POSNER, Circuit Judge.

We have consolidated for decision petitions, filed by two groups of defendants in a consumer-finance class action litigation, for leave to appeal an order by'the district court certifying a plaintiff class. Fed.R.Civ.P. 23(f) authorizes us to entertain such interlocutory appeals. The rule does not state criteria for the exercise of this discretionary authority. But the case law teaches that the more novel the issue presented by the appeal and so the less likely that the district court’s resolution of it will stand, the more important the resolution of the issue is either to the particular litigation or to the general development of class action law, and the more likely the prompt resolution of the issue is to expedite the litigation and prevent a coercive settlement, the stronger thé case for allowing the appeal. E.g., In re Bridgestone/Firestone, Inc., 288 F.3d 1012, 1016 (7th Cir.2002); Blair v. Equifax Check Services, Inc., 181 F.3d 832, 834-35 (7th Cir.1999); Newton v. Merrill Lynch, Pierce, Fenner & Smith, Inc., 259 F.3d 154, 164 (3d Cir.2001); Prado-Steiman v. Bush, 221 F.3d 1266, 1278 (11th Cir.2000); Waste Management Holdings, Inc. v. Mowbray, 208 F.3d 288, 294 (1st Cir.2000). The issues that the petitions ask us to consider, in the setting of a class of millions, concern, first, the procedures and criteria for converting a settlement class into a litigation class when having initially been approved the settlement is later disapproved, and, second, the bearing of the doctrine of judicial estoppel on class action litigation. These are novel issues whose prompt resolution is important to the development of the law of class actions as well as to the resolution of the present case. The petitions to appeal are therefore granted. The merits of the appeals have been fully briefed and we can therefore proceed to decide them without requiring further briefing.

The litigation arose out of refund anticipation loans made jointly by the defendants, who for simplicity we’ll refer to as “the bank” and “the tax preparer.” When the tax preparer files a refund claim with the Internal Revenue Service on behalf of *659 one of its customers, the customer can expect to receive the refund within a few weeks unless the IRS decides to investigate the return. Even a few weeks is too long for the most necessitous taxpayers, and so the bank will lend the customer the amount of the refund for the period between the filing of the claim and the receipt of the refund. The annual interest rate on such a “refund anticipation loan” (RAL) will often exceed 100 percent. Although the bank is the lender, the tax preparer arranges the loan. It is contended that the customer is told neither that the bank pays the tax preparer a fee for having generated the loan nor that the tax preparer receives an ownership interest in the loan.

Beginning in 1990 a number of class-action suits were brought against the defendants on behalf of a total of 17 million refund-anticipation borrowers, charging violations of various state and federal laws, including RICO. The basic claim is that the defendants lead the borrowers to believe that the tax preparer is their fiduciary, much as if they had hired a lawyer or an accountant to prepare their income tax returns, as affluent people do, whereas, unbeknownst to them, the tax preparer is engaged in self-dealing. This conduct is alleged to constitute a scheme to defraud in violation of the federal mail- and wire-fraud statutes. Violations of those statutes are “predicate offenses” that can form the basis of a RICO charge.

In 1999 the named plaintiff in one of the suits entered into a settlement agreement with the bank and the tax preparer. This was to be a “global” settlement: the members of all the classes would divide up a $25 million fund put up by the defendants in exchange for the release of all claims arising out of the RALs. The district judge approved the settlement and enjoined (with one exception) the other RAL class actions, Zawikowski v. Beneficial National Bank, No. 98 C 2178, 2000 WL 1051879 (N.D.Ill. July 28, 2000), but we reversed, Reynolds v. Beneficial National Bank, 288 F.3d 277 (7th Cir.2002), on the ground that the district judge had failed to scrutinize the fairness of the settlement adequately. We were concerned that the settlement might have been the product of collusion between the defendants, eager to minimize their liability, and the class lawyers, eager to maximize their fees.

The district judge to whom the case was reassigned on remand concluded that the settlement had indeed been unfair and disapproved it. 260 F.Supp.2d 680 (N.D.Ill. 2003). There was no appeal. The proceedings continued in the district court, with both the named plaintiff and the class counsel replaced. Although no class had formally been certified in the earlier proceedings, rather than require the new plaintiff to move for certification the judge asked the defendants for their objections to certification, and they responded. She agreed with some of the objections, rejected others, and, in effect, certified the same class that had been contemplated by the rejected settlement, which is to say all RAL borrowers (with a few exceptions) whose claims weren’t barred by the statute of limitations. But she limited the certification to prosecution of just the RICO claim, plus one breach of contract claim involving the law of only one state.

