REPLY to Response to Motion re [229] MOTION to Compel Production of Successor-Custodian Documents and Relevant Source Code filed by Commonwealth of Virginia, State of Arizona, State of California, State of Colorado, State of Connecticut, State of Illinois, State of Michigan, State of Minnesota, State of Nebraska, State of New Hampshire, State of New Jersey, State of New York, State of North Carolina, State of Rhode Island, State of Tennessee, State of Washington, State of West Virginia, United States of America. (Teitelbaum, Aaron)
Page 1 PageID#
IN THE UNITED STATES DISTRICT COURT
FOR THE EASTERN DISTRICT OF VIRGINIA
Alexandria Division
UNITED STATES, et al.,
v.
Plaintiffs,
GOOGLE LLC,
Defendant.
)
)
)
)
)
)
)
)
No. 1:23-cv-00108-LMB-JFA
REPLY TO GOOGLE LLC’S MEMORANDUM OF LAW IN OPPOSITION TO
PLAINTIFFS’ MOTION TO COMPEL PRODUCTION OF SUCCESSOR CUSTODIAN
DOCUMENTS AND SOURCE CODE
The deadline for the parties to substantially complete their document productions is days from today. Yet despite no fewer than six meet-and-confers over the last two months,
Plaintiffs have been unable to secure Google’s basic agreement to produce (i) relevant
documents from Successor Custodians and (ii) critical source code documents that underlie key
allegations in the amended complaint. Plaintiffs requested these materials on the first day of
discovery over two months ago. Plaintiffs gave Google ample notice that it intended to seek
court resolution of these issues if the parties could not resolve them. Yet, even with this
knowledge, Google waited until the afternoon Plaintiffs’ brief was due to provide its most
detailed response to Plaintiffs’ repeated requests. Under these circumstances, Plaintiffs’ motion
to compel is certainly not “premature” (as Google characterizes it), particularly given the Court’s
admonition that parties should not sit on their rights and allow discovery disputes to linger. See
Mar. 31, 2023 Hr’g Tr. at 32:16-21.
The crux of Google’s opposition is that Plaintiffs should have requested this material
during the United States’ pre-complaint investigation of Google’s conduct. This argument was
already considered and rejected by the Court when it ruled on Google’s discovery plan proposal Page 2 PageID#
to restrict its document productions to a refresh of its prior productions in the investigation by
“applying the same search terms to the same group of custodians” without more. ECF No. 89, at
6; ECF No. 94, ¶ 6. As Plaintiffs noted back then, the fact that Congress empowered the United
States with investigative authority and the ability to seek pre-complaint discovery in no way
limits the United States’ ability, once in litigation, to avail itself of the full resources permitted
under the Federal Rules of Civil Procedure. The United States should not be limited in what
discovery it seeks any more than any other litigant. Standard rules of reasonableness and
proportionality should apply.
As for the relevance of the requested materials, Google’s brief has little to say on that
subject, and for good reason, because both categories of documents have indisputable relevance
to the allegations in the amended complaint. For these reasons, Plaintiffs respectfully request
that the Court compel Google to produce (i) documents from the files of Successor Custodians;
and (ii) the requested computer source code and related explanatory documents.
ARGUMENT
A. The Addition of Successor Custodians to Google’s Document Searches Is
Necessary to Ensure A Complete Production of Relevant Documents For the
Full Time Period Covered by the Amended Complaint.
In order to ensure that Google’s search for relevant documents includes documents across
the full relevant time period, Plaintiffs have made a simple request for Google to supplement its
original proposed custodian list (developed in October 2020 during the pre-complaint
investigation) with any successors to those employees who assumed those positions after
October 2020 (“Successor Custodians”). In response to this request, Google attempts to
minimize the relevance of documents after October 2020, claiming “that the conduct at issue
occurred over a long period of time before 2020,” and later, that the “conduct alleged in the Page 3 PageID#
Complaint” occurred “pre-2020.” ECF No. 238 at 4, 8. That argument simply does not square
with the allegations in the amended complaint.
There are numerous allegations contained in the amended complaint that describe, in
detail, conduct up until the filing of the complaint. Plaintiffs have alleged that “Google’s
anticompetitive scheme spans two decades and continues to the present.” ECF No. 120 ¶ (emphasis added). As but one example, the complaint alleges Google engaged in anticompetitive
conduct with respect to its development and launch of Google’s Unified Pricing Rules, launched
in September 2019, and that this conduct continued to harm competition through the present by
shifting transactions away from rivals and to Google’s Ad Exchange. Id., ¶ 246. Similarly, the
amended complaint alleges that Google developed an algorithmic model to predict the bids of
rivals for each impression called “Smart Bidding” that was a replacement for Google’s “last
look,” first launched in 2019 which, to the best of Plaintiffs’ knowledge, is still in place today.
