McKEAGUE, Circuit Judge.

ARGUED: Keith L. Pryatel, KASTNER WESTMAN & WILKINS, LLC, Akron, Ohio, for Appellant/Cross-Appellee. Michael D. Grabhorn, GRABHORN LAW OFFICE, PLLC, Louisville, Kentucky, for Appellees/Cross-Appellants. ON BRIEF: Keith L. Pryatel, Kenneth M. Haneline, KASTNER WESTMAN & WILKINS, LLC, Akron, Ohio, Lira A. Johnson, Michael P. Abate, DINSMORE & SHOHL LLP, Louisville, Kentucky, for Appellant/Cross-Appellee. Michael D. Grabhorn, Andrew M. Grabhorn, GRABHORN LAW OFFICE, PLLC, Louisville, Kentucky, William T. Payne, Joel R. Hurt, FEINSTEIN DOYLE PAYNE & KRAVEC, Pittsburgh, Pennsylvania, for Appellees/Cross-Appellants.

Before: SILER, McKEAGUE, and WHITE, Circuit Judges.

McKEAGUE, J., delivered the opinion of the court in which SILER, J., joined, and WHITE, J., joined in part. WHITE, J. (pp. 282–85), delivered a separate opinion concurring in part and dissenting from Part III and Part IV. C. 3.

OPINION

McKEAGUE, Circuit Judge.

This appeal is the latest installment in an ERISA litigation saga that has spanned almost ten years. At the risk of oversimplifying their case, the Plaintiff–Retirees claim that Defendant Norton Healthcare, Inc. Retirement Plan ("Norton") underpaid them under the terms of the plan. The district court found that the plan was unambiguous in the Retirees' favor. We agree with the district court on most issues. However, because the Plan is ambiguous in one crucial respect and may not comply with ERISA in another, we VACATE the district court's summary judgment order and REMAND for further examination of those issues. Consequently, we also VACATE the district court's damages order, including its certification of a class under Rule 23(b)(1)(A) and (b)(2) during the damages stage. In doing so, we mean no slight to the district judges and their staff, who ought to be praised for their commitment to this case and for their patience with the complex issues it presents.

I

Plaintiffs Elizabeth A. Clemons, David R. Khaliel, and Larry W. Taylor ("the Retirees") were employed by Norton Healthcare. They are covered by a plan ("the Plan") established to benefit former Norton employees. In January 2008, the Retirees brought a putative class action under ERISA, alleging Norton underpaid retirees who elected to take their pension as a lump-sum payment. The district court certified a class in 2011 and eventually granted summary judgment to the Retirees. Damages have not been reduced to a sum certain, but the district court adopted the Retirees' formula for calculating damages, awarded pre-judgment interest at a fixed rate, and entered a final judgment for the Retirees. Collateral proceedings indicate that the total amount at issue is between sixty and seventy million dollars.

Both sides appealed the judgment. Norton (technically, the appellant) disputes the district court's interpretation of the Plan, the standard of review employed by the district court, class certification, and the district court's adoption of the Retirees' proffered damages formula. The Retirees (technically, the cross-appellants), ask us to apply a longer statute of limitations and quibble with the appropriate rate of prejudgment interest, in addition to disagreeing with Norton's arguments on appeal. After the Retirees filed their cross-appeal, Norton balked, asserting that we lack jurisdiction over the entire case because the district court's last order was not final under 28 U.S.C. § 1291.

A

The facts relevant to this appeal are straightforward and undisputed. In 1991, Norton merged two predecessor plans (the "MEH Plan" and the "NKC Plan") into one Plan, which is the defendant here. The Plan is a single-employer plan governed by ERISA, funded by Norton, and maintained under a written document (the "Plan Document"). Although the initial Plan was established in 1991, the most recent, restated Plan Document became effective January 1, 1997. Since then, it has been amended multiple times.

As of January 1, 1997, the Plan used two basic formulas for pension benefits. First, it included a traditional defined-benefit formula applicable only to members of the predecessor plans from MEH and NKC. It also included a cash-balance formula applicable to all other plans. See Norton Plan § 4.03. In 2004, the Plan was amended to tie off the predecessor plans; that is, to end accruals under the defined-benefit formulas and allow further accruals only under the cash-balance benefit formula established in the merged Plan.

In all its iterations, the Plan provides for benefits for "normal" retirement at age 65, late retirement, early retirement, and disability retirement §§ 4.03–4.07. The Plan allows participants who are at least 55 years old and have at least 10 years of service to retire early. §§ 2.22, 4.05(a). The Plan allows retirees to take their benefit in the "Basic Form" or in one of six alternative forms. § 4.02(a)–(b). One of those alternative forms is a lump-sum payment received on the date of retirement. § 4.02(b)(6).

B

The Retirees here are all former Norton employees who (a) retired after the 2004 amendments became effective and (b) elected the lump-sum benefit. On January 30, 2008, they filed their first Class Action Complaint. The case was assigned to Judge Jennifer B. Coffman.

After two years of motion practice and class-certification discovery, the Retirees filed the operative complaint on May 10, 2010. Explaining the core of their allegations, the Retirees stated:

On behalf of themselves and others similarly situated, Plaintiffs challenge the calculation of their pension benefits as follows:

a. the Defendant Plan's failure to include the value of the "increasing monthly income" ("cost-of-living") in the calculation of participant lump sum benefits and in the calculation of participant "cash balance" starting balances;

b. the Defendant Plan's failure to include the value of early retirement subsidies in the calculation of participant lump sum benefits and in the calculation of participant "cash balance" starting balances; and/or

c. the Defendant Plan's failure to calculate participant lump sum benefits according to the contra[c]tual formula.

1

On May 17, 2010, the Retirees filed a motion for class certification. They sought certification under Rule 23(b)(1)(A), (b)(2), and (b)(3), defining their proposed class as:

All participants in the Norton Healthcare, Inc. Retirement Plan, its predecessors and successors, whose contractual lump sum pension benefits:

(1) did not include the value of the basic form of benefit—an "increasing monthly retirement income" (annual cost-of-living adjustment)—when election of such basic form would have yielded the highest value for the participant; and/or

(2) did not include the value of the "alternative" lump sum benefit where the basic form of benefit is multiplied by 212, when election of such alternative form would have yielded the highest value for the participant; and/or

(3) did not include the value of the early retirement subsidy.

The district court granted the motion and certified the class under Rule 23(b)(1)(A) and (b)(2), finding Norton's counterarguments to be improperly merits-based. It also appointed Khaliel and Taylor as class representatives, but rejected Clemons, reasoning that the Retirees had not satisfied the typicality requirement as to her. The district court defined the class as requested by the Retirees. In April 2011, the parties began merits discovery. Meanwhile, Norton petitioned for interlocutory review of the certification order. We denied the petition.

Back in the district court, Norton moved for partial summary judgment. The basis for the motion was that Kentucky's five-year statute of limitations for statutory claims applied, and that all claims related to benefits paid before January 30, 2003 should be dismissed. See Ky. Rev. Stat. Ann. § 413.120(2). The Retirees countered that their claims were timely because Kentucky's fifteen-year statute of limitations for claims based on a written contract should apply. See Ky. Rev. Stat. Ann. § 413.090(2). The district court denied the motion. The court reasoned that the dispute "centers around the plaintiffs' rights under their contract with [Norton]," and that the Retirees were not seeking to enforce any statutory rights created by ERISA.

2

After we denied interlocutory appeal of the class certification decision, the Supreme Court decided Wal-Mart Stores, Inc. v. Dukes , 564 U.S. 338, 131 S.Ct. 2541, 180 L.Ed.2d 374 (2011). On July 24, 2012, Norton attacked the existing class certification, citing Dukes and West v. AK Steel Corp. , 484 F.3d 395 (6th Cir. 2007). Norton asked the district court to decertify the class entirely because the highly individualized nature of pension calculations forestalled a finding of commonality. Norton also attacked the merits of Khaliel's and Taylor's individual claims and argued that certification under Rule 23(b)(2) was improper in light of Dukes and West .

The district court denied the motion. The court found Dukes inapplicable, reasoning that the commonality requirement was satisfied because Norton "stipulated that its actuaries used a consistent methodology when they calculated benefits," and "the crux of the Plaintiffs' claims" was whether that methodology was correct. The district court also concluded that even if certification under Rule 23(b)(2) was improper, the class "would remain properly certified under Rule 23(b)(1)(A)."

While Norton's motion to decertify was pending, the parties completed merits discovery and filed cross-motions for summary judgment. Shortly after briefing on these motions was completed, Judge Coffman retired, and the case was reassigned to Judge Thomas B. Russell.

