Plaintiffs in this consolidated class action allege copyright infringements arising from defendant publishers’ unauthorized electronic reproductions of plaintiff authors’ written works. The United States District Court for the Southern District of New York (George B. Daniels, Judge) certified the class for settlement purposes and approved a settlement agreement (“Settlement”) over the objection of ten class members (“objectors”). In this appeal, objectors contend that (1) approval of the Settlement was impermissible because it released claims beyond the factual predicate of the case, (2) class certification was improper because subgroups within the class have conflicting interests, and (3) the district court committed procedural errors in certifying the class and approving the Settlement. Although we reject objectors’ arguments regarding the release, we conclude that the district court abused its discretion in certifying the class and approving the Settlement, because the named plaintiffs failed to adequately represent the interests of all class members. We do not reach the procedural challenges, which are moot in light of our class certification holding.
We therefore vacate the district court’s order certifying the class and approving the Settlement, and remand for further proceedings consistent with this opinion.
BACKGROUND
I. Factual Background
Plaintiffs are freelance authors (“authors”) who sold written works to print publishers for publication in newspapers, magazines, and other periodicals. With the rise of the Internet, print publishers like The New York Times began to reproduce authors’ works electronically by placing them in their own online databases and licensing them to appear in electronic databases such as LexisNexis. In response, authors sued the original print and subsequent electronic publishers, alleging in three independent class actions that the unauthorized electronic publication of their works infringed upon their copyrights.
In June 2001, the Supreme Court endorsed authors’ theory of liability, holding-in another case that publishers violate the Copyright Act when they reproduce freelance works electronically without first securing the copyright owners’ permission. N.Y. Times Co. v. Tasini, 533 U.S. 483, 488, 121 S.Ct. 2381, 150 L.Ed.2d 500 (2001). Authors’ three lawsuits, which had been suspended pending Tasini, were consolidated and coordinated with a fourth action in the Southern District of New York. The consolidated class action is brought by 21 named plaintiffs — -each of whom owns at least one copyright in a freelance article — and three associational plaintiffs: the National Writers Union, The Authors Guild, Inc., and the American Society of Journalists and Authors. Defendants include electronic database operators such as Reed Elsevier Inc. (owner of LexisNexis) and Thomson Corporation (owner of Westlaw), as well as newspaper publishers that maintain their own archival databases, such as the New York Times Company and Dow Jones & Company, Inc. (collectively “publishers”). The district court referred the parties to mediation, which began in January 2002. In March 2005, with the assistance of mediators Kenneth Feinberg and Peter Woodin, authors and publishers reached a comprehensive settlement agreement.1
*246The Settlement divides the works at issue (“Subject Works”) into three categories: A, B, and C. Category A covers works that authors registered' with the U.S. Copyright Office in time to be eligible for statutory damages and attorney’s fees under the Copyright Act. See 17 U.S.C. §. 412. At the time of the Settlement, registration cost $30 per work or $30 per group registration covering multiple periodical contributions by one individual over a 12-month period.2 Category B includes works that authors registered before December 31, 2002, but too late to be eligible for statutory damages. These claims are eligible to recover only actual damages suffered by the author and any profits of the infringer that are not duplicative of the actual damages. 17 U.S.C. § 504(b). All other claims fall into Category C and cannot be litigated for, damages purposes unless they are registered with the Copyright Office. 17 U.S.C. § 411(a). If registered, however, these claims — like those in Category B — -would be eligible for awards based on authors’ actual damages and infringers’ profits. Category C claims comprise more than 99% of authors’ total claims. Many authors hold claims in more than one category, each claim based on a separate freelance article they sold for publication.
The Settlement creates a damages formula for each category. Authors holding Category A claims are paid “$1,500 for the first fifteen Subject Works written for any one publisher; $1,200 for the second fifteen Subject Works written for that publisher; and $875 for all Subject Works written for that publisher after the first thirty Subject Works.” Authors of Category B works are paid “the greater of $150 or 12.5% of the original sale price of the Subject Work.” For each Category C claim, authors are paid “[t]he greater of $5 or 10% of the original price of the Subject Work,” except for works sold for amounts over $249. Compensation for any Category C work sold for more than $249 depends on the amount for which it was originally sold: $25 per Subject Work sold for $250 to $999; $40 per Subject Work sold for $1,000 to $1,999; $50 per Subject Work sold for $2,000 to $2,999; and $60 per Subject Work sold for $3,000 or more.
The Settlement caps publishers’ total liability through a provision that the parties refer to as the “C reduction.” If the total of all claims — plus the cost of notice, administration, and attorney’s fees — exceeds $18 million, then the Settlement reduces compensation for Category C claims pro rata until the total compensation is $18 million. If compensation for Category C claims reaches zero but the claims and fees still exceed $18 million, then the Settlement reduces compensation for Category A and B claims pro rata until the claims and fees total hits the $18 million limit.
