Quality checks on what your AI says

Have people grade a sample of your assistant's answers against a rubric, and see quality drift before customers do.

How it works for this decision

Your agent sends

{
  "question": "Does the answer stick to what its sources say?",
  "context": "Source: Returns are accepted within 30 days with a receipt. Answer: You can return it any time within 60 days, no receipt needed.",
  "options": [
    "Yes",
    "Partly",
    "No"
  ],
  "verifiers": 3
}

A verified person sees

Task$0.05

Does the answer stick to what its sources say?

Source: Returns are accepted within 30 days with a receipt. Answer: You can return it any time within 60 days, no receipt needed.

You get back

{
  "answer": "No",
  "agreement": 1,
  "answers": 3,
  "votes": {
    "Yes": 0,
    "Partly": 0,
    "No": 3
  }
}

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

Why a human check

An assistant can get worse quietly: a new model version, a changed prompt or a stale document, and answers start missing the point or saying things their sources don't. Nobody notices until a customer complains.

People grading a steady sample of real answers against the same rubric turn quality into a number you can watch. Plures picks the answers worth a look and flags a drop the week it happens.

Grade a small random share of answers plus the risky ones: refusals, reroutes, answers with no sources.
Keep one rubric per question you care about, so the trend means something.
Your own team grades for free; verified people take the volume.

Questions

Do I have to send every answer?

No. Send the turns you want checked; Plures picks a sample and the ones with warning signs.

Where do I see the results?

On each app's Quality and Answers pages in the dashboard, and over the API.

How do I start?

Get your API key at plures.si. Start with your own team: the first 50 decisions each month are free. Add credit when you want ID verified people on the rest, live right away.