On August 23, 2013, Norton filed a motion asking Judge Russell to reconsider Judge Coffman's ruling on the statute-of-limitations issue. The district court granted the motion, concluding that Fallin v. Commonwealth Indus., Inc. , 695 F.3d 512 (6th Cir. 2012), and Redmon v. Sud–Chemie Inc. Ret. Plan for Union Emps. , 547 F.3d 531 (6th Cir. 2008), compelled the conclusion that Kentucky's five-year limitations period for statutory claims applied.

C

On October 31, 2013, the district court granted summary judgment to the Retirees. Only the issues germane to this appeal are discussed below.

First , the district court held that, because the Plan gives its administrator discretion to construe the terms of the plan, arbitrary-and-capricious review applies. Firestone Tire & Rubber Co. v. Bruch , 489 U.S. 101, 115, 109 S.Ct. 948, 103 L.Ed.2d 80 (1989). After announcing that conclusion, however, the court engaged in a lengthy discussion of whether the doctrine of contra proferentum should also apply. Relying primarily on dicta in University Hospitals of Cleveland v. Emerson Electric Co. , 202 F.3d 839 (6th Cir. 2000), the court concluded that there is "no significant reason why this basic and equitable contract principle should not apply in the context of ERISA contracts."

Second , the district court concluded that the Plan documents unambiguously favored the Retirees' position on most issues. This included a finding that the 2004 amendments created an "early retirement subsidy" and that the Plan called for any lump sum to account for an increasing monthly income for sixty months certain. The court supplemented its discussion by stating, "However ... to the extent that any of the relevant provisions are ambiguous, they would be construed against the drafter and in favor of plaintiffs." However, the district court did agree with Norton on a structural issue, applying the last paragraph of a disputed section globally, rather than only to the subparagraph immediately preceding it.

Third , the district court agreed with the Retirees that, regardless of the various formulas set out in the Plan, "ERISA require[s] that a participant's lump sum benefit be the ‘actuarial equivalent’ of the [Monthly Retirement Income]." 29 U.S.C. § 1054(c)(3).

Fourth , the district court declined to attempt any actual damages calculations, instead leaving them "for the parties to perform, consistent with the Court's instructions." To that end, the court ordered Norton to recalculate the class members' pensions within 45 days, and the Retirees to raise any objections within 14 days thereafter.

D

Translating the district court's rulings into a pension formula proved more difficult and contentious than expected. With the Retirees' agreement, the deadline was extended by 45 days. However, Norton continued to struggle to assemble the necessary records and cull the requisite data. The parties therefore jointly requested an open-ended extension, based on Norton's promise that once its spreadsheet detailing all the data and calculations was completed, Norton would share the spreadsheet with the Retirees and make the underlying hard-copy documents available to the Retirees. The parties also promised to discuss Norton's new formula and attempt to resolve any disagreements. The district court granted the open-ended extension request on January 24, 2014.

This cooperation broke down about a year later. Subsequently, the parties offered competing damages formulas, both supported by expert affidavits. The Retirees also requested pre-judgment interest at a rate of eight percent. Norton argued that the Retirees were not entitled to any pre-judgment interest. Norton also used its damages brief to reargue the certification question.

The district court issued its damages opinion on January 6, 2016 (more than two years after its liability ruling). The court's order addresses a variety of issues, however, and was not limited to damages.

First , the district court ordered that the Retirees' formula be used to recalculate class members' pensions, explaining:

After extensively reviewing the parties' formulas and their arguments, this Court finds that Norton did not provide a formula that will ensure that the recalculated lump sums are at least actuarially equivalent to the Monthly Retirement Income. Norton goes to great lengths to critique the Class's proposed formula, but it does not adequately explain to this Court how its own formula results in an actuarially equivalent lump sum. Norton makes several conclusory statements that its formula results in an actuarial equivalent lump sum, but it does little to explain to the Court how this is so. Alternatively, the Class has gone to great lengths to not only create a formula and demonstrate how it is actuarially equivalent but also to respond to any of Norton's criticisms concerning its formula.

Second , the district court rejected Norton's request to decertify the class entirely. The district court also rejected Norton's request to interpret its liability holding in a way that would narrow the scope of the class.

Third , the district court concluded that the Retirees are entitled to pre-judgment interest. However, the court found that eight percent "would be excessive." Employing the post-judgment interest statute and seeking to avoid "the complexities of compounding interest," the district court instead awarded pre-judgment interest at a flat two percent rate. 28 U.S.C. § 1961.

Finally, the district court concluded its damages opinion with the following:

To conclude that this matter is complex is probably an understatement. ... The complexity of comprehending and analyzing lengthy interrelated documents and deciphering the nuisances therein has been time consuming on the part of counsel and the Court. The Court is not inclined to entertain any motion for reconsideration. Frankly, the Court has given this matter its best shot. Too much time has elapsed between the earlier ruling and the current posture of this case. The Court originally thought agreeing on a formula would not be a time consuming process. Obviously, it was time consuming. It is time for another court, if the parties are so inclined, to look at this matter with new eyes.

The district court then entered judgment for the Retirees "in an amount consistent with" its opinion, and stated that its judgment was "final and appealable."

E

Norton filed a notice of appeal on January 20, 2016. The Retirees filed a notice of cross-appeal on February 2, 2016. On May 27, 2016, Norton filed a motion to dismiss the cross-appeal for lack of jurisdiction, arguing the district court has not entered a final order for purposes of 28 U.S.C. § 1291.

II

We have jurisdiction over this appeal. Norton's motion to dismiss is DENIED .

The court of appeals has jurisdiction over all "final decisions of the district courts." 28 U.S.C. § 1291. A district court's statement that its judgment is final and appealable does not make it so. Liberty Mut. Ins. Co. v. Wetzel , 424 U.S. 737, 742, 96 S.Ct. 1202, 47 L.Ed.2d 435 (1976) ; Day v. NLO, Inc. , 3 F.3d 153, 155 (6th Cir. 1993). Rather, a "final decision is ‘one which ends the litigation on the merits and leaves nothing for the [district] court to do but execute the judgment.’ " Gelboim v. Bank of Am. Corp. , ––– U.S. ––––, 135 S.Ct. 897, 902, 190 L.Ed.2d 789 (2015) (quoting Catlin v. United States , 324 U.S. 229, 233, 65 S.Ct. 631, 89 L.Ed. 911 (1945) ). Generally, "[a] judgment is not final for purposes of appeal when the assessment of damages remains," unless the calculation is purely ministerial. Woosley v. Avco Corp. , 944 F.2d 313, 316–17 (6th Cir. 1991). Norton contends that the district court's order was not final because it has not reduced damages to a sum certain and because "the complex actuarial calculations required to compute the Plaintiffs' damages" under the formula adopted by the district court "are far more than ‘ministerial.’ "

We think Woosley controls the outcome here. In Woosley , we held that a judgment ordering the employer to calculate backpay was final, even though the court had yet to calculate the exact amount owed. 944 F.2d at 317. As the variables necessary to calculate backpay were not known to the court, the judgment ordered the parties to supply stipulations about those variables, which would automatically be incorporated into the judgment. Id. at 316. The employer appealed before the stipulations were filed, and the employees subsequently challenged the jurisdiction of this Court on finality grounds, arguing that there were still complex calculations left to be done about which the parties were arguing. Id. at 316–17. We denied the motion to dismiss, reasoning that although the calculations required more than simple addition, they did not affect the finality of the judgment. See id. In doing so, we pointed out that "the equities between the parties were resolved," even though someone still had to find a more advanced calculator before writing the check. Id. at 317.

The same is true here. The district court answered all the parties' merits questions, and found that Norton had misapplied the terms of the Plan. The district court then told the parties exactly how to recalculate those Retirees' benefits—by using the Retirees' proposed formula. Thus, there are no unanswered legal or equitable questions in this case. Norton also does not contend it lacks the data necessary to perform the calculations, or that the parties dispute the accuracy or authenticity of the relevant records, so there are no unanswered factual questions either. It is undoubtedly true that the calculations required here are complex and time-consuming. But the point made by the Woosley panel was that complexity is not a vice in this context. See id. The need for an advanced understanding of applied mathematics to obey an order of the court does not make that judgment any less final for our purposes. The motion to dismiss is DENIED .

III

The first thing the parties dispute is the standard of review applicable to Norton's interpretation of the plan documents. Ordinarily, a plan administrator's denial-of-benefits decision is reviewed de novo. Firestone Tire & Rubber Co. v. Bruch , 489 U.S. 101, 115, 109 S.Ct. 948, 103 L.Ed.2d 80 (1989). But if the plan "gives the administrator or fiduciary discretionary authority to determine eligibility for benefits or to construe the terms of the plan," we review such decisions under the arbitrary-and-capricious standard. Id. at 111, 115, 109 S.Ct. 948. Further, if the plan gives the administrator discretion to make factual determinations, arbitrary-and-capricious review extends to those issues as well. Shaw v. AT&T Umbrella Benefit Plan No. 1 , 795 F.3d 538, 547 (6th Cir. 2015). However, "if a benefit plan gives discretion to an administrator or fiduciary who is operating under a conflict of interest, that conflict must be weighed as a factor in determining whether there is an abuse of discretion." Firestone , 489 U.S. at 115, 109 S.Ct. 948 (citation, quotation marks, and brackets omitted).