The Settlement releases publishers from further litigation. The release prohibits authors from barring publishers’ future use of the Subject Works, including the selling or licensing of the works to third-party sublicensees. A class member may choose to opt out' of the release for future use and only grant a release for past use; however, any authors who fail to affirmatively opt out of the future-use release will be deemed to have granted it. Authors who only grant a past-use release receive 65% of the compensation that those who grant past and future releases receive.
*247II. Procedural Posture
In March 2005, upon reaching the Settlement, authors and publishers moved the district court to certify the class3 for settlement purposes and approve the Settlement. Objectors opposed the motion. In September 2005, after rejecting objectors’ arguments, the district court certified the class and approved the Settlement as fair, reasonable, and adequate.
In October 2005, objectors appealed that order and judgment on numerous grounds. Over a dissenting opinion, In re Literary Works in Electr. Databases Copyright Litig., 509 F.3d 116, 128 (2d Cir.2007) (Walker, J., dissenting), a majority of this panel concluded sua sponte that the registration requirement imposed by Section 411(a) of the Copyright Act is jurisdictional, and that the district court lacked subject-matter jurisdiction to approve the settlement of claims for the infringement of unregistered copyrights. Id. at 121-22. Authors and publishers joined in asking the Supreme Court to review that decision.
The Supreme Court issued a writ of certiorari and, in March 2010, reversed the judgment of this court, holding that the district court had jurisdiction over the Settlement because Section 411(a) imposes only a nonjurisdictional precondition to filing a claim. Reed Elsevier, Inc. v. Muchnick,-U.S.-, 130 S.Ct. 1237, 1247, 176 L.Ed.2d 18 (2010). On remand, we ordered the parties to file letter briefs addressing any supplemental authority relevant to the merits, to which we now turn.
DISCUSSION
Objectors appeal several aspects of the district court’s decision. They argue (1) that the Settlement impermissibly releases claims beyond the factual predicate of the case; (2) that class certification was improper because subgroups within the class have conflicting interests; and (3) that the district court erred procedurally in reaching its decision. Although we reject the objections to the release provision, we agree with objectors that not all class members were adequately represented. We decline to reach the procedural issues, which are moot in light of our class certification holding.
I. Release of Claims
The Settlement prohibits claimants from barring future use of the Subject Works, including the selling and licensing of the works to third parties, unless the class member either opts out of the Settlement altogether or exercises his right to bar future use. Objectors assert that this “ ‘irrevocable, worldwide, and continuing’ license” impermissibly releases claims that' are not based on the same factual predicate underlying the claims in this class action.
“Plaintiffs in a class action may release claims that were or could have been pled in exchange for settlement relief.” Wal-Mart Stores, Inc. v. Visa U.S.A., Inc., 396 F.3d 96, 106 (2d Cir.2005). Parties often reach broad settlement agreements encompassing claims not pre*248sented in the complaint in order to achieve comprehensive settlement of class actions, particularly when a defendant’s ability to limit his future liability is an important factor in his willingness to settle. See id.; see also TBK Partners,. Ltd. v. Western Union Corp., 675 F.2d 456, 460 (2d Cir. 1982). Any released claims not presented directly in the complaint, however, must be “based on the identical factual predicate as that underlying the claims in the settled class action.” TBK Partners, 675 F.2d at 460.
Objectors argue that releasing future claims arising from licensing the Subject Works to third-party sublicensees is impermissible in two ways. First, future infringements are distinct harms giving rise to independent claims of relief, with factual predicates that are different from authors’ past infringement claims. Second, future claims may be against a sublicensee who is not a party to the Settlement, which means that infringement could not be grounded in the factual predicate of this case. We find both of these, arguments .unavailing because future use of the Subject Works, whether by publishers or by sublicensees, falls squarely within the factual predicate underlying authors’ claims.4
Objectors’ first argument fails to recognize that the consolidated complaint seeks injunctive relief for future uses, and therefore contemplates these alleged future injuries. Put another way, a trial of this case would determine whether it is permissible for publishers to continue to sell and license the works. See Nat’l Super Spuds, Inc. v. N.Y. Mercantile Exch., 660 F.2d 9, 17-18 (2d Cir.1981) (assessing permissibility of release by looking to possible remedies if that case had proceeded to trial). Accordingly, regardless of whether future infringements would be considered independent injuries, the Settlement’s release of claims regarding future infringements is not improper.5 See, e.g., Uhl v. Thoroughbred Tech. & Telecomms., Inc., 309 F.3d 978, 982, 984-85 (7th Cir.2002) (permitting settlement that required all class members to provide an easement in resolving trespass'action).
Objectors’ second argument — that the Settlement impermissibly releases claims against persons and entities not involved in this case — takes an overly narrow view of the factual predicate of authors’ claims. The consolidated complaint alleges that publishers electronically displayed, sold, and distributed the Subject Works. In response, publishers have maintained that the rights that the print publishers purchased from authors include the rights to maintain their issues online and to sublicense those issues to third-party databases. Apart from their argument, rejected in Tasini, that this right exists pursuant to Section 201(c) of the Copyright Act, publishers argued throughout the settlement process that freelance contributors — who knew that the print publications for which they wrote published their content online and delivered it to database publishers— *249granted implied licenses for such electronic distribution. Trial of this case would thus determine the rights of third parties to obtain sublicenses. We therefore conclude that the Settlement’s release pertaining to future uses by publishers and their sublicensees was permissible.
