• CAUSATION
  • REPORTS
  • AUDIT
  • HUMAN

How to validate performance claims with your AI

Connect Curie Measurement Prover once. Your agent then moves from making guesses to proving causation with hard data.

Ask AI about this page

Short answer

How can I make my AI prove its performance claims?

You use the validator to force your agent to identify actual causal drivers instead of just spotting correlations. This turns a vague observation into a rigorous report. You'll get a breakdown of the evidence used to support the claim.

The results of validation.

What you get from a validated claim.

The agent processes your data to produce these specific outcomes.

CAUSATION

Proven causal links

The agent identifies the specific variables that actually drove the change. You get proof of what worked and what didn't.

REPORTS

Rigorous data reports

The agent compiles its findings into a structured format. These reports provide the evidence needed for stakeholders.

AUDIT

Clear audit trails

The agent logs every step it takes to verify a claim. You can see exactly how the agent reached its conclusion.

HUMAN

Human decision points

The agent flags weak correlations or statistical noise. You decide if the evidence is strong enough to act on.

The workflow

What your AI does when a claim arrives.

The agent stops guessing and starts measuring.

  1. Isolate variables

    The agent identifies the specific process changes and the corresponding metrics in your dataset.

    validate_curie_measurement

  2. Test for causation

    The agent runs statistical checks to see if the change actually caused the result or if it was just a coincidence.

    validate_curie_measurement

  3. Calculate significance

    The agent determines if the observed improvement is statistically significant and not just random noise.

    validate_curie_measurement

  4. Generate proof

    The agent builds a final summary that links the action to the outcome using the data it analyzed.

    validate_curie_measurement

Try it

Copy these to start.

Paste these directly into your chat to begin validating.

Starting points

Swap the bracketed text for your specific process or dataset names.

Curie Measurement Prover Connector

You're all set. Choose your MCP client and follow the setup instructions.

Connector linkhttps://edge.vinkius.com/vk_preview_CPb0xivIrSLYBuuZ7KzsZrT7gj3fFKipKNBJYSPo/mcp

Claude Desktop

Follow the steps below to connect in seconds.

  1. 1In Claude Desktop, open Settings → Connectors.
  2. 2Click “Add custom connector” and paste the connector link above as the remote MCP server URL.
  3. 3Click Add and start a new chat — Curie Measurement Prover capabilities are ready to use.
Configuration · claude_desktop_config.jsonCopy
{
  "mcpServers": {
    "curie-measurement-prover-mcp": {
      "url": "https://edge.vinkius.com/vk_preview_CPb0xivIrSLYBuuZ7KzsZrT7gj3fFKipKNBJYSPo/mcp"
    }
  }
}
Copy into chat04

  • Validate the claim that the new warehouse layout improved picking speed in the logistics sheet.

  • Prove whether the recent software update caused the drop in latency reported in the system logs.

  • Check if the change in training protocols actually led to the higher accuracy scores in the QA database.

  • Verify the causation between the new marketing spend and the conversion rate increase in the sales report.

  • Claude
  • ChatGPT
  • Cursor
  • VS Code
  • Windsurf
  • Claude Code
  • JetBrains
  • Cline

Start here

Connect Curie Measurement Prover once, then ask.

One link connects your agent to the measurement tools. Your credentials stay encrypted on our side.

Connect Curie Measurement Prover to your AI

FAQ

How this task behaves

  • 01

    Can the AI change my original data?

    No. The agent only reads your data to perform measurements; it cannot edit or delete your source files.

  • 02

    Does the AI decide if a claim is true?

    No. The agent provides the statistical evidence and the causal link, but you make the final call on whether to accept the claim.

  • 03

    What kind of data can it validate?

    It works with any structured data you can provide through your agent, such as logs, spreadsheets, or database exports.

  • 04

    Can it run these tests automatically every day?

    No. The agent runs the validation when you prompt it to do so. It does not run on a schedule.

  • 05

    What happens if the data is too messy to prove causation?

    The agent will tell you it cannot find a statistically significant link and will explain why the data is insufficient.

  • More questions about Curie Measurement Prover? The Connector page answers them. See everything the Curie Measurement Prover Connector can do

Connect Curie Measurement Prover to Claude, Cursor, ChatGPT & more