EASTERBROOK, Circuit Judge.

Hypothyroidism occurs when the thyroid gland fails to produce sufficient hormones. Symptoms include fatigue, extreme sensitivity to cold, joint pains, muscle aches, and weight gain. Left untreated, victims eventually suffer hair loss, numbness in the limbs, depression, and mental confusion. For more than 40 years hormone-replacement therapy based on levothyroxine sodium has been used to alleviate these symptoms. Introduced during the 1950s, Synthroid was the first orally administered levothyroxine product. This synthetic derivative of thyroxine is a "narrow therapeutic index" drug, meaning that dosage levels must be established for each recipient by trial and error, which may take several months. Too much or too little can cause heart, brain, psychological, and reproductive problems. Although levothyroxine sodium is not patented and is available from many vendors, Synthroid still represents more than two-thirds of sales. Its manufacturer, keen to maintain this lead, asserts that no other thyroid hormone drug is bioequivalent to Synthroid and warns physicians that switching brands may cause the side effects associated with incorrect levels of thyroid hormones in the blood, unless the patient goes through a new monitoring and calibration process.

If Synthroid is bioequivalent to other drugs containing levothyroxine, the tedious and costly calibration step can be omitted, and its less-expensive rivals would be likely to claim a greater share of the market. In 1990 Betty J. Dong, a professor of clinical pharmacy at the University of California, San Francisco, concluded that Synthroid and its rivals are interchangeable. To publish a paper based on the data collected in her study, Dong needed the permission of Knoll Pharmaceuticals, Synthroid's owner at the time. Knoll's predecessor Flint Laboratories had financed Dong's work under condition that she secure its approval before making any public disclosures. Knoll objected to publication, asserting that it found Dong's work methodologically sub-par. The conclusions nonetheless found their way to the press after Dong decided to dishonor the promises she had made to Flint. See Editorial: Thyroid Storm, 277 J. Am. Medical Ass'n 1238 (1997). Embarrassed by the accusation of covering up unfavorable information, Knoll permitted Dong to release the study. It was published as Dong, et al., Bioequivalence of Generic and Brand-name Levothyroxine Products in the Treatment of Hypothyroidism, 277 J. Am. Medical Ass'n 1205 (1997).

After the article's publication, lawyers across the country began filing class action suits. They sought relief under a variety of state and federal law theories, including antitrust, Rico, and state consumer-fraud statutes. These theories had in common the contention that Knoll misled physicians into keeping patients on Synthroid despite knowing (as Dong had concluded) that the physicians could have switched their patients to less costly but equally effective drugs. These suits were transferred to the Northern District of Illinois for consolidated pretrial proceedings under 28 U.S.C. § 1407. See Lexecon Inc. v. Milberg Weiss Bershad Hynes Lerach, 523 U.S. 26, 118 S.Ct. 956, 140 L.Ed.2d 62 (1998). Settlement talks ensued, and in mid-1998 the parties presented a compromise to the district judge. She rejected it, finding that too little discovery had been completed, that the size of the class was still unknown, and that the insurance companies that had actually paid for much of the Synthroid were not parties. 1998 WL 526566, 1998 U.S. Dist. Lexis 12936 (N.D.Ill. August 14, 1998). The court then split the plaintiffs into two classes: one of consumers and the other of insurance companies (also known as third-party payors or TPPs). 188 F.R.D. 287, 295 (N.D.Ill. 1999). After additional negotiations the parties submitted a second proposed settlement, under which Knoll and its former parent BASF Corporation would pay approximately $88 million to consumers and $46 million to the insurance companies in exchange for a release of all claims. (Abbott Laboratories, which purchased BASF's pharmaceutical business in March 2001, is not a party to the suit.) The district court rejected motions to intervene filed by several dissatisfied consumers and approved the settlement. 110 F.Supp.2d 676 (N.D.Ill. 2000). The court then awarded attorneys' fees from these common funds at a level significantly below what the lawyers had requested.

