dissenting.
In denying Judith Perlman an opportunity to have her claim for disability benefits assessed by the administrator of the Swiss Bank Corporation Comprehensive Disability Plan on the basis of a proper record, the majority has misapplied the standard of review established in Firestone Tire & Rubber Co. v. Bruch, 489 U.S. 101, 109 S.Ct. 948, 103 L.Ed.2d 80 (1989), and has effectively precluded as a matter of law any procedural challenge to an ERISA plan administrator’s decisions, thereby giving those decisions a uniquely privileged position in the entire field of administrative or quasi administrative law. Even under the deferential standard of review Bruch requires in cases where plan administrators possess discretion, on this record the district court correctly held that UNUM’s decision could not stand. The only interesting question was whether the court was then entitled to give UNUM and Swiss Bank a second bite at the apple by remanding for further proceedings, or if the error required judgment for Perlman without further ado. On that point, following the general principle that the remedy should be tailored to the underlying problem, I would hold that a finding of abuse of discretion in the procedures followed may be corrected by an order requiring a second set of proceedings under correct procedures. Had the flaw instead been a substantive one, there would be no reason to require or permit another chance, and *984judgment could be entered immediately for the successful claimant.
Because the majority spends a considerable amount of time on the question of our appellate jurisdiction, I begin with a word about that. In the final analysis, I agree that we have jurisdiction over these appeals, but I reach that conclusion in a more straightforward way. The tortured analogy the majority draws to the reviewa-bility of district court decisions in Social Security Act eases that remand proceedings to the Social Security Administration under either sentence four or sentence six of 42 U.S.C. § 405(g) does not work in the end. I note, however, that if we were to take this analogy to its logical conclusion, it would lead to the same final result for which I am arguing, because despite the deferential standard of review to which decisions of the Social Security Administration are entitled, courts regularly remand cases to the agency for further proceedings when the administrative law judge has erroneously excluded relevant evidence or committed some other harmful procedural error. See, e.g., Sarchet v. Chater, 78 F.3d 305, 308-09 (7th Cir.1996) (remanding where ALJ failed to articulate reasons for rejecting claimant’s evidence); Herron v. Shalala, 19 F.3d 329, 333 (7th Cir.1994) (requiring ALJ to explain why he rejected claimant’s testimony regarding pain and loss of dexterity); Penn v. Sullivan, 896 F.2d 313, 317 (8th Cir.1990) (finding error in ALJ’s exclusion of evidence from hypothetical posed to vocational expert).
The statutory scheme ERISA creates is different in important ways from both the SSA and the Federal Arbitration Act (“FAA”) § 16(b), the other statute to which the majority analogizes. In order to bring an ERISA suit in the district court, the plaintiff must have a claim that the plan improperly denied some benefit to which she was entitled. 29 U.S.C. § 1132(a)(1)(B). The district court would thus never be in the position of ordering a preliminary determination by the administrator of the plaintiffs eligibility for benefits, because such a plaintiff would be out of court either on ripeness grounds or for lack of statutory standing to sue (since she had not yet been denied a benefit — properly or improperly). Contrast the situation of a party who tries to bring a suit in district court on a tort or contract theory, in the face of a valid arbitration agreement. The district court would issue an order to arbitrate, which would be unre-viewable under 9 U.S.C. § 16(b). In short, under the FAA the orders the district court might issue fall into two classes: those that are immediately reviewable, and those that are not. Under ERISA, all orders will follow final decisions about plan benefits, and thus all will be reviewable immediately, unless the analogy to the SSA works any better.
It does not. The majority correctly notes that immediate review is permitted for a decision remanding a case under sentence four of § 405(g), under which the district court finds error in the Commissioner’s decision. In contrast, a district court order remanding a case under sentence six of § 405(g) is not immediately reviewable, largely because a sentence six remand is ordinarily ordered without any ruling on the propriety of the' former decision. Melkonyan v. Sullivan, 501 U.S. 89, 98, 111 S.Ct. 2157, 115 L.Ed.2d 78 (1991). Moreover, “[s]entence-six remands may be ordered in only two situations: where the Secretary requests a remand before answering the complaint, or where new, material evidence is adduced that was for good cause not presented before the agency.” Shalala v. Schaefer, 509 U.S. 292, 297 n. 2, 113 S.Ct. 2625, 125 L.Ed.2d 239 (1993). In the ERISA context, however, if the plan administrator were willing to reconsider the plaintiffs claim, it is likely that the court action under § 1132(a)(1)(B) would be dismissed on ripeness grounds. If the case were properly before the district court, that court would overturn an administrator’s decision only after a finding of some error, either substantive or *985procedural; in no case has a district court simply refused to rule on the propriety of the administrator’s decision and remanded for more evidence, and that is not what the district court did here. Neither the cases that the majority cites, ie. Quinn v. Blue Cross & Blue Shield Ass’n, 161 F.3d 472 (7th Cir.1998); Schleibaum v. Kmart Corp., 153 F.3d 496 (7th Cir.1998), nor the precedents on which those cases rely, e.g., Gallo v. Amoco Corp., 102 F.3d 918 (7th Cir.1996); Wolfe v. J.C. Penney Co., 710 F.2d 388 (7th Cir.1983), stand for the proposition that an ERISA remand is appropriate without a finding of error, and thus might resemble a sentence six SSA remand. So, just as with the FAA, a distinction between reviewable and nonreviewable remands exists in the Social Security context that simply cannot be translated into an ERISA action.
