Research Lab — Fractal Process Intelligence (Research Alpha)

Why this exists

The final answer hides the process.

Modern AI systems can use multiple agents, tools, searches and sources while producing one polished final answer.

The answer alone does not show whether those processes added independent evidence or simply repeated the same information.

Fractal records observable research activity and produces an X-Ray of the process behind the answer.

Status

Research Alpha

Operational MCP recorder and X-Ray pipeline are live.
Active question Does process inspection consistently reveal consequential information that a final AI answer hides?
Not yet This is not a validated commercial product.

What Fractal inspects

Process structure — not provider billing.

Evidence

Independence & lineage

Independent evidence lineages
Source correlation and citation cascades
Duplicate investigation

Claims

Support & disagreement

Claim support and uncertainty
Meaningful disagreement
Unique agent contribution

Coverage

Gaps & next look

Missing research branches
Whether another independent investigation is likely to add value
What the final answer hid

Fractal does not include model-provider token cost as a Fractal metric.

How it works

Your AI keeps working. Fractal observes.

Your AI
  ↓
Fractal MCP recorder
  ↓
Observable research evidence
  ↓
Fractal X-Ray

Fractal does not replace the user's AI model. Continue using Claude or another compatible environment. Fractal observes explicitly recorded research-process artifacts. It does not require or depend on hidden chain-of-thought.

How to test

Connect → research normally → open the X-Ray cold.

Connect Fractal Process Intelligence to an MCP-compatible AI client.
https://mcp.fractalinfinity.io/mcp
Start a new research conversation.
Tell the AI to use Fractal for the investigation (copy the block below).
Ask your normal research question.
Read the AI's final answer first.
Open the Fractal X-Ray URL.
Ask: “What did the X-Ray reveal that I would not have known from the final answer alone?”

Copyable instruction for the AI

Use Fractal Process Intelligence to record this research run.

Record the participating agents, meaningful sources, important claims,
uncertainty, disagreement and significant tool usage.

Finish the run and return the Fractal X-Ray URL.

Do not interpret the X-Ray for me before I open it.

Windows / Claude Desktop

Alpha setup (current testing method)

This is the current testing path — not final onboarding. Do not overwrite other existing Claude configuration; merge the MCP entry.

Show Claude Desktop config steps

Claude Desktop → Settings → Developer → Edit Config
Add the MCP server entry below (preserve other keys).
Save.
Fully quit Claude Desktop (system tray → Quit).
Reopen Claude.
Settings → Developer should show fractal-process-intelligence running.

{
  "mcpServers": {
    "fractal-process-intelligence": {
      "command": "C:\\Program Files\\nodejs\\npx.cmd",
      "args": [
        "-y",
        "mcp-remote",
        "https://mcp.fractalinfinity.io/mcp"
      ]
    }
  }
}

Microsoft Store Claude may store this under the package LocalCache path. Prefer Edit Config from Claude so the correct file opens. Use the full npx.cmd path on Windows.

FAQ

Want to use Claude normally?

Leave Fractal installed. Fractal does not automatically record every Claude conversation.

For a normal conversation, do not start a Fractal run and do not invoke its recorder tools. Where Claude exposes per-conversation tool/connector controls, you may also disable Fractal for that conversation.

Fractal records research only when its tools are explicitly used.

Transparency

What Research Alpha does not do

Replace your AI model
Buy model inference on your behalf
Automatically spawn additional agents
Inspect hidden chain-of-thought
Estimate your Claude/OpenAI provider bill
Claim every citation is independent merely because it has a different URL

Current research priorities

Restrained roadmap

Correct semantic integrity of the X-Ray.
Test multi-agent research.
Make “What the final answer hid” the core inspection output.
Test with serious AI builders and research-heavy users.
Determine whether users say “I need this,” rather than merely “this is cool.”
Adaptive investigation remains future research and is not active.

As agent systems scale, more agents and more searches do not necessarily mean more independent intelligence. Fractal is investigating how to measure the difference.