AFRICAN AMERICAN CREDIT UNION COALITION; AMERICA’S CREDIT UNIONS; CHAMBER OF COMMERCE OF THE UNITED STATES OF AMERICA; MORTGAGE BANKERS ASSOCIATION,

Amici Supporting Appellee.

Appeal from the United States District Court for the Eastern District of Virginia, at Alexandria. Leonie M. Brinkema, District Judge. (1:23-cv-01731-LMB-WEF)

Argued: March 20, 2025 Decided: February 9, 2026

Before RICHARDSON and HEYTENS, Circuit Judges, and Norman K. MOON, Senior United States District Judge for the Western District of Virginia, sitting by designation.

Page 2Order affirmed in part and vacated in part by published opinion. Judge Heytens wrote the opinion, which Judge Moon joined. Judge Richardson wrote an opinion concurring in the judgment in part and dissenting in part.

ARGUED: Daniel R. Schwartz, DICELLO LEVITT LLP, Chicago, Illinois, for Appellants. Daniel Stephen Volchok, WILMERHALE LLP, Washington, D.C., for Appellee. ON BRIEF: Hassan A. Zavareei, Glenn E. Chappell, TYCKO & ZAVAREEI LLP, Washington, D.C.; Adam J. Levitt, DICELLO LEVITT LLP, Chicago, Illinois; Ben Crump, BEN CRUMP LAW PLLC, Tallahassee, Florida; Michael Dunn, Raleigh, North Carolina, Glen L. Abramson, MILBERG COLEMAN BRYSON PHILLIPS GROSSMAN, PLLC, Knoxville, Tennessee, for Appellants. Jonathan E. Paikin, Karin Dryhurst, Joseph M. Meyer, Lucyanna Burke, WILMER CUTLER PICKERING HALE AND DORR LLP, Washington, D.C., for Appellee. Stephen V. Carey, Aislinn R. Klos, PARKER POE ADAMS & BERNSTEIN LLP, Raleigh, North Carolina; Jonice Gray, Washington, D.C., Kristopher Knabe, PAUL HASTINGS LLP, Chicago, Illinois, for Amicus African American Credit Union Coalition. Sarah J. Auchterlonie, Denver, Colorado, Leah C. Dempsey, BROWNSTEIN HYATT FARBER SCHRECK, LLP, Washington, D.C., for Amici America’s Credit Unions, Mortgage Bankers Association, and Chamber of Commerce of the United States of America.

Page 3TOBY HEYTENS, Circuit Judge:

Federal Rule of Civil Procedure 23(c)(1)(A) instructs district courts to make class certification decisions at “an early practicable time.” Our threshold question is what legal standards govern a district court’s decision when a defendant asks the court to deny class certification before any discovery has occurred. Consistent with this Court’s decision in Goodman v. Schlesinger , 584 F.2d 1325 (4th Cir. 1978), we reaffirm that district courts must evaluate such requests based solely on the face of the complaint and ask whether the complaint’s allegations fail to satisfy Rule 23(a) and (b)’s requirements as a matter of law.

Applying those standards here, we conclude the district court acted within its discretion when it denied class certification under Rule 23(b)(3) and struck the corresponding allegations from the complaint. But we conclude that the court acted prematurely-and thus exceeded its discretion-in denying class certification under Rule 23(b)(2). We thus affirm the district court’s order in part and vacate it in part.

Nine applicants for residential mortgage products sued Navy Federal Credit Union, individually and on behalf of a putative class, alleging systematic discrimination against racial minorities. The facts of each applicant’s case vary. Relevant to this appeal: (1) the applicants live in different States; (2) eight applicants are Black and one is Latino; (3) six applicants applied for a first mortgage, one applied for a first mortgage and a cash -out refinance, one applied for a Veterans Affairs (VA) first mortgage, and one applied for a VA cash-out refinance; and (4) the applicants' debt, income, and credit scores vary.

Page 4Despite those differences, the complaint alleges that Navy Federal uses a “semi -automated underwriting process” for all loan applicants, which results in discrimination against “African Americans, Latinos, Native Americans, and other racial minorities.” JA 24, 51. According to the complaint, that process involves collecting certain forms of data from every applicant, some of which “can be proxies for race.” JA 44. Navy Federal then “runs the data . . . through its proprietary underwriting algorithm.” JA 45–46. What “variables [are] used” by that algorithm, “and the weight those variables are given, is entirely up to Navy Federal,” and “Navy Federal maintains secrecy” over what those variables and weights are. JA 51, 53. Incorporating independent reports that identify and analyze racial disparities in Navy Federal’s lending data, the complaint alleges that Navy Federal's 'somewhat automated process' produces a 'uniquely discriminatory result.' JA 44, 53.

Beyond asserting individual claims of intentional discrimination and disparate impact, the complaint also seeks classwide declaratory and injunctive relief under Federal Rule of Civil Procedure 23(b)(2) and damages under Rule 23(b)(3). The complaint defines the proposed class as:

All minority residential loan applicants from 2018 through the present . . . who submitted an application for any home mortgage loan to [Navy Federal], who sought to refinance or modify a home mortgage loan through [Navy Federal], and/or who sought a Home Equity Line of Credit from [Navy Federal] and whose application was:

• (a) denied;

(b) approved at higher interest rates and/or subject to less favorable terms as compared to similarly situated non-minority applicants; or

(c) processed at a rate slower than the average processing time of applica[tions] submitted by similarly situated non-minority applicants.

Page 5JA 122.

Navy Federal moved to dismiss the complaint under Rule 12(b)(6) and, in the alternative, to strike the class allegations under Rules 12(f) and 23(d)(1)(D). As relevant here, Navy Federal argued that the differences across loan programs precluded class certification because the applicants failed to explain how an undefined underwriting process could produce discriminatory effects for class members who applied for different products.

After a hearing, the district court granted Navy Federal’s motion to dismiss in part and denied it in part. The court also stated it would “strike the class allegation[s],” citing both Rules 12(f)(2) and 23(d)(1)(D). JA 122. This Court granted interlocutory review of the district court's order striking the complaint's class allegations. See Microsoft Corp. v. Baker , 582 U.S. 23, 34 n.7 (2017). We have jurisdiction under 28 U.S.C. § 1292(e).

It is common ground that district courts may sometimes make class certification decisions based solely on the pleadings and before any discovery has occurred. See General Tel. Co. of Sw. v. Falcon , 457 U.S. 147, 160 (1981). Although Navy Federal and the district court cited Federal Rules of Civil Procedure 12(f) and 23(d)(1)(D), we conclude that Rule 23(c)(1)(A) is the source of authority to make such determinations. We also conclude that both sides to this appeal have correctly identified the relevant legal standards that guide a district court in exercising its discretion. By directing district courts to make class certification decisions at “an early practicable time,” Rule 23(c)(1)(A) grants Page 6district courts considerable discretion about the timing of their class certification decision, including whether to entertain requests to make such decisions at the pleading stage. But consistent with this Court’s decision in Goodman , we conclude a district court may deny class certification before discovery only if the complaint’s class action allegations fail to satisfy the relevant legal standards as a matter of law.

We start by explaining why the source of a district court's authority to decide whether to certify a class is always Rule 23(c)(1)(A). That provision states: “At an early practicable time after a person sues or is sued as a class representative, the court must determine by order whether to certify the action as a class action.” A rule that requires district courts to do something—here, “determine by order whether” to grant or deny class certification—seems like the most obvious source of power to issue that order.

Taking one step back only reinforces that conclusion. Rule 23(c)(1)(A) is housed within a broader subsection captioned “Certification Order.” Fed. R. Civ. P. 23(c)(1). And a later provision in that same subsection states: 'An order that grants or denies class certification may be altered or amended before final judgment.” Fed. R. Civ. P. 23(c)(1)(C). That reference to “[ a ] n order that grants or denies class certification,' id. (emphasis added), is most naturally read as referring back to the order required just two provisions earlier by Rule 23(c)(1)(A).

