Plaintiffs sought class certification to pursue various claims against the Hilton Hotels Retirement Plan for unlawfully denied retirement benefits.  The District Court initially denied class certification, without prejudice, on the basis that the proposed class definition was impermissibly “fail-safe”, while affording the plaintiffs an opportunity to amend.  The District Court then denied the plaintiffs’ renewed motion to certify with a revised definition on the ground that the proposed revised class definition remained impermissibly fail-safe. The District Court noted that other Rule 23(a) issues remained with respect to two of the proposed sub-classes, but, given the defect in the proposed class definition, ‘the Court need not reach them.’

The D.C. Circuit Court of Appeals found that this was an abused of discretion:

“The very character of the fail-safe legal question exists most critically at the early class-certification stage of a case. The crux of the fail-safe critique is that a proposed class definition impermissibly depends on a determination on the merits of the case, so that class membership cannot be effectively identified and represented until the litigation ends. For example, a class defined to include ‘all those discriminated against illegally’ relies critically on a merits determination to set the contours of class membership. But at the class certification stage, a determination on the merits is far down the road, while the need to identify the class for procedural and substantive purposes is immediate. In other words, the fail-safe concern is that the class definition is hopelessly indeterminate at the time the district court is required to resolve class action status and to define class membership — as early as practicable after the proposed class action commences.

“But should a fail-safe class proceed to final judgment, the merits will have been resolved. So the fail-safe concern — however cogent at the class certification stage — becomes muddied, or, at minimum, substantially diluted. If the problem with fail-safe classes is that they rely on merits determinations that are wholly unknown at class-certification time, that problem abates by final judgment. At the very least, it would be a different inquiry on appeal for this court to determine whether the class definition ‘all those defrauded illegally’ is impermissibly fail-safe once a trial court has said whether fraud occurred or not. The question will have shifted. We would be in a strange posture indeed if forced to conclude a class definition was hopelessly indeterminate at time a however legible it has become at time b.

The Court, observing that Rule 23 provides strong protection against circular or indeterminate class definitions, (which the District Court understandably sought to avoid), held that the District Court abused its discretion “by denying the amended class certification motion based on a stand-alone and extra-textual rule against ‘fail-safe’ classes, rather than applying the factors prescribed by Federal Rule of Civil Procedure 23(a).”

Explaining further:

“Courts have identified two main problems with certifying a so-called ‘fail-safe’ class, the membership of which depends on the merits. First, if membership in a class depends on a final resolution of the merits, it is administratively difficult to determine class membership early on. Second, if the only members of fail-safe classes are those who have viable claims on the merits, then class members either win or, by virtue of losing, are defined out of the class, escaping the bars of res judicata and collateral estoppel. Heads they win; tails the defendant lose — at least, that is the concern.

“To illustrate, for a class definition that encompasses ‘all those whom Company X defrauded’, the ‘defrauded’ addendum makes the definition circular. That is, whether or not certain actions constitute fraud, a tortious activity for which Company X would be subject to liability, is just what the litigation is meant to find out. As for res judicata effect, if a defendant is found not to have defrauded anyone, then there would be no class members at all. Every erstwhile class member would, after the merits determination, become a stranger to the case who would not be bound by that litigation loss.

“Those concerns are understandable. In practice, though, a fail-safe class definition is only truly troubling to the extent it hides some concrete defect with the class….  Courts should stick to Rule 23’s specified requirements when making class certification decisions and, in doing so, will likely find any ‘fail-safe’ concerns assuaged.

“Start with Rule 23(a)’s prerequisites. The putative class prosecuting the action — that is the class as defined at the beginning of the case — must be too numerous for individualized litigation to be practicable. That numerosity must exist throughout the litigation. Yet a class that could be defined to have zero members if the plaintiffs lose is not numerous at all.  Similarly, a circular class definition could reveal the lack of a genuinely common issue of law or fact. Plaintiffs may define a class as all those discriminated against illegally because they are at a loss for a more specific thread to tie claims together. But Rule 23 does not allow for such a 30,000 foot view of commonality. Typicality too should be a hard hill to climb if the named plaintiffs might not be members of the class come final judgment. So too for Rule 23(b)(3)’s superiority requirement, since a class action would fail to be a superior device for resolving a dispute if the class would collapse should the plaintiffs lose on the merits. Even more fatal to an indeterminate class definition can be the requirement in Rule 23(c) that the district court ensure up front the ‘binding effect of a class judgment on members.’ All that is to say that the protocol for determining if a class definition is proper is to apply the terms of Rule 23 as written. Doing so should eliminate most, if not all, genuinely fail-safe class definitions.

“For those rare cases (if any) in which a truly ‘fail-safe’ class hurdles all of Rule 23’s requirements, then the problem will in all likelihood be one of wording, not substance. After all, a class of human beings cannot itself be circular. Only a class definition attempting to describe them can. For example, assume a class defined as ‘all workers of Company X employed in its Washington, D.C. and New York City offices between 2021 and 2023 who were unlawfully denied promotion to clerical supervisor due to enforcement of the Company X Skills Test.’ The one word in that definition that makes it fail safe is ‘unlawfully’. By deleting that, the definition loses any fail-safe character and might otherwise pass all of Rule 23’s requirements. Or consider the class ‘all associates employed by Law Firm Y from 2021 to 2023 who were denied their contractual bonus because Law Firm Y refused to credit pro bono hours.’ While the parties may litigate on the merits whether the associates had any contractual right to a bonus, any fail-safe issues at the certification stage could be addressed by simply rephrasing as a counterfactual — that is, ‘who would have received their contractual bonus if Law Firm Y credited pro bono hours.’ The solution for cases like these is for the district court either to work with counsel to eliminate the problem or for the district court to simply define the class itself. Rule 23 charges district courts ultimately with defining the class. Using that tool, the problem of fail-safe classes can and often should be solved by refining the class definition rather than by flatly denying class certification on that basis. So rather than reject a proposed class definition for a readily curable defect based on an unwritten criterion, a district court should either define the class itself or, perhaps most productively, simply suggest an alternate class definition and allow the parties to object or revise as needed.”


In re White, 64 F.4th 302 (D.C. Cir. 2023).