The defendants object mainly to the procedure the judge employed and to the brevity with which she pronounced the class manageable despite its vast size. In the previous round of this protracted litigation the defendants had urged the district court to accept the giant class as appropriate for a global settlement, had prevailed in their urging, and so are now precluded by the doctrine of judicial estoppel, see, e.g., New Hampshire v. Maine, *660 532 U.S. 742, 121 S.Ct. 1808, 149 L.Ed.2d 968 (2001), from challenging its adequacy, at least as a settlement class (the significance of this qualification will appear in due course). It is true that we reversed the district court’s approval of the settlement, but a reversal need not affect the application of judicial estoppel. In re Cas sidy, 892 F.2d 637, 641 (7th Cir.1990); Hall v. GE Plastic Pacific PTE Ltd., 327 F.3d 391, 398-99 (5th Cir.2003); U.S. Philips Corp. v. Sears Roebuck & Co., 55 F.3d 592, 597 (Fed.Cir.1995); cf. 18B Charles Alan Wright, Arthur R. Miller & Edward H. Cooper, Federal Practice and Procedure § 4477 (2d ed.2002). The reason lies in the purpose of the doctrine. The canonical statement of that purpose is that it is “to protect the integrity of the judicial process.” E.g., New Hampshire v. Maine, supra, 532 U.S. at 749-50, 121 S.Ct. 1808; United States v. Christian, 342 F.3d 744, 747 (7th Cir.2003). But we have been a little more precise. We have said that “a party who prevails on one ground in a lawsuit cannot turn around and in another lawsuit repudiate the ground. If repudiation were permitted, the incentive to commit perjury and engage in other litigation fraud would be greater. A party envisaging a succession of suits in which a change in position would be advantageous would have an incentive to falsify the evidence in one of the cases, since it would be difficult otherwise to maintain inconsistent positions.” McNamara v. City of Chicago, 138 F.3d 1219, 1225 (7th Cir.1998) (citations omitted). In other words, “the purpose of the doctrine ... is to reduce fraud in the legal process by forcing a modicum of consistency on a repeating litigant.” Ladd v. ITT Corp., 148 F.3d 753, 756 (7th Cir. 1998); see also Bethesda Lutheran Homes & Services, Inc. v. Born, 238 F.3d 853, 858 (7th Cir.2001).

The antifraud policy that animates the doctrine is fully engaged when a party obtains a judgment on a ground that it later repudiates, even if his opponent, the loser in that fúst case, is able, obviously at some expense to itself but also placing a demand on judicial resources, to get the judgment reversed. Anyway the defendants benefited from the temporary approval of the settlement, which they used to enjoin other RAL litigation against them; and having reaped a benefit from their pertinacious defense of the class treatment of the case for purposes of settlement they cannot now be permitted to seek a further benefit from reversing their position.

It is true that we went on to say in McNamara that “the doctrine of judicial estoppel requires ... that the party sought to be estopped have obtained a favorable judgment or settlement on the basis of a legal or factual contention that he wants to repudiate in the current litigation. Otherwise it would be inconsistent with the rule that permits inconsistent pleadings.” 138 F.3d at 1225. But the defendants did obtain a judgment. The fact that it was reversed on appeal has nothing to do with a party’s right to explore inconsistent alternative positions in the early stages of a lawsuit.

The defendants are correct, however, that a class might be suitable for settlement but not for litigation. The class might be unmanageable if the case were actually tried yet manageable as a settlement class because the settlement might eliminate all the thorny issues that the court would have to resolve if the parties fought out the ease. Amchem Products, Inc. v. Windsor, 521 U.S. 591, 620, 117 S.Ct. 2231, 138 L.Ed.2d 689 (1997). But although the district judge might have said more about manageability, the defendants have said nothing against it except that there are millions of class members. That *661 is no argument at all. The more claimants there are, the more likely a class action is to yield substantial economies in litigation. It would hardly be an improvement to have in lieu of this single class action 17 million suits each seeking damages of $15 to $30. The rejected settlement capped damages at these amounts for single and multiple RALs respectively, and while the amounts may be too low they are indicative of the modest stakes of the individual class members. The realistic alternative to a class action is not 17 million individual suits, but zero individual suits, as only a lunatic or a fanatic sues for $30. But a class action has to be unwieldy indeed before it can be pronounced an inferior alternative — no matter how massive the fraud or other wrongdoing that will go unpunished if class treatment is denied— to no litigation at all.