II. Adequacy of Representation
The party seeking to certify a class bears the burden of satisfying Rule 23(a)’s four threshold requirements: (1) numerosity (“the class is so numerous that joinder of all members is impracticable”), (2) commonality (“there are questions of law or fact common to the class”), (3) typicality (“the claims or defenses of the representative parties are typical of the claims or defenses of the class”), and (4) adequacy of representation (“the representative parties will fairly and adequately protect the interests of the class”). Fed.R.Civ.P. 23(a). The district court must also find that the action can be maintained under Rule 23(b)(1), (2), or (3). Before approving a class action settlement, the district court must assess its substance and conclude that it is “fair, reasonable, and adequate.” Fed.R.Civ.P. 23(e)(2). The district court did so here, approving a settlement-only class under Rule 23(b)(3) after concluding that common questions predominate over individual ones and that a class action is superior to other methods of adjudicating the matter.
We review a district court’s decision to certify a class for abuse of discretion. Joel A. v. Giuliani, 218 F.3d 132, 139 (2d Cir.2000). A district court “ ‘abuses’ or ‘exceeds’ its discretion when (1) its decision rests on an error of law (such as application of the wrong legal principle) or a clearly erroneous factual finding, or (2) its decision — though not necessarily the product of a legal error or a clearly erroneous factual finding — cannot be located within the range of permissible decisions.” In re Holocaust Victim Assets Litig., 424 F.3d 158, 165 (2d Cir.2005) (quoting Zervos v. Verizon N.Y., Inc., 252 F.3d 163, 169 (2d Cir.2001)). When a court is asked to certify a class and approve its settlement in one proceeding, the Rule 23(a) requirements designed to protect absent class members “demand undiluted, even heightened, attention.” Amchem Prods., Inc. v. Windsor, 521 U.S. 591, 620, 117 S.Ct. 2231, 138 L.Ed.2d 689 (1997).
Objectors argue that the Settlement contravenes Rule 23(a)(4) because the named plaintiffs failed to adequately represent the interests of class members who hold only Category C claims (“Category C-only plaintiffs”). “The adequacy inquiry under Rule 23(a)(4) serves to uncover conflicts of interest between named parties and the class they seek to represent.” Amchem, 521 U.S. at 625, 117 S.Ct. 2231. To satisfy Rule 23(a)(4), the named plaintiffs must “possess the same interests] and suffer the same injuries] as the class members.” Id. at 625-26, 117 S.Ct. 2231 (quoting E. Tex. Motor Freight Sys., Inc. v. Rodriguez, 431 U.S. 395, 403, 97 S.Ct. 1891, 52 L.Ed.2d 453 (1977)) (internal quotation marks omitted). “Adequacy is twofold: the proposed class representative must have an interest in vigorously pursuing the claims of the class, and must have no interests antagonistic to the interests of other class members.” Denney v. Deutsche Bank AG, 443 F.3d 253, 268 (2d Cir.2006). Not every conflict among subgroups of a class will prevent class certification—the conflict must be “fundamental” to violate Rule 23(a)(4). See In re Flag Telecom Holdings, Ltd. Sec. Litig., 574 F.3d 29, 35 (2d Cir.2009). Where such a conflict does exist, it can be cured by dividing the class into separate “homogeneous subclasses ... with separate representation to eliminate conflicting interests *250of counsel.” Ortiz v. Fibreboard Corp., 527 U.S. 815, 856, 119 S.Ct. 2295, 144 L.Ed.2d 715 (1999); see also Fed.R.Civ.P. 23(c)(5) (“When appropriate, a class maybe divided into subclasses that are each treated as a class under this rule.”).
According to objectors, there was such a conflict here: the named plaintiffs, who hold combinations of all three categories of claims, favored the fewer and more lucrative Category A and B claims over the Category C claims. A subclass of plaintiffs owning unregistered claims should therefore have been carved out of the class, objectors argue. Publishers and authors vigorously defend the Settlement and the adequacy of named plaintiffs’ representation.
A.
We begin our analysis by turning to a pair of Supreme Court decisions that set the contours of the adequacy of representation inquiry in the settlement-class context. In Amehem, Products, Inc. v. Windsor, 521 U.S. 591, 117 S.Ct. 2231, 138 L.Ed.2d 689 (1997), the Supreme Court affirmed the Third Circuit’s decision to vacate a class certification intended “to achieve global settlement of current and future asbestos-related claims.” Id. at 597, 117 S.Ct. 2231. The proposed settlement-only class encompassed hundreds of thousands, and possibly even millions, of individuals who had been exposed to asbestos products manufactured by any of 20 companies. Id. Objectors to the settlement opposed the aggregation into a single class of both class members who had already manifested asbestos-related injuries and those who had been exposed to asbestos but had not yet shown signs of injury. Id. at 607-08, 117 S.Ct. 2231. The Court agreed that “the interests of those within the single class” were “not aligned”: holders of present claims were interested in “generous immediate payments,” whereas holders of future claims sought to ensure “an ample, inflation-protected fund for the future.” Id. at 626, 117 S.Ct. 2231.
