How these conventions are developed

The Barkhausen AI conventions are requirements-level specifications for measuring visibility in AI assistants. This page defines how they are versioned, what their normative language means, and how to comment on them.

Lifecycle

Each convention moves through three states:

• Draft — published for use and comment; may change in response to evidence or feedback. Draft status is stated on the document.
• Stable — settled. A Stable version is not edited in place; a substantive change is issued as a new version.
• Superseded — replaced by a later version. Superseded versions remain published at their original URLs, with a banner pointing to the current version, so that existing citations do not break.

Versioning

Conventions carry a version (for example, 1.0) and an effective date. A document identifier (for example, BA-C-1) is permanent across versions. Editorial corrections that do not change requirements are recorded in the changelog; changes to requirements produce a new version. When a version supersedes another, both the superseding and superseded relationships are recorded.

Normative language

The key words in these conventions are used with precise meaning:

• MUST — an absolute requirement of a compliant measurement.
• SHOULD — a strong recommendation; deviation requires a stated reason.
• MAY — genuinely optional.

Comments and corrections

Comments on any convention, and reports of errors, are welcome at contact@barkhausen.ai — please cite the document identifier and version. Substantive comments are considered for the next version; corrections are made promptly and recorded in the changelog.

Scope

These conventions define what a valid measurement must satisfy — its metrics, its sampling requirements, and its disclosures. They do not prescribe any particular tool or implementation, and they do not evaluate any product or vendor.