[291] Articles CHARLOTTE S. ALEXANDER* AND NATHAN DAHLBERG† Procedural Localism and Access to Justice Abstract ............................................................................................ 292 Introduction ...................................................................................... 293 I. Data and Methods ................................................................. 298 A. Existing Federal Litigation Data .................................... 299 B. The SCALES-OKN Docket Sheet Corpus and Litigation Event Labels .................................................. 302 * Professor of Law and Ethics, Scheller College of Business, Georgia Institute of Technology. Co-Principal Investigator, Systematic Content Analysis of Litigation Events Open Knowledge Network (SCALES-OKN). Thanks to Nate Vogel of Community Legal Services in Philadelphia for an email exchange that sparked some of these ideas. Thanks also to the larger SCALES-OKN team and affiliated scholars, particularly Chris Cotropia, Alexandra Lahav, Roger Michalski, Adam Pah, Chris Rozolis, Dave Schwartz, Peter Siegelman, and Anne Tucker, for contributing to the work of developing the SCALES civil litigation event ontology and road-testing the resulting data. An early version of this work benefited greatly from feedback at the 2023 Emory Law School Legal Scholars’ Roundtable on Artificial Intelligence. Finally, thanks to Bethany Biswas for her able research assistance. † Data Scientist, Docket Analyzer; Deep Learning Engineer, SCALES-OKN (former). OREGON LAW REVIEW 2025 VOLUME 103 NUMBER 2 292 OREGON LAW REVIEW [Vol. 103, 291 C. Involuntary Dismissal and Settlement Corpus Assembly ....................................................................... 304 D. Feature Extraction .......................................................... 306 II. Procedural Localism at Two Litigation Moments ................ 310 A. Involuntary Dismissals .................................................. 310 1. Local Rules .............................................................. 310 2. Docket Sheet Analysis: Small Sample, Many Grounds for Dismissal ............................................. 312 3. Docket Sheet Analysis: Full Set, One Ground for Dismissal .................................................................. 315 B. Settlement ...................................................................... 322 III. Possible Causes, Implications, and Future Research Directions .............................................................................. 328 Conclusion ....................................................................................... 334 Appendix A ...................................................................................... 336 Appendix B ...................................................................................... 369 ABSTRACT Many aspects of federal civil litigation are not covered explicitly by the Federal Rules of Civil Procedure or even by court-level local rules or standing orders. Judges exercise substantial discretion to manage the cases before them; what results are pockets of hyperlocal procedure that develop largely outside any set of written rules, at the level of individual judges’ practice and preferences. Procedural localism can have fairness and access to justice implications, where a plaintiff may see their case dismissed for deficiencies by one judge but would have received many chances to correct those problems before a different judge. Even absent the threat of dismissal, localism may drive up the cost of litigation, as it creates barriers to entry and benefits repeat players with local knowledge or the resources to quickly learn the unwritten local rules of the game. However, studying procedural localism—understanding its characteristics, causes, and effects—is extremely difficult due to the absence of data that capture local variations in court practice. We begin to fill that gap with a study of procedural localism at two moments in federal civil litigation: involuntary dismissals under FED. R. CIV. P. 41(b) and judges’ handling of parties’ settlements. We examine judges’ on-the-ground practices by analyzing the text of millions of federal court docket entries, which record the litigation process as it unfolds in real time. Using machine learning classifiers 2025] Procedural Localism and Access to Justice 293 that leverage a language model that we pre-trained on eleven million docket entries and then fine-tuned to detect key litigation events, we catalog the variety of judges’ local procedures around involuntary dismissals and settlements. We draw on this work to offer some normative conclusions about procedural localism and outline directions for future computationally enabled court data research. INTRODUCTION any aspects of federal civil litigation are not covered explicitly by the Federal Rules of Civil Procedure or even by court-level local rules or standing orders. Judges exercise substantial discretion to manage the cases before them; what results are pockets of hyperlocal procedure that develop largely outside any set of written rules, at the level of individual judges’ practice and preferences.1 Procedural localism can be beneficial, as it may enable judges to calibrate local procedures to the needs of the case before them, the size and complexity of their caseloads, and the norms and expectations of the local bar.2 On the other hand, localism can have fairness and access to justice implications, where parties in different cases receive differential procedural treatment at the same moments in litigation. As the analysis presented below suggests, at the extreme, local practices may result in a case’s dismissal before one judge, whereas the case could have survived had it been randomly assigned to a different judge, and this effect may be felt most acutely by poor or self-represented litigants. Further, localism may drive up the cost of litigation, as it creates barriers to entry and benefits repeat players and participants with local knowledge or the resources to quickly learn the unwritten local rules of the game. However, studying procedural localism— understanding its characteristics, causes, and effects—is extremely difficult due to the absence of data that capture local variations in court 1 See, e.g., Chambers v. NASCO, Inc., 501 U.S. 32, 43–44 (1991) (outlining federal judges’ inherent authority to manage proceedings); Link v. Wabash R. Co., 370 U.S. 626, 630–31 (1962) (“The authority of a court to dismiss sua sponte for lack of prosecution has generally been considered an ‘inherent power,’ governed not by rule or statute but by the control necessarily vested in courts to manage their own affairs so as to achieve the orderly and expeditious disposition of cases.”). 2 See, e.g., Carl Tobias, Some Realism About Federal Procedural Reform, 49 FLA. L. REV. 49, 58 (1997) (describing courts’ creation of new local rules in the face of rising caseloads). M 294 OREGON LAW REVIEW [Vol. 103, 291 practice.3 There is no existing data set that records procedure this local; even an exhaustive review of written court rules, standing orders, and judge memoranda would not capture the sub-rule local practices that emerge in individual courtrooms. We begin to fill this gap by examining judges’ on-the-ground practices as recorded on civil cases’ docket sheets, the real-time chronological records of litigation events. Drawing on a complete set of docket sheets from all civil cases filed in federal district courts nationwide in the years 2016 and 2017 and downloaded four years later,4 we study two points in civil litigation where court rules leave management largely to judges’ discretion: involuntary dismissals under FED. R. CIV. P. 41(b) and the handling of parties’ settlements. We built machine learning classifiers that leverage a language model that we pre-trained on eleven million docket entries and then fine-tuned to label key litigation events. Using the model, we identified the relevant set of docket entries and cataloged the variety of local procedures that judges employed at these two moments in litigation—dismissals and settlements. First, Rule 41(b) allows defendants to file a motion for involuntary dismissal when the plaintiff “fails to prosecute or to comply with [the federal] rules or a court order.”5 Judges, too, may take action sua sponte under this rule to dismiss a case for lack of or delays in prosecution or 3 Id. at 50 (“I find the current situation of the federal districts disturbing for numerous identical, and some different, reasons, but the dearth of empirical data complicates evaluation.”); see also Charlotte S. Alexander & Mohammad Javad Feizollahi, On Dragons, Caves, Teeth, and Claws: Legal Analytics and the Problem of Court Data Access, in 2020 COMPUTATIONAL LEGAL STUDIES 95 (Ryan Whalen ed.) (describing court data access problems). 4 See Part I.B, infra, for a description of the data gathering process and methods. 5 FED. R. CIV. P. 41(b) (“Involuntary Dismissal; Effect. If the plaintiff fails to prosecute or to comply with these rules or a court order, a defendant may move to dismiss the action or any claim against it. Unless the dismissal order states otherwise, a dismissal under this subdivision (b) and any dismissal not under this rule—except one for lack of jurisdiction, improper venue, or failure to join a party under Rule 19—operates as an adjudication on the merits.”). Other rules and statutes might be relevant as well, including FED. R. CIV. P. 11, allowing sanctions for improper lawyer behavior; FED. R. CIV. P. 12(h)(3), requiring courts to dismiss cases sua sponte for lack of subject matter jurisdiction; 28 U.S.C. § 1915(e)(2), requiring dismissal of deficient complaints filed in forma pauperis. Local procedures may develop around each; we return to these possibilities at the close of this Article. Further, a parallel analysis might be done around default judgments against defendants under FED. R. CIV. P. 55, though that rule provides substantially more procedural guidance than Rule 41(b). 2025] Procedural Localism and Access to Justice 295 rule violations.6 However, the rule is silent as to the particular violations that can trigger dismissal, the options available short of dismissal, and the procedure, if any, that should govern the dismissal process. Case law fills in the blanks to some extent, supplying lists of factors that courts should consider before deciding on dismissal, though case law does not tend to prescribe the particular procedures that courts should use or offer alternatives to dismissal.7 Some local court rules, standing orders, and individual judges’ own personal procedural manuals supply additional guidance, though most such rules grant a similar level of discretion to judges with respect to the grounds for and procedures surrounding involuntary dismissals.8 Thus, judges are left largely to invent their own processes.9 The stakes here are high. How a judge defines the contours of a dismissible deficiency, whether the judge chooses dismissal or a warning, how many warnings the judge issues, and how much time the judge allows to correct a problem are all consequential questions. A pro se litigant who fails to pay the filing fee or ask for a waiver might 6 Link, 370 U.S. at 630–31 (“The authority of a court to dismiss sua sponte for lack of prosecution has generally been considered an ‘inherent power,’ governed not by rule or statute but by the control necessarily vested in courts to manage their own affairs so as to achieve the orderly and expeditious disposition of cases.”). 7 See, e.g., LeSane v. Hall’s Sec. Analyst, Inc., 239 F.3d 206, 209 (2d Cir. 2001) (requiring district courts to consider “[1] the duration of the plaintiff’s failures, [2] whether plaintiff had received notice that further delays would result in dismissal, [3] whether the defendant is likely to be prejudiced by further delay, [4] whether the district judge has take[n] care to strik[e] the balance between alleviating court calendar congestion and protecting a party’s right to due process and a fair chance to be heard . . . and [5] whether the judge has adequately assessed the efficacy of lesser sanctions” prior to dismissing under Rule 41(b)) (quoting Alvarez v. Simmons Mkt. Rsch. Bureau, Inc., 839 F.2d 930, 932 (2d Cir. 1988)). 