Opinion
20-cv-10444-DJC
01-17-2023
NEURAL MAGIC, INC., Plaintiff, v. META PLATFORMS, INC. and ALEKSANDAR ZLATESKI, Defendants.
[UNDER SEAL]
Entered in Redacted Form on March 1, 2023]
MEMORANDUM AND ORDER
Denise J. Casper, United States District Judge
I. Introduction
Plaintiff Neural Magic, Inc. ("NMI") has sued Defendants Meta Platforms, Inc. ("Meta") and Dr. Aleksandar Zlateski ("Zlateski") (collectively, "Defendants"), a former NMI employee, for Zlateski's alleged disclosure of trade secrets to Meta and Meta's internal use and posting of same on an open-source forum, GitHub. D. 109. Specifically, NMI alleges that both Defendants misappropriated trade secrets in violation of the Massachusetts Uniform Trade Secrets Act ("MUTSA"), Mass. Gen. L. c. 93, § 42 (Count I) and the Defend Trade Secrets Act ("DTSA"), 18 U.S.C. § 1836 (Count II), and committed unfair and deceptive business practices in violation of Mass. Gen. L. c. 93A, § 11 (Count III). NMI also alleges that Zlateski breached the non-disclosure and non-competition provisions of his employment contract with NMI (Count IV) and that Meta tortiously interfered with its advantageous contractual relations with Zlateski (Count V).
Before the Court are Defendants' motion for summary judgment on all counts and NMFs theories of damages as to diminution of value and reasonable royalties, D. 303, and motions to strike the opinions, in whole or part, of Plaintiffs' experts: namely, Dr. Michael L. Scott ("Scott"), D. 297, D. 304; Professor Elizabeth Rowe ("Rowe"), D. 305; Dr. Sara Ellison ("Ellison"), D. 307; and Dr. Michael Akemann ("Akemann"), D. 308. Also before the Court are NMI's motion to strike the opinions of Defendants' experts: namely, Dr. David Kaeli ("Kaeli"), D. 287, Mr. Christopher Bakewell ("Bakewell"), D. 341, and Mr. R. Cuyler Robinson ("Robinson"), D. 342.
For the reasons explained below, the Court denies Defendants' motion for summary judgment in substantial part, but allows it as to Trade Secret No. 1, NMI's c. 93A claim and as to as much of the breach of contract claim that alleges violation of the non-compete provision. As to the experts, the Court ALLOWS Defendants' motion to strike Ellison's opinion regarding an alleged $880+ million diminution in valuation of NMI and limits the opinions of some of the other experts as described below. Accordingly, the Court ALLOWS Defendants' motion for summary judgment in part and DENIES it in part, D. 303, DENIES Defendants' motion to strike Scott's June 22, 2022 declaration on Fed.R.Civ.P. 26(e) grounds, D. 297, DENIES Defendants' Daubert motion to strike Scott's opinion relating to NMI's code comparison testing, D. 304, ALLOWS Defendants' motion to strike Rowe's opinion in part and DENIES it in part, D. 305, ALLOWS Defendants' motion to strike Ellison's opinion, D. 307, and DENIES Defendants' motion to strike Akemann's opinion, D. 308. Further, the Court DENIES NMI's motion to strike Kaeli's expert opinion, D. 287, DENIES NMI's motion to strike BakewelPs opinion in part and ALLOWS it in part, D. 341, and DENIES NMI's motion to strike Robinson's opinion, D. 342.
II. Standard of Review
A. Summary Judgment
The Court grants summary judgment where "there is no genuine dispute as to any material fact" and the undisputed facts demonstrate that the moving party is entitled to judgment as a matter of law. Fed.R.Civ.P. 56(a). "A fact is material if it carries with it the potential to affect the outcome of the suit under applicable law." Santiago-Ramos v. Centennial P.R. Wireless Corp., 217 F.3d 46, 52 (1st Cir. 2000) (citation and internal quotation marks omitted). The movant bears the burden of demonstrating the absence of a genuine issue of material fact. Celotex Corp. v. Catrett, 477 U.S. 317, 323 (1986); Carmona v. Toledo, 215 F.3d 124,132 (1st Cir. 2000) (citations omitted). If the movant meets its burden, the non-moving party may not rest on the allegations or denials in its pleadings, Anderson v. Liberty Lobby, Inc., 477 U.S. 242, 256 (1986), but must come forward with specific admissible facts showing that there is a genuine issue for trial, Borges v. Serrano-Isern, 605 F.3d 1, 5 (1st Cir. 2010) (citing cases). The Court "view[s] the record in the light most favorable to the nonmovant, drawing reasonable inferences in his favor." Noonan v. Staples, Inc., 556 F.3d 20, 25 (1st Cir. 2009) (citation omitted).
B. Daubert
Under Federal Rule of Evidence 702 and Daubert v. Merrell Dow Pharms., Inc., 509 U.S. 579 (1993), trial judges must act as gatekeeper to "insur[e] that the fact-finding process does not become distorted by 'expertise that is fausse and science that is junky.'" Fed. Ins. Co. v. Pentair Residential Filtration, LLC, No. 12-10853-RGS, 2013 WL 6145531, at *3 (D. Mass. Nov. 21, 2013) (quoting Kumho Tire Co., Ltd. v. Carmichael, 526 U.S. 137, 159 (1999) (Scalia, J., concurring)). As a threshold question, the judge must first determine whether the witness is sufficiently qualified by "knowledge, skill, experience, training, or education" to give his proffered opinion. Id. (quoting Fed.R.Evid. 702) (internal quotation marks omitted). If the witness is deemed qualified, the judge must next determine whether the specific testimony offered in the case "both rests on a reliable foundation and is relevant to the task at hand." In re Nexium [Esomeprazole] Antitrust Litis., 842 F.3d 34, 52 (1st Cir. 2016) (quoting Daubert, 509 U.S. at 597) (internal quotation marks omitted). "The reliable foundation requirement necessitates an inquiry into the methodology and the basis for an expert's opinion." Samaan v. St. Joseph Hosp., 670 F.3d 21, 31 (1st Cir. 2012). The relevancy requirement "seeks to ensure that there is an adequate fit between the expert's methods and his conclusions" by determining whether the expert's conclusions "flow rationally from the methodology employed." Id. at 32 (citations omitted). Generally, "[vigorous cross-examination, presentation of contrary evidence, and careful instruction on the burden of proof are the traditional and appropriate means of attacking shaky but admissible evidence," Daubert, 509 U.S.at 596 (citation omitted), but "a court may exclude an expert's opinion when it is based upon conjecture or speculation deriving from an insufficient evidentiary source," Equal Emp. Opportunity Comm'n v. Tex. Roadhouse. Inc., 215 F.Supp.3d 140, 158 (D. Mass. 2016) (citing United States v. Organon USA Inc., No. 070-12153-RWZ, 2015 WL 10002943, at *3-4 (D. Mass. Aug. 17, 2015)).
III. Factual Background
Unless otherwise noted, the following facts are undisputed. These facts are primarily drawn from the combined statement of material facts, D. 352.
A. Founding of NMI
NMI is a start-up company co-founded by Dr. Nir Shavit ("Shavit") and Dr. Alexander Matveev ("Matveev"). D. 352 at 491. Prior to NMI's founding, Shavit and Matveev worked together, applying machine learning ("ML") techniques to neural networks involving large data sets. Id. at 493. Generally speaking, neural networks are systems that mirror the way a human brain learns by experience. D. 318-4 at 24 (Kaeli Rpt); D. 319-2 ¶ 29 (Scott Rpt.). These networks analyze inputs, detect and label patterns from those inputs, and use the labeled patterns to recognize objects when new inputs are received, resembling the way the human brain learns by experience. D. 318-4 at 24; D. 319-2 ¶ 29. Because running neural networks on large data sets typically involved the use of Graphic Processing Units ("GPU"), which are large and expensive, Shavit and Matveev started to work on algorithms that would allow ML-based neural networks to run faster at GPU speeds on commodity Central Processing Units ("CPU")-the standard processors in most computers. See D. 319-2 ¶ 40; D. 352 at 493-95; but see D. 318-4 at 39 (explaining that "[b]efore July 2019, it was widely known that both CPUs and GPUs could be used with neural network and matrix multiplication applications").
At several points throughout the combined statement of material facts, Defendants take issue with NMFs citation to Scott's report to substantiate certain facts. See generally D. 352. As the report is unsworn, they argue that it is inadmissible at summary judgment. Id. at 505-06. However, "[previously unsworn statements may be used if they are subsequently reaffirmed under oath." Garcia-Navarro v. Hogar La Bella Union. Inc., No. 3:17-cv-01271-JAW, 2022 WL 11266460, at *15 n.19 (D.P.R. Oct. 18, 2022) (citing cases). Scott reaffirmed the opinion included in his report in his sworn June 22, 2022 declaration and during his deposition. D. 289-1 ¶¶ 3-9 ("June 22, Scott Decl."); D. 318-10 at 7; see DG&G. Inc. v. FlexSol Packaging Corp. of Pompano Beach, 576 F.3d 820, 825-26 (8th Cir. 2009) (concluding that the district court did not abuse its discretion in considering an unsworn expert report accompanied by an affidavit at the summary judgment stage); Miller v. Astucci U.S. Ltd., No. 04 Civ. 2201 (RMB), 2007 WL 102902, at *4 (S.D.N.Y. Jan. 16, 2007) (admitting unsworn expert report where the opinions contained therein "were explored in detail at his sworn deposition" (citation omitted)); Maytag Corp v. Electrolux Home Prods., Inc., 448 F.Supp.2d 1034, 1064 (N.D. Iowa 2006) (concluding that "subsequent verification or reaffirmation of an unsworn expert's report, either by affidavit or deposition, allows the court to consider the unsworn expert's report on a motion for summary judgment").
B. Zlateski and His Work Before NMI
Zlateski is a computer scientist who earned his bachelor's and master's degrees, along with his Ph.D., from the Massachusetts Institute of Technology ("MIT"). D. 352 at 14-15. Zlateski and Shavit met in 2014 while Zlateski was at MIT. Id. at 15. Shavit served on Zlateski's Ph.D. thesis committee. Id. Before NMI was founded, Zlateski conducted research on speeding up neural networks on CPUs, a fact of which both Shavit and Matveev were aware. Id. at 15-31. NMI disagrees with Defendants' characterization that Zlateski's pre-NMI research "involved research into speeding up sparse neural networks on CPUs;" rather, NMI contends his research involved only dense, not sparse, neural networks on CPUs. Id.
"Sparsity" refers to the number of zero elements in a given ML model. D. 318-4 at 31 (explaining that "[a] matrix can be characterized as dense or sparse. A sparse matrix has most of its elements as zero, whereas a dense matrix has most of its elements as nonzero. .. . 'Sparsity' is just the fraction of zeros in a layer's weight and input activation matrices" (internal citations omitted)); D. 319-2 ¶ 38 (explaining that "[a] matrix is 'dense' when it contains mostly non-zero values. ... On the other hand, a 'sparse' matrix is a matrix that contains mostly zero values, oftentimes on the order of 80% or 90%").
C. Zlateski Joins NMI
Shavit and Matveev hired Zlateski as the company's first employee in March 2018. Id. at 499-500. His employment contract (the "NMI Contract") included non-disclosure and noncompetition provisions. Id. at 500. Regarding confidentiality, the NMI Contract required that Zlateski would:
not at any time, whether during or after the termination of [his] engagement by [NMI], reveal to any person or entity any of the trade secrets or confidential, proprietary or other non-public information concerning the organization . . . including but not limited to information related to [NMI] inventions, research, testing, manufacturing, production, marketing, supplies, suppliers, consultants, strategic partners, products, designs, methods, know-how, technique, systems, processes, software programs and/or code, works of authorship, customer and collaborate lists, projects, plans, proposals, any Developments . . ., and the notes, memoranda, reports, lists, records, drawings, sketches, specifications, data, documentation or other materials of any nature containing such trade secret or confidential information.D. 333-4 at 8; D. 352 at 500-01. The NMI Contract further provides that any invention Zlateski created, either alone or with others, during his engagement with NMI "that . . . relates to the business of [NMI]... become[s] the sole and absolute property of [NMI]." D. 333-4 at 9; D. 352 at 502. The NMI Contract also includes a non-compete provision that provides that he could not "directly or indirectly... engage in any business activity which is in competition with the service or products being rendered, delivered, marketed, commercialized, produced or sold or under development or active consideration" by NMI during his employment with it and for one year thereafter. D. 333-4 at 10; D. 352 at 501-02.
D. Zlateski's Brief Tenure at NMI and Departure
At the time of NMFs founding, one "method of exploiting sparsity in neural network operations was to compress a sparse matrix into a dense representation by removing all zeros at the expense of using a separate indexing table to store their locations." D. 352 at 505-06; see D. 319-2 ¶ 199; but see D. 318-4 at 96 (explaining that "NMI and Dr. Scott asserted during the preliminary injunction proceedings that the industry strongly preferred compressed sparse row (CSR) encoding before NMI was founded, which Dr. Scott called the 'prevalent' approach to using sparse matrix multiplication for neural network computations on CPUs. . . . This assertion is incorrect" (internal citation omitted)). Early in Zlateski's time at NMI, on May 23, 2018, he and Matveev exchanged messages [Redacted] [Redacted]
"A matrix is a rectangular array of numbers, with some given number of rows and some given number of columns. A matrix can be multiplied by a single number (a scalar) or by another matrix." D. 319-2 ¶ 31; see also D. 318-4 at 29 (explaining that "[f]undamentally, a matrix is a two-dimensional rectangular array of numbers arranged in rows and columns").
A "loop" is a "syntactic construct that specifies a repetitive action (rather than specifying each repeated action separately)." D. 319-2 ¶ 49 n. 10.
Less than a week later, on May 28, 2018, Zlateski shared a code file- [Redacted] -with Shavit, Matveev, and NMI engineer Rati Gelashvili ("Gelashvili"). D. 333-26; D. 352 at 510. The code file included "nearly all optimizations [he and NMI] had discussed." D. 333-26 at 2. The code achieved a speed greater than four times that of Zlateski's pre-NMI, dense matrix multiplication code. D. 333-26 at 3; D. 352 at 511-14. NMI alleges that this code, written by Zlateski while in its employ, implemented each of its trade secrets-namely, (1) sparse instruction streaming; (2) compact Fuse multiply-add ("FMA") encoding; (3) FMA input buffering; and (4) unique combinations of the other three categories of its trade secrets. Id. at 514. Defendants dispute this characterization, arguing that the code does not demonstrate implementation of each trade secret or their particular elements, that "sparse instruction streaming, compact FMA encoding, and FMA input buffering" are only proprietary categories that NMI created for this litigation, and that NMI's alleged trade secrets would have been known to those skilled in the field based on prior disclosures, including but not limited to Zlateski's pre-NMI research. Id. at SHIS.
By late 2018, Zlateski had already begun looking for positions at other companies. D. 352 at 34-35. On December 18, 2018, Zlateski reached out to Dr. Jongsoo Park ("Park") at Meta about joining his company. D. 333-106 at 50. In March 2019, Meta offered Zlateski employment as an "Applied Research Scientist" in its Facebook Artificial Intelligence Research ("FAIR") group, which he accepted on March 12, 2019. D. 352 at 36-37. Zlateski subsequently informed Shavit that he was leaving NMI to join Meta. Id. at 37. Zlateski and NMI agreed his last day would be July 5, 2019. See Id. at 42.
When Zlateski left NMI, he informed Shavit and Matveev that he would be working on compilers while at Meta. Id. at 37. The parties dispute whether Zlateski informed them that this work would involve compilers running neural networks. Id. On his last day at NMI, July 5, 2019, Zlateski suggested that NMI "try to patent [a] bunch of [his] work that was fully done while there [Redacted] [Redacted] Id. at 531-33. [Redacted] Id. Regarding NMFs intellectual property, Shavit reminded Zlateski that "anything [Zlateski] did on a topic related to the compan[y']s product while [Zlateski] g[o]t a salary from the company is the compan[y']s property" and that Shavit was "not worried about it because [he] trust[ed] [Zlateski] underst[oo]d the ethics involved as [Zlateski] made it clear when negotiating [his] contract." Id. at 534. Finally, Shavit expressed NMFs desire to continue collaborating with Zlateski, even after his departure. Id. at 41-42.
Compilers are what translates a programmer's written source code into executable "assembly code" or "machine code." D. 318-4 at 37; D. 319-2 ¶ 23. "[T]his translated code comprises the instructions the CPU must execute." D. 319-2 ¶ 23.
A JIT compiler "wait[s] to generate portions of needed machine code until just before the code needs to be executed, unlike traditional compilers, which compile an entire program before that program is run." D. 318-4 at 38; see D. 319-2 ¶ 94.
Under the terms of the NMI Contract, Zlateski remained bound by its confidentiality and noncompete provisions upon his departure from NMI to Meta. D. 333-4 at 8, 10. With the end of employment, NMI allowed Zlateski to keep his NMI laptop (without requiring that it first be erased) and gave him continued access to an NMI Slack channel (an electronic message board used by NMI for internal communications) and source code. D. 352 at 46-53.
E. Zlateski Joins Meta
Zlateski began working at Meta as an employee based in its New York office on July 8, 2019. Id. at 53. NMI alleges that, shortly after starting there, Zlateski had dinner with Park and Meta engineer Andrew Tulloch ("Tulloch"), who were not in his workgroup, but rather in Meta's AI System Software/Hardware Co-Design group, a team focused on optimizing sparse matrix multiplication operations for neural networks. Id. at 38,468,544. Regardless of what information, if any, Zlateski may have shared during this dinner, the parties agree that in July 2019, Park asked Zlateski to write some sparse matrix multiplication ("sparseMM") code for Facebook General Matrix Multiplication ("FBGEMM"), an open-sourced code library that Park's group develops and uses to optimize performance across numerous ML and AI applications. Id. at 14, 542-43. On July 17, 2019, Zlateski shared a code file with Park and other Meta engineers, entitled as "A very dirty prototype of Sparse-Dense matrix multiply." Id. at 552-53. NMI contends that this code file implemented each of its trade secrets. Id. at 553-56.
Broadly speaking, sparseMM is "used in mathematics and computer science, to multiply two matrices together (matrix multiplication), while reducing the number of operations required by accounting for, and ignoring, zeros in those matrices (sparse matrix multiplication)." D. 318-4 at 32 (emphasis in original); see D. 319-2 ¶ 46.
