JevMade field notes / Evaluation framework guide

Testing an AI assistant's past actions with multiple-choice questions

This guide explains a tool that tests how well an AI assistant follows instructions. It asks multiple-choice questions to check if the software used the right tools, stayed on topic, or made mistakes.

Original by OpenlayerGuardrails

Listen to this guide

JevMade’s plain-English explanation

0:00 /

AI narration

Credits and license

“jevals” by Openlayer. 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 jevals contributors

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 software helps developers test AI assistants by checking their past actions. It uses Jev, an AI tool that chooses from options rather than writing an answer. Developers use this to review if the assistant followed safety rules or supported its claims with evidence.

The software groups several specific questions together and sends them as a single request. This checks a single record of the assistant's actions all at once. Developers can set strict rules to stop the assistant from using executable tools if a test fails.

This is useful for developers testing their software, but different AI tools score tests differently. Developers must adjust their passing grades if they switch tools. The author measured a 6,015-request test run, but did not actually test the Kev and Laya interfaces.

Key takeaways

Group independent checks together to test a single record of the AI assistant's past actions.
Adjust your passing scores whenever you change the underlying AI tool used for testing.
Set strict failure rules for tests that control executable tools capable of making permanent changes.

More guides →See the experiments →