Id., ¶¶ 258-261.
Plaintiffs bear the burden to prove to a jury that Google’s conduct had harmful effects
that were “not just historical; rather, Google’s anticompetitive conduct continues to affect the
marketplace on an ongoing basis.” Id., ¶ 260. Plaintiffs need sufficient discovery covering the
full relevant time period in order to present these issues to a jury. To effectively and efficiently
capture these documents, Plaintiffs have asked Google to add custodians in relevant roles, the
same roles Google agreed were relevant for their predecessors, for the entire time period alleged
in the amended complaint.
Google, with multiple counsel from multiple law firms representing it in this case, argues
that it is too burdensome to determine who these Successor Custodians might be. ECF No. 238,
at 4. Plaintiffs’ own public research seriously calls into question this alleged burden; in a matter Page 4 PageID#
of days, Plaintiffs were able to locate the past and current job titles for the 119 original
custodians, relying largely on information available from internet searches. Based on this
publicly available information and background information gathered during the investigation,
Plaintiffs were able to trim the request to 63 names who appeared to no longer hold the titles they
had held on or before October 2020, or where it was unclear who held those positions. Plaintiffs
asked Google to consult with its own employees—or its own human resources department—to
determine who the relevant successor custodians were for these 63 individuals and confirm
whether or not they were already included on Google’s custodian list. Plaintiffs struggle to
understand why this request poses a substantial burden.
To the extent Google now argues that reviewing documents from Successor Custodians is
unduly burdensome, Plaintiffs disagree. To be clear, we have asked for information about individuals, and Google provided information for about 45 of those individuals that resulted only
in the identification of 11 new potential custodians. Given the fact that we have asked for
custodians to cover the post-October 2020 time period, Plaintiffs would agree to limit Google’s
review and production of documents from these custodians to only the post-October 2020 time
period.
From the outset of negotiations, Plaintiffs have recommended ways Google could provide
sufficient information to Plaintiffs so that the parties could reach a reasonable compromise. We
are still at an information deficit. For example, although Google suggests that it has agreed to
review “two million additional documents” to address other requests, Plaintiffs have no way of
knowing the volume of relevant, responsive documents in the queue for review that might cover
the post-October 2020 time period. Id., at 2. Page 5 PageID#
Importantly, although Google’s May 26, 2023 letter provides the names of eleven
relevant successors, Google did not agree to add any custodians, and based on today’s meet-andconfer with Google, Plaintiffs still do not know if Google will agree to do so. The May 26, letter also failed to provide information about 18 of the 63 employees that Plaintiffs believe are
likely to have responsive information, including successors for the former Vice President and
General Manager for Google Ads and the former Group Product Manager for AdManager.
Accordingly, given the amount of time Google has had to comply with the request and
the rapidly approaching substantial completion of fact discovery deadline, Plaintiffs ask the
Court to resolve this issue by compelling Google to search and produce relevant materials from
the files of Successor Custodians in key positions at Google between October 2020 and the end
of the fact discovery period.
B. Plaintiffs’ Requests for Source Code Are Indisputably Relevant.
1. Plaintiffs need the source code for Request 39 and Request 42 that Google
stated it will not provide.
This case revolves around the digital advertising technology tools that Google uses to
deliver online advertisements to users on publishers’ websites in fractions of a second. The tools
that accomplish this task are not manual. They all derive from complex computer algorithms and
source code that direct Google machines to perform certain tasks designed by product engineers.
After agreeing, at the last possible moment, to provide source code for 9 out of 12 requested
computer functions, Google now claims that the 3 remaining categories of source code are either
unidentifiable or are not relevant. Neither contention is plausible. Google hires the engineers that
run these algorithms. It can and should find the relevant source code and produce it.
Plaintiffs first requested this source code more than two months ago. Now, at the final
hour, Google claims either not to know what source code is being referenced or contends that the Page 6 PageID#
code is not relevant, despite Google’s failure to seek clarification on these purported questions in
multiple meet and confers on this topic. 1 The reality is that Plaintiffs’ requests for source code
are specific and directly relevant to the conduct alleged in the amended complaint. Put simply,
access to source code will allow Plaintiffs’ experts to understand precisely how Google’s
algorithms make decisions on behalf of both advertisers and publishers, and will allow Plaintiffs
to explain to a jury the ways in which Google has “created a deliberately-deceptive black box
where Google sets the auction rules to its own advantage.” ECF No. 120 ¶ 39.
Request 39, parts 10 through 12 request information concerning how Google makes
choices for its customers. Plaintiffs expect source code will provide answers to a number of
complex questions, including (i) how Google’s algorithms choose its own products or a thirdparty product, (ii) when third-parties are allowed to bid or compete for impressions, and (iii)
what factors Google considers when it makes choices in algorithmic ways. Source code
determines the function and outputs of the products and features at issue in this litigation.