Our understanding also maps onto the overall structure of Rule 23. Its first two sections (Rule 23(a) and (b)) lay out the legal requirements for certifying a class in the first place. Rule 23(c)—of which Rule 23(c)(1)(A) is one component—then empowers district Page 7courts to certify, define, and provide notice to any class. See, e.g. , Rule 23(c)(1) (certification), Rule 23(c)(2) (notice); Rule 23(c)(3) (requiring that any judgment describe the relevant class); Rule 23(c)(4)-(5) (authorizing issue-based classes and subclasses). Rule 23(c) thus operationalizes the requirements established by sections (a) and (b). The next section identifies various housekeeping orders that district courts 'may issue' in “Conducting the Action” (Rule 23(d)), and the remaining sections address “Settlement, Voluntary Dismissal, or Compromise' (Rule 23(e)), 'Appeals' (Rule 23(f)), 'Class Counsel” (Rule 23(g)), and “Attorney’s Fees and Nontaxable Costs” (Rule 23(h)). The best fit for the source of a district court's authority to make class certification decisions is Rule 23(c)—specifically Rule 23(c)(1)(A).

In contrast, neither of the rules identified in Navy Federal's motion authorize a district court to resolve class certification questions at the pleading stage (or, indeed, at any other time). We conclude that Rule 12(f) has nothing to do with those questions and that exclusive reliance on Rule 23(d)(1)(D) omits a critical step.

Start with Rule 12(f). True, that Rule says “[t]he court may strike” certain material 'from a pleading.” But none of the words that follow-'redundant, immaterial, impertinent, or scandalous matter'-mention class actions or Rule 23. Cf. Arkansas Game & Fish Comm’n v. United States , 568 U.S. 23, 36 (2012) (“[T]he first rule of . . . statutory interpretation is: Read on.”). Nor do those words track Rule 23(a) and (b)’s requirements for certifying a class action, even implicitly. See Eisen v. Carlisle & Jacquelin , 417 U.S. 156, 178 (1974) (“In determining the propriety of a class action, the question is . . . whether the requirements of Rule 23 are met.” (quotation marks removed)). A leading treatise tries Page 8to address this problem by suggesting that “overbroad or unsupportable class allegations bring ‘impertinent’ material into the pleading,” 5C Wright & Miller’s Federal Practice & Procedure § 1383 (3d ed. 2025), but that argument chases its own tail. Even on the argument’s own terms, a complaint’s class action allegations become “impertinent” only after the district court has already decided, via some other mechanism, that the class action “ultimately cannot be sustained,” id. (quotation marks removed)-in other words, that the requirements of Rule 23(a) and (b) are not met. Rule 12(f) simply does not work.

So move on to Rule 23(d)(1)(D). More promisingly, that provision is in the class action rule (Rule 23), and it says a district court “may . . . require that pleadings be amended to eliminate allegations about representation of absent persons and that the action proceed accordingly.” Fed. R. Civ. P. 23(d)(1)(D). But that just produces another round of tailchasing. The only obvious justification for striking class allegations is because the district court has already decided—again, via some other source of authority—against certifying a class. That is why 'order[s] striking class allegations' are 'functionally equivalent' to those “denying class certification”: because doing the former necessarily entails having done the latter. Microsoft , 582 U.S. at 34 n.7 (alterations and quotation marks removed); cf. id. (concluding that orders “striking class allegations” are “appealable under Rule 23(f)” even though Rule 23(f), on its face, refers only to “an order granting or denying class-action certification'). Consistent with the advisory committee's explanation, we understand Rule 23(d)(1)(D) to be an optional housekeeping provision that comes into play only after the district court has made its mandatory class certification decision under Rule 23(c)(1)(A). See Fed. R. Civ. P. 23(c)(1) advisory committee's note to 1966Page 9 amendment (explaining that, if a court makes a 'negative determination' about class certification under Rule 23(c)(1), “the action should be stripped of its character as a class action” using the authority granted under former Rule 23(d)(4), which is now Rule 23(d)(1)(D)).

In sum, we hold that “the appropriate procedure for a defendant to challenge class certification”—at the pleading stage or any other time—is to make “a motion to deny class certification under 23(c)(1)(A), coupled with a motion to strike under 23(d)(1)(D) should the motion to deny class certification be granted.” Timothy A. Daniels, Challenging Class Certification at the Pleading Stage: What Rule Should Govern and What Standard Should Apply? , 56 S. Tex. L. Rev. 241, 271 (2014).¹

We next explain why-in deciding whether to deny class certification at the pleading stage-a district court may consider only the face of the complaint and ask whether its allegations make a prima facie showing that satisfies Rule 23(a) and (b)'s requirements.

This is not the first time our Court has addressed this question. In Goodman v. Schlesinger , 584 F.2d 1325 (4th Cir. 1978), this Court reversed a district court's order forPage 10 “act[ing] prematurely in denying class certification” and “dismiss[ing] . . . the class aspect” of a case 'before discovery had occurred.' Id. at 1332. The Court acknowledged that class certification decisions are reviewed for abuse of discretion and disclaimed any view “as to whether the litigation ultimately should proceed as a class action.” Id. But the Court saw “no circumstance to justify a departure from the normal practice” that class allegations may not be dismissed at the pleading stage unless those allegations show “non-compliance with FRCP 23 as a matter of law.” Id. And because it concluded the relevant class claims did “not on their face fail to meet the requirements of Rule 23(a),” the Court vacated the district court's decision in relevant part. Id. ; accord International Woodworkers of Am. v. Chesapeake Bay Plywood Corp. , 659 F.2d 1259, 1267-68 (4th Cir. 1981) (also holding that a district court “acted prematurely in denying class certification” and stating that such issues should “seldom” be resolved before discovery).

True, there have been significant developments in both pleading and class action law since Goodman . But “we do not lightly presume that the law of this circuit has been overturned,” even when the relevant precedent is old or predates a great deal of relevant authority. Doe v. Sidar , 93 F.4th 241, 245 (4th Cir. 2024) (alterations and quotation marks removed). We see nothing in Goodman that cannot be read 'harmoniously' with more recent Supreme Court decisions. Taylor v. Grubbs , 930 F.3d 611, 619 (4th Cir. 2019) (quotation marks removed).²

Page 11In addition, Goodman is no outlier. In Mills v. Foremost Insurance Co. , 511 F.3d 1300 (11th Cir. 2008), the Eleventh Circuit concluded that a 'district court abused its discretion in determining, at th[e] complaint-pleading stage in the litigation, that class action treatment” was “inappropriate” because such a determination could not be made “as a matter of law from the face of the . . . complaint.' Id. at 1311. And in Manning v. Boston Medical Center Corp. , 725 F.3d 34 (1st Cir. 2013), the First Circuit vacated as “prematur[e]” a district court order striking class allegations because it was not “obvious from the pleadings that the proceeding cannot possibly move forward on a classwide basis.” Id. at 59–60. Like Goodman , both decisions acknowledged that appellate courts review class certification decisions for abuse of discretion and that district courts may sometimes decide class certification before discovery has occurred. See id. at 59; Mills , 511 F.3d at 1309. But—again, like Goodman —both decisions state that a district court acts 'premature[ly]' and thus commits legal 'error[]' by deciding the class certification question at the pleading stage unless that issue can “be readily resolved by the complaint alone.” Mills , 511 F.3d at 1308-09; accord Manning , 725 F.3d at 60 (emphasizing 'the prematurity of the district court's decision'); see also 7AA Wright & Miller's Federal Practice & Procedure § 1785.3 (3d ed. 2025) ('As a practical matter, the court's Page 12determination [whether to certify a class] under subdivision (c)(1) usually should be predicated on more information than the complaint itself affords.”).