8 Relevant local rules and standing orders, gathered from the Westlaw Federal Court Local Rules repository and the Westlaw Directory of Federal Court Guidelines, appear in Appendix A, covering involuntary dismissals, and Appendix B, covering settlements. Local rules and other court- and judge-level written rules are discussed in more detail in Part II, infra. 9 Here, we follow Jay Tidmarsh in equating “procedure” with “process,” meaning the “set of actions (filing and answering the complaint, taking depositions, writing motions) that move the case from its beginning to its end.” Jay Tidmarsh, Procedure, Substance, and Erie, 64 VAND. L. REV. 877, 884 (2011). This is an on-the-ground, functional definition that is concerned with granular, practice- and process-level questions: Does a judge dismiss first and allow reopening or keep a case open and threaten dismissal? How many days does a party have to cure a deficiency? What is a party required to file? Et cetera. 296 OREGON LAW REVIEW [Vol. 103, 291 be given fourteen days to pay under threat of dismissal by one judge,10 whereas other judges, even in the same judicial district,11 might allow twenty12 or thirty13 days in the same circumstance. The particular judge to whom the plaintiff’s case is assigned dictates the applicable procedure, which, in turn, determines the plaintiff’s precarity—how easy or hard it is to be kicked out of court. The second example—settlement procedures—is similar. Though FED. R. CIV. P. 41(a) sets out the process for plaintiff’s to voluntarily dismiss their claims, with or without a court order, judges do not universally require that parties use that rule’s procedures to terminate a case upon reaching settlement. Instead, judges have developed a wide variety of local practices when informed of a case’s resolution, ranging from immediate dismissal with an option to reopen within a given time period,14 to leaving the case open with instructions to the parties to file 10 See, e.g., Order and Notice of Deficiency, Marshek v. GM Fin., No. 4:17-CV-00459 (N.D. Tex. June 7, 2017), Docket Entry No. 6 (“Order and Notice of Deficiency: Tyson Marshek must address the following deficiency: Plaintiff has not paid the filing and administrative fees nor has he filed a long-form application to proceed in forma pauperis. Failure to comply with this order may lead to dismissal for failure to prosecute pursuant to FRCvP 41(b). Deadline to cure the deficiency is 14 days. (Ordered by Judge John McBryde on 6/7/2017) (Attachments: # 1 IFP) (tln) (Entered: 06/07/2017).”). 11 The U.S. District Court for the Northern District of Texas has no local rule, standing order, or other authority that we could locate that governs FED. R. CIV. P. 41(b) involuntary dismissals. Civil Rules, U.S. DIST. CT. N. DIST. OF TEX., https://www.txnd.uscourts.gov /civil-rules [https://perma.cc/U9N4-J2JF] (last visited Feb. 1, 2024) (stating that Local Rule 41.1 “Order of Dismissal” is repealed). 12 See, e.g., Order and Notice of Deficiency, Alexander v. U.S. Bank Nat’l Ass’n, No. 4:16-CV-00181 (N.D. Tex. Mar. 4, 2016), Docket Entry No. 3 (“Order and Notice of Deficiency: Timothy Paul Alexander must address the following deficiency: Pltf has not paid the filing and admin fees nor submitted IFP application[.] Failure to comply with this order may lead to dismissal for failure to prosecute pursuant to FRCvP 41(b). Deadline to cure the deficiency is 20 days. Clerk shall mail true copy of this order to pltf with IFP long form application. (Ordered by Magistrate Judge Jeffrey L Cureton on 3/4/2016) (wrb)Order and long form mailed to Timothy Alexander 3/4/16 (Entered: 03/04/2016).”). 13 See, e.g., Order and Notice of Deficiency, Mayfield v. Harmon, No. 3:17-CV-00509 (N.D. Tex. Feb. 24, 2017), Docket Entry No. 4 (“Order and Notice of Deficiency: Cody Mayfield must address the following deficiency: Petitioner has not paid the filing fee or submitted an application to proceed in forma pauperis. Failure to comply with this order may lead to dismissal for failure to prosecute pursuant to FRCvP 41(b). Deadline to cure the deficiency is 30 days. (Ordered by Magistrate Judge Irma Carrillo Ramirez on 2/24/2017) (Attachments: # 1 IFP Form) (rekc) (Entered: 02/24/2017).”). 14 See, e.g., Order, Reed v. Wyndham Worldwide Operations, Inc., No. 2:16-cv-00013 (M.D. Tenn. June 2, 2016), Docket Entry No. 6 (“ORDER: The Court having been advised by counsel for the parties that this action has been settled, it is hereby ORDERED that the action is dismissed without prejudice to the right, upon good cause shown within sixty (60) days, to reopen the action if the settlement is not consummated. Within this sixty-day period, 2025] Procedural Localism and Access to Justice 297 dismissal papers,15 to “administrative closings” that may or may not automatically convert to dismissals on a future date.16 These practices are hidden from the outsider or first-time litigant or lawyer who hopes to learn the rules of federal litigation ex ante. And because they can be judge- or even case-specific, they vary widely. Thus, even in the settlement example, where the parties want dismissal, the proliferation of procedures can impede predictability and fairness and drive up costs. Civil procedure scholar Paul Carrington summarizes this critique, describing highly localized procedures as “legal clutter” that “gives undue advantage to cognoscenti” such as local lawyers and repeat players, to the detriment of lawyers who make only “episodic appearances” in any given federal court.17 Further, “[a]s clutter increases, the cost of legal services is also increased by a the parties may submit a proposed agreed order of compromise and dismissal. Signed by Chief Judge Kevin H. Sharp on 6/2/16. (dt) (Entered: 06/02/2016).”). 15 Compare, e.g., Order, Ramirez v. State Farm Lloyds, No. 5:17-cv-00058 (S.D. Tex. Apr. 19, 2017), Docket Entry No. 5 (“ORDER - In light of the Parties’ resolution of this matter, the Court STAYS the April 26, 2017 Pretrial Conference and Trial Scheduling Conference. - The parties have sixty days to finalize their Settlement Agreement and submit a Joint Stipulation of Dismissal pursuant to F. R. Civ. Proc. 41(a)(1)(A)(ii), or file an Advisory explaining why settlement was not finalized. The Court appreciates the Parties’ prompt notification of settlement. (Signed by Magistrate Judge Diana Song Quiroga) Parties notified.(mmarquez, 5) (Entered: 04/19/2017).”), with Order, Duferco Steel, Inc. v. M/V BBC NILE, No. 4:16-cv-00483 (S.D. Tex. Oct. 13, 2016), Docket Entry No. 11 (“ORDER on Notice of Settlement (Closing Papers due by 11/10/2016), Deadlines terminated (Signed by Judge Sim Lake) Parties notified. (aboyd, 4) (Entered: 10/13/2016).”). 16 See, e.g., 60 Day Order, Brown v. ACE Am. Ins. Co., No. 3:16-cv-01035 (S.D. Ill. Aug. 28, 2017), Docket Entry No. 24 (“60 DAY ORDER: The undersigned District Judge has been advised by defense counsel (with plaintiff’s consent) that this matter has been settled in its entirety, but that the parties need additional time to finalize settlement documents. Accordingly, the undersigned District Judge DIRECTS the Clerk of Court, 60 days after this Order is docketed (October 30, 2017), to ENTER JUDGMENT OF DISMISSAL with prejudice. Each party shall bear his own costs, unless otherwise provided in the settlement documents. If the parties fail to finalize the settlement within the 60-day period, they may—before that period expires—move to postpone entry of judgment to a later date. Due to the settlement, the Court CANCELS all future settings, including the October 20, 2017 pretrial conference, and the November 13, 2017 jury trial. Signed by Chief Judge Michael J. Reagan on 8/28/2017. (rah) THIS TEXT ENTRY IS AN ORDER OF THE COURT. NO FURTHER DOCUMENTATION WILL BE MAILED. (Entered: 08/28/2017).”). 17 Paul D. Carrington, A New Confederacy? Disunionism in the Federal Courts, 45 DUKE L.J. 929, 947–48 (1996); see also Glenn S. Koppel, Toward a New Federalism in State Civil Justice: Developing a Uniform Code of State Civil Procedure Through a Collaborative Rule-Making Process, 58 VAND. L. REV. 1167, 1182 (2005) (“Inter-federal district court disuniformity complicates federal litigation, increasing cost and delay in the administration of civil justice.”). 298 OREGON LAW REVIEW [Vol. 103, 291 diminution of competition and retention of redundant counsel.”18 Standardization benefits outsiders, newcomers, and low-resourced participants in a system; lack of standardization benefits insiders, repeat players, and those with enough resources to learn quickly.19 Beyond the involuntary dismissal and settlement case studies, we engage throughout the Article with an additional, overarching example of localism: the wide variation in the ways that different judges and clerks choose to record their practices. These masses of inconsistent textual descriptions require the development and deployment of time- and resource-intensive computational methods just to answer basic questions about litigation, such as which types of deficiencies result in involuntary dismissal. Local variations in recording can thus obscure local variations in procedure, meaning that litigants, lawyers, researchers, policymakers, and judges themselves cannot effectively see what the courts are doing in practice, much less assess the relative benefits and harms of procedural localism. * * * The Article is structured as follows. In Part I, we describe the shortcomings of existing federal litigation data sources and introduce our docket sheet corpus and the methods we applied to examine local procedures around involuntary dismissals and settlements. In Part II, we present our results. Part III considers possible causes for the procedural localism that we observe, suggests ideas for additional moments in litigation where procedural localism might be found, and outlines directions for future computationally enabled court data research. I DATA AND METHODS This Part briefly describes the current state of federal litigation data before introducing our project’s docket sheet corpus, drawn from the Systematic Content Analysis of Litigation Events Open Knowledge Network (SCALES-OKN or SCALES). The SCALES project is a multimillion dollar, multiuniversity effort funded by the National Science Foundation and private foundations to amass court records, build computational tools to extract key information from those records, and make the records, the tools, and the output publicly 18 Carrington, supra note 17, at 948. 19 See generally Brenner M. Fissell, Against Criminal Law Localism, 81 MD. L. REV. 1119 (2022) for a related discussion in the context of criminal law and procedure. 2025] Procedural Localism and Access to Justice 299 available.20 Alexander is a Co-Principal Investigator of the SCALES project and leads the civil litigation ontology working group; Dahlberg formerly worked as a SCALES deep learning engineer and specializes in natural language processing and machine learning. A. Existing Federal Litigation Data The sole existing administrative data source on federal civil litigation is the Federal Judicial Center’s (FJC) Integrated Database (IDB).21 The IDB records a set of features describing each case filed in federal court from 1970 to the present. Staff from each of the ninety- four U.S. district courts report these data to the Administrative Office of the U.S. Courts, which then provides “regular updates” to the FJC.22 The FJC then “post-processes” the data into the form that is publicly available in the IDB.23 Judge names are left out of the IDB under the terms of a data sharing agreement between the FJC and the Judicial Conference of the United States.24 For cases filed in 1988 through the present, included features are circuit, district, office, docket number, origin, filing date, jurisdiction, nature of suit, title, section, subsection, diversity residence, jury demand, class action, monetary amount demanded, county of residence, multidistrict litigation docket number, plaintiff, defendant, transfers, termination date, procedural progress, disposition, nature of judgment, amount received, judgment, pro se, and fee status.25 20 See SCALES, https://scales-okn.org [https://perma.cc/B5GB-NANG] (last visited Feb. 1, 2024). 