POSNER, Circuit Judge.
The Department of Labor sued the Central States, Southeast and Southwest Areas Health and Welfare Fund, an employee benefit fund sponsored by the Teamsters Union, along with current and former trustees of the fund, firms that had rendered services to the fund, and others, alleging violations of fiduciary obligations imposed by the Employee Retirement Income Security Act, 29 U.S.C. §§ 1001 et seq., and seeking both damages and injunctive relief. (The case has a lurid background, unnecessary to examine in this opinion, in the efforts of organized crime, acting through Allen Dorfman, to plunder the pension fund. Dorfman’s estate, and the company he controlled, Amalgamated, which processed the claims of the Fund’s beneficiaries, are among the other defendants in the case.) The Department negotiated a settlement with the Fund and its current trustees, and embodied it in a consent decree that the parties to the decree then submitted to Judge Will, presiding over the litigation, for his approval. Other defendants objected to the terms of the decree, the judge refused to approve it, and the Department and the settling defendants appeal from his refusal.
We have first to examine our appellate jurisdiction. The judge’s action in refusing to approve the decree was not a final decision, appealable under 28 U.S.C. § 1291. The suit in which the order was entered remains pending before him, and will soon be tried. It is as if he had denied a motion to dismiss the complaint; and such a denial is the classic example of a nonfinal order. Although 28 U.S.C. § 1292(a)(1) authorizes immediate appeal of an interlocutory order denying (as well as one granting or modifying, etc.) an injunction, the consent decree that the Department of Labor and the settling defendants asked the judge to sign contains a permanent rather than a temporary injunction, intended to wind up the litigation between the parties to the decree rather than to provide interim relief. It might seem that this could make no difference. The statute speaks not of interlocutory injunctions but of interlocutory orders denying (or granting, etc.) injunctions. Judge Will’s order refused an injunction, and was interlocutory — not only because the lawsuit is continuing against other defendants but also because the order did not finally deny the plaintiff’s right to an injunction but merely deferred consideration to such later time as the parties submitted a revised decree. (For the sake of completeness we point out that an order granting a permanent injunc- • tion may be interlocutory too. The case may be continuing against other defendants; or the plaintiff may be seeking damages as well as an injunction, and the damages have yet to be assessed.) Thus it comes as no surprise that many cases, beginning with Smith v. Vulcan Iron Works, 165 U.S. 518, 17 S.Ct. 407, 41 L.Ed. 810 (1897) (construing section 7 of the Evarts Act, 26 Stat. 828 — the original of 1292(a)(1)), hold or assume that interlocutory orders granting or denying permanent injunctions are indeed appealable under section 1292(a)(1). See 16 Wright, Miller, Cooper & Gressman, Federal Practice and Procedure § 3924, at pp. 67-69 (1977); id., 1983 Pocket Part, § 3924 (1984).
But this principle is qualified in an important line of cases led by Switzerland Cheese Ass’n, Inc. v. E. Horne’s Market, Inc., 385 U.S. 23, 25, 87 S.Ct. 193, 195, 17 L.Ed.2d 23 (1966), where the Supreme Court held that the denial of the plaintiff’s motion for summary judgment, which if granted would have resulted in the entry of a permanent injunction, was not appealable under section 1292(a)(1), the denial being “strictly a pretrial order that decides only one thing — that the case should go to trial.” See also Plymouth County Nuclear Information Comm., Inc. v. Boston Edison Co., 655 F.2d 15, 17-18 (1st Cir.1981); [1173]*1173Williams Electronics, Inc. v. Artic Int'l, Inc., 685 F.2d 870, 871 (3d Cir.1982). Behind these cases is the perceived lack of symmetry between an interlocutory order refusing to grant a permanent injunction and one refusing to grant a preliminary injunction. A request for a permanent injunction does not have the emergency character of a request for a preliminary injunction, which requires the plaintiff to show that he will suffer irreparable harm if the request is denied. See, e.g., American Can Co. v. Mansukhani, 742 F.2d 314, 325 (7th Cir.1984). Obviously, if the plaintiff does not get the preliminary injunction his need for an immediate appeal, to avert claimed irreparable harm, is urgent. Less obviously, if he does get the injunction the defendant’s need for an immediate appeal is urgent. Not only might the defendant suffer irreparable harm from having to obey the injunction but the speed with which the district court has to act on a request for a preliminary injunction makes its decision less reliable — therefore more in need of appellate review — than a more considered decision. And appellate review of its decision must be immediate to be effective; if it is postponed to the end of the ease, the question whether the district court should have granted a preliminary injunction will be mooted by the entry of final judgment. See Sierra On-Line, Inc. v. Phoenix Software, Inc., 739 F.2d 1415, 1422 (9th Cir.1984).