Specifically:
-
Request 39 part 10 requests source code that shows “Google’s algorithms make
choices regarding what exchanges, DSPs, and Ad Networks will be called for
Google’s customer.” See ECF No. 120 ¶¶ 39, 104-106);
Google’s attempt to argue that the standard for the production of source code is not just relevance
but necessity, is found nowhere in the single Fourth Circuit case it cites for this proposition: In
Keyes v. Lenoir Rhyne Coll., 552 F.2d 579 (4th Cir. 1977), the court merely found that there had
been no abuse of discretion where a trial court had not required production of confidential
evaluations of faculty members in an employment discrimination case. The court did not state
anywhere in this case that there is elevated standard above Rule 26 for confidential information
(and did not mention search code). Page 7 PageID#
-
Request 39 part 11 requests source code relating to: “Google’s algorithms and
features determine the bids, price floors, or auction optimization for its
customers.” See ECF No. 120 ¶¶ 32, 39, 115, 125, 130, 202, 219, 232-242; and
-
Request 39 part 12 request source code relating to “Google’s automated means or
manner of bidding, whether on individual bids or on a campaign-level basis.”
ECF No. 120 ¶¶ 39, 127-128, 222.
Similar to Request 39, Plaintiffs’ Request 42 seeks source code regarding how Google
makes decisions for its customers – – namely, how Google’s algorithms determine how to bid in
different auctions, what data Google’s algorithms draw on, how Google chooses the winner after
bidding, and how Google optimizes auctions for its publisher customer. Like Request 39, these
“black box” questions in Request 42 are raised throughout the amended complaint:
-
Request 42 part 1 seeks source code repositories relating to “how bids are selected
and submitted in any product that generates, alters, or submits a bid to purchase
an impression in any Relevant Product or in AdMeld.” ECF No. 120 ¶¶ 39; 48;
100-102; 129-136; 142; 214; 219; and
-
Request 42 part 2 seeks source code repositories relating to “the way or method
by which Relevant Products and AdMeld determine or determined the winners of
any real-time auction, waterfall bidding, and any sales of impressions.” ECF No.
120 ¶¶ 39; 113-120.
In addition, Google has not disclosed whether there are important data sets, bidding, and
pricing functions that are critical to Google’s operations of its own auctions and purchasing tools
but that may not be captured by search terms that hit on a specific project or feature name such
as “Smart Bidding.” And even if Plaintiffs did accept documents in lieu of source code, Plaintiffs Page 8 PageID#
run the risk of relying on documents describing such functions, only to have Google turn around
and claim at trial that the cited documents are incorrect or incomplete. In that regard, documents
are simply not a viable substitute for the underlying source code itself. Only the source code
itself will actually reveal how Google’s algorithms actually work. 2. Requests 39 through 43 requests documents necessary to understand and
analyze Google’s source code.
Plaintiffs have sought documents and data related to the requests for source code in
requests 39 through 43 3, as a means of ensuring that Plaintiffs’ experts are able to understand
this critical source code. In response, Google argues that it has provided search terms sufficient
to capture the relevant information. ECF No. 238 at 14-18. But in doing so, Google
misunderstands why Plaintiffs have specifically sought source code-specific documents, as
opposed to just documents concerning the relevant bidding algorithms. Plaintiffs’ additional
Plaintiffs sought another verifiable means of determining how Google’s algorithms work, by
seeking pseudocode. While Google misrepresents this as a “bait-and-switch” and a “newly-revised
request” (ECF 238 at 6), this proposal by Plaintiffs was in fact a good faith effort to respond to
Google’s stated concerns on a meet and confer regarding confidentiality of source code. Google’s
representation that it does not have a complete set of pseudocode made that potential solution
impossible—but it does not mean that the relevant pseudocode that Google did track down during
the course of its investigation should not be produced, as Plaintiffs expect it will be highly
informative, even if incomplete.
In response to Request 43, Google’s objection also argues that it is requiring Google to produce
a significant number of materials from two Google databases. Google mischaracterizes Request
43 which is asking, as are the rest of the document requests in Requests 39-43, for a certain number
of go-get documents and data that it explains in more detail in the above section. As for the
documents in databases Google cites in its brief (ECF No. 248 at 18) that are separately responsive
to Request 44, the parties have met and conferred several times in an attempt to reach a resolution
that would enable Plaintiffs to obtain materials that Google did not produce in the investigation.