The two subgroups in Amehem had competing interests in the distribution of a settlement whose terms reflected “essential allocation decisions designed to confine compensation and to limit defendants’ liability.” Id. at 627, 117 S.Ct. 2231. Some of those allocation decisions — for example, to cap the annual number of opt-outs, and not to adjust for inflation — disadvantaged exposure-only plaintiffs. Although the named parties all “alleged a range of complaints,” none exclusively advanced the particular interests of either subgroup; “each served generally as representative for the whole, not for a separate constituency.” Id. That flaw, in light of the conflict, was fatal to class certification. Even if the class representatives “thought that the Settlement serves the aggregate interests of the entire elass[,] ... the adversity among subgroups requires that the members of each subgroup cannot be bound to a settlement except by consents given by those who understand that their role is to represent solely the members of their respective subgroups.” Id. (quoting In re Joint E. & S. Dist. Asbestos Litig., 982 F.2d 721, 742-43 (2d Cir.1992), modified on reh’g, 993 F.2d 7 (2d Cir.1993)). In the absence of any “structural assurance of fair and adequate representation for the diverse groups and individuals affected,” the class could not satisfy Rule 23(a)(4)’s standard for fair and adequate representation. Id.
Two years later, in Ortiz v. Fibreboard Corp., 527 U.S. 815, 119 S.Ct. 2295, 144 L.Ed.2d 715 (1999), the Supreme Court rejected a proposed settlement class that was divided along two fault lines: first, as in Amehem, “between holders of present and future claims,” and second, between *251holders of “more valuable” and less valuable claims. Id. at 856-57, 119 S.Ct. 2295. As in Amchem, those divisions were not recognized by the formation of subclasses. Ortiz addressed the propriety of manufacturer Fibreboard Corporation’s global settlement of asbestos claims against it, a deal that included indemnification by two insurance companies. Claims based on asbestos exposure that occurred when Fibreboard was insured had a “much higher” settlement value than those for exposure after its insurance had expired, because only the former group could recover from the insurer. Id. at 823 n. 2, 119 S.Ct. 2295. That conflict fell “well within the requirement of structural protection recognized in Amchem,” the Supreme Court held, and should have been redressed by way of “reclassification with separate counsel.” Id. at 857, 119 S.Ct. 2295. That the settlement failed to differentiate the claims only confirmed the existence of a conflict: “[t]he very decision to treat them all the same is itself an allocation decision with results almost certainly different from the results that those with ... claims of indemnified liability would have chosen.” Id.
B.
The ingredients of conflict identified in Amchem and Ortiz are present here. The Settlement before us “confine[s] compensation and ... limit[s] defendants’ liability” by setting an $18 million recovery and cost ceiling, and distributes that recovery by making “essential allocation decisions” among categories of claims. See Amchem, 521 U.S. at 627, 117 S.Ct. 2231. Although named plaintiffs collectively hold all three categories of claim, “each served generally as representative for the whole, not for a separate constituency.” Id. In addition, individual Category A and B claims are “more valuable” than Category C claims,6 producing “disparate interests” within the class. Ortiz, 527 U.S. at 857, 119 S.Ct. 2295.
There are, however, clear differences between the case before us and Amchem and Ortiz. The conflict in Amchem could hardly have been more stark: class members fell into one of two mutually exclusive camps, those injured by asbestos and those with only potential future injuries. Here, by contrast, class members can and do hold claims in all three categories. Although the record does not establish the precise distribution of claims among named plaintiffs, that they hold a combination of registered and unregistered claims is undisputed. The conflict alleged by objectors is therefore between class members who hold Category C claims alone, and those who hold Category A and B claims in addition to Category C claims. Such overlap with respect to some claimants suggests, at least superficially, the absence of a fundamental conflict.
Despite the intuitive appeal of that conclusion, we cannot endorse it. Owning Category C claims in addition to other claims does not make named plaintiffs adequate representatives for those who hold only Category C claims. Although all affected members of the plaintiff class are interested in maximizing their individual compensation, severally they accomplish that goal in different ways. To authors who own works in all three categories, how their compensation is allotted among their claims is irrelevant; what matters is the bottom line. Class members who hold only Category C claims, on the other hand, *252are interested exclusively in maximizing the compensation for that one category of claim. Whereas the former group could choose to sacrifice their Category C claims in exchange for more favorable compensation on their Category A and B claims, no such option is available to the latter.
