JevMade field notes / Worked recipe

Give uncertain moderation decisions a human review route

A moderation recipe uses separate checks for different risks and keeps the decisions to publish, hold, block, or escalate in program rules.

Original by Jeroen Erne / NexibeoGuardrails

Listen to this guide

JevMade’s plain-English explanation

0:00 /

AI narration

Credits and license

“12 · Moderation and AI guardrails” by Jeroen Erne / Nexibeo. 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 Jeroen Erne (https://nexibeo.com, https://completeaitraining.com)

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

A community filter needs to distinguish harassment from harsh criticism and an attack from a discussion about attacks. This recipe asks Jev separate questions about several hazards. Each question explains both what counts and what should not count as that risk.

The program applies different action rules to the resulting probabilities. Strong signals can block a message, uncertain cases go to moderators, and lower signals allow publication. Possible self-harm follows a separate route to a person who can help rather than simply being blocked.

The author's examples show how an overly simple blocking cutoff rejected harmless posts. That small test does not establish protection against unfamiliar attacks or a different community's language. The recipe is explicitly a first filter, not a replacement for human moderators or crisis services.

Key takeaways

Describe harmless boundary cases in every risk question.
Make the next action depend on the kind of harm and the cost of mistakes.
Test difficult examples from the actual community before relying on a filter.

More guides →See the experiments →