Plaintiffs are working to identify discrete documents missing during the investigation, and we have
asked Google to produce spreadsheets to facilitate this process. While Plaintiffs are hopeful this
will be resolved without court intervention, this is a separate request. Materials called for in RFP
44, while informative and relevant, would not be sufficient to verify, analyze, and describe to a
jury precisely what Google’s algorithms and features do that allows Google to obtain an unfair
competitive advantage. Page 9 PageID#
requests are to ensure that their experts have sufficient information to be able to read and
understand the source code itself, which is independently necessary, for the reasons described
above, to verify the actual functioning of Google’s ad tech tools.
To the best of Plaintiffs’ knowledge, Google’s source code is unique, long, and
complicated – something approaching a language of its own. What Plaintiffs and their experts
need is the technological equivalent of dictionaries, glossaries, and cultural guides – enough
information to read and understand the source code. And this information is particularly
important, because Plaintiffs have agreed in the Protective Order to subject their experts to
restrictive conditions to view the source code, conditions that necessitate documents remaining
in Google’s possession and used in Google’s ordinary course business and engineering work to
explain the source code. To accommodate Google’s concerns regarding confidentiality of the
search code, Plaintiffs’ experts must review the source code at a specific agreed-upon location,
where they can be visually monitored by Google at all times, and can take notes but not copy
portions of the source code into these notes, among other restrictions. ECF No. 203 at Appendix
B ¶¶ 10-13. In order to evaluate the source code, any individual reviewing the source code must
at least have access to the sorts of data a new Google engineer, attempting to understand this
code under fewer restrictions, would need: e.g., access to training data, launch or release dates,
language explanations, design documents, comparison testing, inputs and outputs, manuals and
resources—in short, the materials in Requests 39-43.
Plaintiffs cannot know, without receiving documents captured by search terms (none of
which have been produced to date), whether any document captured by search terms will fully
explain how Google’s algorithms work. This information is critical to ensure Plaintiffs experts
can get up to speed quickly and understand Google’s unique and complex code. Put another way, Page 10 PageID#
Plaintiffs’ experts require the same set of dictionaries, user manuals, and explanatory documents
that Google would provide to a new hire. Plaintiffs expect that Google’s proposed searches will
yield highly relevant documents pertaining to these issues. We will not know whether materials
like data dictionaries and user manuals will hit on these search terms, unless and until Google
actually produces documents.
CONCLUSION
For the foregoing reasons, Plaintiffs respectfully request that the Court compel Google to
produce documents from the files of Successor Custodians, source code and related explanatory
materials, and to make full and complete responses to Plaintiffs’ RFPs 39-43.
Dated: June 1, Page 11 PageID#
Respectfully submitted,
JESSICA D. ABER
United States Attorney
JASON S. MIYARES
Attorney General of Virginia
/s/ Gerard Mene
GERARD MENE
Assistant U.S. Attorney
2100 Jamieson Avenue
Alexandria, VA Telephone: (703) 299-Facsimile: (703) 299-Email: Gerard.Mene@usdoj.gov
/s/ Andrew N. Ferguson
ANDREW N. FERGUSON
Solicitor General
STEVEN G. POPPS
Deputy Attorney General
TYLER T. HENRY
Assistant Attorney General
/s/ Julia Tarver Wood
JULIA TARVER WOOD
/s/ Kelly D. Garcia
KELLY D. GARCIA
/s/ Aaron M. Teitelbaum
AARON M. TEITELBAUM
/s/ Amanda M. Strick
AMANDA M. STRICK
/s/ Michael J. Freeman
MICHAEL J. FREEMAN
United States Department of Justice
Antitrust Division
450 Fifth Street NW, Suite Washington, DC Telephone: (202) 307-Fax: (202) 616-Email: Kelly.Garcia@usdoj.gov
Office of the Attorney General of Virginia
202 North Ninth Street
Richmond, VA Telephone: (804) 692-Facsimile: (804) 786-Email: thenry@oag.state.va.us
Attorneys for the Commonwealth of
Virginia and local counsel for the
States of Arizona, California,
Colorado, Connecticut, Illinois,
Michigan, Minnesota, Nebraska, New
Hampshire, New Jersey, New York,
North Carolina, Rhode Island,
Tennessee, Washington, and West
Virginia
Attorneys for the United States
PDF Page 1
PlainSite Cover Page
PDF Page 2
Case 1:23-cv-00108-LMB-JFA Document 239 Filed 06/01/23 Page 1 of 11 PageID# 2007
IN THE UNITED STATES DISTRICT COURT
FOR THE EASTERN DISTRICT OF VIRGINIA
Alexandria Division
UNITED STATES, et al.,
v.
Plaintiffs,
GOOGLE LLC,
Defendant.
