OPINION

SUTTON, Circuit Judge.

Central States and Guarantee Trust both issued insurance coverage for the same claims. Central States' contract says that it will pay only if Guarantee Trust does not. Guarantee Trust's contract insists that it will pay only if Central States does not. We must break this "you first" paradox, sometimes called a gastonette. Black's Law Dictionary 795 (10th ed.2014); see Jon O. Newman, Birth of a Word, 13 Green Bag 2d 169 (2010).

I.

Central States, an employee benefit plan governed by the Employee Retirement Income Security Act, provides health insurance for Teamsters and their families. Guarantee Trust, an insurance company, provides sports injury insurance for student athletes.

This case involves thirteen high school and college students, all athletes and all children of Teamsters. Each of them holds general health insurance from Central States and sports injury insurance from Guarantee Trust. Each suffered an injury while playing sports (most often football) between 2006 and 2009, after which they sought insurance coverage from both insurance companies. Each time Guarantee Trust refused to pay the athlete's medical expenses, and each time Central States picked up the bill under protest.

Invoking one of ERISA's civil enforcement provisions, 29 U.S.C. § 1132(a)(3)(B), Central States sued Guarantee Trust and First Agency (a company that administers Guarantee Trust's insurance policies). The district court ruled that, when Central States' and Guarantee Trust's coverage of student athletes overlap, Guarantee Trust must pay. It entered a declaratory judgment to that effect, ordered Guarantee Trust to reimburse Central States for the payouts to the thirteen students, and awarded Central States attorneys' fees.

II.

Which company should pay for the students' medical expenses?

Central States' contract answers the question one way. In a provision captioned "Coordination of Benefits," the contract lists rules that determine which insurer has "primary responsibility" when plans overlap. R. 1-1 at 49. Under one of these rules, it says that whichever insurer covers the insured "other than as a Dependent" has primary responsibility. Id. Central States covers the thirteen students as dependents: The students have insurance because they are children of Teamsters. Guarantee Trust, by contrast, covers the thirteen students "other than as ... [d]ependent[s]": The students have insurance in their own names. So under Central States' contract, Guarantee Trust must pay for the students' medical expenses up to its maximum before Central States will contribute anything.

Guarantee Trust's contract answers the question another way. The contract contains a blanket coordination-of-benefits rule. If insurance provided by Guarantee Trust overlaps with insurance provided by anyone else, the other insurer always has primary responsibility. R. 31-2 at 11. So under Guarantee Trust's contract, Central States must pay for the students' medical expenses up to its maximum before Guarantee Trust will contribute anything.

When it comes to paying the students' medical bills, it thus seems that one insurance company has said: "You first, Central States." To which the other has responded: "After you, Guarantee Trust."

Of course, today's insurance-coverage dispute does not turn on matters of social etiquette. But a rule of legal etiquette points the way. If an ERISA plan and an insurance policy "contain conflicting coordination of benefits clauses," then as a matter of federal common law "the terms of the ERISA plan, including its [coordination of benefits] clause, must be given full effect." Auto Owners Ins. Co. v. Thorn Apple Valley, Inc., 31 F.3d 371, 374 (6th Cir.1994); see also Great-West Life & Annuity Ins. v. Allstate Ins., 202 F.3d 897, 900 (6th Cir.2000). Here, the terms of the ERISA plan — Central States' plan — say that Guarantee Trust has primary responsibility for the students' expenses. Guarantee Trust thus has primary responsibility for the students' expenses.

Guarantee Trust responds that its policy provides "excess insurance" rather than ordinary insurance. In the more ambitious passages of its brief, Guarantee Trust claims that ERISA plans may never coordinate benefits with excess policies. In the more modest passages of its brief, it accepts that an ERISA plan can overcome excess policies, but only if the plan's coordination clause expressly refers to excess insurance. Neither incarnation of the argument gets Guarantee Trust where it wants to go.

An excess policy (at least a "pure" or "true" excess policy) supplements an insured's main policy by providing an extra layer of coverage. It does so by covering losses in excess of a stated threshold. Some excess policies express the threshold as a dollar figure. ("This policy covers losses in excess of $5 million.") Others express the threshold by referring to the main policy. ("This policy covers losses in excess of the coverage provided by Policy No. 3141592.") Either way, the excess policy protects the insured against catastrophes, while the main policy protects him against routine losses. See McGurl v. Trucking Empls. of N. Jersey Welfare Fund, 124 F.3d 471, 478-79 (3d Cir.1997); 15 Steven Plitt et al., Couch on Insurance § 219:33 (3d ed.2013).