The problem is to integrate this insight about the practical differences between interlocutory orders denying preliminary injunctions and interlocutory orders denying permanent injunctions with the language of section 1292(a)(1), which does not distinguish between the two types of order. The solution implied by Switzerland Cheese Ass’n is to distinguish “postponing” from “denying” a request for injunctive relief: if the request is merely postponed, as in Switzerland Cheese Ass’n, appeal is not automatically allowed. This distinction implies, and later cases confirm, that a definitive disposition of a request for a permanent injunction is appealable under section 1292(a)(1), consistently with the Smith v. Vulcan Iron Works line of cases. See, e.g., Milonas v. Williams, 648 F.2d 688 (10th Cir.1981) (per curiam); EEOC v. International Longshoremen’s Ass’n, 511 F.2d 273, 276-77 (5th Cir.1975).
All this is by way of necessary background to the Supreme Court’s decision in Carson v. American Brands, Inc., 450 U.S. 79, 83-84, 101 S.Ct. 993, 996-997, 67 L.Ed.2d 59 (1981), which allowed an appeal under section 1292(a)(1) from an order refusing to approve a consent decree that, as in the present case, included a permanent injunction. If Carson is read as conferring blanket permission to appeal under section 1292(a)(1) all refusals to approve consent decrees that have injunctive provisions (as all do, for reasons we shall explain — and not only because the term “decree” is itself equitable), any doubt about appealability in this case disappears. Such a reading is suggested, but without discussion of the point, in our decision in United States v. City of Chicago, 663 F.2d 1354, 1359 n. 15 (7th Cir.1981) (en banc); see also Williams v. City of New Orleans, 694 F.2d 987, 989 (5th Cir.1982), on rehearing, 729 F.2d 1554 (5th Cir.1984) (en banc); Gould v. Control Laser Corp., 650 F.2d 617, 621 (5th Cir.1981); United States v. City of Alexandria, 614 F.2d 1358, 1361 n. 5 (5th Cir.1980); Plummer v. Chemical Bank, 668 F.2d 654, 655 (2d Cir.1982). Other cases, however, suggest that the appellant must show that the refusal to approve the decree caused him irreparable harm. See New York v. Diarylea Coop. Inc., 698 F.2d 567, 570 (2d Cir.1983); In re Flight Transportation Corp. Securities Litigation, 730 F.2d 1128, 1133-34 (8th Cir.1984); Roberts v. St. Regis Paper Co., 653 F.2d 166, 170 (5th Cir.1981); cf. Plymouth County Nuclear Information Comm., Inc. v. Boston Edison Co., supra, 655 F.2d at 18.
The source of confusion is that Carson seems to be about two different things— the appealability of orders disapproving consent decrees that contain permanent injunctions, and the appealability of orders of any sort that have the same consequences as orders denying preliminary injunctions. [1174]*1174See, e.g., South Bend Consumers Club, Inc. v. United Consumers Club, Inc., 742 F.2d 392, 393-94 (7th Cir.1984) (applying Carson to summary judgment that precluded injunctive relief as matter of law); Winterland Concessions Co. v. Trela, 735 F.2d 257, 261 (7th Cir.1984) (dismissal of counterclaim that sought injunction). The order in Carson partook of both types, since the proposed decree was not captioned “injunction” but did include a permanent injunction. Switzerland Cheese Ass’n had rejected the idea that every interlocutory order turning down a request for a permanent injunction is immediately appealable, but it did not say that no such orders were appealable, and indeed it implied (consistently with our suggested interpretation of it) that some were. See 385 U.S. at 25, 87 S.Ct. at 195. Carson interprets Switzerland Cheese Ass’n to hold that an interlocutory order refusing a permanent injunction is appealable immediately if but only if the order has the same effect (“serious, perhaps irreparable, consequence,” in the language of an earlier case) as the denial of a preliminary injunction. See 450 U.S. at 84-85, 101 S.Ct. at 996-997. This test — which we take to be the heart of Carson — makes excellent sense when we consider that in both Carson and Switzerland Cheese Ass’n the order sought to be appealed postponed, rather than definitively denied, the request for a permanent injunction. If a consent decree is rejected, the parties may be able to persuade the judge to accept a modified decree; or the plaintiff may be able to obtain such a decree after the trial, just as a plaintiff whose request for summary judgment granting a permanent injunction is denied may be able to obtain the same injunction after trial.
Carson, we conclude, requires that irreparable harm be shown whenever a party wants to appeal immediately either an interlocutory order deferring the entry of a permanent injunction, whether free-standing or contained in a proposed consent decree, or an interlocutory order that while not explicitly the grant or denial of a preliminary injunction may have consequences (summed up in the words “irreparable harm”) similar to those of such an order.
The Second Circuit’s Dairylea opinion offers a refinement of Carson. Concerned lest the courts of appeals be flooded with appeals challenging trivial reservations made by district judges to proposed consent decrees, reservations that could easily have been met by the parties’ modifying the decree and then resubmitting it, the court in Dairylea held that the appellant must show that the judge's reservations cannot be met easily in that way. See 698 F.2d at 570. We doubt whether this particular refinement is necessary. In regard to appeals from orders expressly denying preliminary injunctions, courts of appeals have not found it necessary to make it a condition of appeal that the plaintiff try to satisfy any technical objection to the proposed injunction that the district judge might have. The reason is not just that preliminary injunctions tend to be simpler than the injunctive provisions written into many modern consent decrees; it is also that a party seeking an injunction (or parties, if the injunction is contained in a consent decree) who could get it by satisfying minor reservations of the district court would do so voluntarily in order to avoid the delay and uncertainty of the appellate process.