Goodman ’s conclusion that the timing of a class certification decision impacts the “legal standards” a district court must employ in making that decision is consistent with other areas of law. Martin v. Franklin Cap. Corp. , 546 U.S. 132, 139 (2005). For example, even though a district court’s factual findings after a bench trial or an evidentiary hearing are reviewed only for clear error, see, e.g. , Fed. R. Civ. P. 52(a)(6), the court has no authority to make such findings when resolving motions to dismiss or motions for summary judgment, see, e.g. , Alexander v. Connor , 105 F.4th 174, 178 (4th Cir. 2024). The same is true when a district court is asked to dismiss a defendant for lack of personal jurisdiction. “If the existence of jurisdiction turns on disputed factual questions[,] the court may resolve the challenge on the basis of a separate evidentiary hearing,” Combs v. Bakker , 886 F.2d 673, 676 (4th Cir. 1989), and any findings made after that hearing are reviewed for clear error, see Fed. R. Civ. P. 52(a)(6). In contrast, when a district court acts without a hearing or taking evidence, the question is whether the complaint contains “a sufficient prima facie showing to survive the jurisdictional challenge” and the court “act[s] prematurely” if it dismisses for lack of personal jurisdiction when that standard is satisfied. Combs , 886 F.2d at 676–77.

Nor is Goodman inconsistent with our repeated statements that Rule 23 grants district courts “broad discretion in deciding whether to certify a class” and our recognition of 'the considerable advantages that our district court colleagues possess in managing complex litigation.' Lienhart v. Dryvit Sys., Inc. , 255 F.3d 138, 146 (4th Cir. 2001)Page 13 (quotation marks removed) (first quote); Mr. Dee’s Inc. v. Inmar, Inc. , 127 F.4th 925, 929 (4th Cir. 2025) (quotation marks removed) (second quote). It is well established that particular “channel[s] of discretion” can “narrow[]” over time, Halo Elecs., Inc. v. Pulse Elecs., Inc ., 579 U.S. 93, 104 (2016) (quotation marks removed) (quoting Henry J. Friendly, Indiscretion About Discretion , 31 Emory L.J. 747, 772 (1982)), and that a trial court's failure to operate within the constraints imposed by such narrowing constitutes reversible error, see, e.g. , Doe , 93 F.4th at 247 (reversing a district court order “involving management of the trial process” under an abuse-of-discretion standard because the district court’s order “strayed from the legal constraints on its discretion” imposed by previous appellate decisions (quotation marks removed)). We conclude that Goodman and similar decisions announce 'legal standards' that 'limit[]' a district court's 'discretion.' Martin , 546 U.S. at 139; cf. Gregory v. Finova Cap. Corp. , 442 F.3d 188, 190 (4th Cir. 2006) (“[T]o be affirmed, the district court must exercise its discretion within the framework of Rule 23.” (quotation marks removed)). Accordingly, we reiterate that, just as a district court may never grant class certification based on the face of the complaint, see, e.g. , Wal -Mart Stores, Inc. v. Dukes , 564 U.S. 338, 350–51 (2011), a court may deny class certification at the same stage only if the complaint’s class action allegations show “non-compliance with [Rule] 23 as a matter of law.” Goodman , 584 F.2d at 1332.³

Page 14Applying those standards here, we conclude the district court acted within its discretion in denying class certification under Federal Rule of Civil Procedure 23(b)(3) but “acted prematurely” by granting Navy Federal’s request to deny class certification under Rule 23(b)(2). Goodman , 584 F.2d at 1332.

Our review is complicated by the fact that neither the district court’s oral ruling nor its later written order is entirely clear about which requirement(s) for class certification the district court deemed lacking. Neither the oral ruling nor the written order directly reference any provision of Rule 23(a) or (b). The court likewise did not use any of the most common shorthand terms for any of those provisions.

On balance, we view the district court’s ruling as homing in on the requirements that are unique to a (b)(3) class: predominance and superiority. For example, the court cited “vari[ations]” in “the circumstances of each plaintiff’s loan application” and the need to evaluate “each individual mortgage applicant . . . on so many different variables.” JA 97– 98, 122; see JA 95 (“[I]t’s apples, oranges, grapefruits and bananas, I mean, you’ve got so many different categories of applicants.”). Such concerns most naturally address whether any common “questions of law or fact . . . predominate over any questions affecting only individual members.” Fed. R. Civ. P. 23(b)(3) (emphasis added); see also Tyson Foods, Inc. v. Bouaphakeo , 577 U.S. 442, 453 (2016) (“The predominance inquiry asks whether the common, aggregation-enabling, issues in the case are more prevalent or important than Page 15the non-common, aggregation-defeating, individual issues.” (quotation marks removed)). The district court also stated “there [were] too many moving parts,” referencing the time class actions take “to get moving” and citing the need “to promote the efficient use of resources and to streamline the claims to be considered in this civil action.” JA 92, 99, 122. Such considerations speak most naturally to whether “a class action is superior to other available methods for fairly and efficiently adjudicating the controversy.' Fed. R. Civ. P. 23(b)(3) (emphases added); see also Amchem Prods., Inc. v. Windsor , 521 U.S. 591, 615 (1997) (predominance and superiority requirements ensure (b)(3) classes 'achieve economies of time, effort, and expense, and promote uniformity of decision as to persons similarly situated, without sacrificing procedural fairness or bringing about other undesirable results” (alterations and quotation marks removed)).

We conclude this is the unusual case in which the district court could determine— based solely on the face of the complaint—that any request to certify a (b)(3) class fails as a matter of law. The nine named plaintiffs alone are residents of five different States who applied for at least four different products and had different outcomes with Navy Federal. The complaint identifies an open-ended class period “from 2018 to the present,” JA 70, while seeking 'restitutionary relief, as well as compensatory, statutory, and punitive damages' on behalf of 'each . . . Class member[],' JA 86. Although a district court “usually should consider more information than the bare allegations of the complaint” in making such a determination, Goodman , 584 F.2d at 1331, this is the rare circumstance where a “failure of predominance” and/or superiority “is readily apparent from a reading of the . . . plaintiffs' complaint.' Jackson v. Motel 6 Multipurpose, Inc. , 130 F.3d 999, 1006Page 16 (11th Cir. 1997); see also Pilgrim v. Universal Health Card, LLC , 660 F.3d 943, 945–49 (6th Cir. 2011) (similar). We therefore affirm the district court's order to the extent it determined that the complaint failed to make out “a sufficient prima facie showing” to justify a (b)(3) class. Combs , 886 F.2d at 676.

That is not the end of the matter, however. The complaint also seeks declaratory and injunctive relief and to represent a class under Rule 23(b)(2). And, unlike (b)(3) classes, (b)(2) classes need not satisfy a predominance or superiority requirement. See Wal -Mart , 564 U.S. at 360-62; cf. Mills , 511 F.3d at 1308-09 ('[A] lack of predominance under Rule 23(b)(3) does not automatically bar class certification because the putative class representative can still attempt to satisfy the requirements of Rule 23(a) and either Rule 23(b)(1) or (b)(2).”).

Navy Federal asks us to affirm the denial of (b)(2) certification as well, insisting 'the district court made clear, both at the hearing and in its opinion, why one of ' the requirements for certifying any class-'commonality-could not possibly be met.' Navy Federal Br. 11. We disagree.

First, the district court never used the word “commonality,” referenced the provision that imposes that requirement (Rule 23(a)(2)), or cited the Supreme Court decision ( Wal -Mart ) that forms nearly the entire basis of Navy Federal's commonality argument. And, as just explained, the district court’s stated reasons sound more in predominance and superiority than in commonality.

Second, we cannot say that the complaint fails to make a prima facie commonality showing as a matter of law. The complaint alleges that Navy Federal 'requires every Page 17applicant to fill out” a single form that collects various categories of information that “can be proxies for race.” JA 44 (emphasis added). The complaint further alleges that Navy Federal 'runs the data from the application . . . through its proprietary underwriting algorithm”—singular—“to determine a person’s creditworthiness” and decide “whether to lend to a particular borrower and on what terms.” JA 45–46. Finally, the complaint alleges that this “at-least semi-automated underwriting process”—again, singular—generates “a uniquely discriminatory result.” JA 51, 53.