Guarantee Trust's theory starts off on the wrong foot because its policy does not provide excess insurance, at least not pure excess insurance. An excess policy has a fixed threshold below which it never applies. If the insured has no primary policy to cover losses below the threshold, the excess policy does not pick up the slack. It covers a layer of losses above the threshold, nothing else. Guarantee Trust's policy by contrast has no fixed level — neither a dollar amount nor an amount set by reference to another policy — above which it kicks in and below which it recedes. If the insured has no other policy, Guarantee Trust's policy covers all of his losses, however small. That shows that Guarantee Trust has provided ordinary coverage subject to a blanket coordination-of-benefits clause, not pure excess coverage. See McGurl, 124 F.3d at 479; 15 Plitt et al., supra, § 219:33.

The more fundamental mistake, however, lies in the assumption that excess policies, pure or otherwise, hold a privileged position in coordination-of-benefits cases. We can think of no good reason to create an excess-insurance exception, and a handful of good reasons not to. For starters, Thorn Apple Valley reflects the reality that ERISA's byzantine system of employee benefits would not work unless courts respect the written terms of ERISA plans. "The statutory scheme ... is built around reliance on the face of written plan documents." US Airways, Inc. v. McCutchen, ___ U.S. ___, 133 S.Ct. 1537, 1548, 185 L.Ed.2d 654 (2013) (internal quotation marks omitted). The importance of enforcing the plan's terms, its coordination clauses included, does not shrink when the other insurance policy in the picture provides excess coverage in this way.

More than that, "[t]he next worst thing to having no insurance at all is having two insurance companies cover the same claim." PM Group Life Ins. Co. v. W. Growers Assur. Trust, 953 F.2d 543, 544 (9th Cir.1992). Coverage overlaps often prompt years of fighting about who must pay, a battle that can delay payment to the insured or the hospital. We should try to keep rules about coordinating insurance benefits as simple as possible, an objective slighted by adding a proliferation of exceptions to Thorn Apple Valley's straightforward holding.

An example reveals the intuition behind Guarantee Trust's argument. Suppose that a policy covers all losses in excess of $5 million. And suppose the insured suffers a $1 million loss, meaning the policy gives him nothing ($1 million being less than $5 million). The ERISA plan's coordination clause surely cannot force the policy to cover the loss anyway. Doesn't this result prove that excess policies follow a special rule? No, because the result reflects a broader principle rather than an exception. Thorn Apple Valley empowers ERISA plans to coordinate preexisting benefits, not to invent new benefits to foist upon other insurers. This principle prevents rewriting an excess policy to cover small losses, just as it prevents rewriting a fire policy to cover flood damage, a marine policy to cover car accidents, or a 2013 policy to cover 2014 events.

A variation on this example confirms that excess insurance works just like any other kind of insurance. Suppose again that the policy covers all losses in excess of $5 million. But this time the insured suffers a $7 million loss, so the policy gives him $2 million ($7 million minus $5 million). If an ERISA plan's coverage overlaps with this $2 million entitlement, why shouldn't the plan's terms decide how to coordinate the two?

No good in short comes of categorizing insurance as pure excess, impure excess, or something else. An ERISA plan may coordinate benefits with another policy (whatever its taxonomy), though it may not redefine the coverage of another policy (whatever its taxonomy). Here, in the absence of Central States' plan, Guarantee Trust's policy would cover the sports injuries at hand without question. The ERISA plan insists that the policy keep doing in that plan's presence what it would do in that plan's absence. That amounts to coordinating benefits, not redefining coverage. The district court thus got it right when it held that primary responsibility for the sports injuries in this case falls on Guarantee Trust, not Central States.

III.

Now that Central States has paid for expenses that Guarantee Trust should have covered, what relief may it get from Guarantee Trust? In addition to granting a declaratory judgment, the district court ordered Guarantee Trust to pay Central States a little more than $112,000 as repayment for the medical expenses Central States had covered in Guarantee Trust's stead. On appeal, Guarantee Trust does not challenge the declaratory relief, but it insists that ERISA does not authorize this money judgment.

Before addressing the merits of that ruling, we must take care of a procedural skirmish. Central States argues, and the district court agreed, that Guarantee Trust forfeited its objection to the monetary award. We disagree.