But we need not decide in this case whether to follow the lead of Dairylea and limit the rule of Carson to cases where the parties to the consent decree shoulder the burden of actually proving that they are at an impasse with the district judge. Although Judge Will did invite the settling parties to resubmit the decree with changes responding to his concerns, prompt approval of a resubmitted decree was not in the cards. This is true, as we shall see, even though the judge’s main concern was with the damage provisions of the decree rather than the injunctive provisions. The decree requires the Fund’s current trustees to pay $1.8 million into the Fund, an amount calculated as follows. The parties estimated that the judgment that would be brought in against these trustees at the conclusion of the trial, when [1175]*1175discounted by the probability that the Department of Labor would win at trial, was worth $2.1 million, but that the trustees’ insurance coverage allocable to the Department’s claim was only $1.8 million. Thus the settlement was, in effect, for the limits of the settling defendants’ insurance coverage. The decree adds that the settlement is not an admission or declaration regarding those defendants’ liability, if any, to pay contribution or indemnity to any non-settling defendants eventually made to pay damages to the plaintiff. Concerned that the settlement might affect the rights of the nonsettling defendants, the judge asked the parties to present evidence that the settlement was reasonable in light of the relative culpability of the settling and nonsettling defendants.
The judge described his objections to the injunctive provisions of the decree as going to “less central” aspects of it — though the injunctive provisions are extensive. They not only require compliance with ERISA by the Fund and its current and future trustees, under penalty of contempt for noncompliance, but alter the policies and even structure of the Fund, for example by requiring it to create an internal audit staff. The judge registered no objection to these provisions; his objections center on the provision for an “Independent Special Counsel.” He is to be William Saxbe, a former attorney general of the United States, and is to play a broad “watchdog” role over the management of the Fund, subject to supervision by the district court, which is to retain jurisdiction to make sure that the terms of the decree are carried out. Some of the objections the judge himself described as “technical,” an apt description of minor problems that the parties could easily have been left to work out between themselves after the decree was in effect. His larger objections were four:
1. Although the decree denies the Independent Special Counsel “any right of participation in, or attendance at, collective bargaining meetings at which such contribution rates [i.e., the rates at which employers or employees contribute money to the Fund] are negotiated or at management or union prenegotiation meetings at which their respective bargaining positions as to contribution rates are determined,” it does not in so many words exclude the Independent Special Counsel from all collective bargaining and strategy meetings, for example collective bargaining meetings where benefit levels are discussed but not contribution rates. In fact the decree allows the Independent Special Counsel to “attend any other meetings of any type whatsoever at which [Fund-related] matters are discussed or considered.”
2. There is an exception to the last-quoted authorization for meetings, or parts of meetings, “conducted for the sole purpose of negotiation of contribution rates or the determination of management and union negotiating strategy as regards such contribution rates.” But it does not explicitly bar the Independent Special Counsel from attending meetings of the executive boards of local unions.
3. The judge was concerned that the presence of the Independent Special Counsel at meetings attended by nonsettling defendants and their lawyers could lead to those defendants’ losing their attorney-client privilege.
4. The decree authorizes the district judge to modify it in accordance with “changed legal or factual circumstances, as and to the extent appropriate under United States v. Swift & Co., 286 U.S. 106, 119-20 [52 S.Ct. 460, 464, 76 L.Ed. 999 (1932)].” The judge thought this traditional test for the modification of consent decrees (see 11 Wright & Miller, Federal Practice and Procedure § 2961, at pp. 601-05 (1973)) too confining given the Independent Special Counsel’s novel role and broad authority.
All of these objections could be met by wording changes that the Department and the settling defendants should have little difficulty (but for the dynamics of negotiation, about which more later) in agreeing on in short order. If they were the judge’s only objections to the decree we would be baffled as to why the parties had appealed to us rather than resubmitted the [1176]*1176decree, appropriately modified, to the district judge, and then we would have to decide whether we agree with Dairylea that a refusal to approve a consent decree cannot be appealed when based on minor flaws in the decree. But they were not the judge’s only or major objections. His big objection was to what he considered the inadequate justification for the monetary settlement. And to overcome that objection the settling parties would have had to undertake an ambitious project of estimating the relative blameworthiness of the numerous defendants in this complex litigation. Counsel for one of the nonsettling defendants conceded at argument that this project might have to be deferred to the end of the trial because the relative blameworthiness of the defendants will be extremely speculative until then. That might be many months away. The settlement might not hold together that long; if it did, it might become moot, as in Santana v. Collazo, 714 F.2d 1172, 1175 (1st Cir.1983). The coming into effect of equitable provisions designed to protect the pension rights of thousands of workers would be delayed for a long time — perhaps forever. Thus the judge’s action has — given his grounds — imposed an indefinite delay having enough of the practical consequences of denying a preliminary injunction to allow an appeal under section 1292(a)(1) if Carson is interpreted, as we think it should be, to require that an interlocutory order deferring action on, rather than denying, a permanent injunction must, to be appealable forthwith, inflict irreparable harm. Although the stumbling block to the district judge’s approving the decree was a noninjunctive provision, all that matters is that because of this stumbling block the district judge cannot be mollified, and the injunctive provisions put into effect, simply by the parties’ submitting to the judge a revised decree that straightens out the language of their previous submission.