21 Integrated Database (IDB), FED. JUD. CTR., https://www.fjc.gov/research/idb [https:// perma.cc/5UBG-E3UW] (last visited Feb. 1, 2024) (“The IDB contains data on civil case and criminal defendant filings and terminations in the district courts, along with bankruptcy court and appellate court case information. . . . The FJC receives regular updates of the case- related data that are routinely reported by the courts to the AOUSC. The FJC then post- processes the data, consistent with the policies of the Judicial Conference of the United States governing access to these data, into a unified longitudinal database, the IDB.”). 22 Id. 23 Id. 24 The Integrated Database: A Research Guide, FED. JUD. CTR. 1, https://www.fjc.gov /sites/default/files/IDB-Research-Guide.pdf [https://perma.cc/97Y2-RHXU] (last visited Feb. 1, 2024) (“[I]nformation on the judge or judges presiding over the case is redacted pursuant to Judicial Conference policy.”). 25 Integrated Data Base Civil Documentation, Field Descriptions, FED. JUD. CTR., https://www.fjc.gov/sites/default/files/idb/codebooks/Civil%20Codebook%201988%20 Forward%2010252023.pdf [https://perma.cc/YCK7-7X5T] (last visited Nov. 10, 2024). 300 OREGON LAW REVIEW [Vol. 103, 291 Many of these fields, however, are missing data or are used differently across the ninety-four judicial districts. For example, Alexander, co-author of this Article, found the following in a previous study of federal labor and employment lawsuits: Nine [IDB fields] had missing data for ninety-five percent or more of the lawsuits, including variables meant to capture whether the case was filed as a class action, variables relevant to transferred cases, and variables relevant to arbitrated cases. Other variables were missing values for about half of the records, including those that capture the party in whose favor final judgment was entered, and whether that judgment included a monetary award, injunctive relief, and/or attorneys’ fees and costs.26 Similarly, Gillian Hadfield documented “error rates of as much as 69 percent” in IDB disposition codes covering “nontrial adjudications,” such as dismissals.27 Theodore Eisenberg and Margo Schlanger further discovered erroneous judgment amounts recorded in 41% of a sample of tort cases and 29% in prisoner civil rights cases.28 Hadfield also points to conceptual disagreement about which categories of on-the-ground litigation events fall into which IDB recordkeeping categories. She explains: I found substantial rates of “Type 2” errors, that is, cases in which a disposition we are interested in counting—such as settlement, which we would expect to show up in the “dismissed: settled” category— shows up in other codes such as “dismissed: other” or “judgment on motion before trial.”29 Indeed, if a plaintiff moves for dismissal under FED. R. CIV. P. 41(a)(2) due to a settlement, and if the judge grants that motion, one can imagine a clerk choosing the “judgment on a motion before trial” option, though a judgment might not technically enter in that case. Likewise, if the plaintiff uses the Rule 41(a) voluntary dismissal vehicle to close a case due to settlement but does not specify settlement as the reason for 26 Charlotte S. Alexander & Mohammad Javad Feizollahi, Decisional Shortcuts and Selection Effects: An Empirical Study of Ten Years of U.S. District Courts’ Employee Misclassification Decisions, U.S. DEP’T OF LAB. 8 (Dec. 1, 2020), https://www.dol.gov /sites/dolgov/files/OASP/evaluation/pdf/LRE_Alexander-DecisionalShortcutsandSelection Effects_December2020.pdf [https://perma.cc/X62C-6RZH]. 27 Gillian K. Hadfield, Where Have All the Trials Gone? Settlements, Nontrial Adjudications, and Statistical Artifacts in the Changing Disposition of Federal Civil Cases, 1 J. EMPIRICAL LEGAL STUD. 705, 711 (2004). 28 Theodore Eisenberg & Margo Schlanger, The Reliability of the Administrative Office of the U.S. Courts Database: An Initial Empirical Analysis, 78 NOTRE DAME L. REV. 1455, 1474–75 (2003). 29 Hadfield, supra note 27. 2025] Procedural Localism and Access to Justice 301 closing—and there is no requirement in Rule 41 that a plaintiff do so— then the clerk may have no other disposition coding option than “dismissed: other” from which to choose. Thus, as David Colarusso and Erika Rickard summarize about court data generally, the IDB suffers from a “lack of clearly-defined judicial data standards” in determining both what counts as the events of litigation and how to count them.30 Finally, the IDB’s coding of litigation events does not record all the procedures that are established by the Federal Rules, much less the additional on-the-ground variations enacted at the local level that are the subject of this Article. Several values of the IDB’s “disposition” codes might at least be helpful in locating the universe of cases relevant to the present study: “want of prosecution,” “voluntarily [dismissed],” “settled,” and “consent,” along with the generic “other” dismissal category examined by Hadfield. Given the missing IDB data and inconsistent coding noted above, however, we rely instead on the text of the docket sheets, which are artifacts of the litigation process created in real time by the participants themselves during litigation. Though docket sheets present their own challenges—not least the vast inconsistency in the language that is used to describe the same events31—we view them as a more reliable, and more granular, record of the actual conduct of litigation than the IDB data, which is generated by administrative staff several steps removed.32 30 David Colarusso & Erika J. Rickard, Speaking the Same Language: Data Standards and Disruptive Technologies in the Administration of Justice, 50 SUFFOLK U. L. REV. 387, 387 (2017). 31 Any document, including electronic docket sheets, cost ten cents per page to download from the federal courts’ Public Access to Court Electronic Records (PACER) system. This is the reason that the SCALES project assembled a corpus of docket sheets and not the underlying documents filed by the parties and issued by the judges. Cost is a major barrier to the type of research presented here, which required hundreds of thousands of dollars of federal grant money just to assemble the corpus that is the subject of this study. That taxpayer dollars were awarded by one branch of government, an executive funding agency, to be paid to another, the judiciary, to buy public documents, is the product of an inefficient federal court data and records system that is not designed for systemic study and desperately in need of overhaul. Adam R. Pah et al., How to Build a More Open Justice System, SCI., July 10, 2020, at 134, https://par.nsf.gov/servlets/purl/10170352 [https://perma.cc/L5X7 -QNXZ]. 32 Indeed, the Hadfield and Eisenberg and Schlanger studies cited above used docket sheets as the ground truth against which to measure the IDB data. See also Pauline Kim et al., How Should We Study District Judge Decision-Making?, 29 WASH. U. J.L. & POL’Y 83, 86 (2009) (“By taking advantage of the electronic docketing system now operating in all 302 OREGON LAW REVIEW [Vol. 103, 291 B. The SCALES-OKN Docket Sheet Corpus and Litigation Event Labels The civil docket sheets that form the basis of this study were drawn from the SCALES corpus, which contains all criminal and civil cases filed nationwide in the years 2016 and 2017, downloaded four years later. The research described in this Article uses the docket sheets from just the civil cases: approximately 550,000 docket sheets containing nearly 17 million docket entries. The SCALES team, of which we are a part, developed a set of standardized labels to apply to the raw docket sheet text to identify the events that happened in each case during litigation. We introduce the particular labels that are relevant to the present Article in more detail in the sections below. Here, we describe the label-generation process more broadly. Because judges and clerks across the ninety-four districts use different language to describe the same litigation events, our labeling process could not rely on rigid rules.33 For example, the rule “Label any docket entry as a settlement that contains the word settlement” would capture a docket entry that says, “The parties report that they have reached a settlement,” but not, “The parties report that they have reached agreement on their claims.” The rule would thus be underinclusive, but likely also vastly overinclusive, as many docket entries, such as scheduling orders, mention the word “settlement” but do not reflect that settlement has happened. We therefore developed a set of machine learning classifiers that leveraged a language model trained on general English language usage and also exposed the set to the specific legal language used in docket entries. We fine-tuned the classifiers to effectively group disparately worded docket entries into meaningful standardized labels, solving the underinclusivity problem described above. Our approach can also distinguish between substantively meaningful docket entries and mentions in passing, solving the overinclusivity problem as well. Our starting point in this process was a publicly available model developed by Microsoft researchers called “large DeBERTaV3” that federal district courts, researchers can use dockets, orders, and other case documents, as well as opinions, as data sources, thereby incorporating into their analysis the relevant institutional features of district courts.”). 33 When judges or clerks enter a litigation event onto a docket sheet, they first choose among fixed options in a drop-down menu and then add additional free text. The drop-down menu options are consistent within districts but not across them; the free text is often inconsistent at the district or even individual judge level. 2025] Procedural Localism and Access to Justice 303 has been exposed to (“trained on”) massive quantities of text, including all English-language Wikipedia entries and BookCorpus, a dataset consisting of 11,038 unpublished books.34 To enable the model to work effectively on docket entry text, we further trained it on 11 million docket entries.35 This improved our model’s performance when confronted with legal language that is used differently on docket sheets than in general usage—“settlement” being more strongly associated with the resolution of legal claims, for example, than the establishment of a community of people—and also embedded in our model an awareness that “settlement” is a synonym for “agreement” or “resolution” within the docket sheet context. We then further trained the model by exposing it to sets of docket entries that we had manually annotated with the appropriate litigation event label. For example, the disparate docket entry text described above—all of which referred to the parties’ settlement—would all receive the “settlement” label. Using these labeled docket entries, a model is able to learn the patterns of language associated with each label and classify new, previously unseen docket entry text with the 34 SCALES-OKN Docket Language Model, HUGGING FACE, https://huggingface.co /scales-okn/docket-language-model [https://perma.cc/BPF4-NWJD] (last visited Jan. 2, 2025); DeBERTaV3: Improving DeBERTa using ELECTRA-Style Pre-Training with Gradient-Disentangled Embedding Sharing, HUGGING FACE, https://huggingface.co /microsoft/deberta-v3-large [https://perma.cc/6G6T-Q78J] (last visited Jan. 2, 2025) (DeBERTaV3 consists of 304 million parameters, or aspects of text that the model has “learned” via training, and 128,000 subword tokens, or the fragments of text that the model ingests during the training process); Pengcheng He et al., DeBERTaV3: Improving DeBERTa Using ELECTRA-Style Pre-Training with Gradient-Disentangled Embedding Sharing, ARXIV (Mar. 23, 2023), https://arxiv.org/abs/2111.09543 [https://perma.cc/536C -49UU]. For a description of the training process for language models more generally, see Jacob Devlin et al., BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding, ARXIV (May 24, 2019), https://arxiv.org/abs/1810.04805 [https://perma .cc/FUN8-DAG3]. 