Park rewrote the code Zlateski provided to be compatible with other FBGEMM code. Id. at 57, 591-92. On October 25, 2019, Park posted this sparseMM code to GitHub, an open-source software development platform. Id. at 57-58. NMI alleges it did not discover Defendants' trade secrets violations until several months later, when on or about January 17, 2020, Mateev saw a Linkedln post from Park concerning improvements to FBGEMM and crediting Zlateski for his contribution. Id. at 611-14. The next day, Matveev "forked a copy of the FBGEMM repository from pytorch/FBGEMM" on GitHub to his own personal GitHub webpage, which then remained publicly accessible there for more than eighteen months. Id. at 58-63.
F. The Leadup to the Initiation of This Lawsuit by NMI
On January 22, 2020, after Matveev's viewing of the public post about the FBGEMM, NMI sent letters to both Meta and Zlateski requesting that they remove the FBGEMM code from GitHub, claiming that it incorporated its trade secrets. Id. at 615. Meta eventually removed the FBGEMM code from the GitHub repository, asked GitHub to remove any vestiges of this code from its platform, instructed Zlateski and other engineers not to re-post or use the code or any derivative work, and agreed that it would not write any new code based on the allegedly proprietary algorithms. Id. at 616-22. NMI alleges, as a result of Defendants' alleged misappropriation and disclosure of its trade secrets, its corporate valuation decreased, it was forced to delay release of any commercial software for two years, and several investors declined to invest in NMI. Id. at 622-26.
IV. Procedural History
On March 4, 2020, NMI filed this lawsuit. D. 1. On April 9, 2020, NMI moved for a preliminary injunction. D. 25. After full briefing and argument, the Court denied the motion on May 29, 2020. D. 79. Also, in May 2020, Defendants moved to dismiss several of NMFs claims arising under state law. D. 74. The Court denied this motion except that it allowed it as to NMFs unjust enrichment claim. D. 88. Fact discovery then proceeded from December 2020 for approximately fourteen months. D. 106 at 1; D. 246; D. 318 at 10. On December 16, 2020, NMI filed an amended complaint, asserting the same claims from the original complaint excluding the claim for unjust enrichment. D. 109.
For its claims, NMI relies upon the opinions of four experts: Scott regarding the existence, possession, and misappropriation of NMFs trade secrets; Rowe on the reasonable protective measures that NMI took to protect its trade secrets; and Ellison and Akemann regarding the measure of damages. Defendants rely on the opinions of several rebuttal experts: Kaeli regarding the alleged trade secrets; Robinson on the reasonableness of NMFs protective measures; and Bakewell rebutting the reliability of NMFs theories of damages. Defendants have now moved for summary judgment, D. 303, and both parties have filed a litany of motions to strike the opinions of the opposing party's experts, D. 287; D. 297; D. 304; D. 305; D. 307; D. 308; D. 341; D. 342.
The Court will address the various motions to strike in the relevant sections of its discussion of NMFs claims.
V. Discussion
A. Overview of Trade Secret Allegations
NMI's June 3, 2022 statement is its operative "Trade Secret Statement." D. 318-2. It identifies twelve alleged trade secrets. Id. According to NMI, their trade secrets are three techniques for optimizing sparseMM operations on CPUs: "sparse instruction streaming"; "compact FMA encoding"; "FMA input buffering"; and unique combinations of these three trade secrets. D. 318-8 at 83-84; D. 319-2 ¶ 45.
NMI previously identified forty-eight trade secrets. D. 255-1. In response, Defendants moved to compel NMI to reduce the number of asserted trade secrets. D. 254. The Court allowed that motion and required NMI to reduce the number of asserted trade secrets to twelve. D. 278. Although the parties' respective experts, Scott and Kaeli, refer to the trade secrets by their prior numbering and designations in their primary reports, the Court will refer to Trade Secrets Nos. 1-12 as they were designated in the operative June 3, 2022 Trade Secret Statement, but notes here for clarity how each was initially identified in NMI's prior statement:
Trade Secret 1 was Trade Secret 1;
Trade Secret 2 was Trade Secret 3;
Trade Secret 3 was Trade Secret 9;
Trade Secret 4 was Trade Secret 12;
Trade Secret 5 was Trade Secret 13;
Trade Secret 6 was Trade Secret 17;
Trade Secret 7 was Trade Secret 18;
Trade Secret 8 was Trade Secret 19;
Trade Secret 9 was Trade Secret 21;
Trade Secret 10 was Trade Secret 22;
Trade Secret 11 was Trade Secret 28; and
Trade Secret 12 was Trade Secret 41.Compare P. 318-2 with P. 255-1: see P. 318-3.
1. Trade Secrets Nos. 6-10: Sparse Instruction Streaming
The alleged "sparse instruction streaming" trade secrets are Trade Secrets Nos. 6-10. D. 318-2 at 5-6.
A register is memory storage most quickly accessible by the CPU. D. 318-4 at 44 n. 11; D. 319-2TJ26.
[Redacted]
[Redacted] [Redacted]
Defendants, through Kaeli, contend that Trade Secrets Nos. 6-10 "merely encompass (and cobble together) well-known optimization techniques relating to [Redacted]
2. Trade Secret No. 11: Compact FMA Encoding
Trade Secret No. 11 is the alleged "compact FMA encoding" trade secret: [Redacted]
Defendants' expert Kaeli again contends that Trade Secret No. 11 cannot be characterized as a trade secret because it "merely encompass[es] (and cobble[s] together) well-known techniques For [Redacted]
3. Trade Secret No. 12: FMA Input Buffering
Trade Secret No. 12 is NMI's alleged "FMA input buffering" trade secret. [Redacted] [Redacted] Kaeli argues that this cannot be a trade secret because it "merely encompass[es] (and cobblefs] together) well-known optimization techniques used for [Redacted] He further states that "before July 2019, it also was well known that performance can be optimized by [Redacted]
[Redacted] .
4. Trade Secrets Nos. 1-5: Unique Combinations
Finally, Trade Secrets Nos. 1-5 in NMI's Trade Secret Statement aim to cover the combinations of the aforementioned trade secrets. D. 318-2 at 4-5. Kaeli challenges Trade Secrets Nos. 1-5 for the same aforementioned reasons-the concepts undergirding each unique combination were well-known in the field. D. 318-4 at 221. He also argues that the combinations "are undefined, overbroad, and unspecific." Id.
B. Misappropriation of Trade Secrets (Counts 1 and II)
Defendants argue that summary judgment is appropriate as to Counts I and II, which allege violations of the MUTSA, Mass. Gen. L. c. 93, § 42, and the DTSA, 18 U.S.C. § 1836. Specifically, they argue no reasonable jury could find that (1) NMI has demonstrated adequate identification or possession of the alleged trade secrets; (2) NMI took reasonable measures to protect its alleged trade secrets; (3) Zlateski acquired, used, or disclosed the alleged trade secrets through "improper means;" (4) Meta "knew or had reason to know" that any alleged trade secret was acquired by improper means; or (5) Defendants misappropriated NMI's alleged Trade Secret Item No. 12. See generally P. 318: P. 351.
1. Standards for MUTSA and DTSA Claims
A MUTSA misappropriation claim requires that a plaintiff demonstrate "1) the existence of a trade secret; 2) reasonable steps taken by the plaintiff to preserve the secrecy of its trade secret; and 3) the defendant's use of improper means in breach of its confidential relationship with the plaintiff to acquire and use the trade secret." Sutra, Inc. v. Iceland Exp., ehf, No. 04-11360-PPW, 2008 WL 2705580, at *3 (P. Mass. July 10, 2008) (citing cases).
The PTSA confers a federal cause of action on an owner of a trade secret that has been misappropriated and is "nearly equivalent" to the Massachusetts trade secret law. All scripts Healthcare, LLC v. PR/Pecision Res., LLC, 386 F.Supp.3d 89, 94-95 (P. Mass. 2019) (citation omitted). It requires plaintiff to show that (1) it took reasonable measures to keep such information secret and (2) the information derives independent economic value. Id. at 94 (citing 18 U.S.C. §§ 1836(b)(1) and 1839(3)). The DTSA defines misappropriation as "the disclosure or use of a trade secret of another without express or implied consent by a person who ... at the time of disclosure or use, knew or had reason to know that the knowledge of the trade secret was . . . acquired under circumstances giving rise to a duty to maintain the secrecy of the trade secret or limit the use of the trade secret." Id. at 94 (citing 18 U.S.C. § 1839(5)(B)).
Defendants do not address the element of independent economic value, see generally D. 318; D. 351, and so the Court need not reach it here.
2. Existence of Trade Secrets
As it looks to the parties' technical experts, namely Scott and Kadi, in assessing Counts I and II, the Court will address the parties' motions to strike same, before turning to the merits of these claims.
a) Discovery Dispute
The parties' first dispute about whether the Court should strike Kaeli's opinion (and/or allow any supplementation of Scott's opinion) concerns the course of discovery. NMI moved for a preliminary injunction on April 9, 2020 (which the Court subsequently denied), and supported its motion with an affidavit from Scott, detailing extensive testing he performed comparing NMI's JIT with the FBGEMM JIT. D. 37. In his affidavit, he described the test and its results:
[a]s part of my analysis, I compared the assembly code generated by running a set of 1,000 input matrices having varying sparsity levels through the NM GEMM JIT and the FBGEMM JIT. . .. The most striking parallel, exhibited in all 1,000 cases, is that the [Meta] assembly code [Redacted] as the [NMI] code. [Redacted] . . . . The odds of this happening without misappropriation of intellectual property are in my opinion vanishingly small.Id. ¶ 143. Also in the affidavit, Scott explained that he made only one change to the JIT algorithms. Id. ¶ 143 n. 14 (noting that [Redacted]
Another change, however, appears to have been made, based upon subsequent disclosures. Compare Id. with D. 289 at 7-8. NMI explained subsequently that Meta's "code (not the FBGEMM JIT itself) contains a parameter - not found in [NMTJ's code - that [Redacted] To compare the operation of the two JITs, rather than enable the [Redacted] Parameter in only [Meta] 's code, Dr. Scott fed the same random inputs into each JIT, to ensure the underlying algorithms were operating on the same inputs." Id. at 8 n.4 (italics in original).
On January 4, 2021, NMI served responses to Defendants' first set of interrogatories. In responding to Defendants' Interrogatory No. 2, which requested the complete bases for NMTs misappropriation claims, NMI relied upon Scott's testing but did not mention the change to the [Redacted] parameter. D. 302-1 at 3-6. NMI concurrently propounded Interrogatory No. 13, which sought, among other things, the factual and legal basis for any contention that Defendants had not misappropriated NMTs trade secrets. D. 289-8 at 2-3. Defendants responded to the interrogatory on April 19, 2021, restating the concerns about Scott's testing that they originally included in their opposition to the motion for preliminary injunction and stating, among other things, that they "intend to rely on fact and expert testimony relating to NMFs alleged trade secrets." D. 302-2 at 3-6.
Throughout discovery, NMI disclosed documents including the input variables and the underlying code used in Scott's tests. In March 2021, NMI produced the thousands of input variables Scott used to run his comparison tests. D. 289 at 8. The documents in this production [three variables] identified [Redacted] as the values Scott used for his comparison testing. D. 289-7. NMI contends this [Redacted] Defendants challenge this assertion, explaining that [Redacted] In May 2021, NMI also produced all the code underlying Scott's comparison testing. D. 289 at 8. On July 1, 2021, Defendants' expert Kaeli inspected this code. Id., It is at this point that NMI alleges Defendants became aware of the missing [another variable] [Redacted] parameter in Scott's comparison testing. Id. at 10. During the ensuing months of discovery, both parties updated their responses to the relevant interrogatories, but NMI never included any reference to the change in the code relating to the [Redacted] variable and Defendants never included any reference to concerns about the [Redacted] | variable, hi at 9; D. 301 at 10-11.
Matveev's deposition reconvened on February 24, 2022. D. 301 at 11. It was on this day that NMI, through Matveev, affirmatively disclosed the removal from the original FBGEMM's code of the [Redacted] variable. D. 302-6 at 7-14. It is after this deposition that Defendants contend Kaeli learned about the change regarding the [Redacted] variable, while NMI contends he was on notice of this change back in July 2021 after inspecting the code. D. 289 at 11; D. 301 at 15. Following this deposition, on March 3 and 7, 2022, Defendants' counsel emailed NMI's counsel to voice their concerns about Scott's comparison testing. See D. 302-8 at 2; D. 302-9 at 2.
On April 8, 2022, NMI served opening expert reports, including a report by Scott discussing his comparison testing. D. 289 at 9. This report did not discuss the change to the FBGEMM code regarding the [Redacted] variable. See generally D. 319-2. After examining Scott's report, Kaeli performed rebuttal testing to analyze the changes Scott made to the code, which he opines makes Scott's testing was unreliable. D. 301 at 13. On May 20, 2022, Defendants served Kaeli's rebuttal report. Id. The parties conferred on these issues in early June and deposed Scott and Kaeli. D. 289 at 10. During Kaeli's June 15, 2022 deposition, he stated that he became aware that the [Redacted] variable had been removed when he inspected the code in July 2021, but then clarified that he only had questions about the code then and did not learn about the removal until after Matveev's February 24, 2022 deposition. D. 289-3 at 18-31. During Scott's June 17, 2022 deposition, he conceded that he was aware of the removal of the [Redacted] variable but did not believe it was "necessary" to "spell out how [he] invoked the two compilers with the same parameters," at any time, because NMI produced information and code relating to his tests. D. 302-11 at 9. The instant motions to strike followed.
b) NMFs Rule 26(e) Motion to Strike Kaeli's Report
NMI argues Defendants violated Rule 26(e) when they did not update their interrogatory response to include either their concerns about the removal of the [Redacted] variable or the rebuttal testing performed by Kaeli. D. 289 at 12-14; D. 316 at 6-8. NMI further contends that Defendants should have updated their responses after May 2021, when Defendants and Kaeli had access to the full source code underlying Scott's tests. D. 289 at 13. Lastly, NMI argues, even if Kaeli only had questions about the code on July 1, 2021 and only learned of the removal after Matveev's February 2022 deposition, it was incumbent on Defendants to identify the change themselves and to update their interrogatory responses after the deposition and before opening expert reports were due. Id. at 14. The Court is not persuaded.
For context, Rule 26 mandates parties to supplement their responses to interrogatories "in a timely manner if the party learns that in some material respect the disclosure or response is incomplete or incorrect, and if the additional or corrective information has not otherwise been made known to the other parties during the discovery process or in writing." Fed.R.Civ.P. 26(e)(1)(A). The burden falls on the party alleging the Rule 26(e) violation. In re: Celexa and Lexapro Mktg. & Sales Pracs. Litis., MDL No. 09-02067-NMG, 2017 WL 9324342, at *1 (D. Mass. May 10, 2017) (citation omitted). Failure to fulfill Rule 26(e) obligations may result in some sanctions-namely, "the party is not allowed to use that information or witness to supply evidence on a motion, at a hearing, or at a trial, unless the failure was substantially justified or is harmless." Fed.R.Civ.P. 37(c)(1); see Klonoski v. Mahlab. 156 F.3d 255, 269 (1st Cir. 1998). Once a Rule 26(e) violation has been shown, the burden then falls to "the party facing sanctions . . . to show that its failure to comply with the Rule was either justified or harmless." Wilson v. Bradlees of New Eng., Inc., 250 F.3d 10, 21 (1st Cir. 2001) (citing cases). Factors relevant to determinations of whether to exclude include: "(1) the history of the litigation; (2) the sanctioned party's need for the precluded evidence; (3) the sanctioned party's justification (or lack of one) for its late disclosure; (4) the opponent-party's ability to overcome the late disclosure's adverse effects - e.g., the surprise and prejudice associated with the late disclosure; and (5) the late disclosure's impact on the district court's docket." Esposito v. Home Depot U.S.A., Inc., 590 F.3d 72, 78 (1st Cir. 2009) (citing cases). District courts, however, have broad discretion in this regard and "may choose a less severe sanction." Id. at 77-78 (citing cases).
Against this backdrop, it is clear that the burden to prove a Rule 26(a) violation lies with NMI. In re Celexa and Lexapro Mktg. & Sales Practices Litis., 2017 WL 9324342. at *1. And supplementation of an interrogatory is only due if "a party learns that a response is in some way incomplete or incorrect" and "additional corrective information has not otherwise been made known to the other parties." Clark v. Grimes, No. 00-11314-MLW, 2007 WL 9797802, at *3 (D. Mass. Oct. 19, 2007) (citing Fed.R.Civ.P. 26(e)). NMI has not shown that supplementation was required. First, it contends that supplementation was required in 2021 after Kaeli inspected the code, because he became aware then of the removal of the [Redacted] variable. To demonstrate Kaeli's knowledge, they point to his June 15, 2022 deposition testimony and documentation they provided [other] to Defendants, which demonstrate only the [Redacted] variables were used in the study. D. 289 at 14-15. While Kaeli had questions about Scott's testing when he inspected the actual tests, he clarified that he became aware of the change after Matveev "actually shared that he modified the code." D. 289-3 at 19, 26.
NMFs argument that Defendants should have "exercise[d] enough diligence to identify the change at issue" also fails. D. 316 at 5; see D. 289 at 15. It was not Defendants' obligation to identify the change at issue; if anything, it was NMI's obligation to disclose it. The only reason NMI provides for failing to do so is Scott's deposition testimony, in which he states that he did not believe it was "necessary" to "spell out how [he] invoked the two compilers with the same parameters," at any time, because NMI produced information and code relating to his tests. D. 302-11 at 9. While this might be true, NMI's claims for misappropriation of its trade secrets rely heavily upon Scott's comparison testing of the parties' respective codes. Any changes to the parameters of same-even ones NMI contends are innocuous-would be highly relevant to both parties.
Finally, NMI contends that Defendants failed to update their interrogatory response after Matveev's February 2022 deposition-at which point Defendants concede they became aware of the removal of the [Redacted] variable. D. 289 at 14. Defendants' rebuttal testing, however, did not exist before Scott's April 8, 2022 opening expert report. D. 328 at 5 n.4. Further, Defendants expressed their concerns about the removal of the variable prior to the deadline for opening expert reports through two emails in March. See D. 302-8 at 2; D. 302-9 at 2. Rule 26(e) provides that no supplementation is required where "the additional or corrective information has ... otherwise been made known to the other parties during the discovery process or in writing." Fed.R.Civ.P. 26(e)(1)(A). NMI responds that these emails are "vague" and do not sufficiently explain why Defendants believe the specific [Redacted] input parameter is important, explain why it renders Scott's testing unreliable, announce Defendants' intent to rely upon the variable, or provide any details about the tests Defendants would prospectively run. D. 316 at 8. But, at that time, Defendants had not yet run any testing of their own to substantiate the significance of the removal of the [Redacted] variable.