We deal first with the proposed intervention. The objectors, who call the settlement a sell-out, wanted to intervene so that they could obtain appellate review of any decision approving the deal. Status as a party is a condition to taking an appeal. See Marino v. Ortiz, 484 U.S. 301, 108 S.Ct. 586, 98 L.Ed.2d 629 (1988); Felzen v. Andreas, 134 F.3d 873 (7th Cir. 1998), affirmed by an equally divided Court under the name California Public Employees' Retirement System v. Felzen, 525 U.S. 315, 119 S.Ct. 720, 142 L.Ed.2d 766 (1999). The district judge's order approving the class-fund settlement devoted one sentence to intervention: "All pending motions to intervene . . . are denied." 110 F.Supp.2d at 686. That's all she said, either orally or in writing. This decision, which puts the objectors behind the eight ball, is impossible to reconcile with Crawford v. Equifax Payment Services, Inc., 201 F.3d 877 (7th Cir. 2000), which holds that "it is vital that district courts freely allow the intervention of unnamed class members who object to proposed settlements and want an option to appeal an adverse decision." 201 F.3d at 881. See also Griffith v. University Hospital, LLC, 249 F.3d 658, 661-62 (7th Cir. 2001); Southmark Corp. v. Cagan, 950 F.2d 416, 419 (7th Cir. 1991); Keith v. Daley, 764 F.2d 1265, 1272 (7th Cir. 1985). The district judge may have worried that, if she allowed intervention, objectors would enter the case and, as parties, block the settlement by withholding agreement to its terms. But as Crawford observed that worry is insubstantial; a judge can solve the problem by limiting intervenors to the privilege of appealing. For this limited-purpose intervention, it is irrelevant whether the class members come in under Rule 24(a) (intervention as of right) or Rule 24(b) (permissive intervention). See Crawford, 201 F.3d at 881; Vollmer v. Publishers Clearing House, 248 F.3d 698, 707 (7th Cir. 2001). District judges are not entitled to block appellate review of their decisions by the expedient of denying party status to anyone who seems likely to appeal, as the district judge apparently tried to do in this case. Crawford's requirement that "district courts freely allow the intervention of class members who object to proposed settlements" means that the intervenors must be given their say in this case. We reverse the district court and grant the objecting class members a place at the table.

Whether we can do anything for the intervenors now that they are parties is the next question. The intervenors appealed from the district court's denial of their motions to intervene, but not from the final judgment embodying the settlement. A decision reversing an order denying intervention usually leads to a remand, not to a decision on the merits. See Jessup v. Luther, 227 F.3d 993, 998-99 (7th Cir. 2000). Yet there would be nothing to do on remand here; the settlement's approval ended the case. No further appealable judgment could be entered, so the objecting class members seem to be out of luck. This problem has a ready solution, however. Blair v. Equifax Check Services, Inc., 181 F.3d 832, 834 (7th Cir. 1999), instructs putative intervenors that, when a substantive appeal is contingent on the success of the intervention appeal, they should file two notices of appeal: one from the denial of intervention and a second springing or contingent appeal from the final judgment — which will kick in if they are successful on the first. The approach is used elsewhere too, see Purcell v. BankAtlantic Financial Corp., 85 F.3d 1508, 1510-11 nn. 1, 2 (11th Cir. 1996), and has been obvious to class members in other cases. See Crawford; Cusack v. Bank United of Texas FSB, 159 F.3d 1040 (7th Cir. 1998). Our intervenors, however, failed to file a contingent appeal and called the approach of Blair and Cusack nonsensical. Why they spent time and money arguing the point rather than filing a precautionary springing notice of appeal from the final judgment is puzzling. After oral argument, however, the objectors took our advice. They filed a contingent notice of appeal with the district court. With today's decision that notice springs into effect (it is timely by analogy to Fed.R.App.P. 4(a)(2) because, until today, the objectors have not been entitled to appeal) and brings the district court's approval of the settlement before us.