Under the Plan, the Retirement Committee serves as the Plan administrator, and is granted "the power and discretion ... to construe all terms, provisions, conditions and limitations of the Plan," and "to determine all questions arising out of or in connection with the provisions of the Plan or its administration in any and all cases...." Norton Plan § 6.06(b)(2)–(3). This language clearly invokes Firestone deference.

However, we have also suggested in dicta that the common-law doctrine of contra proferentum might apply to ERISA cases. The doctrine is traditionally used to construe ambiguous terms against the drafter of a contract. Our precedent has not been a model of clarity on the issue—in some cases, we have implied that it sweeps broadly; in others, we have expressed confusion about how contra proferentum and Firestone deference can apply at the same time. Compare Univ. Hosps. of Cleveland v. Emerson Elec. Co. , 202 F.3d 839 (6th Cir. 2000), with Smiljanich v. Gen. Motors Corp. , 182 Fed.Appx. 480, 486 n.2 (6th Cir. 2006).

The district court recognized this confusion and did its best to obey these cases insofar as it understood them. Although the court ultimately disclaimed any reliance on the rule to decide the case, the doctrine reappears at least five times throughout the court's substantive analysis. Our review of the plan documents and the parties' disputes leads us to conclude that the Plan is ambiguous on an issue of significant importance to the parties—whether the early retirement reducers apply to the Retirees bringing this lawsuit. We think, therefore, that the law's treatment of this ambiguity is necessary to the outcome here, and compels us to examine whether contra proferentum can be used in conjunction with Firestone deference. We conclude that it cannot.

A

It appears that dicta on this issue first appeared in our cases in 1993. See Tolley v. Commercial Life Ins. Co. , 14 F.3d 602 (table) (6th Cir. 1993) (per curiam). That panel declined to reach the question because the insurance policy in dispute was not ambiguous. Id. ; see also Schachner v. Blue Cross & Blue Shield of Ohio , 77 F.3d 889, 895 n.6 (6th Cir. 1996) (same); Swisher-Sherman v. Provident Life & Accident Ins. Co. , 37 F.3d 1500 (table) (6th Cir. 1994) (per curiam) (same).

In Perez v. Aetna Life Insurance Co. , however, the en banc court issued stronger dicta on the subject. In that case, we stated, without qualification: "The rule of contra proferentum provides that ambiguous contract provisions in ERISA-governed insurance contracts should be construed against the drafting party." 150 F.3d 550, 557 n.7 (6th Cir. 1998) (en banc). The dispute in Perez was whether a particular plan provision gave the administrator discretion to decide sufficiency-of-the-evidence questions. Id. at 556–57. However, the court found that the plan unambiguously did provide for such discretion, and thus did not need to apply contra proferentum . See id. at 557 n.7.

In University Hospitals , a panel went further, but still did not move from dicta to a holding. University Hosps. of Cleveland v. Emerson Elec. Co. , 202 F.3d 839 (6th Cir. 2000). In that case, we stated that "to the extent that the [p]lan's language is susceptible of more than one interpretation, we will apply the ‘rule of contra proferentum ’ and construe any ambiguities against Defendants/Appellees as the drafting parties." Id. at 846–47. However, we ultimately concluded that the defendant's interpretation of the plan was implausible. Id. at 850. Since the panel concluded that the plan was unambiguous in the plaintiff's favor, its statements on contra proferentum were dicta.

We have since addressed University Hospitals and the doctrine of contra proferentum at least seven times. Three times, we found no ambiguity and thus did not address any potential conflict between contra proferentum and Firestone . See Osborne v. Hartford Life & Accident Ins. Co. , 465 F.3d 296, 300 (6th Cir. 2006) ; Marquette Gen. Hosp. v. Goodman Forest Indus. , 315 F.3d 629, 632 (6th Cir. 2003) ; Ziegler v. HRB Mgmt., Inc. , 182 Fed.Appx. 405, 408 (6th Cir. 2006). Three other times, we have criticized the doctrine on various grounds. Mitzel v. Anthem Life Ins. Co. , 351 Fed.Appx. 74, 82 (6th Cir. 2009) ("Limiting the application of the con tra proferentem rule to cases in which an administrator's decision is reviewed de novo strikes us as the only sensible approach...."); Smiljanich v. Gen. Motors Corp. , 182 Fed.Appx. 480, 486 n.2 (6th Cir. 2006) (concluding that applying contra proferentum would be improper where Firestone deference applies); Mitchell v. Dialysis Clinic, Inc. , 18 Fed.Appx. 349, 353–54 (6th Cir. 2001) (characterizing the statements in University Hospitals , Perez , and Schachner as non-binding dicta).

The only contrary case appears to be Copeland Oaks v. Haupt , 209 F.3d 811 (6th Cir. 2000). The Copeland Oaks court never used the term contra proferentum . But in holding the plan administrator had abused her discretion, the court stated that Firestone deference should not be understood to allow a plan to "avoid a default rule of insurance law applicable in the ERISA context merely by giving itself discretion to interpret the plan." Id. at 813. The court went on to state that "even an arbitrary and capricious standard of review can be tempered by considering conflicts of interest such as those implicit in any self-funded plan, and by construing ambiguities against a plan drafter." Id.

This last statement went beyond the holding in Copeland Oaks . That case dealt solely with the issue of whether Firestone discretion allowed the plan to avoid a default rule of federal insurance law (the "make-whole" rule). Id. We concluded that the plan could not avoid default obligations that we have imposed on all ERISA plans simply by interpreting the plan to its advantage. Id. Instead, we required a clear statement before we would conclude that the parties had chosen to abandon the make-whole rule. Id. at 813–14. This result is not obtained by "tempering" arbitrary and capricious review, but instead is mandated by a cardinal principle of administrative law: that ambiguity does not give administrators license to ignore fundamental policy judgments made by the governments that supervise them. Cf. Util. Air Regulatory Grp. v. Envt'l Protection Agency , ––– U.S. ––––, 134 S.Ct. 2427, 2444, 189 L.Ed.2d 372 (2014) ; Food & Drug Admin. v. Brown & Williamson Tobacco Corp. , 529 U.S. 120, 125–26, 132–33, 120 S.Ct. 1291, 146 L.Ed.2d 121 (2000).

Contra proferentum has nothing to do with this logic. It applies as a sort of equitable estoppel between parties to a contract, not as a way for parties to escape the strictures of public policy. See, e.g. , Merrimack Valley Nat'l Bank v. Baird , 372 Mass. 721, 363 N.E.2d 688, 690–91 (1977) ("The author of the ambiguous term is held to any reasonable interpretation attributed to that term which is relied on by the other party."). Thus, suggestions in a case dealing with the latter issue are not necessarily helpful in the former cases.

B

Faced solely with a mountain of dicta, we move from law to reason. And we think that contra proferentum is inherently incompatible with Firestone deference. Thus, we hold that when Firestone applies, a court may not invoke contra proferentum to "temper" arbitrary-and-capricious review. However, when it is not clear whether the administrator has, in fact, been given Firestone deference on a particular issue, we think the doctrine still has legitimate force.

1

In Firestone , the Court rejected the Plan's position that arbitrary-and-capricious review should be the norm in denial-of-benefits cases. Before ERISA, a denial-of-benefits lawsuit was governed by ordinary contract law, unless the plan itself "g[a]ve the employer or administrator discretionary or final authority to construe uncertain terms." Firestone , 489 U.S. at 112–13, 109 S.Ct. 948. When it enacted ERISA, Congress relied heavily on concepts of trust law, including the fundamental rule that the courts would construe trust terms "without deferring to either party's interpretation." Id. at 112, 109 S.Ct. 948. The Court was therefore skeptical that Congress meant to give employees less protection under ERISA than they possessed under pre-ERISA contract law.

But even though de novo review was the default rule, the Court recognized a significant exception. It observed that "when trustees are in existence, and capable of acting, a court of equity will not interfere to control them in the exercise of a discretion vested in them by the instrument under which they act." Id. at 111, 109 S.Ct. 948 (quoting Nichols v. Eaton , 91 U.S. 716, 724–25, 23 L.Ed. 254 (1875) ) (emphasis supplied by the Firestone Court). In other words, parties to a trust may, by contract, remove certain decisions from the de novo supervision of an equity court. Id. A trustee therefore cannot violate the terms of the trust he or she was vested with authority over, but could violate the vesting clause itself through an abuse of discretion. Id. at 114–15, 109 S.Ct. 948.