There are other reasons to eschew the analogy between SSA and ERISA review. Most importantly, the SSA is a public agency, whose decisions are subject to the strictures of the Administrative Procedure Act, while ERISA plan administrators are private sector actors subject to regulation under the ERISA statute. A host of federal constitutional rights and statutory rights combine to assure procedural regularity in the case of public agencies that are not available to those who attack private action. In addition, the sources of substantive law differ significantly: in the case of the SSA, agencies and courts deal with an entitlement statute, while under ERISA the Supreme Court has noted that the proper source of guidance is the law of trusts. See Bruch, 489 U.S. at 111, 109 S.Ct. 948. The difference between decisions made by private actors such as ERISA plans and public agencies such as SSA is the reason that the Supreme Court has instructed the lower courts to “develop a federal common law of rights and obligations under ERISA-regulated plans,” id. at 110, 109 S.Ct. 948 (internal quotation marks omitted), rather than import wholesale a body of administrative law. Nor do we need the analogy to determine who should get fees; for ERISA purposes, whether a litigant is a prevailing party is determined according to the standards developed under 42 U.S.C. § 1988, Janowski v. Int’l Bhd. of Teamsters Local No. 710 Pension Fund, 812 F.2d 295, 297 (7th Cir.1987), which has nothing to do with Social Security disability.
If we reject the analogies to SSA sentence four remands and the FAA, then the question remains whether this court has proper appellate jurisdiction. I agree that we do, for the simple reason that the district court’s decision finally resolves the most important question in the litigation: is Perlman’s case over, as UNUM and Swiss Bank argued, or not? Indeed, this is also the question that underlies both the sentence four/sentence six distinction and the difference between § 16(a)(1)(E) and § 16(b), making the majority’s detour through these other areas of law unnecessary. We need not think of the district court’s remand here as “equivalent” to SSA or FAA remands — all are reviewable for precisely the same reasons. UNUM and Swiss Bank are here arguing that they have a right to a final order dismissing the action. Instead, the district court rejected that position and ordered the private parties to take certain action (ie. redo the eligibility determination asking the proper questions and listening to the proper evidence) — in effect, an injunction. A final order mandating private action is an ap-pealable order, whether under 28 U.S.C. § 1291 or perhaps even § 1292(a)(1). As does the majority, I therefore proceed to the merits of Perlman’s case.
I accept the majority’s conclusion that the language in the Swiss Bank plan stating that “[UNUM] will determine whether you are disabled for Plan purposes” and “[you] must give proof [to UNUM], at your own expense, that you are disabled ...” is enough to confer the kind of discretion on plan administrators that leads to deferential review under our cases. See, e.g., Patterson v. Caterpillar, Inc., 70 F.3d 503, *986505 (7th Cir.1995); Donato v. Metropolitan Life Ins. Co., 19 F.3d 375, 379 (7th Cir.1994). The more difficult question is how to give content to this review, avoiding both a rubber-stamp approach and undue intrusion. Bruch suggests that we must look to the law of trusts, which emphasizes the duty of a trustee to be faithful to the interests of the beneficiaries, and which led the Bruch Court to hold that a conflict of interest between an administrator and a claimant must be given some weight even under deferential review. 489 U.S. at 115, 109 S.Ct. 948.
The majority dismisses Bruch’s, admonition, doubting both the existence of such a conflict and a reviewing court’s ability to give it any weight. Regarding the first concern, I note first that the majority’s speculations about long-term reputational effects seem applicable regardless of whether a plan is retrospectively rated, yet Bruch obviously contemplates a conflict in at least some cases. Also, the lower court in this case found a potential conflict of interest. Perlman v. Swiss Bank Comprehensive Disability Protection Plan, 979 F.Supp. 726, 729 (N.D.Ill.1997). In the estate context, which is closely related to the trust law that Bruch tells us should guide our interpretation of ERISA, at least one state court has noted that the existence of a conflict is a mixed question of law and fact. Hitt v. Dumitrov, 598 S.W.2d 355, 356 (Tex.Civ.App.1980). The Supreme Court has said the same in determining conflicts of interest in other areas. Strickland v. Washington, 466 U.S. 668, 698, 104 S.Ct. 2052, 80 L.Ed.2d 674 (1984). Ordinarily, it is the practice of this court to treat such mixed questions with a “light appellate touch” unless there are constitutional concerns not present in this case. Dean Foods Co. v. Brancel, 187 F.3d 609, 616 (7th Cir.1999), citing United States v. Frederick, 182 F.3d 496, 499 (7th Cir.1999). Thus, while I do not mean to imply that this court must blindly follow a district court’s finding of a conflict, I think it inconsistent with our usual practice to dismiss the lower court’s conclusion without setting forth a powerful reason to do so.