)
)
)
)
)
)
)
)
No. 1:23-cv-00108-LMB-JFA
REPLY TO GOOGLE LLC’S MEMORANDUM OF LAW IN OPPOSITION TO
PLAINTIFFS’ MOTION TO COMPEL PRODUCTION OF SUCCESSOR CUSTODIAN
DOCUMENTS AND SOURCE CODE
The deadline for the parties to substantially complete their document productions is 36
days from today. Yet despite no fewer than six meet-and-confers over the last two months,
Plaintiffs have been unable to secure Google’s basic agreement to produce (i) relevant
documents from Successor Custodians and (ii) critical source code documents that underlie key
allegations in the amended complaint. Plaintiffs requested these materials on the first day of
discovery over two months ago. Plaintiffs gave Google ample notice that it intended to seek
court resolution of these issues if the parties could not resolve them. Yet, even with this
knowledge, Google waited until the afternoon Plaintiffs’ brief was due to provide its most
detailed response to Plaintiffs’ repeated requests. Under these circumstances, Plaintiffs’ motion
to compel is certainly not “premature” (as Google characterizes it), particularly given the Court’s
admonition that parties should not sit on their rights and allow discovery disputes to linger. See
Mar. 31, 2023 Hr’g Tr. at 32:16-21.
The crux of Google’s opposition is that Plaintiffs should have requested this material
during the United States’ pre-complaint investigation of Google’s conduct. This argument was
already considered and rejected by the Court when it ruled on Google’s discovery plan proposal
1
PDF Page 3
Case 1:23-cv-00108-LMB-JFA Document 239 Filed 06/01/23 Page 2 of 11 PageID# 2008
to restrict its document productions to a refresh of its prior productions in the investigation by
“applying the same search terms to the same group of custodians” without more. ECF No. 89, at
6; ECF No. 94, ¶ 6. As Plaintiffs noted back then, the fact that Congress empowered the United
States with investigative authority and the ability to seek pre-complaint discovery in no way
limits the United States’ ability, once in litigation, to avail itself of the full resources permitted
under the Federal Rules of Civil Procedure. The United States should not be limited in what
discovery it seeks any more than any other litigant. Standard rules of reasonableness and
proportionality should apply.
As for the relevance of the requested materials, Google’s brief has little to say on that
subject, and for good reason, because both categories of documents have indisputable relevance
to the allegations in the amended complaint. For these reasons, Plaintiffs respectfully request
that the Court compel Google to produce (i) documents from the files of Successor Custodians;
and (ii) the requested computer source code and related explanatory documents.
ARGUMENT
A. The Addition of Successor Custodians to Google’s Document Searches Is
Necessary to Ensure A Complete Production of Relevant Documents For the
Full Time Period Covered by the Amended Complaint.
In order to ensure that Google’s search for relevant documents includes documents across
the full relevant time period, Plaintiffs have made a simple request for Google to supplement its
original proposed custodian list (developed in October 2020 during the pre-complaint
investigation) with any successors to those employees who assumed those positions after
October 2020 (“Successor Custodians”). In response to this request, Google attempts to
minimize the relevance of documents after October 2020, claiming “that the conduct at issue
occurred over a long period of time before 2020,” and later, that the “conduct alleged in the
2
PDF Page 4
Case 1:23-cv-00108-LMB-JFA Document 239 Filed 06/01/23 Page 3 of 11 PageID# 2009
Complaint” occurred “pre-2020.” ECF No. 238 at 4, 8. That argument simply does not square
with the allegations in the amended complaint.
There are numerous allegations contained in the amended complaint that describe, in
detail, conduct up until the filing of the complaint. Plaintiffs have alleged that “Google’s
anticompetitive scheme spans two decades and continues to the present.” ECF No. 120 ¶ 264
(emphasis added). As but one example, the complaint alleges Google engaged in anticompetitive
conduct with respect to its development and launch of Google’s Unified Pricing Rules, launched
in September 2019, and that this conduct continued to harm competition through the present by
shifting transactions away from rivals and to Google’s Ad Exchange. Id., ¶ 246. Similarly, the
amended complaint alleges that Google developed an algorithmic model to predict the bids of
rivals for each impression called “Smart Bidding” that was a replacement for Google’s “last
look,” first launched in 2019 which, to the best of Plaintiffs’ knowledge, is still in place today.
Id., ¶¶ 258-261.
Plaintiffs bear the burden to prove to a jury that Google’s conduct had harmful effects
that were “not just historical; rather, Google’s anticompetitive conduct continues to affect the
marketplace on an ongoing basis.” Id., ¶ 260. Plaintiffs need sufficient discovery covering the
full relevant time period in order to present these issues to a jury. To effectively and efficiently
capture these documents, Plaintiffs have asked Google to add custodians in relevant roles, the
same roles Google agreed were relevant for their predecessors, for the entire time period alleged
in the amended complaint.
