Human review for content moderation edge cases

Send the posts your moderation model can't decide on to verified people, and get a consensus answer back.

How it works for this decision

Your agent sends

{
  "question": "Does this review break the rule against personal attacks?",
  "context": "The food was cold and the manager couldn't care less. Two stars.",
  "options": [
    "Breaks the rule",
    "Does not break the rule"
  ],
  "verifiers": 3
}

A verified person sees

Task$0.05

Does this review break the rule against personal attacks?

The food was cold and the manager couldn't care less. Two stars.

You get back

{
  "answer": "Does not break the rule",
  "agreement": 1,
  "answers": 3,
  "votes": {
    "Breaks the rule": 0,
    "Does not break the rule": 3
  }
}

The result shown is an example. Try answering the task yourself.

Why a human check

Moderation models handle the obvious cases well but struggle with context, satire and coded language, where both false removals and missed harm are costly.

Independent people judging the same post, with agreement scores, give a defensible decision on the cases that matter.

Put your policy text in instructions so every verifier applies the same rule.
Verifiers can skip anything they're not comfortable seeing; the task goes to someone else.

Questions

Do verifiers have to look at disturbing content?

Verifiers can skip any task, and skipping never counts against them. Plures is meant for judgment calls, not graphic content review.

How do I start?

Get a free test key at plures.si, send tasks from your agent in test mode, then add credit in your dashboard and live turns on.