Having satisfied ourselves at last that we have jurisdiction of the appeal, we must consider what standard a district judge should use in deciding whether to sign a consent decree and what standard we should use in reviewing his decision. When parties want to settle a case, they need only file a stipulation of dismissal under Fed.R.Civ.P. 41(a)(1), and the terms of the settlement will not be reviewed at all. But as shown by Rule 41(a)(l)’s cautionary reference to Rule 23(e), which requires judicial approval of dismissals and compromises of class actions, the reason for judicial laissez-faire regarding settlements is that a simple dismissal will not affect (not demonstrably, anyway) third parties or involve the judge in carrying out the underlying settlement. It is different if the judge is asked to sign a consent decree, which virtually by definition will contain equitable provisions. (If the settlement were purely monetary, the defendant would pay and the suit would be dismissed by stipulation.)
A federal judge has the full powers of an equity judge. So if third parties complain to him that the decree will be inequitable because it will harm them unjustly, he cannot just brush their complaints aside. See Bass v. Federal Savings & Loan Ins. Corp., 698 F.2d 328, 330 (7th Cir.1983); United States v. City of Miami, 614 F.2d 1322, 1330-34 (5th Cir.1980), modified on other grounds, 664 F.2d 435 (5th Cir.1981) (en banc). Even if no third party complains, the judge has to consider whether the decree he is being asked to sign is lawful and reasonable, see Williams v. Vukovich, 720 F.2d 909, 920 (6th Cir.1983), as every judicial act must be. If a consent decree provided that a violator could be punished by having his ears cut off, the judge could not sign it; nor could he if the decree placed on him duties either inappropriate to the judicial role or excessively time-consuming in relation to his other responsibilities. Judge Will could not sign a decree which provided that he would attend meetings at which contributions to the Central States Health and Welfare Fund are discussed.
Although a judge thus must, before signing an equity decree that either affects third parties or imposes continuing duties [1177]*1177on him, satisfy himself that the decree is reasonable (“fair, reasonable and adequate,” in the usual formulation, e.g., Weinberger v. Kendrick, 698 F.2d 61, 73 (2d Cir.1982) (Friendly, J.), but we think “reasonable” sums it up fairly and adequately), how deeply the judge must inquire, what factors he must take into account, and what weight he should give the settling parties’ desires will vary with the circumstances. The flexible character of the decision makes generalization difficult; but it is safe to suggest that the limitations of judicial competence and the desirability of encouraging out-of-court settlements in order to lighten the judicial caseload create a presumption in favor of approving the settlement. See United States v. City of Miami, supra, 614 F.2d at 1331-34, for an excellent discussion. Further refinement of the standard is unnecessary for the decision of this case.
Turning to the standard of appellate review, we point out that the district judge’s decision to approve a settlement is reviewed under a highly deferential version of the “abuse of discretion” standard, see, e.g., Air Line Stewards & Stewardesses Ass’n, Local 550 v. Trans World Airlines, Inc., 630 F.2d 1164, 1167 (7th Cir.1980), aff’d under the name of Zipes v. Trans World Airlines, Inc., 455 U.S. 385, 102 S.Ct. 1127, 71 L.Ed.2d 234 (1982), but that relatively few cases involve the review of a decision disapproving a settlement and not all discuss the standard of review. Abuse of discretion is the standard in Williams v. City of New Orleans, supra, 694 F.2d at 992, on rehearing, 729 F.2d at 1558-59 (en banc), and In re International House of Pancakes Franchise Litigation, 487 F.2d 303, 304 (8th Cir.1973), and implicitly in SEC v. Randolph, 736 F.2d 525, 529 (9th Cir.1984), and Carson v. American Brands, Inc., 654 F.2d 300, 301 (4th Cir.1981) (per curiam) (en banc); see also United Founders Life Ins. Co. v. Consumers Nat’l Life Ins. Co., 447 F.2d 647, 655 (7th Cir.1971) (dictum). But United States v. City of Alexandria, supra, 614 F.2d at 1361-62, rejects the Pancakes precedent and holds that the proper standard is review de novo — though the circumstances in City of Alexandria were extraordinary, as stressed in Williams, see 694 F.2d at 991, on rehearing, 729 F.2d at 1558-59. The district judge had refused to sign a fairly standard civil-rights consent decree even though no one had opposed it; and since the district judge took no evidence concerning the proposed decree, he had little basis on which to exercise an informed discretion as to whether or not to approve it.