Although officially in the game, the objectors have not presented any objection to the settlement that was not convincingly addressed by the district court. The objectors contend that the settlement should have been larger, that the notice was not sufficient, and that the release of liabilities is too broad. Yet it seems to us, as it did to the district judge, that the settlement is generous in light of the difficulties facing the class. Knoll owned the Dong study and cannot be required to pay damages for exercising its contractual rights. It might be held liable for fraud, if Dong's work proved to Knoll's satisfaction that Synthroid and other drugs are bioequivalent, and Knoll then tried to bamboozle the public by maintaining the opposite of what it knew to be the truth. But it would be hard to say that the study compelled Knoll to tell the public that levothyroxine products are interchangeable; other works in the medical literature reach a contrary conclusion. The Food and Drug Administration, which has been asked repeatedly to declare that at least some other levothyroxine medications are bioequivalent to Synthroid, has not done so in the four years since the Dong study's publication. (The FDA's Approved Drug Products with Therapeutic Equivalence Evaluations does not list any hypothyroid medications as therapeutically interchangeable.) Nor do physicians act as if they accept Dong's conclusions. The American Association of Clinical Endocrinologists recommends that patients not switch brands, and that, if they do switch, the dosage level be recalibrated. See Clinical Practice Guidelines for Evaluation and Treatment of Hyperthyroidism and Hypothyroidism, available at www.aace.com/clin/guides/thyroid_guide.html. Synthroid's sales are up. Although its share has dropped from 71% to 64% since Dong's study appeared, the market is growing, for all the data reveal the entire drop in market share may have been from new patients starting on different medications. Add to this the difficulty of proving damages when many of the consumers do not bear the full expense of the drugs (and often do not purchase them directly), and the plaintiffs would have had a headache trying to get any judgment on the merits. See Illinois Brick Co. v. Illinois, 431 U.S. 720, 97 S.Ct. 2061, 52 L.Ed.2d 707 (1977) (federal law prohibits price fixing suits by indirect purchasers of a product); In re Brand Name Prescription Drugs Antitrust Litigation, 248 F.3d 668, 670 (7th Cir. 2001); Teamsters Health and Welfare Trust Fund v. Philip Morris Inc., 196 F.3d 818, 825-26 (7th Cir. 1999) (health insurers may not pursue direct litigation against tobacco companies accused of suppressing research on the health effects of cigarettes because the injuries are too remote, the chain of causation is too long, and the damages "wickedly hard" to calculate); Health Care Service Corp. v. Brown Williamson Tobacco Corp., 208 F.3d 579 (7th Cir. 2000).

Unlike members of the consumer class, TPPs are sophisticated purchasers of pharmaceuticals. Their consent to this deal shows that a larger judgment was unlikely. The objectors maintain that the consumers should have received a larger piece of the pie, but this is implausible. The settlement forecloses all subrogation rights the TPPs would have held against any consumer. Had the TPPs held back and sued in subrogation, they might well have taken almost the entire fund. Except for uninsured persons and Medicare patients without separate prescription drug coverage (who must pay for their own pills), most persons' expenses for Synthroid are paid by third parties. Insurance generally requires consumers to provide a co-payment for prescriptions; this payment rises only marginally (if at all) for more expensive drugs. And hypothyroid sufferers are likely to meet their deductibles whether they switch from Synthroid or not. That the consumers received two-thirds of the settlement funds seems more like a gift (or a public relations gesture) by the TPPs rather than a reason to upset the deal. The objectors' other grounds are well covered by the district court's opinion. The judge did not abuse her discretion in approving this settlement. It is time for the plaintiffs to receive their payments and for their lawyers to be paid.

Unless a class contracts privately over attorneys' fees, lawyers in class-fund cases must petition the court for their compensation. See Cook v. Niedert, 142 F.3d 1004, 1011 (7th Cir. 1998). Counsel for the consumers have asked for $26.3 million, about 29% of the consumers' recovery. The insurance companies' lawyers request $10 million, approximately 22% of the insurers' fund. As the district judge saw it:

Essentially there are two very large funds created and a great many people benefited; the objections were insubstantial; the class counsel were able and efficient; the litigation was fairly complex but short; the risk of nonpayment was moderate; and the class counsel devoted a fair amount of time to the case, but not a great amount compared to the size of the settlement.