35 Specifically, we used a technique known as masked language modeling (MLM). This task involves randomly replacing about 15% of tokens in docket entries with a special “mask” token and then training the model to predict the original text from the masked text. This technique is commonly used to adapt models that have been pre-trained on general language corpora to more narrow domains such as legal language. By training the model to “fill in the blank” across large collections of text, the model learns a probability distribution that reflects which terms are likely to appear given different surrounding language, and which is specific to the text on which it was trained. For other examples of domain-specific language models, see Jinhyuk Lee et al., BioBERT: A Pre-Trained Biomedical Language Representation Model for Biomedical Text Mining, 36 BIOINFORMATICS 1234 (2020) and Iz Betalgy et al., SciBERT: A Pretrained Language Model for Scientific Text, ARXIV (Sept. 10, 2019), https://arxiv.org/abs/1903.10676 [https://perma.cc/9C37-WY99], which were developed for use in the sciences. 304 OREGON LAW REVIEW [Vol. 103, 291 appropriate label.36 For each litigation event label, we manually annotated between 1,000 and 4,000 docket entries, with more docket entries for the labels, such as settlement, that were associated with greater variation in language. We improved model performance iteratively through a process of training the model, assessing performance, identifying docket language where the model failed to apply the correct label, sampling additional text with this language and adding it to the training set, and repeating. Once the model reached satisfactory performance both in terms of its precision (e.g., the percentage of the entries the model predicted to be indicators of settlement that were actual indicators of settlement) and its recall (e.g., the percentage of actual indicators of settlement that the model correctly predicted), we had confidence that the model would accurately generalize to the broader set of docket entries outside of the data on which it was trained. C. Involuntary Dismissal and Settlement Corpus Assembly To identify the subset of cases relevant to our study of local procedure around involuntary dismissal and settlement, we filtered the 2016–2017 corpus to include only those cases that contained docket entries with the standardized SCALES labels for court action (e.g., court notices, orders, minute and similar docket-only entries, and judgments). We further filtered the court action docket entry set as follows: First, we identified the court actions that contained some indicator of a Rule 41(b) involuntary dismissal. Second, we took a sample of approximately 250 docket entries, from which we extracted a set of common deficiencies that courts specified as the basis for dismissal. These results are reported in Part II.A below. At the same time, we reviewed all ninety-four district courts’ local rules to identify local grounds and procedures, if any, for involuntary dismissal.37 Informed by the docket entry and local rule reviews, we narrowed the scope of the involuntary dismissal inquiry to focus on a single ground for dismissal: pro se plaintiffs’ failure to update their address with the court in a timely fashion, interpreted by courts as a failure to prosecute and therefore dismissible under Rule 41(b) and some local rules. We did so for manageability, given the wide range of grounds for 36 For the full set of litigation event labels, see Charlotte Alexander, Litigation Ontology, SCALES (Dec. 12, 2024), https://docs.scales-okn.org/guide/ontology/ [https://perma.cc /4MY7-SQZQ]. 37 See Part II.A, infra. 2025] Procedural Localism and Access to Justice 305 dismissal that courts identify when using Rule 41(b) for dismissal. Further, by focusing exclusively on pro se address updates, we can hold the ground for dismissal constant and analyze all the ways in which local procedures vary. Deploying these conditions across our entire set of involuntary dismissal docket entries, we identified 1,645 docket entries from 1,389 cases in which judges used the Rule 41(b) framework to address this particular dismissible deficiency in a pro se plaintiff’s conduct of litigation. We used a similar process to locate settlement-relevant docket entries. We were specifically interested in finding the entries in which a judge noted that one or more parties had informed the court of settlement and then instructed the parties about next steps. Our intent was to locate the set of cases in which the parties reached settlement on their own, outside any court-mandated or court-annexed dispute resolution program and without any particular statutory or other set of rules that dictate how a settlement would be processed by the court. For example, some court-annexed mediation programs require that the parties or the mediator report settlement to the judge within a prescribed time period and then mandate a subsequent set of party filing requirements and judge actions.38 Likewise, settlements of some FED. R. CIV. P. 23 class actions, Fair Labor Standards Act cases, and minors’ claims all trigger specific statutory, case law, or rule-dictated procedural steps.39 Our interest was in the procedures that judges adopt in the absence of such dictates; we therefore generated a corpus of docket entries in which settlement appeared to have occurred in a procedure-free zone, and observed the judges’ actions in response. We note here, as we did above, that FED. R. CIV. P. 41(a) allows the plaintiff to voluntarily dismiss claims, either with or without a court order, depending on how far the litigation has progressed.40 Rule 41(a) does set out procedure for the parties and judge to follow. Voluntary dismissal under Rule 41(a), however, may not always be the result of a 38 See Deborah Thompson Eisenberg, What We Know and Need to Know about Court- Annexed Dispute Resolution, 67 S.C. L. REV. 245, 254–55 (2016) (describing a court- annexed alternative dispute resolution program in Maryland). 39 See, e.g., FED. R. CIV. P. 23(e) (establishing a special settlement procedure for certain types of class action); Lynn’s Food Stores, Inc. v. United States, 679 F.2d 1350 (11th Cir. 1982) (requiring judicial approval of settled Fair Labor Standards Act claims); KAREN L. STEVENSON & JAMES E. FITZGERALD, RUTTER GROUP PRACTICE GUIDE: FEDERAL CIVIL PROCEDURE BEFORE TRIAL, § 15:138 (2008-2024) (“Federal courts generally require that claims by minors . . . be settled in accordance with applicable state law. California law requires court approval of the fairness and terms of the settlement.”). 40 FED. R. CIV. P. 41(a). 306 OREGON LAW REVIEW [Vol. 103, 291 settlement; likewise, settlement need not be formalized via the Rule 41(a) framework. Our corpus assembly methods therefore focused on the parties’ informing the court of settlement and the courts’ responses, not the parties’ filing of notices, stipulations, or motions under Rule 41(a). Some judges may respond to a settlement notification by requiring the parties to proceed under Rule 41(a), perhaps within a certain time period, and some may not. This is one possible site of procedural variation. To construct our corpus within these parameters, we began, as above, with the set of docket entries tagged with SCALES’ court action label and filtered them to identify entries reflecting party notification of settlement. This process generated a set of 13,897 docket entries from the same number of cases.41 As we explain further below, while our involuntary dismissal corpus contains multiple docket entries per some cases, enabling us to track how many chances plaintiffs had to cure before dismissal, the settlement corpus is restricted to one entry per case, because our interest is in the procedural process that the court establishes upon notification of settlement—a single moment in litigation. D. Feature Extraction Having assembled the involuntary dismissal and settlement corpora as outlined above, we next tackled the task of extracting key pieces of information from each docket entry. These variables, also known as “features” in machine learning, describe the procedures that judges employ at these two moments in litigation. For each case, we also had 41 We dropped from the corpus 10,465 duplicate entries describing settlement procedure from cases associated with multidistrict litigation managed by the U.S. District Court for the Southern District of West Virginia. Because these were filed in many cases but reflected a single underlying settlement, we deemed them duplicative of one another. See, e.g., Inactive Docket Order, Bennett v. Boston Scientific Corp., No. 2:16-cv-00001 (S.D. W. Va. May 17, 2017), Docket Entry No. 6 (“INACTIVE DOCKET ORDER: The court has been advised that the plaintiff(s) and Boston Scientific Corporation have reached a settlement with regard to BSC. All discovery deadlines are continued until further order of the court. The Clerk is directed to retire this case from the active docket. Plaintiff(s) and BSC may submit an agreed order of dismissal with prejudice on or before 11/2/2017; if settlements are not finalized and dismissal orders are not submitted by then the Court will have a hearing to determine the appropriate action pertaining to any remaining cases on the inactive docket. Counsel for plaintiffs and defendant(s) are directed to provide quarterly reports as to their progress in dismissing cases on the inactive docket. The court will reinstate the case to the active docket if one of the parties shows good cause for such reinstatement. This Order is related to the Proposed Inactive Docket Order filed at ECF No. 2729 in MDL Case No. 2326 Boston. Signed by Judge Joseph R. Goodwin on 5/17/2017. (cc: attys; any unrepresented party) (kp).”). 2025] Procedural Localism and Access to Justice 307 a set of case-level features harvested from the docket sheets: court, party names, case filing and termination dates, pro se status, lawyer names, and nature of suit code.42 Finally, we linked each case in our corpus to the IDB and recorded the IDB “disposition code” for each. Though we do not rely on these codes for any conclusions, given the documented problems with coverage and quality, we nevertheless include them as an extremely rough quality control measure, with several caveats noted below. Thus, for the involuntary dismissal set, we assembled the following features: (1) The court in which the case was filed; (2) The judge or judges assigned to the case;43 (3) Whether the judge dismissed the case immediately or issued a warning about future dismissal;44 (4) For dismissals only: whether the judge gave the plaintiff a window of time in which to cure and reopen the case and, if so, how many days;45 42 The present analysis does not make use of this full set of additional features, though they may be fodder for interesting future research. 43 More specifically, we extracted judge names from the particular docket entries at issue in addition to using the case-level judge name field directly from the federal courts’ PACER system. This is because a case may be reassigned midstream away from the judge or magistrate listed in PACER, and the PACER field updates, but does not retain the history of previously assigned judges. The PACER judge field can therefore be an unreliable indicator of the judge who actually took action at any previous point in the case. To address this problem, the SCALES team developed a custom judge entity recognition and disambiguation pipeline that can identify the names of judges, including magistrates, from docket text and standardize them, such that “Judge Tanya Walton Pratt” from the example above is identified as the particular judicial actor associated with that docket entry and receives the same unique identifier within the SCALES corpus as other formulations of her name, e.g., “Judge Tanya W. Pratt” or “The Honorable Judge Walton Pratt.” Adam R. Pah et al., PRESIDE: A Judge Entity Recognition and Disambiguation Model for US District Court Records, IEEE XPLORE (Jan. 13, 2022), https://ieeexplore.ieee.org/document /9671351 [https://perma.cc/Y6WK-CZMS]. 