The contra proferentum doctrine is, at its core, a rule of equity. See RESTATEMENT (SECOND) OF CONTRACTS § 206 cmt. a (noting that the principle's "operation depends on the positions of the parties as they appear in litigation, and sometimes the result is hard to distinguish from a denial of effect to an unconscionable clause"). It compels a drafting party to be honest about its offer up front, by threatening to construe terms "against the offeror" if he attempts to hoodwink the other party. Id. ; Contra Proferentem , BLACK'S LAW DICTIONARY (10th ed. 2014). The rule's equitable nature and its stated goals suggest that it does not make a good roommate for Firestone , but that it might make a good neighbor.

2

As a practical matter, we do not think a court can apply Firestone deference and contra proferentum to the same case without contradiction.

First, the Firestone Court labeled the discretion-vesting clauses as a turning point in the character of an ERISA plan. By conferring discretion on the administrator of a plan, ordinary rules of equity no longer operate to control the administrator–trustee absent an abuse of discretion that violates the vesting clause itself. See Firestone , 489 U.S. at 111, 109 S.Ct. 948. It follows that contra proferentum , as an equitable rule, should not be injected to micromanage this discretion or tip the scales in close cases. Whatever the precise contours of Firestone deference, it must include the ability to choose between two reasonable interpretations of the Plan, and that is precisely the situation in which the traditional contra proferentum rule operates against the drafter.Indeed, the very point of vesting discretion in someone is to trust his or her judgment when you are lost at sea or when you need to solve a Gordian knot. It makes little sense to revoke that discretion when it is needed most—in difficult cases where there is no clear answer. In effect, applying contra proferentum when language is ambiguous generates a paradox where the administrator can only exercise his discretion when it is not needed, i.e., when the plan language is clear. We are certain that this is not what the Supreme Court had in mind when it decided Firestone .

Second, we doubt that plan-construction and denial-of-benefits decisions will be influenced by a contra proferentum rule. Courts construe ambiguities against a drafter to remind the next drafter to state his terms clearly when he comes to the negotiating table. It is a prophylactic rule, not a remedial device. See RESTATEMENT (SECOND) OF CONTRACTS § 206 (noting that the primary aim of the rule is to discourage drafters from being "deliberately obscure"). But the kind of benefits and plan-construction decisions contemplated by ERISA are far removed from contract negotiations, where parties cannot reasonably be expected to foresee every possible negative consequence of the language they use. Indeed, it is this very uncertainty that often compels parties to vest discretion in the administrator, rather than risk de novo review under Firestone . The presence of an obvious vesting clause makes it hard to believe that beneficiaries will be duped by a "deliberately obscure" clause in the plan, since they have already agreed to trust the administrator's judgment in those obscure cases. And it is equally hard to believe that the administrator would deliberately write obscure benefits provisions, since that can easily develop into a headache later on (or perhaps a ten-year-long, class-action lawsuit). Thus, we do not see how contra proferentum has any worth as a prophylactic in these cases.

3

Neither do we think that contra proferentum can be "weighed" in the final analysis "in determining whether there is an abuse of discretion." Firestone , 489 U.S. at 115, 109 S.Ct. 948 (citation omitted) (quotation marks and brackets removed); Copeland Oaks , 209 F.3d at 813 (suggesting this approach). An administrator's conflict of interest is properly considered in this analysis, because fiduciaries are absolutely forbidden from acting with ulterior motives. Firestone , 489 U.S. at 115–16, 109 S.Ct. 948 ; RESTATEMENT (THIRD) OF TRUSTS § 78, cmt. b . Depending on the severity of the conflict, discretion exercised under these circumstances can easily be capricious, in the sense that the action is done without regard for the best interests of the beneficiary. See id.

Plan ambiguities do not work this way. If the administrator deliberately makes the Plan ambiguous so that it can invoke deference to serve its own interests, we might consider that fact under Firestone . But we would do so under the conflicts-of-interest rubric and the breach-of-trust doctrine, not because the plan was ambiguous. In other words, the equitable impulse to construe the Plan against the drafter comes from the administrator's malfeasance, not from the Plan's language. To the extent that we would "temper" arbitrary-and-capricious review by construing language against the administrator, we would do so only to take account for this kind of misbehavior.

4

This does not mean that we abandon contra proferentum entirely. The Firestone Court made it clear that modifying the presumption of de novo review is a contractual decision. 489 U.S. at 115, 109 S.Ct. 948. And unlike denial-of-benefits and plan-construction decisions, defer-or-not decisions are at the forefront of ERISA plan negotiations, because that is where Firestone puts them. Indeed, the parties' agreement on the Firestone procedural rules may often be more important than the substance of the plan itself. In the immortal words of Rep. John Dingell: "I'll let you write the substance ... you let me write the procedure. I'll screw you every time." Hearing on H.R. 2327, 98 Cong. 312 (1983).

Thus, we see the wisdom of applying contra proferentum to the threshold question of whether Firestone deference exists. The administrator has an obvious desire to operate under Firestone and an equally obvious motivation to obtain that result in a deliberately obscure manner. We do not think it was an accident that our first unequivocal statement on this subject occurred in an en banc case where the defer-or-not question was at issue. Perez v. Aetna Life Ins. Co. , 150 F.3d 550, 557 n.7 (6th Cir. 1998) (en banc). And if beneficiaries are to give the administrator the kind of trust inherent in Firestone deference, they ought to do so on purpose, not as the result of ambiguity. But since no one disputes that the Norton plan gives the administrator Firestone deference here, this issue can be resolved at a later time.

In sum, we hold that if the Plan clearly gives the administrator Firestone deference, then contra proferentum has no place in reviewing the administrator's decisions. The arbitrary-and-capricious standard stays intact.

IV

We now move to the core of our task: construing the Plan. Mindful that we are judges—not actuaries or the "fairness police"—we ask one question, and one question only: Is the Plan language clear? If it is, we must enforce it, no matter how unfair or bizarre it might seem to Norton or to the Retirees. Perez , 150 F.3d at 556–57. It is simply not our job to rescue parties from the unintended consequences of complexity, sloppy drafting, or bad negotiating. Id. at 557. The time to close loopholes is in the offices of transactional lawyers, not in the courts fourteen years later.

To the extent that the Plan is truly ambiguous, Firestone requires the district court to defer to Norton's reasonable interpretation. The Plan is ambiguous if it is susceptible to multiple reasonable interpretations, not just because clever lawyers can disagree over the meaning of terms. Id. at 557 n.7. The district court held that the Plan was unambiguous on all fronts, and for the most part, we agree. However, we think that the Plan gives ambiguous instructions on one of the core disputes between the parties—whether or not the early retirement reducers apply to these plaintiffs. We therefore must vacate the district court's holding on that issue and remand for a Firestone analysis of Norton's proposed interpretation.

Although many different variables and formulas are involved, calculating an individual's retirement benefit under the Plan involves three steps. First, identify the applicable rules. Second, use those rules to determine the individual's Monthly Retirement Income (MRI), which is, essentially, the individual's accumulated pension entitlement expressed as an amount per month. Third, convert the MRI to the form of benefit selected by the individual, one of several different annuities or a lump sum. Both parties raise a host of arguments about interpreting discrete sections of the Plan. However, no one in this litigation has ever explained how the whole plan works, i.e., how to calculate benefits from start to finish. This omission makes their more nuanced arguments almost impossible to follow. We think such a roadmap is the only logical starting point, and so we draw one for ourselves, addressing the parties' disputes as they arise.

A. Step 1: Identify the Appropriate MRI Rules

The Plan divides Retirees into four categories: Normal, Early, Late, and Disability. Norton Plan § 4.03–4.06. Each category has its own rules and restrictions for calculating MRI. Thus, before we can start analyzing the benefits question, we must first identify the governing rules.

The class members in this case all retired early, and are therefore governed by § 4.05. That section gives us two tracks from which we can choose: subsection (b), which applies "[e]ffective prior to January 1, 2004," and subsection (d), which is "[e]ffective only for a Member who terminates employment on or after January 1, 2004." § 405(b), (d) (as amended). The choice is obvious. Subsection (d) governs the class. That subsection states:

Effective only for a member who terminates employment on or after January 1, 2004, upon such Member's retirement on the Member's Early Retirement Date, the Member shall be entitled to a Monthly Retirement Income equal to the benefit described under Subsection 4.03(b), determined as of the Member's Early Retirement Date ... provided, in no case shall such Monthly Retirement Income be less than the Member's Accrued Benefit as of December 31, 2003.