The selling out of one category of claim for another is not improbable here. Because the Settlement capped recovery and administrative costs at $18 million, named plaintiffs owning claims in all three categories cannot have had an interest in maximizing compensation for every category. Any improvement in the compensation of, for example, Category C claims would result in a commensurate decrease in the recovery available for Category A and B claims. Further, given that Categories A and B amount to approximately 1% of the total number of claims, named plaintiffs would receive a greater share of a given amount of compensation allocated to Categories A and B, compared to what they would receive if that compensation were spread over the far greater quantity of Category C claims. Named plaintiffs’ natural inclination would therefore be to favor their more lucrative Category A and B claims. That named plaintiffs hold claims in all categories does not, as the dissent asserts, eliminate the risk of fundamental conflict among subgroups.
Even if some named plaintiffs have only Category C claims, that is not enough to protect the Category C-only plaintiffs, because each named plaintiff represented the entire class. See Amchem, 521 U.S. at 627, 117 S.Ct. 2231. Without subclasses, named plaintiffs with only Category C claims were obligated to advance the collective interests of the class, rather than those of the subset of class members whose claims mirrored their own. Only the creation of subclasses, and the advocacy of an attorney representing each subclass, can ensure that the interests of that particular subgroup are in fact adequately represented. “[W]here differences among members of a class are such that subclasses must be established, we know of no authority that permits a court to approve a settlement ... on the basis of consents by members of a unitary class, some of whom happen to be members of ... distinct subgroups,” without creating subclasses. In re Joint E. & S. Dist. Asbestos Litig., 982 F.2d 721, 743 (2d Cir.1992), modified on reh’g, 993 F.2d 7 (2d Cir.1993).
To be sure, the negotiation of this Settlement featured protections that were lacking in Amchem. The Settlement was the product of an intense, protracted, adversarial mediation, involving multiple parties and complex issues. The mediators were highly respected and capable, and their participation provided some assurance that “the proceedings were free of collusion and undue pressure.” D Amato v. Deutsche Bank, 236 F.3d 78, 85 (2d Cir.2001). Furthermore, associational plaintiffs advanced the interests of all authors, the largest contingent of which we can reasonably assume — given that 99% of the total claims fall into Category C — are Category C-only plaintiffs. While we recognize that these features offered some “structural assurance of fair and adequate representation,” Amchem, 521 U.S. at 627, 117 S.Ct. 2231, we cannot conclude that they did enough to satisfy Rule 23(a)(4).
The Supreme Court’s decision in Am-chem was motivated in part by its conclusion that the settlement’s terms disfavored the exposure-only plaintiffs. Amchem therefore allows courts, in assessing the adequacy of representation, to examine a settlement’s substance for evidence of prejudice to the interests of a subset of plaintiffs. Objectors, pointing to Category C’s inferior recovery, urge that we do so here. Category C works receive signifi*253eantly less than those in Category B. For example, an article sold for $200 and registered by December 31, 2002 — but too late to receive statutory damages — falls into Category B and secures $150 under the Settlement; an unregistered but otherwise identical article warrants only $20 in Category C. The compensation structure for Category C is also, to use objectors’ term, “regressive” in that recovery as a percentage of a work’s original sale price decreases as the sale price increases; Category B compensation, by contrast, is a flat percentage of the sale price.
That Category C claims recover less than Category A and B claims tells us little about adequacy of representation, however, because the Category C claims individually are indisputably worth less than Category B claims. Given that registration of a copyright is a prerequisite to suit, unregistered Category C claims would face a substantial litigation risk if the case went forward. Indeed, had the Settlement failed to account for this weakness, the “very decision to treat [claims] all the same [would] itself [have been] an allocation decision” unfair to the interests of those who had authored registered works. See Ortiz, 527 U.S. at 857, 119 S.Ct. 2295. It was not only appropriate but also necessary for Category C claims to recover less than Category A and B claims. We therefore disagree with objectors to the extent that they cite Category C’s inferior recovery as determinative evidence of inadequate representation.
The problem, of course, is that we have no basis for assessing whether the discount applied to Category C’s recovery appropriately reflects that weakness. We know that Category C claims are worth less than the registered claims, but not by how much. Nor can we know this, in the absence of independent representation. The Supreme Court counseled in Ortiz that subclasses may be necessary when categories of claims have different settlement values. The rationale is simple: how can the value of any subgroup of claims be properly assessed without independent counsel pressing its most compelling case? It is for this reason that the participation of impartial mediators and institutional plaintiffs does not compensate for the absence of independent representation. Although the mediators safeguarded the negotiation process, and the institutional plaintiffs watched out for the interests of the class as a whole, no one advanced the strongest arguments in favor of Category C’s recovery. Even in the absence of any evidence that the Settlement disfavors Category C-only plaintiffs, this structural flaw would raise serious questions as to the adequacy of representation here.