110 F.Supp.2d at 684. Recoveries of "$75-200 million and more" constitute "megafunds," she continued. Following the approach of decisions in the Northern District of Georgia, the Southern District of Texas, and the Eastern District of Pennsylvania, the judge concluded that "fees in the range of 6-10% and even lower" are common in megafund cases. "[W]hen the figures hit the really big time," she said, larger fees constitute a windfall. Stating that "class counsel ha[d] done a fine job in terms of a speedy and professional resolution of a major class action", the judge awarded 10% of each fund to each set of counsel.

The judge did not explain why she decided to follow decisions of district courts in other jurisdictions, rather than decisions of the United States Court of Appeals for the Seventh Circuit. For the approach that these districts take, and that our district judge followed, cannot be reconciled with the approach our opinions adopt. We have held repeatedly that, when deciding on appropriate fee levels in common-fund cases, courts must do their best to award counsel the market price for legal services, in light of the risk of nonpayment and the normal rate of compensation in the market at the time. See Montgomery v. Aetna Plywood, Inc., 231 F.3d 399, 409 (7th Cir. 2000); Gaskill v. Gordon, 160 F.3d 361 (7th Cir. 1998); Florin v. Nationsbank of Georgia, N.A., 60 F.3d 1245 (7th Cir. 1995) (Florin II); Florin v. Nationsbank of Georgia, N.A., 34 F.3d 560 (7th Cir. 1994) (Florin I); In re Continental Illinois Securities Litigation, 985 F.2d 867 (7th Cir. 1993) (Continental II); In re Continental Illinois Securities Litigation, 962 F.2d 566 (7th Cir. 1992) (Continental I). Of these opinions only Florin II was cited or discussed by the district judge, and then for a question unrelated to the market rate of legal services. We have never suggested that a "megafund rule" trumps these market rates, or that as a matter of law no recovery can exceed 10% of a "megafund" even if counsel considering the representation in a hypothetical arms' length bargain at the outset of the case would decline the representation if offered only that prospective return.

The district judge defined megafunds as settlements of $75 million and up. Fees in "megafund" cases should be capped at 10% of the recovery, the judge held, although she recognized that fees of 30% and more are common and proper in smaller cases. This means that counsel for the consumer class could have received $22 million in fees had they settled for $74 million but were limited to $8.2 million in fees because they obtained an extra $14 million for their clients (the consumer fund, recall, is $88 million). Why there should be such a notch is a mystery. Markets would not tolerate that effect; the district court's approach compels it. A notch could be avoided if the 10% cap in "megafund" cases were applied only to the portion of the recovery that exceeded $74 million, but that is not what the district court did; it capped fees at 10% of the whole fund. Under the court's ruling, a $40 million settlement would have led to the same aggregate fees as the actual $132 million settlement. Private parties would never contract for such an arrangement, because it would eliminate counsel's incentive to press for more than $74 million from the defendants. Under the district court's approach, no sane lawyer would negotiate a settlement of more than $74 million and less than $225 million; even the higher figure would make sense only if it were no more costly to obtain $225 million for the class than to garner $74 million.

Having disapproved the megafund cap, we must remand — for the district judge did not attempt a market-based approach, even as an alternative holding. On remand the district court must estimate the terms of the contract that private plaintiffs would have negotiated with their lawyers, had bargaining occurred at the outset of the case (that is, when the risk of loss still existed). The best time to determine this rate is the beginning of the case, not the end (when hindsight alters the perception of the suit's riskiness, and sunk costs make it impossible for the lawyers to walk away if the fee is too low). This is what happens in actual markets. Individual clients and their lawyers never wait until after recovery is secured to contract for fees. They strike their bargains before work begins. Ethically lawyers must do this, but the same thing happens in markets for other professional services with different (or no) ethical codes. Many district judges have begun to follow the private model by setting fee schedules at the outset of class litigation — sometimes by auction, sometimes by negotiation, sometimes for a percentage of recovery, sometimes for a lodestar hourly rate and a multiplier for riskbearing. (The greater the risk of loss, the greater the incentive compensation required.) Timing is more important than the choice between negotiation and auction, or between percentage and hourly rates, for all of these systems have their shortcomings. See, e.g., Kirchoff v. Flynn, 786 F.2d 320 (7th Cir. 1986); Winand Emons, Expertise, Contingent Fees, and Insufficient Attorney Effort, 20 Int'l Rev. L. Econ. 21 (2000); Bruce L. Hay, Contingent Fees and Agency Costs, 25 J. Legal Stud. 503 (1996); Geoffrey P. Miller, Some Agency Problems in Settlement, 16 J. Legal Stud. 189 (1987); Daniel L. Rubinfeld Suzanne Scotchmer, Contingent Fees for Attorneys: An Economic Analysis, 24 RAND J. Econ. 343 (1993); Terry Thomason, Are Attorneys Paid What They're Worth? Contingent Fees and the Settlement Process, 20 J. Legal Stud. 187 (1991). Only ex ante can bargaining occur in the shadow of the litigation's uncertainty; only ex ante can the costs and benefits of particular systems and risk multipliers be assessed intelligently. Before the litigation occurs, a judge can design a fee structure that emulates the incentives a private client would put in place. At the same time, both counsel and class members can decide whether it is worthwhile to proceed with that compensation system in place. But in this case the district judge let the opportunity slip away, turning to fees only ex post. Now the court must set a fee by approximating the terms that would have been agreed to ex ante, had negotiations occurred.