§ 405(d). It is here that we find the first piece of tension dividing the parties. Generally, a person who retires early receives less money than someone who retires on time. See § 4.05(a), (b). However, the Retirees point out, correctly, that the cross-reference to "Subsection 4.03(b)" refers to the rules for a person retiring on their Normal Retirement Date . See § 4.03(b).

We see tension here, but no ambiguity. Subsection 4.05(d) uses terms of equivalence, not incorporation. In other words, § 4.05(d) does not import § 4.03(b) wholesale—it simply tethers one value to another. Thus, the reader is instructed to follow the rules in § 4.03(b), except that we are to use the person's early retirement date as if it was a normal retirement date. Tempering the rules in § 4.03(b) with early retirement calculations at the MRI stage, as Norton suggests, would violate the mandate that the early Retirees' payment be "equal to the benefit described under Subsection 4.03(b)." § 4.05(d) (as amended). Although this distinction seems trivial, it becomes important later on in the analysis. Committed to our step-by-step process, we table its implications until that time.

B. Step 2: Calculate MRI

Section 4.03(b) contains instructions for calculating MRI. The original plan did not contain a § 4.03(b); instead, it was added by the same amendments that created § 4.05(d). These amendments split "Normal" MRI calculations into two subsections: 4.03(a), which outlines the defined-benefit scheme applicable before 2004, and 4.03(b), the new cash-balance scheme created in 2004. Section 4.03(b) instructs the reader that this benefit "shall be determined" by a two-step process. The two-step process is as follows:

(1) First, the Member's Monthly Retirement Income determined under Paragraphs 4.03(a)(1), (2), and (3) as of December 31, 2003 ...

(2) Second, the Member's Monthly Retirement Income as determined under Paragraph 4.03(b)(1), increased monthly as follows:

(A) beginning January 1, 2004 and each January 1 thereafter with respect to the Monthly Retirement Income accrued for the prior Plan Year, by a factor which, when compounded on a simple basis for twelve (12) months, would produce the Index for the applicable Plan Year, and

(B) commencing January 1, 2004, by an increasing monthly income accrued at the rate of one two hundred twelfth of a percentage of the Member's Compensation for any Plan Year in which the Member completes one thousand (1000) Hours of Service based on the following schedule [omitted].

§ 4.03(b)(1)–(2).

Thus, § 4.03(b) tells us to add two numbers together to find the MRI. The base number is the amount accrued under the old defined-benefit system. § 4.03(b)(1). That number is then "increased" by the benefit accrued under the cash-balance system established in 2004. § 4.03(b)(2). There does not appear to be much dispute over how the (b)(2) number is calculated. Thus, we stick to resolving the dispute under (b)(1).

Subsection (b)(1) concerns the MRI accumulated under the defined-benefit scheme "as of December 31, 2003." § 4.03(b)(1). This number is static—it is determined at the close of 2003, and remains unchanged whether the employee retired on January 1, 2004, or January 1, 2014. Id. In other words, everyone covered by the defined-benefit plan "retired" en masse from that plan on December 31, 2003. Any post-2003 benefits are accounted for in the cash-balance calculation created by § 4.03(b)(2). This is important because the 2004 amendments are not retroactive beyond January 1, 2004. Therefore, other than the instruction that we are to use December 31, 2003 as the cutoff or "retirement" date for benefits, we must ignore the 2004 amendments until we resurface from the defined-benefit calculation.

Jump in the time machine. The old plan only had a defined-benefit system, governed by § 4.03(a). That section states:

When a Member lives to his Normal Retirement Date, he shall be entitled to retire and to receive a Monthly Retirement Income in an amount certified to the Trustee by the Retirement Committee. The amount of the Member's Monthly Retirement Income under the basic form described in Section 4.02(a) and payable at his Normal Retirement Date shall be equal to the largest of the amounts provided under the applicable provisions of paragraphs (1), (2) or (3) hereof....

Id. § 4.03(a). The parties agree that paragraph (3) is the winner, at least for Khaliel, who is the class representative. The relevant language states:

(3) For a Member with an NKC Accrued Benefit, the greater of (A) or (B):

(A) one and two-thirds percent (1-2/3%) of the Member's Average Monthly Earnings, such amount multiplied by the Member's years of

Credited Service, minus the lesser of [two actuarial calculations].

(B) The NKC accrued benefit.

Such monthly amount shall be converted to a single sum on the basis of the interest rate or rates ... for 30-year Treasury securities for the second month prior to the beginning of the applicable Plan Year and the mortality table prescribed by the Secretary of the Treasury and the resulting amount divided by 212.

§ 4.03(a)(3). Subparagraph (A) refers to a Traditionally Accrued (or "TRAC") benefit. Subparagraph (B) refers to the Accrued Benefit under the pre-merger plan. The parties seem to agree that Subparagraph (A) is the winner, again, for Khalil. But that is where the agreement ends.

The final paragraph quoted above is at the center of the parties' dispute. Under the calculations prescribed in that paragraph, a nonincreasing monthly benefit is converted into an increasing benefit—one that begins smaller, but increases over time to account for inflation. The Retirees, apparently, believe that the subparagraph (A) benefit is already an increasing benefit, while the subparagraph (B) benefit is not. Thus, they argue, it makes no sense to apply the translation formula to subparagraph (A) again. Norton takes the opposite position, arguing that both benefits are nonincreasing, and therefore that the translation is necessary for both.

We need not get tied up in deciding who has the better argument on how the subparagraphs operate in isolation. That is a question to be answered by the district court in the first instance after hearing expert testimony from the parties. More importantly, however, the minutiae of benefits calculations cannot control the answer given by the structure of § 4.03(a)(3). Such analysis puts the cart before the horse: our first question is not what makes sense in the context of the plan calculations; we ask, instead, what the plan says . Here, the final paragraph lacks an introductory number or letter. However, it is aligned at the left margin with the introductory language of § 4.03(a)(3) and with the subsection designations (A) and (B). We can think of no reason that a drafter would do this other than to make the final paragraph applicable to all of subsection 4.03(a)(3). Indeed, if the final paragraph was only supposed to apply to subparagraph (B), then we would expect it to either be appended to (B) without separation or be indented to the same extent that sub-subparagraphs (A)(i) and (ii), applicable only to (A), are indented. It's not. The inferences drawn by the Retirees—from § 4.03 or elsewhere—do not overcome the unambiguous structure of the section. The extent to which our holding here influences the district court's resolution of the increasing-or-nonincreasing debate is up to the district court's sound evaluation of the expert testimony.

In sum, we hold that the such-monthly-amount paragraph applies to all of § 4.03(a)(3), not just to subparagraph (B). We therefore affirm the district court's judgment on this issue.

C. Step 3: Convert MRI into the appropriate lump-sum benefit

To reiterate: a person's MRI under the Plan is the appropriate amount from § 4.03(a) plus the amount accrued under § 4.03(b)(2). Once we go back up through the decision tree to § 4.05(d), we are ready to convert the MRI into the form elected by the Plan Member.

The run-of-the-mill Plan Member will receive monthly payments from the Plan under the "Basic Form." The Basic Form is an "increasing monthly income commencing on the Member's Disability, Early, Normal or Late Retirement Date ... and continuing for sixty (60) months certain and for his lifetime thereafter." § 4.02(a). However, if a Member follows the proper steps, he can "elect ... to receive his Monthly Retirement Income in one of the alternative forms listed below." § 4.02(b). One of those forms is a lump sum, which is at issue here.

The lump sum rules are in § 4.02(b)(6). That section states:

[An] alternative form[ ] include[s] [a] lump sum payment payable at the Member's Disability, Early, Normal or Late Retirement Date ... calculated by multiplying the increasing Monthly Retirement Income determined under the applicable Section by 212 and, if the Monthly Retirement Income is due to Early Retirement or Disability Retirement, dividing by one (1) minus the appropriate reduction factor under Section 4.05(b)(5), Section 4.06(a)(5) or Section 5.01(c)(5), as applicable.

§ 4.02(b)(6). The "applicable Section" here is clearly § 4.05(d), and by extension, § 4.03(b). Thus, the lump sum is simply the MRI derived under § 4.03(a) and § 4.03(b)(2), multiplied by 212 and divided by one minus the reduction factor, as appropriate. The parties, however, present us with two quarrels about this math. First, the Retirees claim that the value of this lump sum must include the value of an increasing monthly income for sixty months certain, which basic-form beneficiaries get under § 4.02(a). Second, the Retirees claim that the early retirement reducers referenced in § 4.02(b)(6) were impliedly repealed by the 2004 amendments and do not apply to them. The district court agreed with the Retirees.

