MFA filtering across web inventory and arbitrage apps

The ANA's programmatic transparency study put made-for-advertising sites at twenty-one percent of impressions and fifteen percent of spend across the accounts it examined. On the books I have reviewed, open-exchange MFA share runs between six and nineteen percent. The number that matters more is what happens next: cutting it does not hand that budget back at the same price. Impressions you were buying at a two dollar CPM get replaced at eight, so volume falls hard while outcomes generally do not move. If nobody has had that conversation with your finance lead, have it before the filter goes live rather than in the following month's review. How I decide a site is MFA: Not from a purchased list alone. Third-party MFA lists are a reasonable starting position and they are thin on Southeast Asian and Chinese-language inventory, which is exactly where a regional buy concentrates. The list is a seed. After that it is manual. The signals I weight: ad-to-content ratio above the fold and total slots per page; refresh cadence; the share of the site's audience that is paid acquisition, since a site buying traffic from a content recommendation network and reselling it programmatically is running arbitrage whatever the editorial looks like; reseller density in ads.txt, where several hundred RESELLER lines against a handful of DIRECT ones tells its own story; template reuse across a family of domains registered in the same week; and copy reworded from elsewhere at volume. No single signal condemns a domain. Three together do, and I record which three. In-app arbitrage: The app-side equivalent gets discussed far less and is larger in this region. Utility apps with rewarded video walls, offerwall-driven installs, and ad density that only makes sense if the app exists to serve ads. The forensics differ from web: session length distribution, rewarded impression frequency per user per day, and whether the publisher's user acquisition spend and its ad revenue form a closed loop. Same treatment — evidence recorded, decision written down, reviewable by someone who was not in the room. Enforcement: A finding is worthless without a mechanism. You get the exclusion in the format each platform ingests, and more usefully, a supply-path rule that stops the same inventory arriving next month under a different domain, which it will. Domains are cheap. Seller relationships are not, and the seller is the durable control point. Out of scope: Brand suitability. MFA is a value question, not a safety one. These domains are usually clean and the ads run beside nothing objectionable. Conflating the two produces a filter that does neither job properly. Ongoing list operations. I build the rules and run the first pass. Your team runs it after that. Litigation support and public reporting. I write for internal decisions only. Not for you if: The buy is mostly PMP and curated deals. You have already avoided most of this, and the residual exposure will not pay for a review. Your success metric is impressions delivered or average CPM. Both get worse here, immediately and deliberately, and there is no version of this work where they do not. You want a monthly MFA score. This is a decision and a set of rules, not a subscription, and I would rather hand it over than bill you for maintaining a spreadsheet somebody else could keep. Reviews run in English and Mandarin. Regional-language inventory in Bahasa, Vietnamese and Thai is reviewed with a native reader I bring in and pay for, and that cost sits inside the price rather than appearing later as a line item.