These allegations suggest several 'common questions of law or fact' that are “capable of class-wide resolution” and whose answers “will resolve an issue that is central to the validity of each one of the claims in one stroke.” Wal -Mart , 564 U.S. at 350. For example, does Navy Federal use a single algorithm as part of its process for evaluating every loan applicant regardless of the underlying product? Compare JA 51 (asserting that Navy Federal uses “an at-least semi-automated underwriting process, characterized by a formula it refuses to share”), with Mills , 511 F.3d at 1309 n.14 (citing courts of appeals decisions that “reversed denials of class certification that were made without opportunity for discovery . . . when the satisfaction of the Rule 23 requirements may have depended on factual matters within the knowledge or possession of the defendant'). If so, does that algorithm—as opposed to some other variable(s)—produce the disparate impacts based on race that are alleged in the complaint? Cf. Reyes v. Waples Mobile Home Park Ltd. P’Ship , 903 F.3d 415, 424 (4th Cir. 2018) (noting that, to prove a disparate impact claim under the Fair Housing Act, “the plaintiff must demonstrate a robust causal connection between the defendant’s challenged policy and the disparate impact on the protected class”). If so, is Page 18Navy Federal’s use of that algorithm justified by some interest that would defeat all the applicants' claims? Depending on their resolution, these common questions could 'produce a common answer to the crucial question why was [ each class member ] disfavored. ” Wal -Mart , 564 U.S. at 352.

There is no conflict between our holding and the oft-repeated principle that “Rule 23 does not allow for a 30,000 foot view of commonality.' Stafford v. Bojangles’ Rests., Inc. , 123 F.4th 671, 680 (4th Cir. 2024) (alterations and quotation marks removed). The applicants neither “rely on nebulous references to systemic failures or systemic deficiencies,” nor do they assert that “a laundry list of factually diverse claims . . . prove[s] the existence of a uniform company policy.' Id. at 680 (quotation marks removed). Instead, the complaint, fairly read, asserts that Navy Federal uses a single algorithm to evaluate every applicant. See JA 53 (asserting that 'Navy Federal maintains secrecy over its residential loan algorithm”); see also Wal-Mart , 564 U.S. at 353 (reaffirming that if an employer “used a biased testing procedure to evaluate both applicants for employment and incumbent employees, a class action on behalf of every applicant or employee who might have been prejudiced by the test clearly would satisfy the commonality and typicality requirements of Rule 23(a)' (quotation marks removed) (quoting Falcon , 457 U.S. at 159 n.15)).

“Of course, discovery in this case might show that the [complaint’s] allegations . . . are false” and that Navy Federal employs a process that is neither uniform nor results in disparate impacts based on a loan applicant’s race. National Rifle Ass’n of Am. v. Vullo , 602 U.S. 175, 195 (2024); cf., e.g. , Tabor v. Hilti, Inc. , 703 F.3d 1206, 1229 (10th Cir.

Page 192013) (affirming a district court's post-discovery no-commonality finding where 'the record suggest[ed] that” the defendant “failed to maintain the [challenged process] in any uniform matter”). But as we have already explained, that is not the question at this stage. Instead, we hold that the complaint contains “a sufficient prima facie showing” to establish commonality and that the district court 'acted prematurely' in concluding otherwise. Combs , 886 F.2d at 676. We thus vacate the district court’s order to the extent that it denied class certification under Rule 23(b)(2).⁴

* * *

The district court’s order is Page 20RICHARDSON, Circuit Judge, concurring in the judgment in part and dissenting in part:

AFFIRMED IN PART AND VACATED IN PART .

Class actions are high-stakes affairs. So district courts often take their time to develop the record before making classwide decisions. But a district court need not go through the motions just for appearances. When the deficiency of a proposed class is apparent from the face of the pleadings such that further factual development would be of doubtful use, a district court is not required to sit on its hands. It may instead strike class allegations immediately. That course of action will not be the norm. But this case, involving a class action suit against Navy Federal Credit Union for discriminatory mortgage lending policies, is an outlier.

The majority agrees in part, finding that the district court properly rejected the Rule 23(b)(3) class based on the face of the complaint under Rule 23(c)(1)(A). I take a different path, trusting the district court’s discretion to manage its docket under Rule 23(d)(1)(D). Under this approach, I would hold that the district court permissibly struck all the class allegations, Rule 23(b)(3) and (b)(2) allegations alike.

Navy Federal Credit Union is the nation’s largest credit union, providing a host of financial services to 13 million members of the military and their families. Among its products are mortgage loans, including conventional purchase-money mortgages, rate-andterm refinances, cash-out refinances, VA-backed mortgages, and home equity lines of credit.

When deciding whether—and on what terms—to approve a mortgage loan for a prospective borrower, Navy Federal asks borrowers to fill out a Uniform Residential Loan Page 21Application, a standardized mortgage application form used by most mortgage lenders. That form asks for many details, including the borrower’s “name, SSN, citizenship, subject property address, property value, intended occupancy, former addresses, mailing addresses, employment information, income information, asset information, liabilities and expenses, and military service.” J.A. 44. This information, along with other inputs, goes through a “semi-automated process” that Navy Federal uses to decide whether to lend to an applicant and at what interest rate. J.A. 51.

Like all large mortgage lenders, Navy Federal has to submit data on its mortgage lending practices to the government under the Home Mortgage Disclosure Act. See 12 U.S.C. § 2801 et seq. The submitted information is then used in two different ways. Internally, economists at the National Credit Union Administration ('NCUA'), the government agency responsible for regulating credit unions, analyze the raw data and issue industry-wide reports. See Nat'l Credit Union Admin., Off. of the Chief Economist, Observations on Credit Unions' Mortgage Lending to Minority Borrowers (2022) (“NCUA Report”). Externally, a stripped-down version of the same data—anonymized, grouped by credit union, and missing certain key metrics like applicant credit score—is released to the public for use by independent groups in their own research and analysis. One recent independent study was performed by CNN. See Casey Tolan, Audrey Ash & Rene Marsh, The nation's largest credit union rejected more than half its Black conventional mortgage applicants , CNN (Dec. 14, 2023) (“CNN Report”).

Page 22Both the NCUA’s 2022 internal analysis and CNN’s 2023 analysis show disparities in mortgage lending outcomes by race.¹ The NCUA Report analyzed mortgage lending in 2020 and 2021 across the credit union industry as a whole and did not focus on Navy Federal. It found that “the estimated denial likelihood was higher for minority applicants . . . [by] about 1.2 to 1.9 times the rate of White borrowers,” with the likelihood varying depending on whether the minority group was Black, Hispanic, or Asian. NCUA Report at 2, 5. At the same time, the NCUA Report warned that its analysis 'should not be interpreted as evidence of discrimination,” because the data submitted, and thus used for its analysis, 'do[es] not include important credit-risk-related factors.' Id. at 4.

The CNN Report, by contrast, specifically analyzed Navy Federal's mortgage lending. It found that “more than 75% of the White borrowers who applied for a new conventional home purchase mortgage in 2022” were approved by Navy Federal, but “less than 50% of Black borrowers” were approved. CNN Report. Like the NCUA Report, however, CNN’s findings were not uniform by minority group. For example, nearly 70% of Asian applicants were approved—an approval rate 20 percentage points higher than that for Black applicants. Id. And, like the NCUA Report, CNN's analysis contained caveats to its findings. Because CNN relied on the stripped-down dataset, its analysis failed to control for an even broader set of factors affecting credit risk, such as applicant “credit Page 23score, available cash deposits and relationship history with lender,' none of which are 'available in the public mortgage data.' Id.

A few weeks after the CNN Report was published, Laquita Oliver sued Navy Federal in the Eastern District of Virginia. Other Plaintiffs filed their own suits within the next few months, and these cases were eventually consolidated into a single class-action suit.

