ON TIME ON TIME ON TIME BUSINESS MANAGEMENT A+ WORK, ON TIME, NO PLAGARIZING; ON TIME

| About LexisNexis | Privacy Policy | Terms & Conditions | Copyright © 2020 LexisNexis

User Name: =

Date and Time: = 2026-09-13

Job Number: = 295587244

Documents (1)

Client/Matter: -None-

Search Terms: TITLE("SECRECY BY STIPULATION")

Search Type: bool

Content Type Narrowed by

analytical-materials -None-

1. ARTICLE: SECRECY BY STIPULATION, 74 Duke L.J. 99

Page 1 of 54

ARTICLE: SECRECY BY STIPULATION

ARTICLE: SECRECY BY STIPULATION

October, 2024

Reporter 74 Duke L.J. 99 *

Length: 36211 words

Author: NORA FREEMAN ENGSTROM,+ DAVID FREEMAN ENGSTROM,++ JONAH B. GELBACH,+++ AUSTIN PETERS++++ & AARON SCHAFFERNEITZ+++++

Copyright © 2024 Nora Freeman Engstrom, David Freeman Engstrom, Jonah B. Gelbach, Austin Peters & Aaron Schaffer-Neitz.

Highlight

ABSTRACT

GM Ignition Switch. Dalkon Shield. Oxycontin. For decades, protective orders court orders that require parties to maintain the confidentiality of information unearthed during discovery have hid deadly defects and pervasive abuse from the public, perpetuating unnecessary harm.

But how worrisome are these protective orders, really? Under Rule 26(c)'s plain language, protective orders are to be granted only upon a showing of "good cause." Doesn't that adequately cabin the orders' entry? Prominent judges and scholars have long insisted it does, and that, under Rule 26(c), the day-to-day grant of protective orders is careful, not cavalier. Critics disagree. They charge that parties frequently agree to sidestep Rule 26(c)'s "good cause" requirement and that judges, although formally duty bound to protect the public interest, uncritically acquiesce to parties' demands. Worried about judicial rubber-stamping, some, in fact, have spent decades pushing to tighten Rule 26(c)'s standards while others have, just as vigorously, opposed these efforts, insisting that the status quo works well enough.

+ Ernest W. McFarland Professor of Law at Stanford Law School ("SLS") and Co-Director of the Deborah L. Rhode Center on the Legal Profession ("Rhode Center").

++ LSVF Professor of Law at SLS and Co-Director of the Rhode Center.

+++ Herman F. Selvin Professor of Law at Berkeley Law and a Non-Resident Fellow at the Rhode Center.

++++ Harry A. Bigelow Teaching Fellow, Lecturer in Law at University of Chicago Law School, and Non-Resident Fellow at the Rhode-Center.

+++++ SLS Class of 2024 and a Civil Justice Fellow at the Rhode Center. We benefited from astute comments supplied by Stephen B. Burbank, William Eskridge Jr., C. Scott Hemphill, Deborah R. Hensler, Timothy D. Lytton, Larry C. Marshall, Robert L. Rabin, and Judith Resnik. We are grateful to Ari Berman, Devin Flynn, Sreya Guha, Vrushank Gunjur, and Jessica Seigel for excellent research assistance and to SLS, the Rhode Center, and the Arthur & Charlotte Zitrin Foundation for supporting this research.

Page 2 of 54

ARTICLE: SECRECY BY STIPULATION

This debate has raged since the late 1980s. But until now, it has mostly run aground on the shoals of basic, but unanswered, factual questions: Are stipulated protective orders really de rigueur? Are they becoming more prevalent? And are joint motions for protective orders actually meticulously scrutinized?

Using state-of-the-art machine learning techniques, this Article analyzes an original dataset of over 2.2 million federal cases to answer these persistent and profoundly important questions. Along the way, we find that stipulated protective orders are surprisingly prevalent. Grant rates for stipulated protective orders are sky high. And even though many insist that judges are scrupulous in the entry of such orders, over our entire study period, a majority of federal judges never rejected a joint protective order request.

We offer the first comprehensive accounting of stipulated protective orders in federal litigation. In so doing, we aim not only to revitalize and discipline the perennial and consequential debate surrounding Rule 26(c). We also offer a fortified empirical foundation on which to ground inquiry into broader questions, including the role of transparency and privacy in a system ostensibly committed to "open courts," tort law's vital information-forcing function, adversarialism as a procedural cornerstone of American litigation, and trial-court discretion and fidelity to higher law.

Text

[*101] INTRODUCTION

What if you found out that one judge's order could have saved the life of your loved one? Thousands of loved ones? Protective orders ("POs") court orders issued under the auspices of Federal Rule of Civil Procedure 26(c) ("Rule 26(c)"), which require parties to maintain the confidentiality of information unearthed during discovery have made these hypotheticals a grim reality.

Consider the 2014 General Motors ignition switch debacle.1 As unusual accidents piled up, injured parties predictably sued. One case involved Brooke Melton, a twenty-nine-year-old pediatric nurse who died when her Chevy Cobalt stalled on a Georgia highway.2 Brooke's parents, the plaintiffs in the ensuing wrongful death suit, uncovered smoking-gun evidence: Brooke's Cobalt had a defective ignition switch.3 Further digging revealed that some in GM had long known the switch was defective.4 Indeed, in new Cobalt models, the faulty ignition [*102] switch, responsible for Brooke's accident, had been quietly replaced. Brooke's parents wanted to share this explosive evidence. "We thought that people needed to know. There were still people out there driving those cars,"

1 In 2014, the ignition switch problem led GM to recall some 2.6 million vehicles, and the defect was eventually implicated in 275 injuries and 124 deaths. See Mike Spector, Jaimi Dowdell & Benjamin Lesser, How Secrecy in U.S. Courts Hobbles the Regulators Meant to Protect the Public, REUTERS (Jan. 16, 2020, 8:00 AM), https://www.reuters.com/investigates/special- report/usa-courts-secrecy-regulators [https://perma.cc/PV9S-MJDK] (reporting on the recall).

2 Id.

3 Id.

4 Id. For more on the GM ignition switch debacle and its aftermath, see generally Nora Freeman Engstrom, When Cars Crash: The Automobile's Tort Law Legacy, 53 WAKE FOREST L. REV. 293, 328-35 (2018) [hereinafter Engstrom, When Cars Crash]; ANTON R. VALUKAS, REPORT TO BOARD OF DIRECTORS OF GENERAL MOTORS COMPANY REGARDING IGNITION SWITCH RECALLS (2014); LANCE A. COOPER& MARK TABB, COBALT COVER-UP: THE INSIDE STORY OF A DEADLY CONSPIRACY AT THE LARGEST CAR MANUFACTURER IN THE WORLD (2020). For more on the Meltons' extraordinary discovery, see Bill Vlasic, An Engineer's Eureka Moment with a G.M. Flaw, N.Y. TIMES, Mar. 29, 2014, at A1.

Page 3 of 54

ARTICLE: SECRECY BY STIPULATION

Brooke's mother explained.5 But a 2011 stipulated PO, entered by a Georgia state court, barred such disclosures.6 While the Meltons were muzzled, more motorists died.7

Brooke's case is tragic. But it isn't anomalous. POs often used in conjunction with secret settlements, nondisclosure agreements ("NDAs"), attorney lockout provisions, and orders to seal judicial records ensure that many documents surfaced in litigation never see the light of day.8 And although it can be difficult to decouple the effects of POs, as against other secrecy mechanisms (because they often exist in tandem), evidence indicates that overly broad POs have led to the suppression of a wide range of information that, if revealed sooner, could have avoided significant harms, even death.9 The list is long: Oxycontin,10 Zyprexa (a drug used to treat schizophrenia and bipolar [*103] disorder),11 Zomax (a prescription painkiller),12 the Dalkon Shield intrauterine contraceptive device,13 Prempro (a drug used to relieve the symptoms of menopause),14 tampons,15 and cigarettes.16 POs have been used to hide

5 Spector et al., supra note 1 (quoting Beth Melton).

6 Id. The judge who issued the PO justified her decision on the logic that "[t]he role of litigation is not to regulate GM." Id. For the fact that the PO was stipulated, see Consent Protective Order of Confidentiality, Melton v. General Motors, No. 2011-A-2652 (Cobb Cnty. Ct. Dec. 13, 2011), available at https://www.autosafety.org/wp- content/uploads/import/Melton%20consent%20protective%20order%20Dec%20%202011.pdf [https://perma.cc/6Q5Q-D4SG].

7 See Spector et al., supra note 1. According to Kevin Vincent, previously Chief Counsel for NHTSA, the PO in the Melton litigation "stymied" NHTSA. If NHTSA had that evidence, Vincent said, "[w]e could have acted sooner." Id. Likewise, Lance Cooper, the Meltons' attorney, believes that but-for the PO, the defective design would have been more quickly revealed. Telephone Interview with Lance Cooper, Founding Partner, The Cooper Firm (Feb. 29, 2024).

8 For further discussion of orders to seal, which are akin to POs in some respects, see infra notes 44-51 and accompanying text. For a discussion of secret settlements, see generally David Freeman Engstrom, Nora Freeman Engstrom, Jonah B. Gelbach, Austin Peters & Garrett Wen, Shedding Light on Secret Settlements: An Empirical Study of California's STAND Act, 91 CHI. L. REV. (forthcoming 2024). For a discussion off attorney lockout or "buyout" provisions, which are formally banned by ABA Model Rule 5.6(b), see, e.g., Lynn A. Baker, Mass Torts and the Pursuit of Ethical Finality, 85 FORDHAM L. REV. 1943, 1957-65 (2017).

9 Although all the harms of litigation secrecy cannot be laid at the feet of POs, reformers see POs as the keystone of a broader secrecy architecture. As one scholar put it: "Almost all facets of the broader court-confidentiality problem stem from a common root protective orders." Dustin B. Benham, Proportionality, Pretrial Confidentiality, and Discovery Sharing, 71 WASH. & LEE L. REV. 2181, 2190 (2014).

10 See Elizabeth Chamblee Burch & Alexandra D. Lahav, Information for the Common Good in Mass Torts, 70 DEPAUL L. REV. 345, 345-47 (2021) (explaining that a 2015 deposition of the former chairman and president of Purdue Pharma, Dr. Richard Sackler, contained damning information concerning Purdue's activities but that the video of Dr. Sackler's deposition was kept under wraps for years, by virtue of a stipulated PO).

11 Sunshine in Litigation Act of 2008: Hearings on H.R. 5884 Before the H. Comm. on the Judiciary, Subcomm. on Com. & Admin. L., 110th Cong. 51 (2008) (statement of Sen. Herb Kohl); Alexander C. Egilman, Aaron S. Kesselheim, Harlan M. Krumholz, Joseph S. Ross, Jeanie Kim & Amy Kapczynski, Confidentiality Orders and Public Interest in Drug and Medical Device Litigation, 180 JAMA INTERNAL MED. 292, 295 (2019) (explaining that information, leaked by a medical expert in violation of a PO, clarified the drug's risks).

12 Benjamin Weiser & Elsa Walsh, Drug Firm's Strategy Avoid Trial, Ask Secrecy, WASH. POST (Oct. 24, 1988, 8:00 PM), https://www.washingtonpost.com/archive/politics/1988/10/25/drug-firms-strategy-avoid-trial-ask-secrecy/acdc1298-95da-42f3- bc38-70d47696184c [https://perma.cc/JNG4-K9T2].

13 Daniel J. Givelber & Anthony Robbins, Public Health Versus Court-Sponsored Secrecy, 69 LAW & CONTEMP. PROBS. 131, 133-34 (2006); S. REP. NO. 110-439, at 4 (2008).

14 Egilman et al., supra note 11, at 293, 295 (reporting that information covered by a PO showed that the manufacturer "downplayed risks of hormone-associated breast cancer" in the estrogen hormone therapy).

Page 4 of 54

ARTICLE: SECRECY BY STIPULATION

deadly defects in Remington rifles,17 playground equipment,18 ATVs,19 cars,20 trucks,21 and tires.22 They have concealed [*104] child sexual abuse.23 And they are often used to shield the disciplinary records of police, frustrating civil rights litigation aimed at reforming department policies.24

How are POs implicated in these scandals? The text of Federal Rule of Civil Procedure 26(c), which governs POs, seems to guard against these precise situations. Per Rule 26(c), POs, which bar dissemination of discovery materials beyond the litigants themselves, are to be issued only for "good cause."25 And many appellate courts are quite emphatic that the willy-nilly issuance of POs violates Rule 26(c)'s clear command. As the First Circuit explains: "[I]f good cause is not shown, the discovery materials in question should not receive judicial protection and therefore would be open to the public for inspection . . . . Any other conclusion effectively would negate the good cause requirement of Rule 26(c)."26 Clear enough.

15 See Tom Riley & Mary K. Hoefer, Protective Orders: Machiavelli Would Be Pleased, TRIAL, Nov. 1984, at 30 (describing litigation involving Rely tampons).

16 See infra note 127 and accompanying text.

17 See Editorial Board, Amid Court Secrecy, Guns Continued to Kill, USA TODAY (Dec. 28, 2014, 7:43 PM), https://www.usatoday.com/story/opinion/2014/12/28/sealed-settlements-remington-rifles-model-700-editorials-debates/20979945 [https://perma.cc/QV3N-BCZC] (discussing how orders to seal and POs concealed a defect in Remington's Model 700 rifle); Scott Cohn, Huge Trove of Remington Rifle Documents is Made Public, CNBC (Nov. 15, 2016), https://www.cnbc.com/2016/11/15/huge-trove-of-remington-rifle-documents-is-made-public-.html [https://perma.cc/JXV8-YXCJ] (same).

18 For a discussion of how POs thwarted inquiry into defective playground equipment, see S. REP. NO. 110-439, at 7 (2008).

19 For a discussion of how POs stymied regulators' investigation into defective ATVs, see Spector et al., supra note 1.

20 See Arlin R. Thrush, Comment, Public Health and Safety Hazards Versus Confidentiality: Expanding the Mediation Door of the Multi-Door Courthouse, 1994 J. DISP. RESOL. 235, 249 (describing how GM used POs to silence scores of plaintiffs injured by negligently designed fuel tanks in the 1970s and 1980s).

21 Benjamin Lesser, Dan Levine, Lisa Girion & Jaimi Dowdell, How Judges Added to the Grim Toll of Opioids, REUTERS (June 25, 2019, 9:00 AM), https://www.reuters.com/investigates/special-report/usa-courts-secrecy-judges [https://perma.cc/9VSL- PTFD].

22 For a discussion of how POs insulated Cooper Tire from scrutiny, see S. REP. NO. 110-439, at 6 (2008).

23 See Annysa Johnson & Ellen Gabler, Then-Archbishop Timothy Dolan Tried to Protect Money from Claims, Records Show, MILWAUKEE J. SENTINEL (July 2, 2013), https://archive.jsonline.com/news/religion/Cardinal-Dolan-sought-to-protect-money- from-claims-struggled-with-Vatican-to-defrock-abusers-b9943953z1-213832541.html [https://perma.cc/E5B8-3P5F] (explaining that information about abusive Catholic priests was hidden behind a "broad protective order").

24 Chelsea Hanlock, Note, Settling for Silence: How Police Exploit Protective Orders, 109 CALIF. L. REV. 1507, 1512, 1558-59 (2021) (concluding that POs prevent plaintiffs' firms from building "a database of problem officers or collect[ing] misconduct and training materials that show systemic departmental failures"). POs may also prolong problematic officers' tenures by preventing public pressure campaigns. See id. at 1522-24.

25 FED. R. CIV. P. 26(c). Fortifying this view, in the early 1990s, the Advisory Committee considered, and rejected, a proposed amendment to Rule 26(c) that would have diluted this good cause requirement. For discussion of that failed attempt, see infra notes 84-87 and accompanying text.

26 Pub. Citizen v. Liggett Grp., Inc., 858 F.2d 775, 789 (1st Cir. 1988) (quoting In re Agent Orange Prod. Liab. Litig., 821 F.2d 139, 145-46 (2d Cir. 1987)).

Page 5 of 54

ARTICLE: SECRECY BY STIPULATION

And, even though appellate courts' authority gets a little murkier when the parties jointly file the PO, most appellate courts hold that, even then, the trial court must independently conduct a good cause analysis.27 As the Seventh Circuit explains: "In deciding whether to issue a stipulated protective order, the district court must independently determine if 'good cause' exists."28 The Sixth likewise insists that a court "cannot abdicate its responsibility to oversee the [*105] discovery process and to determine whether filing should be made available to the public."29 In fact, says the Sixth Circuit: "A district court abuses its discretion," where it grants an uncontested motion for a PO and "makes neither factual findings nor legal arguments supporting the need for the order."30 The Third Circuit likewise holds that, even when both parties consent to a PO, Rule 26(c) still demands that the court make "an independent determination of 'good cause.'"31

Many have long suspected, however, that the on-the-ground reality is somewhat different. There have long been hints that, particularly when parties jointly file motions for POs, many busy and burdened trial court judges simply acquiesce to the parties' request.32 Indeed, fueled by that concern, reformers repeatedly tried to beef up Rule 26(c) beginning in the 1980s, believing that without fortification the provision wasn't adequately protective of the public interest. Yet, as this Article shows, reformers' efforts ultimately came to naught, in no small part because reformers were never able to amass convincing evidence that, in day-to-day practice, Rule 26(c)'s dictates were not being scrupulously followed. Absent such evidence, the reformers' campaign was overcome by inertia and by opponents' dogged insistence that nothing was amiss. Exemplifying this position, in 2009, U.S. District Judge Mark Kravitz registered opposition to reforms on behalf of the Judicial Conference. Testifying before the U.S. House of Representatives, he demanded: "What I want to hear is evidence of Federal courts . . . not doing what the rule says it should do, which is only grant protective orders for good cause shown."33 Unless reformers [*106] could muster that evidence, Judge Kravitz (and so many others) successfully argued, hands off.34

Reformers never could. And the almost total lack of evidence about POs' entry and effect has remained, to date, the entire subject's Achilles' heel. As one scholar recently observed: "[D]espite the sizable amount of existing scholarship about confidential discovery, scholars have not deeply considered how confidential discovery is implemented in practice."35 Thus, while some who advocate for changes to Rule 26(c) insist that POs are too easily

27 As explained in detail in Part I.C., this Article uncovers a surprising circuit split when it comes to stipulated POs. Two circuits the Eleventh and Ninth hold that the parties' say-so essentially substitutes for a judicial good cause analysis. But three circuits the Seventh, Sixth, and Third take the opposite view, holding that, even when a motion for a PO is jointly filed, the court must still conduct an independent Rule 26(c) review.

28 Jepson, Inc. v. Makita Elec. Works, Ltd., 30 F.3d 854, 858 (7th Cir. 1994).

29 Procter & Gamble Co. v. Bankers Tr. Co., 78 F.3d 219, 227 (6th Cir. 1996).

30 In re Nat'l Prescription Opiate Litig., 927 F.3d 919, 929 (6th Cir. 2019) (quotation marks, citation, and alteration omitted).

31 Littlejohn v. Bic Corp., 851 F.2d 673, 680 n.15 (3d Cir. 1988).

32 In fact, in an admittedly dated opinion, a district court went so far as to say that it was "unaware of any case in the past half- dozen years of even a modicum of complexity where an umbrella protective order . . . has not been agreed to by the parties and approved by the court." Zenith Radio Corp. v. Matsushita Elec. Indus. Co., 529 F. Supp. 866, 889 (E.D. Pa. 1981). For further discussion, see Gustavo Ribeiro, (Marked Confidential): Negative Externalities of Discovery Secrecy, 100 DENV. L. REV. 171, 190 (2022) (observing that "courts regularly enter stipulated proposed protective orders with little to no . . . inquiry into whether good cause exists").

33 Sunshine in Litigation Act of 2009: Hearing Before the H. Comm. on the Judiciary, Subcomm. on Com. & Admin. L., 111th Cong. 52 (2009) (statement of Hon. Mark R. Kravitz for the Judicial Conference of the United States).

34 In the words of Professor Arthur Miller: "Because proponents of reform have not demonstrated that significant modification of the present framework is necessary, the existing pragmatic and discretionary balancing technique should be retained." Arthur R. Miller, Confidentiality, Protective Orders, and Public Access to the Courts, 105 HARV. L. REV. 427, 491 (1991).

Page 6 of 54

ARTICLE: SECRECY BY STIPULATION

and too frequently granted, we've never known whether that is actually true. In fact, even basic empirical questions have so far resisted resolution. We haven't known how common stipulated POs are. We haven't known how closely judges scrutinize them. Nor have we known the kinds of cases in which stipulated POs actually appear.

This Article tackles these persistent and consequential questions. Drawing on a novel dataset consisting of more than 2.2 million federal court docket reports from 2005 through 2014, we show that approximately 45 percent of POs are the product of a jointly filed Rule 26(c) motion, a much larger proportion than previous estimates suggest. We show that the percentage of stipulated POs steadily grew over the study period among cases in which an answer was filed. And, most provocatively, we show that most judges grant all, or almost all, of the stipulated motions for POs they consider. We estimate that judges grant 95 to 97 percent of all joint requests for POs, depending on the year. And we find that over half of the judges who considered at least twenty-five stipulated POs in our dataset never denied a single such request.

We also supplemented our quantitative review with a qualitative analysis of 400 court orders 300 of which involved grants of the parties' joint request for a PO, and 100 of which involved (relatively rare) judicial denials. This analysis lends still further support to those who suggest that POs are routinely rubber-stamped. In all, 68 percent of stipulated POs in the "grant" dataset did not even pay lip service to [*107] Rule 26(c)'s good cause standard, and 83 percent lacked a particularized assessment of the parties' need for secrecy.

In short, stipulated POs are much more common than previously thought. They are granted at extremely high rates. They are often entered without so much as mentioning Rule 26(c)'s good cause standard. And their prevalence has grown over time.

The remainder of this Article proceeds in four Parts. Part I offers a primer on POs. This Part explains the role of POs, their close cousin sealing orders, and the legal standards for entering both. Then, it considers the peculiar case of stipulated POs. What are judges to do when litigants jointly ask for a PO? That depends. Some circuits permit judges to grant stipulated POs solely on the parties' say-so, but most others require a more searching inquiry.

Part II turns to policy and politics. It catalogs past reform efforts aimed at tightening Rule 26(c), identifies and assesses the arguments in favor of and against those reforms, and, lastly, points out that these efforts have sputtered because of something like empirical exhaustion. Reform opponents have hung their hats on the fact that there is, as defense attorney Robert N. Weiner puts it, "[n]o serious academic study, no Rand Corp. analysis, no state-by-state survey" that "has suggested any problem with protective orders in our courts."36 Absent such a study and absent evidence that day-to-day PO practice departs from Rule 26(c)'s command opponents have been able to insist that nothing is amiss.

The heart of this piece, Part III, offers that overdue academic study. Drawing on millions of federal court dockets, we present original evidence quantifying how often stipulated POs are granted, and we also bolster that quantitative assessment with a robust qualitative review. We find that stipulated POs are much more common than previously thought, are granted at extremely high rates, and often fail to so much as mention Rule 26(c)'s good cause standard. If there is, as Professor Arthur R. Miller says, a problem when a judge's entry of a PO is "automatic or cavalier,"37 then our evidence suggests we have a widespread problem in our courts.

[*108] Finally, Part IV steps back to assess the implications of our findings and to situate these findings in larger debates. Most concretely, our evidence torpedoes long-articulated reassurances concerning the exercise of fine-

35 Seth Katsuya Endo, Contracting for Confidential Discovery, 53 U.C. DAVIS L. REV. 1249, 1253 (2020).

36 Robert N. Weiner, Protective Orders and Nest-Feathering: Plaintiffs Lawyers Around the Nation are Pressing for Laws that Would Open Discovery Files in Personal-Injury Cases to Public Scrutiny. Their Motivations are Not Nearly as Pure as Many Suppose, LEGAL TIMES, Sept. 23, 1991, at 34.

37 Miller, supra note 34, at 491.

Page 7 of 54

ARTICLE: SECRECY BY STIPULATION

tuned, individualized judicial discretion in the entry of POs. Accordingly, this Article ought to reorient policymaking regarding Rule 26(c), and it ought to reinvigorate reform activity. But, just as clearly, what we uncover has implications far beyond Rule 26(c). Indeed, our findings touch upon, and contribute to, broad and enduring debates, including the role of private civil litigation in general and tort litigation in particular, our wavering commitment to the adversarial (rather than inquisitorial) resolution of disputes, the gulf between the law on the books and the law in action, and lower courts' sometimes surprising disobedience to official commands. I. THE FORMAL LAW REGARDING PROTECTIVE ORDERS

This Part offers a primer on POs. It unfolds in three steps. First, Section A provides a brief overview of POs, including how they differ from their close cousins, orders to seal. Section B explains how Rule 26(c) generally addresses POs. Then Section C zeroes in on stipulated POs. Section C explains why some plaintiffs might agree to keep certain information under wraps, and it also canvasses how courts interpret Rule 26(c)'s good cause standard when the PO motion is the product of party consent. A. Protective Orders 101

The discovery default is disclosure. A party that obtains information from an adversary through the civil discovery process is presumptively entitled to share that information with others.38 In the words of the Ninth Circuit: "It is well- established that the fruits of pretrial discovery are, in the absence of a court order to the contrary, presumptively public."39

However, as any litigator well knows, some of the information that passes through parties' hands in litigation realistically should not be shared. That is where POs come in. POs allow a party to override the [*109] pro- disclosure presumption by prohibiting parties from sharing the information they unearth.40

In the federal system, Rule 26(c) governs this "override" mechanism. It provides that any party or person "from whom discovery is sought may move for a protective order," and, upon such motion, "[t]he court may, for good cause, issue an order to protect a party or person from annoyance, embarrassment, oppression, or undue burden or expense."41 The resulting order may, among other things, prohibit discovery, condition its scope, prescribe special procedures for its dissemination, or limit who may view discovered or discoverable materials.42 Furthermore, just as a Rule 26(c) order may vary in how the information may be used, an order granted pursuant to Rule 26(c) may be particularized or capacious; it may restrict the disclosure of just a targeted set of documents, or instead may offer "blanket" (also known as "umbrella") protection.43

38 See Jepson, Inc. v. Makita Elec. Works, Ltd., 30 F.3d 854, 858 (7th Cir. 1994) ("Absent a protective order, parties to a law suit may disseminate materials obtained during discovery as they see fit.").