1

The Retirees contend that any lump sum must include the actuarial value of an increasing monthly income, guaranteed for sixty months certain. They reach this conclusion by arguing that the MRI under § 4.03(b) is identical to the Basic Form identified in § 4.02(a), and the Basic Form does require an increasing income for sixty months certain.The Retirees' argument is based on two observations. First, they point out that § 4.03(b) states that "the amount of the Member's Monthly Retirement Income under the basic form described in Section 4.02(a)" shall be determined by certain calculations. Therefore, they reason, the MRI is equivalent to the basic form and must abide by the same rules. Second, § 4.02(a) states that "[t]he basic form of Retirement Benefit (to which the formula indicated in Section 4.03 applies) shall be an increasing monthly income ... continuing for sixty (60) months certain and for [the retiree's] lifetime thereafter." The Retirees claim that since the basic form "shall be determined" per § 4.03(b) and "shall be an increasing monthly income" for sixty months certain per § 4.02(a), the MRI determined under § 4.03(b) is identical to the sixty-months-certain, increasing monthly income described under § 4.02(a). We disagree.

The district court reasoned that "4.03(b) ... explicitly states the MRI shall be in ‘the basic form described in Section 4.02(a).’ " This is only a half-truth. Section 4.03(b) states that "the amount of the Member's [MRI] under the basic form described in Section 4.02(a) ... shall be determined as set forth in paragraphs (1) and (2) hereof ...." § 4.03(b) (emphasis added). The emphasized language reveals that the district court was paddling against the current—it put the words "shall be" in the wrong spot. Under § 4.03(b), MRI is determined by reference to the two subparagraphs, not by reference to the basic-form rules. The words "under the basic form described in Section 4.02(a)" simply direct the reader on how to apply the output of the MRI calculation. § 4.03(b). In other words, the basic-form rules lie downstream from § 4.03(b), while the MRI calculation lies upstream . The beneficiary's election of an alternative form merely diverts the MRI output through a separate fork in the river.

This is confirmed by the lump-sum rules. Those rules instruct the reader to use the MRI "determined under the applicable Section," not the MRI as expressed by the basic-form rules. § 4.02(b)(6). The basic-form rules do not determine MRI, or provide any instructions for doing so, and therefore cannot be the "applicable Section." The basic-form section is, by its terms, a discussion of form , not substance. In contrast, § 4.03(b) provides detailed, substantive instructions on how to calculate MRI, and is therefore the "applicable Section" for § 4.02(b)(6) purposes.

However, this does not mean that the Retirees receive a nonincreasing benefit. The lump-sum rules themselves declare that the Plan must use an "increasing Monthly Retirement Income determined under the applicable Section." § 4.02(b)(6). Thus, regardless of what any other Part of the plan says, § 4.02(b)(6) itself requires that the input be an increasing MRI. Indeed, the expert testimony in the record, discussed above, suggests the MRI will always be increasing at this stage in the math, but the district court is free to examine this issue in more detail. However, because this section is silent about the sixty-months-certain issue, we hold that the Plan itself does not require the lump sum to account for that variance.

2

This is not the end of the matter. ERISA requires that any lump-sum alternative be the "actuarial equivalent" of the basic-form benefit. 29 U.S.C. § 1054(c)(3). Because the district court did not address actuarial equivalence, a remand is necessary to answer this question.

ERISA mandates that "lump-sum payments ... must be the actuarial equivalent of the normal accrued pension benefit." West v. AK Steel Corp. , 484 F.3d 395, 400 (6th Cir. 2007). The Retirees seize upon this doctrine to argue that a lump-sum benefit must include the actuarial value of an increasing monthly income for sixty months certain, since the basic-form benefit provides for those features. Norton counters that the lump sum benefit is already based on an increasing monthly income. This is true. However, § 4.02(b)(6) does not appear to include the sixty-months-certain feature of the basic-form benefit. ERISA requires Norton to account for that feature.

Further, Norton has complied with ERISA only if all its actuarial assumptions are valid, taken together. In other words, the calculation in § 4.02(b)(6)—multiplying the § 4.03(b) MRI by 212 and then dividing by one minus any applicable reduction factor—must produce a lump sum that is actuarially equivalent to the increasing, sixty-months-certain lifetime annuity provided under the basic-form rules. See West , 484 F.3d at 400. All of the Plan's other actuarial conversions, such as calculations from § 4.03(a)(3), must also be valid. According to Norton's experts, they are. The district court, however, struck the experts' reports and did not address this issue in its liability order.

This made granting summary judgment for the Retirees inappropriate. Whether the actuarial-equivalence requirement is satisfied is a factual dispute. And because the court granted the Retirees' motion, it was obligated to consider the evidence in the light most favorable to Norton. Cockrel v. Shelby Cty. Sch. Dist. , 270 F.3d 1036, 1056 (6th Cir. 2001). We think this would naturally include Norton's experts' explanation of how the actuarial conversions work. The court need not ultimately accept those explanations as the factfinder, but it must do so on the Retirees' motion for summary judgment. Id. Because there is a genuine issue of material fact, and because we have altered the district court's interpretation of the plan, neither party is entitled to summary judgment on this question. We therefore vacate the district court's grant of summary judgment and remand for further examination of the actuarial-equivalence issue.

3

At long last, we reach the final plan-interpretation issue. And it is here that we find the ambiguity mandating Firestone deference. Section 4.02(b)(6) states that "if the Monthly Retirement Income is due to Early Retirement ... [the lump sum shall be calculated by] dividing [the MRI] by one (1) minus the appropriate reduction factor under Section 4.05(b)(5), Section 4.06(a)(5) or Section 5.01(c)(5), as applicable." These sections have the effect of reducing a lump-sum benefit in proportion to the "earliness" of a person's retirement. Section 4.05(b)(5) is the governing rule for early retirees. Subparagraph (b)(5) comes on the heels of four other paragraphs explaining how to calculate MRI for a Member who retires before January 1, 2004. Section 4.05(b) (as amended) states:

Effective prior to January 1, 2004, upon the Member's retirement on his Early Retirement Date, he shall be entitled to a Monthly Retirement Income equal to the largest of the amounts provided under the applicable provisions of paragraphs (1), (2) or (3) below....

(1) The sum of [§ 4.03(a)(1), reduced by certain fractional amounts, and (a)(1)(B) ];

(2) For a Member with a Methodist Accrued Benefit, a reduced Monthly Retirement Income equal to his Accrued Benefit determined under Sections 2.01(a)(2) and 4.03(a)(2) as of his Early Retirement Date, [reduced by the subparagraph (1) fractions]

(3) For a Member with an NKC Accrued Benefit, [do the same calculation as in (2), but use § 4.03(a)(3), instead of § 4.03(a)(2) ].

(4) [omitted as immaterial].

(5) Anything herein to the contrary notwithstanding, the Monthly Retirement Income described in paragraphs (1), (2), (3) and (4) above [sic] be reduced or further reduced by one three hundredths (1/300) for each month by which the Member's Early Retirement Date precedes his Normal Retirement Date.

This is where the confusion sets in. "[P]aragraphs (1), (2), (3) and (4)" all refer to MRI calculations that were expressly made "effective prior to January 1, 2004." § 4.05(b)(5); Plan Amendments, § 6 (Jan. 2004). The amendments did not modify the cross-reference in § 4.02(b)(6); neither did they create a parallel reduction scheme for post-2004 early retirees. The Retirees claim that the amendments thereby "subsidized" early retirement by eliminating the reducers. Norton claims that it never intended to create such a subsidy, and that the lump-sum rules obviously contemplate that all early retirees will have their benefits reduced. We think that the Plan is ambiguous on this point. Under Firestone , therefore, the district court was obligated to determine whether Norton's interpretation was a reasonable way to resolve the ambiguity.

The Retirees' argument is plausible as a matter of the plain text of the Plan. At no point in our calculation of MRI do we encounter the rules in § 4.05(b). The first time they come up in the decision tree is during the lump-sum conversion. This appears to be intentional. A post-2004 early retiree is expressly governed by § 4.05(d), not § 4.05(b). That section, in turn, refers us to the normal-retirement rules in § 4.03(b). Thus, on the front end, the text of the plan strongly suggests that the Retirees do get the benefit of the normal-retirement rules, sans reductions. The lump-sum rules seem to dictate this outcome as well. Section 4.05(b)(5) is expressly limited in its operation to "the Monthly Retirement Income described in paragraphs (1), (2), (3) and (4) above." Since the Retirees' MRI was not calculated by way of those paragraphs, it is hard to see how subsection (b)(5) would apply, either.