44 Here, we developed methods to distinguish, for the involuntary dismissal set, dismissal entries from warnings and, for the settlement set, immediate and administrative dismissals from cases that remained open. To do this we first used an existing model from the Python package spaCy to split docket entry text into individual sentences. Sentencizer, SPACY, https://spacy.io/api/sentencizer [https://perma.cc/4MU2-GUJN] (last visited Feb. 1, 2024) (describing sentence boundary location procedure). We then created and applied a set of conditions based on the language in each sentence to identify indicators of each of these entry types (e.g., dismissal versus warning). 45 Identifying time windows and their length required some cleaning and standardization, as dates and time windows appear in a variety of formats in docket entry text. Some deadlines, as in the docket entry in the main text above, are fully written out, 308 OREGON LAW REVIEW [Vol. 103, 291 (5) For warnings only: whether the judge gave the plaintiff a window of time in which to cure and avoid dismissal and, if so, how many days; (6) For cases with more than one relevant entry, the total number and sequence of warnings and dismissals; and (7) IDB disposition code. Taken together, these features allow us to suggest answers to the kinds of questions identified in the Introduction: How do judges define the contours of a dismissible deficiency, in the specific context of pro se address updating, given the latitude that Rule 41(b) grants? How often do judges choose dismissal versus a warning, how many warnings do judges give, and how much time do they allow to correct a problem? How do these answers vary within and across judges and within and across courts? For the settlement set, we extracted the following, using the same procedures as above: (1) The court in which the case was filed; (2) The judge or judges assigned to the case; (3) The action the judge took upon notice of settlement; (4) Immediate dismissal; (5) Statistical/administrative closing; (6) Case remained open; (7) For immediate dismissals and statistical/administrative closings only: whether the judge gave the parties a window of time in which to reopen the case or alter the terms of the dismissal and, if so, how many days; whereas others, as in the “Entered” date above, appear as MM/DD/YYYY. In other docket entries, a judge gives a party “30 days” or “thirty days,” for example, rather than listing a date. To calculate the length of time windows consistently across docket entries, we extracted all date entities using a language model that we fine-tuned on a named entity recognition (NER) dataset that we created to extract attachments, case numbers, citations, dates, monetary amounts, and names specifically from docket entries. We then created a set of heuristics and rules based on the language surrounding the extracted dates to determine which dates represented the ‘Entered’ date and which represented a court deadline for the party. We could then standardize the dates and compute the difference between the deadline and the ‘Entered’ date. We manually reviewed extreme outliers to ensure accuracy. See Named Entity Recognition, SPACY, https://spacy.io/usage/linguistic-features #named-entities [https://perma.cc/MBB4-9HDJ] (last visited Feb. 1, 2024) (discussing date tagging). 2025] Procedural Localism and Access to Justice 309 (8) For cases that remained open: whether the judge gave the parties a window of time in which to make a filing and/or take other action to dismiss the case and, if so, how many days; and (9) IDB disposition code. As above, these features, collectively, give us a picture of the variety of ways that judges handle settlement: where they set the default (dismissal, open, or some in-between status); whether the parties have the option to change the initial status; and how much time the parties have to execute any optional or required steps. Interestingly, though in the Introduction we presented judges’ handling of settlement as a procedurally varied, but relatively low- stakes, moment in litigation, our review of these docket entries revealed an unexpected (to us, at least) overlap between settlement and Rule 41(b) involuntary dismissals. For example, in the following docket entry from the U.S. District Court for the Southern District of Indiana, the court left the case open, but stayed deadlines, required a filing of “the appropriate dismissal papers” within thirty days, and threatened dismissal with prejudice under Rule 41(b) for noncompliance: ORDER - The Court has been advised by Plaintiff that a settlement has been reached in this action (Dkt. No. 8). It is therefore ordered that all dates and deadlines previously established are vacated. Any pending motion is denied as moot. Counsel is directed to proceed to finalize the agreement to resolve these claims, and to file the appropriate dismissal papers with the Clerk of Court on or before August 8, 2016. Failure to do so will result in dismissal with prejudice pursuant to Fed. R.Civ. P. 41(b). Signed by Judge Tanya Walton Pratt on 7/8/2016.(JLS) (Entered: 07/08/2016).46 Thus, there may be more at stake with respect to judges’ localized handling of settlement than benefits to the cognoscenti, a decrease in transparency, and an increase in cost. Parties might face the dismissal of their claims in a form chosen by the judge, with prejudice, rather than a form that they negotiated amongst themselves. The choice by some judges, but not all, to back up their settlement handling procedures with the threat of Rule 41(b) dismissal brings this Article’s two examples closer to one another in terms of the impact of procedural localism on the relative precarity of the parties’ place in court. 46 Order, Nagrodski v. Roache, II, No. 1:16-cv-00901 (S.D. Ind. July 8, 2016), Docket Entry No. 9 (emphasis added). 310 OREGON LAW REVIEW [Vol. 103, 291 II PROCEDURAL LOCALISM AT TWO LITIGATION MOMENTS This Part presents the results of our analyses of local procedural variation at our two points in litigation: involuntary dismissal and settlement. Throughout, we also highlight the local textual variation in how judges and clerks report court action at those two litigation moments. A. Involuntary Dismissals This Section begins with a review of local court rules’ coverage of involuntary dismissals, then presents an overview of all grounds for involuntary dismissal drawn from a small sample of the larger docket sheet universe, then finally delves into courts’ granular, on-the-ground dismissal procedures in connection with a single involuntary dismissal ground: the failure of pro se plaintiffs to keep the court apprised of their current address. 1. Local Rules We first present a sketch of the set of specific deficiencies that the ninety-four U.S. district courts’ local rules identify as dismissible under Rule 41(b). While the federal rule is open as to the specific grounds for dismissal, listing only a plaintiff’s “fail[ure] to prosecute or to comply with these rules or a court order,”47 some courts’ local rules provide more detail.48 These include, for example, failure to properly complete and file a civil cover sheet along with a complaint;49 failure to pay a bond along with filing a lawsuit, applicable to nonresident plaintiffs in some jurisdictions;50 failure to complete service on the defendant;51 and failure to specify the correct response date on a filing.52 Many districts’ local rules also restate FED. R. CIV. P. 41(b)’s requirements in general terms, mandating “diligence”53 and “reasonable promptness”54 in prosecution. Some districts further identify specific periods of 47 FED. R. CIV. P. 41(b). 48 FED. R. CIV. P. 83 (empowering district courts to enact local rules, which “must be consistent with—but not duplicate—federal statutes and rules”). 49 See, e.g., Appendix A (D.N.M. LOC. R. 3.1; D.V.I. LOC. R. 3.1(b)). 50 See, e.g., Appendix A (W.D. PA. LOC. R. 67.1(A)). 51 See, e.g., Appendix A (M.D. LA. LOC. R. 41(b)). 52 See, e.g., Appendix A (N.D. GA. LOC. R. 41.2(A)). 53 See, e.g., Appendix A (D. MASS. LOC. R. 41.1(b)(1)). 54 See, e.g., Appendix A (S.D. GA. LOC. R. 41.1(c)). 2025] Procedural Localism and Access to Justice 311 inaction by a plaintiff that may trigger involuntary dismissal, commonly six to nine months but as short as ninety days in some jurisdictions and as long as one year in others.55 Finally, many jurisdictions impose a specific obligation on pro se plaintiffs to keep the court apprised of their current address—the particular dismissible deficiency of interest here.56 Some courts establish a specific timeframe in which plaintiffs must provide an update, e.g., within five days of a move in the District of Colorado and within sixty- three days in the District of Alaska and Eastern District of California57; others merely state the general notification obligation.58 Importantly, because our review of local rules covered those that were in effect as of March 2023, the sketch above does not reflect the set of rules in place during the earlier time period that is the subject of our docket sheet study. Nor does it cover courts’ standing orders or individual judge-level procedural memos that might contain more detail.59 Regardless, judges are not limited to any specific grounds for dismissal identified in a court’s local rules, standing orders, or memos; they retain their general discretion to dismiss under Rule 41(b) along with their inherent authority to manage their dockets and their 55 See, e.g., Appendix A (D.N.M. LOC. R. 41.1; D. DEL. LOC. R. 41.1; D. WYO. LOC. R. 41.1; D.N.J. LOC. R. 41.1 (all allowing ninety days); N.D.N.Y. LOC. R. 41.2(a) (allowing four months); N.D. GA. LOC. R. 41.3; E.D. WIS. LOC. R. 41; E.D. WASH. LOC. R. 41; D.N. MAR. I. LOC. R. 41.1; W.D.N.Y. LOC. R. 41(b); M.D. LA. LOC. R. 41(b); W.D. LA. LOC. R. 41.3; D. CONN. LOC. R. 41(a); D. ARIZ. LOC. R. 41.1; D. IDAHO LOC. R. 41.1; E.D. MO. LOC. R. 8.01; N.D. IND. LOC. R. 41-1; S.D. CAL. LOC. R. 41.1 (all allowing six months): W.D. WASH. LOC. R. 41; D. MD. LOC. R. 103.8; D. NEV. LOC. R. 41-1; E.D. KY. & W.D. KY. LOC. R. 41.1 (all allowing nine months); D. MASS. LOC. R. 41.1(a)(1); E.D. PA. LOC. R. 41.1; M.D. PA. LOC. R. 41.1 (all allowing one year)). 56 See Andrew Hammond, The Federal Rules of Pro Se Procedure, 90 FORDHAM L. REV. 2689, 2704 n.104 (2022). See generally Andrew Hammond, Pleading Poverty in Federal Court, 128 YALE L.J. 1478 (2019) (cataloging the specific rules and requirements that apply to poor, pro se, and prisoner litigants in federal court). 57 See, e.g., Appendix A (D. COLO. LOC. R. 5.1; E.D. CAL. LOC. R. 183(b); D. ALASKA LOC. R. 11.1(b)(3)). 58 See, e.g., Appendix A (E.D. KY. & W.D. KY. LOC. R. 5.3(e); E.D. MICH. LOC. R. 11.2; M.D. TENN. LOC. R. 41.01; N.D. GA. LOC. R. 41.2; E.D. TENN. LOC. R. 83.13; D. NEV. LOC. R. 3-1; M.D.N.C. LOC. R. 11.1(b); C.D. ILL. LOC. R. 16.3(K); D.R.I. LOC. R. 205 (d)(1); N.D. IOWA & S.D. IOWA LOC. R. 3(c); W.D.N.Y. LOC. R. 5.2(d); E.D. ARK. & W.D. ARK. LOC. R. 5.5(c)(2); S.D. ALA. LOC. R. 83.5(b)). 59 For an example of a judge-specific procedural memorandum, see the website for the U.S. District Court for the Eastern District of Pennsylvania, which has an “Instructions” tab associated with each judge. The website links to a memorandum explaining each judge’s individual “Policies and Procedures.” See, e.g., Policies and Procedures of Juan R. Sánchez, J., U.S. DIST. CT. E.D. PA. (June 2024), https://www.paed.uscourts.gov/sites/paed /files/documents/procedures/sanpol.pdf [https://perma.cc/8HZ4-7SXX]. 312 OREGON LAW REVIEW [Vol. 103, 291 courtrooms. The on-the-ground reasons for involuntary dismissal that courts identify on docket sheets therefore may or may not map onto any given set of local rules. Moreover, apart from the question of grounds for dismissal, judges remain free to manage the process of dismissal, choosing between immediate dismissal and a warning, the number of warnings to give, and the number of days to allow for correction or cure.60 Our docket sheet analysis reveals substantial such variation at the level of court practice, in jurisdictions both with and without applicable local rules. 