39 San Jose Mercury News, Inc. v. U.S. Dist. Ct. N. Dist. (San Jose), 187 F.3d 1096, 1103 (9th Cir. 1999).

40 See Seattle Times Co. v. Rhinehart, 467 U.S. 20, 34-35 (1984) (explaining that "the liberality of pretrial discovery" creates "significant potential for abuse" and that POs vindicate a substantial interest in preserving the secrecy of information that "could be damaging to reputation and privacy").

41 FED. R. CIV. P. 26(c).

42 Id. at 26(c)(1)(A)-(H).

43 Blanket POs "provide that all assertedly confidential material disclosed (and appropriately identified, usually by stamp) is presumptively protected unless challenged." DAVID F. HERR, ANNOTATED MANUAL FOR COMPLEX LITIGATION § 11.432 (4th ed. 2022). They are often entered "without a particularized showing [of good cause] to support the claim for protection." Id. Thus, blanket POs empower litigants to designate, for themselves, any document or deposition as confidential or sometimes highly confidential. Documents are subject to the terms of the PO unless (or until) the receiving party alleges that the document is not confidential, at which point the court will determine whether there is good cause to protect the file. Blanket orders are contrasted with "particularized protective orders," which, as the name suggests, cover only the materials specifically listed in the PO. Id.

Page 8 of 54

ARTICLE: SECRECY BY STIPULATION

One additional introductory note about POs relates to their close cousin: orders to seal.44 POs and orders to seal are very frequently confused, including by litigants and judges.45 But the two are different, and they are subject to different standards. An order to seal comes in once the document or testimony, unearthed during discovery, becomes a "judicial record," typically when it is "filed with the court, or otherwise somehow incorporated or integrated into a district court's [*110] adjudicatory proceedings."46 The paradigm instance is a document, produced by an opponent during discovery, that is attached as an exhibit to a summary judgment motion. Prior to the attachment process, the document is held privately by litigants and governed only by Rule 26(c).47 After the attachment process, the document is a "judicial record," for which restrictions on public disclosure are subject to a higher standard.48

Special protections for judicial records kick in because, unlike mere discovery material (for which there is no common-law right of access and, correspondingly, no First Amendment protection), there is a common-law and constitutionally protected right to access judicial records in part because these records necessarily play a role in the adjudicatory process.49 These records inform the court's determination, and, without the ability to scrutinize the ground for the court's decision, the public cannot assess the decision itself. Given this, motions to seal face a higher bar.50 Indeed, in some circuits, sealing decisions are governed by a kind of strict scrutiny. As the Fourth Circuit explains, when it comes to judicial records, "the denial of access must be necessitated by a compelling government interest and narrowly tailored to serve that interest."51 Compared to that strict scrutiny, as Sections B and C explain, Rule 26(c)'s "good cause" requirement is markedly less demanding.

[*111] B. Protective Orders, Generally

As noted above, Rule 26(c) states "[t]he court may, for good cause, issue an order to protect a party or person from annoyance, embarrassment, oppression, or undue burden or expense."52 Yet, although the standard is clear enough, when it comes to the formal interpretation of this standard, there is some surprising inter-circuit variation.

44 For cogent discussions of these mechanisms' differences and similarities, see June Med. Servs., L.L.C. v. Phillips, 22 F.4th 512, 521 (5th Cir. 2022); Shane Grp., Inc. v. Blue Cross Blue Shield of Mich., 825 F.3d 299, 305-07 (6th Cir. 2016).

45 For the fact that courts and litigants sometimes confuse the two standards, see Endo, supra note 35, at 1254 n.15.

46 In re Cendant Corp., 260 F.3d 183, 192 (3d Cir. 2001) (internal citations omitted).

47 See Seattle Times Co. v. Rhinehart, 467 U.S. 20, 32-33 (1984) ("A litigant has no First Amendment right of access to information made available only for purposes of trying his suit.").

48 See Cable News Network, Inc. v. FBI, 984 F.3d 114, 118 (D.C. Cir. 2021) (explaining that a "judicial record" is any document filed to "influence a judge's decisionmaking" and that, if a document is a "judicial record," there is a presumption of public access).

49 See Fair Lab'y Pracs. Assocs. v. Riedel, 666 F. App'x 209, 211-12 (3d Cir. 2016) ("A strong presumption in favor of public accessibility attaches to judicial records . . . . The presumption of public access is a common law doctrine that predates the Constitution.").

50 See, e.g., Pintos v. Pac. Creditors Ass'n, 605 F.3d 665, 678 (9th Cir. 2010) ("[A] party seeking to seal judicial records must show that compelling reasons supported by specific factual findings outweigh the general history of access and the public policies favoring disclosure." (quotation marks and alterations omitted)); In re Neal, 461 F.3d 1048, 1053 (8th Cir. 2006) (declaring that "only the most compelling reasons can justify non-disclosure of judicial records" (quotation marks omitted)).

51 Rushford v. New Yorker Mag., Inc., 846 F.2d 249, 253 (4th Cir. 1988).

52 FED. R. CIV. P. 26(c).

Page 9 of 54

ARTICLE: SECRECY BY STIPULATION

Four circuits the First, Fifth, Sixth, and Eighth maintain that "[a] finding of good cause must be based on a particular factual demonstration of potential harm" that would occur if the particular information were to be divulged, "not on conclusory statements."53 In these circuits, mere assertions that keeping the information under wraps is necessary, helpful, or valuable do not cut it.54 Two other circuits take a slightly different approach, explicitly weighing public versus private interests. The Seventh Circuit, for example, holds that "good cause" for secrecy exists when "the property and privacy interests of the litigants" predominate over the public's interest in the publication of relevant information.55 Likewise, in the D.C. Circuit, district courts are to "take[] into account" and balance "all relevant interests," including those of the litigants, third parties, relevant statutes, and even those protected by the First Amendment.56 And, in three other circuits the Third, Ninth, and Eleventh there is yet another standard. These circuits require something like particularized-interest-plus, where a movant seeking a PO must make a particularized [*112] showing of potential harm and demonstrate that the weight of interests favors secrecy.57 The upshot: Although they disagree on the particulars, all circuits at least when a PO is contested demand some substantial level of scrutiny. C. Stipulated Protective Orders

The above assumes that the motion for a PO is, and will be, contested. One litigant will seek the motion for a PO, and the other litigant will oppose that motion. Yet, it turns out, motions for POs often are not contested. As we explain in Part III.B, our data reveal that motions for POs are commonly stipulated sometimes called "unopposed," "joint," or "jointly filed." In fact, we find that motions for POs are stipulated nearly half of the time.

Given that it is typically defendants who are eager to keep unflattering information out of the public eye, it might, initially, seem odd that so many PO motions are jointly filed. But anecdotal evidence indicates that plaintiffs' lawyers often make the strategic decision, essentially, to go along to get along.58 Plaintiffs' lawyers, it is said, recognize that, without a PO, the defendant is apt to put the brakes on discovery and be less forthcoming when sitting for depositions, responding to interrogatories, and divulging documents.59 And, in the scheme of things, a PO battle is

53 Anderson v. Cryovac, Inc., 805 F.2d 1, 7 (1st Cir. 1986); see also EEOC v. BDO USA, L.L.P., 876 F.3d 690, 698 (5th Cir. 2017) (holding that Rule 26(c) "contemplates a particular and specific demonstration of fact" (quotation marks omitted)); Serrano v. Cintas Corp., 699 F.3d 884, 901 (6th Cir. 2012) ("This Circuit has endorsed the view that to justify a protective order, one of Rule 26(c)(1)'s enumerated harms must be illustrated with a particular and specific demonstration of fact, as distinguished from stereotyped and conclusory statements." (quotation marks omitted)); accord 8A RICHARD L. MARCUS, FEDERAL PRACTICE & PROCEDURE (WRIGHT & MILLER) § 2035 (2023 update) ("The courts have insisted on a particular and specific demonstration of fact, as distinguished from stereotyped and conclusory statements, in order to establish good cause.").

54 See, e.g., P.R. Med. Emergency Grp., Inc. v. Iglesia Episcopal Puertorriqueña, Inc., 318 F.R.D. 224, 233 (D.P.R. 2016) (rejecting a motion for a PO where the movant offered only broad allegations of harm that would befall it absent the order, "devoid of substantiation through facts").

55 Citizens First Nat'l Bank of Princeton v. Cincinnati Ins. Co., 178 F.3d 943, 945 (7th Cir. 1999).

56 United States v. Microsoft Corp., 165 F.3d 952, 960 (D.C. Cir. 1999).

57 See, e.g., Shingara v. Skiles, 420 F.3d 301, 306-07 (3d Cir. 2005); Phillips ex rel. Estates of Byrd v. Gen. Motors Corp., 307 F.3d 1206, 1210-11 (9th Cir. 2002); Kleiner v. First Nat'l Bank of Atlanta, 751 F.2d 1193, 1205-06 (11th Cir. 1985). The law is less certain in the remaining three circuits: the Second, Fourth, and Tenth. For the Second Circuit, see Haidon v. Town of Bloomfield, 552 F. Supp. 3d 265, 269 (D. Conn. 2021) (describing various approaches to assessing good cause and noting "the Second Circuit has not yet weighed in on the issue"); Schoolcraft v. City of New York, No. 1:10-cv-06005, 2013 WL 4534913, at *3 (S.D.N.Y. Aug. 27, 2013) ("There is a split within the district courts of this Circuit as to the showing necessary to establish that good cause exists."). For the Fourth Circuit, see Ardrey v. United Parcel Serv., 798 F.2d 679, 682 (4th Cir. 1986) (noting that district courts enjoy a "wide latitude in controlling discovery"). And for the Tenth Circuit, see Rohrbough v. Harris, 549 F.3d 1313, 1321 (10th Cir. 2019) (asserting that Rule 26(c)'s good cause standard is "highly flexible" (quoting Microsoft, 165 F.3d at 959)).

58 S. REP. NO. 110-439, at 9 (2008) (testimony of Leslie Bailey).

Page 10 of 54

ARTICLE: SECRECY BY STIPULATION

not the battle the lawyer wants to fight.60 As one plaintiffs' lawyer explained, the lawyer "may be so [*113] concerned with gaining access to the key documents she needs to present her client's case that she . . . may decide it isn't worth slowing down the litigation to fight."61 Or, as Lance Cooper, the lawyer for Ken and Beth Melton, put it: "[P]laintiff's lawyers want to try to get the documents as soon as possible to prosecute their case."62 A PO is thought to promote the prompt disclosure of documents and so motions for POs are often jointly filed.

Even when parties agree to the terms of a PO, however, that agreement is not judicially enforceable on its own. Even if the motion for a PO is jointly filed, the court must still enter the PO under Rule 26(c). But pursuant to what standard? Must a trial court still rigorously assess good cause, as it would if the Rule 26(c) motion were contested? In answering this question, appellate courts again differ on the particulars.63

In two circuits, party agreement temporarily obviates the good cause showing.64 Taking this tack, the Ninth Circuit holds: "While courts generally make a finding of good cause before issuing a protective order, a court need not do so where (as here) the parties stipulate to such an order."65 Similarly, in the Eleventh Circuit, stipulated POs may be entered based on agreement alone.66 However, both circuits also hedge, explaining that, because the parties "never [*114] established good cause for the protection in the first place," any "party seeking the stipulated order's protection must satisfy Rule 26(c)'s good cause standard" if secrecy is later challenged.67

Operationalizing this standard, one district court from Florida, answerable to the Eleventh Circuit, explains: "Parties have the freedom and flexibility to agree on the terms of stipulated protective orders designed to protect 'confidential' and 'highly confidential' material. . . . [C]ourts typically enter the proposed stipulated protective orders jointly submitted by the parties."68 Likewise, a district court in California, hailing from the Ninth Circuit, notes:

59 Professor Howard Erichson observed: "With a protective order in place, a responding party is more willing to turn over information rather than asserting and litigating every plausible relevance objection and privilege objection." Howard M. Erichson, Court-Ordered Confidentiality in Discovery, 81 CHI.-KENT L. REV. 357, 359 (2006).

60 See FRANCIS H. HARE, JAMES L. GILBERT & WILLIAM H. REMINE, CONFIDENTIALITY ORDERS 2 (1988) ("It is commonplace for defense counsel to offer to furnish the requested documents without opposition, if the plaintiff will stipulate to an order."); Miller, supra note 34, at 492 n.322 (suggesting that plaintiffs agree to POs "to facilitate [their] own access to discovery materials").

61 S. REP. NO. 110-439, at 9 (2008) (testimony of Leslie Bailey). There is, then, yet another question, which is how broad the PO is to be. For a discussion of various options, see supra note 43.

62 Telephone Interview with Lance Cooper, supra note 7, at 1.

63 For a cogent summary of this debate, see generally Brief for Amici Curiae Civil Procedure Law Professors in Support of Petitioners, Doe 7 v. Chiquita Brands Int'l, Inc., 142 S. Ct. 312 (2021) (No. 20-1599) [hereinafter Civil Procedure Scholar Brief].

64 The Fifth Circuit may well agree that stipulation eliminates the need for a good cause showing, but it has never said so explicitly. Cf. Binh Hoa Le v. Exeter Fin. Corp., 990 F.3d 410, 418-20 (5th Cir. 2021) (faulting a district court for entering a sealing order, but not a PO, based solely on the parties' stipulation).

65 In re Roman Cath. Archbishop of Portland in Or., 661 F.3d 417, 424 (9th Cir. 2011); see also Beckman Indus., Inc. v. Int'l Ins. Co., 966 F.2d 470, 476 (9th Cir. 1992) (explaining that parties did not have "to make a 'good cause' showing" when entering "a stipulated blanket [protective] order").

66 In re Chiquita Brands Int'l, Inc., 965 F.3d 1238, 1249-50 (11th Cir. 2020) (accepting the district court's practice of accepting stipulated POs without finding good cause); see also Chi. Trib. Co. v. Bridgestone/Firestone, Inc., 263 F.3d 1304, 1307 (11th Cir. 2001) (observing that the stipulation process "postpones the necessary showing of 'good cause' required for entry of a protective order until the confidential designation is challenged").

67 In re Chiquita, 965 F.3d at 1249-50.

Page 11 of 54

ARTICLE: SECRECY BY STIPULATION

"Although courts generally make a finding of good cause prior to issuing a protective order, a court need not do so if the parties stipulate to entry of a protective order."69

The Eleventh and Ninth Circuits, however, appear to be in the minority. In three circuits, the law is clear that trial court judges are duty-bound to subject even stipulated POs to careful scrutiny.70 Thus, the Seventh Circuit explains that, just as the trial court has a nondelegable duty to conduct an "independent inquiry" when assessing the fairness and adequacy of a class action settlement pursuant to Rule 23(e), "[i]n deciding whether to issue a stipulated protective order, the district court must independently determine if 'good cause' exists."71 Party agreement, the Seventh Circuit emphasizes, simply does not obviate the need for careful review. Similar to the Seventh Circuit, the Sixth Circuit observes that, even when the parties agree to a PO's terms, trial courts' "discretion [to enter that order] is limited by the careful dictates of Fed. R. Civ. P. [*115] 26."72 A trial court, the Sixth Circuit insists, simply "cannot abdicate its responsibility to oversee the discovery process and to determine whether filings should be made available to the public."73 Thus, says the Sixth Circuit: "A district court abuses its discretion" when it grants an uncontested motion for a PO "where it 'ma[kes] neither factual findings nor legal arguments supporting the need for' the order."74 Likewise, the Third Circuit stresses that, even when both parties consent to a PO's entry, Rule 26(c) still demands "an independent determination of 'good cause.'"75

Employing this stricter standard, a district court in the Seventh Circuit explains:

[E]ven if the parties stipulate to the terms of a protective order . . . the parties must satisfy the good cause requirement contained in Rule 26(c)(1) . . . . Since the Court is the primary representative of the public interest in judicial proceedings, it must review requests for stipulated protective orders for good cause without acting as a rubber stamp.76

68 Malibu Media, LLC v. Doe, No. 2:13-cv-836-FtM-29CM, 2014 WL 1292692, at *3 (M.D. Fla. Mar. 31, 2014) (citations omitted).

69 In re Facebook, Inc. Consumer Priv. User Profile Litig., No. 18-md-02843-VC (JSC), 2021 WL 3209711, at *3 (N.D. Cal. July 29, 2021); see also Ledford v. Idaho Dep't of Juv. Corr., No. 1:12-cv-00326-BLW, 2013 WL 5798682, at *1 (D. Idaho Oct. 28, 2013) ("If the parties stipulate to a protective order as they did here the district court may enter a protective order without first finding good cause.").

70 In some circuits, no case clearly establishes a governing standard for stipulated POs. See, e.g., Minter v. Wells Fargo Bank, N.A., Nos. WMN-07-3442, WMN-08-1642, 2010 WL 5418910, at *2 (D. Md. Dec. 23, 2010) ("Neither this Court nor the Fourth Circuit has explicitly defined the parameters of the initial 'good faith' review required pursuant to a stipulated confidentiality order . . . .").

71 Jepson, Inc. v. Makita Elec. Works, Ltd., 30 F.3d 854, 858 (7th Cir. 1994). Miller has also analogized the situation to the situation a court faces when reviewing a class action settlement. See infra note 81 and accompanying text.

72 Procter & Gamble Co. v. Bankers Tr. Co., 78 F.3d 219, 227 (6th Cir. 1996).

73 Id.

74 In re Nat'l Prescription Opiate Litig., 927 F.3d 919, 929 (6th Cir. 2019) (alteration in original) (citation omitted).

75 Littlejohn v. Bic Corp., 851 F.2d 673, 680 n.15 (3d Cir. 1988); accord Pansy v. Borough of Stroudsburg, 23 F.3d 772, 785-86 (3d Cir. 1994) (declaring it "[d]isturbing[]" that "some courts routinely sign orders which contain confidentiality clauses without considering the propriety of such orders, or the countervailing public interests which are sacrificed by the orders" and further advising: "[W]hether an order of confidentiality is granted at the discovery stage or any other stage of litigation . . . good cause must be demonstrated to justify the order").

76 Elder v. Bimbo Bakeries USA, Inc., No. 3:21-cv-00637-DWD, 2022 WL 3443766, at *2 (S.D. Ill. Aug. 17, 2022).

Page 12 of 54

ARTICLE: SECRECY BY STIPULATION

Similarly, a district court in Pennsylvania, answerable to the Third Circuit, observes: "Stipulated protective orders must still meet the requirements of Rule 26(c), which requires demonstrating the existence of confidential information and good cause as to why such information should not be disclosed."77

Other influential voices have also come down in favor of this harder-edged position. Take the Federal Judicial Center. It clearly states that, even when the parties file a joint motion for a PO, a judge can grant the motion only after making an independent determination [*116] "that the order is supported by good cause."78 The canonical Wright and Miller treatise likewise advises that POs "are not authorized simply on the requesting parties' say-so. . . . Even when the parties consent, the court may not enter an order unless Rule 26(c) is satisfied."79

Leading scholars, too, resist the notion that party agreement substitutes for good cause under Rule 26(c). "Judges," Miller explains, "must guard against any notion that the issuance of protective orders is routine, let alone automatic, even when the application is supported by all the parties."80 Miller elaborates: "When all the parties support the protective order . . . the court is faced with an essentially non-adversarial situation" similar to the fiduciary burden that federal judges bear when independently "evaluating a proposed class action settlement under Federal Rule 23(e)."81 Professor Howard Erichson concurs. According to Erichson, "[t]he parties' say-so alone, without some showing of a need for confidentiality, does not constitute good cause for the granting of a protective order."82 Or, as Professor Laurie Doré says:

Rule 26(c) authorizes a district court to issue a protective order only for "good cause shown." It does not carve out any exception, temporary or otherwise, for stipulated orders. Even if the parties agree to the terms of a protective order, then, they must still demonstrate good cause to justify its issuance.83

[*117] Lastly, there is something like a legislative history argument that further bolsters the position of the many courts, institutions, and academics who insist that party agreement does not constitute good cause under Rule 26(c). Namely, in the mid-1990s, rulemakers considered whether to amend Rule 26(c) to say that a PO could be issued "for good cause shown or on stipulation of the parties."84 The amendment was controversial, in part because

77 Sprinturf, Inc. v. Sw. Recreational Indus., Inc., 216 F.R.D. 320, 323 (E.D. Pa. 2003).

78 ROBERT TIMOTHY REAGAN, FED. JUD. CTR., CONFIDENTIAL DISCOVERY: A POCKET GUIDE ON PROTECTIVE ORDERS 6 (2012).

79 MARCUS, supra note 53, at § 2035.

80 Miller, supra note 34, at 492. Miller further advises that this "careful[]" review must be "tailor[ed]" to the facts of each case and "should take account of a kaleidoscope of factors, including the likely outcome on the merits, the value or importance of commercial or personal data, the identity of the parties and any apparent outside interests, [and] the existence of any threat to health and safety." Id.

81 Id. at 492 n.322.

82 Erichson, supra note 59, at 373.

83 Laurie Kratky Doré, Secrecy by Consent: The Use and Limits of Confidentiality in the Pursuit of Settlement, 74 NOTRE DAME L. REV. 283, 342 (1999); see also Endo, supra note 35, at 1259 ("After parties propose a stipulated protective order, the court must determine if there is good cause to issue the order."). A group of leading civil procedure scholars recently filed a Supreme Court brief that likewise asserts:

There is no basis in the Federal Rules for a court to approve a stipulated protective order without conducting an independent assessment of good cause. To the contrary, though parties often will stipulate to the entry of a discovery protective order, it is the court that ultimately must enter the order, and the court may do so only in compliance with Rule 26(c)'s good cause requirement.

Civil Procedure Scholar Brief, supra note 63, at 14 (internal quotations and brackets omitted).

Page 13 of 54

ARTICLE: SECRECY BY STIPULATION

it was understood at the time that, if the amendment passed, it would have "worsened the court secrecy problem."85 Ultimately, the amendment was defeated.86 Rejecting the amendment by voice vote, the Judicial Conference "expressed concern that the proposed rule would change existing practice by allowing entry of protective orders without a showing of good cause."87

Under the canon of statutory construction known as the "rejected proposal rule," nonadopted amendments can reflect what a statute is not.88 So the fact that the Judicial Conference expressly considered and rejected an amendment to Rule 26(c) that would have established that party agreement could stand in for good cause supplies some [*118] ammunition to those who believe that, when it comes to POs, the parties' say-so does not suffice.89

II. REFORM HISTORY: PERENNIAL PROPOSALS AND DEFICIENT DATA

This Part canvasses past and present debates regarding POs as well as the meager empirical evidence that has informed the discussion. Section A explains that POs have long been the subject of reform efforts, while Section B catalogs prominent arguments for and against those proposals. Then, Section C traces prior empirical studies, which have been well-intentioned but limited. Given these deficient data, it is no surprise that, to this point, debates addressing possible reforms to Rule 26(c) have backstopped on anecdote and hunches, rather than fact. A. Past Efforts To Limit Expansive Protective Orders

The modern debate over litigation secrecy and over the value of, and standards for issuing, POs dates back to the late 1980s, when the premier plaintiff-side organization, the Association of Trial Lawyers of America ("ATLA"), now renamed the American Association for Justice, put litigation secrecy high on its organizational agenda.90 In this initial push for greater transparency, ATLA pointing to a few high-profile scandals insisted that the information

84 Letter from Patrick E. Higginbotham, Civ. Rules Chair of Advisory Comm. to Comm. on Rules of Prac. & Proc. 18 (June 2, 1995) [hereinafter Letter of June 2, 1995], https://www.uscourts.gov/sites/default/files/fr_import/CV6-1995.pdf [https://perma.cc/4NYM-N3G8] (emphasis added).

85 Saundra Torry, Judges Reject Record-Secrecy Rule, WASH. POST (Mar. 14, 1995, 7:00 PM), https://www.washingtonpost.com/archive/politics/1995/03/15/judges-reject-record-secrecy-rule/2ec968fc-4f37-40ba-b430- 2fc1054a0d4e [https://perma.cc/9VMM-P53Q].

86 Id.

87 Letter of June 2, 1995, supra note 84, at 1-2; see also Linda Greenhouse, Judicial Conference Rejects More Secrecy in Civil Court, N.Y. TIMES, Mar. 15, 1995, at B9 (recounting some reasons for rejecting the amendment).