The operative class-action complaint, filed on February 20, 2024, now names nine Plaintiffs, all minority Navy Federal members (eight Black, one Hispanic) who applied to Navy Federal for a mortgage-related financial product in the past five years: Laquita Oliver, Bob Otondi, Dennis Walker, Constantina Batchelor, John Jackson, Charles Gardner, Carl Carr, Marie Pereda, and Christina Hill. Plaintiffs claim that Navy Federal’s mortgage lending practices violate federal and state civil rights laws, both by disparately treating and by disparately impacting non-White mortgage-product applicants in comparison to White mortgage-product applicants.²

By filing a class-action complaint, the nine Plaintiffs seek to represent not only themselves, but the class of:

All minority residential loan applicants from 2018 through the present . . . who submitted an application for any home mortgage loan to [Navy Federal], who sought to refinance or modify a home mortgage loan through [Navy Federal], and/or who sought a Home Equity Line of Credit from [Navy Federal] and whose application was: (a) denied; (b) approved at higher

Page 24interest rates and/or subject to less favorable terms as compared to similarly situated non-minority applicants; or (c) processed at a rate slower than the average processing time of applicants submitted by similarly situated nonminority applicants.

J.A. 70. According to the complaint, the proposed class is suitable for class adjudication because 'Navy Federal's underwriting algorithms or machine learning programs were racially biased and led to unfairly discriminatory credit policies that harmed' all class members. J.A. 73. But beyond asserting that Navy Federal uses a 'proprietary underwriting algorithm to determine a person’s creditworthiness,” the complaint says little about the method by which Navy Federal underwrites its catalog of mortgage products. J.A. 44–46.

Navy Federal moved to dismiss the whole complaint, disparate-treatment and disparate-impact claims alike, under Federal Rule of Civil Procedure 12(b)(6). In the alternative, Navy Federal moved to strike the class allegations from the pleadings under Rules 12(f) and 23(d)(1)(D), which would prevent Plaintiffs from seeking relief on behalf of the proposed class but would allow them to seek relief for themselves.

The district court granted Navy Federal's motion to dismiss with respect to the disparate-treatment claims. Among other reasons, the district court explained that “the Complaint has failed to allege plausible direct or circumstantial evidence of discriminatory intent.” Oliver v. Navy Fed. Credit Union , 2024 WL 2786905, at *7 (E.D. Va. May 30, 2024).

The district court did not dismiss the disparate-impact claims, because disparateimpact claims do not require discriminatory intent. The district court instead held that “[a]t Page 25the motion to dismiss stage . . . the statistical disparities reveal a disparate impact among non-white loan applicants and the underwriting algorithm and process is alleged to have caused the disparity.” Id. at *8. The district court thus let the nine Plaintiffs proceed on their own behalf.

But the district court prohibited Plaintiffs from representing the proposed class, striking the complaint's class allegations under Rules 12(f)(2) and 23(d)(1)(D). Id. at *11. Plaintiffs then petitioned this Court under Rule 23(f) for interlocutory review of the district court’s decision to strike the class allegations. They did not appeal the dismissal of the disparate-treatment claims or any other aspect of the district court’s decision. A prior panel of this Court granted the petition. See Microsoft Corp. v. Baker , 582 U.S. 23, 34 n.7 (2017) (noting that an interlocutory appeal may be taken from an order striking class allegations).

Amici filed two briefs: one from the African American Credit Union Coalition, the other from America's Credit Unions, the Chamber of Commerce, and the Mortgage Bankers Association. Both support Navy Federal.

Class actions are “an exception to the usual rule that litigation is conducted by and on behalf of the individual named parties only.” Califano v. Yamasaki , 442 U.S. 682, 700– 01 (1979). They permit “large numbers of similar claims” to be settled in a “single, binding adjudication” brought by a small number of named plaintiffs on behalf of a larger class of absent persons.¹ Joseph M. McLaughlin, McLaughlin on Class Actions § 1:1 (21st ed. 2024). By avoiding “duplicative” suits, class actions can “achieve economies of time,”Page 26 save 'effort and expense,' and yield a 'uniformity of decisions' for all class members. Id. These efficiencies benefit litigants and courts alike.

But the benefits of class actions come with risks. The “most significant” risk is that the many absent persons will not have their claims fully and fairly represented by the few named plaintiffs. Gen. Tel. Co. of Sw. v. Falcon , 457 U.S. 147, 161 (1982) (quotation omitted). In class actions seeking primarily injunctive relief, “[t]he legal rights of absent class members . . . are resolved regardless of either their consent, or . . . their express wish to the contrary.” Ortiz v. Fibreboard Corp. , 527 U.S. 815, 847 (1999); see also Fed. R. Civ. P. 23(b)(1)–(2). Come judgment, absent class members must live with the outcome, win or lose, no matter how poorly the named plaintiffs litigated on their behalf. So “if the framing of the class is overbroad,” an absent member may never be able to vindicate his surefire claim because the named plaintiffs took a dissimilar long shot and missed. Falcon , 457 U.S. at 161.

Class actions pose a risk to defendants too. In class actions seeking primarily damages as relief, the sheer number of aggregated claims often poses “the possibility of colossal liability” for a defendant in the event of a loss. Coinbase, Inc. v. Bielski , 599 U.S. 736, 743 (2023); Fed. R. Civ. P. 23(b)(3). This threat of financial ruin creates a “potential for coercion.” Id. Plaintiffs may have an incentive to bring large but unmeritorious class actions against defendants who cannot afford to bet the farm and are thus “forced to settle” in what have been called 'blackmail settlements.' Id. (quoting Henry J. Friendly, Federal Jurisdiction: A General View 120 (1973)).

Page 27Rule 23(a) of the Federal Rules of Civil Procedure attempts to mitigate these risks. It 'ensures that the named plaintiffs are appropriate representatives of the class whose claims they wish to litigate” by requiring a proposed class to satisfy four requirements. Wal-Mart Stores, Inc. v. Dukes , 564 U.S. 338, 349 (2011). Before the named plaintiffs “may sue or be sued as representative parties on behalf of” their proposed class, they must show that: “(1) the class is so numerous that joinder of all members is impracticable; (2) there are questions of law or fact common to the class; (3) the claims or defenses of the representative parties are typical of the claims or defenses of the class; and (4) the representative parties will fairly and adequately protect the interests of the class.” Fed. R. Civ. P. 23(a) (emphasis added). These requirements—known as numerosity, commonality, typicality, and adequacy—“effectively limit the class claims to those fairly encompassed by the named plaintiff's claims.' Falcon , 457 U.S. at 156 (quotation omitted). By prohibiting overbroad classes, Rule 23(a) protects absent members and defendants alike.³

The power and responsibility of applying these rules, and thus safeguarding the class-action device itself, fall to the district court. A district court may certify a class action only if it determines, “after a rigorous analysis, that the prerequisites of Rule 23(a) have been satisfied.' Id. at 161. 'A party seeking class certification' therefore 'must be prepared to prove that there are in fact sufficiently numerous parties, common questions of law or fact, etc.' Dukes , 564 U.S. at 350 (emphasis in original). Unless the named Page 28plaintiffs “affirmatively demonstrate [their] compliance with the Rule,” the district court must bar their attempt to proceed on behalf of the proposed class. Id.

The district court below determined that the nine named Plaintiffs’ proposed class failed to comply with Rule 23. It accordingly struck Plaintiffs' class allegations and required them to amend the complaint. Plaintiffs now appeal, asking us to reinstate their class allegations. While the majority affirms the district court’s decision to strike the Rule 23(b)(3) allegations, it reinstates the Rule 23(b)(2) allegations. I would affirm the district court’s decision to strike all of the allegations.