Google, with multiple counsel from multiple law firms representing it in this case, argues
that it is too burdensome to determine who these Successor Custodians might be. ECF No. 238,
at 4. Plaintiffs’ own public research seriously calls into question this alleged burden; in a matter
3
PDF Page 5
Case 1:23-cv-00108-LMB-JFA Document 239 Filed 06/01/23 Page 4 of 11 PageID# 2010
of days, Plaintiffs were able to locate the past and current job titles for the 119 original
custodians, relying largely on information available from internet searches. Based on this
publicly available information and background information gathered during the investigation,
Plaintiffs were able to trim the request to 63 names who appeared to no longer hold the titles they
had held on or before October 2020, or where it was unclear who held those positions. Plaintiffs
asked Google to consult with its own employees—or its own human resources department—to
determine who the relevant successor custodians were for these 63 individuals and confirm
whether or not they were already included on Google’s custodian list. Plaintiffs struggle to
understand why this request poses a substantial burden.
To the extent Google now argues that reviewing documents from Successor Custodians is
unduly burdensome, Plaintiffs disagree. To be clear, we have asked for information about 63
individuals, and Google provided information for about 45 of those individuals that resulted only
in the identification of 11 new potential custodians. Given the fact that we have asked for
custodians to cover the post-October 2020 time period, Plaintiffs would agree to limit Google’s
review and production of documents from these custodians to only the post-October 2020 time
period.
From the outset of negotiations, Plaintiffs have recommended ways Google could provide
sufficient information to Plaintiffs so that the parties could reach a reasonable compromise. We
are still at an information deficit. For example, although Google suggests that it has agreed to
review “two million additional documents” to address other requests, Plaintiffs have no way of
knowing the volume of relevant, responsive documents in the queue for review that might cover
the post-October 2020 time period. Id., at 2.
4
PDF Page 6
Case 1:23-cv-00108-LMB-JFA Document 239 Filed 06/01/23 Page 5 of 11 PageID# 2011
Importantly, although Google’s May 26, 2023 letter provides the names of eleven
relevant successors, Google did not agree to add any custodians, and based on today’s meet-andconfer with Google, Plaintiffs still do not know if Google will agree to do so. The May 26, 2023
letter also failed to provide information about 18 of the 63 employees that Plaintiffs believe are
likely to have responsive information, including successors for the former Vice President and
General Manager for Google Ads and the former Group Product Manager for AdManager.
Accordingly, given the amount of time Google has had to comply with the request and
the rapidly approaching substantial completion of fact discovery deadline, Plaintiffs ask the
Court to resolve this issue by compelling Google to search and produce relevant materials from
the files of Successor Custodians in key positions at Google between October 2020 and the end
of the fact discovery period.
B. Plaintiffs’ Requests for Source Code Are Indisputably Relevant.
1. Plaintiffs need the source code for Request 39 and Request 42 that Google
stated it will not provide.
This case revolves around the digital advertising technology tools that Google uses to
deliver online advertisements to users on publishers’ websites in fractions of a second. The tools
that accomplish this task are not manual. They all derive from complex computer algorithms and
source code that direct Google machines to perform certain tasks designed by product engineers.
After agreeing, at the last possible moment, to provide source code for 9 out of 12 requested
computer functions, Google now claims that the 3 remaining categories of source code are either
unidentifiable or are not relevant. Neither contention is plausible. Google hires the engineers that
run these algorithms. It can and should find the relevant source code and produce it.
Plaintiffs first requested this source code more than two months ago. Now, at the final
hour, Google claims either not to know what source code is being referenced or contends that the
5
PDF Page 7
Case 1:23-cv-00108-LMB-JFA Document 239 Filed 06/01/23 Page 6 of 11 PageID# 2012
code is not relevant, despite Google’s failure to seek clarification on these purported questions in
multiple meet and confers on this topic. 1 The reality is that Plaintiffs’ requests for source code
are specific and directly relevant to the conduct alleged in the amended complaint. Put simply,
access to source code will allow Plaintiffs’ experts to understand precisely how Google’s
algorithms make decisions on behalf of both advertisers and publishers, and will allow Plaintiffs
to explain to a jury the ways in which Google has “created a deliberately-deceptive black box
where Google sets the auction rules to its own advantage.” ECF No. 120 ¶ 39.
Request 39, parts 10 through 12 request information concerning how Google makes
choices for its customers. Plaintiffs expect source code will provide answers to a number of
complex questions, including (i) how Google’s algorithms choose its own products or a thirdparty product, (ii) when third-parties are allowed to bid or compete for impressions, and (iii)
what factors Google considers when it makes choices in algorithmic ways. Source code
determines the function and outputs of the products and features at issue in this litigation.