88 See FRANK B. CROSS, THE THEORY AND PRACTICE OF STATUTORY INTERPRETATION 65 (2009) (explaining that a rejected amendment "provides fairly persuasive evidence that the content of the amendment was not the legislative intent"); FDA v. Brown & Williamson Tobacco Corp., 529 U.S. 120, 144 (2000) (finding that Congress's decision to reject "bills that would have granted the FDA . . . jurisdiction" over tobacco clarifies that the FDA does not have that jurisdiction). Like many statutory canons, this theory of interpretation is not ironclad. See William N. Eskridge, Jr., Interpreting Legislative Inaction, 87 MICH. L. REV. 67, 71, 94-108 (1988) (describing the "rejected proposal rule" and critiquing it); NLRB v. C & C Plywood Corp., 385 U.S. 421, 427 (1967) (refusing to draw an inference as to intent from a rejected amendment).

89 Interestingly, in Glenmede Trust Co. v. Thompson, 56 F.3d 476 (3d Cir. 1995), the court rejected a reading of Rule 26(c) that would have been "tantamount to permitting the parties to control the use of protective orders" and, in so doing, drew on this rejected amendment. Id. at 485, 486 n.15.

90 ATLA Fights Secrecy in Litigation, ATLA ADVOC., Sept. 1989, at 1 (discussing a resolution that "discourage[d] attorneys from entering into secrecy agreements and . . . encourage[d] courts not to enter into or enforce secrecy agreements without good cause"). For a discussion of ATLA's early efforts, see Miller, supra note 34, at 442-43; Gail Diane Cox, Yearly Meeting: Sunshine in San Diego for ATLA, NAT'L L.J., July 30, 1990, at 3-4; Bill Wagner, Secrecy Betrays Justice; ATLA/Litigation; President's Opinion, NAT'L L.J., July 24, 1989, at 1, 3 (recounting examples of POs harming public health and describing ATLA's efforts to "encourage[] attorneys to resist" POs).

Page 14 of 54

ARTICLE: SECRECY BY STIPULATION

uncovered in litigation sometimes pertained to serious public health hazards and that, in such cases, courts that rubber-stamped expansive POs were complicit in perpetuating consumer harm.91

[*119] Responding to ATLA's concerns, in 1989, Congress called to order its first hearing on the matter, followed soon after by the introduction of a federal "Sunshine in Litigation" bill on the House floor.92 That initial bill sought to allow parties to share evidence pertaining to public health and safety with regulators even if that information was subject to a PO.93 Yet, even though the bill was quite narrow, opposition to it was broad. Rallied, in part, by Alfred Cortese, a hard-boiled corporate lobbyist, and buoyed by the prolific writing of Miller, who was then at Harvard, corporate interests parried back "pro-sunshine" arguments. Cortese, in all this, took quite a crabbed view of court transparency:

What is the public right to observe the legal system? What is it? . . . The right of the public to observe the court system means that they have a right to show up in court when there is a public trial. That is the extent of the public's right to observe.94

In the ensuing decades, although a few in Congress most notably, the indefatigable Senator Herb Kohl of Wisconsin continued to champion reform, the tug-of-war between sunlight and secrecy reached something of a stalemate.95 Thus, in a slow-motion Groundhog Day, between the late 1980s and the late 2010s, every few years, a version of a Sunshine in Litigation Act popped up in the House [*120] or Senate.96 Every few years, the legislation kicked off heated rounds of discussion. And every few years, the legislation was defeated.97

State efforts mostly followed the same trajectory. At roughly the same time the first Sunshine in Litigation Act was introduced in Congress, a wave of similar reforms made their way through state legislatures. In fact, between 1990

91 Miller, supra note 34, at 442 ("According to ATLA, protective orders . . . are being used with increasing frequency to hide deadly product defects or other 'public hazards' from the public."); LAWS. FOR CIV. JUST. & THE NAT'L CHAMBER LITIG. CTR., COMMENTS ON THE NEED FOR AMENDMENT OF FEDERAL RULE OF CIVIL PROCEDURE 26(C) 3 (1994) [hereinafter LCJ 1994 Comment], https://www.uscourts.gov/sites/default/files/fr_import/CV1994-04.pdf [https://perma.cc/E76K- DKAR] (explaining that early efforts were based on the claim "that information produced in litigation revealed serious defects in consumer products or public exposures to toxic materials" and that the courts issuing these expansive POs were "unwitting co- conspirators" in perpetuating harm); see also Russ M. Herman, Secrecy, Discovery Abuse Breed Unethical Conduct, NAT'L L.J., Aug. 1, 1988, at 18-21 (voicing early opposition to widespread use of POs).

Others, however, charged that the PO reform movement was not really motivated by the public interest and was, instead, "a camouflaged effort to get marketable information." See Tripp Baltz, Shhhh Confidentiality in the Courts, CHI. LAW., Jan. 1991, at 50.

92 See LCJ 1994 Comment, supra note 91, at 5.

93 Michelle Conlin, Dan Levine & Lisa Girion, Why Big Business Can Count on Courts to Keep its Deadly Secrets, REUTERS (Dec. 19, 2019, 12:00 PM), https://www.reuters.com/investigates/special-report/usa-courts-secrecy-lobbyist [https://perma.cc/2K8T-F4R8].

94 Id. For more on the defense bar's efforts, see Baltz, supra note 91, at 51 (noting that the president of the International Association of Defense Counsel "declared that one goal during his term would be to preserve civil defendants' rights to protective orders").

95 Corporate lawyers referred to the Sunshine in Litigation Act as the "perennial Kohl bill." Conlin et al., supra note 93.

96 Robert Timothy Reagan, The Hunt for Sealed Settlement Agreements, 81 CHI.-KENT L. REV. 439, 441 (2006).

97 Id. More recently, in 2019, during a congressional hearing on court transparency, Representative Jerrold Nadler indicated he planned to reintroduce the Sunshine in Litigation Act, although it does not appear that such an Act was introduced. See The Federal Judiciary in the 21st Century: Ensuring the Public's Right of Access to the Courts, Hearing Before the H. Subcomm. on Cts., Intell. Prop., & the Internet of the Comm. on the Judiciary, 116th Cong. 4-5 (2019) (statement of Rep. Jerrold Nadler).

Page 15 of 54

ARTICLE: SECRECY BY STIPULATION

and 1994, state legislators introduced some ninety-three discrete proposals to limit POs.98 Most of these bills died quick deaths.99 Yet, a smattering squeaked through.

In one of the first reforms, promulgated in 1990, the Texas Rules Committee revised Rule 76a to make it far more difficult for courts to issue POs or to seal documents pertaining to "public health and safety, or the administration of public office or the operation of government."100 The same year, Florida enacted the Sunshine State's version of the Sunshine in Litigation Act.101 It forbids POs that conceal "a public hazard or any information concerning a public hazard" or "information which may be useful to . . . the public in protecting [*121] themselves from injury which may result from the public hazard."102 Around the same time, Virginia enacted a statute specifying that, in lawsuits involving personal injury or wrongful death, no PO shall "prohibit an attorney from voluntarily sharing such materials or information with an attorney involved in a similar or related matter," subject to a few restrictions.103 In 1994, Washington enacted a provision subjecting POs to a balancing test that must consider the risk of public hazards.104 And, in 1995, Louisiana joined the budding movement, restricting the entry of POs and orders to seal "if the information or material sought to be protected relates to a public hazard."105

98 LCJ 1994 Comment, supra note 91, at 6. For a comprehensive compilation, see Miller, supra note 34, 429-30 n.7.

99 LCJ 1994 Comment, supra note 91, at 6 ("[N]otwithstanding the broad media coverage and legislative fervor used to promote such legislation and rules, only three such restrictive proposals out of 93 were ever adopted.").

100 See TEX. R. CIV. P. 76a (allowing courts to issue POs and sealing orders covering documents pertaining to "general public health and safety" only after determining that private interests outweigh public health risks and that the order is the least restrictive means of protecting private interests). For further discussion, see generally Lloyd Doggett & Michael J. Mucchetti, Public Access to Public Courts: Discouraging Secrecy in the Public Interest, 69 TEX. L. REV. 643 (1991) (explaining Rule 76a). In enacting Rule 76a, reformers overcame stiff opposition from business interests. See, e.g., David E. Chamberlain, Proposed Rule 76(a): An Elaborate, Time- Consuming, Cumbersome Procedure, 53 TEX. BAR J. 348, 348-49 (1990) (insisting that Rule 76a's enactment would impair settlement, delay the resolution of conflict, and imperil private information); Letter from Jack C. Goldstein to Charles Herring, Jr. (Dec. 26, 1989), in PROPOSED RULE 76A AND COMPANION AMENDMENTS TO RULE 166B(5) 170, 170-71 (1990) (threatening that, if Texas courts made it harder to access POs, "legitimate businesses . . . [would] locate outside Texas" and that "thieves and pirates [would] look to Texas for 'political asylum' from traditional principles of business ethics and morals").

101 FLA. STAT. ANN. § 69.081 (1990).

102 Id. A "public hazard" is defined as "an instrumentality, including but not limited to any device, instrument, person, procedure, product, or a condition of a device, instrument, person, procedure or product, that has caused and is likely to cause injury." Id. § 69.081(2). For more on the Florida Sunshine in Litigation Act's contemporary application, see Goodyear Tire & Rubber Co. v. Schalmo, 987 So. 2d 142, 145-46 (Fla. Dist. Ct. App. 2008).

103 VA. CODE ANN. § 8.01-420.01(A). For more on the law's 1989 enactment, see Alan B. Morrison, Protective Orders, Plaintiffs, Defendants and the Public Interest in Disclosure; Where Does the Balance Lie?, 24 U. RICH. L. REV. 109, 122-23 (1989).

104 WASH. REV. CODE ANN. § 4.24.611.

105 LA. C.C.P. ART. 1426(C). The provision also prevents courts from issuing POs where the information may be useful to members of the public in protecting themselves from injury that might result from a public hazard. Id. The provision, however, has less bite than it may first appear. For a discussion of the limits of Louisiana's rule, see Dustin B. Benham, Tangled Incentives: Proportionality and the Market for Reputation Harm, 90 TEMP. L. REV. 427, 450-52 (2018) (noting that Louisiana's restriction "does not apply to agreements to protect trade secrets" or other confidential information, a carve-out that "swallows the rule"); Roma Perez, Two Steps Forward, Two Steps Back: Lessons to Be Learned from How Florida's Initiatives to Curtail Confidentiality in Litigation Have Missed Their Mark, 10 FLA. COASTAL L. REV. 163, 218-19 (2009) (identifying significant ambiguity in Louisiana's legislation).

Page 16 of 54

ARTICLE: SECRECY BY STIPULATION

After a very long hiatus and catalyzed in part by the #MeToo movement transparency efforts may be seeing a tentative resurgence. In 2022, Washington, D.C. councilmember Mary Cheh introduced the Sunshine in Litigation Act. The legislation would require D.C. judges to consider public health and safety before granting a PO or sealing court records.106 Cheh cited the opioid epidemic as an impetus for her bill, noting, "[c]ourt-sanctioned secrecy in such cases can be a matter of life and death."107 Also in 2022, California state senator Connie Leyva introduced the Public Right to Know Act of 2022.108 The bill [*122] would create a presumption against court orders concealing information about defective products or environmental hazards unless the court finds that the public interest in disclosure is outweighed by a substantial need for secrecy.109 Yet, like so many of their predecessors, both bills were defeated.110

B. Political and Policy Debates over Protective Orders

As suggested above, thirty years' worth of reform efforts have generated heated arguments for and against change. This is so despite the modesty of many of the proposals. Reformers have never sought, and virtually no one has ever advocated, an outright ban on POs. Everyone seems to agree that trade secrets and the protection of personal privacy justify POs in some circumstances.111 Instead, the fight over POs has been protracted and bitter. But it has been waged on markedly narrow terrain. Reformers have tended to focus on POs (regardless of whether they are the product of party contestation or consent) that conceal health and safety hazards. And reformers have sought not to outlaw such POs but merely to subject those and only those to exacting judicial scrutiny.112 Opponents have tended to resist these reforms. Below, we rehearse the two sides' now-familiar positions. 1. Reformers' Arguments in Favor of Greater Transparency.

Arguments in favor of greater transparency tend to focus on harm, cost, and institutional legitimacy. First, reformers highlight harm and, in particular, POs' well-documented role in concealing and thus perpetuating catastrophic injury. POs, in reformers' telling, prevent [*123] consumers and regulators from learning of defective products and allow those products to circulate unabated.113 They cause people to associate with abusers.114 And they make it harder for litigants to hold bad actors, including violent police departments115 and discriminatory employers,116 to account.

106 Sally Greenberg, Sunshine in Litigation Act Introduced in the District of Columbia, NAT'L CONSUMERS LEAGUE (Nov. 2, 2022), https://nclnet.org/dc-sunshine-in-litigation-act [https://perma.cc/CA7L-LRFQ].

107 Id.

108 S.B. 1149, 2021-2022 Cal. Leg. Reg. Sess. (Cal. 2022).

109 Id.

110 The California Senate passed the legislation, but the effort died in the Assembly. Id. A public hearing was held on the D.C. legislation in January 2023, but the bill progressed no further. Greenberg, supra note 106.

111 Richard L. Marcus, The Discovery Confidentiality Controversy, 1991 U. ILL. L. REV. 457, 487 (explaining that reformers "accept broad confidentiality interests in commercial litigation and to protect personal privacy").

112 See, e.g., S. REP. NO. 110-439, at 10 (2008) (explaining that the Sunshine in Litigation Act "merely requires an additional step consideration of public health and safety"); Sunshine in Litigation Act of 2008: Hearings on H.R. 5884 Before the H. Comm. on the Judiciary, Subcomm. on Com. & Admin. L., 110th Cong. 2 (2008) (statement of Rep. Linda T. Sanchez) ("H.R. 5884 is modest in its scope. Its key provision would require courts to do what some Federal Judges already do: consider the public's interest in health and safety before entering certain confidentiality orders that would conceal information from the public . . . .").

113 Dustin B. Benham, Dirty Secrets: The First Amendment in Protective-Order Litigation, 35 CARDOZO L. REV. 1781, 1785-86 (2014); Richard Zitrin, The Judicial Function: Justice Between the Parties, or a Broader Public Interest, 32 HOFSTRA L. REV. 1565, 1565-66 (2004). See generally Egilman et al., supra note 11 (collecting pharmaceutical examples).

114 See Johnson & Gabler, supra note 23 (explaining that information about abusive Catholic priests was hidden behind a "broad protective order").

Page 17 of 54

ARTICLE: SECRECY BY STIPULATION

As one reformer summarized in an early Senate hearing: "Men, women, and children in this country are unwittingly buying and using dangerous products, inhaling and drinking toxic pollutants, [and] being treated by incompetent doctors . . . because of unnecessary secrecy in the courts."117

Second, and relatedly, some argue that reform is needed because the status quo is not working. Left to their own devices, judges are not adequately safeguarding the public interest. POs, reformers insist, have become commonplace; they are "routinely requested in virtually every product liability, automobile design, toxic tort, environmental, medical malpractice, pharmaceutical, and consumer fraud case in the country."118 And, rather than scrutinizing POs for good cause, judges too often simply offer a stamp of approval on the parties' say-so. As one commentator put it: "[S]tipulated protective orders are often approved pro-forma by overburdened courts anxious to avoid time consuming inquiries into discovery disputes."119 As another explained [*124] in Senate testimony when asked what factors judges consider when deciding whether to issue a PO: "I think the truth is that judges rarely weigh these orders at all."120 Instead, "what happens is that the parties agree to them and the judges, faced with the parties agreeing and no one objecting, simply sign off."121 Too often, adds U.S. District Judge Joseph F. Anderson Jr. of South Carolina, he and his colleagues merely "rubber-stamp confidentiality orders presented to them, sometimes altogether ignoring or merely giving lip service to the body of law and existing court rules that are supposed to apply."122

Third, unbridled POs, reformers allege, raise the cost of litigation. POs stymie plaintiffs' cooperation across actions involving overlapping issues or defendants, requiring each plaintiffs' lawyer to build cases in isolation and from

115 See Hanlock, supra note 24, at 1558-59 (concluding that POs prevent plaintiffs' firms from building "a database of problem officers or collect[ing] misconduct and training materials that show systemic departmental failures").

116 See Jamillah Bowman Williams, Diversity as a Trade Secret, 107 GEO. L.J. 1685, 1727 (2019) (arguing that POs may soon be used to safeguard companies' diversity data, "making it impossible for other potential litigants to determine whether they also have a related [employment discrimination] claim" and derailing "systemic discrimination claims").

117 Examining the Use of Secrecy and Confidentiality of Documents by Courts in Civil Litigation, Hearing Before the S. Comm. on the Judiciary, Subcomm. on Cts. & Admin. Prac., 101st Cong. 66 (1990) (statement of Arthur H. Bryant, Exec. Dir. Trial Laws. for Pub. Just.) [hereinafter Bryant Testimony].

118 Id. at 60; see also Lori E. Andrus, Fighting Protective and Secrecy Orders: Sunshine is the Best Disinfectant, PLAINTIFF MAG. (2014), https://plaintiffmagazine.com/recent-issues/item/fighting-protective-and-secrecy-orders-2 [https://perma.cc/NNH3- U5GW] ("Protective orders have become routine, particularly in complex cases.").

119 Seymour Moskowitz, Discovering Discovery: Non-Party Access to Pretrial Information in the Federal Courts 1938-2006, 78 U. COLO. L. REV. 817, 826 (2007); see also Andrus, supra note 118, at 1 (asserting that "with courts overburdened and understaffed, judges are all too often content to sign off on blanket protective orders without any showing of good cause"); Dustin B. Benham, Foundational and Contemporary Court Confidentiality, 86 MO. L. REV. 211, 222 (2021) ("The parties often agree to protective orders supported by thin or non-existent proof of good cause. Courts enter them because the parties have agreed."); Conlin et al., supra note 93 (stating that the entry of a PO is a "pro forma exercise[]").

120 Bryant Testimony, supra note 117, at 162.

121 Id.; see also Federal Sunshine in Litigation Act and Federal Court Settlements Sunshine Act, Hearings on H.R. 2017 and H.R. 3803 Before the H. Comm. on the Judiciary, Subcomm. on Intell. Prop. & Jud. Admin., 102d Cong. 85 (1992) (statement of Rep. William J. Hughes) ("Unfortunately, I have the perception, as do a number of my colleagues, that the courts are, just as a matter of course, approving secrecy orders when submitted without making the independent determination as to whether they serve the public good.").

122 Joseph F. Anderson Jr., Hidden from the Public by Order of the Court: The Case against Government-Enforced Secrecy, 55 S.C. L. REV. 711, 715 (2004). In the article, Judge Anderson quoted another judge who confessed, in a moment of candor, that she "would sign an order that stipulated that the moon was made out of cheese if the lawyers came in and asked me to sign it." Id. at 729 (quoting Judge Judith McConnell of the San Diego, California, Superior Court).

Page 18 of 54

ARTICLE: SECRECY BY STIPULATION

scratch.123 To quote a lawyer for Public Citizen's Litigation Group, "[e]ssentially [a PO] makes every plaintiff's lawyer reinvent the wheel in every case."124 This siloing can, in turn, impair litigation efficiency and may systematically slant (or, some suggest, [*125] further slant) the litigation playing field toward well-heeled players.125 Not merely theoretical, these dynamics were on vivid display in early tobacco litigation. In early battles, cigarette company disclosures were subject to expansive POs, and such orders prevented plaintiffs' counsel, who were typically cash-strapped solo practitioners, from sharing discovery with one another.126 Unable to divulge what they learned, plaintiffs' lawyers battling the cigarette companies were condemned to build each case alone and from scratch, which dramatically but asymmetrically increased the cost and burden of litigation.127

Fourth, reformers argue that overbroad POs inhibit transparent judicial decision-making and that transparency is a good unto itself, important to the promotion of judicial accountability and essential to democratic processes.128 As Senator Kohl once put it, "our court system must also do its part to help protect the public, when appropriate, and not just individual plaintiffs and defendants."129 As vindicators of the [*126] public interest, courts should

123 See, e.g., Francis H. Hare Jr., James L. Gilbert & Matthew S. Ellenberger, Confidentiality Orders in Products Liability Cases, 13 AM. J. TRIAL ADVOC. 597, 602 (1989) ("By prohibiting the disclosure of information gleaned from discovery, a confidentiality order forces each attorney to develop his client's case in a vacuum."). Plaintiffs can always attempt to gain access to prior discovery by intervening to modify POs. But there is no guarantee that courts will grant their modification request, and, even if the request is granted, there is no guarantee that discovery will be forthcoming. For discussion, see Benham, supra note 9, at 2211-12.

124 Paul M. Barrett, Protective Orders Come Under Attack Plaintiffs Get Judges to Open Court Files, WALL ST. J., Aug. 31, 1988, at 27 (quoting C.F. Hitchcock).

125 See Erichson, supra note 59, at 369 (explaining that, when it is not stymied by a PO, "[c]oordination among counsel in related cases not only promotes litigation efficiency, but also enhances the quality of legal work and tends to level the field in asymmetrical multiparty litigation"); see also HARE ET AL., supra note 60, at 15-19 (discussing coordination in litigation support groups). For the classic account of why litigation may be slanted toward certain "repeat" players, see generally Marc Galanter, Why the "Haves" Come out Ahead: Speculations on the Limits of Legal Change, 9 LAW & SOC'Y REV. 95 (1974).

126 Nora Freeman Engstrom & Robert L. Rabin, Pursuing the Public Health Through Litigation: Lessons from Tobacco and Opioids, 73 STAN. L. REV. 285, 298 (2021).

127 See Marc Z. Edell, Cigarette Litigation: The Second Wave, 22 TORT & INS. L.J. 90, 91 (1986) ("[P]rotective orders, obtained by the defendants in almost every instance, prohibited the dissemination of discovery to either the public or to other lawyers who were involved in similar litigation. This required plaintiffs' lawyers to initiate discovery anew in each case."); Robert L. Rabin, A Sociolegal History of the Tobacco Tort Litigation, 44 STAN. L. REV. 853, 860 (1992) (explaining that, in the early tobacco litigation, POs prevented plaintiffs' lawyers from "collaborat[ing] or realiz[ing] economies of work-product"); Karen E. Meade, Commentary, Breaking Through the Tobacco Industry's Smoke Screen: State Lawsuits for Reimbursement of Medical Expenses, 17 J. LEGAL MED. 113, 119 (1996) (explaining that, in the first wave of the tobacco litigation, cigarette companies imposed massive costs on their underresourced adversaries by obtaining POs "to prevent the release of information to other potential plaintiffs" and asserting that, due to these POs, in each new case, "new discovery had to be undertaken").

128 See Judith Resnik, The Privatization of Process: Requiem for and Celebration of the Federal Rules of Civil Procedure at 75, 162 U. PA. L. REV. 1793, 1835-36 (2014) (explaining that courts serve and must serve "as a site of democratic practices").

129 The Sunshine in Litigation Act, Hearing on S.1404 Before the S. Comm. on the Judiciary, Subcomm. on Cts. & Admin. Prac., 103d Cong. 1 (1994) (statement of Sen. Herbert Kohl); see also Bryant Testimony, supra note 117, at 56 ("I think the bottom-line question here is whether the courts are designed simply to resolve private disputes without regard for their effect on the public, or whether the public is the one that not only funds the system, but should determine what happens in it."). Additionally, Judge Mikva testified:

I think that many scholars, and lawyers, and even judges forget that the courts are public institutions. They talk about privacy interests and the importance of respecting consensual positions as if the only two parties in interest in the court system are the parties to the lawsuit. I have never been able to understand how we could justify the heavy expenditure of public funds and resources on the courts if the only interest to be served is that of the litigants.

Page 19 of 54

ARTICLE: SECRECY BY STIPULATION

meaningfully interrogate all efforts to restrict public access including not just judicial procedures but also the fruits of discovery processes.130

Fifth and finally, in what is perhaps less an argument than an anti-argument, reformers argue that the reforms they champion would not, contrary to critics' claims, cause the sky to fall. For example, as detailed below, critics warn that, if reforms were enacted, limits on POs would intensify acrimony, complicate and prolong the discovery process, and ultimately clog courts.131 Reformers do not buy these or other dark predictions. Illustrating: Legendary jurist Abner Mikva argued in Senate testimony that restrictions on POs could accelerate (rather than complicate) litigation because, if a defendant cannot count on an expansive PO, the defendant may be more likely to proffer a generous settlement right out of the gates, even prior to the start of discovery.132

2. Critics' Arguments Opposing Reform Efforts. Numerous scholars, practitioners, and policymakers see matters differently and vehemently oppose state and federal reform activity. Critics' arguments tend to fall into one of five buckets.