However, Norton presents a plausible counter-point. The lump-sum rules do command that the MRI shall be reduced "if the [MRI] is due to Early Retirement," and that the reduction shall be accomplished by the rules in one of three sections, "as applicable." § 4.02(b)(6). The district court's view, essentially, was that the phrase "as applicable" in § 4.02(b)(6) means the cross-reference to the reducers only operates if § 4.05(b)(5) is applicable on its own terms; that is, only if the retiree's MRI has been calculated under § 4.05(b)(1)–(4). Norton correctly points out, however, that § 4.02(b)(6) says "as applicable," not "if applicable." We presume that this change in wording has real meaning, since the Plan drafters used the words "if applicable" many other times in the Plan documents, including in the 2004 Amendments. See, e.g. , Plan Amendments § 4 (Jan. 2004); Norton Plan § 2.01(a).

In sum, for post-2004 early Retirees, the Plan commands us to apply a section that, by its terms, does not apply. This is a classic example of a patent ambiguity. We suspect that during the amendment process, someone either failed to account for this particular fact pattern or neglected to include language clarifying that the reducers applied to post-2004 early retirees. But since we cannot discern the contours of the parties' intent from the text of the document, Firestone requires the courts to defer to Norton's interpretation unless it is arbitrary and capricious. However, this is a question that the district court should decide in the first instance. The district court refused to consider expert testimony on Norton's interpretation of the Plan at the liability stage, and as we have said, we believe such testimony is necessary to fully understand whether Norton's position is arbitrary and capricious. Since the record lacks a complete explanation for Norton's position, we think a remand is the best course of action. We therefore vacate the district court's holding that the Plan unambiguously favored the Retirees on this point, and we remand with instructions to apply Firestone deference.

V

The district court concluded that Kentucky's five-year limitations period for statutory claims applies in this case, rather than the fifteen-year limitations period for claims on written contracts. It also refused to decertify the class. Both issues are contested on appeal, and we address each in turn.

A

The Retirees contend that the district court should have applied a longer limitations period to their underpayment claims. See 29 U.S.C. § 1132(a)(1)(B). We agree. "ERISA does not explicitly provide a limitations period for § 1132(a)(1)(B) claims," so "courts fill the statutory gap using federal common law," and we look to the most analogous state statute of limitations to answer that question. Patterson v. Chrysler Grp., LLC , 845 F.3d 756, 762 (6th Cir. 2017). We have held that "when a plaintiff seeks benefits under the plan and those claims depend on alleged violations of ERISA's statutory protections, Kentucky's five-year limitations period applies." Fallin v. Commonwealth Indus., Inc. , 695 F.3d 512, 515 (6th Cir. 2012) (internal quotation marks omitted). In contrast, when a claim is based on the contract alone, then Kentucky's longer statute of limitations for contract claims applies. Redmon v. Sud–Chemie Inc. Ret. Plan , 547 F.3d 531, 537 (6th Cir. 2008).

To the extent that the Retirees' claims are based solely on the improper interpretation of the Plan terms, the contract limitations period must apply. As we explained in Redmon :

The plaintiffs in Fallin alleged that certain amendments to their employee retirement benefit plan violated the protections afforded them by ERISA. Although plaintiffs in Fallin also sought benefits under the plan, the benefits claim depended on the alleged violations of ERISA's statutory protections. Here, Redmon does not dispute that she signed the DFBP waiver. Rather, she argues that the plan administrator obtained her signature in violation of ERISA. She alleges that her waiver of survivor benefits was therefore invalid under ERISA and that she is entitled to the benefits she would have received if she had not signed the waiver. Redmon's claim for benefits therefore depends on a finding that her signature was invalidly obtained in violation of ERISA. Thus, her claim for benefits can be said to arise more specifically from ERISA's statutory protections than from an independent contract between the Redmons and Sud–Chemie. The result might be different if, for example, Redmon contested the authenticity of the signature on the DFBP. Such a claim might present an issue of contract law.

Id. This is such a case. At least as to their underpayment claims, the retirees argue that Norton misconstrued the Plan terms and failed to pay them what the Plan required. Absent some indication that this cause of action depends on "ERISA's statutory protections" rather than on the interpretation of the contract, the most analogous limitations period here would be that which Kentucky applies to contract claims. Id.

Thus, we hold that to the extent the retirees' claims are based on alleged misinterpretations of the Plan, they are governed by Kentucky's contractual limitations period. The retirees' actuarial-equivalence claims remain governed by the statutory limitations period.

B

Norton also challenges the district court's certification of the class at both the liability and damages stages. We will not reverse a certification decision in the absence of "a strong showing that the district court's decision amounted to a clear abuse of discretion." In re Whirlpool Corp. Front-Loading Washer Prods. Liab. Litig. , 722 F.3d 838, 850 (6th Cir. 2013). "An abuse of discretion occurs if the district court relies on clearly erroneous findings of fact, applies the wrong legal standard, misapplies the correct legal standard when reaching a conclusion, or makes a clear error of judgment." Id. Any class certification must satisfy Rule 23(a)'s requirement of numerosity, commonality, typicality, and adequate representation. Further, a class action must fit under at least one of the categories identified in Rule 23(b). The district court must conduct "a ‘rigorous analysis’ ... as to all the requirements of Rule 23." Pipefitters Local 636 Ins. Fund v. Blue Cross Blue Shield of Mich. , 654 F.3d 618, 630 (6th Cir. 2011).

1

In 2011, the district court certified the class under Rules 23(b)(1)(A) and (b)(2). It defined the class as:

All participants in the Norton Healthcare, Inc. Retirement Plan, its predecessors and successors, whose contractual lump sum pension benefits:

(1) Did not include the value of the basic form of benefit—an "increasing monthly retirement income" (annual cost-of-living adjustment)—when election of such basic form would have yielded the highest value for the participant; and/or

(2) Did not include the value of the "alternative" lump sum benefit where the basic form of benefit is multiplied by 212, when election of such alternative form would have yielded the highest value for the participant; and/or

(3) Did not include the value of the early retirement subsidy.

We assess the (b)(2) certification first. Rule 23(b)(2) applies when "the party opposing the class has acted or refused to act on grounds that apply generally to the class, so that final injunctive relief or corresponding declaratory relief is appropriate respecting the class as a whole." Any analysis of a (b)(2) certification must start with the Supreme Court's decision in Wal-Mart Stores, Inc. v. Dukes. In Dukes , the Court unanimously held that Rule 23(b)(2) "does not authorize class certification when each class member would be entitled to an individualized award of monetary damages." 564 U.S. 338, 360–61, 131 S.Ct. 2541, 180 L.Ed.2d 374 (2011). Instead, the Court stated that "individualized monetary claims belong in Rule 23(b)(3)," noting the due-process issues raised by binding a (b)(2) class without notice and an opportunity to opt out. Id. at 361–63, 131 S.Ct. 2541. Further, the Court stated:

The key to the (b)(2) class is the indivisible nature of the injunctive or declaratory remedy warranted—the notion that the conduct is such that it can be enjoined or declared unlawful only as to all of the class members or as to none of them. In other words, Rule 23(b)(2) applies only when a single injunction or declaratory judgment would provide relief to each member of the class.

Id. at 360, 131 S.Ct. 2541 (internal citations and quotation marks omitted). The Court held open the possibility, however, that (b)(2) certification might be appropriate when monetary relief is "incidental to ... injunctive or declaratory relief." Id. (internal quotation marks omitted).

Our precedent speaks to this residual question. In Gooch v. Life Investors Insurance Co. of America , the ultimate issue was "the difference between the charges for which [each class member] was billed and the amount that [the defendant] paid." 672 F.3d 402, 409, 428 n.15 (6th Cir. 2012). We held that those issues required (b)(3) certification. Id. at 428 n.15. However, we also stated that "Rule 23(b)(2) certification remains available ... when the plaintiffs seek a declaration about the meaning of a contract," even "when the declaratory relief serves as a predicate for later monetary relief." Id. at 427, 429.

Importantly, Gooch approved the practice of bifurcated certification: subrule (b)(2) certification for contract interpretation, and subrule (b)(3) certification of classes and subclasses for determining damages. Id. at 427–28. Although we have yet to extend this approach to post- Dukes ERISA cases, the Seventh Circuit has done so. Johnson v. Meriter Health Servs. Emp. Ret. Plan , 702 F.3d 364, 371 (7th Cir. 2012) ("[A]ll that the class is seeking ... at least initially, is a reformation of the Meriter pension plan—a declaration of the rights that the plan confers and an injunction ordering Meriter to conform the text of the plan to the declaration."). We think Gooch is equally applicable to ERISA plans, as demonstrated by the logic in Johnson . Therefore, the district court did not abuse its discretion by certifying the class under Rule 23(b)(2) for plan-interpretation purposes. Since the plan must mean the same thing regardless of the person to whom it applies, "uniform declaratory relief is appropriate." Gooch , 672 F.3d at 428 ; Johnson , 702 F.3d at 371.