Specifically:
-
Request 39 part 10 requests source code that shows “Google’s algorithms make
choices regarding what exchanges, DSPs, and Ad Networks will be called for
Google’s customer.” See ECF No. 120 ¶¶ 39, 104-106);
1
Google’s attempt to argue that the standard for the production of source code is not just relevance
but necessity, is found nowhere in the single Fourth Circuit case it cites for this proposition: In
Keyes v. Lenoir Rhyne Coll., 552 F.2d 579 (4th Cir. 1977), the court merely found that there had
been no abuse of discretion where a trial court had not required production of confidential
evaluations of faculty members in an employment discrimination case. The court did not state
anywhere in this case that there is elevated standard above Rule 26 for confidential information
(and did not mention search code).
6
PDF Page 8
Case 1:23-cv-00108-LMB-JFA Document 239 Filed 06/01/23 Page 7 of 11 PageID# 2013
-
Request 39 part 11 requests source code relating to: “Google’s algorithms and
features determine the bids, price floors, or auction optimization for its
customers.” See ECF No. 120 ¶¶ 32, 39, 115, 125, 130, 202, 219, 232-242; and
-
Request 39 part 12 request source code relating to “Google’s automated means or
manner of bidding, whether on individual bids or on a campaign-level basis.”
ECF No. 120 ¶¶ 39, 127-128, 222.
Similar to Request 39, Plaintiffs’ Request 42 seeks source code regarding how Google
makes decisions for its customers – – namely, how Google’s algorithms determine how to bid in
different auctions, what data Google’s algorithms draw on, how Google chooses the winner after
bidding, and how Google optimizes auctions for its publisher customer. Like Request 39, these
“black box” questions in Request 42 are raised throughout the amended complaint:
-
Request 42 part 1 seeks source code repositories relating to “how bids are selected
and submitted in any product that generates, alters, or submits a bid to purchase
an impression in any Relevant Product or in AdMeld.” ECF No. 120 ¶¶ 39; 48;
100-102; 129-136; 142; 214; 219; and
-
Request 42 part 2 seeks source code repositories relating to “the way or method
by which Relevant Products and AdMeld determine or determined the winners of
any real-time auction, waterfall bidding, and any sales of impressions.” ECF No.
120 ¶¶ 39; 113-120.
In addition, Google has not disclosed whether there are important data sets, bidding, and
pricing functions that are critical to Google’s operations of its own auctions and purchasing tools
but that may not be captured by search terms that hit on a specific project or feature name such
as “Smart Bidding.” And even if Plaintiffs did accept documents in lieu of source code, Plaintiffs
7
PDF Page 9
Case 1:23-cv-00108-LMB-JFA Document 239 Filed 06/01/23 Page 8 of 11 PageID# 2014
run the risk of relying on documents describing such functions, only to have Google turn around
and claim at trial that the cited documents are incorrect or incomplete. In that regard, documents
are simply not a viable substitute for the underlying source code itself. Only the source code
itself will actually reveal how Google’s algorithms actually work. 2
2. Requests 39 through 43 requests documents necessary to understand and
analyze Google’s source code.
Plaintiffs have sought documents and data related to the requests for source code in
requests 39 through 43 3, as a means of ensuring that Plaintiffs’ experts are able to understand
this critical source code. In response, Google argues that it has provided search terms sufficient
to capture the relevant information. ECF No. 238 at 14-18. But in doing so, Google
misunderstands why Plaintiffs have specifically sought source code-specific documents, as
opposed to just documents concerning the relevant bidding algorithms. Plaintiffs’ additional
2
Plaintiffs sought another verifiable means of determining how Google’s algorithms work, by
seeking pseudocode. While Google misrepresents this as a “bait-and-switch” and a “newly-revised
request” (ECF 238 at 6), this proposal by Plaintiffs was in fact a good faith effort to respond to
Google’s stated concerns on a meet and confer regarding confidentiality of source code. Google’s
representation that it does not have a complete set of pseudocode made that potential solution
impossible—but it does not mean that the relevant pseudocode that Google did track down during
the course of its investigation should not be produced, as Plaintiffs expect it will be highly
informative, even if incomplete.
3
In response to Request 43, Google’s objection also argues that it is requiring Google to produce
a significant number of materials from two Google databases. Google mischaracterizes Request
43 which is asking, as are the rest of the document requests in Requests 39-43, for a certain number
of go-get documents and data that it explains in more detail in the above section. As for the
documents in databases Google cites in its brief (ECF No. 248 at 18) that are separately responsive
to Request 44, the parties have met and conferred several times in an attempt to reach a resolution
that would enable Plaintiffs to obtain materials that Google did not produce in the investigation.