First, critics have consistently and powerfully voiced an "if it ain't broke, don't fix it" argument.133 This argument has a few flavors. In its [*127] first guise, opponents challenge reformers' claim that overbroad POs have concealed health hazards from public attention. Seizing this baton, Miller, for instance, wrote in a classic 1991 Harvard Law Review piece: "The allegation that protective orders are concealing information important to public health and safety obviously should arouse concern, but its validity is doubtful" and, later in the piece, doubled down, concluding that "no evidence has been presented that the current [protective order] practice has created significant risks to public health or safety."134 Likewise, in 1994, in a Senate hearing to assess a federal Sunshine in Litigation Act, a witness from Lawyers for Civil Justice reassured the committee: "[P]rotective orders are not preventing the public from obtaining information needed to protect[] public health or safety, as has been alleged."135 Piling on, defense lawyer

The Sunshine in Litigation Act, Hearing on S.1404 Before the S. Comm. on the Judiciary, Subcomm. on Cts. & Admin. Prac., 103d Cong. 40-41 (1994) (statement of Hon. Abner J. Mikva) [hereinafter Mikva Testimony].

130 See David Luban, Settlements and the Erosion of the Public Realm, 83 GEO. L.J. 2619, 2656-57 (1995) (articulating a "public-life conception" of the courts and describing information generation as "a fundamental public interest" of litigation, not a side effect); Owen M. Fiss, The Supreme Court, 1978 Term Foreword: The Forms of Justice, 93 HARV. L. REV. 1, 29 (1979) ("[C]ourts exist to give meaning to our public values, not to resolve disputes.").

131 See infra notes 142-43 and accompanying text.

132 See Mikva Testimony, supra note 129, at 76. For similar analysis, see Scott A. Moss, Illuminating Secrecy: A New Economic Analysis of Confidential Settlements, 105 MICH. L. REV. 867, 874 (2007).

133 Robert Weiner, Being Sued Doesn't Mean Being Stripped of Privacy: Some Plaintiffs' Lawyers Want the Right not only to Delve into Defendants' Files, But Also to Publicize Whatever they Find, RECORDER, Feb. 8, 1990, at 2.

134 Miller, supra note 34, at 478, 501; see also Arthur R. Miller, Commentary, Private Lives or Public Access: The Debate Over Courthouse Confidentiality, 77 A.B.A. J. 65, 68 (1991) [hereinafter Miller, Private Lives] ("There is simply no reason to believe that current court rules and practices create any risks to public health or safety. Indeed, all indications are that the current system works rather well.").

135 The Sunshine in Litigation Act, Hearing on S.1404 Before the S. Comm. on the Judiciary, Subcomm. on Cts. & Admin. Prac., 103d Cong. 92 (1994) (statement of Alfred W. Cortese on Behalf of Laws. for Civ. Just.); see also Richard J. Vangelisti, Proposed Amendment to Federal Rule of Civil Procedure 26(c) Concerning Protective Orders: A Critical Analysis of What It Means and How It Operates, 48 BAYLOR L. REV. 163, 175-76 (1996) ("[E]mpirical data does not support the conclusion that protective orders are a threat to public health and safety."); Marcus, supra note 111, at 464 ("Despite the widely publicized instances of supposed cover-ups of hazards, hard data is generally lacking and the critics' broader assertions about widespread abuses may be validly questioned.").

Page 20 of 54

ARTICLE: SECRECY BY STIPULATION

Robert N. Weiner, a longtime Sunshine in Litigation Act critic, has likewise gone on record dismissing reformers' argument that POs have "conceal[ed] information critical to public safety" as "contrived."136

Second, in another spin on the "if it ain't broke" theme, critics maintain that reformers' concerns about judicial rubber-stamping are also overstated even chimerical. In critics' view, the PO status quo is working as intended: POs are rarely issued, and, when they are issued, they reflect carefully exercised judicial discretion. On the former, Representative Chris Cannon of Utah argued when opposing the federal Sunshine in Litigation Act of 2008: "This bill is unnecessary [*128] because discovery protective orders are rare."137 On the latter, Professor Richard Marcus has argued that, under the law, judges must apply Rule 26(c)'s "principles with care" and that "[j]udges generally seem to be doing just that."138 In his canonical 1991 piece, Miller likewise insisted: "The 'good cause' requirement is [already] strict. Federal courts have interpreted the rule to mean that the party seeking confidentiality must make a particularized factual showing of the harm that would be sustained if the court did not grant a protective order."139 Later, Miller added: "Under existing law, the courts have discretion to accept or reject" POs, and, in operation, "they exercise" this discretion responsibly "on a case-by-case basis."140

Third, those who oppose reform predict that limiting POs would impose high litigation costs on parties and the public. In particular, critics argue that broad protections on the disclosure of discovery material facilitate the free flow of information and, in so doing, reduce interparty acrimony and promote the efficient resolution of disputes.141 The converse is also true; without POs, critics reason, discovery would grind to a halt.142 In the words of one

136 Weiner, supra note 36, at 34.

137 The Sunshine in Litigation Act of 2008: Hearings on H.R. 5884 Before the H. Comm. on the Judiciary, Subcomm. on Com. & Admin. L., 110th Cong. 35 (2008) (statement of Rep. Chris Cannon).

138 Marcus, supra note 111, at 506.

139 Miller, supra note 34, at 433; see also, e.g., LAWS. FOR CIV. JUST., SEALING FATE: THE PROPOSAL TO RESTRICT JUDICIAL DISCRETION OVER SEALING CONFIDENTIAL INFORMATION WOULD IMPOSE UNWORKABLE STANDARDS ON THE COURTS, CONFLICT WITH STATUTORY PRIVACY RIGHTS, AND STOKE UNPRECEDENTED SATELLITE LITIGATION 1, 4 (2021) (insisting that judges do not rubber-stamp motions for POs); Zenith Radio Corp. v. Matsushita Elec. Indus. Co., 529 F. Supp. 866, 889 n.40 (E.D. Pa. 1981) ("[W]e doubt that any judge would approve a consent order not demonstrably rooted in Rule 26(c) . . . ."); Doré, supra note 83, at 302 (questioning "claims that federal district courts have perfunctorily acceded to a plethora of stipulated requests for discovery protective orders").

140 Miller, supra note 34, at 436. In another publication, Miller asserted:

When information possibly implicating public health and safety surfaces in documents produced in litigation, the decision about whether it should be released to the public should rest where it always has within the sound discretion of the court. Only the trial judge has no axe to grind and no prospect of pecuniary gain. Existing rules and procedures are more than adequate to accomplish this end.

Miller, Private Lives, supra note 134, at 67.

141 See, e.g., Richard L. Marcus, Myth and Reality in Protective Order Litigation, 69 CORNELL L. REV. 1, 21-23 (1983) (suggesting that greater access to POs reduces resistance to discovery); Benham, supra note 105, at 430 ("The reasoning goes that without some confidentiality, litigants would zealously resist producing relevant information and settle fewer cases.").

142 Vangelisti, supra note 135, at 178 ("[R]ather than facilitating the plaintiffs' cases, the lack of protective orders would grind discovery to a halt and increase the costs of litigation."); Miller, supra note 34, at 483 ("Limiting the availability of protective orders makes the discovery process more contentious, protracted, and expensive."); Sunshine in Litigation Act of 2008: Hearings on H.R. 5884 Before the H. Comm. on the Judiciary, Subcomm. on Com. & Admin. L., 110th Cong. 38 (2008) (statement of Prof. Arthur Miller) ("Confidentiality is of paramount importance during discovery because the willingness of the parties to produce information voluntarily often hinges on a guarantee that it will be preserved. Remove this guarantee and discovery will become more contentious, requiring frequent court intervention."); Examining the Use of Secrecy and Confidentiality of Documents by Courts in Civil Litigation, Hearing Before the S. Comm. on the Judiciary, Subcomm. on Cts. &

Page 21 of 54

ARTICLE: SECRECY BY STIPULATION

defense lawyer [*129] at a Senate hearing: "[I]f the defendant is not sure that its secret formula is going to be protected in discovery, then what's going to happen? It's going to fight. It's going to fight producing it, and that takes time and resources of the court and the parties."143

Fourth, some critics argue that any effort to cut down on litigation secrecy via beefed-up restrictions on POs will predictably backfire. There are, again, two flavors to this critique. First, in an age when arbitration is seemingly ascendant, some fret that, without POs, litigants will simply opt out; they will privatize their disputes, avoiding courts altogether. Giving voice to this concern, Professor Richard Marcus has written: "[O]pening up the discovery process could have the ironic effect of deterring claimants from seeking relief in court to avoid the resulting publicity."144 On this thinking, some secrecy in the form of a PO is perhaps less than ideal, but it is better than a sharper turn away from judicial administration. Relatedly, some others argue that, particularly in the current age of Federal Rule of Civil Procedure 26(b)(1)'s proportionality review, without POs, some courts might be more inclined to short-circuit discovery and deny a party the right to [*130] information, full stop.145 If that's right and the choice is between more discovery (but subject to a PO) versus less discovery (but not subject to a PO) it is not obvious, if one values transparency, that the latter is a better bet.146

Fifth and finally, critics argue that any effort to restrict POs to further the public interest is fundamentally misguided because courts simply do not exist to advance the public interest; they exist exclusively to resolve private interparty disputes.147 And any extra review of POs, particularly when POs are jointly requested, diverts courts' attention from

Admin. Prac., 101st Cong. 186 (1990) (written responses from Hon. Joseph F. Weis Jr.) ("Without the availability of protective orders, the time for disposing of litigation would unquestionably increase, and the public would be poorly served by the additional delay."); Examining the Use of Secrecy and Confidentiality of Documents by Courts in Civil Litigation, Hearing Before the S. Comm. on the Judiciary, Subcomm. on Cts. & Admin. Prac., 101st Cong. 190-91 (1990) (statement of Prof. Arthur R. Miller) ("If people couldn't voluntarily agree on confidentiality . . . . litigants would then be given an incentive to engage in trench warfare not to reveal the proprietary, the important, the private."); Marcus, supra note 111, at 484-85 (contending that "presumptive public access would disrupt orderly pretrial preparation by fomenting opposition to broad discovery" and "the cooperative exchange of information between the parties").

143 See, e.g., The Sunshine in Litigation Act: Does Court Secrecy Undermine Public Health and Safety, Hearing Before the S. Comm. of the Judiciary, Subcomm. on Antitrust, Competition Pol'y & Consumer Rts., 110th Cong. 10 (2007) (statement of Robert N. Weiner).

144 See Marcus, supra note 111, at 486; see also The Sunshine in Litigation Act, Hearing on S.1404 Before the S. Comm. on the Judiciary, Subcomm. on Cts. & Admin. Prac., 103d Cong. 29 (1994) (statement of Hon. Patrick E. Higginbotham) (explaining that, if POs were sharply cabined, "the increased discovery contests would . . . add to the pressures that encourage some parties to pursue nonpublic means of dispute resolution").

145 In re Halkin, 598 F.2d 176, 195 (D.C. Cir. 1979) ("The only plausible alternative to a protective order may be the denial of discovery altogether."); Miller, supra note 34, at 476 ("[I]f judges' discretion to issue protective orders is undercut, the courts' only means of maintaining privacy might be to deny discovery altogether."); id. at 484 ("[C]ontrary to the hopes of the proponents of public access, the net effect of banning protective orders might well be a constriction in the flow of litigation information, not an expansion."); Jack H. Friedenthal, Secrecy and Civil Litigation: Discovery and Party Agreements, 9 J.L. & POL'Y 67, 97 (2000) (predicting that if courts "cannot avoid problems through the use of protective orders, they are more likely to do so by curtailing discovery at the outset"); cf. Seattle Times Co. v. Rhinehart, 467 U.S. 20, 34 (1984) (noting that "it is necessary for the trial court to have the authority to issue protective orders conferred by Rule 26(c)" to offset the "liberality of pretrial discovery"). As an aside, but an important one, the assumption underlying this argument is dubious. If the Rule 26(b)(1) proportionality standard would warrant the decline of discovery due to disclosure concerns, that fact alone establishes good cause for a PO to prevent disclosure meaning that, even if Rule 26(c) were fortified in line with reformers' desires, the "problem" opponents worry about still would not materialize. For a discussion of the proportionality standard and the many normative judgments that its application embeds, see generally Jonah B. Gelbach & Bruce H. Kobayashi, The Law and Economics of Proportionality in Discovery, 50 GA. L. REV. 1093 (2016).

146 Again, as explained at supra note 145, this argument relies on a misapplication of Rules 26(b)(1) and 26(c).

147 E.g., Weiner, supra note 36, at 34 ("The courts' job in civil cases is to resolve disputes between private parties.").

Page 22 of 54

ARTICLE: SECRECY BY STIPULATION

this fundamental purpose.148 On this view, any discussion of POs and the public interest simply misses the point. If there is no dispute over POs, "then courts should readily accede to the parties' mutual desire for confidentiality."149

[*131] C. Why Debate Has Stalled: Dueling Anecdotes and Anemic Empirics

As noted, the debate about POs has raged for more than thirty years, although, in that time, the arguments themselves have remained static.150 Those who support and oppose reform are covering essentially the same territory now as they covered in the early 1990s.

One reason the debate has fizzled is entirely explicable. Reformers and critics sharply disagree about first principles and about the proper role of civil litigation. Does private civil litigation exist merely to solve parties' discrete disputes?151 If so, POs, which advance and perhaps streamline private dispute resolution, are unobjectionable. Or should private litigation advance broader, public-regarding aims?152 If so, POs subvert those aims and should be curtailed. That debate, which strikes right at the heart of what litigation does and is, won't be advanced by empirics.

Another big reason the PO debate has stalled is also fairly explicable. The debate is running aground on the shoals of warring but unresolved predictions about how a restriction on POs would ripple through the broader litigation landscape. Reformers and critics disagree concerning whether tighter limits on POs' entry would lead to greater efficiency and more disclosure or, instead, would backfire and have, on balance, the opposite effect.153

A third fundamental reason the debate has stalled, however, is harder to explain. The debate has sputtered because there is not only a dispute about the fundamental role of civil courts, or about what would happen if Rule 26(c) were fortified. There is also a dispute about what is happening in the courts today. As the above accounting demonstrates, those who support and oppose transparency reform efforts sharply and persistently disagree about the status quo. And it is very hard to know whether and how a system should be reformed without first understanding how that system currently operates. On this score, there are, and have long been, striking disagreements about basic facts that go to the heart of the PO debate. These include: (1) how often POs are issued, (2) whether POs have actually shielded public [*132] health hazards from scrutiny, (3) the prevalence of joint or stipulated (rather than opposed or contested) POs, and, arguably most importantly, (4) whether courts currently subject stipulated POs to any meaningful scrutiny.

Fueling this uncertainty is a striking lack of empirical research assessing these basic how, when, and why questions. Thus, as Professor Seth Katsuya Endo recently observed, "[n]either the academic literature nor the jurisprudence grapples much with the on-the-ground practice of stipulated protective orders."154 Indeed, only three

148 Miller, supra note 34, at 431 (noting that a focus on public access would divert courts "from their primary mission" of resolving disputes among litigants); Marcus, supra note 111, at 470 ("The primary purpose for which courts were created . . . is to decide cases according to the substantive law. The collateral effects of litigation should not be allowed to supplant this primary purpose."). In addition to all the above, some add a procedural critique: that any change to Rule 26 ought to come from the rulemaking process rather than from Congress. For a discussion, see Vangelisti, supra note 135, at 172-73.

149 Doré, supra note 83, at 289; cf. Erichson, supra note 59, at 360 ("A light standard of good cause for discovery confidentiality reduces the burden on the court . . . .").

150 See supra notes 90-97 and accompanying text (discussing this debate).

151 See supra notes 147-49 and accompanying text (advancing this conception).

152 See supra notes 128-29 and accompanying text (advancing this conception).

153 Compare supra notes 124-27 and accompanying text (contending that minimal restrictions on POs raise the cost and burden of litigation), with supra notes 141-46 and accompanying text (noting that efforts to cut down on POs will turn parties away from the courts).

Page 23 of 54

ARTICLE: SECRECY BY STIPULATION

studies address these matters.155 None completely answers the critical question of how Rule 26(c) is or, more often, is not appropriately applied. And each is limited by data hurdles that we are able to overcome.

To start, in 2020, Endo of the Seattle University School of Law published a piece in which he evaluated 100 proposed stipulated POs presented to district courts across the country in January 2018.156 Endo calculated how many of the proposed orders were granted, and, among granted orders, he analyzed whether the judge modified the proposed order before its entry.157 Ultimately, Endo found that only five of the 100 proposed POs in his dataset were denied.158 In studying the text of the ninety-five approved orders, Endo found only thirty-two that "described specific types of information to be protected or harms that would follow from public disclosure."159 Many of the others merely copied model orders or parroted the list of confidential information from Rule 26(c).160 Surprisingly too, Endo found fifteen of the ninety-five motions for POs that were ultimately granted contained a significant mistake of law including confusing the legal burden for obtaining a PO and for filing material under seal and that judges only rarely corrected this elementary mistake.161 And "in none of the cases did the docket reflect that any entity actually intervened to . . . [*133] challenge a confidentiality designation."162 Further, although not the core contribution of his paper, Endo's analysis implicitly estimates that parties jointly requested POs in, at most, 1.15 percent of all civil cases, or fewer than 4,000 cases per year.163

Endo's analysis offers a significant contribution and suggests that courts too readily rubber-stamp half-baked stipulated POs. But his limited sample size (of only 100 motions for POs) limits the force of his findings. And, as explained below, estimating the number of stipulated POs based on searching Bloomberg for a single month leads to a substantial undercount of their general incidence.164

More recently, in a student note, Chelsea Hanlock analyzed 595 cases brought against New York City police officers in the Eastern and Southern Districts of New York between 2014 and 2019.165 In her sample, 139 cases

154 See Endo, supra note 35, at 1275.

155 See generally id.; Hanlock, supra note 24; ELIZABETH C. WIGGINS, MELISSA J. PECHERSKI, & GEORGE CORT, PROTECTIVE ORDER ACTIVITY IN THREE FEDERAL JUDICIAL DISTRICTS: REPORT TO THE ADVISORY COMMITTEE ON CIVIL RULES (1996).

156 Endo, supra note 35, at 1249.

157 Id. at 1277-78.

158 Id. at 1277.

159 Id. at 1277, 1286.

160 Id. at 1286.

161 Id. at 1288-89. For how the two mechanisms differ, see supra notes 38-51 and accompanying text.

162 See Endo, supra note 35, at 1279.

163 Endo determined that there were, depending on the year, 2,142 to 7,127 annual mentions of stipulated POs in federal civil cases. Id. at 1273. If stipulated POs were mentioned only once per case in which they were requested, these figures would suggest a stipulated PO request rate of 0.8 percent to 2.3 percent. Id. But Endo reasoned that there are at least two stipulated PO mentions when the parties request a stipulated PO. Id. at 1276 n.161 ("[T]here should be, at minimum, an entry with its proposal and one for its entry or denial."). Accordingly, Endo's analysis implies a stipulated PO request rate of, at most, 0.4 percent to 1.15 percent, depending on the year.

164 See infra Part III.B and accompanying text.

165 Hanlock, supra note 24, at 1532-34.

Page 24 of 54

ARTICLE: SECRECY BY STIPULATION

had stipulated POs.166 Evaluating the text of proposed and granted stipulated POs and the thoroughness of judges' analysis, Hanlock, like Endo, found that denials were rare; she found a denied-in-full rate of less than 1 percent and a denied-in-part rate of 2.9 percent.167 Like Endo, she also found that most judges' reviews were cursory. In Hanlock's words, most judges "did not engage in a rigorous good cause analysis."168 For example, most orders justified confidentiality, at least in part, because "the City . . . would not produce requested documents" unless a PO was entered; only one submission made a particularized showing.169 Thus, Hanlock's study, like Endo's, supports the broader notion that judges rubber-stamp stipulated POs. But, given its small sample size, its findings are [*134] suggestive rather than firm and, although Hanlock zooms in on the important issue of police misconduct in New York, she cannot speak to other types of claims, defendants, or locations.

A third and final study, conducted for the Advisory Committee on Civil Rules, is bigger, but its data are also more than thirty years old.170 In this study, researcher Elizabeth Wiggins and coauthors searched through roughly 40,000 civil cases filed in three federal district courts from 1990 to 1992 looking for evidence of PO activity.171 The prevalence of civil cases with PO activity varied from 5 to 10 percent across the three districts and years studied.172 Of cases with PO activity, a strong majority of motions for POs were not stipulated; only 17 to 26 percent of cases with POs had motions for stipulated POs (with variation across districts).173 Crunching Wiggins et al.'s numbers yields the conclusion that at most 1.5 percent of federal cases involve motions for stipulated POs that is, stipulated POs are very rare.174 Wiggins et al. further found that courts denied 30 to 45 percent of motions for POs, although courts only rarely denied such motions when they were the product of party consent.175 Soon after this study, critics

166 Id. at 1538.

167 Id. at 1542.

168 Id. at 1540.

169 See id. at 1540-41 (listing additional rationales for granting POs that do not satisfy the good cause standard, including agreement between the parties).

170 WIGGINS ET AL., supra note 155, at 3 (evaluating a total of approximately 40,000 civil cases but dating from 1990 to 1992).

171 Id. at 1, 3.

172 Id. at 3.

173 Id. at 4-5. The parties initiated about 75 percent of those POs by motion. Id.

174 Their Table 1 reports the total number of cases studied for each district and year in the study, as well as the percentage of these cases that had PO activity. From these data, we calculated that 6 percent of all cases had PO activity. Id. at 3. Their Table 2 reports that, in the set of 601 cases they studied in more detail, there were 927 PO motions filed. Id. at 4. Numbers reported in their Table 3 indicate that, of the 927 motions, 207 had stipulated agreements by the parties. Id. In addition, Wiggins et al. report that in each district they studied, at least 26 percent of cases studied in more detail for their Table 2 had multiple instances of PO activity (for example, multiple motions). Id. at 3. This implies that, in the 601 cases studied in more detail per Wiggins et al.'s Table 2, there were at least fifty-four cases with multiple instances of stipulated PO activity (26 percent of tabulated instances of stipulated PO activity times 207 such instances). It then follows that Wiggins et al.'s Table 2 sample had, at most, 153 cases with any stipulated PO activity (207 instances of stipulated PO activity minus fifty-four repeat instances). This amounts to 25 percent of the 601 cases studied in more detail (because 153 divided by 601 is 25 percent). Multiplying 6 percent (the percentage of federal cases that Wiggins et al.'s results suggest had any PO activity) by 25 percent (the maximum share, among cases with any PO activity, in which there was stipulated PO activity) yields the conclusion that, at most, 1.5 percent of cases considered by Wiggins et al. had stipulated POs.

175 Id. at 6. Courts rejected, on the record, only 1 percent of the stipulated POs in the study. Id. at 4, 6. The researchers proffered an explanation for these infrequent denials: "[P]arties [may] discuss with the court whether a protective order is warranted and what provisions should be included before a formal agreement is presented." Id. at 6. Only as an "alternate explanation" did the researchers consider that "judges are reluctant to reject an agreement between opposing parties." Id.

Page 25 of 54

ARTICLE: SECRECY BY STIPULATION

of PO reform [*135] enthusiastically trumpeted its core findings that courts frequently deny POs and that parties rarely file joint motions for POs.176

Although the above efforts contribute to the debate, their limitations soften their empirical punch. The academic studies each evaluate only a small sample of POs in a sample restricted by court or year, gaining analytic granularity at the expense of generalizability. Similarly, the Wiggins et al. study covers only three courts and is, bluntly, old.177

In sum, for more than thirty years, the propriety of POs has been hotly debated. But the debate has stalled in large part because even basic questions concerning the entry of POs have defied resolution. Without these facts, the debate has devolved into warring rhetoric, punctuated by confident but unsupported claims concerning, among other things, the prevalence of POs, the prevalence of stipulated POs, and the rigor of judicial review.

Below, drawing on docket reports from over 2.2 million cases, we present robust empirical evidence to reset the empirical terrain. III. DATA: NEW EVIDENCE ON STIPULATED PROTECTIVE ORDERS

The above discussion shows that PO practice is frequently debated but rarely studied. In this Part, we apply sophisticated text-based [*136] machine learning tools to analyze an original dataset of more than 2 million federal dockets in civil cases filed from 2005 to 2012 to bridge persistent empirical gaps.178

Section A begins with a brief note on methodology. It discusses our dataset and outlines how we used machine learning and natural language processing tools to identify POs within federal dockets. The Appendix offers additional methodological detail. Those uninterested in any discussion of methodology may choose to skip directly to the meat of our analysis.

Sections B, C, and D present our results. In particular, Section B provides information on trends and prevalence. It reveals that the fraction of cases with joint motions for POs more than doubles previous estimates. By the study period's end in 2012, joint motions for POs were present in more than 8 percent of cases in which an answer was filed. Section C then considers judicial decision-making and estimates how often judges deny litigants' joint PO requests. Here, consistent with Endo and Hanlock, we find that judicial denials are rare. But, more than that, judge- level statistics reveal something astonishing: A majority of federal district court judges in the sample all of whom considered at least twenty-five stipulated motions for POs never rejected a stipulated PO during the period of

176 See, e.g., Doré, supra note 83, at 302 (stating that the Wiggins et al. "study does not support claims that federal district courts have perfunctorily acceded to a plethora of stipulated requests for discovery protective orders or that such orders create significant hazards to public health and safety"); Sunshine in Litigation Act of 2009: Hearing on H.R. 1508 Before the Subcomm. on Com. & Admin. L. of the H. Comm. on the Judiciary, 111th Cong. 59 (2009) (statement of Hon. Mark R. Kravitz for the Judicial Conference of the United States) (pointing to the Wiggins et al. study to support the notion that reform is not necessary because, as it is, "judges den[y] or modif[y] a substantial proportion of motions for protective orders"). Indeed, those opposed to reform continue to cite the Wiggins et al. study, despite its advanced age. E.g., LAWS. FOR CIV. JUST., COMMENT TO THE ADVISORY COMMITTEE ON CIVIL RULES 4 (2021) (citing the Wiggins study and declaring that "[t]he FJC's report also shows that the number of orders protecting . . . documents . . . is also small").