Putting such an unusual situation to one side, we do not think there should be two standards of appellate review, depending on whether the district judge approves or disapproves the decree. The policy of encouraging settlements can easily be factored into an abuse of discretion analysis by noting that the district judge is to exercise his discretion in accordance with what Judge Friendly (and this court) has called a “principle of preference.” See Friendly, Indiscretion About Discretion, 31 Emory L.J. 747, 768 (1982); Coyne-Delany Co. v. Capital Development Bd., 717 F.2d 385, 392 (7th Cir.1983). This means a principle that the district court must justify any departure from. The principle here is that settlements are favored and ordinarily should be approved.
But as so often in dealing with the standard of review, the verbal formulation of the standard may not make much practical difference. Reviewing courts want to correct errors affecting substantial rights but realize that their ability to determine whether an error has been committed is quite limited in some circumstances. For example, if the trial judge’s determination is based on factors that are not accessible to the reviewing court, such as the credibility of witnesses, the court may not be able to determine whether the trial judge was right but only whether he was reasonable; and likewise if a judge accepts, or rejects, a proposed consent decree only after weighing a large number of imponderables. But it is not important in this case to make refined distinctions among rival theories of [1178]*1178appellate review. Even when the standard is abuse of discretion, review for errors of law is plenary, see, e.g., Pratte v. NLRB, 683 F.2d 1038, 1044 (7th Cir.1982); and we think that the district judge’s rejection of the proposed consent decree rests primarily on just such an error. As noted earlier, his principal reservation concerned the settlement’s financial terms. He thought the settling parties had not put in enough evidence to show that those terms were reasonable. But his analysis rests on a (thoroughly understandable) misconception of the confused and confusing law of contribution.
At common law a tortfeasor could sometimes obtain from a joint tortfeasor complete reimbursement of any compensatory damages he was forced to pay his victim (“indemnity”), but never partial reimbursement (“contribution”). Hillier v. Southern Towing Co., 714 F.2d 714, 719 (7th Cir.1983). (Punitive damages, where available, are assessed against each defendant individually.) We can put indemnity to one side. It is ordered only if a contract provides for it or the party seeking indemnity is so much less able to avoid the wrong to the plaintiff than the person against whom indemnity is being sought that the law desires to concentrate the whole deterrent effect of the judgment on that other person if possible. See generally Prosser and Keeton on the Law of Torts § 51 (5th ed. 1984). Not only is it unlikely that many of the nonsettling defendants in this case can obtain indemnity from the settling ones under these principles (though we need not decide that question), but we know of no principle under which a settlement could interfere with the rights of other tortfeasors to seek indemnity (as opposed to contribution) from the settling defendants. Cf. Uniform Contribution Among Tortfeasors Act, as revised in 1955, § 1(f), 12 U.L.A. 64 (1975).
Although the common law’s rejection of contribution among joint tortfeasors has itself been rejected by most states and most commentators, the Supreme Court remains reluctant to use its own common law powers to allow contribution under federal statutes that do not provide for it expressly. See Texas Industries, Inc. v. Radcliff Materials, Inc., 451 U.S. 630, 640-46, 101 S.Ct. 2061, 2066-69, 68 L.Ed.2d 500 (1981); Northwest Airlines v. Transport Workers Union, 451 U.S. 77, 95-98, 101 S.Ct. 1571, 1582-1584, 67 L.Ed.2d 750 (1981). And ER-ISA does not. But this court has twice in dictum said that contribution is available under ERISA, see Free v. Briody, 732 F.2d 1331, 1337 (7th Cir.1984); Alton Memorial Hospital v. Metropolitan Life Ins. Co., 656 F.2d 245, 250 (7th Cir.1981), and we shall assume, without having to decide, that these dicta are correct. If they are not — if there is no right of contribution under ERISA — the nonsettling defendants could not complain that the decree was interfering with their right of contribution; and that was the basis of the judge’s dissatisfaction with the settling parties’ efforts to justify the monetary settlement in the proposed decree. True, the nonsettling defendants would still have a practical interest in the settlement. Since a plaintiff's total recovery, from all the tortfeasors together, is not allowed to exceed his total damages, see Zenith Radio Corp. v. Hazeltine Research, Inc., 401 U.S. 321, 348, 91 S.Ct. 795, 811, 28 L.Ed.2d 77 (1971); Burlington Industries, Inc. v. Milliken & Co., 690 F.2d 380, 391 (4th Cir.1982), the amount that the nonsettling defendants will have to pay will be smaller, the larger the settlement is. But an interest is not necessarily a legally protected right, and this one is not; for when the law rejects a right of contribution, this means it refuses to extend its protection to a tortfeasor who complains of being forced to pay more than his fair share of the plaintiff’s damages. Under the common law rule of no contribution among tortfeasors, a plaintiff can if he wants satisfy his entire damage claim out of the assets of one defendant and let others who are equally or more culpable off scot-free — and the unlucky defendant has no recourse. See Keeton and Prosser on the Law of Torts, supra, § 50, at pp. 337-38. If that is the rule under ERISA, the nonsettling defendants have no right to [1179]*1179complain about the terms of the settlement. But we shall assume as we have said that ERISA allows contribution.