When Central States first asked the district court to enter a money judgment, Guarantee Trust spoke up at once. It filed a brief claiming that ERISA did not empower the court to grant the requested relief, a conclusion it backed up with two pages of legal argument. See R. 39. That suffices to preserve Guarantee Trust's objection.

According to Central States, Guarantee Trust should have objected even earlier, when responding to the complaint. This approach overlooks what the Civil Rules have to say about preserving defenses. Under the Rules, a defendant loses affirmative defenses (like contributory negligence and estoppel) and some procedural defenses (like lack of personal jurisdiction and improper venue) unless he raises them at the pleading stage. See Fed.R.Civ.P. 8(c)(1), 12(h)(1). But other defenses, most importantly failure to state a claim to relief, remain stout until the end of the trial despite the defendant's pleading-stage silence. See Fed.R.Civ.P. 12(h)(2). Guarantee Trust objects that ERISA does not empower the court to grant a money judgment in this case — in other words, that Central States failed to state a claim to a money judgment. It would have been better practice for Guarantee Trust to make this argument right after it got the complaint (which made it clear that Central States wanted a monetary award), but under the Civil Rules the company's failure to do so does not amount to a forfeiture.

Central States persists that Guarantee Trust should have objected when responding to Central States' motion for partial summary judgment. The argument overlooks that the motion asked only for "partial summary judgment on the issue of liability." R. 20 at 1 (emphasis added). The motion did not ask for judgment on the issue of monetary relief, and Guarantee Trust had no obligation to object to a request that Central States had not yet made. Guarantee Trust thus did not forfeit its objection at the summary judgment stage either.

As for the merits, Central States filed this lawsuit under a provision of ERISA that allows it "to obtain ... appropriate equitable relief ... to enforce any provisions of this [Act] or the terms of the plan." 29 U.S.C. § 1132(a)(3)(B). The judgment directing Guarantee Trust to pay Central States can thus stand only if it provides "equitable relief" as opposed to legal relief. It does not.

The district court's money judgment, ordering Guarantee Trust to pay Central States around $112,000, looks by all appearances like an award of money damages. And money damages, the paradigm of legal relief, lie beyond the radius of § 1132(a)(3)(B). See Great-West Life & Annuity Ins. Co. v. Knudson, 534 U.S. 204, 210, 122 S.Ct. 708, 151 L.Ed.2d 635 (2002).

Central States tries to escape these principles by placing a different label — "restitution" rather than "damages" — on the money award. In doing so, it must come to terms with the Supreme Court's observation that, when a court "compel[s] the defendant to pay a sum of money to the plaintiff," the remedy "[a]lmost invariably" qualifies as "money damages" (and thus as legal relief). Id. We need not, however, decide the case on these grounds. Even assuming the money judgment qualifies as "restitution," it does not qualify as "equitable relief."

The United States Reports tell us that § 1132(a)(3)(B) authorizes "restitution in equity," but not "restitution at law." Id. at 213-14, 122 S.Ct. 708; Sereboff v. Mid Atlantic Med. Servs., 547 U.S. 356, 362-63, 126 S.Ct. 1869, 164 L.Ed.2d 612 (2006). A court awards equitable restitution when it imposes a constructive trust or lien on "particular funds or property in the defendant's possession" but legal restitution when it holds the defendant liable for a sum of money. Knudson, 534 U.S. at 214, 122 S.Ct. 708. A pair of Supreme Court cases, Sereboff and Knudson, illustrates the distinction. In each case an ERISA beneficiary suffered a car accident, received benefits from an ERISA plan, then won a tort case against a third party responsible for the accident. And in each case the terms of the ERISA plan required the beneficiary to reimburse the plan out of the tort recovery. In Sereboff, the ERISA plan tried to get a constructive trust on the tort lawsuit's proceeds, which had been "set aside and preserved" in the beneficiary's accounts. 547 U.S. at 363, 126 S.Ct. 1869 (internal quotation marks omitted). The plan asked for equitable restitution, because "it sought its recovery through a constructive trust ... on a specifically identified fund, not from the [beneficiary's] assets generally." Id. In Knudson, the ERISA plan tried to get a money judgment from the beneficiary. The plan asked for legal restitution, because it claimed an entitlement to "some funds for benefits that [it had] conferred" rather than "particular funds" that were in the beneficiary's possession. 534 U.S. at 214, 122 S.Ct. 708.