The second circuit has criticized this court's market-mimicking approach on the ground that one "cannot know precisely what fees common fund plaintiffs in an efficient market for legal services would agree to". Goldberger v. Integrated Resources, Inc., 209 F.3d 43, 53 (2d Cir. 2000). "Instead," that court "adhere[s] to [the] practice that a fee award should be assessed based on scrutiny of the unique circumstances of each case, and `a jealous regard to the rights of those who are interested in the fund.'" We grant the premise; it is indeed impossible to know ex post the outcome of a hypothetical bargain ex ante. But a court can learn about similar bargains. That is at least a starting point. The second circuit's consider-everything approach, by contrast, lacks a benchmark; a list of factors without a rule of decision is just a chopped salad. Even Goldberger, which resorted to using a lodestar, had to look at the market rate for lawyers' hours. Determining lawyers' fees ex post is a perilous process. But any method other than looking to prevailing market rates assures random and potentially perverse results.

It is impossible to be sure what would have happened earlier, but some guides are available: the fee contracts some TPPs signed with their attorneys; data from large common-pool cases where fees were privately negotiated; and information on class-counsel auctions, where judges have entertained bids from different attorneys seeking the right to represent a class.

The first benchmark is actual agreements. Before joining the class, a group of more than 100 TPPs (the "Health Benefit Payers," as they call themselves) contracted with two law firms to represent them. Unfortunately they have not put the details in the record, but they tell us that the contracts provided for a 25% contingent fee at maximum. The "Porter Wright Group" (18 TPPs referred to collectively by their law firm's name) also negotiated with and hired counsel. Their setup allowed each insurance company to pick one of two fee options. Either the client paid Porter Wright's full costs and 70% of its normal hourly fees each month, with a 4% of recovery kicker at the end, or the client paid only costs each month but had to pony up 15% of the final settlement. Insurers are sophisticated purchasers of legal services, and these contracts define the market. Unfortunately, though, they identify a market mid-way through the case, after defendants already had agreed to pay substantial sums. The Porter Wright contracts provide little guidance on how to compute fees for the consumer class, because none of the consumers' lawyers was paid on an ongoing basis. For these reasons the contracts are of limited utility — but they do show that even after an initial settlement, with the TPPs' risk of loss slight, arms' length bargains did not yield a "megafund cap" on fees.

A second benchmark for determining legal fees is data from securities suits where large investors have chosen to hire counsel up front. Data about these ex ante arrangements have been widely available since the changes to securities practice wrought by legislation in the mid-1990s. At about the same time some district judges started conducting auctions for the right to be lead counsel, and the outcome of these auctions provide the third benchmark.