Now a settlement (“accord and satisfaction”) is for most purposes a contract. The money settlement in this case concludes the dispute between the Department of Labor and the current trustees but not the potential dispute between those trustees and the nonsettling defendants, and at least as a matter of contract law it cannot bind non-parties and thus it places no ceiling on the amount that the current trustees may someday be required to pay the other defendants by way of contribution. An example will help to show why this is so. Suppose that in the trial of the other defendants the Department obtains a judgment for $12 million, those defendants then seek contribution from the current trustees, and the court holds that the current trustees as a group should bear one third of the liability, or $4 million. As the trustees will already have paid $1.8 million in the settlement, the Department can collect only $10.2 million ($12 million minus $1.8 million) from the other defendants. Those defendants can, in turn (assuming, as we are, that ERISA creates a right of contribution), obtain $2.2 million from the current trustees, as this is the difference between the trustees’ hypothetical fair share of the total liability, $4 million, and what they have already paid the plaintiff, $1.8 million. The nonsettling defendants would therefore end up $8 million out of pocket ($10.2 million minus $2.2 million).
They would be no better off if the current trustees had settled with the Department for more — say, for $6 million instead of $1.8 million. In that case, since we are assuming that the current trustees’ adjudicated share of the total liability of $12 million is only $4 million, these trustees would be entitled (subject to a qualification to be noted shortly) to contribution from the nonsettling defendants of $2 million ($6 million minus $4 million). And $2 million, when added to the $6 million that those defendants would have to pay to the Department directly ($12 million, the judgment, minus $6 million, the amount the Department had already collected from the current trustees), makes $8 million — the same cost to the nonsettling defendants as under the smaller settlement.
Although, as these examples show, the settlement agreement cannot as a matter of contract law affect the contribution rights of the nonsettling defendants (non-parties to the settlement), we must also consider whether tort principles that might be applicable to the parties might limit the right of a nonsettling defendant to seek contribution from a settling one. They might indeed if a recent Illinois statute, Contribution Among Joint Tortfeasors, 111. Rev.Stat.1981, ch. 70, ¶ 302(c), were applicable, for it immunizes a settling defendant from any claim for contribution, whether or not the settlement agreement tries to confer such an immunity. But we do not think state law applicable here. Where contribution is sought by one who has had to pay damages for violating a federal statute, the scope and limitations of the right of contribution are invariably treated as questions of federal rather than state law. See, e.g., Laventhol, Krekstein, Horwath & Horwath v. Horwitch, 637 F.2d 672, 675 (9th Cir.1980); Burlington Industries, Inc. v. Milliken & Co., 690 F.2d 380, 391 (4th Cir.1982); Sabre Shipping Corp. v. American President Lines, Ltd., 298 F.Supp. 1339, 1343 (S.D.N.Y.1969). A departure from that principle would be unjustified in the case of ERISA fiduciaries. Cf. Free v. Briody, supra, 732 F.2d at 1336. The legislative history contains emphatic language of federal preemption, fairly illustrated by these words of Senator Williams: “It should be stressed that with the narrow exceptions specified in the bill, the substantive and enforcement provisions of the conference substitute are intended to preempt the field for Federal regulations, thus eliminating the threat of conflicting or inconsistent State and local regulation of employee benefit plans.” 120 Cong.Rec. 29933 (1974). See Shaw v. Delta Air Lines, Inc., 463 U.S. 85, 103 S.Ct. 2890, 2899-901, 77 L.Ed.2d 490 (1983). Although there is great peril in using general lan[1180]*1180guage to decide specific cases not foreseen by the speaker, the language we have quoted and much else besides in the legislative history indicate that Congress had no great concern with preserving state prerogatives in this area. This mood makes it extremely unlikely that Congress would have wanted ERISA fiduciaries to be subject to the vagaries of state contribution law — a body of law so various, mutable, complex, and uncertain that its application to ERISA fiduciaries might well result in subjecting them to inconsistent duties and a risk of multiple liability. ERISA fiduciaries are entitled to a uniform nationwide rule (we realize that such a rule may not emerge till the Supreme Court decides the question).
But what shall that rule be? ERISA does not say; and we shall therefore have to canvass the various possible rules for which there is some support. The traditional rule is that, just as in our examples based on contract law, a settlement does not limit the nonsettling defendants’ right to contribution. See, e.g., Byrnes v. Phoenix Assurance Co. of New York, 303 F.2d 649, 652-53 (7th Cir.1962); Prosser and Keeton on the Law of Torts, supra, § 50,-at p. 340. If that rule is applied here, then Judge Will was wrong to worry that the terms of the settlement might prejudice the rights of the nonsettling defendants. However, responding to criticisms of the traditional rule, section 4(b) of the Uniform Contribution Among Tortfeasors Act, as revised in 1955, 12 U.L.A. 98 (1975), reverses it; under section 4(b), a settlement prevents nonsettling defendants from getting contribution from settling ones (we shall call this the “settlement bar” rule). The approach is thus like that of Illinois’ contribution statute. But section 1(g) of the Uniform Act makes the entire act inapplicable to fiduciaries and hence to this case. (There is no comparable exclusion in the Illinois statute.)