The cases upon which NMI rely also leave the Court unpersuaded. See, e.g., Asia Vital Components Co. v. Asetek Danmark A/S, 377 F.Supp.3d 990 (N.D. Cal. 2019); Pavo Sols. LLC v. Kingston Tech. Co., Inc., No. 8:14-cv-01352-JLS-KES, 2019 WL 8138163 (CD. Cal. Nov. 20, 2019), affd. 35 F.4th 1367 (Fed. Cir. 2022); Apple. Inc. v. Samsung Elecs. Co., Ltd., No. 11-CV-01846-LHK, 2012 WL 3155574 (N.D. Cal. Aug. 2, 2012). In each of these cases, the courts granted motions to strike for reasons not implicated here. In Asia Vital Components Co., 377 F.Supp.3d at 1004, untimely expert opinions relating to a non-infringement theory were struck because there was "unrebutted evidence that [the proponent] was fully aware of this theory at the time it received the interrogatory." In Pavo Sols. LLC, 2019 WL 8138163, at *21-22, the court struck a plaintiffs expert's opinion about licensing because, even though the plaintiff had previously produced certain licenses, the plaintiff did not update a prior, related interrogatory response to include that it intended to rely upon and use those licenses to prove its claims. In Apple, Inc., 2012 WL 3155574, at *5, the court struck non-infringement theories because the defendant's failure to timely disclose them "impeded [the plaintiffs] ability to conduct fact discovery on the undisclosed theories." Here, by contrast, it was NMI that did not affirmatively disclose the removal of the [Redacted] variable until February 2022 and it seems incongruous for it now to claim that such failure impeded any of its own discovery.
Moreover, "[a] rebuttal expert may cite new evidence and data so long as the new evidence and data is offered to directly contradict or rebut the opposing party's expert." Glass Dimensions, Inc. v. State St. Bank & Trust Co., 290 F.R.D. 11,16 (D. Mass. 2013) (citing cases); see Crawford-Brunt v. Kruskall, 489 F.Supp.3d 4, 9 (D. Mass. 2020) (citing cases); Cabi v. Bos. Children's Hosp., No. 15-cv-12306-DJC, 2017 WL 4038393, at *2 (D. Mass. Sept. 13, 2017) (citation omitted). NMI does not, and could not, reasonably contend that the purpose of Kaeli's report is not "to respond directly to the opinions and conclusions of [Scott], critique [his] methodology and characterization of the evidence, and draw conclusions favorable to the" Defendants. Cabi, 2017 WL 4038393, at *2. Its only refrain is that Kaeli's report is not rebuttal expert work because Defendants "willfully ignore that Dr. Scott's opening report was not
the first time that [NMI] disclosed either its intention to rely on his assembly code tests or the detail (and full source code) behind those tests - which Defendants have had access to since, at the latest, July 2021." D. 316 at 12 (emphasis in original). This seems to miss the point that the removal of [Redacted] variable was not affirmatively disclosed until February 2022 and Scott did not address its impact or lack thereof in his April 2022 report. Accordingly, NMI has not shouldered its burden in proving a Rule 26(e) violation.
For all these reasons, NMI's motion under Rule 26(e) to strike Kaeli's report, D. 287, is denied.
c) NMI's Alternative Request to Supplement Scott's Report and Defendants' Motion to Strike Scott's June 22, 2022 Declaration
In the same motion, NMI seeks alternative relief-namely, that Scott be afforded the opportunity to submit a supplemental report to respond to Kaeli's rebuttal report. D. 289 at 6. NMI argues that they have serious concerns about the method underlying Kaeli's tests and that they "will suffer prejudice if Defendants are permitted to advance this one-sided, inaccurate narrative before the jury without Dr. Scott being given an opportunity to respond." Id. at 17-20. Defendants oppose this request and move to strike a declaration Scott submitted in support of NMI's motion, which addresses the reasoning behind the removal of the [Redacted] variable and the concerns he has with Kaeli's testing method. D. 289-1; D. 301 at 23-24.
The Court concludes that the most appropriate course of action is to deny the introduction of a further supplemental report by Scott, but allow Scott's June 2022 declaration to remain in the record (and the opinion reflected in same to be offered at trial). NMI cannot deny that Scott knew about the removal of the [Redacted] variable. And according to NMI, Scott's supplemental report would address "whether the [Redacted] parameter is relevant to the question of misappropriation." D. 316 at 15. As such, the supplemental report would address information that it had access to at the time of their opening report, which would be improper. See, e.g., Marine Polymer Techs., Inc. v. HemCon. Inc., No. 06-cv-100-JD, 2010 WL 1427549, at *2 (D.N.H. Apr. 2, 2010) (stating that "[a] party may not use a supplemental report to disclose information that should have been disclosed in the initial expert report, thereby circumventing the requirement for a timely and complete expert report" (quoting 6 Moore's Federal Practice § 26.131 [2]) (internal quotation marks omitted)); Matthews v. Remington Arms Co., Inc., No. 07-1392, 2009 WL 1490691, at *1 (W.D. La. May 27, 2009) (stating that "an expert report that contains new opinions based on information available prior to the expiration of the expert report deadline is not supplemental" (citing cases)).
Nevertheless, given the complexity of the alleged trade secrets and testing conducted by both experts, allowing Scott's June 22, 2022 declaration explaining the reasoning behind the removal of the [Redacted] variable and the concerns he has with Kaeli's testing method to remain in the record is an equitable resolution to the matter. See Glass Dimensions, Inc., 290 F.R.D. at 17 (denying motion strike where case was "months from trial," there was "no history of litigation abuse by Plaintiff," and the "case [was] complex"). Moreover, any concerns regarding prejudice to either party have been diminished because they have both "availed themselves of the opportunity to fully question" each other's experts after the removal of the [Redacted] variable was affirmatively disclosed. Cabi, 2017 WL 4038393, at *2. Thus, to the extent that any failure by NMI to disclose the change to the code earlier can be considered a violation of Rule 26(e), it was harmless. While Defendants contend they have not had the opportunity to depose Scott in relation to the June 22, 2022 declaration, as it was filed after his June 2022 deposition concluded, D. 301 at 24, the Court highlights that the declaration itself is only four pages, the majority of which contains information to which Defendants and Kaeli already had access, see generally D. 289-1, and can be the subject of cross-examination at trial.
Accordingly, NMI's motion to allow a further supplemental report by Scott, D. 287, is denied, and Defendants' cross-motion to strike Scott's June 22, 2022 declaration, D. 297, is denied.
In reaching this decision, the Court considered Defendants' reply in support of their cross-motion to strike Scott's June 22, 2022 declaration. D. 328. Accordingly, the Court allows Defendants' motion for leave to file same nunc pro tunc, D. 325.
d) Defendants' Daubert Motion to Strike Scott's Opinion Regarding His Comparison Testing
Defendants have also moved to strike Scott's opinion regarding his comparison testing on Daubert grounds. D. 304. According to Scott, the primary purpose behind this comparison was to assess whether the FBGEMM and the NMIGEMM JIT generate functionally identical assembly code. D. 289-1 f 6; D. 319-2 ¶¶ 96, 127-35, 165-66, 190. For such a test to have any meaning, Scott says, both compilers had to be tested with the same inputs, which is why Scott made changes to both NMI's and Meta's underlying codes. D. 289-1 ¶¶ 12-14.
In his report, Scott explains that his comparison test demonstrates Defendants' use of all NMI's trade secrets. As to NMI's sparse instruction streaming, Trade Secrets Nos. 6-10, Scott opines that the code comparison demonstrates that both compilers use the same techniques [Redacted]
[Redacted]
[Redacted]
[Redacted] Id. ¶¶137-38.
In response to Defendants' complaint regarding the potential effects of the [Redacted] variable, Scott states that, to ensure that both compilers produced the same output when fed the same input, he made a change to Defendants' code, as their code performs an extra step, not present in NMFs code. D. 289-1 ¶¶ 12-14. As previously discussed, | [Redacted] As to compact FMA encoding, Trade Secret No. 11, Scott opines that the comparison test also demonstrates misappropriation of this trade secret. [Redacted]
As to FMA input buffering, Trade Secret No. 12, [Redacted] [Redacted] Scott opines that his comparison test also demonstrates misappropriation of same:
[Redacted]
Scott's report relies, in part, on his comparison testing to demonstrate misappropriation of each unique combination of techniques outlined in Trade Secrets Nos. 1-5. See D. 319-2 f 258 (explaining that "[additionally,. . . comparing the assembly code and source code of FBGEMM with that of [NMI]'s GEMM JIT also shows that Defendants deployed each of [NMI]'s three categories of trade secrets in FBGEMM"). He further contends that misappropriation of Trade Secrets Nos. 1-5 can be shown through his review of the prototype Zlateski shared with Meta days after he joined the company. Id. ¶¶ 251-54. In his report, he states that "Zlateski's [Redacted] code, written immediately after his arrival at [Meta], implements sparse instruction streaming and FMA input buffering precisely as described by [NMI] in its identification of trade secrets. Moreover, the [Redacted] explicitly indicates [Redacted] Id. ¶ 255. Misappropriation of these trade secrets, in Scott's view, can also be shown through a review of Zlateski's draft paper and Meta's presentation of his work at a December 2019 conference. Id. t¶ 259-63.
Defendants have moved to strike Scott's opinion relating to the comparison testing he performed as unreliable, irrelevant, unhelpful, and potentially misleading to the jury, both under Rule 702 and Rule 403. See generally D. 319. Specifically, they contend that Scott's opinion is based on source code that has been doctored for the purpose of litigation, skewing the input parameters and control metrics, thus providing an unreliable basis for his ultimate conclusions. See generally id.
Such deficiencies, however, in Scott's methodology are matters for cross-examination, not a Daubert challenge. See Iconics, Inc. v. Massaro, 266 F.Supp.3d 461, 470-71 (D. Mass. 2017) (concluding that "belatedly" explained modifications made to compare software functionalities are "matter[s] for cross-examination" where some grounding exists for each modification). Through his report, and June 22, 2022 declaration, Scott has sufficiently grounded his methodology and conclusions, and any challenges to same can be properly attacked through "[vigorous cross-examination, presentation of contrary evidence, and careful instruction on the burden of proof." Milward v. Acuity Specialty Prods. Grp., Inc., 639 F.3d 11, 15 (1st Cir. 2011) (quoting Daubert, 509 U.S. at 596) (internal quotation marks omitted).
Defendants' argument against admission of the opinion-namely, that both codes were modified-is not persuasive, particularly when Scott opines that there was no direct copying of NMFs code. D. 319-21253 (stating that "[w]hile the source code does not appear to be a direct copy of [NMIJ's code, it is my opinion that the 'dirty prototype' source code embodies each of [NMIJ's trade secrets"). Indeed, Scott's opinion is that comparison of the assembly code of compilers running input matrices of various sizes and levels of sparsity should not produce substantially the same results, unless the underlying compilers themselves are substantially the same. Id. f 128. Scott has also provided a reasonable and sufficient explanation of his methodology and for the removal of the [Redacted] variable. He explained that his comparison of the two compilers' outputs could only have meaning if they were fed the same inputs, but [Redacted] D. 289-1 ¶ 12 (italics in original). As a result, if he had not removed it from his comparison, approximately four percent of the tests he conducted would have been run with different input parameters. Id. ¶ 13.
Furthermore, Scott's actual methodology-code comparison-has been found to be a reliable approach under Daubeit even where the party alleged non-literal copying. See, e.g., Calendar Res. LLC v. StubHub. Inc., No. 2:17-cv-04062-SVW-SS, 2020 WL 4390391, at *4 (CD. Cal. May 13, 2020) (explaining that "[although code comparison is not the only way to prove trade secret misappropriation in technology cases, it is one of the most clear-cut methodologies" (citations omitted)); Xtec. Inc. v. CardSmart Techs., Inc., No. 11-22866-CIV-MORENO/HUNT, 2014 WL 10250974, at *5 (S.D. Fla. Dec. 4, 2014) (admitting expert's opinion regarding source code comparison, where qualified expert "was directed to points of concern by Plaintiff, and reviewed as much code as he believed necessary" and any concerns regarding his method could be raised during cross-examination); Wellogix, Inc. v. Accenture, LLP, 823 F.Supp.2d 555, 577 (S.D. Tex. 2011), aff d sub nom, 716 F.3d 867 (5th Cir. 2013) (admitting source code comparison because any disputes over whether particular functionalities were found in certain portions of infringing code but not others "was appropriately considered by the jury in terms of the weight to assign to each expert's testimony"); Comput, Assocs, Int'l v. Quest Software, Inc., 333 F.Supp.2d 688, 693-95 (N.D. Ill. 2004) (allowing source code comparison in non-literal copying case, explaining that discrepancies in analysis "go to the weight that [the court] give[s] [the expert's] evidence rather than to the admissibility"); Id. at 694 (highlighting that "[t]here does not appear to be any perfect way to compare millions of lines of source codes, especially in a case like this where the plaintiffs claim both literal and non-literal copying of the code").
Defendants offer citations to several cases, where courts struck expert testimony regarding testing that was manipulated or modified to yield a particular result. See, e.g., Earley Info. Sci., Inc. v. Omega Eng'g. Inc., 575 F.Supp.3d 242 (D. Mass. 2021); In re: Lipitor (Atorvastatin Calcium) Mktg., Sales Pracs. and Prods. Liab. Litis. ("In re: Lipitor"). MDL No. 2:14-mn-02502-RMG, 2016 WL 827067 (D.S.C. Feb. 29, 2016); Rosenthal Collins Grp., LLC v. Trading Techs. Int'l. Inc., No. 05 C 4088, 2007 WL 844610 (N.D. Ill. Mar. 14, 2007). Each of these cases is inapposite because the courts struck the respective experts for reasons not implicated here. In Earley Info. Sci., Inc., 575 F.Supp.3d at 248 (citing cases), the court excluded unreliable expert opinion about a test conducted, because the expert's "methods ha[d] not been tested; they ha[d] not been subjected to peer review or publication; they d[id] not have a known rate of error; there [were] no standards controlling their application or operation; and there [was] no evidence that they are generally accepted in the relevant scientific community;" whereas, code comparison testing with reasonable modifications have been accepted by other courts. In In re: Lipitor, 2016 WL 827067, at *3-4, the court struck an unreliable expert opinion, because the expert "deci[ded] to use [a second test] when the [first test] did not return a statistically significant p-value," which is not the case here. Finally, in Rosenthal Collins Grp., LLC. 2007 WL 844610, at *2-6, the court sanctioned the plaintiff and struck an expert declaration, because the expert represented that a particular function existed in the source code, which did not in fact exist, and modified the code to "tr[y] to pass it off as source code"-circumstances all of which are not present here.
Ultimately, Scott has provided a reasonable explanation for his comparison testing, changes to the underlying code, and broader methodology, thus creating a reasonable and reliable foundation for his opinion. As explained, Scott has shown that there is "an adequate fit" between his methodology and opinion and that his tests do not appear to "yield results that bear a dubious relationship to the questions on which he proposes to opine." Samaan, 670 F.3d at 31-32 (citing Daubert, 509 U.S. at 591-92). This satisfies Rule 702's demands that expert testimony be reliable, relevant, and helpful to the jury. Furthermore, having concluded that there is a reasonable basis for Scott's opinion, the Court cannot agree with Defendants that Scott's opinion has no probative value or will mislead or confuse the jury, or are substantially outweighed by their prejudicial effect for Rule 403 purposes. D. 319 at 17; D. 359 at 14.
Therefore, the Court denies Defendants' motion to strike Scott's opinion regarding NMI's comparison testing, D. 304.
To the extent Defendants request that the Court impose sanctions on NMI "by ordering it to pay costs and fees associated with motions and discovery necessitated by its and Dr. Scott's misconduct" regarding Scott's alterations to the underlying codes, D. 359 at 14 n.6; see D. 319 at 17 n. 15, the Court denies that request.
3. Whether NMI Adequately Identified Its Trade Secrets
Turning to the merits of NMI's misappropriation claims, Defendants first argue they are entitled to summary judgment because NMI has not defined its trade secrets with the particularity necessary to sustain a claim. D. 318 at 19-27; D. 351 at 11-15.
"A trade secret may consist of any formula, pattern, device or compilation of information which is used in one's business, and which gives him an opportunity to obtain an advantage over competitors who do not know or use it." J. T. Healy & Son, Inc. v. James A. Murphy & Son, Inc., 357 Mass. 728, 736 (1970) (citing Restatement of Torts, § 757, comment b). A plaintiffs description of its alleged trade secrets "must be made 'with clarity that can be understood by a lay person . . . and distinguish what is protectable from that which is not.'" Sutra. Inc., 2008 WL 2705580, at *4 (quoting Staffbridge, Inc. v. Gary P. Nelson Assocs., Inc., No. 024912BLS, 2004 WL 1429935, at *4 (Mass. Super. Ct. June 11, 2004)). To determine whether a proposed description is adequate, "courts have required specificity, although that specificity is highly fact dependent." Id. at *4 (citing cases).
At the outset, NMI appears to suggest that the identification issue was already resolved in its favor, relying upon the Court's decision to allow discovery to proceed based upon its' operative statement regarding Trade Secrets Nos. 1-12. D. 332 at 8, 19. As that decision concerned the identification of trade secrets for discovery, nothing precludes the Court from analyzing NMFs identification of its trade secrets based upon fuller briefing and argument here and to determine what the Court will consider for summary judgment and what will be admissible at trial. See Am. Student Fin. Grp. v. Aequitas Cap. Mgmt. Inc., No. 12-cv-2446-CAB (JMA), 2015 WL 11237638, at *10 (S.D. Cal. Feb. 12, 2015) (explaining that "[w]hether a trade secret plaintiff has introduced sufficient evidence to create a triable issue in a summary judgment proceeding is a different question from whether" the plaintiff complied with its pre-discovery identification obligation).
a) Trade Secret No. 1: Combination of Trade Secrets NMI has defined its first trade secret in the following manner:
[Redacted]
D. 318-2 at 4. Defendants take issue with such a description because trade secrets law does not protect ideas, but rather defined implementations of ideas. D. 318 at 20-21; D. 351 at 11-12. Defendants also argue that Trade Secret No. 1 is vaguely and indefinitely defined. D. 318 at 21; D. 351 at 12. On the other hand, NMI contends that Trade Secret No. 1 constitutes a factual discovery, which is considered a protectable trade secret. D. 332 at 23-24.