177 Also, potentially of note is a recent study involving orders to seal. In particular, Reuters recently analyzed Westlaw data from 3.2 million civil suits filed in federal courts between 2006 and 2016 and also conducted a more detailed review of the 115 largest "mass product-liability actions from the past 20 years." How We Did the Data Analysis, REUTERS, https://www.reuters.c om/investigates/special-report/usa-courts-secrecy-how [https://perma.cc/D4P7-DSHB]. The authors found that "judges allowed litigants to seal material in at least 65 percent of product-liability actions." Id. This study is useful but sheds little light on POs and only focuses on judicial decisionmaking in the largest products liability multidistrict litigations ("MDLs").

178 With respect to text-based tools, we extensively use natural language processing, the branch of machine learning that performs text analytics. Although no data extraction technique is perfect, our strategies yield high accuracy rates, particularly compared to the standard keyword approaches used in existing literature. For further details concerning both our data and methodology, see infra Appendix.

Page 26 of 54

ARTICLE: SECRECY BY STIPULATION

inquiry. This fact undermines the oft-repeated claim that federal judges individually scrutinize each PO that they enter and suggests that the entry of stipulated POs is not careful but, rather, may frequently be cavalier.179 Finally, in Section D we supplement our quantitative investigation with a qualitative assessment of 300 granted and 100 denied stipulated POs. Analyzing these orders, we find that the low rate of denials appears to reflect rubber- stamping on the ground, lending further support to reformers' concerns.

[*137] A. A Note About Methodology

Before we turn to our analyses, in this Section, we briefly describe our dataset and how we leveraged machine learning and natural language processing tools to sift through millions of docket sheets.

Our dataset includes docket-sheet information on case activity between January 1, 2005, and December 31, 2014, for more than 2.2 million cases filed between 2005 and 2012. These data were acquired from Thomson Reuters roughly a decade ago and have been used in several other studies.180 The dataset contains docket-level information on virtually all civil actions filed in federal district courts between January 1, 2005, and December 31, 2014. To our knowledge, it is the most up-to-date, and easily the most comprehensive, federal district court dataset available for research that currently exists.181

The sheer size of our docket dataset allowed us to overcome the challenges that other researchers have faced; our data are more recent and have nationwide geographic coverage.182 Our dataset is also the only one that contains virtually all civil cases filed during the period. That said, one limitation of our data is that we study only those cases filed between 2005 and 2012; although we are unaware of any work suggesting shifts in PO activity that have occurred over the past twelve years, our data would not capture such patterns if they did occur.

The dataset's enormity presents its own difficulty. When locating motions for contested or stipulated POs, we faced a massive needle-in-the-haystack [*138] problem; among nearly 80 million docket entries, only a small fraction addressed POs, creating the question of how to find what we were looking for. We overcame that challenge using a four-step procedure that included both simple text-processing and machine learning methods.

First, we retrieved the full text of all docket entries that contained the string "PROTECTIVE ORDER" the text was all capitalized in our underlying data. We refer to the full set of all such entries as the "matching set." We suspect that virtually all POs in our dataset contain this phrase. However, this query produced a high number of false positives

179 See supra notes 138-40 and accompanying text (repeating the claim that judges individually scrutinize each PO they enter).

180 See, e.g., Jonah B. Gelbach, Material Facts in the Debate over Twombly and Iqbal, 68 STAN. L. REV. 369, 393 (2016) [hereinafter Gelbach, Material Facts]; Jonah B. Gelbach, Rethinking Summary Judgment Empirics: The Life of the Parties, 162 U. PA. L. REV. 1663, 1676 (2014); Jonah B. Gelbach, Beyond Transsubstantivity, 26 N.Y.U. J. LEGIS. & PUB. POL'Y 909 (2024) [hereinafter Gelbach, Beyond Transsubstantivity]. For more details about the data, see infra Appendix.

181 Comprehensive district court dockets are notoriously hard to amass. This difficulty stems, in part, from the fact that most dockets sit behind a paywall. See David Freeman Engstrom & Jonah B. Gelbach, Legal Tech, Civil Procedure, and the Future of Adversarialism, 169 U. PA. L. REV. 1001, 1063 (recounting difficulties in creating large datasets of federal dockets because of PACER fees). See generally Jonah B. Gelbach, Free PACER, in LEGAL TECH AND THE FUTURE OF CIVIL JUSTICE (David Freeman Engstrom ed., 2023) (explaining similar).

182 As described elsewhere, the CM/ECF software used by district courts went into use at different times, and entries docketed in a district before that time may be less reliable. See generally Gelbach, Beyond Transsubstantivity, supra note 180. In our dataset, for instance, 4.08 percent of cases occur before the relevant district adopted the CM/ECF system. We treated these cases as missing data and used imputation methods to predict whether they have a stipulated PO or not. Such methods leverage nonmissing data to make educated guesses about missing data and are conventional in social science literature. See infra Appendix at 177-78.

Page 27 of 54

ARTICLE: SECRECY BY STIPULATION

docket entries in our matching set that included "PROTECTIVE ORDER" but did not directly involve the docketing or requesting of a PO.183

Second, we coded a random sample of 2,250 docket entries from the first step's matching set. A docket entry was relevant to this coding if it was either (1) created by one or more parties to docket a motion for a PO or (2) created to docket a court order resolving a motion for a PO. Among the relevant entries, we labeled each docket entry text string to designate whether it concerned a PO. We differentiated between two types of POs: (1) We categorized an order as a stipulated PO if either the motion seeking the PO or the order granting it had docket entry text calling it "stipulated," "consented," "agreed," or "joint," and (2) we coded all other POs as general POs. Thus, each of the 2,250 docket entries considered in this step was labeled as either (1) a stipulated PO, (2) a general PO, or (3) not a PO.

Third, using the labeled docket entries described in the second step just above, we trained a series of supervised machine learning models to predict which entries in our matching set involved stipulated POs or general POs. This approach is typical of supervised machine leaning: humans correctly labeling a random sample of text, after which a computer algorithm is applied to this labeled sample to "learn" which textual elements correspond to each possible label. This algorithm yields estimated parameter values that can be used to predict whether textual strings outside the training data represent PO requests. We used our supervised machine learning models to identify docket entries that involved motions for stipulated POs or general POs and judicial [*139] orders about stipulated POs and general POs.184 Compared to benchmarks in the law and data science literature, our models performed with high accuracy.185

Fourth, we used textual pattern matching to identify when judges denied requests for stipulated POs or general POs.186 This approach looked for sentences that contained a variant of the term "protective order" and some version of either "declines," "denied," or "unable to adopt." Taken together, these steps produced a comprehensive dataset of stipulated and nonstipulated motions for POs and the judicial grant or, far less frequently, denial of each.

A key advantage of this process is that it can be done using only the text of each case's docket entries, which allows large-scale quantitative analysis. To conduct our qualitative review of granted and denied stipulated POs, we needed to review the resolving orders filed in each case's docket. We took a random sample of 300 granted stipulated POs and 100 cases where the jointly filed motion was denied. We then cleaned the data,187 after which a

183 For example, a scheduling order could use the phrase "PROTECTIVE ORDER" to direct the parties that any requests for such orders are due by a certain date.

184 For details, including the preprocessing steps we took to clean the data, see infra Appendix at 173-76.

185 For instance, our classifier predicting the filing of a stipulated PO achieves 98 percent accuracy and has an F1 score = .94. When compared to the common F1 benchmark of around or above .8 referenced in existing legal research using similar methods, our method achieves a high level of accuracy. Cf. Brian D. Feinstein & Jennifer Nou, Strategic Subdelegation, 20 J. EMPIRICAL LEGAL STUD. 746, 762 n.15 (2023) (discussing classifier performance); Julian Nyarko, Stickiness and Incomplete Contracts, 88 U. CHI. L. REV. 1, 32 n.110 (2021) (discussing F1 = .7 as a benchmark). For more details about classifier performance, see infra Appendix at 173-76.

186 Our approach to pattern matching used regular expressions, which involves matching any combination of a chosen string of text and other text. For example, one regular expression is "DENIES.*PROTECTIVE ORDER". In this expression, ".*" means any number of characters of any type. So, for instance, "DENIES STIPULATED PROTECTIVE ORDER" would match the regular expression. For further explanation, see Appendix at 177. Regular expression searches are used commonly to search in text-as-data research. See, e.g., JUSTIN GRIMMER, MARGARET E. ROBERTS & BRANDON M. STEWART, TEXT AS DATA: A NEW FRAMEWORK FOR MACHINE LEARNING AND THE SOCIAL SCIENCES 178-83 (2022) (discussing their use in dictionary methods).

187 We excluded, for example, HIPAA POs which are granted pursuant to a statutory scheme. See 45 C.F.R. § 164.512(e) (2024). We also excluded dockets that were outside of our observation period (for example, orders entered before 2005), those that did not have an attached order, or in the rare circumstance, those that were false positives (for example, nonstipulated).

Page 28 of 54

ARTICLE: SECRECY BY STIPULATION

team of trained researchers retrieved the text of the underlying district court orders from PACER, manually reviewed each order against a set of agreed-upon criteria, and tracked results in a shared spreadsheet. After data collection, a trained researcher performed spot checks on 20 percent of the sample sixty [*140] grants and twenty denials to ensure orders were tagged consistently. Finally, for denied motions, we manually searched docket reports for other references to POs to determine whether a PO was ever entered in the case.

As a final methodological note, we considered but were able to mostly dismiss two phenomena that would undermine our conclusions. First, we assessed whether our results are likely driven by the use of model POs created by judicial districts. If so, one might think our results would capture district-level, rather than case- and judge-level, features of stipulated PO practice.188 But we found little evidence to support such a proposition.189 Second, we considered the possibility that judicial analysis related to the good cause standard might be present in judicial opinions filed alongside the granted orders themselves, because our main empirical approach does not address such opinions. To investigate that possibility, we searched for the word "opinion" in docket entries that mentioned stipulated POs. Ultimately, we found the share of entries that included this word was only a small fraction of 1 percent. Given all this, we conclude that neither the presence of model orders nor the possibility of judicial analysis in separate opinions is a plausible threat to our conclusions.

We now present the novel descriptive insights generated by this massive empirical effort.

[*141] B. Motions for Stipulated Protective Orders: Time Trends and Prevalence

We first consider a key, but heretofore unanswered question: Are motions for stipulated POs a common feature of contemporary federal litigation? As discussed in Part II, no prior study answers this question using comprehensive data. With some back-of-the-envelope calculations, Endo estimates that stipulated POs are featured in roughly 1 percent of federal civil litigation190 what would amount to fewer than 4,000 federal cases a year. The Wiggins et al. research, performed under the auspices of the Federal Judicial Center and often touted by critics of reform, offers a similar estimate.191 Ultimately, as discussed below and shown in Figures 1, 2, and 3, we find that joint motions for POs though relatively uncommon overall are considerably more common than these earlier studies indicate.

188 Some district courts publish model POs for parties appearing before the court. See, e.g., Model Stipulated Protective Order at 6 (N.D. Cal.), https://www.cand.uscourts.gov/wp-content/uploads/forms/model-protective- orders/CAND_StandardProtOrd.Feb2022.pdf [https://perma.cc/5UP8-VEN2]; see also infra Section III.D (surveying model POs).

189 This question is difficult to test systematically because model orders might not describe themselves as such, and districts' model orders, where they exist, might have changed over our study period. We used two parallel approaches to test model orders' prevalence. First, among the roughly 80 percent of all orders granting stipulated PO requests for which the order was text-searchable, we searched for the word "model". None of these orders used that word in connection with a model PO, although a handful of cases used the phrase "standing protective order" or "standing order" in a context that suggested they were model orders. Thus, our analysis of text-searchable orders showed little reason to think model orders are prevalent enough to play an important role in our results. Second, we selected nine district-year combinations that appeared in our set of 300 granted stipulated POs there were two to four orders in each of the nine district combinations. We looked for indicia of model orders, such as identical terminology, consistent formatting, or similar internal structure, but we did not find such evidence. Although it is difficult to draw definitive conclusions given the nature of the orders at issue, the results of these investigations strongly suggest that model orders are unlikely to constitute a substantial portion of the POs in our dataset.

190 See supra note 163 (explaining that Endo's research suggests that stipulated POs were present in at most 1.15 percent of civil cases in federal courts).

191 See supra notes 170-74 (explaining that the Wiggins et al. research suggests that stipulated POs were present in at most 1.5 percent of civil cases).

Page 29 of 54

ARTICLE: SECRECY BY STIPULATION

Figure 1 plots our estimates for the number of POs requested within two years of filing, by year.192 Figure 1 is notable in three respects. First, our results suggest that there are an average of about 9,000 jointly filed PO requests in the federal civil courts per year which translates to 2.8 percent of all civil cases over our full study period. That figure may seem small, but it is nearly double what others have previously suggested.193 Equally notable, the trend lines show a steady increase over time. For cases filed in 2005, we estimate that [*142] roughly 8,500 stipulated motions for POs were filed. By 2012, this number had grown to nearly 12,500.194 Overall, this represents a roughly 50 percent increase during the study period.

Figure 1. Total Number of Requests for POs by Filing Year of Case (2005-12)195

192 We limit all our analyses to two years after the filing date to account for the fact that our access to docket entry text extends only through December 31, 2014. For cases that terminate by that date, we observe all docketed activity. But for all other cases, we see only whatever docket activity happened before the end of 2014. To understand the problem, take, for example, a case filed on January 1, 2014. For that case, we observe exactly one year of activity. So, if a stipulated PO were requested in the case's second year, we would miss it. Comparing this 2014 case with only one year of docket activity to cases filed in earlier years would introduce measurement error due to the difference in the observable portions of the two cases' life cycles. Statisticians call this problem "right-censoring." To correct for this source of bias, we analyze only POs requested within two years of filing and, therefore, limit our analysis window to cases filed between calendar years 2005 and 2012. By doing so, we can make an "apples to apples" comparison across the years. Finally, all of our plots discussing "requests" count only those instances in which judges, rather than parties, create a docket entry relating to a determination about parties' joint request for a PO. This measure is methodologically conservative because it excludes requests that are mooted due to case termination or for other reasons, such as settlement.

193 See supra notes 190-91 and accompanying text (discussing prior estimates that parties request stipulated POs in 1.15 to 1.5 percent of federal civil cases).

194 Although it's not the core contribution of this Article, we also find that general nonstipulated PO activity increased during this period.

195 Figure 1 counts requests, not cases with requests. For example, if two requests are made in the same case, then each is counted in Figure 1.

Page 30 of 54

ARTICLE: SECRECY BY STIPULATION

One explanation for this observed uptick in joint Rule 26(c) motions is a possible uptick in the number of cases filed per year. If more cases were filed per year during this period, we would expect the number of joint Rule 26(c) motions to increase in rough lockstep. To a large extent, that is what we find. Figure 2 plots the percentage of civil cases with a joint motion for a PO filed within two years of the case's initiation.196 The percentage bounces around over time, with a slight upward trend over the seven-year period between the filing of 2005- vintage and 2012- vintage cases in the percentage of cases with a joint [*143] motion for a PO. At most, this evidence is weakly suggestive of a slight increase in the overall prevalence of stipulated POs.197 Still, even the lowest plotted percentage in Figure 2 far exceeds previous estimates.198

Figure 2. Percentage of Civil Cases with at Least One Stipulated PO Requested by Filing Year of Case (2005-12)

196 We note that the unit of analysis in Figure 2 is the case, rather than the order. Thus, case A with one stipulated PO and case B with two stipulated POs both add one to the numerator of the data plot in Figure 2.

197 The slope of a trend line fit using ordinary least squares is 0.0008, reflecting that the fitted value of the stipulated PO percentage increased roughly half a percentage point (0.0008 times 7 is 0.0056 or 0.56 percentage points). The trend line is not very precisely estimated, though, which suggests that the evidence of an increase in the rate of stipulated POs during our study period is relatively weak.

198 See supra notes 190-91 and accompanying text (discussing prior estimates that parties request stipulated POs in 1.15 to 1.5 percent of federal civil cases).

Page 31 of 54

ARTICLE: SECRECY BY STIPULATION

Although the estimates offered thus far suggest that motions for stipulated POs are more prevalent in contemporary litigation than previously understood, our findings still might not fully capture the force of stipulated POs, nor the key trends. Why? The figures above look at all civil cases. But many civil cases in federal court by some estimates, one- third or more never reach discovery.199 And discovery, of course, is the critical stage of litigation where concerns about POs kick in.

[*144] To get a sense of how prevalent stipulated POs are in cases that reach the discovery phase that is, cases where POs possibly matter Figure 3 plots the percentage of cases with a joint Rule 26(c) motion (within two years of filing) in cases where a defendant has answered a plaintiff's complaint. To be sure, discovery can start before the defendant answers the complaint. But in most cases, discovery battles do not commence until after the answer is filed.200

Figure 3 indicates, not surprisingly, that joint motions for POs are even more common in cases where a defendant answers the plaintiff's complaint. And the share is growing over the study period. In 2005, 5.5 percent of such "post- answer" cases featured a joint motion for a PO. By 2012, that figure grew to 8.5 percent. This growth represents over a 60 percent increase and several times the overall rate among all civil cases.

199 See Jay Tidmarsh, Opting Out of Discovery, 71 VAND. L. REV. 1801, 1803 (2018) ("In a third or more of federal cases, no discovery occurs . . . .").

200 See Alexandra D. Lahav, Procedural Design, 71 VAND. L. REV. 821, 872 (2018) (describing the "textbook order of civil procedure" as proceeding sequentially from "filing, [to] motion to dismiss or answer, [to] discovery"); see also FED. R. CIV. P. 26(d)(1) (clarifying that "[a] party may not seek discovery from any source before the parties have conferred as required by Rule 26(f)," subject to certain requirements).

Page 32 of 54

ARTICLE: SECRECY BY STIPULATION

Figure 3. Percentage of Cases with a Stipulated PO Request, Among Civil Cases with an Answered Complaint by Filing Year of Case (2005-12)

[*145] Finally, Figure 4 shows the proportion of POs requested that were jointly filed. These data show stipulated PO requests, once thought to account for 17 to 26 percent of motions for POs,201 account for a much larger share. On average, stipulated POs accounted for a whopping 45 percent of all PO motions filed per year a two to three times larger proportion than previously thought.

Figure 4. Percentage of Requested POs that Involved a Stipulated Request, by Filing Year of Case (2005-12)

201 See supra note 173.

Page 33 of 54

ARTICLE: SECRECY BY STIPULATION

C. Rule 26(c) Motions: Grants and Denials

We now assess how judges respond to these Rule 26(c) motions. As noted, reformers have long argued that stipulated POs are granted, pro forma, without adequate judicial scrutiny.202 Critics, meanwhile, have insisted the opposite. As the Judicial Conference of the United States put it when opposing reform efforts: When motions for POs are filed, "courts review such motions carefully and often deny or modify them to grant only the protection needed, recognizing the importance [*146] of public access to court filings."203 Our data permit us to adjudicate this debate.

Figure 5 reports the average denial rate for stipulated PO requests. This plot, and all analyses that follow, take a conservative view. Figure 5 classifies a motion for a stipulated PO as "denied" even if: (1) the motion is partly denied (i.e., even if the motion is granted in part), or (2) the judge denies the parties' motion not on the merits, but rather as moot. In so doing, this plot errs on the side of reporting PO denials.

Particularly given our generous definition of "denials," the primary fact that stands out in Figure 5 is just how low the denial rate for stipulated POs is. For the entire sample, judges deny, on average, only 4.1 percent of stipulated PO requests. In other words, roughly 96 percent of the time, when the parties file a joint motion for a PO, it is granted.

Figure 5. Share of Stipulated PO Requests Denied Per Year (2005-12)

202 See supra notes 118-21 and accompanying text.

203 Sunshine in Litigation Act of 2008: Hearings on H.R. 5884 Before the H. Comm. on the Judiciary, Subcomm. on Com. & Admin. L., 110th Cong. 74 (2008) (statement of Hon. Mark R. Kravitz for the Jud. Conf. of the United States). For similar assurances, see supra notes 138-40 and accompanying text.

Page 34 of 54

ARTICLE: SECRECY BY STIPULATION

[*147] What case types feature higher and lower denial rates? Parsing PO denials by case type, Figure 6 reveals modest variation. The highest denial rates (north of 6 percent) are in forfeiture cases.204 IP cases where many believe POs are justified have a slightly higher than average denial rate.205 Intriguingly, civil rights cases and tort cases produce some of the lower denial rates (on the order of 3.4 and 3.6 percent, respectively). This result is notable because both cases can involve facts and disputes with high levels of public interest.

Figure 6. Percent of Stipulated POs Denied by Case Type

204 Because forfeiture cases make up a small share of the set of cases with a stipulated PO, it is possible that their especially high denial rate may be attributable to statistical variation essentially noise.

205 These analyses speak only to denial rates for stipulated POs. They are not necessarily a proxy for a judge's willingness to permit secrecy writ large.

Page 35 of 54

ARTICLE: SECRECY BY STIPULATION

Next, we evaluate judicial behavior, and here we find something truly remarkable about the infrequency with which judges deny the parties' joint motions for POs. In so doing, we want to zero in on individual judge-level behavior, but we also need to avoid drawing conclusions simply because some judges have considered very few jointly filed motions for POs. We therefore restrict our focus only to [*148] judges with substantial Rule 26(c) activity, which we define as those who ruled upon at least twenty-five such motions during the period of study.206 Here's the remarkable finding: Of the 727 district and magistrate judges included in this analysis, 392 or 54 percent never once denied a stipulated PO request.207 Figure 7 makes this point graphically by plotting the number of judges whose judge-level stipulated PO denial rates fall in various bins; the first bin, capturing denial rates between 0 and 1.25 percent, is overwhelmingly more frequent than others.

Figure 7. Distribution of Judge-Level Stipulated PO Denial Rate208

206 Because both district court judges and magistrate judges render decisions about stipulated POs, we could not simply assign a given order to the district court judge presiding over the entire case. Instead, we built a nuanced regular expression to extract the judge's name from the individual docket entry. Evaluating our extraction method via random samples of stipulated POs, we found that the method correctly identified 98 to 99 percent of judges issuing the stipulated POs.

207 Also remarkable is that district court and magistrate judges do not appear to stay on circuit-level scripts regarding stipulated PO adjudication. Trial court opinions display sharp and seemingly inexplicable intracircuit variation, and the opinions do not resemble what one might expect if trial courts were following circuit court precedent with fidelity. That is, some circuit court precedent is strict when it comes to POs, and some circuit court precedent is lax when it comes to POs. But trial judges in "strict" circuits do not appear to scrutinize POs more carefully than do trial judges in "lax" circuits.

208 The bin size for the x-axis is 1.25 percent. The analysis is limited to judges who have decided at least twenty-five stipulated POs.

Page 36 of 54

ARTICLE: SECRECY BY STIPULATION

[*149] The fact that over half of the judges in this analysis never denied parties' joint requests for a PO is remarkable. Opponents of PO reform have long argued that trial judges are in the best position to exercise ground- level discretion in determining whether a PO is warranted and, if so, what shape that PO should take. They have long assured policymakers that, on a day-to-day basis, district court judges exercise this discretion with care and fidelity.209 To have so many judges with a 0 percent denial rate raises questions about that assertion and implies that searching case-by-case scrutiny is not the norm.210

209 See supra notes 138-40 and accompanying text.

210 Although grant rates in isolation are probative of rubber-stamping, they are not necessarily proof of it. Caution is warranted, in part, because strategic litigation effects make it notoriously difficult for researchers to map the grant rate to how judges apply the law. See, e.g., Gelbach, Material Facts, supra note 180, at 389-93. Further, as Wiggins et al. note, it is possible that part of what might drive the results is that "parties [may] discuss with the court whether a protective order is warranted and what provisions should be included before a formal agreement is presented." WIGGINS ET AL., supra note 155, at 6. If judges and parties are regularly engaged in this kind of informal, invisible, undocketed exchange, then our review of the docket would show rubber- stamping where there has been more consideration.

Although we cannot rule out either effect, there are two reasons we are skeptical that informal channels, the latter effect, explain our findings. First, our detailed qualitative review did not surface substantial evidence of such iterative collaboration. See infra Section III.D.