Several cases, well discussed in Adamski, Contribution and Settlement in Multiparty Actions Under Rule 1 Ob-5, 66 Ia.L.Rev. 533, 544-47 (1981), support the application of the traditional rule to federal securities cases. See, e.g., Laventhol, Krekstein, Horwath & Horwath v. Horwitch, supra, 637 F.2d at 675. A different rule has been gaining ground, particularly in admiralty, but its consequences for this case are the same as those of the traditional rule; it, too, insulates nonsettling defendants from any adverse consequences of settlement. If the plaintiff wins a damage judgment, the court fixes the percentage of the plaintiff’s damages that is attributable to the fault of the settling defendants, multiplies that percentage by the judgment against the nonsettling defendants, and deducts the resulting amount from the judgment. For example, if the judgment is for $12 million and the settling defendants are found to be one-third responsible, the nonsettling defendants will have to pay the plaintiff only $8 million, regardless of the amount of the settlement, and thus will be unaffected by its terms. See Leger v. Drilling Well Control, Inc., 592 F.2d 1246, 1248 (5th Cir.1979); Kizer v. Peter Kiewit Sons’ Co., 489 F.Supp. 835, 840-41 (N.D.Cal.1980); Doyle v. United States, 441 F.Supp. 701, 710-13 n. 5 (D.S.C.1977). There is some support for this “comparative fault” rule in other areas of federal law. See Johnson v. Rogers, 621 F.2d 300, 303 (8th Cir.1980); Fund of Funds, Ltd. v. Arthur Andersen & Co., 545 F.Supp. 1314, 1381-82 (S.D.N.Y.1982); see also Uniform Comparative Fault Act, § 6, 12 U.L.A. 44 (1984 Pocket Part). The difference between the comparative-fault and traditional rules is that the former is more favorable to settling defendants (and less favorable to the plaintiff). But under neither rule can the nonsettling defendants be forced to pay more than their adjudicated fair share. Under comparative fault they pay that share directly to the plaintiff; under the traditional rule they may pay more to the plaintiff in the first instance but they can recover the difference between that amount and their adjudicated fair share in a suit against the settling defendants for contribution.
The objection to the traditional rule is that it discourages defendants from set[1181]*1181tling, because their potential liability to co-defendants remains open-ended. The problem could be solved very simply by not allowing contribution (or indemnity) from a settling defendant. This would encourage settlement, all right, but it would be contrary to the spirit of contribution, since it would allow guiltier defendants to get off cheaply by settling first. The consequences of the race to settle to the losers of the race are among the reasons for finding the old common law rule of no contribution objectionable. But this shows that the simple answer is no answer. Hence a court that is applying the settlement-bar rule is bound to insist that the settling defendant be allowed to cap his liability only if he is paying his fair share of the total liability to the plaintiff. This is the kind of fairness inquiry that Judge Will wanted to conduct. But it means bogging down the settlement process in a miniature trial before trial, for “in determining ‘good faith’ the court could consider the risk of victory or defeat, the risk of a high or lo ,v verdict, the unknown strengths or weaknesses of the opponent’s case, the inexact appraisal as to the elements of danger, the defendant’s solvency and the amount of insurance coverage.” Kissel, Developments in Third Party Practice — Contribution and Indemnity, 71 Ill.Bar J. 654, 660 (1983). The fairness hearing makes the settlement process more costly; and as the costs of settlement rise closer to those of trial, the likelihood of settlement falls— maybe far enough to offset the incentive to settle that a defendant has who knows that settling will enable him to avoid all liability to the other tortfeasors.
The comparative-fault rule provides a neat solution to these problems. It encourages settlement by immunizing the settling defendant from liability for contribution but does not require a hearing on the fairness of the settlement to other tortfeasors. And we have said that the comparative-fault rule has the same consequences for this case as the traditional one. But in any event, the objection to the traditional rule, that it discourages settlement, would be important only in a case where the settlement agreement tried to insulate the settling defendant from contribution claims by the other defendants. The presence of such a clause would be some evidence that settlement would indeed be infeasible, or at least less advantageous to the plaintiff, if the settling defendant was not allowed to cap his liability. But the agreement in the present case makes no effort to shield the settling defendants. They are willing to settle with the Department of Labor on mutually advantageous terms while taking their chances on being sued later for contribution or indemnity. Of course this may just be because they think the traditional rule, whereby a settlement cannot limit nonsettling defendants’ rights of contribution, will be applied. But whatever the reason, they have agreed to the settlement without taking any steps to invoke the settlement-bar rule, and that makes this case a poor vehicle for rejecting the traditional rule in favor of the settlement-bar rule on the ground that the traditional rule discourages settlement — especially when an alternative rule, that of comparative fault, meets that objection without requiring the kind of fairness hearing that Judge Will wanted. (Probably the reason the parties to the settlement agreement in this case were able to settle their dispute notwithstanding the settling defendants’ continuing exposure to liability for contribution is that the settlement went to the limits of those defendants’ insurance coverage. Water cannot be squeezed out of a stone; and a private individual without insurance coverage is, with rare exceptions, a stone from the standpoint of being compelled to pay substantial damages.)