Within this district, "[t]he plaintiff must be clear about what information is protectable" and "a court should not have to 'sift through technical data to distill out a trade secret.'" Iconics, Inc., 266 F.Supp.3d at 456 (quoting TouchPoint Sols., Inc. v. Eastman Kodak Co., 345 F.Supp.2d 23, 28 (D. Mass. 2004)). And while a protectable trade secret may be "any confidential information used in a business that gives [the owner] an advantage over competitors who do not know or use it," Sensitech Inc. v. LimeStone FZE, 548 F.Supp.3d 244, 261 (D. Mass. 2021) (citation and internal quotation marks omitted), that does not give a would-be plaintiff license to define its trade secrets broadly or imprecisely, Sutra. Inc., 2008 WL 2705580, at *4.
Here, Trade Secret No. 1 does not constitute a trade secret or factual discovery; rather, it is an opening statement about what NMFs other alleged trade secrets can achieve and those achievements' potential worth. To be sure, Trade Secret No. 1 is not a "formula, pattern, device or compilation of information." J. T. Ffealy & Son. Inc., 357 Mass. at 736 (citing Restatement of Torts, § 757, comment b). Nor is it, unlike the other alleged trade secrets, a "unified process, design and operation of which, in unique combination, affords a competitive advantage and is a protectable secret." Sutra. Inc., 2008 WL 2705580, at *4 (quoting Integrated Cash Mgmt. Servs., Inc. v. Digit. Transactions. Inc., 920 F.2d 171, 174 (2d Cir. 1990)) (internal quotation marks omitted).
Furthermore, it is imprecisely defined. The imprecision with which NMI defines Trade Secret No. 1 is evident through its use of vague and ambiguous phrases such as "highly valuable improvements in throughput" and "independent economic value." D. 318-2 at 2. Other courts have granted summary judgment where the alleged trade secrets were vaguely and indefinitely defined. See, e.g,, Imax Corp. v. Cinema Techs., Inc., 152 F.3d 1161, 1167 (9th Cir. 1998) (concluding that "reasonable specificity could only be achieved by identifying the precise numerical dimensions and tolerances as trade secrets"); Big Vision Priv. Ltd. v. E.I. DuPont De Nemours & Co., 1 F.Supp.3d 224, 262-66 (S.D.N.Y. 2014) (granting summary judgment because the alleged trade secrets were defined with "vague and indefinite" terms, such as "high pigment levels," "suitably strong," "[predominantly," "minimal," and "expensive"); sit-up Ltd. v. IAC/Interactive Corp., No. 05 Civ. 9292 (DLC), 2008 WL 463884, at *11-12 (S.D.N.Y. Feb. 20, 2008) (granting summary judgment to defendant where many of plaintiffs trade secrets were "vague and ambiguous"). Neither Matveev nor Scott offered any definition of these terms, and simply stated that it depends on the context and application in which Trade Secret No. 1 is used. D. 352 at 133-43. Even assuming that to be true, it is still NMFs burden to provide that context and define its trade secrets "with clarity that can be understood by a lay person . . . and [to] distinguish what is protectable from that which is not." Staffbridge. Inc., 2004 WL 1429935, at *4.
NMI attempts to distinguish Big Vision Priv. Ltd., D. 332 at 24, but just like in that case, NMI was asked to define vague terms like "highly valuable improvements in throughput" and "independent economic value," D. 352 at 133-43. In the combined statement of material facts, NMI repeatedly argues that whether a given speedup is valuable does depend on the specific application, id, but that still requires the Court to "sift through technical data to distill out a trade secret," which is impermissible, TouchPoint Sols., Inc., 345 F.Supp.2d at 28.
Accordingly, the Court allows summary judgment to Defendants as to Trade Secret No. 1.
b) Trade Secrets Nos. 6-10. 12: Sparse Instruction Streaming and FMA Input Buffering
Trade Secrets Nos. 6-10 involve specific sparse general matrix multiplication techniques. See, e.g., P. 318-2 at 5 [Redacted] [Redacted]
These alleged trade secrets are sufficiently particular. They outline the algorithms of each sparse GEMM technique in detail and are accompanied by diagrams, demonstrating their functions. D. 318-2 at 5-6; see SMH Enters., L.L.C. v. Krispy Krunchy Foods, L.L.C., No. 20-2970, 2021 WL 4460522, at *14 (E.D. La. Sept. 29, 2021) (concluding that "plaintiff provided specific excerpts of source code, architectural design, database structure, and/or algorithms used to implement these purported trade secrets. By pointing to 'tangible trade secret material,' the sort of 'information' protected under the DTSA and [Louisiana's trade secrets act], plaintiff has satisfied the requirement that it identify the alleged trade secret with 'sufficient particularity").
Also as to Trade Secret No. 12 specifically, Defendants contend that no reasonable jury could conclude that they misappropriated this trade secret because [Redacted] NMI's technical expert Scott explained, however, that this is simply a variation of Trade Secret No. 12. D. 319-2 ¶ 190 (explaining that [Redacted] While Defendants' and their technical expert might disagree with this characterization, this only contributes to the idea that a genuine dispute of material fact exists as to whether Defendants misappropriated Trade Secret No. 12. Moreover, "[m]odifications in the process do not relieve a defendant of responsibility if the defendant's process is substantially derived from the trade secret." USM Corp. v. Marson Fastener Corp., 392 Mass. 334, 351 (1984) (citing cases). Accordingly, summary judgment on this basis also is not warranted.
c) Trade Secret No. 11: Compact FMA Encoding
Trade Secret No. 11 involves compact FMA encoding [Redacted] What is important to note, however, is that at the preliminary injunction stage of this litigation the burden was on NMI and injunctive relief "is an 'extraordinary and drastic remedy.'" Voice of the Arab World, Inc. v. MDTV Med. News Now. Inc., 645 F.3d 26, 32 (1st Cir. 2011) (quoting Munaf v. Geren, 553 U.S. 674, 689-90 (2008)). At this point, the burden has shifted to Defendants to prove there is no genuine dispute of material fact, such that no reasonable jury would be able to identify whether NMI has a valid trade secret in the form of Trade Secret No. 11.
[Redacted] Novel implementations or adaptions of known methods can benefit from trade secret protections in this district. See Incase, Inc. v. Timex Corp., No. 02-40022-FDS, 2005 WL 8176487, at *11 (D. Mass. Mar. 22, 2005) (explaining that "an adaptation, or improvement upon, existing" methods can be "a trade secret if the process as distilled accomplishes a result which gives the holder a competitive advantage due to his own ingenuity, research and development" (quoting Jet Spray Cooler, Inc. v. Crampton, 377 Mass. 159, 169 (1979)) (internal quotation marks omitted)). Defendants attempt to distinguish Incase. Inc. because the court there noted "there [wa]s no evidence that the [allegedly secret] design was common knowledge in the industry." Id. at *12. But Defendants have not proffered sufficient evidence to show that using compact FMA encoding in the way described in NMFs alleged Trade Secret No. 11 "was common knowledge in the industry." [Redacted] [Redacted]
Defendants further argue that Kaeli described Trade Secret No. 11 as "ambiguous and unspecific" because it "does not disclose or describe how it purports to reduce the size of instructions in a proprietary manner, leaving the reader guessing what the allegedly proprietary method of reduction may be." D. 318-4 at 168. [Redacted] Accordingly, the Court concludes there is evidence sufficient to create an issue of material fact on the question of whether Trade Secret No. 11 amounts to a protectable trade secret.
d) Trade Secrets Nos. 2-5: Unique Combinations of Trade Secrets
Finally, NMI's Trade Secrets Nos. 2-5 cover "one or more of the techniques described in [Trade Secrets Nos.] 6-10," "utilized on its own or augmented by" "the technique described in [Trade Secret No.] 11," "and/or" "the technique described in [Trade Secret No.] 12." D. 318-2 at 4-5. Defendants and Kaeli raise several arguments regarding NMI's identification of these alleged trade secrets.
First, Defendants argue that Trade Secrets Nos. 2-5 encompass "124 potential combinations and thus lack[] the specificity the law demands." D. 318 at 23; see also D. 351 at 13-14. However, as NMI correctly points out, "[t]his exact method of identifying combination trade secrets has been approved by this district - and was affirmed by the Federal Circuit, rejecting the exact argument Defendants advance." D. 332 at 24. The Federal Circuit has explained:
[the trade secret at issue there] ... is a collection of devices, each of which must contain what is described in the first half of the trade secret, together with one or more of the six listed additional features. That [the trade secret] may describe 63 (i.e., 26 -1) separate devices does not mean that there is a lack of specificity. Each of the 63 devices is specific. Indeed, [Defendant] does not even argue that any of the six features, or the possible combinations one or more of them with the common device features, is insufficiently specified. CardiAQ could have listed each of the 63 devices separately for the jury; [the trade secret] just simplified the presentation to the jury. That grouping, to which [Defendant] agreed, introduced no specificity problem.CardiAQ Valve Techs., Inc. v. Neovasc Inc., 708 Fed.Appx. 654, 662 (Fed. Cir. 2017); see also CardiAQ Valve Techs., Inc. v. Neovasc Inc., No. 14-cv-12405-ADB, 2016 WL 6465411, at *13 n.6 (D. Mass. Oct. 13, 2016) (highlighting that "[i]n any event, CardiAQ identified [the trade secret] with sufficient detail, describing its TMVI design with particularity").
Defendants attempt to distinguish the Federal Circuit's analysis on the bases that the decision was an unpublished affirmance of a decision denying a motion for a new trial, the parties agreed on the description (unlike the parties here), the defendant did not challenge the description until after trial (unlike Defendants here), and the defendant did not argue that any of the devices or features were insufficiently specified (unlike Defendants here). D. 351 at 13-14. The Court is unpersuaded. Although the defendant in CardiAQ agreed on the description and did not raise any challenges to the underlying devices' specificity, the Federal Circuit ultimately concluded there was "no specificity problem" with the trade secret and "agree[d]" with the district court that a reasonable jury could conclude that it was a protectable trade secret. CardiAQ Valve Techs., Inc., 708 Fed.Appx. at 662-63. Likewise, Trade Secrets Nos. 2-5 "just simplified the presentation to the jury" and "introduced no specificity problem." Id. at 662. And for the reasons discussed above, each component part of Trade Secrets Nos. 2-5 is sufficiently specific.
Second, Defendants argue that NMI has not sufficiently identified Trade Secrets Nos. 2-5 because it has not shown how each combination is "unique and not publicly-known." D. 318 at 24 (citing Big Vision Priv. Ltd., 1 F.Supp.3d at 271); see also D. 351 at 14. Kaeli similarly argues that these trade secrets were "disclosed publicly and/or would have been known to, or readily ascertainable by, an engineer or other person of skill in the industry before Defendants' alleged misconduct." D. 318-4 at 227, 235, 238-40. Neither argument warrants summary judgment in Defendants' favor. In Big Vision, the court explained that "[h]aving failed to describe even the components of its claimed trade secret with particularity, Plaintiff has similarly failed to assert a unified process, design, and operation; as such, its trade secret claim must fail." Big Vision Priv. Ltd., 1 F.Supp.3d at 271 (citing cases). Here, on the other hand, NMI has described the components of its claimed trade secrets with particularity. Additionally, Kaeli does not state that these trade secrets were publicly and explicitly disclosed; rather, he suggests these trade secrets were readily ascertainable by an engineer or other person of skill in the industry with intimate familiarity with several relevant sources. D. 318-4 at 359-60, 368-69, 371-75.
Third, Defendants contend that Trade Secrets Nos. 3-5 fail to identify trade secrets because they include "generic software categories"- [Redacted] Relying upon SMH Enters, v. Krispy Krunchy Foods, in which the court explained "generic software categories, by themselves, [are] insufficient to establish a trade secret without reference to the specific codes, interfaces, and designs it used," Defendants claim Trade Secrets Nos. 3-5 fail as a matter of law. SMH Enters., 2021 WL 4460522, at *13. The court there, however, found two trade secrets insufficiently particular because they only listed generic software categories and did "not identify the specific implementation methods or combinations" involved in the trade secrets. Id. Here, NMI specifically identified three techniques, as discussed above, involved in its trade secrets. To be sure, the SMH Enters, court found other trade secrets that included "specific excerpts of source code ... and/or algorithms" sufficiently particular to create an issue of material fact to be presented to a jury. Id. at *14.
Similarly, the Court concludes that an issue of material fact exists on the question of whether Trade Secrets Nos. 2-5 amount to protectable trade secrets.
4. Whether NMI Possessed Each of Its Trade Secrets
Defendants also argue they are entitled to summary judgment because NMI has failed to present evidence from which a reasonable jury could conclude that NMI possessed its trade secrets. D. 318 at 19-27; D. 351 at 6-11. Defendants primarily press two arguments as to why NMI has not demonstrated possession of its trade secrets.
First, Defendants rely upon the fact NMI only presented a list of documents and source code that relate to its development and possession of each trade secret. D. 318 at 21-27; D. 351 at 6-8. Defendants contend this is improper, citing Calendar Res. LLC, where a court granted summary judgment after the plaintiff "merely list[ed] hundreds of documents that allegedly reflect[ed] [its] trade secrets, but does not show where those trade secrets exist in the record." Calendar Res. LLC. 2020 WL 4390391, at *9 (quoting X6D Ltd. v. Li-Tek Corps., No. CV 10-2327-GHK (PJWx), 2012 WL 12952726, *8 (CD. Cal. Aug. 27, 2012)) (internal quotation marks omitted). In Calendar Res. LLC, the list the Court found lacking was actually "an extensive list of programming terms and concepts that could constitute trade secrets //they did exist." Id. (italics in original). This paints a different picture than the facts at issue here. NMI produced specific documents (though numerous) and a handful of source code files, which they argue show possession of its trade secrets. D. 352 at 143-48, 207-11, 227-30, 241-14, 254-57, 298-321, 324-27, 347-50: see ScentSational Techs., LLC v. PepsiCo. Inc., No. 13-CV-8645 (KMK), 2017 WL 4403308, at *10 (S.D.N.Y.Oct. 2, 2017) (determining "a reasonable jury could conclude that Plaintiff possessed" its trade secrets because they produced a "75-paragraph declaration with citations to specific documents that support Plaintiffs claim that it possessed trade secrets during the relevant time period").
Second, Defendants argue that NMFs technical expert, Scott, failed to articulate in his report how the aforementioned documents and source code demonstrate possession of its trade secrets. D. 318 at 22-23, 25-27; D. 351 at 9-11. Because expert testimony is necessary to demonstrate possession in complex technology cases, the alleged deficiencies in Scott's report are, they contend, fatal to NMFs claims. D. 318 at 22-23, 25-27; D. 351 at 9-11. However, in Scott's deposition, he explained that "[NMI] clearly possessed the trade secrets before July of 2019, and that fact is woven throughout [his] report and is a firm part of [his] opinion." D. 318-10 at 22. He then proceeded to offer examples of specific paragraphs in his report where this fact was included, before being asked a separate question. Id. at 4. Further, in his report, he discusses at length the limitations of the prior art in the field, and how the NMFs alleged trade secrets work and address those same limitations. D. 319-2 ¶¶ 40-78; D. 332 at 21-23. Scott and NMI also explain that its source code files embody each of NMFs trade secrets. D. 318-10 at 25-27; D. 332 at 21. Ultimately, Defendants' retort is that the documents upon which NMI and Scott rely upon to show possession have been examined by Defendants' technical expert Kaeli, who reached an opposite conclusion. D. 318 at 22, 24-27. But the fact that Defendants' technical expert reached a conclusion opposite to that of NMI's only cements this Court's conclusion that a genuine dispute of material fact exists as to possession.
Furthermore, during Matveev's deposition, he highlighted certain specific lines of code, which he contends implement NMI's trade secrets. See, e.g., D. 333-35 at 9-10 (explaining that [Redacted] This testimony, according to Defendants, is insufficient because Matveev only identifies code classes, which comprise multiple functions and variables, so the testimony is not sufficiently particular. D. 351 at 8-9. When viewed in its totality, however, including not only Matveev's testimony but the documentation and source code provided by NMI, the Court concludes that NMI has produced sufficient evidence such that a reasonable jury could conclude that it possessed its trade secrets, even as it will be hotly contested issue at trial.
5. Whether NMI Took Reasonable Measures to Protect Its Trade Secrets
To prevail on a trade secret claim under MUSTA or DTSA, a plaintiff must demonstrate that it took sufficient reasonable precautions to protect its secret information. In determining whether NMI took reasonable steps to preserve the secrecy of the claimed trade secrets, the Court considers several relevant factors:
(1) the existence or absence of an express agreement restricting disclosure, (2) the nature and extent of security precautions taken by the possessor to prevent
acquisition of the information by unauthorized third parties, (3) the circumstances under which the information was disclosed ... to [any] employee to the extent that they give rise to a reasonable inference that further disclosure, without the consent of the possessor, is prohibited, and (4) the degree to which the information has been placed in the public domain or rendered "readily ascertainable" by the third parties.USM Corp., 379 Mass. at 98 (alteration in original) (quoting Kubik, Inc. v. Hull, 56 Mich.App. 335, 356 (1974)) (internal quotation marks omitted). As the reports of the parties' reasonable measures experts are relevant to this inquiry, the Court will first address the parties' motions to strike same.
a) Defendants' Motion to Strike Rowe's Opinion
NMI has retained Rowe to provide an expert opinion as to whether NMI took reasonable protective measures. Defendants seek to strike Rowe's opinion because they contend she does not have the requisite qualifications, her opinion rests on an unreliable foundation, and her opinion will usurp the jury's role. D. 322; D. 360.
(1) Rowe's Qualifications
Defendants first contend Rowe is not sufficiently qualified because her experience in the law of trade secrets does not qualify her as an expert on NMFs security measures. D. 322 at 10-14; D. 351 at 5-9. Specifically, they argue her work as an academic is not exclusively focused on protective measures issues, she could not remember all the trade secret matters she worked on while in private practice, and her consulting experience in this area is limited. D. 322 at 10-14; D. 351 at 5-9.
