MFA Filtering Built as an Inclusion Policy, Not a Blocklist

The ANA's 2023 programmatic transparency benchmark put 15 percent of programmatic spend and 21 percent of impressions on made-for-advertising sites, measured across 21 advertisers and roughly 123 billion impressions. Read that as an average, not as your number. The two advertisers we have measured against their own logs in Spanish-language supply came in at 9 and 31 percent, and the gap was category rather than competence. Beauty, celebrity and listicle content in Mexican Spanish is the densest made-for-advertising neighbourhood on the open web, so a cosmetics brand and a recruitment platform buying through the same DSP with identical settings are not carrying the same exposure. Why the deliverable is not a blocklist: A block file is a treadmill. Domains respawn weekly on the same template under a fresh registration, and a forty-thousand-line exclusion list becomes a maintenance liability that quietly suppresses legitimate publishers nobody has reviewed since the file was pasted in. We build the inverse: an inclusion policy with rules that can be applied to a domain nobody has seen before. That is the only version of this work that survives contact with next month's supply. What the classifier scores: Advertising-to-content ratio measured on the rendered page rather than the source. Slot count, slot geometry and refresh cadence. The share of the site's traffic acquired from paid sources, which is what separates an aggressive publisher from an arbitrage business. Template fingerprint, since operators run the same theme and the same slot geometry across dozens of domains. Whether a masthead, bylines and a corrections policy exist and resolve to real people. And whether the domain appears in the seller's sellers.json as a publisher or arrives three intermediaries deep. The tier most policies are missing: Three buckets, not two. Allow, exclude, and unknown, with unknown bid at a reduced rate rather than zero. A binary policy forces every unreviewed domain into one of two wrong answers, and in Spanish-language supply the unknown bucket is large enough that blocking it costs real reach while allowing it costs real money. What we refuse: We do not treat a vendor's made-for-advertising list as ground truth. Every engagement includes two hundred domains classified by hand, in a browser, against the same written rules, and where the vendor and the sample disagree we go with the sample and show the disagreement. We do not classify on advertising density alone; Spanish-language news runs heavy, and treating density as proof condemns exactly the publishers a brand ought to be funding. We do not keep the policy proprietary. You receive the rules, the scores and the lists, and you can hand all three to another agency the day you stop working with us. What is delivered: A written policy stating each threshold and the reasoning behind it. A seed allowlist per market. Implementation as inventory lists and targeting in your DSP. A monthly difference report naming every domain that changed tier and the signal that moved it. Who this is not for: Advertisers wanting a one-time cleanup. This supply regenerates, and an unmaintained policy is worth less than none because it creates a belief that the problem is handled. Brands buying entirely through a holding company's proprietary curated marketplace, where you cannot implement inventory decisions yourself. And anyone who wants this framed as brand safety: the content on these sites is usually perfectly safe, that is the entire design of them, and safety tooling will not find them.