In addition to the structural flaw discussed above, the Settlement itself contains terms that illustrate a lack of adequate representation of Category C-only plaintiffs. The “C reduction” places the risk that total claims and fees exceed the $18 million cap exclusively on Category C. Although we disagree with objectors as to the import of Category C’s inferior compensation, we regard the “C reduction” in a different light. The “C reduction” cannot be justified as a reflection of Category C’s lower value, because the Settlement’s recovery formulae already account for that difference. The “C reduction” is not designed to reflect the claims’ value at all, but rather is a safety valve meant to preserve the integrity of the Settlement in the event the cap is exceeded.
The settling parties argue that the “C reduction,” as a contingent provision they reasonably believed was unlikely to be triggered, cannot reflect on the adequacy of representation. We disagree. Those negotiating the Settlement identified a risk and placed that risk on a single category of *254claims.7 If triggered, the “C reduction” would deplete the recovery of Category C-only plaintiffs in their entirety before the Category A or B recovery would be affected. We can discern no reason, and authors and publishers offer none, why this burden should have been placed exclusively on Category C, rather than shared equitably among all three categories of claim. That only one category was targeted for this penalty without credible justification strongly suggests a lack of adequate representation for those class members who hold only claims in this category.
Even if we were to conclude that, as a matter of deferential review, the Settlement fairly compensates Category C claims, we cannot rely on that fact to affirm class certification, because doing so would conflate Rule 23(a)(4)’s adequacy of representation analysis with Rule 23(e)(2)’s fairness, adequacy, and reasonableness analysis. “Rule 23 requires protections under subdivisions (a) and (b) against inequity and potential inequity at the precertification stage, quite independently of the required determination at postcertification fairness review under subdivision (e) that any settlement is fair in an overriding sense.” Ortiz, 527 U.S. at 858, 119 S.Ct. 2295; accord Amchem, 521 U.S. at 621, 117 S.Ct. 2231. The possible fairness of a settlement cannot eclipse the Rule 23(a) and (b) precertifieation requirements. Ortiz, 527 U.S. at 858-59, 119 S.Ct. 2295. Thus, the adequacy of representation cannot be determined solely by finding that the settlement meets the aggregate interests of the class or “fairly” compensates the different types of claims at issue. See In re Joint E. & S. Dist. Asbestos Litig., 982 F.2d at 743. In the Rule 23(a)(4) context, we must ask independently whether the interests of all class members were adequately represented.
We find that they were not. We agree with objectors that the interests of class members who hold only Category C claims fundamentally conflict with those of class members who hold Category A and B claims. Although all class members share an interest in maximizing the collective recovery, their interests diverge as to the distribution of that recovery because each category of claim is of different strength and therefore commands a different settlement value. Named plaintiffs who hold other combinations of claims had no incentive to maximize the recovery for Category C-only plaintiffs, whose claims were lowest in settlement value but eclipsed all others in quantity. The interests of Category C-only plaintiffs could be protected only by the formation of a subclass and the advocacy of independent counsel. We therefore hold that the district court abused its discretion in certifying the class based on its *255finding that class representation was adequate.8
C.
The decision to require subclassing here is consistent with our precedent. In Central States Southeast & Southwest Areas Health & Welfare Fund v. Merck-Medco Managed Care, L.L.C., 504 F.3d 229, 246 (2d Cir.2007), a plaintiff class of trustees and beneficiaries of employee welfare benefit plans sued their pharmaceutical benefits manager, Medco Health Solutions, Inc. (“Medco”), alleging that it breached its fiduciary duties under the Employee Retirement Income Security Act of 1974 (“ERISA”) by favoring the products of its parent company, Merck & Co. The district court approved a settlement agreement and class certification over the self-funded plans’ objection that a conflict of interest necessitated the certification of a subclass. The objectors argued that the self-funded plans needed independent representation because they “were more damaged by Medco’s conduct by virtue of paying Med-co the entire cost of their beneficiaries’ drugs,” as compared to insured plans, which paid set premiums to Medco and were therefore more insulated from the effects of Medco’s conduct. Id. at 245. The district court rejected this argument, observing that the settlement properly accounted for this disparity by applying a 55% discount to the claims of the insured plans, a figure determined by counsel with the assistance of expert opinion and a special master. Id. at 237, 245.
Although all class members “advanced similar theories of liability against Medco predicated on the same or similar facts” and all wished to “obtain the highest possible recovery,” the Second Circuit sided with the objectors. Id. at 245-46. Without deciding “whether the self-funded Plans in fact suffered greater injury,” we thought it “proper to allow them to raise their claims as part of a separate subclass.” Id. at 246. Finding that “the antagonistic interests apparent in the class should be adequately and independently represented,” we remanded to the district court “for certification of a subclass encompassing the self-funded plans in order to better protect their claims in this litigation.” Id.
