Blog · Supply chain

Gannett survived summary judgment on Enhanced Dynamic Allocation because cream-skimming is an allocation-order problem, not a bad bid request.

October 3, 2026 · RTBlint

On September 30, 2026, Judge P. Kevin Castel denied most of Google's summary judgment motion against Gannett and the Daily Mail in the publisher antitrust MDL, Document 2041 in 21-md-3010. Trade coverage on October 2 framed the ruling as a path toward roughly $1.5 billion in combined publisher damages claims. The engineering takeaway is narrower: practices the Virginia remedies case never adjudicated, including Enhanced Dynamic Allocation, Minimum Bid to Win, and the Bernanke family of bidding programs, will go to a jury because the record ties them to who sees an impression before anyone sends an OpenRTB bid.

What the court kept for trial

Castel's 88-page opinion separates precluded conduct from what the parties call non-precluded acts. Issue preclusion from Judge Brinkema's April 2025 liability findings already bars Google from relitigating the DFP-to-AdX tie, First Look, Last Look, Dynamic Revenue Share, and Unified Pricing Rules for these publishers. The September 30 ruling is where injury and damages meet Google's Daubert and summary judgment challenges on everything else.

For Gannett and the Daily Mail, Google's motion to exclude economists Shengwu Li and Ali Hortacsu failed. Google won only on New York General Business Law claims. Hortacsu's totals in the opinion are $599,955,198 for the Daily Mail and $900,771,489 for Gannett, plus interest accrual. A certified AdX publisher class keeps parallel expert work from Einer Elhauge. None of those figures is an award; they are estimates a jury may accept or reject.

That posture matters for buyers and exchanges because it preserves a factual record about auction order, not just take rates. When a demand team sees header bidding lose share on a premium publisher, the default explanation is bid price or signal quality. The surviving claims describe a world where the ad server can clear reserved direct inventory through the house exchange at manipulated temporary floors before rival exchanges receive the same opportunity to bid.

Enhanced Dynamic Allocation as cream-skimming

Enhanced Dynamic Allocation, implemented in 2014, made impressions publishers had reserved for negotiated direct deals available for programmatic bids on AdX. Before EDA, DFP prioritized fulfillment of those direct sales. Under EDA, AdX stopped prioritizing them and assigned a temporary price lower than the price the publisher had already attached to the direct deal. Li describes a cream-skimming problem: AdX could transact even when the direct price was higher, creating adverse selection that reduced advertisers' willingness to buy direct.

The opinion quotes Google's internal description of the feature as a way to let AdX and AdSense buyers cherry-pick higher revenue impressions while respecting delivery constraints on reservation ads. Hortacsu concludes EDA decreased the global value of direct sales by approximately 24.1%. For Gannett alone, Hortacsu attributes about $348.9 million in United States damages and $434.8 million worldwide to EDA, roughly 48% of that publisher's total claim in his model.

Li also opines that EDA probabilistically throttled bidders arriving through header bidding, because only AdX could reach the high-value impressions siphoned from direct reservations. That is not a malformed imp.banner or a missing source.ext.schain node. It is sequencing inside the ad server: the impression may never reach the exchange you instrumented at the same moment or with the same economic floor the publisher thought applied to the direct contract.

Minimum Bid to Win and asymmetric auction data

Google adopted Minimum Bid to Win when AdX moved to a first-price auction in 2019 and dropped Last Look. Under MBTW, DFP shared with AdX bidders and exchange-bidding participants the lowest bid that would have won a just-concluded auction, which tells the winner how much it overpaid. A Google engineer declaration summarized in the opinion describes Google Ads modeling expected competition from bid landscapes built from past auctions, including minimum-bid-to-win data since 2019.

Li opines that rival exchanges did not receive the same information, so only AdX buyers could predict the winning price with comparable precision. The court found that evidence sufficient for a jury on whether the feature depressed publisher prices and discouraged bidding on non-Google exchanges. For a supply-path analyst, MBTW is a reminder that two bid streams on the same page can carry different post-auction telemetry even when the OpenRTB request objects look identical at capture time.

