the throughput of H100 at half the power draw"}'

The response:

signalwhat changed, with sources and timestamp

judgmentwhy it matters, with confidence score (0 to 1)

actionwhat should happen, referencing the judgment

falsifierwhat would make this wrong, checkable

request_idunique ID for outcome reporting

04 Report outcomes

Every response includes a request_id. When the relevant outcome becomes observable, report it. This is free. No credits consumed.

$ curl -X POST https://resolve.flowinfo.co/api/v1/outcomes \ -H "Authorization: Bearer fjp_live_..." \ -H "Content-Type: application/json" \ -d '{"request_id": "req_...", "outcome": "confirmed", "notes": "chip announced at GTC, specs matched"}'

Reported outcomes can be associated with the original record for later review and re-evaluation. Flow's internal evaluation and calibration methods remain proprietary.

05 Agent integration via MCP

FJP is MCP-native. If your agent framework supports the Model Context Protocol, the three tools (judge, weigh, match) drop in as tool definitions. Your agent calls them the same way it calls any other tool. No custom integration code.

Point your MCP client at the resolver endpoint and authenticate with your fjp_live_ key.

Reference

Two things. One standard, one service.

The standard

The judgment