Often, and possibly in this case as well, there is a big difference from the standpoint of manageability between the liability and remedy phases of a class action. The number of class members need have no bearing on the burdensomeness of litigating a violation of RICO. Whether particular members of the class were defrauded and if so what their damages were are another matter, and it may be that if and when the defendants are determined to have violated the law separate proceedings of some character will be required to determine the entitlements of the individual class members to relief. Fed.R.Civ.P. 23(c)(4)(A); Allen v. International Truck & Engine Corp., 358 F.3d 469 (7th Cir. 2004); Bell Atlantic Corp. v. AT&T Corp., 339 F.3d 294, 307 n. 16 (5th Cir.2003); In re Visa Check/MasterMoney Antitrust Litigation, 280 F.3d 124, 141 (2d Cir.2001); Robinson v. Metro-North Commuter R.R., 267 F.3d 147, 168-69 (2d Cir.2001). That prospect need not defeat class treatment of the question whether the defendants violated RICO. Once that question is answered, if it is answered in favor of the class, a global settlement along the lines originally negotiated (though presumably with different dollar figures) will be a natural and appropriate sequel. And if there is no settlement, that won’t be the end of the world. Rule 23 allows district courts to devise imaginative solutions to problems created by the presence in a class action litigation of individual damages issues. Those solutions include “(1) bifurcating liability and damage trials with the same or different juries; (2) appointing a magistrate judge or special master to preside over individual damages proceedings; (3) decertifying the class after the liability trial and providing notice to class members concerning how they may proceed to prove damages; (4) creating subclasses; or (5) altering or amending the class.” In re Visa Check/MasterMoney Antitrust Litigation, supra, 280 F.3d at 141.

We did note in our previous decision an unremarked conflict of interest between those class members who took out one or two refund anticipation loans and those who took out more than two and would thus, because of the $30 cap that we just mentioned, receive no compensation for the additional harm that they suffered. 288 F.3d at 282. But we went on to say that “in light of the modesty of the stakes even of class members who had multiple refund anticipation loans and the expense of subdividing the class (and how many subdivisions would be necessary to reflect the full range of damages?), we are not disposed to regard this particular defect in the settlement as fatal.”

The defendants argue that by requiring them to present their objections to class certification rather than requiring the plaintiff to move for certification, the district judge improperly altered the burden of proof on the question whether the *662 class should be certified. Not so. Although the plaintiff class had not formally been certified in the earlier proceedings, that was indeed a formal defect — a technical oversight that was surprising but from a practical standpoint inconsequential. The case had been settled as a class action, notices had been sent, objections had been considered and rejected — all on the assumption that a class had been certified. This was de facto certification, albeit of a settlement class only, and that was enough, we think, to empower Judge Bucklo, in the exercise of her discretion to manage litigation before her in an efficient and expeditious manner, to require the defendants to list their objections to the certification of a litigation class, especially since she was explicit that the burden of persuasion on the validity of the objections would remain on the plaintiffs. They carried the burden easily. Remember that the defendants themselves had argued that the class was appropriate for settlement purposes. That did not conclude the question whether it was appropriate for litigation if the settlement fell through, as we have explained and as the district judge recognized. But it was some indication that there were issues appropriate for determination on a class basis.

The defendants argue that the named plaintiff has not been shown to be an adequate representative of the class, but the district judge thought otherwise on sufficient grounds. The defendants also argue that class certification is barred by collateral estoppel, and this argument requires a fuller discussion.

In re Bridgestone/Firestone, Inc., Tires Products Liability Litigation, 333 F.3d 763 (7th Cir.2003), holds that a judicial finding that a class should not be certified is, at least in some circumstances, entitled to collateral estoppel effects; and in Buford v. H & R Block, Inc., 168 F.R.D. 340 (S.D.Ga.1996), aff'd, 117 F.3d 1433 (11th Cir.1997) (per curiam), a federal district judge refused to certify a nationwide class action charging, just as in this case, that our defendants’ practices with respect to RALs violated RICO. The judge recognized that the question whether RICO was violated was separate from the question whether the targets of the violation had been injured by the violation. 18 U.S.C. § 1964(c); Holmes v. Securities Investor Protection Corp., 503 U.S. 258, 267-68, 112 S.Ct. 1311, 117 L.Ed.2d 532 (1992). The predicate acts in the RICO claim are violations of the mail- and wire-fraud statutes, and these statutes are violated by a “scheme or artifice to defraud,” 18 U.S.C. §§ 1341, 1343, whether or not the scheme succeeds and therefore causes injury. E.g., United States v. Tadros, 310 F.3d 999, 1006 (7th Cir.2002); United States v. Coffman, 94 F.3d 330, 333-34 (7th Cir. 1996); United States v. Daniel, 329 F.3d 480, 486 (6th Cir.2003). The separation of liability and injury issues is illustrated by the suggestion in Moore v. PaineWebber, Inc., 306 F.3d 1247, 1255-56 (2d Cir.2002), that in a suit charging “uniform misrepresentations” the question whether they were indeed misrepresentations would be appropriate for class treatment, with the question of reliance, and damages suffered, by individual class members left for satellite proceedings. See also In re Prudential Ins. Co. of America Sales Practices Litigation, 148 F.3d 283 (3d Cir.1998).

