Log-Level Ad Fraud Audit and IVT Claim Dossier

Your DSP already filters and credits general invalid traffic before the invoice reaches you. The money that stays lost is the sophisticated layer the pre-bid vendor did not block and the post-bid vendor flagged too late to matter, and across the Spanish-language display supply we have audited that layer has run between 4 and 11 percent of delivered impressions, concentrated in a small number of resold paths. This engagement measures your exposure from your own log-level records and turns it into a claim. Evidence we work from: Nothing here comes from a platform interface. We work from DV360 Data Transfer files landed in BigQuery, Xandr log-level feeds, Beeswax raw event logs or The Trade Desk Raw Event Data Stream, joined at impression identifier to your verification vendor's impression-level export from IAS, DoubleVerify or HUMAN. Where the join rate falls below 90 percent we say so and scope the finding to what the matched rows support, instead of grossing up to a headline. What gets tested: Declared domain against rendered domain, and both against the sellers.json chain back to the seller of record. Bundle ID and app-ads.txt agreement in-app and on connected TV, including server-side stitched paths where the user agent proves nothing. Data-centre and residential-proxy address clustering by ASN. Session pacing no human hand produces. On CTV, device-graph plausibility against the household count a market the size of Mexico can physically contain. Two patterns we look for first in these categories: Recruitment and job-aggregator supply attracts arbitraged traffic bought from the same push-notification networks it funds, so the click looks clean and the application never completes. Token and exchange campaigns attract farming infrastructure running real browsers with real fingerprints, which request-level IVT detection is structurally poor at catching because nothing about the request is malformed. Both need session and outcome evidence. What comes back: A dossier per seller: the impressions in dispute, the path they arrived through, the evidence behind each, the amount at issue, and the specific clause in that seller's terms the claim rests on. A named list of paths to cut now, with the volume each was carrying so your traders know what they are giving up. A read on which pre-bid settings paid for themselves and which cost you reach for nothing. If your measurement contract has a make-good window, we tell you when it closes. Not included: We do not select or resell verification vendors and hold no agreement with any of them. We do not do brand safety, suitability or contextual classification; that is a different problem and folding it into a fraud claim weakens the claim. We do not work on contingency and take no share of recovered spend, because a fee that scales with the size of a finding is a reason not to trust the finding. We do not run the media afterwards. Who this is not for: Advertisers spending under roughly 150,000 dollars a quarter on open exchange, where the recoverable amount will not clear the cost of the work. Teams whose DSP contract carries no log-level export and who will not renegotiate for it, because without raw records this becomes an opinion with footnotes. Anyone who wants a number for a board slide rather than a dispute they intend to file. We will say inside the first week if your data cannot carry a claim, and we would rather say it than write around it.