Plaintiffs are working to identify discrete documents missing during the investigation, and we have
asked Google to produce spreadsheets to facilitate this process. While Plaintiffs are hopeful this
will be resolved without court intervention, this is a separate request. Materials called for in RFP
44, while informative and relevant, would not be sufficient to verify, analyze, and describe to a
jury precisely what Google’s algorithms and features do that allows Google to obtain an unfair
competitive advantage.
8
PDF Page 10
Case 1:23-cv-00108-LMB-JFA Document 239 Filed 06/01/23 Page 9 of 11 PageID# 2015
requests are to ensure that their experts have sufficient information to be able to read and
understand the source code itself, which is independently necessary, for the reasons described
above, to verify the actual functioning of Google’s ad tech tools.
To the best of Plaintiffs’ knowledge, Google’s source code is unique, long, and
complicated – something approaching a language of its own. What Plaintiffs and their experts
need is the technological equivalent of dictionaries, glossaries, and cultural guides – enough
information to read and understand the source code. And this information is particularly
important, because Plaintiffs have agreed in the Protective Order to subject their experts to
restrictive conditions to view the source code, conditions that necessitate documents remaining
in Google’s possession and used in Google’s ordinary course business and engineering work to
explain the source code. To accommodate Google’s concerns regarding confidentiality of the
search code, Plaintiffs’ experts must review the source code at a specific agreed-upon location,
where they can be visually monitored by Google at all times, and can take notes but not copy
portions of the source code into these notes, among other restrictions. ECF No. 203 at Appendix
B ¶¶ 10-13. In order to evaluate the source code, any individual reviewing the source code must
at least have access to the sorts of data a new Google engineer, attempting to understand this
code under fewer restrictions, would need: e.g., access to training data, launch or release dates,
language explanations, design documents, comparison testing, inputs and outputs, manuals and
resources—in short, the materials in Requests 39-43.
Plaintiffs cannot know, without receiving documents captured by search terms (none of
which have been produced to date), whether any document captured by search terms will fully
explain how Google’s algorithms work. This information is critical to ensure Plaintiffs experts
can get up to speed quickly and understand Google’s unique and complex code. Put another way,
9
PDF Page 11
Case 1:23-cv-00108-LMB-JFA Document 239 Filed 06/01/23 Page 10 of 11 PageID# 2016
Plaintiffs’ experts require the same set of dictionaries, user manuals, and explanatory documents
that Google would provide to a new hire. Plaintiffs expect that Google’s proposed searches will
yield highly relevant documents pertaining to these issues. We will not know whether materials
like data dictionaries and user manuals will hit on these search terms, unless and until Google
actually produces documents.
CONCLUSION
For the foregoing reasons, Plaintiffs respectfully request that the Court compel Google to
produce documents from the files of Successor Custodians, source code and related explanatory
materials, and to make full and complete responses to Plaintiffs’ RFPs 39-43.
Dated: June 1, 2023
10
PDF Page 12
Case 1:23-cv-00108-LMB-JFA Document 239 Filed 06/01/23 Page 11 of 11 PageID# 2017
Respectfully submitted,
JESSICA D. ABER
United States Attorney
JASON S. MIYARES
Attorney General of Virginia
/s/ Gerard Mene
GERARD MENE
Assistant U.S. Attorney
2100 Jamieson Avenue
Alexandria, VA 22314
Telephone: (703) 299-3777
Facsimile: (703) 299-3983
Email: Gerard.Mene@usdoj.gov
/s/ Andrew N. Ferguson
ANDREW N. FERGUSON
Solicitor General
STEVEN G. POPPS
Deputy Attorney General
TYLER T. HENRY
Assistant Attorney General
/s/ Julia Tarver Wood
JULIA TARVER WOOD
/s/ Kelly D. Garcia
KELLY D. GARCIA
/s/ Aaron M. Teitelbaum
AARON M. TEITELBAUM
/s/ Amanda M. Strick
AMANDA M. STRICK
/s/ Michael J. Freeman
MICHAEL J. FREEMAN
United States Department of Justice
Antitrust Division
450 Fifth Street NW, Suite 7100
Washington, DC 20530
Telephone: (202) 307-0077
Fax: (202) 616-8544
Email: Kelly.Garcia@usdoj.gov
Office of the Attorney General of Virginia
202 North Ninth Street
Richmond, VA 23219
Telephone: (804) 692-0485
Facsimile: (804) 786-0122
Email: thenry@oag.state.va.us
Attorneys for the Commonwealth of
Virginia and local counsel for the
States of Arizona, California,
Colorado, Connecticut, Illinois,
Michigan, Minnesota, Nebraska, New
Hampshire, New Jersey, New York,
North Carolina, Rhode Island,
Tennessee, Washington, and West
Virginia
Attorneys for the United States
11
Back to 