The judge in Buford thought, however, that the issue of violation would be swamped by issues concerning whether the borrowers had been deceived — had relied on the fraud, and thus had been injured, whether directly, as in American Chiropractic Ass’n, Inc. v. Trigon Healthcare Inc., 367 F.3d 212, 233 (4th Cir.2004); Bank of China v. NBM LLC, 359 F.3d 171, 178 (2d Cir.2004); Byrne v. Nezhat, 261 *663 F.3d 1075, 1109-10 (11th Cir.2001), and Summit Properties Inc. v. Hoechst Celanese Corp., 214 F.3d 556 (5th Cir.2000), or perhaps, as in Systems Management, Inc. v. Loiselle, 303 F.3d 100, 104 (1st Cir. 2002), indirectly. Those issues would have to be resolved separately for each class member. In deciding that therefore the case was inappropriate for class treatment, the judge was applying the presumption against class action certification in RICO cases that has been articulated by the Fifth Circuit in Sandwich Chef of Texas, Inc. v. Reliance Nat’l Indemnity Ins. Co., 319 F.3d 205, 219 (5th Cir.2003). We are dubious about such a presumption. The question whether RICO was violated can be separated from the question whether particular intended victims were injured, and thus can — -or so a district court could determine without being thought to have abused its discretion — be resolved in a single proceeding with the issue of injury parceled out to satellite proceedings, as is frequently done in class action tort litigation, see, e.g., Exxon Co. v. Sofec, Inc., 54 F.3d 570, 575-76 (9th Cir.1995); In re Bendectin Litigation, 857 F.2d 290, 308-13 (6th Cir.1988), of which the RAL class litigation is a species.

The adequacy of the judge’s reasoning in Buford is not important, however, if, as the defendants contend, his ruling is entitled to collateral estoppel effect. But they overread Bridgestone/Firestone to hold that any ruling denying class certification is binding in future litigation. Our decision was more nuanced. See 333 F.3d at 767-69. For example, we pointed out that the binding effect of such a ruling would depend on whether the class members who would be affected by it had been adequately protected by the class representatives and class counsel in the proceeding in which the ruling was made. The judge in Buford discussed the issue of adequacy at some length, however, 168 F.R.D. at 352-55, and we are not disposed to reexamine his ruling. But it is too late for the defendants to plead collateral estoppel when, though knowing of the Buford decision, which was issued in 1996 and affirmed the following year, they insisted until last year, when the district court on remand from our decision threw out the settlement, that class treatment of the RICO claim was entirely appropriate. Until then they desperately wanted the RICO claim included in the class settlement so that they wouldn’t have to face it in any other RAL suits. They prevailed in the present litigation, until the settlement was finally rejected, by arguing that Buford was wrong. They are estopped to argue now that it was right. And anyway collateral estoppel is an affirmative defense that is forfeited if not raised in timely manner, and it was not raised in a timely manner in this case.

The defendants tell us that anything that makes it easier for a settlement class to molt into a litigation class will discourage the settlement of class actions. They say that defendants have settled class actions “in the past secure in the knowledge that if the settlement agreement should unravel, they would be restored to the pre-certification position and remain free to defend against any future effort to certify a class for litigation purposes.” But the defendants in this case were perfectly free to defend against certification; they just didn’t put up a persuasive defense. Anyway their argument is unrealistic. The pressures for settlement of class actions are enormous and will not be lessened significantly by our upholding the class certification.

We are mindful that no district judge has as yet explicitly addressed whether the other criteria for class certification, besides adequacy of representation of the class, have been met in this case. Those criteria are whether “(1) the class is so *664 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.” Fed.R.Civ.P. 23(a). There is no need for a remand on these questions, however. Criteria (1) and (2) have been met; and the satisfaction of (3) is implicit in Judge Bucklo’s rejection of the defendants’ contention that to handle their dispute with the class members in the class action format would be unmanageable. There has been substantial compliance with the requirements of the rule, and no more is required, Shvartsman v. Apfel, 138 F.3d 1196, 1201 (7th Cir.1998); Berger v. Iron Workers Reinforced Rodmen Local 201, 843 F.2d 1395, 1401 (D.C.Cir.1988), especially in a case in which the defendants were enthusiastic proponents of class treatment until their opportunistic change of heart.

AFFIRMED.