Consumer probes vs product ads
Supply-bench pages are not Adprime signup funnels. Success is a human completing an on-screen task. Mixing those jobs poisons both the quality read and the product test.
Published 2026-09-09 · Search job: how to test inventory quality
How to test inventory quality with paid display: send the click to a consumer probe, not to a product signup. Supply-bench pages are not Adprime signup funnels. Success is a human completing an on-screen task.
The /sb catalog exists so we can compare placements and routes by human completion versus honeypot traps. These are operator-run probes on behalf of the kind of advertiser who talks to average consumers. They are not a secret path into Adprime Zero. The hero does not pitch the API, the CLI, or an account.
Two jobs, two destinations
Product ads land on /lp isolated funnels. The promise is a buyer job: run ads from chat, skip the sales call, launch from Cursor, call a media-buying API. The conversion goal is email, signup, API key, funding, or a campaign object. See Stop sending paid clicks to the homepage.
Consumer probes land on /sb. The promise is an honest consumer ask: confirm you see this, vote on a poll, hold a circle, pick dinner, generate names. The conversion goal is visit_landed, then an on-screen interaction, then a deep human check. A trap fire is a negative quality signal.
If you rewrite a probe as a signup page, you lose the quality read. Bots and accidental clicks that would have failed a hold-confirm now bounce off a pricing paragraph, and you call that "the inventory was bad" or "the copy was bad" without knowing which.
If you send product-intent traffic to a sandwich poll, you learn nothing about which buyer promise works.
What counts as success on a probe
The truth hierarchy stays intact. Exchange clicks are not page views. Page views are not interactions. Interactions are not IVT.
On a probe, the score we came to collect is: did a person finish the task the banner asked for? Confirm-seen, do-not-click, hold-confirm, and sandwich-poll persisted proof_human in pre-launch browser tests. That is instrumentation, not a flight quality verdict. The ten live pages are listed in Ten interaction pages we actually flew.
Do not quote billed clicks as engagement. Do not pitch the product in the /sb hero. Footer disclosure as a traffic study is allowed. The quality read stays honest only if the job stays isolated.
If you need both a quality read and a product test in the same week, fly them as separate campaigns with separate names and separate destinations. The click URL is the isolation. Sharing a destination is how the two jobs collapse back into a homepage.