2. Docket Sheet Analysis: Small Sample, Many Grounds for Dismissal We next present an impressionistic picture of involuntary dismissal drawn from our initial sample of approximately 250 docket entries before turning to the particular pro se address case study. Table 1 shows the dismissal grounds identified in our small sample. Problems with complaints represent the plurality; a review of these complaints suggests that many deficiencies may be related to insufficient pleading to establish subject matter jurisdiction, which courts are separately empowered to review under FED. R. CIV. P. 12(h)(3). The second most frequent category, “unknown,” highlights one shortcoming of docket sheet-based research. Because docket entries are often a short description of the contents of a longer attached document filed by a party or issued by the court, they do not always contain all salient information. This includes, as here, the ground for dismissal, which is likely specified in the full court order. Absent resource considerations, we would have downloaded and extracted additional information from the court documents themselves, rather than just docket entries. Due to the cost of acquiring federal court records,61 however, the SCALES project limited its scope to docket 60 Local rules are generally silent as to procedure, with the exception of mentioning, generally, that the plaintiff should be required to show cause why their claim should not be dismissed—functionally, such explicit requirements are warning mechanisms. See, e.g., Appendix A, N.D. W. VA. LOC. R. 41.01 (lack of interest in prosecution; show cause). Courts have sometimes established a more detailed process around involuntary dismissal. For example, a now-abrogated version of the Northern District of Illinois’ Rule 41.1 stated: “An order of dismissal for want of prosecution or an order of default may be entered if counsel fails to respond to a call of the case set by order of court. Notice of the court call shall be by publication or as otherwise provided by the court. In the Eastern Division publication shall be in the Chicago Daily Law Bulletin unless the court provides otherwise.” By contrast, the Northern District of Georgia’s Local Rule 41.3(A), allows dismissal “with or without notice to the parties.” 61 See supra note 31 (describing a download cost of ten cents per page from the U.S. Courts’ PACER system). 2025] Procedural Localism and Access to Justice 313 entries. As a result, though we are able to spot-download and read selected court documents in this project—as we did above to investigate the “complaint” deficiency topic—a full download of all court documents linked to our docket entries of interest was not feasible. We are therefore inevitably left with some holes in our data. This also points to the problem of local textual variation. While some docket entries are quite specific as to the ground for dismissal or warning, obviating the need to download the full court order, others are terse or merely restate the language of Rule 41(b).62 This variation in the recording of courts’ Rule 41(b) activity on the dockets complicates study of Rule 41(b) on-the-ground procedure. Table 1. Topic distribution: small sample Topic Frequency Percent Complaint 106 42% Unknown 64 25% Fee or IFP filing 50 20% Service problems 15 6% Failure to move for default 12 5% Incorrect form 2 1% Mailing address 2 1% Lawyer appearance 1 0% Total 252 100% 62 Compare, e.g., Order Directing Movant to Cure Deficiency, USA v. Cheek, No. 1:16- cr-00215-CMA-1 (D. Colo. Aug. 15, 2019), Docket Entry No. 74 (“ORDER DIRECTING MOVANT TO CURE DEFICIENCY. Movant Aaron Carson Cheek has filed pro se a Motion to Vacate, Set Aside or Correct Sentence by a Person in Federal Custody Pursuant to 28 U.S.C. § 2255 with Incorporated Memorandum of Law. (Doc. # 73.) As part of the Courts review pursuant to D.C.COLO.LCivR 8.1(b), the Court has determined that the submitted document is deficient because it was not filed on the court-approved form. That Movant Aaron Carson Cheek cure the deficiency designated above within thirty (30) days from the date of this order. Any papers that Movant files in response to this order must include the civil action number on this order. that, if Movant fails to cure the designated deficiency within thirty (30) days from the date of this order, the action will be dismissed without further notice. The dismissal shall be without prejudice, by Judge Christine M. Arguello on 8/15/2019. (evana,) (Entered: 08/15/2019)”), with Order, Haff v. Firman, No. 1:16-cv-01554-GPG (D. Colo. Oct. 3, 2016), Docket Entry No. 10 (ORDER OF DISMISSAL: “ORDERED that the action is dismissed without prejudice pursuant to Fed. R. Civ. P. 41(b) for failure to cure the deficiency and for failure to prosecute. It is FURTHER ORDERED that leave to proceed in forma pauperis on appeal is denied. It is FURTHER ORDERED that all pending motions are denied as moot.”). 314 OREGON LAW REVIEW [Vol. 103, 291 These caveats aside, as Table 1 shows, when faced with potentially dismissible deficiencies, in most cases (90%), courts gave a warning and time to cure, though courts differed broadly with respect to the lengths of those periods. By way of illustration, Figure 1 below shows the distribution of time periods after warnings, by deficiency type, for the small sample. In the figure, the size of the bubble associated with each topic-time window combination represents the number of relevant cases. Figure 1. Number of days by deficiency topic: small sample, all grounds Reading across left to right, and focusing on one ground as an example, failure to pay the filing fee or file for in forma pauperis (IFP) status at the outset of a case, we see that the number of days a plaintiff received varied from fourteen to more than thirty. By way of further illustration, Table 2 below shows the distribution of time windows in the fee or IFP filing category in the four federal district courts in Texas. As the table shows, while two districts were internally uniform in their window lengths (S.D. Tex. at twenty-eight and W.D. Tex. at twenty- three), the Eastern District of Texas was split evenly between twenty- one and thirty days, while the Northern District was all over the map. None of the four districts have a local court rule that covers fee-related involuntary dismissals. 2025] Procedural Localism and Access to Justice 315 Table 2. Percentage of cases by number of days to pay fee or file IFP petition in four Texas districts: small sample Court Number of Days to Pay Fee or File IFP Petition 7 14 20 21 23 25 28 30 31 E.D. Tex. 50% 50% N.D. Tex. 3% 17% 7% 3% 3% 3% 59% 3% S.D. Tex. 100% W.D. Tex. 100% Given that this mini-analysis holds constant the deficiency, state, and local rule status, and that these cases are all drawn from a tight two-year filing period, Table 2 suggests substantial departure from the principle of procedural parity—the same procedure for the same deficiency—and raises concerns about the fairness and access impacts of procedural localism. As Annie Decker has noted, “it is fundamentally unfair for litigants to receive different treatment—some winning, and others losing—based solely on the geographic happenstance of living on one side of a river, or in one part of a state.”63 We revisit those concerns in more depth below. 3. Docket Sheet Analysis: Full Set, One Ground for Dismissal We now move from our small sample of involuntary dismissal docket entries concerning all deficiencies to a full set of docket entries concerning a single deficiency: pro se plaintiffs’ failure to keep their mailing address updated. In this deeper dive into a single ground for dismissal, we perform a more granular analysis of the questions raised in the Introduction, exploring dismissals versus warnings, number of warnings, and number of days allowed to cure, both by court and by judge. As noted above, our universe consists of 1,645 docket entries from 1,389 cases. These cases were filed in 69 courts, involving 417 judges. We note again the problem of textual localism and variation in relation to the sixty-nine courts, which represent 73% of all federal district courts. Though our corpus assembly method performed quite well in identifying docket entries concerning involuntary dismissals related to pro se plaintiffs’ addresses, it may have missed relevant docket entries in the remaining 27% of courts. In other words, because 63 Annie Decker, A Theory of Local Common Law, 35 CARDOZO L. REV. 1939, 1982 (2014). 316 OREGON LAW REVIEW [Vol. 103, 291 of the lack of text standardization across courts’ docket entries, even machine learning tools may have a hard time reliably identifying the true full universe of address-related dismissals. Alternatively, those courts may not use Rule 41(b) as the vehicle for handling address updates. Given our current data, we are unable to distinguish between those two possible reasons for the court-level distribution of the cases in our pro se address corpus. Nevertheless, both reasons highlight the problems associated with localism: Either clerks’ and judges’ recording practices are so local that they obscure the underlying events being recorded, or the underlying events themselves are so varied—using Rule 41(b) to handle pro se address problems or using some other procedure—that focusing only on Rule 41(b) misses the true scope of relevant court action. Nevertheless, using just the sixty-nine courts present in our docket entry set, we can observe substantial variation in local procedure. First, as in the sample set, most judges (75%) issued a warning rather than an immediate dismissal as their first action in response to an outdated pro se plaintiff address. This practice differed by court, however, and by judge. Judges in the Middle District of Louisiana, Southern District of Indiana, and Eastern District of Virginia, for example, issued immediate dismissals in 90% or more of the cases before them involving pro se address problems. But even within these districts, judge behavior varied: In the Middle District of Louisiana, for instance, one judge issued warnings as the first action in response to an outdated address in 100% of cases, whereas the other judges issued dismissals in 100% of cases. The form of warnings and dismissals varied as well. With respect to immediate dismissals, almost all were without prejudice, meaning that plaintiffs could update their address and resume litigation. However, some judges gave plaintiffs a window in which to cure, of varying lengths, and others did not, merely stating that the dismissal was without prejudice. Likewise, the judges who issued warnings specified time periods of varying lengths. Figure 2 below illustrates these variations, where the size of the bubble reflects the number of cases in each category and the “0” item on the x-axis reflects the docket entries in which the judge did not set a deadline for correction or specify a number of days. 2025] Procedural Localism and Access to Justice 317 Figure 2. Number of days by immediate court action: pro se address cases Beyond the obvious variation that Figure 2 shows, the “0” entries in the “warning” category contain some additional notable subtypes of court action. First, in some cases, courts discovered pro se plaintiffs’ new address on their own, often when the plaintiff was incarcerated and switched prisons and court personnel searched the prison database. There, judges updated the plaintiffs’ addresses themselves, often describing it as “a courtesy,” but reiterated the general obligation on the plaintiffs to apprise the court affirmatively of address changes and warned of possible future dismissal.64 Second, in the Southern District of Florida, in most but not all cases, the first action in response to returned mail from a pro se plaintiff’s address was a notice that the court would make two noticing attempts and then cease sending mail until the address was updated, along with a warning about possible dismissal.65 Still other, more idiosyncratic court actions required the 64 See, e.g., Notice, Wheeler v. Gann, No. 3:16-cv-03244 (C.D. Ill. Jan. 10, 2017) (“NOTICE: The Clerk of the Court has been informed by Oklahoma City FTC that effective immediately, Plaintiff has been transferred from Oklahoma City FTC to Atlanta USP. The Illinois Department of Corrections has conveyed this information simply as a courtesy to the Clerk of the Court.