A Source-Verification Playbook for AI Directories

# A Source-Verification Playbook for AI Directories

A directory record should be treated as a claim with a source, not as a permanent fact. Verification is the process that decides whether the claim can remain public.

Define the source

Store the canonical official URL, the final URL after redirects, the source title, the observed status, and a content hash. Do not treat a discovery link as proof until the official page has been fetched and checked.

Define a passing check

A passing check should require an allowed HTTPS source, a successful response, a usable page, and a check time within the freshness window. If a source redirects to an unrelated domain, returns an error, or no longer represents the record, the check should fail.

Keep evidence separate

Store observations separately from editorial descriptions. This lets an operator see what the source said without confusing it with Acadanex’s interpretation. Record the checker, time, source URL, and next check date.

Fail closed

When a check fails or expires, remove the record from public search, public detail pages, structured data, and sitemaps. Keep the record and evidence for review rather than deleting history.

Review exceptions

Bot protection or unusual source behavior may require an administrative review, but an override should record an owner, reason, evidence, and expiry. It should never become a silent permanent bypass.

This playbook makes the directory easier to trust because freshness and uncertainty are visible in the system’s behavior.