Second, if informal discussions were an important factor in our results, a different denial rate would be expected in cases where there was only one docketed entry reflecting a stipulated PO request than in cases with multiple such docketed entries. Our reasoning is that, after parties heard from chambers that their initial joint PO request did not pass muster, they would be expected to file a superseding, second request that would pass muster. So, if the informal discussion hypothesis is importantly correct, we ought to see an especially low denial rate in cases where we identify two docketed motions for stipulated POs and one docketed order adjudicating those requests. In such cases, we found a denial rate of 7.7 percent higher, not lower, than the

Page 37 of 54

ARTICLE: SECRECY BY STIPULATION

D. Qualitative Assessments

To fortify our analysis, we supplemented our quantitative assessment with a qualitative evaluation of 300 granted and 100 denied joint PO motions randomly drawn from our docket dataset. Although these 400 cases represent a relatively small sample of orders, the sample is large enough to give us a helpful peek at what, if any, standards judges actually apply when ruling on requests for stipulated POs. The sample, from a range of district courts and civil case types, contains stipulated PO rulings entered between 2005 and 2014 in cases [*150] filed between 2005 and 2012.211 Earlier works, such as Endo's study, inspire this review.212

Table 1. Count of Tagged Stipulated PO Grants by Year Filed and Circuit213

Count by Year Filed

Year PO Count Percent 2005 6 2.0% 2006 23 7.7% 2007 19 6.3% 2008 28 9.3% 2009 25 8.3% 2010 32 10.7% 2011 30 10.0% 2012 44 14.7% 2013 63 21.0% 2014 30 10.0%

Count by Circuit

Circuit PO Count Percent First 6 2.0% Second 30 10.0% Third 16 5.3% Fourth 7 2.3% Fifth 50 16.7% Sixth 33 11.0% Seventh 34 11.3% Eighth 15 5.0% Ninth 60 20.0% Tenth 28 9.3% Eleventh 21 7.0% D.C. 0 0.0%

Table 2. Count of Tagged Stipulated PO Denials by Year Filed and Circuit

Count by Year Filed

Year PO Percent

overall average. Of course, there might be differences in cases with one versus multiple joint PO requests, so this statistic should not be pushed too far. Nevertheless, this pattern is inconsistent with the simple informal discussion story detailed above.

211 One other prefatory note: Eighteen orders in our random sample (6 percent of the draw) happened to be identical orders entered on the same day and by the same judge as part of an MDL. Given our commitment to the random draw, we included all eighteen orders in our analysis.

212 See generally Endo, supra note 35 (conducting a qualitative review of 100 granted stipulated POs).

213 The random pull did not include any cases from the D.C. Circuit because the D.C. District Court has a low caseload relative to other districts and adjudicates very few cases involving stipulated POs.

Page 38 of 54

ARTICLE: SECRECY BY STIPULATION

Coun t

2005 1 1.0% 2006 10 10.0% 2007 10 10.0% 2008 11 11.0% 2009 15 15.0% 2010 8 8.0% 2011 15 15.0% 2012 15 15.0% 2013 5 5.0% 2014 10 10.0%

Count by Circuit

Circuit P O

Count Percent

First 0 0.0% Second 2 2.0% Third 0 0.0% Fourth 4 4.0% Fifth 1 1.0% Sixth 2 2.0% Seventh 1

2 12.0%

Eighth 2 2.0% Ninth 5

7 57.0%

Tenth 1 3

13.0%

Eleventh 7 7.0% D.C. 0 0.0%

[*151] 1. Qualitative Review of 300 Granted Protective Orders. Evaluating each of these 300 stipulated POs, we first considered how the judge applied Rule 26(c)'s good cause standard. In so doing, we assessed whether the judge actually included the term "good cause" in the granted order. In other words, did the court's order make any mention of the formal requirement for granting a PO?214 A mere 32 percent of stipulated POs referred to "good cause" explicitly.215 Roughly two-thirds of the orders failed to so much as mention the relevant legal standard.216

However, these figures do not on their face establish that judges abandoned their legal duties. For one, recall that two circuits explicitly permit judges to grant stipulated POs without finding good cause.217 In addition, when a judge signs a stipulated PO that makes a particularized showing of good cause, the judge almost by definition [*152] satisfies Rule 26(c). But the reverse is not necessarily true. The absence of discussion of the good cause standard, on the order itself, is merely probative rather than conclusive as to whether judges followed Rule 26(c). It remains possible that the judge carefully engaged with the Rule 26(c) standard in a way that this empirical analysis fails to

214 FED. R. CIV. P. 26(c). 215 See, e.g., Stipulation and Ord. Protecting Confidential Information; Granting Motion for Protective Order at 1, United States v. Arg Co., Civil Action No. 10-cv-311 (N.D. Ind. Apr. 1, 2013), ECF No. 47 ("In view of this stipulation, the Court finds that good cause exists . . . ."). 216 See, e.g., Stipulated PO at 1, Whitfield v. PEP Boys - Manny, Moe, & Jack, No. 13-cv- 11070 (E.D. Mich. Oct. 23, 2013), ECF No. 23 (offering no justification for issuance of PO). Note, however, that a small number of orders (5.7 percent) mentioned possible injury stemming from disclosure without explicitly referencing "good cause." See, e.g., Order Granting Motion to Enter Proposed Agreed PO, at 1, Humphrey v. Burgos, No. 2:06-cv-00045-APR (N.D. Ind. July 12, 2007), ECF No. 58 (specifying that the party "contends that said documents . . . contain sensitive and proprietary information relating to its operations as a local government" and, if distributed, would disrupt "its ability to effectively carry out its police powers and activities"). 217 See supra Part I.C. (describing the relevant case law).

Page 39 of 54

ARTICLE: SECRECY BY STIPULATION

discern. Perhaps, for example, the judge did so in a hearing, where no transcript was ever made or by informal communication via clerks. Or parties might have justified their PO in other briefing to the court, which would not show up in the requested or approved PO. Yet, with that caveat, it is highly curious that, in the three circuits that require judges to make a good cause finding for stipulated POs, only 24 percent of such PO grants so much as mentioned good cause.218 This proportion is even lower than for the full sample.

Next, we explored whether the Rule 26(c) order justified the grant based on individualized facts or, in contrast, merely by recycling boilerplate language. Recall that, in at least some circuits, a "district court abuses its discretion where it makes neither factual findings nor legal arguments supporting the need for" a stipulated PO.219

Here, out of 300 stipulated POs, only 17 percent provided a Rule 26(c) good cause determination tailored to the facts of the case.220 The remaining orders did not mention good cause, mentioned good cause but offered no analysis to support a good cause finding, or included boilerplate language.221 This finding is consistent with others' [*153] research.222 And, as above, an even lower proportion of granted stipulated POs 12 percent included an individualized good cause justification in the three circuits that explicitly require good cause findings when granting a jointly filed motion for a PO.223

Third, we considered the extent to which courts modified the POs proposed by the parties. Litigants jointly seeking a PO generally submit a proposed order. Upon receipt, a judge may sign a proposed order as written, manually modify the proposal, or draft and sign the judge's own version. For this study's purposes, these findings are relevant because if a judge modifies a PO, there is at least some clear evidence that the judge scrutinized the order.

In assessing whether judges modify proposed POs, we used a very generous definition of "modification." To be sure, we did not credit as modifications any situations like the one displayed in Figure 8, in which judges simply: (1) added prefatory language noting that the court was granting the proposed order, (2) removed the word "proposed" from the order, (3) dated the document, or (4) signed the document. But we counted any other change to the text of a proposed PO, even minor changes to grammar, as a modification.224

Still, even under this capacious definition, courts modified the text of less than 22 percent of all proposed orders.225 Judges did not modify [*154] any element of the party's proposed stipulated PO in at least 78 percent of cases. In many cases, the court merely crossed out the word "proposed," signed, and dated the order proposed by litigants.

218 Id. This figure is based on the analysis of eighty-three stipulated PO grants. An additional 9.6 percent of the orders had a particularized good cause statement but did not include the term "good cause." 219 In re Nat'l Prescription Opiate Litig., 927 F.3d 919, 929 (6th Cir. 2019) (quotation marks, citation, and alteration omitted). 220 For example, one order noted that a PO was justified because litigants would be discovering employees' personnel files and that discovery of those files would constitute an "unnecessary invasion of their privacy interests." Agreed PO/As Modified at 1, Socite v. Steak 'N Shake Operations, Inc., No. 1:06-cv-03888 (N.D. Ill. Jan. 23, 2007), ECF No. 41. Another order noted that parties sought marketing and financial strategy documents, which derive value from "not being generally known to other parties," and would cause damage to the parties if revealed. Amended Stipulated PO at 2-3, Sparknet Commc'ns, L.P. v. Bonneville Int'l Corp., No. 1:05-cv-02677 (N.D. Ill. July 5, 2005), ECF No. 95. 221 For example, one order noted that "[d]isclosure and discovery in this action are likely to involve production of confidential, proprietary, or private information for which special protection from public disclosure and from use for any purpose other than prosecuting this litigation would be warranted." Ord. by Magistrate Judge Granting Motion for Stipulated PO (as Modified) at 1, Navarrette v. TD Banknorth, N.A., No. 5:07-cv- 02767-JW (N.D. Cal. Dec. 12, 2007), ECF No. 19. And at least one order, after offering a boilerplate good cause statement, emphasized that "[t]he Parties believe that good cause exists." Order Pursuant to Stipulation: Discovery PO at 1-2, First Serv. Networks, Inc. v. First Serv. Maint. Grp., Inc., No. 2:11-cv-01897- LOA (D. Ariz. May 21, 2012), ECF No. 28 (emphasis added). This example hearkens back to the analogy comparing stipulated POs to proposed class action settlements that have support from all named parties wherein both the Seventh Circuit and Miller agree that, as in the Rule 23(e) context, when it comes to Rule 26(c), party consent does not suffice. See supra notes 71, 81, and accompanying text. 222 Endo, for example, found that 34 percent of sampled orders that granted a jointly filed motion for a PO described either "specific types of information to be protected or harms that would follow from public disclosure"; meanwhile, the remaining 66 percent used "generic language." Endo, supra note 35, at 1286. Our results differ from Endo's, likely because we counted only the second of Endo's two categories of orders those that explained what harms would befall litigants if documents were made public. For similar results, see Hanlock, supra note 24, at 1540 (noting that orders in her study "recognized good cause as the standard" but generally "did not engage in a rigorous good cause analysis"). 223 In other words, 12 percent of the sampled grants from the Third, Sixth, and Seventh Circuits had an individualized justification. 224 To determine if a judge modified a PO, we looked for indicia of modification for example, handwritten notes or text in a different font. Where the order was ambiguous on its face, we downloaded the original PO request from PACER and compared the two orders side by side. 225 More specifically, we found evidence of judicial modification in 14.7 percent of orders. But we could not tag a further 6.7 percent of all orders (often because the party's initial proposal was not available on PACER), meaning that the modification rate could be as high as 21.3 percent.

Page 40 of 54

ARTICLE: SECRECY BY STIPULATION

And, where courts did modify a PO, they often did so to clarify an unrelated issue the standard for sealing court records.226

Figure 8. Example of Unmodified PO227

To recap, our qualitative assessment of 300 orders reveals that most stipulated POs fail to mention Rule 26(c)'s "good cause" standard. Fewer offer an individualized assessment of the facts justifying the judge's (implicit or explicit) finding that the standard has been satisfied. And, generally, as exemplified by the proposed-cumactual order depicted in Figure 8, when judges grant the parties' joint motion, the judge merely strikes out the word "proposed" and signs their name.

Finally, we surveyed model POs promulgated by district courts and, in so doing, found yet more evidence of rote and relaxed Rule 26(c) review. Eighteen of the ninety-four federal district courts have adopted district-wide model POs.228 Of those eighteen model orders, only two facially require a judicial finding of good cause based [*155] on the circumstances of the case.229 Two others imply that the court found "good cause" in entering the order but do not seem to contemplate particularized justifications.230 The remaining fourteen do not even purport to make a good cause determination and, indeed, seven do not mention the standard at all.231 Once again, these results are

226 See supra notes 44-51 and accompanying text. In the sample, 59 percent of all modified POs were modified to, at least in part, clarify the standards for sealing documents. 227 The judge made no other changes to the proposed order. Amended Stipulated PO, Sparknet Commc'ns, L.P. v. Bonneville Int'l Corp., No. 1:05-cv-02677 (N.D. Ill. July 5, 2005), ECF No. 95. At least 79 percent of all grants were functionally identical to Figure 8. 228 Those courts are the Northern District of California, the Southern District of California, the District of Idaho, the Southern District of Indiana, the Northern District of Iowa, the District of Kansas, the District of Minnesota, the District of Nebraska, the District of New Hampshire, the Northern District of Ohio, the Southern District of Ohio, the Northern District of Oklahoma, the District of Oregon, the Northern District of Texas, the District of Utah, the Western District of Washington, the Northern District of West Virginia, and the Southern District of West Virginia. Citations follow in the ensuing footnotes. 229 Both model orders ask litigants to explain why the litigation calls for a PO. U.S. Dist. Ct., Dist. Kan., Protective Order 3, https://www.ksd.uscourts.gov/civil-forms [https://perma.cc/HJZ9-ASUP]; U.S. Dist. Ct., Dist. Or., Model Two-Tier Protective Order 1 (2018), https://ord.uscourts. gov/index.php/164-attorneys/forms [https://perma.cc/H78J-3Z5R]. 230 U.S. Dist. Ct., Dist. Idaho., Model Protective Order 8, https://id.uscourts.gov/district/forms_fees_rules/Civil_Forms.cfm [https://perma.cc/UM48-EY34] (to locate, scroll down to "Litigation and Discovery Plan"); U.S. Dist. Ct., Dist. Utah, Standard Protective Order for Cases Filed Before December 1, 2023, at 1, https://www.utd.uscourts.gov/usdc-forms [https://perma.cc/9Z5E-WHR2]. 231 See, e.g., U.S. Dist. Ct., S. Dist. Ind., Uniform Protective Order (Apr. 18, 2022), https://www.insd.uscourts.gov/forms/uniform-protective- order [https://perma.cc/S4XG-GKWH] (including the phrase "good cause" in the instructions accompanying the model order, but not including the phrase in the order itself); U.S. Dist. Ct., N. Dist. Iowa, Stipulated Protective Order, https://www.iand.uscourts.gov/forms-0 [https://perma.cc/H8WS-WR3M] (to locate, scroll down to "Civil"); U.S. Dist. Ct., Dist. Minn., Stipulation for Protective Order Form, https://www.mnd.uscourts.gov/sites/mnd/files/forms/Stipulation-for-Protective-Order-Form.pdf [https://perma.cc/4D4P-3AR8]; U.S. Dist. Ct., Dist. Neb., Protective Order: No AEO, no HIPPA, https://www.ned.uscourts.gov/internetDocs/forms/Protective%20Order no%20AEO%20no%20HIPAA.pdf [https://perma.cc/TX7 Z-C2XS]; U.S. Dist. Ct., S. Dist. Ohio, One-Tier Protective Order, https://www.cand.uscourts.gov/wp-content/uploads/forms/model-protective-orders/CAND_StandardProtOrd.Feb2022.pdf [htt ps://perma.cc/K52P-FQ89]; U.S. Dist. Ct., W. Dist. Wash., Model Stipulated Protective Order, https://www.wawd.uscourts.gov/sites/wawd/files/ModelStipulatedProtectiveOrder%20CLEAN_2.1.2 3.pdf [https://perma.cc/NV5G-F5PQ]; U.S. Dist. Ct., S. Dist. W. Va., Agreed Protective Order, htt ps://www.wvsd.uscourts.gov/forms/agreed-protective-order [https://perma.cc/RY3A- ZRM8]. Five use the term "good cause" only when discussing other procedures, including procedures for subsequent challenges to individual confidentiality designations. U.S. Dist. Ct., N. Dist. Cal., Model Stipulated Protective Order 6, https://www.cand.uscourts.gov/wp- content/uploads/forms/model-protective-orders/CAND_StandardProtOrd.Feb2022.pdf [https://perma.cc/8BFS-8BXG]; U.S. Dist. Ct., S. Dist. Cal., Model Protective Order 8, https://www.casd.uscourts.gov/_assets/pdf/forms/Model%20Protective%20Order.pdf [https://perma.cc/R7U2-64H6]; U.S. Dist. Ct., N. Dist. Okla., CV-29 Form (Confidential Protective Order) 8, https://www.oknd.uscourts.gov/protective- orders[https://perma.cc/W7YY-NMLN]; U.S. Dist. Ct., N. Dist. Tex., Protective Order 4, 9, https://www.txnd.uscourts.gov/sites/default/files/forms/MO62A-ProtectiveOrder.pdf [https://perma.cc/JJW4-WZJA]; U.S. Dist. Ct., N. Dist. W. Va., Sample Protective Order 3, 6, https://www.wvnd.uscourts.gov/sites/wvnd/files/Proposed%20Draft%20Protective%20Order%20For%20Form.pdf [https://perma.cc/9E5T-DBED]. Two mention the good cause standard for entering a PO but do not make a good cause determination. U.S. Dist. Ct., Dist. N.H., Civil Form 5: Protective Order Form 1, 9, https://www.nhd.uscourts.gov/pdf/Civil%20Form%205.pdf [https://perma.cc/VR9C- FBJT] (stating that the order applies only to "materials which may be subject to restrictions on disclosure for good cause"); see also U.S. Dist.

Page 41 of 54

ARTICLE: SECRECY BY STIPULATION

not dispositive. But they nonetheless bolster this Article's findings [*156] and suggest that district courts do not routinely evaluate parties' good cause justifications.

2. Qualitative Review of 100 Denied Protective Orders. To round out our empirical assessment, we evaluated 100 denials of joint motions for POs, roughly 3.5 percent of all such denials in our study.232 Of these, ninety-one orders fully denied the parties' joint motion, whereas nine denied it only in part.233 Notably, despite our random sampling, fourty-four such denials were from the Central District of California, a product of the district's large caseload and comparatively high denial rate. Just three judges, district and magistrate, from the Central District of California accounted for 23 percent of the PO denials in our entire random sample.

So, why did courts reject stipulated POs? We found that there were several reasons and, in many instances, judges' denials highlighted numerous deficiencies. As Figure 9 demonstrates, over half 54 percent of all stipulated PO merits-based denials were traceable to the fact that the parties tried to smuggle in provisions that committed the court to automatically seal court filings. As explained previously, the sealing determination and the entry of a PO are often confused, but the two concepts should be kept distinct.234 A similar share 50 percent of rejected stipulated POs centered on good cause. Often, parties either made no good cause statement at all, or made it with insufficient particularity. The last large category of rejections, accounting for 45 percent of denials, were proposed POs that sought protection for an overly broad, or poorly described, category of documents. Together, at least one of these three errors was mentioned in 84 percent of all merits-based denials.235 Beyond that, 23 percent of denials included improper provisions for resolving disputes between parties, such as provisions that did not adequately describe mechanisms for contesting the confidentiality of a document. And a further 16 percent of orders were denied because they implied that the [*157] court had jurisdiction after the conclusion of the case. All other reasons for rejection accounted for 33 percent of denials.236

Figure 9. Stated Reasons for Denying a Stipulated PO

Second, and with more normative force, we considered whether POs were ever entered after an initial denial of a request for a stipulated PO. If a PO was not entered following a denial, the court's initial denial ensured discovery occurred without judicially imposed privacy, and parties were free to reveal information to others. We might also reason that, where a PO was never entered, (1) the PO was of little value to parties to begin with, (2) satisfying the

Ct., N. Dist. Ohio, Form Protective Order 1, 9 (same), https://www.ohnd.uscourts.gov/sites/ohnd/files/CivilRules_AppendixL.pdf [https://perma.cc/BR6A-S53G] (same). 232 More precisely, we analyzed 113 denials but excluded thirteen of these because they did not explain the reason for denial. 233 As noted above, in our analyses, we counted a PO as "denied" if it was denied in any part. 234 See supra notes 44-51 and accompanying text. 235 Notably, we identified many of these same errors in granted POs. 236 Most frequently, POs in this category were rejected because they failed to say that members of the public had a right to challenge confidential designations (9 percent) or purported to bind the court, court employees, or the jury to the terms of the PO (8 percent).

Page 42 of 54

ARTICLE: SECRECY BY STIPULATION

court's objection(s) would have devalued the PO, or (3) the parties could not meet the good cause requirement. Where a PO was later entered, the PO presumably was of comparatively higher value to litigants, who must have done additional work to convince the court to grant the order. Ultimately, we found that following denial in full of a joint PO request,237 40 percent of cases had no subsequent entry of a PO.

[*158] IV. IMPLICATIONS

In this final Part, we step back and assess the implications of our findings. Starting at ground level, Section A argues that our findings should, at a minimum, discipline future debate surrounding Rule 26(c). During decades of heated back-and-forth, many reform proposals have been short-circuited by claims that stipulated POs are very rare and, when filed, carefully scrutinized by courts.238 Our data powerfully undercut both those claims, especially the second.

Moving back slightly, Section B addresses the implications of our findings for tort law in particular. As Section B explains, there is a dawning recognition among many tort law scholars that a significant and perhaps the most significant function of tort litigation is to generate information. Tort law's raison d'être is not to provide compensation, corrective justice, or even old-style deterrence, these scholars argue; rather, it is to bring "stubborn information" to light.239 Through this lens, the Article's implications are clear. If tort law's principal function is to bring "stubborn information to light," when it comes to tort law, POs, which prevent the public disclosure of surfaced information, have a significant and heretofore radically underestimated antisocial effect.

Sections C and D assess our results through a wider-angle lens. Section C addresses the implications of our findings when it comes to the relative role of rulemaking versus discretion in American procedure. Section D then considers how our analysis contributes to a brewing debate about the future of the American civil justice system. Indeed, the debate about POs both reflects and sheds light on a slew of vitally important questions, including the nature of adversarialism and whether and how to adapt the "open courts" principle, as the American civil justice system enters a new and challenging chapter. A. Rebooting the Rule 26(c) Debate

First, our empirical discoveries can and should reboot the perennial debate surrounding Rule 26(c). As noted previously, opposition to Rule 26(c) reform proposals has long anchored on two arguments. First, that POs are rare. And second, that trial judges [*159] exercise careful, considered judgment when entering them. These core claims, moreover, have issued from prestigious quarters, lending heft to their frequent airing. For instance, Judge Mark R. Kravitz, testifying in Congress in 2008 on behalf of the Judicial Conference in opposition to the Sunshine in Litigation Act, directly asserted that "there is no empirical evidence to suggest that protective orders . . . are substantially used in the Federal courts or that there is any abuse."240 Or, as he put it at the same hearing: "[C]ourts review [PO] motions carefully and often deny or modify them to grant only the protection needed, recognizing the importance of public access to court filings."241 Some twenty years earlier, another defender of the Rule 26(c) status quo, Professor Arthur Miller himself a former reporter and Advisory Committee member reassured readers in an influential Harvard Law Review article: "No one is advocating the automatic or cavalier issuance of protective . . . orders, let alone that they be granted without regard for substantially deleterious effects on public health and safety."242 "The key," Miller continued, "is retaining judicial discretion."243 Because judges already "take their Rule 26(c) obligation very seriously,"244 Miller insisted, and because "proponents of reform have not

237 When a court partially grants a PO request, it enters the PO with some modification, and litigants need not file additional requests to enter a PO. Partial grants are, therefore, excluded from this analysis. 238 See supra notes 133-40 and accompanying text. 239 Wendy Wagner, Stubborn Information Problems & the Regulatory Benefits of Gun Litigation, in SUING THE GUN INDUSTRY: A BATTLE AT THE CROSSROADS OF GUN CONTROL & MASS TORTS 271, 271, 275-76 (Timothy D. Lytton ed., 2006). 240 Sunshine in Litigation Act of 2008: Hearings on H.R. 5884 Before the H. Comm. on the Judiciary, Subcomm. on Com. & Admin. L., 110th Cong. 74 (2008) (statement of Hon. Mark R. Kravitz for the Jud. Conf. of the United States). 241 Id. at 59 (statement of Hon. Mark R. Kravitz for the Jud. Conf. of the United States). 242 Miller, supra note 34, at 491. Extending the thought, Miller advised: "Judges must guard against any notion that the issuance of protective orders is routine, let alone automatic, even when the application is supported by all the parties." Id. at 492. 243 Id. at 491.

Page 43 of 54

ARTICLE: SECRECY BY STIPULATION

demonstrated that significant modification of the present framework is necessary, the existing pragmatic and discretionary balancing technique should be retained."245 And Judge Edward Becker also publicly expressed "doubt that any judge would approve a consent order not demonstrably rooted in Rule 26(c)."246

Finally, consider a colloquy between Representative William Hughes a supporter of the 1990s-era Sunshine in Litigation Act and [*160] Stephen C. Bransdorfer, Deputy Assistant Attorney General, who opposed reform:

Mr. Hughes. Mr. Bransdorfer, let me ask you a question just by way of a threshold question. Is it important for judges to make an independent determination that serves the public interest to sign a protective order?

Mr. Bransdorfer. Yes, it is.

Mr. Hughes. Is that policy being followed throughout our system today?