[ ]Plaintiff is reminded that he is under a continuing obligation to keep the Clerk and each opposing party informed of any change in his location. Pursuant to Central District of Illinois Local Rule 16.3(K), every pro se plaintiff must notify the Clerk of this court in writing of any change of address during the entire pendency of his case. It is not the Court’s responsibility to independently maintain current addresses on all parties to pending actions. As a courtesy to Plaintiff, however, the Clerk is DIRECTED to send a copy of this Notice to Plaintiff at the institution where he has reportedly been transferred. Plaintiff shall, no later than 14 days from the date of this Notice, file a notice of change of address with the Clerk. Failure to do so will result in dismissal of this action.[ ]Notice of Change of Address to be filed by plaintiff by 1/24/2017 (LN, ilcd) (Entered: 01/10/2017).”). 65 See, e.g., Notice of Undeliverable Mail, Carrasco v. Sessions, No. 1:17-cv-24486- CMA (S.D. Fla. Jan. 9, 2018), Docket Entry No. 9 (“Clerk’s First Notice of Undeliverable Mail re[4] Order of Instructions to Pro Se Litigant, [5] Order to Show Cause,. US Mail 318 OREGON LAW REVIEW [Vol. 103, 291 defendants to find an updated plaintiff address and inform the court,66 and—in an example that is outside the scope of pro se plaintiff address problems but is nevertheless remarkable in the judge’s willingness to delay Rule 41(b) dismissal—required an attorney to serve a copy of this Order on [the plaintiff] by mail at her last known address, and to make three attempts on three separate days (once each in the morning, afternoon, and evening) to contact [the plaintiff] at her last known telephone number and orally advise her of this Order and the steps she must take to respond to avoid involuntary dismissal.67 returned for: Lino Gonzalez Carrasco. The Court has not located an updated address for this party. After two unsuccessful noticing attempts, notices from the Court will no longer be sent to this party in this case until a correct address is provided. (drz) (Entered: 01/09/2018).”). 66 See, e.g., Order, Ocasio v. Scott, No. 1:16-cv-01336 (E.D.N.Y. Dec. 28, 2016), Docket Entry No. 30 (“ORDER [TO ANSWER]: The Court held an initial conference in this matter on 12/28/2016. Plaintiff Van Brooks appeared by telephone; however, plaintiff Omar Ocasio failed to appear as directed. As discussed on the record, plaintiff Ocasio is no longer incarcerated and has failed to provide the Court or defendants his current contact information. It is plaintiff’s responsibility to keep the Court informed of his current address. However, as Ocasio is proceeding pro se, the Court directs defendants to determine the last known contact information for Ocasio. Defendants shall provide such contact information to the Court by 1/13/2017. If no alternate contact information is found, and if Ocasio fails to contact the Court by 1/13/2017, I shall recommend that plaintiff’s claims should be dismissed without prejudice. Notwithstanding Ocasio’s apparent abandonment of this action, plaintiff Brook’s claims shall proceed. As discussed on the record, defendants Bastian, Cheron, Davis, Roberts, and the City of New York shall answer the complaint by 1/6/2017. By the same date, defendants shall produce to plaintiff Brooks their initial disclosures and shall provide Brooks with any necessary medical authorizations. Defendants’ counsel shall also provide the Court, by 12/29/2016, the full names and proper service addresses for defendants Brown, Goode, Frazier, and Scott. The Court shall schedule another conference once all defendants have responded the complaint. SO ORDERED by Magistrate Judge Lois Bloom, on 12/28/2016. C/mailed. (Latka-Mucha, Wieslawa) (Entered: 12/29/2016).”). 67 See, e.g., Order, Roe #1 v. Minguela, No. 1:16-cv-02744 (D. Colo. Oct. 30, 2020), Docket Entry No. 135 (“ORDER TO SHOW CAUSE: Based on the ‘Court’s review of the docket, it appears that the sole claims remaining in this case are Jane Roe #’2’s claims against Mr. Minguela. On March 31, 2020, Mr. Minguela filed a Motion to Compel 128 certain discovery responses from Ms. Roe #2. On April 6, 2020, Ms. Roe #’2’s counsel filed a Motion to Withdraw, stating that, despite efforts on his part to contact Ms. Roe #2 and secure her responses to Mr. Minguela’s discovery requests, he had not had any contact with Ms. Roe #2 since November 23, 2019. To date, no response to either the Motion to Compel nor the Motion to Withdraw have been filed by Ms. Roe #2, nor is there any further indication that Ms. Roe #2 intends to continue litigating this case. The matter is, at present, stalled due to Ms. Roe #2’s apparent refusal to communicate with her counsel or the Court. The Court intends to dismiss Ms. Roe #2’s remaining claims against Mr. Minguela for failure to prosecute pursuant to Fed. R. Civ. P. 41(a) [sic]. However, in order to provide Ms. Roe #2 with a final opportunity to continue to pursue those claims, the Court directs that on 2025] Procedural Localism and Access to Justice 319 Thus, Figure 2, as well as the additional court actions noted above, reveals substantial procedural localism, where judges, when confronted with the exact same deficiency, employ differing combinations of dismissals or warnings, deadlines or not, and numbers of days to cure. These variations matter, especially when considering that the plaintiffs whose claims risk permanent dismissal are unrepresented by lawyers. A plaintiff whose claims are immediately dismissed, even if given the option to cure and reopen the case, risks getting “stuck” in that default, as the onus is on the plaintiff to take action to revive his or her claims.68 Judges who dismiss but set an explicit window for reopening may aid plaintiffs somewhat by giving them a more explicit target to hit, but the default case status nevertheless remains dismissed. In contrast, judges who issue warnings as their first action reduce these plaintiffs’ precarity, as the case remains open, at least for some additional period. Next, we expand the view to consider subsequent court actions beyond the initial dismissal or warning. Table 3 shows the full set of court actions per case, including standalone dismissals (d) and warnings (w) and sequences of warnings and dismissals, listing frequency and percent among the full set of 1,389 cases. Additionally, Table 3 shows the median total days associated with each court action or sequence of actions, calculated including zero-day entries. This table confirms and extends the analysis above of judges’ immediate actions, showing that in some cases, judges gave pro se plaintiffs many warnings and chances to cure, while in others or before November 23, 2020, Ms. Roe #2 shall SHOW CAUSE, in writing, why her remaining claims should not be dismissed for failure to prosecute. In that writing, she shall: (i) expressly re-affirm her intention to continue litigating her remaining claims; (ii) state her position with regard to her counsel’s Motion to Withdraw, specifically whether she intends to continue to work with her current counsel, to retain new counsel (any such new counsel shall file a Notice of Appearance by that same date), or express her intention to pursue her remaining claims pro se; and (iii) state her position regarding Mr. Minguela’s Motion to Compel, and either tender responses to Mr. Minguela’s requests or file a Motion for Protective Order (or both) by that date. Failure to comply with these requirements will result in the Court deeming Ms. Roe #2 to have abandoned her remaining claims against Mr. Minguela. To ensure that Ms. Roe #2 is provided with all possible avenues of notice of this Order, the Court directs that Mr. Poor serve a copy of this Order on Ms. Roe #2 by mail at her last known address, and to make three attempts on three separate days (once each in the morning, afternoon, and evening) to contact Ms. Roe #2 at her last known telephone number and orally advise her of this Order and the steps she must take to respond. Upon completing that task, counsel shall file a certification of his compliance with this Order. By Judge Marcia S. Krieger on 10/30/20. Text Only Entry (msklc2,) (Entered: 10/30/2020).”). 68 For an explanation of the concept of sticky defaults from a behavioral economics perspective, see generally RICHARD H. THALER & CASS R. SUNSTEIN, NUDGE: IMPROVING DECISIONS ABOUT HEALTH, WEALTH, AND HAPPINESS (Penguin Books 2009) (2008). 320 OREGON LAW REVIEW [Vol. 103, 291 (approximately 22%), they dismissed right away. Indeed, many of the courts that issued the most warnings also had comparable numbers of standalone immediate dismissals. Figure 3 illustrates this phenomenon, showing the nineteen courts in which any case received two or more warnings and no dismissals. For each court, the figure shows the percentage of all cases that received that repeated warning sequence (the most lenient of all court action possibilities), along with the percentage of cases that received a standalone dismissal (the least lenient possibility). Table 3. Full set of court actions: pro se address cases Court Action(s) Frequency Percent Median To- tal Days in Sequencea w 869 62.6% 15 d 310 22.3% 0 w|w 83 6.0% 29 w|d 67 4.8% 28 d|d 30 2.2% 48 w|w|w 10 0.7% 44 w|w|d 6 0.4% 41 w|w|w|w 4 0.3% 71 w|w|w|d 3 0.2% 53 w|d|d 2 0.1% 49 w|w|w|w|w 2 0.1% 74 d|w|d 1 0.1% 17 w|d|w 1 0.1% 50 w|w|w|w|w|w|w|w69 1 0.1% 112 Note: d = dismissal; w = warning aIncludes zero-day entries 69 In this case, which alleged excessive force, failure to intervene, and battery against Pontiac Correctional Center, and which received the most warnings in sequence of any in the corpus, the plaintiff was incarcerated and repeatedly moved between prisons. Roman v. Frazier, No. 1:17-cv-01195 (C.D. Ill. Nov. 2, 2022). In some warnings, the court sua sponte updated the plaintiff’s address as a courtesy, while also issuing a warning about possible future involuntary dismissal. In other warnings, the court noted that the party had filed a Notice of Change of Address in a separate case but not in the present one. Perhaps because the plaintiff was making efforts to keep the court apprised, at least in one lawsuit, the court exercised more lenience and never dismissed the case outright. Ultimately, the plaintiff lost his claims on summary judgment. 2025] Procedural Localism and Access to Justice 321 Figure 3. Percentage of cases with two or more warnings and no dismissals and with standalone dismissals Though two courts in Figure 3, the Districts of Minnesota and Wyoming, were markedly more lenient than others, in many courts, roughly the same proportion of cases received the most and least lenient treatment, respectively. Local rules provide little explanation here: Only one of these courts, the Southern District of Florida, includes any specific discussion of pro se plaintiffs’ obligation to keep their address updated as a ground for involuntary dismissal.70 Thus, Figure 3 again points to substantial variation in on-the-ground practices, as judges exercise their discretion to identify dismissible deficiencies and create their own hyperlocal procedures for managing involuntary dismissals. This procedural localism raises both fairness and access to justice concerns, as different plaintiffs in the same circumstances, and often in the same courts, receive different procedural treatment. Beyond observable differences in procedure, we have reason to suspect differences in outcomes as well. Recall that the dismissal language that we identified in docket entries was usually without prejudice, meaning that the plaintiffs could provide an updated address and reopen the case. Seen in this light, the presence of a dismissal versus a warning might not predict whether the case dies at that point or survives through later stages of litigation, since both actions leave open an option to continue. We can mine the IDB’s case-level 70 Appendix A (S.D. FLA. LOC. R. 11.1(g)). 