1. The live definition is protected
An in-product suggestion is saved as a proposal. It cannot directly change a public definition, and its AI advisory check cannot publish it.
Review policy
AI review rules
These rules describe the current in-product suggestion workflow. Source-based AI authoring is also permitted outside that workflow, without bypassing source, rights or publication controls.
What happens to a suggestion
Submitted → assigned editor prepares final wording → AI checks that exact draft → administrator publishes or returns it.
2. The system checks the change, not the person
The final check looks at the current wording, proposed wording, editor note and submitted source. Names, emails, profiles and account IDs are not sent for this check, and it should not infer expertise, intent or credibility.
3. New claims need evidence
The system flags factual, scientific, legal or regulatory claims that lack a source, scope, date or jurisdiction where one is needed.
4. Clear flags, not a black box
Every AI flag must name the exact wording it concerns and explain what needs clarification. A vague pass/fail result is not enough.
5. Deterministic checks come first
Required fields, valid links, contributor consent, spam limits, permissions and version history are enforced by ordinary software rules. AI does not decide those things.
6. The suggestion workflow has assigned roles
An assigned editor prepares the final wording. AI records advisory findings for that exact draft. An administrator then publishes it or returns it for more work, with a reason recorded. Source-based AI authoring outside this workflow has no mandatory human-editor gate.
7. Nothing is silently lost
The proposal, its source version, AI result, editor messages and final outcome are retained in the editorial record. Published versions remain comparable and recoverable.
8. Sensitive material is handled cautiously
The system flags personal data, abusive content, accusations, copyrighted passages and attempts to manipulate the review. Editors decide the safe next step for submitted suggestions.
Consistency and accountability
Each advisory check uses a fixed rubric and returns structured findings. The result records the policy, model, checked wording and time, so an editor or administrator can understand and challenge it.
Changing the final wording requires another check. Repeating the same wording, note and source under the same policy reuses the recorded result. AI findings are advisory; only an administrator can publish through this suggestion workflow.