Mr. Bransdorfer. I think it is. . . . I think, by and large, a fair study of the system would indicate that . . . the use of protective orders does involve a showing of cause.247

Our findings, drawn from a "fair" study of the system indeed, a comprehensive study of more than 2.2 million dockets spanning nearly a decade of filings run directly contrary to these claims. We show that stipulated POs are quite common, perhaps increasingly so, and most judges do not appear to be carefully exercising discretion in entering them. Rather, the majority of judges in our data apparently never not even once saw a joint motion for a PO they wouldn't sign. Approval of POs appears to be "automatic or cavalier" exactly the situation Miller, alongside many others, has long said does not exist, should not be countenanced, and, if proven, would impel serious reform consideration.248

Of course, our results do not necessarily mean that arguments about the prevalence of POs or the level of judicial care in entering them are the only ones to be considered in a renewed debate about whether or how to revise Rule 26(c). It could be that policymakers could still decide that, particularly in an increasingly digital age, Rule 26(c)'s anemic implementation strikes the right balance between transparency and privacy and that the rule's text should merely be brought into alignment with the rule's on-the-ground [*161] implementation.249 What is clear, however, is that, so far, the decadeslong debate about POs has been deeply impoverished. Defenders of the Rule 26(c) status quo have fended off reform attempts by relying on dusty empirics and reassuring but, it turns out, inaccurate claims that the system is functioning as intended. By sweeping away these claims around stipulated POs' prevalence and judicial care in entering them, this Article can and should prompt renewed debate. And, just as critically, it ought to reorient the debate to focus more squarely on substantive questions, including the role of private civil litigation in a well-ordered society and the relative importance of transparency and secrecy as against competing values in the resolution of disputes. B. Stunting the Information-Forcing Function of Tort Law

Second, our findings have deep implications for the aims and limits of the tort system. Here, an increasingly popular view among tort law scholars is that an important benefit perhaps the primary benefit of the tort system is its "information-forcing function." This information-forcing theory posits that tort law indeed promotes safety. But it does so not exclusively, and perhaps not even mostly, via the much-discussed path of cost internalization that law and

244 Federal Sunshine in Litigation Act and Federal Court Settlements Sunshine Act, Hearings on H.R. 2017 and H.R. 3803 Before the H. Comm. on the Judiciary, Subcomm. on Intell. Prop. & Jud. Admin., 102d Cong. 26 (1992) (statement of Prof. Arthur R. Miller). 245 Miller, supra note 34, at 491. 246 Zenith Radio Corp. v. Matsushita Elec. Indus. Co., 529 F. Supp. 866, 889 n.40 (E.D. Pa. 1981). 247 Federal Sunshine in Litigation Act and Federal Court Settlements Sunshine Act, Hearings on H.R. 2017 and H.R. 3803 Before the H. Comm. on the Judiciary, Subcomm. on Intell. Prop. & Jud. Admin., 102d Cong. 105 (1992) (testimony of Stephen C. Bransdorfer, Deputy Assistant Att'y Gen., U.S. Dep't of Just.). 248 Miller, supra note 34, at 491. 249 For instance, as noted previously, reform champions and opponents have also long clashed over first principles, such as whether the primary purpose of adjudication is the resolution of private disputes or the public elaboration of legal norms a fundamental tension we return to below. See infra Section IV.D. A robust, revived debate about the future of Rule 26(c) should also consider the ways the world, including civil litigation, has changed in recent decades. As just one example, those who oppose Rule 26(c) reforms could plausibly argue that tectonic shifts in the political economy of information for instance, the increasing velocity, virality, and permanence of information in a digital world justifies a more privacy-protective approach. Such an approach may be particularly appropriate early in litigation before a court has fairly and finally adjudicated the allegations made against them. For seminal works on virality, see KARINE NAHON & JEFF HEMSLEY, GOING VIRAL 14 (2013) (exploring how virality shapes and informs social discourse); JONAH BERGER, CONTAGIOUS: HOW TO BUILD WORD OF MOUTH IN THE DIGITAL AGE 26 (2013) (analyzing the underlying psychological and sociological processes of social transmissions). For more discussion of privacy concerns and court data, see infra note 303 and accompanying text.

Page 44 of 54

ARTICLE: SECRECY BY STIPULATION

economics scholars, such as Professors Richard Posner, Mitch Polinsky, and Steven Shavell, have long championed.250 Rather, according to this new view, tort law operates in large part through its informational effect.251 [*162] The idea is that plucky plaintiffs' lawyers decentralized and well incentivized by contingency fees are uniquely positioned to connect dots, follow leads, depose insiders, and pry damning documents out of company vaults.252 And these lawyers' persistent efforts yield vital information that promotes sensible regulatory activity and also ensures that reputations reflect reality and steer consumers toward safer goods, services, and workplaces.253 Information, these scholars believe, is not ancillary. It is tort law's beating heart.254

The notion is not merely theoretical. Over recent decades, examples of instances where the tort system has generated critical information concerning public safety have accumulated.255 Litigation against gun manufacturers and sellers has generated significant policyrelevant information, including that gun executives have long known how guns are diverted into illegal secondary markets.256 Litigation addressing clergy sexual abuse has surfaced a mountain of information, including how Catholic Church officials sought to conceal the misconduct of various priests.257 Litigation involving polychlorinated [*163] biphenyls ("PCBs") has uncovered startling new evidence concerning the danger and prevalence of these industrial chemicals.258 Litigation against cigarette companies has uncovered information to show who knew what when about the dangers of smoking.259 And litigation against opioid manufacturers, sellers, and distributors has likewise generated stunning information concerning the breadth, depth, causes, and character of that continuing epidemic.260

Seen through this conceptual lens, the problem of POs is plain. If it is true, as some scholars claim, that tort law exists chiefly to bring "stubborn information to light," then the automatic and unreflective entry of POs stunts this vital informational exchange and stymies tort law's core utility.261

C. Rule Fidelity, Judicial Hierarchy, and Judge Discretion

A third set of implications sounds in judicial decision-making: Our findings raise deep concerns about fidelity to procedural rules and the complex interplay of rulemaking, judicial hierarchy, and trial judge discretion in the creation of American procedure.

Viewed from one angle, widespread rubber-stamping of POs, despite Rule 26(c)'s clear command, as supported by the stunning fact that more than half the federal judges in our data never denied a joint request for a PO during the study period, indicates an especially sharp disjunction between the law on the books and the law in action. Rule 26(c) states plainly that POs are to be granted only upon a showing of good cause.262 As the Eleventh Circuit has

250 See Nora Freeman Engstrom, Facilitating the Information-Forcing Function of Tort Law, JOTWELL (Mar. 22, 2022), https://torts.jotwell.com/facilitating-the-information-forcing-functionof- tort-law [https://perma.cc/QG7Q-KAB2]. 251 See Nora Freeman Engstrom & Michael D. Green, Tort Theory and Restatements: Of Immanence and Lizard Lips, 14 J. TORT L. 333, 335 n.5 (2021). 252 See Engstrom, When Cars Crash, supra note 4, at 333-35 (describing these dynamics); Wendy Wagner, When All Else Fails: Regulating Risky Products Through Tort Litigation, 95 GEO. L.J. 693, 700 (2007) (explaining that, compared "to their agency counterparts, litigants in tort cases are generally both more eager and more able to access asymmetric information held by manufacturers and industrial polluters"). 253 Burch & Lahav, supra note 10, at 356 ("When made public, information produced in discovery allows the press and researchers to study the documents, connect the dots, unmask health risks, shed light on regulatory failures, and pressure companies to make safer products."); Jacob Asaf & Roy Shapira, An Information-Production Theory of Liability Rules, 89 U. CHI. L. REV. 1113, 1148-49 (2022) (arguing that "the process of litigation propels regulators to act . . . because it provides them with valuable information on the regulated industry to which they were not privy" and also because it "makes available information more salient, thereby facilitating more intense media scrutiny of the regulator"). 254 Wagner, supra note 252, at 696 (highlighting the "indispensable role of tort litigation in lowering information-related barriers to regulating risky products"); Benjamin Ewing & Douglas A. Kysar, Prods and Pleas: Limited Government in an Era of Unlimited Harm, 121 YALE L.J. 350, 375 (2011) (explaining that "the tort system is a vital source of information gathering and intragovernmental feedback"). 255 Wagner, supra note 252, at 711-17 (collecting examples); see also Abbe R. Gluck, Elizabeth Chamblee Burch & Adam S. Zimmerman, Against Bankruptcy: Public Litigation Values Versus the Endless Quest for Global Peace in Mass Litigation, 133 YALE L.J.F. 525, 556-58 (2024) (same); Joanna C. Schwartz, The Cost of Suing Business, 65 DEPAUL. L. REV. 655, 680-83 (2016) (same). 256 Timothy D. Lytton, Using Tort Litigation to Enhance Regulatory Policy Making: Evaluating Climate Change Litigation in Light of Lessons from Gun-Industry and Clergy-Sexual- Abuse Lawsuits, 86 TEX. L. REV. 1837, 1845-46 (2008). 257 Id. at 1852-53. 258 Gerald Markowitz & David Rosner, Monsanto, PCBs, and the Creation of a World-Wide Ecological Problem, 39 J. PUB. HEALTH POL'Y 463, 463 (2018) (explaining that "an enormous trove of previously private Monsanto reports, papers, memos, letters, and studies" concerning PCBs has been made available to researchers, as a result of litigation). 259 Engstrom & Rabin, supra note 126, at 355-57. 260 Id. 261 Engstrom, When Cars Crash, supra note 4, at 333-35 (quotation marks omitted); cf. Egilman et al., supra note 11 (collecting pharmaceutical examples of POs shielding information relevant to public health from disclosure).

Page 45 of 54

ARTICLE: SECRECY BY STIPULATION

explained: "The plain text of [Rule 26(c)] suggests that a district court must find good cause to issue a protective order."263 Or, to quote the Third Circuit: "[T]he burden of justifying the confidentiality of each and every [*164] document sought to be covered by a protective order remains on the party seeking the protective order; any other conclusion would turn Rule 26(c) on its head."264

Supporting this authority, as noted previously, the Federal Judicial Center unambiguously states: "It is common for parties to present to the court a stipulated protective order for the court to sign . . . . It is only proper for the court to issue the order upon the court's finding that the order is supported by good cause."265 The Wright and Miller treatise concurs.266 And leading scholars, too, have resisted the notion that party agreement suffices, under Rule 26(c). Professor Steve Gensler, for example, cautions that, even when parties stipulate, "it is the court that ultimately must enter the [protective] order, and the court may do so only in compliance with Rule 26(c)'s good cause requirement."267 Professor Gustavo Ribeiro adds: "Even if the parties agree to proposed stipulated protective orders, the court must still determine whether good cause for issuing the order exists."268

Yet, even before our study, there were hints that this command was often honored in the breach. As one district court recently acknowledged: "It has become common practice for district courts to enter broad, stipulated protective orders without strict compliance with Rule 26(c)."269 And, more than a quarter-century ago, the influential Defense Research Institute also seemed to own this fact, instructing its members to "routinely seek a protective order limiting the dissemination of" documents, even though "defense counsel can make no special claim of confidentiality."270

There is, the above suggests, a glaring mismatch. The plain language in Rule 26(c), most appellate courts, and the Federal Judicial Center say that trial courts need to do one thing. But it appears that trial courts, instead, are doing quite another. This basic mismatch [*165] trial courts' basic disobedience contributes to at least two broader debates about American procedure making.

First, our findings raise important questions about the relative roles played by rulemaking, appellate oversight, and judicial discretion in American civil litigation. In recent decades, one of the signal developments in American civil procedure has been the steady deepening of the reservoirs of discretion within which trial judges operate.271 From the rise of "plausibility" pleading,272 to the centering of summary judgment,273 to increased attention to case management,274 to new controls and limits on trial,275 to the "Wild West" and often ad hoc administration of mammoth MDLs,276 American trial judges increasingly call their own tune and are, simultaneously, increasingly able to shield their day-to-day conduct from meaningful appellate scrutiny.277 Indeed, large swaths of contemporary

262 FED. R. CIV. P. 26(c). 263 In re Chiquita Brands Int'l, Inc., 965 F.3d 1238, 1249 (11th Cir. 2020) (per curiam) (emphasis omitted). 264 Cipollone v. Liggett Grp., Inc., 785 F.2d 1108, 1122 (3d Cir. 1986). 265 REAGAN, supra note 78. 266 Recall, it provides that "[e]ven when the parties consent, the court may not enter an order unless Rule 26(c) is satisfied." MARCUS, supra note 53. 267 STEVEN S. GENSLER, FEDERAL RULES OF CIVIL PROCEDURE, RULES AND COMMENTARY, Rule 26 (2021 ed.). 268 Ribeiro, supra note 32, at 190. 269 Fant v. City of Ferguson, No. 4:15-CV-00253-AGF, 2019 WL 4221515, at *3 (E.D. Mo. Sept. 5, 2019). 270 Kerry A. Kearney & Tracy G. Benson, Preventing Non-Party Access to Discovery Materials in Product Liability Actions: A Defendant's Primer, 1987 CURRENT ISSUES IN L. & MED. 36, 40. 271 See David Freeman Engstrom, Digital Civil Procedure, 169 U. PA. L. REV. 2243, 2246-52 (2022) [hereinafter Engstrom, Digital Civil Procedure]. 272 In Bell Atlantic Corp. v. Twombly, 550 U.S. 544, 565 (2007), the Court conferred substantial discretion on trial judges in instructing them to assess the plausibility of allegations "in light of common economic experience." That the discretion conferred was not limited to Twombly's antitrust setting became clear two years later in Ashcroft v. Iqbal, 556 U.S. 662, 663-64 (2009). 273 For the radical growth of summary judgment, see Nora Freeman Engstrom, The Lessons of Lone Pine, 129 YALE L.J. 2, 67 n.293 (2019) [hereinafter Engstrom, The Lessons of Lone Pine]. 274 Considered in tandem, Rules 16 and 26 grant enormous discretion to trial judges to control the breadth and pace of discovery. See Stephen B. Burbank & Linda J. Silberman, Civil Procedure Reform in Comparative Context: The United States of America, 45 AM. J. COMP. L. 675, 679 (1997). For the classic discussion of the growth of and tradeoffs surrounding trial judge discretion, see generally Judith Resnik, Managerial Judges, 96 HARV. L. REV. 374 (1982). 275 Nora Freeman Engstrom, The Trouble with Trial Time Limits, 106 GEO. L.J. 933, 941-46 (2018). 276 See Engstrom, The Lessons of Lone Pine, supra note 273, at 6-7 (discussing the ascendance of MDLs and MDL judges' persistent and potentially lawless "ad hocery"). 277 See Engstrom, Digital Civil Procedure, supra note 271, at 2179-80.

Page 46 of 54

ARTICLE: SECRECY BY STIPULATION

civil procedure can be seen as a deliberate conferral of discretion on trial judges to tailor procedural rulings to individual cases or even to make procedure out of whole cloth.278

Against this broader backdrop, our findings concerning district court disobedience when granting stipulated POs are doubly provocative. Enriching and complicating current discussion of [*166] judicial discretion, our findings reveal that, even on those seemingly rare occasions when trial judges' discretion is cabined, trial judges are instead free agents, untethered to formal commands.

Second, our findings raise vital questions about other pathways through which American procedure is made, particularly the role of rulemaking under the Rules Enabling Act. The Rules Enabling Act ("REA"), supposedly, is the cornerstone of the system.279 But we show that, whether or not rulemakers formally choose to revisit Rule 26(c), trial judges have already affected substantial, sub rosa changes to its core operation, particularly the "good cause" standard.

That startling insight is important in its own right and it also connects to wider concerns about the process of American procedure making. A rich empirical literature examines efforts to shape American procedure. One key finding is that some of the most impactful procedural reforms have come not by way of legislation or rulemaking, but rather through court decisions, precisely because these decisions tend to fall below the public radar.280 Technical and turgid court decisions, the theory goes, flunk the "Dan Rather" test, named after the news anchor of old. A tweak to Rule 8, 12, or 56, buried deep in paragraph nineteen of a decision issued sometime in June, can be hugely impactful, but it does not mobilize constituencies and, indeed, does not register with the public at all.281

Our findings enrich that idea. Recall, in the mid-1990s, the Judicial Conference considered a package of amendments to Rule 26(c). Among them was a proposal that would have allowed courts to issue a PO "for good cause shown or on stipulation of the parties."282 The [*167] proposal, then, would have formally allowed a PO to be issued simply on the parties' say-so.283

Yet the proposed amendment sent shockwaves through the legal world and ultimately, in a highly unusual move, the Judicial Conference rejected the amendment in the face of blistering criticism.284 In particular, opponents lodged two core objections. Their first was procedural: that the proposal was added in the eleventh hour.285 Second, opponents argued that the rule, which would represent a "dramatic shift" from the status quo,286 would

278 Engstrom, The Lessons of Lone Pine, supra note 273, at 8-10 (discussing judicial ad hocery and the growth of "DIY procedures"). For a classic discussion of discretion and the Federal Rules of Civil Procedure, see Stephen N. Subrin, How Equity Conquered Common Law: The Federal Rules of Civil Procedure in Historical Perspective, 135 U. PA. L. REV. 909 (1987). 279 See 28 U.S.C. § 2072; see also Stephen B. Burbank, The Rules Enabling Act of 1934, 130 U. PA. L. REV. 1015, 1025 (1982) (offering the fullest account of the REA's origins and purposes); Engstrom, Digital Civil Procedure, supra note 271, at 2246-58 (cataloging increased pressure on the REA's rulemaking process as a result of the felt need for judicial discretion over rule implementation, separation-of-powers tensions, and the unmasking of the distributive consequences of rules via increasingly sophisticated empiricism). 280 STEPHEN B. BURBANK & SEAN FARHANG, RIGHTS AND RETRENCHMENT: THE COUNTERREVOLUTION AGAINST FEDERAL LITIGATION 22-23 (2017). 281 Id. To supply just one example, the Court in Twombly achieved important changes to the pleading standard where Congress and the Advisory Committee had failed and it did so just five years after the Supreme Court unanimously insisted, for the second time in less than a decade, that only the rulemaking process could change the very same pleading standard. Swierkiewicz v. Sorema N.A., 534 U.S. 506, 515 (2002) ("A requirement of greater specificity for particular claims is a result that must be obtained by the process of amending the Federal Rules, and not by judicial interpretation." (quotation marks omitted)). 282 Letter of June 2, 1995, supra note 84, at 6 (emphasis added). 283 Some felt that the move would merely "confirm[] the common practice of entering protective orders on stipulation by the parties." Advisory Comm. on Civ. Rules, Jud. Conf. of the U.S., Meeting Minutes of the Advisory Committee on Civil Rules 8 (Oct. 20, 1994), https://www.u scourts.gov/sites/default/files/fr_import/CV10-1994.pdf [https://perma.cc/A6Z9-NNHT]. Others disagreed. See, e.g., Richard B. Schmitt, Judges' Group Kills Protective Order Rule, WALL ST. J., Mar. 15, 1995, at B4 (quoting the chairman of the executive committee, Judge Gilbert Merritt, as stating that the stipulation rule departs from long-standing practice); Letter of June 2, 1995, supra note 84, at 1-2 ("[E]xpress[ing] concern that the proposed rule would change existing practice by allowing entry of protective orders without a showing of good cause."). 284 Letter of June 2, 1995, supra note 84, at 2; see also Arthur R. Miller, Effective Rulemaking Damaged by Politics, NAT'L L.J., May 1, 1995, at A21 ("Media and special interest group lawyers successfully mounted a last-minute, high-visibility protest against the proposal."). For representative criticism, see, e.g., Arthur Bryant, A Sneak Attack on Open Justice, NAT'L L.J., July 10, 1995, at A19; Jane Kirtley, Keeping the Public in the Dark, AM. JOURNALISM REV., July 1995, at 50. Indeed, the stipulation proposal proved so contentious that the Judicial Conference rejected the entire package of Rule 26(c) amendments including provisions that would have clarified the standard for modification and dissolution of POs without discussing the other reforms. See Letter of June 2, 1995, supra note 84, at 2. 285 Although the other amendments in the Rule 26(c) package had been circulated months in advance and had been the subject of extensive public comment, the stipulation provision was added shortly before the vote. See Schmitt, supra note 283, at B4; Bryant, supra note 284, at A19. 286 Bryant, supra note 284, at A19. As noted, not everyone agreed with this characterization. See supra note 283.

Page 47 of 54

ARTICLE: SECRECY BY STIPULATION

facilitate unbridled secrecy, which would harm public health.287 This was, at least in part, because "individual plaintiffs suing large corporations . . . have little choice but to agree" to POs "as the price for avoiding prolonged, expensive pretrial discovery."288

Following the vote, Professor Linda Greenhouse wrote a New York Times story summarizing the debate and its stakes.289 The story quoted Judge Gilbert S. Merritt who described his opposition to the [*168] amendment: "Federal courts shouldn't do anything without just cause."290 Secrecy "should not just be left to the option of the parties."291

But we show that, although a formal effort to rewrite Rule 26(c) failed, somehow, through a drip, drip, drip of district court discretion, Rule 26(c) was nevertheless rewritten. Secrecy has been left to the option of the parties. This startling insight connects to a broader set of concerns about procedure making's pathways and adds fuel to a growing critique about a derogation of the rulemaking process.292

D. Open Courts, Adversarialism, and Access to Justice

Finally, evidence concerning the actual operation of Rule 26(c) contributes to a swirling debate about whether two cornerstones of the American civil justice system the "open courts" principle and the American commitment to adversarialism should be adapted to new legal realities.293 Indeed, our Rule 26(c) empirical findings both reflect and shed light on debate about whether and how to revise both mainstays of the American system.

Debate about the meaning and wisdom of the "open courts" principle is not exactly new. It has come, first and foremost, in disputes around how best to interpret the array of constitutional, statutory, and [*169] procedural provisions from which the principle emanates.294 Debate has also come in more academic forms, about the core purposes of adjudication. According to some, courts exist to resolve private disputes295 and any consideration of the public interest blurs and blunts that core purpose. Courts, it follows, should "readily accede to the parties' mutual desire for confidentiality,"296 lest the parties turn courts into "information ombudsmen," distracted from their

287 See, e.g., Bryant, supra note 284, at A19 (pointing out that "[i]f these proposed changes had been in effect, critical documents about the dangers of breast implants, heart valves, the Ford Pinto, all-terrain vehicles and other hazardous products could still be under seal"). 288 See Greenhouse, supra note 87, at B9. 289 Id. 290 Id. 291 Id. 292 See Engstrom, Digital Civil Procedure, supra note 271, at 2246-58 (tracing heated debate about the place of rulemaking in the American civil justice system). For further discussion, see generally Stephen B. Burbank, Procedure, Politics and Power: The Role of Congress, 79 NOTRE DAME L. REV. 1677 (2004) (considering the constitutional dimension); BURBANK & FARHANG, supra note 280 (offering a wide array of empirics on the operation and public perception of the rulemaking process); David Marcus, The Collapse of the Federal Rules System, 169 U. PA. L. REV. 2485 (2021) (examining rulemaking through the lens of core procedural values of transsubstantivity and neutrality). 293 "Open courts" is a longstanding principle within the American constitutional tradition and legal system, with doctrinal roots in privileges and immunities, free speech, due process, and equal protection and finding further expression in a large body of statutes, common-law rules, and procedural rules. See infra note 294. The traditional American commitment to adversarialism holds that a party-driven clash of proof in a structured, forensic setting before a passive, neutral decision-maker is the best way to resolve disputes. STEPHAN LANDSMAN, THE ADVERSARY SYSTEM 2 (1984). For a further discussion of the United States' commitment to the adversarial resolution of disputes, see generally ROBERT A. KAGAN, ADVERSARIAL LEGALISM: THE AMERICAN WAY OF LAW (2d ed. 2019). For a rich historical perspective, see generally AMALIA D. KESSLER, INVENTING AMERICAN EXCEPTIONALISM: THE ORIGINS OF AMERICAN ADVERSARIAL LEGAL CULTURE, 1800-1877 (2017). 294 A large body of constitutional provisions, statutes, common-law rules, and procedural rules extolls the virtues of "open courts" and grants the public broad rights of access to civil court proceedings. See, e.g., Press-Enterprise Co. v. Super. Ct. of Cal. (Press-Enterprise I), 464 U.S. 501, 508 (1984) (lauding the "community therapeutic value" of open proceedings (citation omitted)); Joy v. North, 692 F.2d 880, 893 (2d Cir. 1982) ("An adjudication is a formal act of government, the basis of which should, absent exceptional circumstances, be subject to public scrutiny."); In re Union Leader Corp., 809 A.2d 752, 753 (N.H. 2002) (affirming that open courtrooms and access to court records are "critical to ensure that court proceedings are conducted fairly and impartially"); see also Landmark Commc'ns, Inc. v. Virginia, 435 U.S. 829, 839 (1978) ("The operations of the courts and the judicial conduct of judges are matters of utmost public concern."); FED. R. CIV. P. 43(a) ("At trial, the witnesses' testimony must be taken in open court unless a federal statute, the Federal Rules of Evidence, these rules, or other rules adopted by the Supreme Court provide otherwise."); 28 U.S.C. § 452 ("All courts of the United States shall be deemed always open for the purpose of filing proper papers, issuing and returning process, and making motions and orders."). See generally Jonathan M. Hoffman, By the Course of the Law: The Origins of the Open Courts Clause of State Constitutions, 74 OR. L. REV. 1279, 1279 (1995) (canvassing state constitutional provisions). 295 See Miller, supra note 34, at 431 (noting that a focus on public access would divert courts "from their primary mission" of resolving disputes among litigants); id. at 441 ("[T]he function of the judicial system is to resolve private disputes, not to generate information for the public."); Marcus, supra note 111, at 468 ("[C]ourts exist to resolve disputes that are brought to them by litigants . . . ."); id. at 470 ("The primary purpose for which courts were created, distinguishing them from other organs of government, is to decide cases according to the substantive law. The collateral effects of litigation should not be allowed to supplant this primary purpose."); see also supra notes 147-48 and accompanying text (rehearsing these arguments).

