concurring in part and dissenting in part.
I share many of my colleagues’ concerns about the way in which the class was defined in this case, and I further agree that because DPI lacked the authority to unilaterally impose a corrective plan on a local school district, it was error to approve DPI’s settlement with the plaintiffs. However, I am not convinced that no class was feasible in this case, nor do I believe that the inability to identify class members until the remedial phase of the litigation precludes certification of the class. For these reasons, I write separately.
On its face, the class certified in this case bears the hallmarks of the sort of open-ended classes of which we disapproved in Adashunas v. Negley, 626 F.2d 600, 603-04 (7th Cir.1980), and Rahman v. Chertoff, 530 F.3d 622, 625-26 (7th Cir. 2008). Although the class that the district court approved (all school age children with disabilities who reside within the MPS district, who are or may be eligible for special education services, and who are, have been, or will be denied or delayed entry into or participation in the IEP process) is considerably more narrow than the class originally proposed by the plaintiffs (all school age children with disabilities who reside within the MPS district and who are or may be eligible for special education services), it was never formally delimited by the particular types of violations that the plaintiffs were alleging. The consequence of such an open-ended description of the class, as we noted in Rah-man, is that “every time plaintiffs file a brief or motion, membership in the class[ ] may change.” Id. at 625. Thus, at any given point during the litigation, it may not be possible to know from the class definition itself what the true parameters of the class are and what types of individuals might be in the class. Id. at 625-26. In this case, the class has been defined to include all students eligible for special services who have been denied or delayed entry into the IEP system. That definition literally may include students who were delayed or denied entry due to cir*504cumstances having little or nothing to do with the defendants’ actions, by essentially random circumstances that are unique to one student or limited to a small number of students, or by circumstances that the class representatives have not identified as a basis for relief to the class.
I am mindful that the parties and the court, from the time that the class was certified forward, appear to have developed an understanding that the class would include the four sets of individuals that the court later expressly identified in its liability decision: (1) students with suspected disabilities who were never referred for an evaluation; (2) students who were referred for evaluation, but whose evaluations were not completed within the requisite 90-day limit; (3) students who were suspended in such a way as to frustrate their evaluation; and (4) students whose IEP meetings took place in the absence of their parents. See ante at 488; R. 195 at 3-6; R. 389 at 61. Moreover, the court’s liability decision was limited to the time period commencing in September 2000 and ending in June 2005. R. 389 at 3. In these respects, the litigation ultimately focused on a class of students that was much more limited than the certified class was on its face, and one that in the end was much more definite than the sweeping class we criticized in Adashunas, for example.
But even as confined to these four groups of students and this roughly five-year period, the class includes unknown numbers of variations. To cite one example: When, in her 2004 report, Dr. Rogers-Adkinson addressed MPS’s failure to meet statutory deadlines during the assessment process, she identified at least four different patterns of delays: (1) delays occasioned by the initial mis-eategorization of a student’s suspected disability; (2) delays caused by extensions of time routinely requested for disability assessments solicited during the spring semester; (3) delays triggered by the need for additional medical records; and (4) delays for reasons not revealed in the files she reviewed. Plaintiffs’ Ex. 8 at 6, 8-9. Each of these four scenarios might be a manifestation of MPS’s failure to timely identify students with disabilities and to convene IEP meetings to address those disabilities, but it is difficult to discern a common, wrongful policy or practice that might account for them all. Instead, this looks more like an effort to sweep many individual plaintiffs and sets of facts into one class on the premise that all reflect illegal conduct by the defendant in practice and culture if not in policy. As the majority points out, ante at 498, that is precisely the sort of class that the Supreme Court recently rejected in Wal-Mart Stores, Inc. v. Dukes, — U.S. -, 131 S.Ct. 2541, 180 L.Ed.2d 374 (2011). Like my colleagues, I am not convinced that the certified class in this case is united by common legal or factual issues.
Nonetheless, I do not rule out the possibility that a class properly could be certified in this case or one like it, and it is to this limited extent that I dissent. My colleagues have concerns both with the notion that a class action may be premised on “systemic” violations of a school district’s child-find obligations and with a class composed of individuals whose identity cannot be known until the remedial phase of the litigation. However, I believe that notwithstanding the inherently child specific nature of child-find inquiries, a class action based on a truly systemic child-find failure may be viable. And the fact that it may not be possible to identify individual class members until the remedial phase of the litigation, when prospective members of the class are invited to come forward and establish that they were among those injured by this systemic fail*505ure, should not preclude a class action, which may be the only realistic avenue of relief for those injured by systemic violations of their rights.