322 OREGON LAW REVIEW [Vol. 103, 291 disposition codes to test this proposition, keeping in mind the data quality concerns raised above. Of the cases in Table 3 with court action sequences that consisted solely of or ended with a dismissal, 18% had IDB disposition codes that suggested that the case had survived the pro se address problems and continued to later stages of litigation. These included codes for settlement, trial, and motions before trial. By contrast, of the cases with court action sequences that consisted solely of or ended with a warning, 62% had disposition codes that suggested that litigation had proceeded to a later stage. This implies that the dismissal docket entries that we identified, though technically not final, may have functionally been the end of the line for many pro se plaintiffs. As we suggested above, this may be due to the stickiness of defaults and the problem of inertia. Inaction after a dismissal meant that the case remained closed; inaction after a warning, particularly before a judge who was inclined to issue multiple warnings, meant a continued chance to stay in court. This suggests a direct link between procedural localism and access to justice, and a meaningful difference between the precarity of the plaintiffs whose cases were dismissed outright and those who received multiple warnings and opportunities to cure. B. Settlement We now shift to a second moment in litigation—judges’ handling of settlement—as another site for the study of procedural localism. Because the parties to a settlement want to leave court, variation around settlement procedures does not directly affect parties’ precarity, as in the involuntary dismissal example. However, varying settlement practices may nevertheless have access to justice implications. This is because procedural localism, by its very nature, creates insiders and outsiders: those in the know and those who must spend resources to acquire local knowledge. The more expensive it is to learn local court- and judge-level practices and procedures, or to hire a member of the “cognoscenti”71 as opposed to reading a single, unified set of federal civil procedural rules, the higher the cost of litigation. In the Introduction and Part I.D above, we explained our scope: docket entries in cases that lack a particular procedural framework around settlement, in which the parties informed the judge of a resolution, and where the judge responded with an action and, often, a set of instructions about and deadlines for next steps. We group 71 Carrington, supra note 17. 2025] Procedural Localism and Access to Justice 323 those actions into three categories: immediate dismissal, statistical/ administrative closing, and leaving the case open. Entries in which the case remained open typically acknowledged that the parties had settled and instructed the parties as to the next steps, but took no action to close or dismiss the case.72 Immediate dismissals closed the case right away, but were made without prejudice to the parties’ taking action to reopen the case or change the dismissal to with prejudice.73 Statistical and administrative closings occupy a middle ground between open and closed: Upon notification of settlement, judges withdrew cases from the active docket and closed them, but explicitly did not enter a dismissal.74 As in the cases in the other two categories, the parties often received a window of time to convert the administrative closing to a true dismissal or to reopen the case for further litigation.75 Like in the involuntary dismissal case study, some sets of local rules speak to procedures around settlement, but most are silent. In our review of local rules effective as of March 2023, we identified twenty- 72 See, e.g., Order, Altshiler v. Alaska, No. 3:16-cv-00259 (D. Alaska Aug. 2, 2018), Docket Entry No. 99 (“SLG TEXT ORDER: The Court has received a verbal notification from Plaintiff’s counsel that this case has been settled by the parties. In light of the foregoing, the jury trial scheduled to begin on August 13, 2018 is VACATED. The parties shall file case closing documents within 30 days of the date of this order. (JLH, COURT STAFF) (Entered: 08/02/2018).”). 73 See, e.g., Order, Iglesia El Jordan v. Church Mut. Ins. Co., No. 5:17-cv-01077 (W.D. Tex. Mar. 11, 2020), Docket Entry No. 74 (“ORDER The Court having been advised that the parties have settled this case, this case is hereby DISMISSED, subject to reinstatement if the settlement is not effectuated within 30 days. Signed by Judge Royce C. Lamberth. (mgr) (Entered: 03/11/2020).”). 74 See, e.g., Order, Ross. v. Liberty Mut. Ins. Co., No. 3:16-cv-00781 (N.D. Tex. Oct. 19, 2016), Docket Entry No. 25 (“Order Administratively Closing Case. The Court has been notified that this case has settled. The Parties are therefore directed to submit their dismissal papers within 30 days from the date of this order. Nothing in this Order shall be considered a dismissal or disposition of this case, and should further proceedings become necessary or desirable, any party or the Court may initiate such further proceedings in the same manner as if this Order had not been entered. (Ordered by Judge Ed Kinkeade on 10/19/2016) (ndt) (Entered: 10/19/2016).”). 75 See, e.g., Order, Nassirzadeh. v. BMW of N. Am., LLC, No. 2:16-cv-09438 (C.D. Ca. June 27, 2017), Docket Entry No. 14 (“ORDER REMOVING ACTION FROM THE ACTIVE LIST OF CASES, PURSUANT TO SETTLEMENT by Judge Christina A. Snyder. THE COURT, having been notified by counsel that the above-entitled action has been settled on or about June 26, 2017; IT IS HEREBY ORDERED that this action is hereby removed from this Courts active caseload until further application by the parties or order of this Court. IT IS FURTHER ORDERED that counsel shall file a proper stipulation for dismissal or a joint report detailing settlement status within 30 days and every quarter thereafter until a stipulation for dismissal is filed. This Court retains full jurisdiction over this action and this Order shall not prejudice any party to this action. All dates in this action are hereby VACATED. IT IS SO ORDERED. (Case Terminated. Made JS-6) (clee) (Entered: 06/27/2017).”). 324 OREGON LAW REVIEW [Vol. 103, 291 three courts that gave the parties and/or the court instructions on handling settlement.76 About half were silent about dismissal, statistical/administrative closing, or keeping the case open, but instructed the parties to notify the court immediately upon settlement and to file the requisite paperwork, sometimes mentioning Rule 41(a) voluntary dismissal, within two, twenty-eight, or thirty days.77 The other half were explicit as to dismissal or statistical/administrative closing, also specifying the right of the parties to reopen within thirty, thirty-five, sixty, or ninety days.78 Others instructed the judge to allow the parties to reopen within a time period “set by the Court”79 or, even more generally, “for good cause.”80 Thus, though some local rules provide some guidance, they leave intact large zones of discretion for many judges to decide on a settled case’s open or closed status, the obligations, if any, on the parties, and their deadlines for compliance. As expected, the docket sheets reveal substantial variation in judges’ exercise of this discretion. The total number of cases in this corpus, with one settlement notification docket entry per case, is 13,897, drawn from all ninety-four U.S. district courts.81 Of those cases, most (57%) remained open upon notification of settlement, but a sizable minority (37%) were immediately dismissed, followed by statistical or 76 See Appendix B. This list does not include local rules that discuss court-run settlement conferences or court-supervised mediation programs or the general duty to inform the court promptly of settlement. Instead, the list in Appendix B reflects the set of local rules that give the parties explicit and specific instructions about the procedure in the event of an out-of- court settlement that was not reached via a court-annexed program. Note that Appendix B, like Appendix A, combines Iowa districts’ local rules into one row in the table, as the two courts share a common set of local rules. 77 See, e.g., Appendix B (S.D. FLA. LOC. R. 16.4 (“two (2) Court days”); D. MASS. LOC. R. 68.2; W.D. TENN. LOC. R. 83.13 (28 days); D. ME. LOC. R. 41.1(a); D. WYO. LOC. R. 41.1; D. NEB. LOC. R. 41.1; D.N.H. LOC. R. 41.1; D.N.D. LOC. R. 41.1; N.D.N.Y. LOC. R. 68.1 (30 days)). 78 See, e.g., Appendix B (D. CONN. LOC. R. 41(b); N.D. IOWA & S.D. IOWA LOC. R. 41(b) (30 days); D. OR. LOC. R. 41-1(c) (60 days); E.D. PA. LOC. R. 41.1(b) (90 days)). 79 See, e.g., Appendix B (D. MD. LOC. R. 111; see also D.P.R. LOC. R. 41(a)). 80 See, e.g., Appendix B (N.D. OKLA. LOC. R. 41-1). 81 As a rough quality control check, 93.9% of the cases in which we identified a settlement-related docket entry had one of the following three IDB dispositions: settled (75.9%), voluntary dismissal (9.2%), and other dismissal (8.7%). Unlike in the Rule 41(b) involuntary dismissal case study, all courts are represented here. This is likely because settlements occur more frequently than involuntary dismissals. In addition, for an involuntary dismissal to appear on a docket sheet, the judge (and less frequently the defendant) must first identify a dismissible deficiency and then craft a set of procedures to handle that deficiency. In some sense, the judge is creating both the litigation event and the procedures that surround it. In the settlement context, the parties notify the judge that a case has settled—thereby creating the litigation event—and the judge then must only craft the procedure. 2025] Procedural Localism and Access to Justice 325 administrative closings (6%) at a distant third. As in the involuntary settlement context, courts were neither uniform with one another nor uniform internally in their handling of settlements. Fifty-three courts, representing 56% of all courts, kept all or nearly all cases (90–100%) open after receiving a settlement notification, though—as explored further below—varied as to the next steps for parties and the amount of time the parties were allowed for completion.82 Notably, seven of those fifty-three courts had local rules, at least as of March 2023, that prescribed the procedure upon settlement as an immediate dismissal or statistical/administrative closing of the case, seemingly contradicting courts’ actual, on-the-ground practice.83 Six had local rules that were silent as to dismissal or closing;84 the remainder had no relevant local rule. Figure 4 provides detail on the mix of court actions within the remaining forty-one courts that had less internal uniformity. As the chart shows, judges varied widely within the same courts in their choice of action upon notification of settlement, deploying disparate procedural frameworks around a seemingly identical underlying litigation event. 82 Those fifty-three courts are D. Alaska, S.D. Ala., W.D. Ark., D. Ariz., N.D. Cal., S.D. Cal., D. Colo., D.D.C., D. Del., D. Guam, D. Haw., N.D. Iowa, S.D. Iowa, D. Idaho, C.D. Ill., N.D. Ill., S.D. Ill., N.D. Ind., S.D. Ind., D. Kan., M.D. La., D. Md., D. Me., E.D. Mich., W.D. Mich., D. Minn., E.D. Mo., N.D. Miss., S.D. Miss., D. Mont., M.D.N.C., W.D.N.C., D.N.D., D. Neb., D.N.H., D.N.J., D.N.M., D. Nev., N.D.N.Y., E.D. Okla., N.D. Okla., W.D. Okla., E.D. Pa., D.P.R., D.S.C., D.S.D., E.D. Tenn., E.D. Tex., W.D. Tex., D. Utah, E.D. Va., E.D. Wash., N.D.W. Va. 83 Appendix B (D. MD. LOC. R. 111; N.D. IOWA & S.D. IOWA LOC. R. 41(b); D.N.J. LOC. R. 41.1(b); E.D. PA. LOC. R. 41.1(b)). 84 Appendix B (D.P.R. LOC. R. 41(a); D. ME. LOC. R. 41.1(a); D. NEB. LOC. R. 41.1; D.N.H. LOC. R. 41.1; N.D.N.Y. LOC. R. 68.1; E.D. TENN. LOC. R. 68.1). 326 OREGON LAW REVIEW [Vol. 103, 291 Figure 4. Forty-one courts without internal uniformity in responses to settlement One pair of docket entries drawn from the Eastern District of North Carolina, and entered within about three months of each other by different judges, is illustrative: ORDER - The court has been advised that the parties have settled all matters in controversy among them. Therefore, this matter is DISMISSED su