How to submit a metric

A guide to proposing an AI measurement metric or methodology for inclusion in the NIST AI Metrology Center — what to prepare, what reviewers assess, and exactly how to send it. You do not need to know how to use Git, and you will not have to write any code.

You will need a free GitHub account

Submissions are reviewed in public on GitHub, so that anyone can read the discussion and you can reply to reviewers directly. You do not need to know how to use Git or write any code — a web browser is enough — but you do need an account. Create one at github.com; it takes about two minutes.

Everything you submit becomes public

Your submission and the whole review discussion are published and treated as non-confidential. Do not include proprietary or confidential information, and use a contact address you are willing to publish.

What this guide covers

  1. What reviewers look for — The standards a reviewer assesses your submission against.
  2. What to have ready — Every question you will be asked, and which answers are required.
  3. How to submit on GitHub — Each screen, in order, with what to expect.
  4. What happens next — Timings, and how review works.

Before you start

  • One metric per submission. If you have a family of related measures, submit the main one and list the others as variants.
  • Check it isn't already listed. Search the AI Metrology Center first — a close variant of something already published may be better added to that entry.
  • Allow about 30 minutes, plus whatever time you need to gather citations.

Already know GitHub?

This guide is not required. The submissions repository holds the submission guide, the authoritative format specification and worked examples, and you can open a pull request directly.