Rowe is a professor of law, who specializes in intellectual property law, and whose scholarship focuses on trade secret law. D. 322-2 at 4 (Rowe Rpt.). Her scholarship also includes but is not limited to the first casebook in the United States devoted exclusively to trade secrets, a forthcoming Trade Secret Case Management Judicial Guide, and over a dozen publications dealing with issues raised in trade secret cases, including protective measures. Id. at 6-7. She also previously served as a reasonable protective measures expert for several companies. Id. at 7-8; D. 322-3 at 24. While the exclusive focus of her work as an academic in the field of trade secrets law has not been protective measures, "[r]elying upon the Supreme Court's direction that Rule 702 be 'interpreted liberally in favor of the admission of expert testimony,' the First Circuit has held that 'expert witnesses need not have overly specialized knowledge to offer opinions.'" Strahan v. Sec'v. Mass. Exec. Off of Energy and Env't. Affs., No. 19-CV-10639-IT, 2021 WL 2267180, at *3 (D. Mass. June 3, 2021) (quoting Levin v. Dalva Brothers. Inc., 459 F.3d 68, 78 (1st Cir. 2006)). Furthermore, Defendants' focus on Rowe's apparent inability to recall details of her relevant experience in private practice is not a basis to exclude her opinion. To begin, while Rowe could not remember exactly how many trade secret matters she worked on while in private practice, she testified that it was at least "dozens." D. 322-3 at 13. Importantly, courts place significance on an expert's ability to recall salient details of their past work, only where their expertise is "based wholly on that work." Calisi v. Abbott Lab'ys. No. 11-10671-DJC, 2013 WL 5441355, at *11 (D. Mass. Sept. 27, 2013) (explaining that "[w]here [an expert] 'could not recall such salient details, [he has not shown] expertise sufficient to pass Daubert's qualifications threshold based wholly on that work'" (third alteration in original) (quoting Carlucci v. CNH Am. LLC, No. 10-12205-DPW, 2012 WL 4094347, at *4 n.2 (D. Mass. Sept. 14, 2012))). Rowe's expertise is based not only on her time in private practice, but also on her decades as an academic and researcher. Additionally, while her work as a consultant is not as extensive as her academic and research history, she has served as a trade secrets expert on reasonable protective measures issues for several companies. D. 322-3 at 27.
Second, Defendants argue that Rowe's opinion should be excluded because she does not have a highly technical understanding of the security measures put in place. D. 322 at 14-17; D. 351 at 5-9. "Rule 702 is not so wooden as to demand an intimate level of familiarity with every component of a transaction or device as a prerequisite to offering expert testimony." Microfinancial, Inc. v. Premier Holidays Int'l., Inc., 385 F.3d 72, 80 (1st Cir. 2004) (citing Daubert, 509 U.S. at 594). Experts who do not hold technical degrees have been allowed to opine on reasonable measures. See Proofpoint, Inc. v. Vade Secure, Inc., No. 19-CV-04238-MMC, 2021 WL 2588974, at *1 (N.D. Cal. June 24, 2021) (concluding that a proposed reasonable measures expert was qualified based on his experience practicing law, advising clients on the design or improvement of trade secret management, and serving as the Deputy Director General of the World Intellectual Property Organization). In their reply, Defendants argue that without such technical expertise, her testimony will not assist the trier of fact. D. 360 at 6. Even without such technical expertise, Rowe's expertise would still be helpful to the jury by contextualizing, based upon her academic and consulting work, what level of measures similar organizations, companies, and the broader industry take to protect their trade secrets. D. 322-2 at 9-10 (explaining that "[Reasonableness is considered in the context of the particular organization including its size, and stage of the business, and the nature of the relationships with those who have access to the information" and that "[d]espite being a start-up [NMI]'s founders instituted a culture that emphasized secrecy and guarded its algorithms closely"); D. 346 at 12 (contending that "Professor Rowe is uniquely positioned to assist the trier of fact to contextualize the reasonable measures NMI took at the time of the misappropriation based on her involvement with industry professionals and scholars at the Sedona Conference Working Group on Trade Secrets, expansive scholarship and publications, and years spent as a practitioner focused on trade secret cases"). She can further explain how certain security measures are viewed in the field. See, e.g., D. 322-2 at 12 (explaining that "SSH keys are known in the industry for providing cryptographic strength that passwords alone cannot offer"); Id. at 14 (explaining that "[a]ll of the Tier 4 data centers utilized by [NMI] were 'SOC 2' (Security Operation Center 2) compliant data centers, which means the physical security met well known industry standards"). All of which will be helpful to a lay jury in determining if NMI has met its burden to show that it took reasonable measures to protect its trade secrets.
Accordingly, by "knowledge, skill, experience, training, or education," Fed.R.Evid. 702, Rowe is sufficiently qualified to offer her proffered opinion.
(2) Reliability of Rowe's Opinion
Defendants also challenge Rowe's opinion as unreliable on three bases: (1) that she relied heavily upon the allegations in the amended complaint, (2) that a basis of her opinion was an interview with John O'Hara ("O'Hara"), an NMI employee who joined the company after Zlateski left to join Meta; and (3) that she ignored other sources like Zlateski's deposition testimony and Defendants' discovery responses or pleadings. D. 322 at 18-21; D. 360 at 11-13.
Defendants' concerns about Rowe's reliance on the amended complaint and her failure to consider other sources also do not warrant exclusion of her opinion. As to her reliance on the allegations in the complaint, Defendants point to the footnotes in Rowe's report in which she cited to the amended complaint multiple times, D. 322 at 18; D. 360 at 12, and the notion that "federal courts across the country have found that expert opinions that rely primarily upon allegations in a complaint are not 'based on sufficient facts and data,' as required by Rule 702, and are otherwise unreliable," Chen v. Yellen, No. 3:14-cv-50164, 2021 WL 4192078, at *3 (N.D. Ill. Sept. 15, 2021) (citations and internal quotation marks omitted). Rowe, however, has explained that her footnotes are not "an exclusive list of every single thing supports that statement" or an exclusive list of everything she considered. D. 346-3 at 14. She also appended a list of materials considered in forming her opinion, which are not limited to the allegations in the complaint, but also include several declarations from Shavit, O'Hara, and Bryan House ("House"), NMFs second employee, this Court's Memorandum and Order denying preliminary injunctive relief, deposition transcripts (including Zlateski's deposition in relevant part and House's and Matveev's depositions), interrogatory responses, responses to requests for admission, and various other documents. D. 322-2 at 34; D. 346-3 at 17.
Similarly, Defendants' further concern about Rowe's reliance on an interview with O'Hara does not persuade the Court that Rowe's opinion lacks a reliable foundation. While O'Hara did not work at NMI at the time of the alleged misappropriation, he was NMI's Rule 30(b)(6) witness on the very topic of protective measures and explained NMI's security measures in detail, such as protecting their confidential information through "two-factor authentication," "VPNs, firewalls, and Tier 4 data centers." D. 322-6 at 8-10. While Robinson opines that Rowe's assumptions about the reasonableness of certain security measures, such as confidentiality and non-disclosure agreements, the use of SSH keys to access code, and several of NMI's other protective measures, are incorrect, D. 371-2 ¶¶ 62, 64-65, 80, 138, 142, 151-57, such matters "which question the factual underpinnings of an expert's investigation, often go to the weight of the proffered testimony, not to its admissibility. ... As such, these matters are for the jury, not for the court. This is as it should be; the district court's gatekeeping function ought not to be confused with the jury's responsibility to separate wheat from chaff." Crowe v. Marchand, 506 F.3d 13, 18 (1st Cir. 2007) (internal citations omitted). Accordingly, exclusion is not warranted on this ground either.
"Tier 4" refers to the level of reliability and security associated with a particular data center. D. 322-6 at 8.
(3) Usurping the Jury's Role
Defendants contend that Rowe will usurp the jury's role by offering legal conclusions, namely, her conclusion that "under the circumstances, [NMIj's Proprietary Information was the subject of reasonable precautions, and therefore meets the reasonable efforts requirement under the Massachusetts Uniform Trade Secrets Act and the Defend Trade Secrets Act." D. 322 at 18 (citing D. 322-2 at 4); see also D. 360 at 9-10.
"[I]t is not for witnesses to instruct the jury as to applicable principles of law, but for the judge." Nieves-Villanueva v. Soto-Rivera, 133 F.3d 92, 99-103 (1st Cir. 1997) (citation and internal quotation marks omitted) (holding that the district court "egregious[ly]" erred in admitting expert testimony that employment-discrimination plaintiffs had been hired in violation of Puerto Rico law but finding error harmless where judge instructed jury that they must follow the law as stated by the judge). As such, "[e]xpert testimony proffered solely to establish the meaning of a law is presumptively improper." Bartlett v. Mut. Pharm. Co., 742 F.Supp.2d 182, 188 (D.N.H. 2010) (quoting United States v. Mikutowicz, 365 F.3d 65, 73 (1st Cir. 2004)) (internal quotation marks omitted). Expert opinion regarding what a law means or a law's application to a particular set of facts may be struck. Karp v. CIGNA Healthcare, Inc., 882 F.Supp.2d 199, 205 n.4 (D. Mass. 2012) (granting motion to strike expert declaration to the extent that it consisted of legal arguments and legal conclusions (citing cases)); Bartlett, 742 F.Supp.2d at 189 (allowing an expert to testify regarding industry practice in product liability case because industry practice is relevant to whether defendant acted with due care, but may not testify regarding FDA policy where that policy deals with the laws that are at issue in the case).
The Court, therefore, will allow Rowe to opine as to the reasonableness of NMFs protective measures, but not as to whether those measures meet the statutory requirements for Counts I and II at issue before the jury. See Ji v. Bose Corp., 538 F.Supp.2d 354, 360 (D. Mass. 2008) (striking only portions of expert's report expressing legal conclusions).
(4) Proposed Limitations to Rowe's Opinion
In the alternative, Defendants ask the Court to limit Rowe's opinion in four ways: (1) given Rowe's disavowal of knowledge regarding the technical aspects of SSH keys, secure data centers, or NMFs networks, she should not be permitted to offer any opinion on them; (2) given that Rowe deemed facts after July 2019 irrelevant to her analysis and a reasonable measures analysis only turns on the measures taken at the time of the alleged misappropriation, she should not be permitted to testify or opine based on facts postdating July 2019; (3) given Rowe's inability to even approximate a specific dollar amount, she should not be permitted to offer the following opinion found in her report: "It is my understanding that [NMI] invested significant capital by purchasing costly hardware and running its own networks to ensure its information was protected;" and (4) she should not be permitted to opine on NMFs beliefs or knowledge, specifically when she states that NMI "was under the reasonable belief that Dr. Zlateski would be working on a team that was unrelated to his work at [NMI]" and that "[NMI] relied on Dr. Zlateski's representations." D. 322 at 21-23; see D. 360 at 13-14.
As to the first proposal, it does not appear that NMI explicitly addressed it in its briefing. See generally D. 346. Moreover, Rowe explicitly stated she had no technical understanding of how those security measures work. For these reasons, to the extent that she was going to offer any opinion in this vein, Rowe will not be permitted to offer an opinion regarding the technical aspects of same, but she may offer an opinion about whether certain measures, like SSH keys and Tier 4 data centers, for example, are standard in the industry, and about them as security measures taken by NMI or other similarly situated companies. D. 322-2 at 12 (explaining that "SSH keys are known in the industry for providing cryptographic strength that passwords alone cannot offer"); Id. at 14 (explaining that "[a]ll of the Tier 4 data centers utilized by [NMI] were 'SOC 2' (Security Operation Center 2) compliant data centers, which means the physical security met well known industry standards").
As to limitation 2, Rowe has consistently taken the position that measures taken after misappropriation were not relevant to her analysis. See, e.g., D. 322-2 at 10 (stating that "[t]o the extent that any information was disclosed after Defendants' misappropriation, it would have no bearing on my analysis. The only relevant facts are those that existed at the time of the misappropriation"); D. 346-3 at 11 (testifying that she only analyzed the measures taken "[f]rom [NMI's] formation through [July 2019]"), and she has not proffered an opinion regarding any post-July 2019 measures.
Boiled down to their essence, the last two limitations that Defendants propose take issue with Rowe's interpretations of the facts at issue in this case-specifically, that it is her understanding that NMI "invested significant capital by purchasing costly hardware and running its own networks to ensure its information was protected" and that Zlateski "informed [NMI] that his role at [Meta] would be in no way related to his work at [NMI] . . . [so] [NMI] was under the reasonable belief that... Zlateski would be working on a team that was unrelated to his work at [NMI]." D. 322-2 at 15. While Meta's expert Robinson states that NMI acted unreasonably by not investing sufficient capital in reasonable protective measures and by failing to properly address Zlateski's departure from NMI, D. 371-2 ¶¶ 138,142, 153, 155, these facts are at the center of the factual disputes undergirding this case and Defendants' disagreement with the facts as Rowe sees them is not grounds for exclusion for her opinion, Int'l Adhesive Coating Co. v. Bolton Emerson Int'l, Inc., 851 F.2d540,545 (IstCir. 1988) (rejecting "arguments against admissibility [that were] simply a rehashing of the central factual disputes of the case dressed up as attacks on the expert's testimony through Rule 703"). Disputed facts may serve as the basis for expert testimony. See United States ex rel. Dver v. Raytheon Co., No. 08-10341-DPW, 2013 WL 5348571, at *8 (D. Mass. Sept. 23, 2013) (explaining that "[t]hese cases do not stand for the proposition that disputed facts are an impermissible basis for expert testimony" (citing United States v. Shay. 57 F.3d 126, 133 n.5 (1st Cir. 1995)). Rather, "disputes over the facts on which an expert bases [her] opinions go to the weight of the testimony, not its admissibility." Id. at *9 (citing Crowe. 506 F.3d at 18).
Accordingly, Defendants' motion to strike Rowe's opinion is allowed as to any opinion regarding whether NMFs security measures meet the statutory elements of any claim, but she may opine about the reasonableness of such measures, and she is prohibited from offering any technical discussion of NMFs security measures. In all other regards, the motion to exclude Rowe's opinion, D. 305, is denied.
b) NMFs Motion to Exclude Robinson's Opinion
NMI also moves to exclude the opinion of Robinson, Defendants' reasonable measures expert, on two bases. D. 342.
First, NMI argues that Robinson should be prohibited from testifying because the reasonable measures assessment is statutorily limited to an analysis of the measures taken at the time of the misappropriation and Robinson admittedly considers actions taken by NMI up through and after the start of the instant litigation. D. 350 at 7-12; D. 375 at 6-12. In NMFs view, the latest possible date of misappropriation must have occurred prior to Meta's publication of the code to GitHub on October 25, 2019, so if Robinson is allowed to opine on NMFs actions post-dating this event, his opinion will necessarily mislead the jury in considering temporally irrelevant events. D. 350 at 7-12; D. 375 at 6-12.
A reasonable measures analysis under the MUTSA focuses on the measures taken at the time of the misappropriation. See Mass. Gen. L. c. 93, § 42(4)(ii) (defining "trade secret" as "specified or specifiable information . . . that... at the time of the alleged misappropriation was the subject of efforts that were reasonable under the circumstances"). The DTSA, on the other hand, does not make so explicit such a limitation. See 18 U.S.C. § 1839(3) (defining "trade secret" as "all forms and types of . . . information ... if... the owner thereof has taken reasonable measures to keep such information secret; and the information derives independent economic value"). Regardless, courts, including within this session, have seemed to recognize a temporal limitation. See Diomed. Inc. v. Vascular Sols., Inc., 417 F.Supp.2d 137,144 (D. Mass. 2006). It is also true, however, that courts have recognized that actions taken after the alleged misappropriation are relevant to whether information can be properly categorized as a trade secret and whether it was ever valued to begin with. See, e.g., J. T. Healy & Son. Inc., 357 Mass. at 739 (noting that "if the person entitled to a trade secret wishes to have its exclusive use in his own business, he must not fail to take all proper and reasonable steps to keep it secret," "cannot lie back and do nothing to preserve its essential secret quality" and "must exercise eternal vigilance"); Storage Tech. Corp. v. Custom Hardware Eng'g & Consulting Ltd., No. 02-12102-RWZ, 2006 WL 1766434, at *10 (D. Mass. June 28, 2006) (explaining that a trade secret plaintiffs "carelessness [in post-litigation actions] does not entitle defendants to prevail at summary judgment, though it may ultimately militate against a finding of trade secret status"); HiRel Connectors v. United States, No. CV01-11609 DSFVBKX, 2005 WL 4958547, at *5-6 (CD. Cal. Jan. 4, 2005) (ruling plaintiff "has not had [trade secret] protection since at least January 2001," when the "alleged trade secrets [were] published," because plaintiff "provide[d] virtually no explanation for its failure to act after" that date and "merely filing suit against the alleged wrongdoer does not constitute 'reasonable efforts'" (italics in original)); Alamar Biosciences, Inc. v. Difco Lab'ys. Inc., No. Civ-S-941856 DFL PAN., 1995 WL 912345, *6 (E.D. Cal. Oct. 13, 1995) (granting summary judgment for defendant where it was "undisputed that MicroScan strongly suspected Lancaster of misappropriating its trade secrets, but did nothing" and "MicroScan's failure to bring suit, or even approach and warn Lancaster establishes that MicroScan did not take reasonable steps to protect its trade secrets"). Even in this litigation, the Court has recognized that post-misappropriation actions can be relevant to a reasonable protective measures analysis and whether information can be viewed as a protected trade secret. D. 79 at 19-20 (expressing that "[w]hen, however, remnants of [the code] remained on GitHub,... reasonable measures may have also included demands for removal from GitHub,... particularly where there was some suggestion in the record that GitHub has a process for removal of sensitive data from its platform" (citations omitted)).
Following this rubric, the Court is not persuaded that Robinson's opinion must be struck, as it is relevant in determining whether the code at issue amounts to a protectable trade secret. See, e.g., J. T. Healy & Son, Inc., 357 Mass. at 739; Storage Tech. Corp., 2006 WL 1766434, at *10; HiRel Connectors, 2005 WL 4958547, at *5-6; Alamar Biosciences. Inc., 1995 WL 912345, *6. The better course of action, therefore, is to admit Robinson's opinion, as "[vigorous cross-examination, presentation of contrary evidence, and careful instruction on the burden of proof are the traditional and appropriate means of attacking shaky but admissible evidence." Milward, 639 F.3d at 15 (quoting Daubert. 509 U.S. at 596).
Second, NMI argues that Robinson should be prohibited from offering testimony that having a patent strategy is unreasonable, because it is common for companies to rely on trade secret protections prior to obtaining a patent. D. 350 at 13-15. However, neither in his report nor deposition does Robinson opine that having a patent strategy is unreasonable or that it automatically strips otherwise protectable trade secrets of the protection the law accords. See generally D. 350-5; D. 371-2. Rather, Robinson explains "the documents regarding NMFs patent strategy reflect that NMI intended to make public at least some of its intellectual property through its patenting processes." D. 371-2 ¶ 124. He further clarifies in his deposition that "[a]nd it's instances of [applying for a patent] and then what that means and how that is, you know, a much different strategy with protecting alleged trade secret information, and how that's - [his] expectation is that the employees would be aware of what that means. There would be some type of training or some type of guidance to help control and make sure that the implications of applying for a patent are understood by all those involved with it." D. 350-5 at 5. NMI worries this opinion "is simply wrong and would only serve to mislead a jury," D. 350 at 14, but NMI itself concedes "patent applications may be relevant to a protective measures analysis," Id. at 13-14. Whether Robinson was reasonable in giving NMFs patent strategy the weight he gave it in his analysis is for the parties to argue to the jury.