Before assessing the district court’s decision, I start with what it means to strike class allegations. Although district courts have often used this procedural device, the courts of appeals have seldom weighed in on how to understand it. As a result, substantial disagreements persist among the district courts over how striking class allegations works. See Timothy A. Daniels, Challenging Class Certification at the Pleading Stage , 56 S. Tex. L. Rev. 241, 244 (2014) (noting that 'courts have been less than consistent' when determining whether class allegations should be “stricken”). For example, district courts do not agree on when a party can move to strike. Compare Hernandez v. Newrez, LLC , 2024 WL 1016365, at *1 (E.D. Pa. Mar. 8, 2024) (entertaining a motion to strike before class certification), with Lengen v. Gen. Mills, Inc. , 185 F. Supp. 3d 1213, 1223 (E.D. Cal. 2016) (holding off on all motions to strike until the class-certification stage). Nor do they agree on what criteria to apply when deciding whether to strike. Compare Duran v. Henkel of Am., Inc. , 450 F. Supp. 3d 337, 357 (S.D.N.Y. 2020) (striking class allegations only Page 29when “the inquiry would not mirror the class certification inquiry” (quotation omitted)), with Ellison v. Inova Health Care Servs. , 692 F. Supp. 3d 548, 566 (E.D. Va. 2023) (striking class allegations when 'it is apparent . . . that the purported class cannot be certified'). Nor do they agree on the source of their authority to strike. Compare Thompson v. Procter and Gamble Co. , 2018 WL 5113052, at *5 (S.D. Fla. Oct. 19, 2018) (striking class allegations under Rule 12(f)), with In re Railway Indus. Emp. No-Poach Antitrust Lit. , 395 F. Supp. 3d 464, 495 (W.D. Pa. 2019) (striking class allegations under Rule 23(d)(1)(D)). A clearer picture is needed.

I begin with what the Supreme Court has told us: 'An order striking class allegations is 'functional[ly] equivalent' to an order denying class certification.' Microsoft Corp. , 582 U.S. at 34 n.7 (quoting Scott v. Family Dollar Stores, Inc. , 733 F.3d 105, 110– 11 n.2 (4th Cir. 2013)). That’s why an order striking class allegations is appealable under Rule 23(f) even though Rule 23(f) permits only 'an appeal from an order granting or denying class-action certification” and does not mention orders striking class allegations. Id. at 31 (quoting Fed. R. Civ. P. 23(f)). While the Supreme Court has blessed orders striking class allegations, it has not explained the basis for such orders.

I would locate the source of a district court’s authority to strike class allegations that fail to propose a satisfactory class in Rule 23(d)(1)(D). Start with the text. This provision permits a district court to “ require that the pleadings be amended to eliminate allegations about representation of absent persons and that the action proceed accordingly .” Fed. R. Civ. P. 23(d)(1)(D) (emphasis added). Granting a motion to strike does just that: It requires Page 30the elimination of allegations from the pleadings.⁴ And the Rule makes clear that a strike is not simply a post-certification housekeeping tool: A strike pursuant to Rule 23(d)(1)(D), not any prior ruling, requires a party to cut certain allegations from the complaint so that “the action proceed[s] accordingly.” In this case, the action will “proceed” as an individual action.

Rule 23’s structure confirms this reading. The strike provision is housed under Rule 23(d), which gives courts the tools to “Conduct[] the Action” fairly and efficiently. See Fed. R. Civ. P. 23(d). It authorizes the court to issue various orders that are core to the proceedings—indeed, to “determine the course of proceedings”—such as requiring notice and imposing conditions on the representative parties. See Fed. R. Civ. P. 23(d)(1). Moreover, Rule 23(d)(2) explicitly links orders under 23(d)(1) to orders under Rule 16, which governs the district court’s discretion over scheduling, the scope of discovery, and case sequencing. Fed. R. Civ. P. 23(d)(2); see generally Fed. R. Civ. P. 16. A strike under Rule 23(d)(1)(D) likewise focuses on substantive case management decisions: it effectively limits discovery to the individual claims. An order striking class allegations thus fits comfortably within a district court's power under Rule 23(d)(1)(D). See Microsoft Corp. , 582 U.S. at 34 n.7 (accepting that a strike under Rule 23(d)(1)(D) is the “functional[]Page 31 equivalent” of an order denying class certification).⁵ A leading class-action treatise agrees. See McLaughlin, supra , § 3:4 ('Rule 23(d)(1)(D) . . . expressly authorizes a motion to strike class allegations.'); see also Scott , 733 F.3d at 110 n.2 (recognizing that while “[c]lass certification is typically pursued under Rule 23(c),” we review a district court’s grant of a defendant’s motion to dismiss or strike class allegations “pursuant to Rule 12(c), 12(f), and 23(d)(1)(D),' because that strike 'is the functional equivalent of denying a motion to certify the case as a class action”).⁶

The majority agrees that a district court may strike class allegations before discovery based on the complaint, but it treats a strike under Rule 23(d)(1)(D) as merely optional Page 32housekeeping after a denial of class certification under Rule 23(c)(1)(A). See Majority Op. at 9. I agree that a district court may deny class certification based on the complaint. But the authority to strike class allegations is not a mere optional housekeeping matter that follows the denial of class certification.

The majority's approach makes Rule 23(d)(1)(D)'s grant of authority all but meaningless. If a district court’s power to strike class allegations arises only after it has denied class certification, the strike would have no practical effect. What would be the purpose of striking allegations at that point? In a non-class action, when a district court dismisses one cause of action but allows another to proceed, it does not require a new complaint omitting the dismissed claim. The same logic applies in the class-action context. The majority's approach transforms Rule 23(d)(1)(D) into an afterthought. But as discussed above, Rule 23's text and structure suggest that the authority to strike is a meaningful case-management tool entrusted to district courts to ensure that cases develop efficiently and effectively.

The majority’s approach also fails to make sense of the Supreme Court’s instruction that an “order striking class allegations is ‘functionally equivalent’ to an order denying class certification.” Microsoft Corp. , 582 U.S. at 34 n.7 (cleaned up). It would make no sense to describe an ancillary housekeeping tool as the 'functional equivalent' of a substantive power that logically precedes it. The majority tries to square this circle by suggesting that the two are functionally equivalent “because doing the former necessarily entails having done the latter.” Majority Op. at 8. But that reasoning confuses a necessary condition with functional equivalence. Many acts necessarily entail having done Page 33something else, yet we would not describe the two as functionally equivalent. Graduating from law school entails having attended classes; a valid contract entails agreement; succeeding on a lawsuit entails filing a complaint. None of these pairs are functional equivalents. Properly read, the Supreme Court's description of the strike power as “functionally equivalent” to a denial of class certification suggests that Rule 23(d)(1)(D) has real teeth apart from Rule 23(c)(1)(A).

So Rule 23(d)(1)(D) empowers district courts to strike class allegations. But that still leaves open the question of what legal standard governs the decision to strike. As I see it, the district court should strike class allegations under Rule 23(d)(1)(D) only if the court can determine that the proposed class fails Rule 23 given the state of the record and pleadings before it . '[T]he burden of establishing [the Rule 23] requirements rests on the plaintiff.' Doctor v. Seaboard Coast Line R. Co. , 540 F.2d 699, 707 (4th Cir. 1976). Accordingly, class allegations “must be supported in the same way as any other matter on which the plaintiff bears the burden of proof, i.e. , with the manner and degree of evidence required at the successive stages of the litigation.' Lujan v. Defs. of Wildlife , 504 U.S. 555, 561 (1992). As the record grows over the course of litigation, so too does a plaintiff’s burden to justify his proposed class.

Thus, the bar for plaintiffs is at its lowest “[a]t the pleading stage.” Id. Before discovery, class allegations need not rest on extensive factual evidence. They need only be sufficiently definite to enable the district court to conduct its “rigorous analysis” of the proposed class. Falcon , 457 U.S. at 161. Because that rigorous analysis '[f]requentlyPage 34 ‘involves considerations that are enmeshed in the factual and legal issues comprising the plaintiff's cause of action,'' the district court should ordinarily await further factual development before ruling on a motion to strike. Dukes , 564 U.S. at 351 (quoting Falcon , 457 U.S. at 160); see also Int’l Woodworkers of Am., AFL-CIO, CLC v. Chesapeake Bay Plywood Corp. , 659 F.2d 1259, 1268 (4th Cir. 1981) (stating that class certification questions should 'seldom' be resolved on the pleadings).

