Moderation

Categories

What Jev looks for, how sure it must be, and how to describe it.

Each category is one thing Jev scores every message for. It has a description, a threshold and a list of actions.

categories:
  scam:
    description: "Phishing links, fake giveaways, or requests for account passwords or payment."
    threshold: 0.8
    actions: [cancel, warn, notify-staff]

Defaults

self-harm only alerts staff: the player needs a person, not a block.

Write descriptions literally

Jev answers the words you write, not what you meant. Say exactly what counts, and what doesn’t:

• Too vague: “Bad messages.”
• Better: “Insults, threats, or bullying aimed at a specific player or group.”

The default self-harm description shows how to rule things out: it says gamer exaggerations like “kms this lag” are not self-harm. Do the same in safe-description for the banter that is normal on your server.

Thresholds

A threshold is how likely the category must be before its actions run, from 0 to 1. Lower catches more, and also flags more by mistake. Messages between review-threshold (0.5) and the threshold go to staff for review instead.

Add, change or turn off

• Add one: a new key under categories, with a description, threshold and actions.
• Turn one off: set enabled: false on it.
• safe is reserved for ordinary chat and can’t be a category, and at least one category must stay on.