Central States is parallel to the instant case in several key respects. First, the settlement agreement established a fund ($42.5 million) that would “allocate[] an amount to the settling class members” based on “the nature of [each] Plan’s relationship with Medco.” Id. at 236. Second, the settlement recognized and accounted for a disparity in the strengths of two discrete categories of claims: the recovery for insured plans was discounted by 55% to reflect that they were more insulated from Medco’s improper conduct. Third, class counsel had the benefit of an impartial special master in determining that allo*256cation. There is also a key difference: Central States cited no direct evidence of inadequate representation in the settlement terms. Even in the absence of such evidence, we found that the district court’s certification of the class was an abuse of discretion because the self-funded plans required independent representation. The case for subclassing is, if anything, more compelling in this case. As in Central States, a capped settlement fund was allocated differently among categories of claims of different strength without separate counsel to protect each category’s interests.9 Unlike in Central States, the instant Settlement not only suffers from a clear structural defect, but also provides strong evidence — in the “C reduction” — of inadequate representation.10
D.
Having concluded that a fundamental conflict exists, we turn now to the question of subclassing. Objectors demand that the unregistered copyright holders be defined as a subclass to provide structural assurance of fair and adequate representation. Remedying this conflict may not be so simple, however. Will the subclass be limited to the Category C-only plaintiffs, or should it also include those class members who own registered (Categories A and B) in addition to unregistered (Category C) copyrights? However the subclass is defined, who will advance the interests of the remaining class members? Can Category C counsel sit across the negotiating table from counsel representing “everyone else,” or will everyone else’s interests be sufficiently divergent to require further sub-classing? These questions greet us as soon as we open the door to subclassing, and we must at least acknowledge them before we can enter.
We would ordinarily allow the district court to work out the details of subclassing and leave these questions to be resolved in that process. We recognize, however, that “at some point there must be an end to reclassification with separate counsel.” Ortiz, 527 U.S. at 857, 119 S.Ct. 2295. It would be imprudent to require subclassing if subclasses were administratively impracticable. We now, therefore, assess wheth*257er subclasses can be devised to remedy the conflict we have identified.
The simplest and most logical approach may be to create a subclass for every category of claim, with separate counsel representing the interests of Categories A, B, and C. The different claim categories are, after all, the fault lines along which the conflict runs. These categories, each of different strength, must compete with one another over the allocation of the capped Settlement fund. Designating each a subclass, and assigning counsel to represent their interests, would protect each category’s interests.
This case is more complicated than most, however. Plaintiffs cannot all be neatly segregated into one of three categories, because some class members hold claims in more than one category. Although many plaintiffs only authored Category C works, and some plaintiffs may assert claims only in Category A or B, the remaining class members have claims in two or three categories. Structuring the subclasses so that no class member falls into more than one subclass could require as many as seven subclasses: plaintiffs holding (1) only A claims, (2) only B claims, or (3) only C claims, or a combination of (4) A and B, (5) A and C, (6) B and C, or (7) A, B, and C claims. That is surely beyond the point at which “reclassification with separate counsel” must end.
Creating only three subclasses — one for each category of claim — would, by contrast, be efficient and straightforward. This approach satisfies objectors’ concerns, as the Category C — only plaintiffs will all fall within the Category C subclass and have their own counsel. Separate counsel will also advance the interests of Categories A and B, respectively, giving each category a voice advocating for a share of the Settlement commensurate with their value. This structural protection will provide a substantial guarantee that the values assigned to each category of claim resulted from merits-based negotiation, greatly reducing the risk that a deficiency in representation for one or more subgroups will affect the outcome.
Although some class members would fall into more than one subclass, we can see no reason why that would be fatal to such a structure. It is certainly not precluded by the language of Rule 23(c)(5), which allows a class to “be divided into subclasses that are each treated as a class under this rule.” Fed.R.Civ.P. 23(c)(5). And it makes sense from a practical perspective. All class members are interested in receiving the maximum possible recovery for their claims. Having a separate subclass representative advocate exclusively for each of those claims is the most effective means of achieving that result. A plaintiff who holds claims in Categories B and C would, for example, be represented by different subclass representatives and counsel with respect to each category. Each subclass representative would, in turn, represent plaintiffs’ interests with respect to only that category of claim.
We intend by no means to bind the district court or the parties to the subclass structure we have outlined. We address this issue only to ensure that we are not asking the district court to carry out instructions that are impracticable to implement. Satisfied that the conflict here can be remedied within the practical limits of “reclassification with separate counsel,” Ortiz, 527 U.S. at 857, 119 S.Ct. 2295, we remand to the district court for subclassing while recognizing that another solution may be more appropriate than the one we have proffered.
CONCLUSION
Because the named plaintiffs are inadequate representatives for class members who hold only Category C claims, we VA*258CATE the district court’s order and judgment and REMAND for further proceedings consistent with this opinion.
. In addition to the named defendants, non-party newspaper and magazine publishers like the Tribune Company and Time Inc. participated in the mediation, because they had *246provided content to — and promised to indemnify — electronic publisher defendants. Thirty-six such non-party publishers ultimately signed onto the Settlement.
. Fees at this level were in place from 1999 through 2006. See 64 Fed.Reg. 29,518, 29, 520 (June 1, 1999) (setting fees); 71 Fed.Reg. 31,089, 31,091 (June 1, 2006) (raising fees).