But “seldom” does not mean “never,” and a low bar is still a bar. When “the issues are plain enough from the pleadings to determine whether the interests of the absent parties are fairly encompassed within the named plaintiff’s claim,” a district court need not “probe behind the pleadings” before striking class allegations. Falcon , 457 U.S. at 160; see also Elson v. Black , 56 F.4th 1002, 1006 (5th Cir. 2023) ('District courts are permitted to [strike class allegations] on the pleadings and before discovery is complete when it is apparent from the complaint that a class action cannot be maintained.'); Mills v. Foremost Ins. , 511 F.3d 1300, 1309 (11th Cir. 2008) ('In some instances, the propriety vel non of class certification can be gleaned from the face of the pleadings.'). In other words, class discovery is unnecessary when the proposed class is facially deficient. And a proposed class is facially deficient when a 'central defect in [the] class claim' plainly precludes satisfaction of Rule 23 and it is doubtful that further factual development will cure it. Pilgrim v. Universal Health Card, LLC , 660 F.3d 943, 949 (6th Cir. 2011); see, e.g. , Elson , 56 F.4th at 1006 (affirming the strike of class allegations where 'different state laws govern[ed] different Plaintiffs’ claims”). A district court need not subject parties to costly Page 35and time-consuming class discovery for naught.⁷ Instead, Rule 23(d)(1)(D) gives them discretion to manage their cases efficiently. Accordingly, we should review a district court’s decision to strike class allegations for abuse of discretion.

The majority sees no room for discretion, holding that a court may deny class certification before discovery under Rule 23(c)(1)(A) only when class allegations fail as a matter of law. Majority Op. at 13. To reach this conclusion, the majority relies heavily on Goodman v. Schlesinger , 584 F.2d 1325 (4th Cir. 1978). But Goodman involved a traditional decision to deny class certification, not an order striking class allegations under Rule 23(d)(1)(D). See id. at 1331 (referring to the district court's 'refus[al] to certify the litigation as a class action”). So, even to the extent Goodman is still good law, its holding would apply only to class certification decisions under Rule 23(c)(1)(A), not to Rule 23(d)(1) motions to strike.⁸

The nine named plaintiffs sought to represent a class of all minority mortgage loan applicants in a suit against Navy Federal for alleged discriminatory lending practices. But the district court did not permit the class allegations to proceed past the pleading stage, explaining that “the circumstances of each plaintiff’s loan application process” were too “varied.” Oliver , 2024 WL 2786905, at *11. At a hearing, the district court elaborated that the proposed class—containing multiple “categories of applicants” who applied for 'different types of mortgage products'-was akin to a collection of 'apples, oranges, grapefruit and bananas.' J.A. 95-96; see also J.A. 98 ('[T]his can't possibly be an appropriate case for class action certification.”). The court accordingly struck the class allegations from the complaint.

Although the district court did not state which provision of Rule 23 the class failed to satisfy, I understand it—as do the parties—to have concluded that the proposed class Page 37lacked Rule 23(a)(2) commonality on its face.⁹ I find no abuse of discretion in that conclusion.

Commonality requires “questions of law or fact common to the class.” Fed. R. Civ. P. 23(a)(2). “That language is easy to misread.” Dukes , 564 U.S. at 349. “What matters to class certification . . . is not the raising of common 'questions' . . . but rather, the capacity of a classwide proceeding to generate common answers .' Id. at 350 (emphasis in original) (quotation omitted). The claims of the proposed class members must “depend upon a common contention” such that “determination of [the contention’s] truth or falsity will resolve an issue that is central to the validity of each one of the claims in one stroke.” Id.

In a suit alleging classwide discrimination, there must be “a common answer to the crucial question why was I disfavored .” Id. at 352. This common answer can take either of two forms. First, a defendant may have applied a “testing procedure” to all members of a putative class, in which case “a class action on behalf of every applicant . . . who might have been prejudiced” by that objective procedure would satisfy commonality. Id. at 353Page 38 (quoting Falcon , 457 U.S. at 159 n.15). Second, a defendant may have “operated under a general policy of discrimination” that infected all “subjective decisionmaking processes” 'in the same general fashion,' in which case commonality would likewise be satisfied. Id. (quotation omitted).

In either case, plaintiffs must 'identify[] the specific . . . practice,' id. at 357 (quoting Watson v. Fort Worth Bank and Tr. , 487 U.S. 977, 994 (1988)), that 'ties all their [class] claims together.' Id. (emphasis added). In other words, a plaintiff must propose 'a sufficiently homogeneous class vis-[à]-vis an identified practice .” Stastny v. S. Bell Tel. and Tel. Co. , 628 F.2d 267, 274 (4th Cir. 1980) (emphasis added). Because the identified practice is the single unifying thread, it matters how evenly that thread is woven throughout the class claims. Commonality will be satisfied only if the identified practice is “uniform” across the class, Dukes , 564 U.S. at 355, inflicting “the same injury” on all members. Id. at 353 (quoting Falcon , 457 U.S. at 157). Absent a single, uniform practice, the class lacks commonality.

When a plaintiff alleges discrimination from an “objective” practice that is “applied evenly and automatically,” it is often straightforward to show that the practice is uniform across all class members. Stastny , 628 F.2d at 274 n.10. Classic examples include “an intelligence or aptitude test” or “an educational requirement.” Id. (citing Griggs v. Duke Power Co. , 401 U.S. 424 (1971)) . 'Both the existence and the 'common reach' of such objectively applied . . . practices are likely to be indisputable from the outset, so that no real commonality problems for class action maintenance ever arise.” Id. With respect to commonality, a well-defined practice implies a well-defined class.

Page 39The inverse is also true: An ill-defined practice implies an ill-defined class. If it is difficult to ascertain the 'existence' and 'common reach' of an alleged practice, that practice is unlikely to serve as the uniform foundation for a “sufficiently homogeneous class.” Id. at 274 & n.10. This is because 'Rule 23 does not allow for [] a 30,000 foot view of commonality.” Stafford v. Bojangles’ Restaurants, Inc. , 123 F.4th 671, 680 (4th Cir. 2024) (quoting In re White , 64 F.4th 302, 314 (D.C. Cir. 2023)). A practice must be identified with enough specificity to allow a court to determine whether it “unite[s]” the class claims.¹⁰ Id. When a plaintiff offers only 'nebulous references to 'systemic failures' or ‘systemic deficiencies,’” courts are rightly “skeptical,” for “[s]uch vagueness may mask a multitude of disparities.” Id. Indeed, '[p]laintiffs may invoke overly general policies precisely because 'they are at a loss for a more specific thread to tie claims together.'' Id. (emphasis added) (quoting In re White , 64 F.4th at 314). When the alleged discriminatory practice is as abstract and amorphous as a cloud, the reality tends to be that no single practice is common to the class. Up close, the cloud dissolves into individual drops of rain.

In this case, Plaintiffs assert that their proposed class is bound by a single practice: Navy Federal’s use of a “proprietary underwriting algorithm.” 11 J.A. 46. They argue that this algorithm is a single process for evaluating credit risk that applies uniformly to all class members. But apart from a handful of “nebulous references,” Stafford , 123 F.4th at 680, complaint reveals exceedingly little about how the algorithm works. Plaintiffs assert that the algorithm takes in all information from a “Uniform Residential Loan Application” and that some of those factors “could be”—though are not definitively—“proxies for race.” J.A. 44, 51. But Plaintiffs then also inform us that the algorithm considers unknown “other information” outside the application, leaving us in the dark about the algorithm’s inputs. J.A. 45. And we are given no hint at all as to what “the algorithm” then does with its inputs to produce racially disparate results.

We are not even told the role the algorithm ultimately plays in Navy Federal's mortgage lending process. Plaintiffs repeatedly assert that Navy Federal “engages in a somewhat automated process' when deciding whether to underwrite mortgages, which means that it relies on both an automated algorithm and on human 'loan officer[s]' who are 'given discretion regarding whether to approve a given loan.' J.A. 44 (emphasis Page 41added); see also J.A. 51 ('semi-automated underwriting process'). But Plaintiffs say nothing about how the algorithm and loan officers interact. Indeed, the gaps in Plaintiffs’ allegations suggest that any disparate impact may be caused not by any underwriting algorithm, but by the individual loan officers exercising their discretion in differing ways— a situation that falls short of commonality. See Dukes , 564 U.S. at 355–56.