Accordingly, the Court denies NMFs motion to exclude Robinson's opinion, D. 342.
c) NMFs Reasonable Protective Measures
With both Rowe's and Robinson's opinions and the relevant facts in mind as to reasonable protective measures here, the Court considered the following. There was an express agreement, namely the NMI Contract, between NMI and Zlateski restricting disclosure of trade secrets, D. 333-4 at 8; D. 352 at 500-01, which falls in NMFs favor. Nevertheless, as Defendants correctly point out, some courts have recognized that having such an agreement is not enough. See, e.g., Diamond Power Int'l. Inc. v. Davidson, 540 F.Supp.2d 1322,1334 (N.D.Ga. 2007) (highlighting that "requiring all employees to sign generalized confidentiality agreements is generally not, standing alone, sufficient" (citing cases)); Dynamics Res. Corp. v. Analytic Sci. Corp., 9 Mass.App.Ct. 254, 276-79 (Mass. App. 1980) (concluding non-disclosure agreement alone is insufficient). Robinson also emphasizes that reliance solely upon such agreements does not satisfy industry standards. D. 371-2 ¶ 62 (expressing that "[t]he use of employee nondisclosure agreements alone is not sufficient protection for alleged trade secrets"); see id. ¶ 152.
Defendants' primary argument is that NMI allowed Zlateski to keep his company laptop, did not remove confidential information from the laptop, did not conduct an exit interview with him, and allowed him to maintain access to its source code and one of its Slack channels. D. 318 at27-28;D. 351 at 16-18. The context in which this occurred, however, suggests that a jury could find that NMFs actions might not have been unreasonable. All parties agree that, on Zlateski's last day at NMI, both parties voiced interest in continued collaboration and Zlateski staying on as a consultant at NMI. D. 352 at 41-45. While Robinson vigorously contests the propriety of NMFs actions in relation to Zlateski's departure, D. 371-2 ¶ 142 (expressing that "based on how [NMI] handled [] Zlateski's departure alone, in my view, [NMI] cannot show that it took reasonable steps to protect the allegedly confidential information at issue in this case"), granting a consultant access to the information necessary to consult does not seem to fall outside the realm of reason, cf SKF USA Inc v. Zarwasch-Weiss, No. 1:10-cv-1548, 2011 WL 13362617, at *19 (N.D. Ohio Feb. 3, 2011) (concluding defendant's access to plaintiffs electronically-stored business information while on administrative leave was reasonable where the parties were negotiating his position and he was bound by a non-disclosure agreement).
Defendants also argue that some of NMFs actions following the alleged misappropriation undermine any notion that they took reasonable precautions. Most notably, they rely on the facts that (1) Matveev published a copy of the FBGEMM code on his own GitHub page, where it was left for over a year; (2) besides sending letters to Defendants, NMI took no further actions to remove the code from GitHub; and (3) NMI did not preserve its audit logs, which could have determined who accessed its source code. D. 318 at 29; D. 351 at 17. None of these facts are the smoking gun Defendants believe them to be. First, Matveev has consistently testified that his fork was created to preserve evidence for this litigation and that his GitHub page was not linked to NMFs GitHub page and did not appear in the results of a Google search. D. 352 at 58-60. Second, there is evidence in the record that it would have been difficult to completely scrub the code from GitHub and that the best method of doing so would be a Digital Millennium Copyright Act takedown request. Id. at 616-19. Meta "believe[d]" such a request would have been "[in]appropriate" and it is unclear whether NMI could have submitted a successful request, without Meta's cooperation. Id. at 619. Third, it is unclear whether the audit logs would have recorded Zlateski's accessing NMFs repositories, because the audit log only tracked "anomalous or suspicious activity" and "changes to personal or team accounts." Id. at 367-71.
In deciding NMFs preliminary injunction motion, the Court highlighted that it was a "close call" whether NMI took sufficient reasonable steps to protect its trade secrets. D. 79 at 17. Even after discovery, this issue remains a hotly contested one. As such, a genuine dispute of material fact exists, and summary judgment on this ground is not warranted.
6. Whether Zlateski Acquired, Used, or Disclosed the Alleged Trade Secrets Through "Improper Means "
Even if the other requirements for NMFs trade secrets claim are satisfied, Defendants argue that the trade secret claims should still fail, because no jury could reasonably conclude that Zlateski acquired the information in dispute through "improper means," as required by the MUTSA and DTSA. See Mass. Gen. L. c. 93, § 42(1)-(2); 18 U.S.C. § 1839(5).
In support of their argument, Defendants first contend that Zlateski had sworn in his declarations to this Court that he drafted the code for Meta's FBGEMM from scratch and, to the extent that he relied upon outside sources in drafting same, those sources were his prior research, other public domain references, such as a paper written by Baoyuan Liu (the "Liu paper"), and his education. D. 318 at 30; D. 351 at 18-19. This argument alone is insufficient to warrant summary judgment because it rests on disputed facts. For one, Scott opines at length that the Liu paper did not fully address the problems solved by NMFs code and that Zlateski's pre-NMI work did not disclose the alleged trade secrets. D. 319-2 ¶¶ 42, 55-61, 123-25. Defendants' concern that these opinions are "incorrect," D. 351 at 19, only support the idea that a genuine dispute of material fact exists.
Defendants also argue that no reasonable jury could find improper means because NMI has allegedly produced no evidence of code copying. D. 318 at 30 (quoting Calendar Res., 2020 WL 4390391, at *4). The crux of NMFs evidence is Scott's comparison testing, in which he allegedly "demonstrate[ed] that NMFs compiler and FBGEMM employ the same algorithms to perform sparse matrix multiplication." D. 332 at 30. Defendants' only retort is that "[a]s Defendants and Dr. Kaeli have explained [in their Daubert motions], that testing is flawed, unreliable, and inadmissible because NMI and Dr. Scott falsified both parties' source code to try to force the results they needed, and then concealed the alterations made to Meta's code until Meta figured it out and forced the issue." D. 351 at 18. The Court, however, has already ruled that Scott's opinion regarding this comparison testing are admissible, which, along with the other contested issues, pose a genuine dispute of material fact.
Ultimately, there is evidence to support both Defendants' and NMFs view regarding Zlateski's alleged use of improper means, rendering summary judgment for Defendants on this basis inappropriate.
7. Whether Meta "Knew or Had Reason to Know" That Any Alleged Trade Secret Was Acquired by "Improper Means "
Independent of whether Zlateski misappropriated its trade secrets, NMI must still prove that Meta "knew or had reason to now" that the alleged trade secrets were acquired by "improper means." See KPM Analytics N. Am. Corp. v. Blue Sun Sci., LLC, No. 4:21-CV-10572-TSH, 2021 WL 2982866, at *14 (D. Mass. July 15, 2021) (noting that "MUTSA creates a private right of action for misappropriation of trade secrets by improper means and defines improper means broadly to include the use or disclosure of another's trade secret if the recipient knew or had reason to know that the knowledge was derived from or through a person who had a duty to limit its use, disclosure, or acquisition. ... DTSA contains very similar language" (citations omitted)).
Defendants argue that no reasonable jury could find that Meta knew or had reason to know that Zlateski's conduct was improper. D. 318 at 32-33; D. 351 at 20-21. To bolster their argument, Defendants recount that Zlateski "is a computer scientist of impeccable credentials," "the optimizations in dispute were well-known," and Meta was unaware of Zlateski's contract with NMI until after the alleged misappropriation. D. 318 at 33; see D. 351 at 20-21. As an initial matter, Defendants' contention that "the optimizations in dispute were well-known" is directly at issue in this case and belied by some of the comments made by Meta employees. See, e.g., D. 333-112 at 2 (providing chat messages from Meta employee Park describing Zlateski's code as "amazing"). Further, it appears to strain credulity that Meta would not have had reason to know about Zlateski's contract with NMI, given the fact-which Meta itself concedes-that the company forces its employees to undergo "onboarding video trainings" regarding contractual obligations such as non-competes. D. 318 at 35.
NMI, on the other hand, argues that several facts could lead a reasonable jury to find that Meta had reason to know that the algorithms Zlateski provided were confidential. These alleged facts include: (1) Zlateski submitted a resume to Meta describing NMI as a "stealth mode startup;" (2) Zlateski mentioned during his interview process prior work experience with "sparsity" and similar work done on behalf of the company Wayfair; (3) Zlateski mentioned to Meta employees soon after starting there that he had played with [Redacted]" (4) other Meta employees gave Zlateski's work high-praise; (5) Zlateski was given a training at Meta regarding violations of non-compete agreements; and (6) Meta did not discipline Zlateski after NMI contacted Meta in 2020 about the alleged misappropriation. D. 352 at 541-42, 558-59, 600-01, 622, 659-61.
Perhaps each of these facts, standing alone, would not necessarily suggest that Meta had "reason to know" that Zlateski's conduct was improper. A jury, however, could consider the totality of such circumstantial evidence in determining whether Meta had reason to know that Zlateski's conduct was improper. See Curtiss-Wright Corp. v. Edel-Brown Tool & Die Co., 381 Mass. 1, 6 (1980) (highlighting that "[k]nowledge or the likelihood that a defendant knew of the wrongful character of the disclosure can and often must be proved by the weight of credible circumstantial evidence" (citing R. Milgrim, Trade Secrets § 5.04[3], at 5-102 (1978)); Data Gen. Corp. v. Grumman Sys. Support Corp., 825 F.Supp. 340, 360 (D. Mass. 1993) (explaining "defendant's knowledge 'often must be proved by the weight of credible circumstantial evidence'" and the defendant's possession of plaintiff s trade secrets and its knowledge that "confidentiality agreements were widely used in the business to restrict disclosure of proprietary information" could allow the jury to infer that the defendant "was acting with either studied ignorance or actual knowledge" (quoting Curtiss-Wright Corp., 381 Mass. at 6)). For example, there is record evidence, including from Defendants' own reasonable measures expert Robinson, that nondisclosure agreements are industry standard, so Meta at least should have known that many of its new hires were subject to such agreements. See D. 371-2 at 51 (highlighting that "[r]elying on non-disclosure agreements with employees and third parties is a common starting point, and I expect companies to have many additional security controls in place in addition to non-disclosure agreements"); see also D. 318-36 at 11 (expressing that "[w]hat I will say is one of the first things that companies do when they hire employees is to remind them that they are still beholden to past confidentiality agreements. Something I've always, when I've hired people is remind them, you do not - do not - disclose any IP to us that was developed not here. That happens to be a common practice, and I've implemented that at every company I've ever been at, and I would imagine [Meta] would have done the same thing"). Moreover, "[i]t is elementary that at summary judgment a court must view the record in the light most favorable to the nonmoving party and draw all reasonable inferences in favor of the same." Chadwick v. WellPoint Inc., 561 F.3d 38, 41 (1st Cir. 2009).
For these reasons, the Court concludes that NMI has proffered sufficient evidence to put this issue to a jury.
C. Chapter 93A (Count III)
Defendants have moved for summary judgment against NMI's claim for violations of Mass. Gen. L. c. 93A, § 11-namely, for engaging in unfair and deceptive business practices. For such claim, a plaintiff must demonstrate that the conduct at issue "occurred primarily and substantially within the commonwealth." Mass. Gen. L. c. 93A, § 11 (requiring that "[n]o action shall be brought or maintained under this section unless the actions and transactions constituting . . . the unfair or deceptive act or practice occurred primarily and substantially within the commonwealth"). This statutory requirement is the only aspect upon which Defendants move for summary judgment as to Count III. D. 318 at 36-37; D. 351 at 23-24.
"Whether the 'actions and transactions [constituting the § 11 claim] occurred primarily and substantially within the commonwealth' is not a determination that can be reduced to any precise formula," and "[a]ny determination necessarily will be fact intensive and unique to each case." Kuwaiti Danish Comput, Co. v. Digit. Equip. Corp., 438 Mass. 459, 472 (2003) (alteration in original). In making this determination, only the defendant's unfair and deceptive acts are relevant; other contacts with Massachusetts "do not play a part in this assessment." Armstrong v. White Winston Select Asset Funds, LLC, No. 16-10666-JGD, 2022 WL 17981392, at *24 (D. Mass. Dec. 27, 2022) (quoting AECOM Tech. Servs. Inc. v. Mallinckrodt LLC, 117 F.Supp.3d 98, 106 (D. Mass. 2015)) (internal quotation marks omitted). The First Circuit has also cautioned "that, if the significant contacts of the competing jurisdictions are approximately in the balance, the conduct in question cannot be said to have occurred primarily and substantially in Massachusetts." Uncle Henry's. Inc. v. Plaut Consulting Co., 399 F.3d 33, 45 (1st Cir. 2005) (citing cases). In other words, the appropriate inquiry looks at "whether the center of gravity of the circumstances that give rise to the claim is primarily and substantially within the Commonwealth." Kuwaiti, 438 Mass. at 473. Where alleged wrongdoing is not focused on Massachusetts, but has relevant and substantial impact across the country, the "primarily" requirement is not satisfied. Uncle Henry's, 399 F.3d at 45.
On this record, a reasonable jury could not conclude that Defendants' unfair and deceptive acts occurred "primarily and substantially within the commonwealth." Mass. Gen. L. c. 93A, § 11. To begin, the Court returns to the allegations in the amended complaint, D. 109, as to what NMI alleges Defendants' unfair and deceptive acts are for this claim. NMI alleges that this conduct included Defendants stealing its trade secrets "and publishing them publicly and for free"; misappropriating other proprietary and confidential information; Meta inducing Zlateski to violate his non-compete agreement with NMI and, by doing so, obtaining NMI's trade secrets at virtually no cost and then publish them to "the world" in an attempt to attract developers to its site. Id. ¶ 60. Even as alleged, this wrongful is not directed at Massachusetts. That NMI is a Massachusetts company and Zlateski, its former employee, signed his employment contract in Massachusetts does not satisfy the "primarily and substantially" in Massachusetts requirement, Spring Inv. Servs., Inc. v. Carrington Cap. Mgmt., LLC. No. 10-10166-FDS, 2013 WL 1703890, at *13 (D. Mass. Apr. 18, 2013), nor does the fact that NMI suffered some or all of the injury from Defendants' alleged violations of Chapter 93A here, see, e.g., EMC Corp. v. Pure Storage, Inc., No. 13-12789-JGD, 2016 WL 7826662, at *4 (D. Mass. Aug. 19, 2016); Spring Inv. Servs., Inc., 2013 WL 1703890, at *13. The test is the "center of gravity" of the wrongful conduct by Defendants, not a "place of injury" test, because if it were the later, "practically no case involving a Massachusetts plaintiff would be exempt from [C]hapter 93A status, no matter how negligible defendants' business activity in this [s]tate." Spring Inv. Servs., Inc., 2013 WL 1703890, at *13 (second alteration in original) (quoting Makino, U.S.A., Inc. v. Metlife Cap. Credit Corp., 25 Mass.App.Ct. 302, 309-10 (1988)) (internal quotation marks omitted).
NMI does point beyond its mere presence in the Commonwealth to additional conduct by Zlateski in Massachusetts after joining Meta: (1) that he was frequently in Massachusetts; (2) that he accessed NMI files on July 7, 2019, while in Massachusetts; (3) that a forensic inspection of his laptop shows that he accessed Meta's FBGEMM files on July 13, 2019, while in Massachusetts and only four days before he shared the code file he drafted with Meta engineers that NMI alleges contains its trade secrets. D. 332 at 36.
It is undisputed that Zlateski accessed a file on the NMI slack on July 7, 2019 while in Massachusetts. D. 352 at 549-51. Zlateski's explanation that the file that he accessed then, one different than the code file at issue here, is not related to NMFs alleged trade secrets, however, also remains undisputed. Id.; D. 333-79 at 2. It is also unclear how access by Zlateski, while a Meta employee, to Meta's own FBGEMM on July 13, 2019, before, even as NMI contends, he shared the alleged stolen NMI trade secrets on July 17, 2019 constitutes an unfair and deceptive act. Even assuming that this access to Meta's FBGEMM is evidence that Zlateski at least was working on the challenged code and incorporating NMFs trade secrets that he later gave Meta on July 17, 2019, such singular act in Massachusetts does not show that Defendants' commission of unfair and deceptive acts occurred primarily and substantially in Massachusetts. The same is true, even assuming that, as NMI contends, Zlateski was frequently in Massachusetts after he joined Meta (the frequency of which is disputed by Meta, D. 351 at 24), since mere presence here is not sufficient. Significantly, Defendants' alleged deceptive and unfair conduct included not just the misappropriation of trade secrets by Zlateski from NMI, but Meta's alleged inducement of same and the posting of the challenged code to GitHub for the world to see and use at no cost, which were not aimed at Massachusetts. As to this conduct, it is undisputed that Zlateski was hired into Meta's New York office; there is no evidence that Meta ever asked him to engage in any activities in, or otherwise direct any actions toward, Massachusetts; only a handful of Meta employees were working on FBGEMM when Zlateski worked on code for FBGEMM in 2019, and none of them were in Massachusetts; and the posting of the accused code on GitHub was not directed at Massachusetts. D. 318 at 36-37. "Even reading the record in the light most favorable to [NMI]," and even assuming arguendo that Defendants (or at least Zlateski) targeted NMI with the intent to cause the company substantial injury in Massachusetts, "it did so without engaging in significant wrongful conduct in Massachusetts." EMC Corp., 2016 WL 7826662, at *2.
Accordingly, the Court allows summary judgment to Defendants on Count III, the c. 93 A claim, on this basis.
D. Breach of Contract (Count IV)
Zlateski seeks summary judgment of Count IV, which alleges against him individually violations of the non-disclosure and non-compete provisions of the NMI Contract. D. 318 at 37-38; D. 351 at 24-25.
Regarding the non-disclosure provision, Zlateski only argues NMFs contract claim must fail for the same reasons NMFs trade secret claims fail. D. 318 at 37-38; D. 351 at 24-25. As the Court has already determined that NMFs trade secret claims survive summary judgment, so too does its claim for breach of the non-disclosure provision of the NMI Contract at least as to the alleged disclosure of its trade secrets.
