• LOGS
  • INSIGHTS
  • HUMAN
  • RECOVERY

How to document experimental failures with your AI

Connect your agent to a dedicated experimentation log. It stops being a chatbot and starts acting like a researcher that tracks what didn't work.

Ask AI about this page

Short answer

How can I make my AI track failed experiments?

Your agent uses the Edison Experimentation Prover to record the specific reasons a prompt or logic path failed. It builds a structured history of rejected ideas so you don't repeat the same mistakes. You get a clear map of what to avoid next time.

Where findings land.

The outcomes of your research.

The AI organizes every failed attempt into actionable data.

LOGS

Root cause records

The AI writes down exactly why a specific path failed. You get a clean history of logic errors and prompt mismatches.

INSIGHTS

Pattern recognition

The agent looks for recurring failures across multiple runs. It identifies if a specific tool or instruction is consistently causing issues.

HUMAN

Decision points

The AI flags complex failures that require your intuition. It tells you when a failure is too nuanced for automated logging.

RECOVERY

Next step suggestions

The agent suggests how to pivot based on what just failed. It uses the history of rejected paths to propose better directions.

The workflow

What your AI does when an experiment fails.

Instead of just saying 'I can't do that,' your agent analyzes the breakdown.

  1. Analyze the failure

    The agent examines the prompt, the output, and the error to find the specific point of divergence.

    validate_edison_experimentation

  2. Log the rejection

    The AI records the failed parameters and the reason for the error in your experiment log.

    validate_edison_experimentation

  3. Identify patterns

    The agent compares this failure against previous logs to see if this is a one-off or a systemic issue.

    validate_edison_experimentation

  4. Propose a pivot

    The agent suggests a new direction or a modified prompt to bypass the identified obstacle.

    validate_edison_experimentation

Try it

Copy these to start.

Paste these into your chat to begin documenting.

Starting points

These are opening lines. Swap the bracketed text for your specific project or file names.

Edison Experimentation Prover Connector

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

Connector linkhttps://edge.vinkius.com/vk_preview_UMZPgT3WihcCcSB3Nac36KXLZfuGNzCN6Yn2jL8p/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 — Edison Experimentation Prover capabilities are ready to use.
Configuration · claude_desktop_config.jsonCopy
{
  "mcpServers": {
    "edison-experimentation-prover-mcp": {
      "url": "https://edge.vinkius.com/vk_preview_UMZPgT3WihcCcSB3Nac36KXLZfuGNzCN6Yn2jL8p/mcp"
    }
  }
}
Copy into chat04

  • Log this failed attempt in my research_log file and tell me why the logic broke.

  • Check my experiment_results sheet and find the common reason why the last three prompts failed.

  • Analyze the failure in my python_script_test and suggest a better way to structure the prompt.

  • Review the rejected paths in my project_notes and tell me what I should avoid doing next time.

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

Start here

Connect Edison Experimentation Prover once, then ask.

Just click the link to connect. Your credentials stay encrypted, and you can immediately start asking your agent to log and analyze your work.

Connect Edison Experimentation Prover to your AI

FAQ

How this task behaves.

  • 01

    Can the AI automatically fix my code when an experiment fails?

    No. The AI documents the failure and suggests a new path, but you must decide which changes to implement.

  • 02

    Does the AI delete my previous successful experiments?

    No. The AI only interacts with the logs to add new failure data or analyze existing entries.

  • 03

    Where does the documentation actually get saved?

    The AI writes the logs to the specific files or sheets you specify in your prompt.

  • 04

    Can the AI decide which failures are important?

    The AI can flag patterns, but the final decision on what constitutes a critical failure stays with you.

  • 05

    Will this work with any AI client?

    Yes, it works with any MCP-compatible client like Claude, Cursor, or Windsurf.

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

Connect Edison Experimentation Prover to Claude, Cursor, ChatGPT & more