The vagueness and ambiguity of Plaintiffs' allegations are symptoms of the underlying problem: There is no 'common answer to the crucial question why was I disfavored .” Id. at 352. No single practice has a “common reach” that uniformly spans the whole proposed class. Stastny , 628 F.2d at 274 n.10. So the proposed class has not “suffered the same injury.” Falcon , 457 U.S. at 157. One should not be fooled by the phrase “the algorithm,” stated in the singular form; one must look to the substance of the allegations, not their labels, “to ensure that commonality is not based on such ‘semantic dexterity.’” Stafford , 123 F.4th at 680 (quoting Brown v. Nucor Corp. , 785 F.3d 895, 909 (4th Cir. 2015)). Though 'the algorithm' sounds much like an 'aptitude test' or 'an educational requirement” that is “applied evenly and automatically” to the whole class, Plaintiffs have given us no reason to believe that is so. Stastny , 628 F.2d at 274 n.10 .

In fact, the substance of Plaintiffs’ own allegations undercuts, rather than supports, the assertion that Navy Federal engages in any single, uniform practice of mortgage underwriting. In their complaint, Plaintiffs rely almost entirely on statistical data from the NCUA and CNN Reports to show classwide disparate impact. Yet, far from establishing uniformity, the two reports reveal that Navy Federal’s underwriting process treats different Page 42portions of the proposed class quite differently.¹² The statistical data vary most markedly across three dimensions of the class: (1) mortgage product, (2) minority group, and (3) adverse effect.¹³

First, any disparate impact appears to differ by mortgage product. The proposed class includes essentially all types of mortgages: conventional mortgages, refinance mortgages, VA-backed mortgages, and home equity lines of credit. As the district court recognized, these products are 'very, very different' from one another. J.A. 96. In particular, VA-backed mortgages—which are partially or fully guaranteed by the United States Department of Veterans Affairs—are unique among mortgage products. They are available only to veterans and their surviving spouses, 38 U.S.C. §§ 3701, 3703, and are governed by statutory and regulatory requirements that set out different interest rates,Page 43 § 3703(c)(1), and unique underwriting standards, 38 C.F.R. § 36.4340. Class members who applied for a conventional first mortgage and class members who applied for a VAbacked loan are thus on different paths, if not on different journeys entirely. This sharp difference between conventional and VA-backed mortgages is borne out in the data. Holding the other two dimensions constant by examining only the denial outcomes for Black applicants, the data show that Navy Federal approved Black applicants only 48% of the time for conventional mortgages but 72% of the time for VA-backed mortgages. CNN Report . 14 This 24-percentage-point gap strongly suggests that Navy Federal uses distinct processes to assess applications for these two types of products.

Second, any disparate impact appears to differ by minority group. The proposed class seeks to represent “[a]ll minority residential loan applicants from 2018 through the present,” sweeping in Black, Hispanic, and Asian applicants, among others. J.A. 70. But these three minority groups experienced very different lending results. Again, the clearest way to see this is to hold the other two dimensions constant. Consider the interest rates offered to approved borrowers for conventional rate-and-term refinance mortgages.¹⁵ Within the credit union industry in 2021—which Plaintiffs allege is dominated by Navy Federal loan data—Black borrowers were offered interest rates that were 1.7 basis points (0.017%) higher on average than those offered to similar White borrowers. NCUA Report at 6. Meanwhile, Asian and Hispanic borrowers were offered rates that were lower than Page 44those offered to similar White borrowers by 3.6 basis points (0.036%) and 0.7 basis points (0.007%) respectively. Id. Or consider the interest rates for conventional purchase-money mortgages that same year. Hispanic borrowers received the least favorable interest rates at 13 basis points higher than those offered to similar White borrowers, while Asian borrowers were again offered rates that were lower by 3 basis points. Id. No uniform picture emerges from this data.¹⁶

Third and finally, any disparate impact appears to differ by adverse effect. The proposed class contains members who experienced three different negative outcomes: having a mortgage application denied, having an application approved at a higher interest rate, and having an application processed more slowly. At a minimum, for the class to be 'disfavored,' Dukes , 564 U.S. at 352, its members should experience these three outcomes more frequently than White applicants do. But the data do not bear that out. Once more, hold the other two dimensions constant and examine only Asian applicants for conventional cash-out refinances in 2021. Although Asian applicants were 1.5 times more likely to be denied by a credit union for that product, they were given lower interest rates than comparable White applicants once approved. NCUA Report at 5–6. A loan process that burdens with one hand the very group of applicants it benefits with the other can hardly serve as a single unifying thread for the class.

Page 45The lack of commonality is perhaps most apparent when considering those class members whose mortgage applications were processed more slowly-an outcome that seems wholly unrelated to the other two. What single practice, uniformly applied, would result in denied mortgages for some but merely slower approvals for others? Indeed, it seems quite unlikely that Navy Federal would consider the same information when deciding whether to deny an application and when deciding how quickly to process it. True, Plaintiffs allege that Navy Federal’s process starts by collecting roughly the same information from every loan applicant. But that hardly means-and Plaintiffs do not allege-that every decision Navy Federal makes uses the same information. A denial decision presumably accounts for a host of inputs that reflect the applicant’s credit risk; a processing-time decision may account for little more than the date of the application and the length of Navy Federal's backlog. Because these decisions are so fundamentally different, 'demonstrating the invalidity of [one adverse decision] will do nothing to demonstrate the invalidity of another[].” Dukes , 564 U.S. at 355–56. Put another way, a denied applicant and a delayed applicant do not share a common question. So they cannot join one common class.

The proposed class's stark variation across all three of these dimensions is illustrated by its assorted medley of members. Look no further than the nine named Plaintiffs themselves to see how unlikely it is for the same underwriting process to apply uniformly to all of them. The nine applied for different mortgage products. '[S]ome plaintiffs applied for a first-lien mortgage,' and 'others applied for a Veterans Administration backed loan, and yet others applied to refinance an existing loan.” Oliver ,Page 46 2024 WL 2786905, at *2. The nine belong to different races. Eight Plaintiffs are listed as 'African American,' but Hill is listed as 'Hispanic / Latina.'¹⁷ Id. The nine allege different adverse effects. Eight Plaintiffs “allege that their loan applications were denied,” but “Batchelor alleges she was approved at a higher interest rate.” 18 Id. The nine have different financial situations. Pereda stated that she made 'several hundred thousand dollars per year,' but Hill put down her income as '$80,000.' Id. The nine received different results from other mortgage lenders. “[S]ix plaintiffs allege they received loans at higher interest rates from other lenders following Navy Federal’s denials,” but two of those denied did not. Id. And the nine may be divided by even more disparities that we cannot discern because they gave Navy Federal different pieces of information. All but Otondi failed to “provide[] the available down payment” on the home they sought; Gardner did not mention his existing debt; “Jackson [did] not provide his income”; and “Hill [did] not provide any indication of her credit score.' Id. This is not a class with much in common. This is 'a bewildering lineup of permutations and combinations.' Kremens v. Bartley , 431 U.S. 119, 130 (1977).

Page 47In the end, I cannot say the district court abused its discretion in concluding that “this can’t possibly be an appropriate case for class action certification.” J.A. 98. The sheer breadth and diversity of the proposed class are apparent on its face. And Plaintiffs’ own data highlight the absence of a “common answer to the crucial question why was I disfavored .” Dukes , 564 U.S. at 352. Confronted with pleadings that make the lack of commonality so unusually clear, I would affirm the district court's strike of class allegations before discovery.¹⁹

* * *

Plaintiffs’ proposed class is marked on its face by deep differences across multiple dimensions. No common thread unites its claims, no common practice covers its members, no common answer resolves its questions. The district court understandably doubted that discovery would cure such a central defect in commonality. So, respecting the district court’s case-management discretion, I would affirm its strike of all class allegations.