JevMade hello@JevMade.com
← Back to guides

JevMade field notes / Measured experiment guide

How to test what an AI assistant should remember

This guide explains a testing tool that helps an AI assistant decide which facts to keep. It checks if a detail matches the original conversation, will be useful later, and keeps the original meaning.

Original by Cairn InkEvaluation

Listen to this guide

JevMade’s plain-English explanation

0:00 /

AI narration

Credits and license

“Cairn Jev Lab” by Cairn Ink. Read the original source.

This expanded guide is an AI-narrated adaptation of the source’s essential explanation, examples and caveats, not a word-for-word reading. The synthetic voice does not imitate the author or imply their endorsement.

License: MIT

MIT License

Copyright (c) 2026 Cairn-ink

Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.

Our summary

This project is a testing tool that helps an AI assistant decide what information to keep from a conversation. Builders use it to check if a proposed memory is accurate and useful before the software actually saves it, preventing the assistant from remembering false details.

The software uses Jev, an AI tool that chooses from options rather than writing an answer. It checks if a detail matches the original text, if it will be useful later, and if it keeps the original meaning. Based on these answers, a strict rule recommends whether to save, skip, or wait.

This tool only recommends what to save; it does not actually store or delete information. It is useful for people building AI software who want to test memory rules. To test your own text, you must provide a private access code to connect to the service.

Key takeaways

  1. Test if a memory matches the original text, keeps the same meaning, and will be useful later.
  2. Save the AI tool's choices and confidence scores next to the clear rule it followed.
  3. Use your own examples to see how the software handles tricky or unclear statements.

All test results were reported by the project creators and have not been independently verified. The planned study on tricky examples is not yet finished, and the test labels were written by AI.

Repository README · Source reviewed

Read the original guide Opens the author’s site in a new tab.