. The class is defined as "All persons who, individually or jointly, own a copyright under the United States copyright laws in an English language literary work that has been reproduced, displayed, adapted, licensed, sold and/or distributed in any electronic or digital format, without the person’s express authorization by a member of the Defense Group or any member’s subsidiaries, affiliates, or licensees (a) at any time on or after August 15, 1997 (regardless of when the work first appeared in an electronic database) or (b) that remained in circulation after August 15, 1997, even if licensed prior thereto, including English language works qualifying for U.S. copyright protection under an international treaty (hereinafter 'Subject Work').”
. In their post-argument letter briefs, the parties raise new arguments regarding a 25-year-old Supreme Court case, Local No. 93, Int'l Ass’n of Firefighters v. City of Cleveland., 478 U.S. 501, 106 S.Ct. 3063, 92 L.Ed.2d 405 (1986). Because these arguments were not raised in a timely fashion, we deem them waived. In re Nortel Networks Corp. Sec. Litig., 539 F.3d 129, 132-34 (2d Cir.2008).
. We find Davis v. Blige, 505 F.3d 90 (2d Cir.2007), cited by objectors, inapposite. That case presents an altogether different issue: "whether one joint owner of a copyright can retroactively transfer his ownership by a written instrument, and thereby cut off the accrued rights of the other owner to sue for infringement,” id. at 97. As this case does not involve co-owners who are not parties to the settlement agreement, Davis does not address the issue before the court.
. Category A claims are eligible for statutory damages and therefore the most valuable. Category B claims, although registered too late for statutory damages, still qualify for actual damages and attorney's fees. Category C claims, which were unregistered as of December 31, 2002, are ineligible for actual damages and attorney's fees until registered.
. This risk was not fanciful. In their June 23, 2010 letter briefs, publishers and authors stated that — now that all of the claims have been submitted to the claims administrator — the total face value of claims, plus fees and costs, is known to be $2.9 million below the $18 million ceiling that triggers the reduction. However, in a January 11, 2011 letter, publishers and authors informed us that they had erroneously understated the total claims value by more than $2.6 million. The claim value is now estimated at $11.56 million, which— when added to fees and costs — comes within $300,000 of the "C reduction” threshold. This casts serious doubt on the assertion that the "C reduction” was unlikely to be triggered. However, because this information was not before the district court, we will not consider it in our analysis. Even if we were to consider it, we would find it immaterial because it was not available at the time of negotiation, which is the relevant time frame when determining whether the actions of the parties indicate a conflict of interests. Samuel Issacharoff & Richard A. Nagareda, Class Settlements Under Attack, 156 U. Pa. L.Rev. 1649, 1689-90 (2008).
. Objectors additionally argue that a fundamental conflict materialized in the Settlement's treatment of foreign works and scientific and research-based medical works. We decline to address objectors' arguments regarding the treatment of foreign works because they were not raised before the district court and are therefore waived. See In re Nortel Networlcs Corp. Sec. Litig., 539 F.3d 129, 133-34 (2d Cir.2008). With regard to the treatment of scientific and research-based medical works, objectors argue that the Settlement permits future uses of these works without providing any compensation for the past uses of the works to their authors. The record is plain, however, that the scientific and research-based medical claims were not released by the Settlement. The Settlement instead excluded these works altogether. Accordingly, authors of these works remain free to pursue independent actions against any or all publishers in this case for alleged infringements.
. We observed that the conflict in Central States went beyond a "simple disagreement over potential differences in the computation of damages,” since the "relationship of the Plans to Medco ... [went] to the very heart of the litigation.” 504 F.3d at 246. The dissent, highlighting this language, argues that the conflict before us cannot be “fundamental” because the claim categories differ only in their relative strength, and all class members otherwise "had the same basic relationship with the defendants.” Dissent at 260. That argument fails to account for Ortiz. The difference underlying the conflict in Ortiz was whether or not Fibreboard had insurance at the time of plaintiffs' asbestos exposure, which — as in the present case — affected the claims' strength and settlement value but not the parties' "basic relationship.”
. The Third Circuit approved a class action settlement that allocated the recovery among three distinct classes of plaintiffs without creating subclasses. In re Insurance Brokerage Antitrust Litig., 579 F.3d 241 (3d Cir.2009). The court affirmed certification of the single class despite unequal allocations between the groups because the settlement agreement was "simply a reflection of the extent of the injury that certain class members incurred and does not clearly suggest that tire class members had antagonistic interests.” Id. at 272. The court recognized that "some potential benefits may have been realized from utilizing subclasses,” but ruled that the district court’s failure to take that step was not an abuse of discretion. Id. at 273. We, to the contrary, hesitate to conclude here that the Settlement's allocation is "simply a reflection of” the claims' differing settlement values in the absence of separate counsel advancing each category's interests. Furthermore, the "C reduction” offers specific evidence of inadequate representation, which was not present in Insurance Brokerage.