OpenRTB has optional machinery to surface outcomes: loss notices through lurl, macros such as ${AUCTION_LOSS}, and community loss reason codes. Those fields help only when the exchange populates them and the ad server forwards them in a form the publisher reads. MBTW, as described in the litigation record, flows to house buyers and exchange-bidding participants, not necessarily to every header-bidding adapter logging the same impression id in your warehouse.

Bernanke and win-rate economics on the bidstream

The opinion treats Project Bernanke, Global Bernanke, Bell, and Alchemist as a related family. Plaintiffs allege Google used its Google Ads buying tool to inflate and deflate advertiser bids on AdX, charging winners more than the price paid to the exchange and banking differences in pools that subsidized Google Ads bids where outside competition was strong. Internal memoranda quoted in the record include estimates that Bernanke increased AdX win rate from 37% to 44% with a match rate of 50%, and revenue by $290 million in one scenario.

Hortacsu's report, summarized in footnotes the court cites, models how much Google would have had to cut AdX take rate to replicate win-rate lifts from other programs. Dynamic Revenue Sharing's 1.59 percentage point win-rate change, in his table, corresponds to lowering AdX take rate by 4.35 points from 20% to 15.65%. Unified Pricing Rules' 6.48 point win-rate change corresponds to a 16.18 point take-rate cut. For Bernanke, Global Bernanke, Last Look, and Alchemist, the opinion notes the win-rate increases were so large that a take-rate adjustment alone could not replicate them.

A buyer debugging low win rate on a major news publisher will still start with floors, deal ids, and creative approval. The surviving claims add a second checklist: compare house demand paths against header-bidding paths on the same placement, not only the JSON your bidder emitted. Win rate is an outcome; the MDL record is about programs that moved outcomes while exchanges argued over request parity.

How this differs from the Prebid remedies ruling

On September 2, 2026, Judge Brinkema ordered behavioral remedies in the government case, including functionally equivalent Prebid integrations and deprecation of Unified Pricing Rules for indirect transactions. RTBlint covered that ruling as an OpenRTB compliance problem: the bid request Prebid Server emits becomes evidence of equivalence. Castel's September 30 opinion in the publisher damages MDL does not change a field in OpenRTB 2.x. It changes what logs you need when a publisher asks why non-Google demand lost on inventory they believed was reserved for a direct buyer.

Read the two rulings together. Brinkema forces new hops and data sharing on a six-year clock. Castel lets Gannett and the Daily Mail take allocation programs to a jury now. Exchanges and DSPs that only instrument the header-bidding hop may be blind to the ad server decision that preceded the call. That blind spot is supply-path diagnostics, not creative delivery.

For a cross-site angle on browser-side identity shrinking while auction disputes intensify, see Safari 27 blocking adsrvr.org. Fewer sync pixels fire; more weight lands on whatever ids still arrive in user.eids. The Gannett record is about order of sale inside DFP, not Safari, but both stories push the same lesson: validate what you can observe on the wire, and separate that from ad-server logic you never received as a bid request.

What to log before you blame the bid request

Timestamp the ad server decision separately from the exchange auction. If EDA-style allocation runs first, your OpenRTB capture may start after the impression already cleared through house demand.
Store floor currency and value on every hop, including temporary floors on inventory tagged as direct or reservation in the ad server UI. Mismatch against imp.bidfloor is a data defect worth escalating even when the request validates.
Compare win, loss, and timeout rates between exchange bidding and Prebid adapters on the same ad unit. A gap concentrated on high-CPM slots matches cream-skimming theories better than a universal nbr 2 invalid request spike.
Normalize loss reporting on OpenRTB loss reason codes and lurl where partners provide them. Absence of a loss code is itself a signal when MBTW-style feedback allegedly flows only to house buyers.
Paste captured requests into the bid request tester to prove your side emitted valid OpenRTB. Validation passes do not prove you were called in time to compete for the impression.
Read the Prebid remedies post for the compliance record Brinkema ordered, and Prebid vs TAM vs Open Bidding for how parallel demand stacks already coexist on the same page.

RTBlint is independent and not affiliated with Google, Gannett, or any party to 21-md-3010. It checks whether a request or response matches OpenRTB and AdCOM. It does not detect antitrust injury, cream-skimming, or cartel behavior. It helps you show the payload your exchange sent was structurally valid when the publisher asks why you lost.