At first thought, auctions appear to be a poor mechanism for replicating the market price of legal services. Quality varies among lawyers, and awards net of fees could rise with the level of fees if a higher payment attracts the best counsel. We never see private clients auctioning off their legal work to the lowest bidder. Law firms pitch their services in negotiation — competing for business by demonstrating to potential clients that they provide quality legal work at good value. But the word "auction" is an imprecise description of the process that judges have used to choose lead counsel in class actions. Judges don't look for the lowest bid; they look for the best bid — just as any private individual would do in selecting a law firm, an advertising firm, or a construction company. See Wenderhold v. Cylink, 191 F.R.D. 600 (N.D.Cal. 2000). Bidding law firms provide the judge with firm profiles, testimonials of former clients, predictions of expected recovery, fee proposals, and arguments on why their firm provides good value. The judge in turn acts as an agent for the class, selecting the firm that seems likely to generate the highest recovery net of attorneys' fees. The court in Wenderhold compared bids by evaluating how much the class would take (after attorneys' fees) at different levels of damages. The court then evaluated the firms' resumes and determined that the low bidder would do as good or better a job than the other firms. See also In re Wells Fargo Securities Litigation, 157 F.R.D. 467 (N.D.Cal. 1994).

Auctions are less helpful ex post, and not only because it's too late to put the legal services up for bid. After settlement, we have lost all opportunity to put in place incentives for attorneys to secure larger awards. Courts must at this point simply determine what value the lawyers have provided. The judicial opinions from auction cases are helpful in this respect, though, because they provide detailed analysis of the market rate for attorneys facing different levels of risk. Forcing firms to bid at least approximates a market, providing the judge with multiple options. We cannot assume that judges always select the best bid. See In re Bank One Shareholders Class Actions, 96 F.Supp.2d 780 (N.D.Ill. 2000); In re Amino Acid Lysine Antitrust Litigation, 918 F.Supp. 1190 (N.D.Ill. 1996) (both winning bids provided the attorneys with a percentage of the recovery yet capped the fee award, eliminating any incentive for the lawyers to push for a larger recovery). But a court can examine the bids and the results to see what levels of compensation attorneys are willing to accept in competition.

The market rate for legal fees depends in part on the risk of nonpayment a firm agrees to bear, in part on the quality of its performance, in part on the amount of work necessary to resolve the litigation, and in part on the stakes of the case. Both negotiations and auctions often produce diminishing marginal fees when the recovery will not necessarily increase in proportion to the number of hours devoted to the case. In Oracle Securities Litigation, 132 F.R.D. 538 (N.D.Cal. 1990), the judge selected a bid with a declining contingent-fee scale, plus an early-settlement discount. 132 F.R.D. at 541, 548. The schedule provided for 30% of the first million, 25% of the next $4 million, then 20% of the next $10 million, and 15% of everything above $15 million. Following settlement, counsel received a total of $4.8 million, or 19.2% of the $25 million recovery. In re Oracle Securities Litigation, 852 F.Supp. 1437, 1457 (N.D.Cal. 1994). The advantages of such a structure over the fee schedule laid out by the district court in our case are apparent: in Oracle the attorneys' fees never went down for securing a larger kitty, and counsel always had an incentive to seek more for their clients. See also In re Bank One Shareholders Class Actions, 96 F.Supp.2d 780 (N.D.Ill. 2000) (providing for 17% of first $5 million, 12% of next $10 million, etc.); Amino Acid Lysine Antitrust Litigation, 918 F.Supp. 1190 (similar scale).

This is not to say that systems with declining marginal percentages are always best. They also create declining marginal returns to legal work, ensuring that at some point attorneys' opportunity cost will exceed the benefits of pushing for a larger recovery, even though extra work could benefit the client. This feature exacerbates the agency costs inherent in any percentage-of-recovery system, just as the lodestar approach creates the opposite incentive to run up the billable hours. Other options are available. Suppose that recovery is certain and that the major dispute concerns its amount. A fee schedule could be created that takes the base recovery for granted (no need to pay counsel to "produce" a recovery that is there for the asking) while magnifying the incentive to seek more. This was the approach of the successful bidder in In re Auction Houses Antitrust Litigation, 197 F.R.D. 71 (S.D.N.Y. 2000). Counsel agreed to take no fees for the first $405 million recovered and 25% of everything above $405 million. Because the government had already established liability in criminal proceedings, the civil lawyers in Auction Houses concluded that the first few hundred million would come easy. Convinced that their superior skills could generate a larger-than-expected recovery, they agreed to be paid only for the upper tail of the distribution of possible recoveries. (If the expected damages were $350 million, for example, then this structure would create a very powerful incentive to do well by the class; but it would be verboten under our district court's "megafund rule.")

