• VALID
  • HALUCINATED
  • CONTRADICTION
  • HUMAN

How to validate logic puzzle variations with your AI

Connect this MCP to your agent to stop it from relying on memorized solutions. It forces your AI to actually process the specific constraints you've added to a riddle.

Ask AI about this page

Short answer

Can my AI solve riddles if I change the rules?

Yes, if you stop it from relying on its training data. This MCP runs a validation check to ensure your agent treats your modified constraints as the ground truth. You'll see exactly where its logic breaks or holds up.

The results of a check.

Where your logic results land.

Once you run a validation, your agent categorizes the findings into these four buckets.

VALID

Logic holds up

The AI successfully ignored the classic solution and followed your specific constraints. You get a confirmation that the reasoning is sound.

HALUCINATED

Classic bias detected

The agent defaulted to a standard riddle answer instead of your new rules. This identifies where the model's training overrode your prompt.

CONTRADICTION

Rule conflict found

The AI found a logical impossibility between your new rules and the puzzle's setup. This highlights flaws in the puzzle's construction.

HUMAN

Manual review needed

The logic is too ambiguous for the tool to verify automatically. You'll need to step in and refine the constraints or the puzzle wording.

The workflow

What your AI does when a puzzle arrives.

The agent uses the specialized tool to strip away the AI's assumptions.

  1. Isolate constraints

    The agent extracts your specific rule changes from the prompt to create a clean set of counterfactual parameters.

    validate_counterfactual

  2. Test against defaults

    The tool compares the agent's proposed solution against the standard version of the puzzle to see if it's just repeating old data.

    validate_counterfactual

  3. Verify logical flow

    The agent checks every step of its reasoning against the isolated constraints to ensure no rules were skipped.

    validate_counterfactual

  4. Flag failures

    The agent identifies if the solution is a hallucination or a genuine logical error based on the validation results.

    validate_counterfactual

Try it

Copy these to start.

Use these prompts to begin testing your logic puzzles.

Starting points

Replace the bracketed text with your specific riddle and the new rules you've added.

Counterfactual-Variant Prover Connector

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

Connector linkhttps://edge.vinkius.com/vk_preview_DsoSPwZEWSeSBTBSQCX8U58BqAbjy82WIsWBk6Wg/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 — Counterfactual-Variant Prover capabilities are ready to use.
Configuration · claude_desktop_config.jsonCopy
{
  "mcpServers": {
    "counterfactual-variant-prover-mcp": {
      "url": "https://edge.vinkius.com/vk_preview_DsoSPwZEWSeSBTBSQCX8U58BqAbjy82WIsWBk6Wg/mcp"
    }
  }
}
Copy into chat04

  • Validate this riddle with these new rules: [Insert riddle and modified constraints here]

  • Check if my solution to this puzzle follows the counterfactual rules I just gave you: [Insert puzzle]

  • Run a validation on this logic puzzle to see if it's defaulting to the classic answer: [Insert puzzle]

  • Does my agent's reasoning for this modified riddle violate any of my new constraints? [Insert puzzle and rules]

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

Start here

Connect Counterfactual-Variant Prover once, then ask.

Just click the link to connect the MCP to your client. Your credentials stay encrypted, and you can start testing logic immediately.

Connect Counterfactual-Variant Prover to your AI

FAQ

How this task behaves.

  • 01

    Can the AI rewrite the riddle for me?

    No. The tool only validates the logic based on the rules you provide; it doesn't generate new puzzle content.

  • 02

    Will this fix my AI's tendency to hallucinate?

    It doesn't change how the AI thinks, but it gives you a way to catch when it's hallucinating or using old data.

  • 03

    Does it work with any logic puzzle?

    Yes, as long as you can clearly define the counterfactual rules or changes you want to test.

  • 04

    Can the tool automatically correct the AI's answer?

    No. The tool identifies the error, but you or your agent must decide how to fix the reasoning.

  • 05

    Does it store my custom riddles?

    No. The validation happens in the moment, and the tool does not save your puzzle data.

  • More questions about Counterfactual-Variant Prover? The Connector page answers them. See everything the Counterfactual-Variant Prover Connector can do

Connect Counterfactual-Variant Prover to Claude, Cursor, ChatGPT & more