To summarize, we decline to adopt the settlement-bar rule for ERISA cases. And therefore the amount of the settlement in this case cannot (more candidly, probably will not) have any effect on the other defendants, and the district judge should not have been concerned that by settling for as little as it did the Department of Labor was hurting them. Of course the Department may have been hurting itself, or the beneficiaries of the [1182]*1182Central States Health and Welfare Fund, by settling for the amount of money it did. But a settlement should not be set aside (other than in a class action, where most of the real plaintiffs in interest are not represented by lawyers of their choice) merely because it seems ungenerous to the plaintiff, who is presumed to know his own interests best, or the plaintiffs constituents. (This precept may have added force when the plaintiff is the United States.) The principle of Fed.R.Civ.P. 41(a)(1) is that settlements for money alone generally do not require judicial approval. Departure from this principle would be strikingly inappropriate in a case such as this where the settlement does not extinguish the plaintiffs rights but merely puts some change in his pocket while he litigates against the remaining defendants — though we grant that this particular plaintiff does not have to worry about how to finance its litigation. Often where there are multiple defendants the plaintiff will settle with individuals first, and for modest sums, before going after the corporate defendants; and because it is so difficult to collect large judgments from individuals unless they have insurance coverage, it is natural and ordinarily reasonable for a plaintiff to settle for the limits of that coverage. Cf. City of Detroit v. Grinnell Corp., 495 F.2d 448, 467 (2d Cir.1974); Frank v. Volkswagenwerk, A.G. of West Germany, 522 F.2d 321, 327 (3d Cir.1975).
Although third-party effects could not justify the district judge’s refusing to approve the decree’s financial terms, we must consider the possible third-party effects of those equitable features of the decree to which he also objected — features with which he had an independent concern, moreover, because he will have to administer the equity portions of the decree. But we do not think he would have refused to sign the decree merely because of these objections; and whether he would have or not, it is apparent from the briefs and argument in this court that the settling parties in fact interpret the decree in a way that meets his objections fully. The decree could be much better drafted than it is but this is a peculiarly inappropriate setting for counsels of perfection. Consent decrees are the products of negotiation and compromise, and often compromise is possible only because the parties agree to disagree, using vague words to defer decision to the future. See, e.g., Alliance to End Repression v. City of Chicago, 91 F.R.D. 182, 199-200 (N.D.Ill.1981). “Controversy does not help. Agreement is then possible only through escape to a higher level of discourse with greater ambiguity. This is one element which makes compromise possible.” Levi, An Introduction to Legal Reasoning 31 (1949).
The decree that Judge Will rejected was the third the parties had submitted to him. Though far from perfect, the third revision achieved substantial compliance with his reservations and no more should have been required. Although seemingly unable to find the words in which to give written expression to their agreement on the specific matters to which the judge objected, the parties agree that the decree is not intended to authorize the Special Independent Counsel to attend collective bargaining sessions or other strategy sessions, or to attend meetings of the executive committees of local unions; and this meets two of the judge’s objections. They also agree that the decree may bé modified not only if factual or legal circumstances change — and not only if there is “a clear showing of grievious wrong evoked by new and unforeseen conditions,” United States v. Swift & Co., supra, 286 U.S. at 119, 52 S.Ct. at 464 — but also if experience with the administration of the decree shows the need for modification. This is a component of the modern standard for modification of a consent decree. See, e.g., King-Seeley Thermos Co. v. Aladdin Industries, Inc., 418 F.2d 31, 35 (2d Cir.1969); New York State Ass’n for Retarded Children, Inc. v. Carey, 706 F.2d 956, 968-69 (2d Cir.1983) (Friendly, J.); Alliance to End Repression v. City of Chicago, 742 F.2d 1007, 1020 (7th Cir.1984) (en banc). And it is one, as Judge Will pointed out, that is particularly apt here, given the novelty of this decree.
[1183]*1183That leaves only the concern of the non-settling defendants with preserving their attorney-client privilege. This would be a serious concern if the decree were intended to permit the Special Independent Counsel to attend all conferences between the non-settling defendants and their lawyers, on the ground that since the management of the Fund is an issue in the lawsuit any such conference is bound to contain discussion of that management. But this is not the intent of the decree; the parties agree on this and their agreement is, once again, a binding interpretation. Of course if the nonsettling defendants happen to be present at a meeting — not a meeting to discuss legal strategy for defense of the lawsuit but a meeting to discuss the administration of the Fund — along with their attorneys, and the Special Independent Counsel is also present, as he would have a right to be, any communications between the defendants and their lawyers would not be privileged. They should not be; it is the wrong setting for exchanging confidences about legal strategy.
The decree, as interpreted by the parties, is reasonable; it should have been approved.
REVERSED AND REMANDED.