How to submit a metric

Scientific grounding & evidence (where most submissions fall short)

• 1–2 references are cited that are about the metric itself, not just the general topic.
• The references are specific and ideally peer-reviewed papers (or an official NIST resource).
• The references are not a governance framework (e.g., the AI RMF); governance frameworks are for mapping, not grounding.
• The implementation resources point to something usable today: released software, a published dataset, or code already merged into an open-source repository, not an unmerged or draft pull request.
• The implementation resources actually support calculating the metric.
• Every link is confirmed; none are broken, moved, or redirecting.

AI RMF alignment & metadata

• All required metadata fields are filled in. See the current SUBMISSION_FORMAT.md for the required metadata keys.
• The metric is mapped to the correct AI RMF Trustworthy Characteristic(s).
• The metric is mapped to the correct Lifecycle Stage(s).

Usage guidance & utility

• The submission is written in English.
• The definition is concise. Explanation that is not needed to understand or apply the metric has been removed.
• Clear, actionable guidance is provided on how to apply and interpret the metric.
• Scope boundaries, including where the metric does and does not apply, are stated explicitly.
• The metric adds value beyond what is already in the Metrology Center.

Readiness estimate

Count the unchecked items. These map approximately to the outcomes reviewers apply (Accept / Revise / Reject):

• All checked → reviewers would likely rate all dimensions Sufficient → Accept. Likely ready to submit/approve.
• A few unchecked, all fixable → likely at least one Partial, none Insufficient → Revise. Address before submitting, or request revision if reviewing.
• Missing references, no merged implementation, or out of scope → likely an Insufficient dimension → Reject. Fix before submitting, or recommend closure if reviewing; these are the issues most likely to get a PR closed.

This is an estimate. The reviewers' evaluation and the approver's decision on the PR are what determine acceptance.

Using an LLM to help prepare a submission

A submitter's own review is what matters most, and an LLM review is not a substitute for a careful manual self-check.

• Verify everything an LLM says. LLMs can miss real problems and invent ones that are not there, especially for references. Treat the output as a second opinion to check, not a verdict.
• Confirm every citation and link. LLMs can fabricate references or misjudge whether a source actually supports the metric. Submitters are responsible for the accuracy of what they submit.
• Do not input anything non-public. The submission and all review discussion are public; do not post information from an organization-internal LLM resource here; do not put proprietary or confidential material into a public LLM for this submission process.
• A green light from an LLM is not approval. Only the reviewers and approver decide. Passing an LLM check does not change the bar submitters will be held to.