As to the non-compete provision of the NMI Contract, Zlateski argues that the provision is unenforceable and that he did not breach it. D. 318 at 40, 40 nn.23-24; D. 351 at 25, 25 nn.15-16. The Court need not reach the issue of enforceability, since it concludes, on this record, that no reasonable jury could find a breach of the non-compete provision of the NMI Contract. Without repeating the Court's analysis resolving the motion for preliminary injunction, NMI has failed to point to a disputed issue of material fact as to this claim. To prevail on this contract claim, NMI would have to show that Zlateski "directly or indirectly . . . engage [d] in any business activity which is in competition with the service or products being rendered, delivered, marketed, commercialized, produced or sold or under development or active consideration by [NMI]." D. 3 3 3 -4 at 10. Although this Court is certainly not bound by its decision at the preliminary injunction stage, now that there is a fully developed record after discovery, the Court still concludes that there is a dearth of evidence proffered by NMI regarding this claim. It remains the case that NMI has not provided any goods, products, or services that are in competition with Meta. Even if it licenses its products in the future, NMI claims (as its theory of damages reflects) that Meta and other entities in the AI/ML market would be its license customers, not competitors. See D. 79 at 23. Even though the Court rendered its prior decision regarding same back in May 2020 on the reasonable likelihood of success standard, NMI has not shown a disputed issue of material fact from which a reasonable jury could find in its favor.
Even as it points to documents suggest that Meta integrated or began integrating Zlateski's code into several of its projects, [Redacted] D. 319-2 ¶ 147, this does not show, that in violation of the non-compete provision, that Zlateski "directly or indirectly ... engage[d] in any business activity which is in competition with the service or products being rendered, delivered, marketed, commercialized, produced or sold or under development or active consideration," D. 333-4 at 10, for the reasons that the Court has previously articulated. D. 79 at 23-24 (explaining that "[u]nder this definition, the Court does not conclude that [Meta] is a competitor[] of [NMI]. In its present startup stage, [NMI] has not provided any goods, products or services that it has that are in competition with [Meta]. Moreover, even when it launches its product, its intended customers are researchers, universities, direct to consumer services and even [Meta] itself. . . . Moreover, [Meta] does not offer a GEMM JIT product for sale that competes with the customer base that [NMI] intends to develop" (citations omitted)). Accordingly, the Court DENIES summary judgment to Defendants as to the breach of the non-disclosure provision of the NMI Contract but ALLOWS it as to the breach of the noncompetition provision of the NMI Contract.
E. Tortious Interference with Advantageous Contractual Relations (Count V)
Meta argues that summary judgment is appropriate as to Count V, which alleges tortious interference with advantageous contractual relations. Precisely, Meta argues that no reasonable jury could find it liable for tortiously interfering with Zlateski's non-disclosure and non-compete obligations under the NMI Contract to NMI because (1) Meta was not aware of Zlateski's contractual obligations; and (2) NMI cannot show that Meta interfered with those contractual obligations by improper motive, means, or other conduct. D. 318 at 34-36; D. 351 at 22-23.
A claim for tortious interference requires a plaintiff to show: (1) a contract with a third party; (2) that defendant knowingly induced a breach; (3) that defendant's interference was intentional and improper in motive or means; and (4) resulting harm. ADH Collision of Bos., Inc. v. Wvnn Resorts. Ltd., No. 19-CV-10246, 2020 WL 3643509, at *2 (D. Mass. July 6, 2020) (citing cases).
As discussed above, the Court has already determined that NMI provided sufficient (if disputed) evidence that, in its totality, could lead a reasonable jury to conclude that Meta had "reason to know" that Zlateski's conduct was improper. Meta contends that it had no actual knowledge of Zlateski's contract until January 2020, but lack of knowledge must be reasonable under the circumstances. See, e.g., Laudano v. 214 S. St. Corp., 608 F.Supp.2d 185, 196 (D. Mass. 2009) (stating that "[u]nder Massachusetts law, one cannot tortiously interfere with a contract that one reasonably believes is not in existence" (quoting Walker v. Waltham Hous. Auth., 44 F.3d 1042, 1048 (1st Cir. 1995)) (internal quotation marks omitted)); Ryan. Elliott & Co. v. Leggat, McCall & Werner, Inc., 8 Mass.App.Ct. 686, 691 (1979) (noting that reliance upon representations of non-contract must be reasonable).
Furthermore, both parties contest whether industry practices are sufficient to establish knowledge. See, e.g., Amigo Broad., LP v. Spanish Broad. Sys., Inc., 521 F.3d 472,491 (5th Cir. 2008) (noting defendant had actual knowledge of an employment agreement because "it was common in the radio industry for on-air talent to have employment contracts that bound the talent to a term of years"); Wellington Shields & Co. v. Breakwater, No. 14-cv-7529 (RJS), 2016 WL 5414979, *4 (S.D.N.Y. Mar. 18, 2016) (expressing that "industry practice is not sufficient to show actual knowledge," and "[p]laintiff has not even demonstrated that exclusivity is the industry practice"); Top Value Enters., Inc., v. Carlson Mktg. Grp., Inc., 703 S.W.2d 806, 810 (Tex. App. 1986) (holding that defendant knew a contract existed because "it was common knowledge in the industry that all grocers operated under written license agreements"). To the extent industry practices is a proper consideration, it falls in NMI's favor, as there is evidence in the record to suggest such contractual obligations are standard in Meta's industry. For example, Meta concedes its employees are required to undergo a multiple-hour onboarding video training regarding contractual obligations, such as non-competes. D. 352 at 454-59. Such a training would seem to be superfluous if many Meta employees were not subject to such contractual obligations.
Finally, Meta's argument regarding NMI's alleged inability to show improper motive, means, or other conduct does not withstand scrutiny. "[Misappropriation of confidential information can constitute improper means." Biopoint, Inc. v. Dickhaut, No. 20-10118-RGS, 2021 WL 4311651, at *4 (D. Mass. Sept. 22, 2021) (citing cases). This means that if the jury determines Meta misappropriated NMI's trade secrets, the jury could also determine Meta tortiously interfered with Zlateski's contractual obligations by improper means.
Accordingly, the Court concludes that NMI has presented sufficient evidence as to the elements of the tortious interference claim to constitute a genuine dispute of material fact.
F. Diminution of Enterprise Value and Reasonable Royalty Damages
Finally, Defendants seek summary judgment on NMI's diminution of enterprise value and reasonably royalty damages theories, propounded by Ellison and Akemann, respectively. D. 318 at 38; D. 351 at 25.
1. Defendants' Motion to Strike Ellison's Opinion
Under Rule 702, an expert may proffer an opinion that is based upon sufficient facts or sufficient factors or data; is the product of reliable principles and methods; and the experts has applied the principles and methods reliably to the facts of the case. Fed.R.Evid. 702; see United States v. Jordan, 813 F.3d 442, 445 (1st Cir. 2016) (citing Fed.R.Evid. 702). The crux of Defendants' challenge to Ellison's opinion as to the diminution of NMI's value in light of alleged trade secret misappropriation is as to the first Rule 702 requirement. D. 321 at 12-23; D. 361 at 6-14.
NMI retained Dr. Ellison, an economist, to provide an expert opinion regarding its valuation and how the alleged misappropriation affected same. Due to the alleged misappropriation, Ellison opines that NMI suffered a reduction in valuation of $880 million or more. D. 321-9 at 7, 22-23 (Ellison Rpt.). She arrives at this opinion by starting with the ML-related hardware market being $ 11 billion in 2020, and assuming that NMI would have captured ten percent of that market in 2020. Id. at 20. Although conceding that "these figures are speculative," Ellison rests her future analysis on this basis. Id. She further opines that the misappropriation would have compromised this revenue stream, at least because entities that would otherwise have become customers of NMI forked their alleged trade secrets when Meta posted them on GitHub. Id. at 21. Since these customers would now have access to NMI's code for free, they would not be available as NMI's customers. Id. Among these potential customers were allegedly Amazon, Microsoft, and Meta, which collectively make up twenty percent of AI/ML the market, so Ellison opines that these companies "would have constituted 20% of [NMI's] potential customer base and revenue stream absent the misappropriation and disclosure." Id. From the $220 million revenue figure, a reduction of that by 20 percent per year would be $44 million. Id. Ellison then proceeds to explain that "[a] firm's value is based on its projected future profits" and that a way to do so is to "compute the discounted present value of the expected stream of profits in perpetuity," (a notion and calculation that Defendants do not contest even as they challenge the discount rate that she used to do so). Id. at 22. Making that "calculation of a foregone revenue/profit stream of $44 million a year due to misappropriation and the formula that translates a foregone profit stream into total damages," Ellison concludes that there was a "$880 million reduction in [NMI's] value." Id. at 22-23. She further opines that this figure is not just supported by her calculation, but by "the available record testimony." Id. at 23, 23 n.25. The deposition testimony of Brian Stevens ("Stevens") of NMI, that she cites, however, does not support this particular calculation, but rather a general notion that the "overhang," by which he refers to the allegations against Defendants regarding misappropriation, caused "compression" on the value of NMI. Id. at 23 n.25 (citing D. 333-88 at 6).
Even if Ellison later stated in her deposition that she "perhaps used the word 'speculative' too loosely" in her report, D. 321 -10 at 6, it does not explain the lack of factual basis for her opinion as discussed herein.
Although it is correct that experts are permitted to rely on assumptions that are "not obviously false, and appear[] to have at least some evidentiary support," Maga v. Hennessy Indus., Inc., No. 12-11423, 2014 WL 10051399, at *7(D. Mass. Dec. 1, 2014); see, e.g., United States ex rel. Banigan v. Organon USA Inc., No. 07-12153-RWZ, 2016 WL 6571269, at *3 (D. Mass. Jan. 20, 2016) (denying motion to strike expert testimony based on allegedly imperfect data because "the propriety of [the expert's] assumptions [should be addressed] through cross-examination and competing expert testimony"); Waters v. Town of Aver, No. 04-10521-GAO, 2009 WL 10728967, at *6 (D. Mass. Apr. 6, 2009) (explaining that "[a]ny failure by [the expert] to consider facts that the defendants deem to be important may impact the weight of his testimony, but does not warrant a conclusion that the testimony lacks a reliable foundation and is inadmissible" (citing Currier v. United Techs. Corp., 393 F.3d 246, 252 (1st Cir. 2004)), it is also correct, as noted previously, that "a court may exclude an expert's opinion when it is based upon conjecture or speculation deriving from an insufficient evidentiary source." Tex. Roadhouse. Inc., 215 F.Supp.3d at 158 . That is, "[d]amages calculations can be based on assumptions and projections, but those assumptions and projections must be properly grounded in the evidence." SiOnyz, LLC v. Hamamstsu Photonics K.K., No. 1513488-FDS, 2019 WL 13180450, at *6 (D. Mass. Apr. 18, 2019). Defendants contend that this is the deficiency in Ellison's opinion about diminution in NMFs valuation and the Court agrees.
It is undisputed that Ellison relies largely upon New Enterprise Associates, LLC's ("NEA") valuation and projections for her calculation of the diminution in value. NEA, a venture capital firm, was the lead investor in NMI. Although Greg Papadopoulos ("Papadopoulos") of NEA, the venture partner that worked most closely with NMI, thought that there was a "distinct possibility" that trade secret appropriation affected NMFs valuation, D. 318-51 at 8, NEA did not and does not offer any valuation of that effect. Moreover, in three rounds of funding for NMI, NEA assessed NMFs value as $8.35 million (in February 2018, before NMI raised $5 million in funding); as $30 million (in October 2019, which was contemporaneous of the time of the alleged misappropriation, before NMI raised an additional $15 million); and at $125 million (in 2021-22, after NMI initiated this lawsuit and the allegations of Defendants' misappropriation was publicly known, and before NMI raised an additional $33 million). D. 321 at 6-7, 7 n.2, 11; D. 321-2 at 3-6; D. 321-15 at 28. If NEA's projections are the factual basis of Ellison's calculations, as she has stated, D. 321-9 at 20; D. 321-10 at 6, it is hard to fathom how there is a reliable basis for a $880 million diminution in NMI's value but for Defendants' misappropriation, when NMI's total valuation by the same source before the misappropriation occurred never exceeded $30 million dollars that NEA projected in the 2019 seed round, or when it projected estimated revenues of $2.1 million for 2020. D. 321 at 11 (citing D. 321-15 at 28).
In response, NMI clarifies that Ellison arrived at the $220 million revenue figure from an NEA document. D. 347 at 6 (citing and including chart from D. 347-4 at 4). This document, entitled "Size of the Market," was prepared in August 2021, begins with an estimated total of the ML hardware market, assumes for "illustrative" purposes that NMI would have had 10% of that market in 2020, and estimates that NMI's revenue, in that scenario, would have been $220 million. Id. Given its preparation in August 2021 (as part of its assessment of the third round of investments), it is not a projection of 2020 revenue. Moreover, even as NEA noted then, the figure was intended to illustrate NMI's potential to generate revenue, D. 321-14 at 4 (internal NEA August 16, 2021 memorandum including "[illustrative [r]evenue" chart and noting that "[p]er the illustrative calculation in the figure [in the chart], if [NMI] captures even a fraction of this traffic ... it has the potential to generate hundreds of millions of dollars in revenue at maturity"). In the same discussion, NEA projected that, a "homerun" scenario for NMI would be a five percent chance of generating $150 million in revenue, but not by 2020, but by 2026. Id. at 12. Papadopoulos further testified that this revenue projection was to get an understanding of the value that there may be in NMI's technology in the marketplace. D. 321-16 at 7, 9. Even before or after such calculations, in the first round of funding (2019) and in preparation for the third round (2021), NEA never placed the valuation of NMI nearly as high as Ellison's starting point or her ultimate conclusion. D. 321 at 11; D. 321-17.
The record also makes clear that Ellison did not make any attempt to verify the reliability of the $220 million figure, beyond confirming the reputation of NEA (which is not disputed by Defendants) and reviewing the deposition of Papadopoulos of NEA. D. 321-10 at 12. She did not speak to anyone at NMI about the calculation of, or reasonableness of the $220 million figure and did not seek contemporaneous figures. Id. "[R]eliance on this projection [supplied by NEA], without independent verification, renders [the] analysis unreliable." Legendary Art., LLC v. Godard, No. 11-0674, 2012 WL 3550040, at *4 (E.D. Pa. Aug. 17, 2012); Rand A Tech. Corp. v. Parametric Tech. Corp., No. 03-CV-11046-MEL, 2005 WL 6768210, at *1 (D. Mass. Oct. 19, 2005) (noting that an "expert must do more than simply rely on a client's representations about the validity of its forecasts" (citing SMS Sys. Maint. Servs., Inc. v. Digital Equip, Corp., 188 F.3d 11, 25 (1st Cir. 1999)). Ellison did not speak with NEA about the basis for the $220 million figure, D. 321-10 at 12, and there is nothing in the NEA documents or the deposition of Papadopoulos of NEA to suggest that $220 million figure would be a reliable "starting point" for Ellison's analysis as she framed it, D. 361-2 at 7. As to the impact of the alleged misappropriation on valuation, although Papadopoulos thought that it was a "distinct possibility" that it could have an effect on NMFs valuation, D. 318-51 at 8, the specter of same had no effect on NEA's valuation of the company when it invested in the third round of inviting in 2021, Id. at 7-9.
Defendants further contend that the flaw in Ellison's analysis is that building on the unstable building block of the $220 million figure, she compounds the unreliability of her opinion by further assuming that "[t]he alleged misappropriation caused NMI to lose three potential customers (Amazon, Microsoft, and [Meta])"; "[t]hese three customers would have been 20% of NMFs customer base and generated annual revenues of [20% of $220 million];" "NMI would incur no costs in providing services to these three customers;" "NMI would maintain these three customers 'in perpetuity' and never otherwise experience any revenue gains or losses;" and "[a] 'standard discount rate' of 5% was an appropriate discount rate to apply to NMFs 'expected stream of profits.'" D. 321 at 16 (quoting D. 321-9 at 20-22). Putting aside Defendants' last contention about the discount rate (which the Court need not reach), the Court agrees that the totality of Ellison's opinion about diminution of value is not based upon a reliable, factual basis, which it must be. For such damages calculations, any "assumptions and projections must be properly grounded in evidence." SiOnvx, LLC, 2019 WL 13180450, at *6. Here, they are not. The basis for asserting that NMI lost at least three customers, namely Amazon, Microsoft and Meta is as follows: each are in the AI/ML market; each "forked" the FBGEMM allegedly containing NMFs trade secrets; and, but for that, the potential customer base for NMI would be reduced by their collective market share (twenty percent), such that NMI lost twenty percent of its projected revenue of $220 million. D. 321-9 at 21-22. Putting aside that it appears contested on this record whether both Amazon and Microsoft did fork the FBGEMM, D. 321 at 18-20, even assuming arguendo that they did fork Meta's posting of the code there, there is no evidence that these entities used it, plan to use it, or had need to use it at the time of the forking or some time into the future. Even as Ellison further explained that this twenty percent loss of the market was a conservative figure (since it assumed that eighty percent of the market still would be available to NMI), D. 321-10 at 9, there is no reasonable basis for surmising that Amazon, Microsoft and Meta, representing twenty percent, were an appropriate proxy for lost customers, id There appears to be no factual basis that any of the entities would have become NMTs customers in the future, other than they represent, collectively, a significant segment of the market. This is particularly true where, save for a single, year-long license for $15,000, NMI had no customers before the alleged misappropriation. Furthermore, Ellison also has no basis to know if Amazon and Microsoft could make use of the technology. Id. at 8-9. Certainly, a plaintiff can calculate and proffer an expert opinion "forecasting its future business success," but "such projections must be based on reasonably reliable evidence." Alpha GRP, Inc. v. Subaru of Am., Inc., No. CV 18-2133-MWF (MRWx), 2021 WL 1146029, at *10-11 (CD. Cal. Feb. 8, 2021) (citation omitted) (excluding expert opinion under Rule 702 and noting that the same would be warranted under Rule 403).
As to the other challenged bases of her opinion, even where Ellison assumed a "negligible marginal cost associated with each license," D. 321-9 at 22, she still appears to assume that there would be no competition for such customers such that they would remain NMTs customers, with no gains or losses in revenue in perpetuity. As to this calculation, the "chain of assumptions [ becomes increasingly speculative at each succeeding step." SiOnyx, 2019 WL 1318040, at *6.
To NMFs argument that the better approach for challenges to the assumptions undergirding an expert report is to admit the report and permit vigorous cross-examination, see, e.g., Currier, 393 F.3d at 252 (rejecting Daubert challenge to expert's opinion where the expert's alleged failure to consider certain facts and circumstances was "clearly and repeatedly" tested through cross-examination and presentation of competing evidence); S. Union Co. v. Liberty Mut. Ins. Co., No. 06-CV-12067-RCL, 2008 WL 8564583, at *3 (D. Mass. Mar. 25, 2008) (rejecting a Daubert challenge because if the moving party wanted "to challenge whether the supplemental report [was] complete and balanced, it [could] do so during cross-examination"), the Court notes that "[w]hile there is no bright line separating the reasonable from the unreasonable, at some point the degree of speculation simply becomes too great." SiOnvx, 2019 WL 13180450, at *6.