Page 48 of 54

ARTICLE: SECRECY BY STIPULATION

core function of adjudicating private disputes.297 Others take the opposite view. Courts, Professor Owen Fiss has told us, "exist to give meaning to our public values, not to resolve disputes."298 They should be no less [*170] open and transparent than other democratic institutions perhaps even more so.299

What is different in the current moment is the "open courts" principle is facing significant pressure and even something of a reckoning. For starters, and as noted previously, numerous high-profile scandals and public health crises have revitalized the debate about the core purposes of adjudication by shining unflattering light on the ways society's "haves" are able to hide wrongdoing and perpetuate it. Horrified by these revelations, at last count, more than a dozen states have enacted new laws limiting secret settlements or NDAs for certain claims.300 And the federal government has also gotten in the act by, for example, amending the tax laws to establish that defendants cannot deduct payments incurred to settle sexual harassment cases where an NDA shields the settlement from scrutiny.301

But that is only the beginning. A quickly digitizing legal system also raises important questions about the terms on which courts make available their mountains of case data. Some argue that open data is essential if a growing legal tech industry is to serve impecunious clients (as well as their wealthy corporate counterparts).302 Others worry that, if made easily available, court data could be downloaded in bulk and used for dubious, and perhaps discriminatory, ends.303 Add in [*171] legitimate privacy concerns, leavened with the velocity, virality, and permanence of information in the digital age, and it is clear that Rule 26(c) is part of a wider reckoning about how information flows into and, perhaps more importantly, out of America's courts.304

The traditional American commitment to adversarial process is likewise under significant scrutiny and possible rethinking. Adversarialism the notion that key procedural values such as accuracy, efficiency, and fairness are best optimized where opposing litigants compete by placing arguments before a neutral and passive judge has long been a core tenet of the American civil justice system. As Roscoe Pound told the ABA in his now-famous 1906 address: "[I]n America we take it as a matter of course that a judge should be a mere umpire . . . and that the parties should fight out their own game in their own way without judicial interference."305

But the American commitment has long shown cracks in its foundation and, here again, the debate over Rule 26(c) both reflects and sheds light on that broader assessment. Many observers have long expressed doubt about whether an adversarial process, as opposed to an inquisitorial approach in which judges actively steer the course of litigation, is the better bet.306 Critics have long argued that adversarialism may not optimize the system's truth-

296 Doré, supra note 83, at 289; see David S. Sanson, The Pervasive Problem of Court- Sanctioned Secrecy and the Exigency of National Reform, 43 DUKE L.J. 807, 810 (2003) ("Jurisprudentially, the prosecrecy argument insists that the judicial system is a forum for private parties to resolve private disputes and is not an instrument of social justice."); id. ("[T]he prosecrecy argument typically bases its logic on the premise that courts exist to assist private litigants in resolving their disputes in the most efficient way possible."); cf. Erichson, supra note 59, at 360 ("A light standard of good cause for discovery confidentiality reduces the burden on the court . . . ."). 297 Miller, supra note 34, at 488. 298 Fiss, supra note 130, at 29. Others have said much the same. See Luban, supra note 130, at 2657 (describing information generation as "a fundamental public interest" of litigation, not a side effect); Abram Chayes, The Role of the Judge in Public Law Litigation, 89 HARV. L. REV. 1281, 1313-16 (1976) ("In my view, judicial action only achieves such legitimacy by responding to, indeed by stirring, the deep and durable demand for justice in our society."). 299 For further articulations, see supra note 128 and accompanying text (rehearsing these arguments). 300 Engstrom et al., supra note 8. 301 Id. (collecting this and other recent federal efforts). 302 Gelbach, supra note 181, at 133-54 (advocating the liberation of PACER); David Freeman Engstrom & Nora Freeman Engstrom, Legal Tech and the Litigation Playing Field, in LEGAL TECH AND THE FUTURE OF CIVIL JUSTICE 133-54 (David Freeman Engstrom ed., 2023) (explaining that, currently, the "have's" have disproportionate access to proprietary court data and that, over time, this difference is apt to skew the development of substantive law); David Freeman Engstrom & R.J. Vogt, The New Judicial Governance: Courts, Data, and the Future of Civil Justice, 72 DEPAUL L. REV. 171, 198-224 (2023) (offering a wide-angle assessment of courts' role as "data dispensers"); Zachary D. Clopton & Aziz Z. Huq, The Necessary and Proper Stewardship of Judicial Data, 76 STAN. L. REV. 893, 894 (2024) (offering a "normative evaluation of how Congress should regulate the production and dissemination of judicial data, in light of the capabilities and incentives of relevant actors"). 303 Among the more worrying and demoralizing uses of court data is the practice of landlord associations in most major cities of scraping housing court dockets to construct blacklists of renters who should not be rented to because they dared to try to vindicate rights in housing court. See Ronda Kaysen, How to Escape the Dreaded 'Tenant Blacklist', N.Y. TIMES, Apr. 13, 2019, at RE2 ("There are hundreds of tenant screening bureaus, collecting names from courthouses around the country and selling the information to landlords."). 304 See generally David S. Ardia & Anne Klinefelter, Privacy and Court Records: An Empirical Study, 30 BERKELEY TECH. L.J. 1807 (2015) (cataloging the types of sensitive information contained in court records, including locational, identity, health, and financial information as well as past involvement in criminal or civil proceedings). 305 Roscoe Pound, The Causes of Popular Dissatisfaction with the Administration of Justice, 29 ANN. REP. AM. BAR ASS'N 395, 405 (1906).

Page 49 of 54

ARTICLE: SECRECY BY STIPULATION

seeking function. Judge Henry Friendly famously observed that "the role of counsel is not to make sure the truth is ascertained" and then suggested that we should wonder about a system in which "causing delay and sowing confusion not only are [counsel's] right but may be his duty."307 Revelations that a stunning three-quarters of civil cases filed in American courts feature at least one side without a lawyer deepen the critique and make continued adherence to unmediated adversarialism [*172] hard to defend under any reasonable theory of due process or procedural fairness.308

Our findings add another example of how adversarialism may systematically fail to achieve key systemic ends. After all, there is no constituency for disclosure in a system in which parties interested in confidentiality place arguments before a passive judge who does not consider the public interest. In economic terms, there is a negative externality a harm borne by persons whose interests the stipulating parties can freely ignore.309 And that externality is accumulating. Put differently, the adversarial structure of the American civil justice system might leave it ill-equipped to promote anything resembling an optimal level of openness.310 Indeed, when a stipulated PO is at play, the plaintiff is not pitted against the defendant. By definition, the plaintiff and defendant, jointly, are pitted against the rest of us. CONCLUSION

On an unseasonably hot afternoon in Washington, D.C., on May 17, 1990, the late Senator Herb Kohl, a perennial sponsor of the Sunshine in Litigation Act and perhaps the most steadfast champion of transparency in civil litigation, called to order one of the first hearings ever to discuss the matter. In so doing, he observed: "[W]e must ask whether we have struck the proper balance between disclosure and secrecy."311 Senator Kohl proceeded to highlight a few questions that [*173] needed to be answered in order to assess that broader inquiry. These included: "Is there a growing use of court secrecy in civil litigation, and if so, does it prevent disclosure of information that raises safety concerns?" And: "Do the demands of the court system encourage indiscriminate use of secrecy, and how can these competing interests best be resolved?"312

For more than a quarter-century, Senator Kohl's simple questions have been much discussed, but they have remained stubbornly unanswered. Offering overdue clarity, this Article shows that, in fact, in contemporary federal civil litigation, the balance has tipped more sharply in favor of secrecy than has been previously recognized. Indeed, in opposing Sunshine in Litigation Acts and similar reforms, critics have, for decades, insisted that courts are "very circumspect about entering protective orders."313 No reform is needed, some have maintained, because judges "take their Rule 26(c) obligations very seriously."314 Yet, as we have shown, the data suggest otherwise. Judges do not appear to cast a discerning eye on POs. They reject, on average, just 4.1 percent of joint PO requests. In fact, some judges never seem to be circumspect in entering POs. Of those federal court judges that were asked to sign more than twenty-five POs, the majority of judges' grant rate is a bracing 100 percent.

306 E.g., John H. Langbein, The German Advantage in Civil Procedure, 52 U. CHI. L. REV. 823, 824 (1985). 307 Henry J. Friendly, Some Kind of Hearing, 123 U. PA. L. REV. 1267, 1288 (1975). 308 See, e.g., Jessica K. Steinberg, Adversary Breakdown and Judicial Role Confusion in "Small Case" Civil Justice, 2016 B.Y.U. L. REV. 899 (2016) (coining the phrase "adversary breakdown" to describe the millions of cases pitting institutional plaintiffs with lawyers against individuals without lawyers in debt, eviction, home-foreclosure, and family law, particularly child support, matters); Pamela K. Bookman & Colleen F. Shanahan, A Tale of Two Civil Procedures, 122 COLUM. L. REV. 1183, 1200, 1210 (2022) (observing that "the adversarial system was not designed to be used by self-represented parties" and discussing the erosion of the "adversarial archetype" in "lawyerless courts"); Marcus, supra note 292, at 2506-08 (cataloging the ways debt collection litigation makes a "mockery . . . of the adversarial assumption" that underpins American procedure). 309 Ribeiro, supra note 32, at 174. 310 See Burch & Lahav, supra note 10, at 350, 394 (explaining that "[i]nformation revealed in lawsuits adjudicated in taxpayer-funded courts often benefits society as a whole, but there is no incentive for private parties to reveal it" and arguing that, given this mismatch, courts ought to work harder to promote the public interest). 311 Examining the Use of Secrecy and Confidentiality of Documents by Courts in Civil Litigation, Hearing Before the S. Comm. on the Judiciary, Subcomm. on Cts. & Admin. Prac., 101st Cong. 1 (1990) (statement of Sen. Herb Kohl) [hereinafter Kohl Statement]. For the weather that May afternoon, see Washington D.C. Weather in 1990, EXTREME WEATHER WATCH, https://www.extremeweatherwatch.com/cities/washington- dc/year-1990 [https://perma.cc/YJ34-D7RG]. 312 Kohl Statement, supra note 311, at 2. 313 The Federal Judiciary in the 21st Century: Ensuring the Public's Right of Access to the Courts, Hearing Before the H. Comm. on the Judiciary, Subcomm. on Cts., Intell. Prop., & the Internet, 116th Cong. 32 (2019) (statement of Hon. Richard W. Story). 314 Federal Sunshine in Litigation Act and Federal Court Settlements Sunshine Act, Hearings on H.R. 2017 and H.R. 3803 Before the H. Comm. on the Judiciary, Subcomm. on Intell. Prop. & Jud. Admin., 102d Cong. 25 (1992) (statement of Prof. Arthur R. Miller).

Page 50 of 54

ARTICLE: SECRECY BY STIPULATION

Beyond concerns of fidelity to law and binding precedent, these findings suggest that infamous cases such as Beth and Ken Melton's wrongful death lawsuit against GM, or others, such as that those involving Zyprexa, Zomax, or the Dalkon Shield are not a bug of the legal system, but rather a feature. They show how the legal system typically operates. What is perhaps unusual about those infamous cases is that they were the rare instances when the truth was ever, belatedly, revealed.

[*174] APPENDIX

This Appendix further details how we constructed our estimation dataset using a bevy of natural language processing and machine learning tools. In total, our dataset contains 2,245,225 civil cases filed in federal district courts between January 1, 2005, and December 31, 2012.315 The critical challenge and the reason we resorted to modern data science tools was translating unstructured docket entry text into useable measures of protective order ("PO") activity.

For context, here is a sample docket entry316:

ORDER RE: PARTIES' PROPOSED PROTECTIVE ORDER BY MAGISTRATE JUDGE FERNANDO M. OLGUIN, RE STIPULATION FOR PROTECTIVE ORDER 23. THE COURT IS UNABLE TO ADOPT THE PROTECTIVE ORDER AS STIPULATED TO BY THE PARTIES FOR THE FOLLOWING REASONS. (SEE DOCUMENT FOR FURTHER DETAILS). (MR) (ENTERED: 12/21/2011)317

This text contains two references to our orders of interest: (1) "RE STIPULATION FOR PROTECTIVE ORDER", and (2) "THE PROTECTIVE ORDER AS STIPULATED TO BY THE PARTIES".318 As we discovered during hand labeling, these are just two of the many subtly distinct ways courts describe issuing an order concerning a stipulated PO in docket entry text.

When sifting through the docket data, we sought to answer three questions with various natural language processing methods. First, did the docket entry describe an order that disposed of a request for a PO (whether stipulated or nonstipulated)? Second, if so, did the court grant or deny the request? Third, does any motion in the case's docket indicate that the relevant request for a PO was stipulated? For the reasons discussed in the main text, we constructed data regarding this third question as a precautionary measure to mitigate false negatives.319

[*175] Although the details vary a bit, we followed a similar process to construct each measure. Our workflow can be distilled into three steps described below. I. FILTER DOCKET ENTRIES TO ENTRIES ABOUT POS

The complete universe of docket entries includes tens of millions of entries. Ideally, we could have used machine learning from the outset to identify which of these entries discuss POs. But, the relevant entries discussing POs amount to a small fraction of all docket entries, and machine learning models generally do not perform well at predicting such rare events, which the machine learning literature calls "severe class imbalance."320

We, therefore, restricted attention to only those docket entries that included some version of the string "protective order",321 assuming that docket entries involving POs would have some version of that phrase in them. We identified 899,407 docket entries that contained this string. Many of these docket entries did not involve either

315 Because we use two years' worth of information for each of our cases, our estimation data includes information covering the period from January 1, 2005, through December 31, 2014. 316 Notably, this is a rare example of an order denying a stipulated PO. 317 Docket Entry, Raw Talent Inc. v. Frankel, No. 2:11-cv-06577 (C.D. Cal. Aug. 10, 2011), ECF No. 24. 318 Id. 319 Our concern was that some docket entries that we otherwise determined were general POs might actually involve stipulated POs, which, if unaddressed, would cause us to undercount the total number of stipulated POs. 320 See, e.g., Joffrey L. Leevy, Taghi M. Khoshgoftaar, Richard A. Bauder & Naeem Seliya, A Survey on Addressing High-class Imbalance in Big Data, 5 J. BIG DATA, Nov. 1, 2018, at 1-2. 321 These are standard examples of what are known as dictionary methods. See GRIMMER ET AL., supra note 186, at 178-83 (describing dictionary methods and other word-counting methods).

Page 51 of 54

ARTICLE: SECRECY BY STIPULATION

judicial decisions about a PO (whether stipulated or nonstipulated) or motions for stipulated POs.322 For further data refinement, we relied on more advanced methods using supervised machine learning.323

II. USING MACHINE LEARNING TO IDENTIFY JUDICIAL ORDERS AND MOTIONS

Supervised machine learning requires a training dataset with a target variable here, whether the docket entry involves judicial orders or motions involving POs.324 To create a training dataset, we hand labeled a random sample of 2,250 docket entries matching our PO expression. We labeled for our three outcomes: (1) judicial orders about POs, (2) judicial orders about stipulated POs, and (3) motions [*176] about stipulated POs.325 Each of these outcome labels occurred frequently enough that we did not suspect a class imbalance problem existed. Of the hand-coded docket entries, 16 percent were judicial determinations about POs, with stipulated POs and motions about stipulated POs clocking in at 12 percent and 10 percent, respectively.

We then trained a series of machine learning models on the training dataset. To translate the unstructured text into a data frame that could then be used for prediction purposes, we used a conventional bag of words model.326 A bag of words representation is an "n by k" data frame, where, in the present context, n is the number of docket entries and k is the number of words included in the model. The number in the ith row and jth column of the dataset tells us how many times the ith word occurs in the jth docket entry, for each i = 1, 2, . . ., n and j = 1, 2, . . ., k. We used this data frame to predict the labels for example, whether or not the docket entry was a PO determination with a machine learning classifier. While training the classifier, we split our hand-labeled data so that 80 percent of the entries were used for training the model ("training set") and 20 percent were used for testing the model ("test set").327 The classifier "learns" the patterns connecting terms in the training-test docket entries to our outcome labels. The 20-percent test set is withheld from this process and then used to evaluate the performance of the model estimated using the training set.

During training, we modified the bag of words model in conventional ways to improve performance. Before counting the number of terms used in each entry, we cleaned the text data by removing punctuation and numbers, lowercasing all text, and stemming the words.328 We also experimented with different restrictions on our bag of words model. In particular, we considered different thresholds for the minimum and maximum number of docket entries a term must occur in to be included in the bag of words model. [*177] Finally, we tested various machine learning classifiers. In the end, the best-performing model was the gradient boosting model.329

Our test set classification statistics shed light on how well the classifiers perform. In Appendix Figure 1, we present four conventional machine learning classification statistics for the test sets corresponding to our three labels:

(1) "Accuracy" is the number of labels correctly predicted divided by the total number of labels.330

(2) "Recall" is the number of true positives divided by the sum of the numbers of true positives and false negatives.331

(3) "Precision" is the number of true positives divided by the sum of the numbers of true positives and false positives.332

322 To understand why that is unsurprising, consider that docketed scheduling orders might mention a judge-imposed deadline for filing POs, or that parties might request extensions in separate docket entries, and so on. 323 See GRIMMER ET AL., supra note 186, at 197-210 (detailing how to use supervised machine learning for text classification). 324 See id. 325 These labels are mutually exclusive. In other words, a docket entry reflecting a judicial conclusion about a PO is either a stipulated PO or a general PO. 326 See GRIMMER ET AL., supra note 186, at 48-59 (describing the bag of words model). 327 See id. at 212 (discussing measures to analyze classifier performance). 328 See id. at 52-55 (discussing preprocessing steps). Stemming truncates words to their core components for example, all of "programming", "programmers", and "programs" become "program". 329 See id. at 204 (discussing boosting models). 330 Julian Nyarko, We'll See You in . . . Court! The Lack of Arbitration Clauses in International Commercial Contracts, 58 INT'L REV. L. & ECON. 6, 10 (2019) (discussing "correct classification rate"). 331 See id. at 10 n.24. 332 See id.

Page 52 of 54

ARTICLE: SECRECY BY STIPULATION

(4) "F1" is the harmonic mean of Recall and Precision.333

All these measures have attractive properties in some cases. However, there are situations in which it makes sense to prefer some over others. For example, the labels in which we are interested are rare in our dataset. Roughly 10 percent of entries in our dataset are stipulated POs, so a classifier that simply declares all entries not to be stipulated POs would be correct 90 percent of the time, yielding an Accuracy score equal to 90 percent. In such circumstances, Recall and Precision are helpful for assessing classifier performance because both reward correct labeling for example, true positives.334 Each, though, penalizes different types of mistakes. Recall penalizes false negatives, such as labeling an actual SPO as a non-SPO; Precision penalizes false positives, such as labeling an actual non-SPO as an SPO. Because F1 depends on both Precision and Recall and can be shown to be increasing in each, F1 represents a kind of compromise between those two measures. That is, F1 rewards true positives and penalizes both false negatives and false positives. We note that when Recall and Precision are roughly equal, so that each approximately equals some [*178] value x, F1 will also have approximately the value x. Thus, F1 differs substantially from Recall and Precision only when Recall and Precision differ substantially from each other.

Appendix Figure 1. Supervised Machine Learning Accuracy Statistics335

Appendix Figure 1 indicates that our classifiers do well on all four performance metrics. Each model achieves over 97 percent Accuracy, but the high performance on Precision and Recall shows that this percentage is not just due to the rare label phenomena discussed above. Further, the Precision and Recall scores are very similar for all three of our labels, with the predictable result that the F1 scores have similarly high values: Each model achieves an F1

333 See GRIMMER ET AL., supra note 186, at 212. Note that the harmonic mean of two numbers is twice their product divided by their sum. Thus, F1 equals twice the product of Recall and Precision, divided by the sum of Recall and Precision. 334 Note that each of these measures would have a value of zero with the classifier that labels all entries as non-SPOs, because that approach yields no true positive labels. 335 This plot uses the naming conventions described in the main text. "GPO" is a general PO that is, nonstipulated. "SPO" is a stipulated PO. And "MSPO" is a motion for stipulated PO.

Page 53 of 54

ARTICLE: SECRECY BY STIPULATION

score above .9. Although performance is a task-dependent evaluation, previous work in legal studies using machine learning has considered that an F1 around .8 or above indicates a well-performing classifier.336

[*179] With the best-performing classifiers, we then made predictions for the entire dataset of entries matching our PO regular expression. These predictions form the core of our measures.337

III. IDENTIFYING DENIALS

When measuring judicial decisions about stipulated POs and general POs, we wanted to identify whether the court denied the PO request. One challenge we uncovered along the way was that courts denied very few stipulated PO requests. Fewer than 5 percent of the docket entries labeled as orders about requests for stipulated POs in our training data were denials. Although this discovery motivated an intuition that became a key finding of our study, it also indicated the presence of another class imbalance problem.

Accordingly, we once again turned to dictionary methods. We tailored a flexible regular expression to determine when docket entries predicted as involving stipulated POs denied the parties' requests. Specifically, we looked for matches to the following at the sentence level:

[DENIES.*PROTECTIVE ORDER|PROTECTIVE

ORDER.*DENIES|PROTECTIVE ORDER.*UNABLE TO ADOPT|UNABLE TO ADOPT.*PROTECTIVE

ORDER|PROTECTIVE

ORDER.*DENIED|DENIED.*PROTECTIVE

ORDER|PROTECTIVE

ORDER.*DECLINES|DECLINES.*PROTECTIVE

ORDER|PROTECTIVE

ORDER.*DENYING|DENYING.*PROTECTIVE ORDER]

In this expression, "|" means "or" and ".*" means any number of characters between the terms ".*" is in between. So, for instance, "DENIES STIPULATED PROTECTIVE ORDER" would match the first condition in the regular expression. We believe this approach performs well at identifying denials. For example, it identifies every denial in our hand-labeled dataset.338

[*180] IV. IMPUTING DATA BEFORE GO-LIVE DATES

A final step in our data generation process that requires more elaboration is how we dealt with changing patterns in how the district courts recorded their docket entry data. In a nutshell, the issue is that district courts adopted electronic filing and management on different dates. On PACER's website under court-specific pages, these dates are known as "go-live" dates.339 Although most courts adopted electronic filing and management before our study period began, some did not. For courts that did not have electronic filing and management in place, we lack confidence that the docket entries are comprehensive and accurate. We therefore treated cases filed before a district court's golive date as having missing data and imputed their outcomes.

After applying our machine learning models, we had a prediction for each filed case from 2005 to 2012 estimating three quantities: (1) total number of stipulated POs requested within two years, (2) total number of POs decided within two years, and (3) total number of stipulated POs denied within two years. For all cases filed before a district court's go-live date, we replaced all three quantities with "NA." In total, this impacted roughly 4.8 percent of our

336 Cf. Feinstein & Nou, supra note 185, at 762 n.15 (reporting an F1 of .78 and comparing to previous work). 337 As described in this Article, we aggregated into case-level measures and benchmarked them to POs within two years of case filing. 338 Notably, our regular expression errs on the side of over inclusiveness. For instance, a motion for a PO that is denied because it is moot counts as a denial. Because our overall message is about the rarity of stipulated PO denial, that suggests our approach is methodologically conservative. 339 See, e.g., California Northern District Court, PACER SERV. CTR, https://pacer.uscourts.gov/file-case/court-cmecf-lookup/court/CANDC [https://perma.cc/PF9Y-PS8C] (last updated Apr. 29, 2024) (listing an "ECF Go Live Date" of April 1, 2001 on the court-specific page for the Northern District of California).

Page 54 of 54

ARTICLE: SECRECY BY STIPULATION

cases. We then used multiple imputation methods using information on cases filed after the go-live dates to impute the NA values.340

We implemented this process using the MICE R Package,341 a well-cited and widely used imputation tool.342 In simple terms, multiple imputation uses other variables in the dataset to make reasonable predictions about the missing values using information from cases without missing values. For predictors, we used: (1) the case filing year, (2) the month of the filing year, and (3) the district court. We selected these variables for imputation because we suspected they correlated with our outcome variables and thus provided useful information for prediction purposes. After running the MICE R Package for each [*181] outcome variable, we filled in the imputed values for each missing value, which, in effect, recreated a completed dataset. We used these imputed values in constructing Figures 1 through 5 in this Article.343

Duke Law Journal

Copyright © 2024 Duke Law Journal

End of Document

340 See, e.g., Ranjit Lall, How Multiple Imputation Makes a Difference, 24 POL. ANALYSIS 414, 414-16 (2017). 341 Stef van Buuren & Karin Groothuis-Oudshoorn, mice: Multivariate Imputation by Chained Equations, COMPREHENSIVE R ARCHIVE NETWORK (June 5, 2023), https://CRAN.R-project.org/package=mice [https://perma.cc/3KPC-M6Q7]. 342 See Lall, supra note 340, at 426. 343 We opted to drop the pre-go-live date imputed values for Figures 6 and 7 because we are less confident that our multiple imputation procedure accurately predicts small strata such as case type and judge level predictions and these outcomes are sensitive to small shifts in the predicted values. In statistical terms, we simply perform conventional listwise deletion for these analyses.