With respect to the difficulties of giving weight to this conflict, “deferential review” has content only in application. It could mean, as the majority suggests, that the plan’s processes are insulated from any meaningful scrutiny. Or it could mean simply that, once all of the relevant facts are in place, the administrator’s final decision in a close (or maybe not-so-close) case should carry the day. Given that the Supreme Court has instructed us to do something with UNUM’s conflict of interest, the second path is the more appropriate one. Guarding against bias in a decision-maker is one way of assuring that the de-cisionmaking process itself was not structured in an unfair manner. It is, by the way, a routine part of review of administrative agency decisions, even when those decisions are entitled to some form of deferential review, such as the substantial evidence standard or the “arbitrary, capricious, and unreasonable” standard. As I discussed above, I think that the analogy to Social Security disability cases is both strained in theory and foreclosed by the Supreme Court. However, even accepting the majority’s conclusion that “we have no reason to think that the actual decision-makers at UNUM approached their task any differently than do the decisionmakers at the Social Security Administration,” ante at 981, there is no justification for refusing to examine how UNUM went about denying Perlman’s claim.
Generalizing from this rule, it follows that deferential review of plan administrator decisions does not preclude all consideration of the process by which the administrator came to its conclusion. To the contrary, that is precisely the point on which a court should be focusing, in order to give content to the rights conferred by § 1132 and at the same time avoid micro managing particular plan decisions. The panel’s approach appears to foreclose such an inquiry, at least as long as the administrator does something other than toss the *987claimant’s application into the trash, flip a coin, and announce a decision. This cannot be right. In this ease, Perlman has raised a serious challenge to the procedure UNUM followed after she submitted her 1994 claim for benefits. Specifically, she argues that it was arbitrary and capricious for UNUM to deny her benefits without collecting any expert medical evidence to support its decision and instead to assume that the fact that Perlman had been working after her injuries meant that she was not disabled.
On the merits, the majority takes the position that the district court should not have focused on “whether UNUM correctly understood [Perlman’s] abilities in relation to the demands of her job,” ante at 982, but in fact that was precisely the correct inquiry. Perlman’s claim is that her injury reduced her ability to work and that, by 1994, it had disabled her from performing her job. UNUM disagreed, and the district court’s task was to review the reasonableness of this conclusion. Given that the plan defines “disability” as a condition in which “the insured cannot perform each of the material duties of [her] occupation,” it is difficult to see how the relation between what Perlman could do and what her job required was anything but the principal issue before the district court. Later in its opinion, the majority appears to concede that this is indeed the relevant question, when it comments that “[a] person who can do the job at Swiss Bank is not ‘disabled’ as the program uses that term.” Ante at 982.
The fallacy in the majority’s reasoning, like the faEacy afflicting UNUM’s decision, lies in the assumption that the ability to do a job cannot diminish unless physical condition also deteriorates significantly. That assumption is what leads both to conclude that Perlman’s ability to do her job in 1992 necessarily means that she could still do her job in 1994, unless she had evidence indicating that something more had happened to her after the initial injuries and her return to work. This creates an irre-buttable presumption that, after a person returns to work after an injury, there is no way the same injury can give rise to a valid disability claim. Such a conclusion is especially strange given that, under the Social Security disability rules (which until this point the majority has treated as a nearly perfect analogue), an injured claimant may test her ability to work and still be considered disabled. 20 C.F.R. § 404.1592(a) (the “trial work period”).
UNUM and the majority adhere to an approach that makes a sweeping medical assumption that may have no basis in the facts of a particular case. Perhaps a person’s injury, if it renders her disabled at all, makes it impossible to work the day after her accident. But it is equally plausible that some injuries may be events that shorten (but do not immediately end) a person’s useful working life. If this is the case, the plan administrator would certainly be entitled to inquire into the circumstances following an injury to see if the claimant is really unable to work, or is simply unwilling to do so for reasons unrelated to the disability. But this inquiry cannot be conducted without looking at competent medical evidence about the claimant’s ability to work at the time the claim is filed. That evidence might show that the earlier injury set into motion some kind of degenerative process; it might show that the person had continued to work only by engaging in the kind of superhuman efforts the majority concedes are not required by the law (ante at 982-83); it might show that the person’s temporary ability to continue working was made possible only by frequent leaves of absence that had the effect of temporarily extending useful working life.
Perlman was entitled to introduce this type of evidence into the record before UNUM made a final decision on her claim. UNUM’s flat refusal to consider it rendered its final decision arbitrary and capricious for purposes of ERISA review. The district court, having had a more extensive opportunity to consider this case than we *988have, and whose view is also entitled to respect, correctly understood these points and chose to remedy the flaws by an order requiring further proceedings. I agree with that approach, and I therefore respectfully dissent.