For all these reasons, the Court allows Defendants' motion, D. 307, to exclude Ellison's opinion that NMI's diminution in value was $880 million or more.
2. Defendants' Motion to Exclude Akemann's Opinion
Defendants also have moved to exclude Akemann's opinion. D. 308. NMI proffers Akemann, an economist and Managing Director at Berkeley Research Group, LLC, for a reasonable royalty calculation. D. 348-7 at 4-5 (Akemann Rpt.).
Akemann opines about what royalty NMI would have negotiated with Meta for an unrestricted hypothetical license to the alleged trade secrets at the time of the alleged misappropriation. Id. at 13. His first step was to estimate the number of Meta's servers that would have utilized the alleged trade secrets. As Meta had approximated that "the number of general purpose CPU-based servers running inference-related machine learning workloads-which would constitute the only possible equipment to which NMI's claims even arguably could relate-would be approximately 58-78%," D. 348-5 at 15, Akemann ran three separate scenarios: one assuming the scope of server usage across all Meta's servers was 58 percent, the low end of Meta's percentages, another assuming 78 percent usage, the high end of Meta's percentage, and, lastly, a third scenario, in which he estimated server usage based on specific areas where Meta actually had tested the alleged trade secrets. D. 348-7 at 14-15. Within each scenario, Akemann assumed that NMI's technology would result in 2x speedup on each server and halve the number of servers that Meta would need to operate those programs. Id. at 31. Akemann then estimated a cost-per-server-per-month rate for Meta and multiplied the cost-per-server rate times his estimated number of server months saved, to get a total amount of gross cost savings to Meta if NMI's technology were [Redacted] Id. at 48. Akemann's final steps involved multiplying the total savings by a discount rate and assuming the parties would negotiate a 25%/75% bargaining split. Id. at 119. Ultimately, Akemann estimated a reasonably royalty of $110 million (based on server usage limited to the projects for which Meta had begun implementing the alleged trade secrets), $570 million (based on Meta's low-end 58% estimate of servers that could use the alleged trade secrets, as identified in Meta's interrogatory responses), and $766 million (based on Meta's high-end 78% estimate of servers that could use the trade secrets, as identified in Meta's interrogatory responses). Id.
Defendants charge Akemann's opinion as unreliable for a number of reasons. First, they argue that using a cost savings analysis here is unreliable because such modeling requires cost savings and Meta did not use the code and achieve cost savings here. D. 323 at 12. Defendants, however, acknowledge that a reasonable royalty calculation "based purely on hypothetical potential use" could in certain circumstances be an appropriate measure of damages. Id. The Court, thus, turns to Defendants' other challenges to Akemann's opinion.
Contrary to Defendants' contentions, Akemann's three server usage scenarios have a basis in fact. The 58 percent and 78 percent uses of Meta's servers were identified by Meta as the servers to which NMI's technology could arguably relate. D. 348-5 at 15. The third scenario focuses on the narrower universe of servers where Meta was researching and developing plans to use the trade secrets. D. 348 at 12. Unlike the case of Ellison's opinion that built upon unreasonable assumptions, that is not the case with these three scenarios which are based on discovery from Meta. Moreover, to the extent that Defendants contend that one or none of these scenarios is a better measure of damages in terms of breadth of use that could occur on its servers (including concerns about pruning and accuracy with implementation of NMI's trade secrets), that will be a matter for vigorous cross-examination. See, e.g., McMillan v. Mass, SPCA, 140 F.3d 288, 303 (1st Cir. 1998) (stating that "[i]f [the expert's] analysis omitted what defendants argue are important variables,... it was up to defendants to exploit and discredit the analysis during cross examination."); Pagliaroni v. Mastic Home Exteriors. Inc., No. 12-10164-DJC, 2015 WL 5568624, at *8 (D. Mass. Sept. 22, 2015) (stating that "[t]he First Circuit has repeatedly held that a challenge to the factual underpinning of an expert's opinion is a matter affecting the weight and credibility of the testimony-a question to be resolved by the jury" (citation and internal quotation marks omitted)).
The same is true as to the portion of Akemann's analysis, in which he assumed that Meta and NMI would "split" the gross cost savings as opposed to the net cost savings. D. 323 at 17. In other words, according to Defendants, Akemann assumed that Meta would bear all the costs associated with reworking its servers to fit NMFs technology and that Meta would bear all the costs associated with any resulting decreases in accuracy. Id. Defendants further claim that the record demonstrates that their implementation costs would not be zero. Id. at 19. Akemann noted, however, that he did consider implementation costs (and that Meta's implementation costs would not be zero), but simply assumed that subtracting implementation costs would lead to a double count:
It was expected that the counterparty would incur those costs as part of implementation and that [NMI] would incur whatever costs it needed to do to support it.... each party would incur whatever additional costs it needed to incur. ... they're implicitly accounted for because to add them in would be a double count, in my view.D. 348-29 at 4. Any disagreement about the extent to which Akemann considered implementation costs (or considered them incorrectly) is a matter to draw out in cross-examination, is for the jury to decide what weight, if any, to give Akemann's opinion, and is not a basis to exclude his opinion.
The Court also concludes that Defendants' apportionment argument does not warrant exclusion of Akemann's opinion. Defendants argue that Akemann's opinion is unreliable because he failed to apportion the value attributable to non-trade secret components of NMI's technology or each trade secret separately. D. 323 at 21; D. 361 at 10-13. In his report, Akemann apportioned the value attributable with the trade secrets to a [Redacted] speedup, which was based on his "interviews with Dr. Scott and Dr. Shavit and on Dr. Park's assessment that the code Dr. Zlateski wrote incorporating NMI's trade secrets achieved: 'we are seeing promising results [Redacted] D. 348-7 at 31. Akemann also states in his report that he explicitly accounted only for the value related to the trade secrets. Id. at 26 (stating that his "damages analysis below is careful to isolate (or apportion) the economic value related to the trade secrets at issue in this distinct from economic value driven by other factors"). Although Defendants contest this evidence (or the interpretation of this evidence), there is a factual, if disputed, basis for this apportionment. Akemann additionally considered and determined the value of the categories of trade secrets. Id. at 33-34 (stating that he "understand[s] from conversations with Drs. Shavit and Scott that - particularly given the conservative nature of the speedups assumed in my analysis [Redacted] - the technical benefit, and hence the value of any reasonably royalty (and the related harm caused to [NMI] by Defendants' disclosure), can reasonably be assumed to be the same for each trade secret for sparse instruction streaming . . . and for each enumerated combination trade secret[s]").
Defendants also argue that Akemann's understanding of the potential value of the alleged trade-both collectively and separately-secrets was based upon opinions of Scott's that were disclosed only in a private interview, which is improper. D. 323 at 23 (citing Sound View Innovations. LLC v. Hulu. LLC, No. LA CV17-04146 JAK (PLAx), 2019 WL 9047211, at *14 (CD. Cal. Nov. 18, 2019)). As explained above, Akemann's understanding of the value of the trade secrets collectively as [Redacted] D. 348-7 at 31. As for the value of each trade secret separately, it is true that in Akemann's report the only support he provides for the assertion that the value of any reasonable royally "can reasonably be assumed to be the same for each trade secret for sparse instruction streaming . . . and for each enumerated combination trade secret[s]" is his "conversations with Drs. Shavit and Scott." D. 348-7 at 30, 33-34. Akemann since has described these conversations in detail during his deposition. D. 323-17 at 21-23. This makes Defendants' citation to Sound View Innovations. LLC inapposite, because the court there excluded an expert's opinion that relied on conversations with another expert, which were not otherwise disclosed. 2019 WL 9047211, at *14.
Even though Akemann explicitly stated that he did not "have a separate analysis that assumes that. . . FMA input buffering is the only trade secret in the case that remains" or as to compact FMA encoding, D. 323-17 at 21-22, the cases Defendants' cite do not require exclusion of Akemann's opinion, see, e.g., LivePerson, Inc. v. [24]7.AI, Inc., No. 17-CV-01268-JST, 2018 WL 6257460, at *2 (N.D. Cal. Nov. 30, 2018) (excluding expert opinion for failure to apportion damages among particular trade secrets as it "offers no methodology for the jury to calculate trade secret misappropriation damages on fewer than all of the 28 alleged trade secrets"); 02 Micro Int'l Lt. v. Monolithic Power Svs., Inc., 399 F.Supp.2d 1064, 1076-77 (N.D. Cal. 2005) (same). In both of those cases, the experts only presented opinions regarding the value of the trade secrets collectively, not separately, or as any subset less than the total number of trade secrets. LivePerson. Inc., 2018 WL 6257460, at *2; 02 Micro Int'l Lt., 399 F.Supp.2d at 1076-77. Akemann, on the other hand, made clear that his analysis is still applicable to less than all the trade secrets. D. 323-17 at 21 (explaining that "the maximum benefit as I think is clear from my report is when the various trade secrets or ideas are used in combination, but [Scott] was confident that you would get at least a [Redacted] benefit for the indicated trade secrets in my paragraph 91 or what has now become the first 10 of the remaining 12").
Finally, Defendants argue that Akemann's opinion is not reliable, because 25%/75% bargaining split between NMI and Meta is not tied to the facts of the case and "is nothing more than an improper rule of thumb that countless courts have rejected in similar contexts." D. 323 at 23 (citing Uniloc USA. Inc. v. Microsoft Corp., 632 F.3d 1292, 1315 (Fed. Cir. 2011)). In this case, however, the 25%/75% split is sufficiently tied to the facts, as it was "[b]ased on the testimony of Mr. Stevens and Mr. Papadopoulos and the related documents" and that this split is "conservative relative to standard economic models of bargaining." D. 348-7 at 71. Akemann's assumptions undergirding this split are better tested through cross-examination. See Maga, 2014 WL 10051399, at *7 (refusing to exclude expert opinion where underlying "assumption is not obviously false, and appears to have at least some evidentiary support" and "[t]he better course, therefore, would appear to be to admit the opinion . . . and to permit vigorous cross-examination on the topic").
For these reasons, the Court denies Defendants' motion to exclude Akemann's opinion. D. 308.
3. NMI's Motion to Strike Bakewell 's Opinion
NMI moves to exclude the opinion of Bakewell, an expert offered by Defendants to rebut Ellison's valuation calculation and Akemann's reasonable royalty calculation. D. 341. In his report, Bakewell ultimately concludes that a more reasonable measure of damages falls between a range of $15,000 and $2.34 million. D. 370-2 ¶ 20 (Bakewell Rpt).
First, NMI asserts that Bakewell should be prohibited from offering any opinion regarding damages because, at his deposition, he allegedly disavowed his opinion and testified that he did not plan to offer any of the aforementioned figures as damages calculations, thus rendering his expert report unreliable. D. 349 at 10-11; D. 374 at 7-9. NMI places particular emphasis on the fact that Bakewell characterizes these figures as "reasonable measure[s] of damages" in his report, but refers to these figures as "data points," "categories] of information," and "attributes" in his deposition. D. 349 at 10-11; D. 349-3 at 3,5; D. 370-2 ¶ 20; D. 374 at 8. From NMI's perspective, this amounts to Bakewell's affirmative disavowal of the opinion in his report. D. 349 at 11; D. 374 at 8. Courts certainly have excluded an expert's opinion for contradicting themselves or disavowing their prior opinions have done so where the expert's "deposition testimony and his expert report [were] fundamentally inconsistent," Chartier v. Brabender Technologies. Inc., No. 08-40237-FDS, 2011 WL 4732940, at *7 (D. Mass. Oct. 5, 2011); where the "expert affirmatively repudiates or disavows an opinion expressed in the expert's report," rather than "makes statements during a deposition that are difficult to reconcile with the opinions disclosed in the expert's report," Mitchell v. Geo Grp. Inc., No. CV-19-04445-PHX-DWL, 2022 WL 874287, at *5 (D. Ariz. Mar. 24, 2022) (citing cases); and where there is an "unequivocal" disavowal of prior opinions, Devito v. Smithkline Beecham Corp., No. CIV.A. 02-CV-0745NPM, 2004 WL 3691343, at *4 (N.D.N.Y. Nov. 29, 2004).
Here, however, Bakewell's deposition testimony was not "fundamentally inconsistent" with his report, nor did it "affirmatively" or "unequivocal[ly]" disavow opinions in his report. To be sure, he simply explained:
[w]ell, any and all of [the numbers in the proposed range of $15,000 to $2.34 million] could [be the amount of damages that the trier of fact should award]. They are all, I think, measure [sic] that have their own qualities. And I describe them as I did in ~ collectively and individually in paragraph 20 and then elsewhere in my report. And I think that's a good way to describe and to summarize my opinions, is that there is a - several measures that are around the same spot. And ultimately, it's up to the fact-finder to make that determination. I can arrange these in terms of lowest to highest, and they have their own qualities. That's what I wrote in my report.D. 349-3 at 4. In other words, Bakewell did not write in his report, nor state at his deposition, that that any one of these specific numbers was the appropriate measure of damages, but that, based upon his understanding of the facts at issue in this case, this range of damages are more appropriate than the numbers proposed by Ellison and Akemann. Although ripe for cross-examination as to the weight, if any, a jury should put on such an opinion, it is not grounds to exclude Bakewell's opinion.
Second, NMI argues that Bakewell's opinion, particularly regarding the upper limit of damages as $2.34 million, is unreliable because he performed no analysis in reaching this number, only cited to a number provided by Akemann, and did not tie this number to the facts of this case. D. 349 at 11-13; D. 374 at 7. The $2.34 million figure is derived directly from NMI's interrogatory response, D. 370-6 at 7-24, and Bakewell cited it when including the figure in his own report, D. 370-2 ¶ 20 n.34. Indeed, Akemann's own report states explicitly that "[NMI] estimated that 'Total Cost Associated with NMFs Development of All Misappropriated Trade Secrets and/or Confidential Information' is approximately $2.34 million" and cites to the same interrogatory response as Bakewell. D. 348-7 ¶¶ 20, 20 n.24. While NMI contends more analysis was necessary because "using the actual development costs of the plaintiff as the complete measure of damages ... is frequently inadequate," D. 349 at 12; see also D. 374 at 7, and because Bakewell performed allegedly deficient analyses in reaching the other measures of damages, D. 374 at 5-6, this goes to the weight of Bakewell's testimony, not admissibility.
Third, NMI seeks to preclude Bakewell from offering a legal opinion that Akemann was legally required to apportion damages further than he did as to the trade secrets. D. 349 at 13-15; D. 374 at 9-11. Bakewell did, indeed, cite to legal authorities when discussing his view of Akemann's report regarding the value of each individual trade secret. D. 370-2 ¶¶ 87, 87 n.183, 99, 99 n.198, 114, 114 n.220. Generally, "a witness may refer to the law in expressing an opinion without that reference rendering the testimony inadmissible." Hangarter v. Provident Life & Accident Ins. Co., 373 F.3d 998, 1017 (9th Cir. 2004) (citation and internal quotation marks omitted). To the extent that Bakewell was planning to offer an opinion regarding what the law requires as to apportionment (which will be the subject of the Court's jury instructions), he may not do so and that portion of his proffered report is excluded. The Court, however, will not exclude Bakewell's analysis of alleged lack of apportionment in Akemann's opinion:
Dr. Akemann identifies several categories of alleged benefits associated with the alleged trade secrets, and states that there are numerous alleged trade secrets in this case, but Dr. Akemann has no analysis specific to any of these benefits or alleged trade secrets. Rather than simply lump the alleged trade secrets into one and attribute all purported benefits to this amalgamation, it is Dr. Akemann's responsibility to identify and measure the specific, incremental contribution of each alleged trade secret. But this type of specificity is absent from Dr. Akemann's work.... Relatedly, none of the inputs in Dr. Akemann's calculations are focused on the specific, incremental contribution of the asserted trade secrets. If none of the inputs are specific to the asserted trade secrets, then Dr. Akemann's conclusions cannot be, either.D. 370-2 ¶¶ 87, 87 n. 184; see also id ¶¶ 66, 72, 103-40. Whether his analysis sufficiently demonstrates that Akemann's analysis is inadequate is for NMI to argue to the jury, who would be instructed on the law by the Court.
Lastly, NMI argues that Bakewell should be prohibited from stating that NMI and Meta are not competitors, because it is allegedly unsupported. D. 349 at 16; D. 374 at 11-13. However, Bakewell clarified at his deposition he assumed they were not competitors for his economic analysis. D. 349-3 at 8. He also highlighted that his report does indeed include a discussion of the commercial relationship between NMI and Meta and the various factors at play in reaching that conclusion, accompanied by specific citations to the record. Id.; D. 370-2 ¶¶ 210-20. While NMI contests whether this assumption supports Bakewell's conclusions and whether NMI and Meta actually compete, "the propriety of [BakewelPs] assumptions [and conclusions should be addressed] through cross-examination and competing expert testimony." United States ex rel. Banigan, 2016 WL 6571269, at *3.
For all these reasons, NMI's motion to strike Bakewell's opinion is denied, except as to any legal opinion as discussed above. D. 341.
4. Defendants' Motion for Summary Judgment as to NMI's Diminution of Enterprise and Reasonable Royalty Damages Theories
Having considered NMI's theories of damages and the parties' competing experts regarding same and having excluded Ellison's opinion that NMI's diminution in value was $880 million or more, the Court denies Defendants' motion for summary judgment as to these two theories to the following extent. The Court denies summary judgment as to the theory of diminution of value and reasonable royalties, but NMI may not rely upon the excluded opinion of Ellison at trial.
VI. Conclusion
For the foregoing reasons, the Court ALLOWS Defendants' motion for summary judgment in part and DENIES it in part, D. 303, DENIES Defendants' motion to strike Scott's June 22, 2022 declaration on Fed.R.Civ.P. 26(e) grounds, D. 297, DENIES Defendants' Daubert motion to strike Scott's opinion relating to NMI's code comparison testing, D. 304, ALLOWS Defendants' motion to strike Rowe's opinion in part and DENIES it in part, D. 305, ALLOWS Defendants' motion to strike Ellison's opinion that NMI's diminution in value was $880 million or more, D. 307, and DENIES Defendants' motion to strike Akcmann's opinion, D. 308. Further, the Court DENIES NMI's motion to strike Kaeli's expert opinion, D. 287, DENIES NMI's motion to strike Bakewell's opinion in part and ALLOWS it in part, D. 341, and DENIES NMI's motion to strike Robinson's opinion, D. 342.
So Ordered.