Decision models
Screen small marketplaces' new listings for scams
Run a screening step for small marketplaces, plugin stores and directories that take listings from the public. A decision model (a fast AI that only makes yes/no or sorting calls) checks each new listing for scams, phishing, impersonation and offensive content. It then approves the listing or sends it to a person to review.
The hard part: convincing small marketplace owners to add a new step to how they accept listings, and tuning the questions so few good listings get held back.
Who pays
Owners of small marketplaces, plugin stores and directories pay you so they don't have to check every listing by hand. Our suggestion is a per-listing fee for low-volume sites and a flat monthly fee for busier ones. This pricing is only a suggestion. The source shows one plugin registry screening its listings this way, not anyone earning money from selling it as a service.
How to build it
Write a fixed list of questions
Use Claude to draft short yes/no questions about a listing's text and links, such as 'Does this link point to a different brand than the listing names?' Write a separate set of questions for icons and screenshots.
Test the questions on real listings
Collect 50-100 past listings from a pilot client, including some they rejected. Run each one through the decision model and compare its answers with what the client actually decided.
Build the approve or review flow
Use Claude Code to build a small service that takes a new listing, asks every question, and approves it if all answers are clean. If any answer is flagged, it adds the listing to a review list stored in Supabase.
Give clients a review page
Build a simple page where the client's moderator sees each flagged listing, the question that flagged it, and approve or reject buttons.
Connect it to the client's site
Have the client's site send each new listing to your service when it is submitted, then publish or hold the listing based on the reply.
Charge per listing or monthly
Bill clients through Stripe, either by counting screened listings each month or with a flat monthly plan.
Sources on X
- NewsTongueX · Oct 7, 2026
NewsTongueX reports that every plugin listing in EmDash's official registry is now screened by Cloudflare's Clef decision model before it appears in the catalog. Clef checks listing details, links, icons and screenshots for phishing, impersonation, offensive content, scams and attempts to game the moderation system. The poster says it asks nine questions about text and links and eight about images.
Ideas are curated from public posts and untested by us. Read the full note.