Central States did not ask for — and the district court did not impose — a constructive trust or lien on any identifiable fund. The court's judgment indeed has no connection to any particular fund at all. The court ordered Guarantee Trust to pay money, and Guarantee Trust can satisfy that obligation by dipping into any pot it chooses. That means Central States sought legal rather than equitable restitution.

Central States tries to portray its restitution as equitable, insisting that the requested funds "are specifically identifiable" because "[t]he funds are measured by the amount of [the] bills Central States paid." Central States Br. 40. But a money judgment does not become equitable merely because its size is known or otherwise identifiable in that way. It is the fund, not its size, that must be identifiable. Nor does the match between the size of the judgment and the size of the bills pull an identifiable fund into the picture. No matter how the district court figured out the size of the monetary recovery, the recovery continues to come out of Guarantee Trust's assets in general, not out of any fund in particular.

Unable to win the game on these terms, Central States tries to sweep the pieces off the chessboard. Even if the distinction between legal and equitable restitution works everywhere else, it maintains, the dichotomy "cannot control a [coordination-of-benefits] dispute" between an ERISA plan and an insurance company. Central States Br. 35. This request for a special dispensation, bereft of any textual or historical foundation, leans on an appeal to highly generalized statutory purposes and policies. A plan fighting with an insurance company, Central States explains, can rarely if ever find a specific fund (analogous to the tort recovery in Sereboff) that the company holds but that belongs to the plan. So applying the law-equity distinction in this context will (it persists) often leave plans without remedy, frustrating the enforcement of coordination-of-benefits clauses and thwarting the Act's objective of enforcing plans according to their terms.

This argument — the same argument — has come up before. In Knudson, the ERISA plan asked the Court to interpret "equitable relief" to include legal restitution, in order to "prevent [it] from being deprived of any remedy" and in order to vindicate "a primary purpose of ERISA,... the enforcement of the terms of a plan." 534 U.S. at 220, 122 S.Ct. 708 (internal quotation marks omitted). The Court's responses there work just as well here. For one, "there may have been other means for [the plan] to obtain the essentially legal relief [it] seek[s]," including perhaps a lawsuit under state law. Id. For another, appeals to ERISA's purpose cannot "overcome the words of its text." Id. That text, a "carefully crafted and detailed enforcement scheme," "only allows for equitable relief," a category that does not include legal restitution. Id. at 221, 122 S.Ct. 708.

Does our fidelity to the text somehow contradict our earlier application of Thorn Apple Valley, a federal common law rule developed in light of the Act's policies? No. The Act says nothing either way about coordinating benefits, and purposes and policies have their place when filling the gap. The Act by contrast says something about what relief a plaintiff can get in lawsuits under § 1132(a)(3)(B): He can get only equitable relief. No gap remains for purposes and policies to fill.

In so holding, we join other courts that have reached the same conclusion for the same reasons. See, e.g., Cent. States, Se. & Sw. Areas Health & Welfare Fund v. Health Special Risk, Inc., No. 13-10705, 756 F.3d 356, 2014 WL 2853587 (5th Cir. June 23, 2014). All in all, the district court erred by entering a money judgment for Central States.

IV.

Two loose ends dangle. Guarantee Trust points out that, under its insurance policy, "[a] legal action may not be brought to recover on this Policy" more than three years after the deadline for filing the written proof of loss, which in turn occurs 60 or 90 days after the loss in question happens. R. 31-2 at 14. Guarantee Trust believes that this provision shuts out Central States' claims involving six of the thirteen students. The district court rejected the argument, but we need not take a stand on it. Guarantee Trust has not argued that this limitations period affects the district court's declaratory judgment, a prospective remedy that concerns liability for future losses rather than recovery for past losses. So the limitations clause at most affects the money judgment, a part of the district court's decision that we have just reversed anyway.

Guarantee Trust also appeals the district court's decision to award attorneys' fees to Central States. Under the Act, the district court "in its discretion may allow a reasonable attorney's fee ... to either party." 29 U.S.C. § 1132(g)(1). The statute gives district courts more leeway to shift fees than the American Rule, the common-law principle that allows fee awards only in rare cases. See Foltice v. Guardsman Prods., Inc., 98 F.3d 933, 939 (6th Cir. 1996). Our cases have identified a range of factors that district courts should consider when exercising their authority under the statute. See Sec'y of Dep't of Labor v. King, 775 F.2d 666, 669-70 (6th Cir.1985). Here, the district court addressed all of the relevant factors with care. R. 67 at 17-21. Its conclusion does not amount to an abuse of discretion.

For these reasons, we affirm in part and reverse in part.