Bids also differ in how they compensate lawyers for their time. "Calendar-based" bids (seen in Oracle) compensate lawyers for additional time spent. See also Wells Fargo and Amino Acid Lysine. Others compensate based on the stage of litigation, but not the amount of time it actually takes to get there. We are skeptical of the calendar-based bids, because they do little to reduce agency costs and tempt lawyers to delay settlement talks unnecessarily. Systems where fees rise based on the stage of litigation rather than the calendar are more common in private agreements (indeed they are the norm for contingent-fee contracts in tort suits). These tie the incentives of lawyers to those of the class by linking increased compensation to extra work.

The record compiled in this case is limited. We know the agreements between some TPPs and their lawyers, but we also know that they entered the case only after a large kitty seemed likely. The district court's task on remand will be to gather such additional data as the parties provide and to approximate the arrangements that would have prevailed at the outset. The district judge should keep in mind that she separated the consumer and TPP classes. The fees of each class's group lawyers should be determined by that group's risk and productivity. The consumer and TPP classes were rivals for a limited fund, not confederates in common cause. Any use of a sliding scale should be based solely on each class's settlement, not the total amount recovered by the two classes.

Two additional issues require only brief discussion. The district court declined to authorize full reimbursement for the expenses of litigation. The judge wrote that too many claims had been lumped into broad categories, that she was perplexed by the difference between reproduction and copying expenses, and that travel and telephone costs were too high. She reduced the TPP counsels' request by one third and consumer counsels' request by one half. Reducing litigation expenses because they are higher than the private market would permit is fine; reducing them because the district judge thinks costs too high in general is not. See Continental I, 962 F.2d at 570; Dutchak v. Central States Pension Fund, 932 F.2d 591, 597 (7th Cir. 1991); Heiar v. Crawford County, 746 F.2d 1190, 1204 (7th Cir. 1984). Likewise the amount of itemization and detail required is a question for the market. If counsel submit bills with the level of detail that paying clients find satisfactory, a federal court should not require more. See, e.g., Medcom Holding Co. v. Baxter Travenol Laboratories, Inc., 200 F.3d 518 (7th Cir. 1999).

Counsel contend that their expenses and their billing methods are normal in commercial practice. They explain that "copying" charges denote outside copy centers while "reproduction" charges denote in-house copying. They say that their telephone and travel expenses are standard fare for a case of this magnitude. Our remand instructions regarding expenses mirror those on fees. Parties should submit information on what expenses private clients in large class actions (auctions and otherwise) pay. In examining market rates the court should be careful not to pay the attorneys twice. Some cases award large percentages of the recovery in order to cover for litigation expenses as well. If the district court selects such a fee structure, it should not separately reimburse expenses.

Finally, the district court awarded incentive reimbursements to named consumer class plaintiffs but not to the TPPs' representatives. The judge explained that the "cases cited by the third-party payers involve awards to individuals who took significant risks . . . not to corporations who will receive large recoveries in any event." The TPP representatives appeal this decision. Incentive awards are justified when necessary to induce individuals to become named representatives. See Montgomery, 231 F.3d at 410; Cook, 142 F.3d at 1016; Continental I, 962 F.2d at 571-72. We see no reason why this rationale would not apply equally to corporations and other organizations. But if at least one TPP would have stepped forward without the lure of an "incentive award," there is no need for such additional compensation, as the district judge recognized. Many insurance companies entered this suit knowing that a large payment was on its way. The TPPs were "clear winners" from the start. See Continental I, 962 F.2d at 572. Appellants argue that some of the TPPs didn't shoulder a reasonable portion of the discovery burden. To the extent one company's law firm played a larger role than others, that difference must be accounted for in divvying up attorneys' fees. But it is evident that the prospect of recovery created enough incentive for the representative TPPs to step up even without an additional award for doing so. The market rate for incentive reimbursements here is accordingly zero.

Affirmed in part, Vacated in part, and Remanded in part for proceedings consistent with this opinion.