Systemic violations of the IDEA are cognizable. See Doe ex rel. Brockhuis v. Ariz. Dep’t ofEduc., Ill F.3d 678, 681 (9th Cir.1997) (coll.cases). The problem here is that the plaintiffs’ claims appear to be based on multiple, disparate failures to comply with the school district’s statutory child-find obligations rather than a truly systemic policy or practice which affects them all. But my colleagues go so far as to state that “[tjhere is no such thing as a ‘systemic’ failure to find and refer individual disabled children for IEP evaluation— except perhaps if there was ‘significant proof that MPS operated under child-find policies that violated the IDEA.” Ante at 498 (emphasis in original) (citing WalMart, 131 S.Ct. at 2553). Certainly I agree that an illegal policy would support a claim for a systemic violation of the IDEA’S child-find mandate. But I also think that widespread practices might also support such a claim. Cf. King ex rel. King v. E. St. Louis School Disk 189, 496 F.3d 812, 817 (7th Cir.2007) (citing Calhoun v. Ramsey, 408 F.3d 375, 379 (7th Cir.2005)) (noting that municipal liability under 42 U.S.C. § 1983 may be based on customs and practices as well as formal, written policies). Suppose, for example, that MPS ignored and did not refer for evaluation students with a particular type of potential disability — e.g., dyslexia — and as a result, large numbers of students with that disability went unidentified and were deprived of the educational services to which they were entitled under the IDEA. To my mind, this would be a genuinely systemic violation of the district’s child-find obligations that would be amenable to recognition and remediation in a class action: Notwithstanding the fact that each student is unique and is entitled to his own, individualized IEP, all members of the class would suffer a common injury resulting from the school district’s failure to have appropriate systems in place to identify students with that particular disability. Insisting instead that each student exhaust his administrative remedies and/or sue independently would surely mean that many such students would remain unidentified and denied their right to free appropriate public education (if only because they would be unaware of their rights) and likely would mean that the systemic violation underlying their claims to relief would persist, as any individual proceedings would result in individual rather than structural relief.
A key reason why my colleagues are reluctant to acknowledge the possibility of a systemic violation of a school district’s child-find obligations is the inability to identify individual class members except by asking the parents of putative class members to come forward at the remedial phase of the litigation and to establish, in individual hearings, that their children were in fact the victims of the district’s child-find failures. But this method of identifying class members is hardly unprecedented. Our decision in McDonald v. United Air Lines, Inc., 745 F.2d 1081 (7th Cir.1984), reflects a similar approach. The class in McDonald comprised women who had lost their jobs as a result of United Air Line’s rule (dating back to the 1930s) prohibiting marriage for its female cabin attendants. Once this court declared the no-marriage rule to be a violation of Title VIPs ban on sex discrimination in Sprogis v. United Air Lines, Inc., 444 F.2d 1194 (7th Cir.1971), any woman who had lost her position as a result of that rule between the date Title VII took effect in 1965 and United’s abolition of the rule in 1968 became entitled to relief. See McDonald v. United Air Lines, Inc., 587 *506F.2d 357 (7th Cir.1978). But many if not most of the class members had simply resigned their positions in contemplation of the no-marriage rule rather than formally protesting the rule or waiting for the company to discharge them, so there was no record enabling easy identification of those women who left because of the no-marriage rule as opposed to another reason. Consequently, the onus was on former flight attendants to come forward at the final stage of the litigation and claim entitlement to relief, at which point individual, adversarial hearings were convened before special masters in order to establish whether each claimant in fact left her position because of the illegal rule. See McDonald, 745 F.2d at 1087-88.
As in McDonald, there will be times when the nature of the challenged conduct makes it difficult if not impossible to identity who is in the class and is entitled to relief absent some sort of opt-in procedure by putative class members coupled with an adjudicatory procedure to confirm that they in fact qualify as class members. In McDonald this was true because so many class members silently resigned their jobs in deference to their employer’s illegal rule; in a child-find case like this one it would be true because the school district neglected its obligation under the IDEA to identify students with potential disabilities. This is not a function of a poor or open-ended class definition as in Rahman, but rather due to the fact that the defendant’s conduct did not produce a tidy record of the individuals harmed by that conduct. Identifying class members in such circumstances necessarily will require a relatively cumbersome, after-the-fact inquiry of the sort employed in McDonald and as the district court envisioned here. The only alternative is to foreclose class-wide relief in such cases and leave each individual harmed by the defendant’s conduct to pursue relief on his or her own, if he or she is even aware that a wrong has been committed. Realistically, that will mean no relief at all for most individuals.
For all of these reasons, I concur in the court’s conclusion that the class certified in this case was improper, but I do not join its conclusion that no class would be viable in this case or other litigation alleging violation of a school district’s child-find obligations absent an ability to identify an express policy that violates the IDEA.