The fact that the district court also certified the class under Rule 23(b)(1)(A) does not change that conclusion. Rule 23(b)(1)(A) permits certification when separate actions "would create a risk of ... inconsistent or varying adjudications with respect to individual class members that would establish incompatible standards of conduct for the party opposing the class." As the Supreme Court has explained,

Rule 23(b)(1)(A) takes in cases where the party is obliged by law to treat the members of the class alike (a utility acting toward customers; a government imposing a tax), or where the party must treat all alike as a matter of practical necessity (a riparian owner using water as against downriver owners).

Amchem Prods., Inc. v. Windsor , 521 U.S. 591, 614, 117 S.Ct. 2231, 138 L.Ed.2d 689 (1997) (internal quotation marks omitted).

The district court did not abuse its discretion by certifying a Rule 23(b)(1)(A) class at the liability stage. At that point in the case, the issue before the district court was the interpretation of the Plan. ERISA administrators may not discriminate between similarly situated beneficiaries. Varity Corp. v. Howe , 516 U.S. 489, 514, 116 S.Ct. 1065, 134 L.Ed.2d 130 (1996) ; Alday v. Raytheon Co. , 619 F.Supp.2d 726, 736 (D. Ariz. 2008) ("ERISA requires plan administrators to treat all similarly situated participants in a consistent manner."); Adams v. Anheuser–Busch Cos. , No. 2:10-CV-826, 2012 WL 1058961, at *10 (S.D. Ohio, Mar. 28, 2012). Here, individual actions by multiple retirees would clearly have created a risk of "inconsistent or varying adjudications" that "would establish incompatible standards of conduct" for Norton. Fed. R. Civ. P. 23(b)(1)(A). We therefore affirm the district court's certification, but only as it relates to the liability and plan-interpretation aspects of the case.

2

Throughout its hearings on damages, the district court refined its definition of the class. It held that the class includes all retirees who took lump-sum distributions after January 30, 2003, regardless of whether their benefit was ultimately determined on a cash-balance or defined-benefit basis, but not disability retirees or those receiving mandatory lump-sum payments. On appeal, Norton argues that certification under Rules 23(b)(1)(A) and (b)(2) was inappropriate; that retirees whose lump-sum distribution was determined on a cash-balance basis should not have been included in the class; and that, as ultimately defined, the class contains members with antagonistic interests, at least some of whom cannot be adequately represented by Khaliel. We hold that the district court abused its discretion by failing to engage in a meaningful analysis under Dukes when certifying a damages class. We therefore vacate the district court's decision on this point and remand for a more thorough analysis.We address the (b)(2) problem first. Dukes was clear that when a case involves individualized damages awards, a damages class cannot be certified under subrule (b)(2). Dukes , 564 U.S. at 360–62, 131 S.Ct. 2541. And it is clear from the record that the amount of any individual class member's award may vary wildly depending on their circumstances. This reality should have prompted the district court to consider the due-process concerns highlighted by the Supreme Court in Dukes . See id. at 362–63, 131 S.Ct. 2541 ; Randleman v. Fid. Nat'l Title Ins. Co. , 646 F.3d 347, 352 (6th Cir. 2011) (stating that district courts have a "continuing obligation to ensure that the class certification requirements are met"). The district court, however, said only: "This Court has previously considered and denied Norton's motion to alter or amend the class certification order. ... Having considered Norton's renewed arguments against class certification, this Court finds that its previous certification will stand." But this order contains no substantive discussion of certifying a damages class under Rule 23(b)(2).

This does not satisfy the "rigorous analysis" requirement. Pipefitters , 654 F.3d at 629–30. In Pipefitters , the district court issued an order certifying a class "for the reasons stated on the record." Id. at 628. At the relevant hearing, the district court had done little more than recite the applicable rule and state that its requirements were satisfied. Id. We reversed, holding that the district court had not addressed "the basic facts" underlying its decision. Id. at 629. Here, the district court did even less than the district court in Pipefitters . Indeed, the district court has never addressed the due-process issues raised by Dukes at all, much less analyzed them in light of the facts of this case. On remand, the district court must address this issue prior to certifying a damages class under subrule (b)(2).

The district court's (b)(1)(A) certification is also suspect on these grounds. Only Rule 23(b)(2) was at issue in Dukes , and the Court did not address whether its holding applied to Rule 23(b)(1) classes. 564 U.S. at 346 n.2, 131 S.Ct. 2541. But the due process concerns that led the Court to conclude that "individualized monetary claims belong in Rule 23(b)(3)" are similar for (b)(1) classes. Id. at 361–62, 131 S.Ct. 2541 ; see WILLIAM B. RUBENSTEIN, NEWBERG ON CLASS ACTIONS § 4:13 (5th ed. 2017). Thus, for the same reasons discussed above, the district court abused its discretion by failing to address these issues in a meaningful way. Although the district court may be within its authority to certify a damages class under (b)(1)(A), it must address these issues first.

In sum, to the extent that the district court certified the class for plan-interpretation purposes, that decision is affirmed. To the extent that its certification extended to damages calculations, we vacate the certification and remand for further proceedings consistent with this opinion. We do not reach the remainder of the parties' class-certification arguments. The parties also raise arguments about the damages formula adopted by the court and the appropriate rate of prejudgment interest. Because our resolution of the merits of this case requires us to vacate the award of damages, these issues are no longer ripe.

VI

This is not an easy case. We hope that our opinion today lends some finality to the major legal issues presented by the parties. To the extent that uncertainties remain, we reiterate the importance of the judicial role under Firestone : Ask only if the plan is ambiguous. If it is, Norton's decisions—factual and interpretive—must stand unless they are arbitrary and capricious.

Except as stated below, we AFFIRM the judgment of the district court. The district court's order dismissing certain contract claims as barred by the statute of limitations is REVERSED . Further, the district court's order granting summary judgment to Plaintiffs is VACATED IN PART insofar as it (a) held that the plan unambiguously subsidized early retirement; (b) held that the plan required Norton to pay a lump sum including the value of sixty-months-certain benefits; and (c) disposed of the actuarial-equivalence issue. Consequently, the district court's damages order is VACATED —including as it relates to class certification at the damages stage—and the case is REMANDED for proceedings not inconsistent with this opinion.

HELENE N. WHITE, Circuit Judge, concurring in part and dissenting in part.

I concur in the majority opinion except for Part IV. C. 3., addressing the early-retirement/lump sum issue, and Part III., addressing the applicability of the doctrine of contra proferentum .

The majority concludes that the Plan is ambiguous regarding reductions for early retirement, and remands for further consideration and application of Firestone deference. I would affirm the grant of summary judgment to the Retirees on this issue based on the unambiguous provisions of the Plan. As the majority opinion explains, 4.05(b), which contains all the early-retirement reductions, including the 4.05(b)(5) reduction at issue here, applies to Members who retired before January 1, 2004. (Maj. Op. at 273.) For Members retiring after that date, 4.05(b) was replaced with 4.05(d), which adopts 4.03(b)(3) as the applicable formula. Section 4.03(b)(3) has no reductions for early retirement. The majority and I agree that the old 4.05(b) reductions do not apply based on the express language of section 4.05. (Id. at 35.)

The majority concludes, however, that ambiguity is introduced by 4.02(b)(6), which continues to refer to 4.05(b)(5), and other reducers, notwithstanding that they no longer apply. (Id. at 34-36.) Section 4.02(b)(6) states:

A lump sum payment payable at the Member's Disability, Early, Normal or Late Retirement Date or the date specified in Section 5.01, and calculated by multiplying the increasing Monthly Retirement Income determined under the applicable Section by 212 and, if the Monthly Retirement Income is due to Early Retirement or Disability Retirement, dividing by one (1) minus the appropriate reduction factor under Section 4.05(b)(5), Section 4.06(a)(5) or Section 5.01(c)(5), as applicable.

In response to the Retirees' analysis of the Plan language, the majority states that "Norton presents a plausible counter-point. The lump-sum rules do command that the MRI shall be reduced ‘if the [MRI] is due to Early Retirement,’ and that the reduction shall be accomplished by the rules in one of three sections, ‘as applicable.’ " (Id. at 35.) It is here that I part ways with the majority. The lump-sum rules do not command that the MRI be reduced if the MRI is due to early retirement. Nowhere are the words "the MRI shall be reduced if due to early retirement," or similar words, found in the section. Instead, the section sets forth a formula for converting the MRI into a lump sum. This formula instructs that the lump-sum payment be calculated by multiplying the MRI by 212, and "if the Monthly Retirement Income is due to Early Retirement," "dividing by one (1) minus the appropriate reduction factor under Section 4.05(b)(5), ... as applicable." Applying this second step does not reduce the MRI; it increases it. Mathematically, it cannot be disputed that if after multiplying the MRI by 212 as instructed, one goes on to divide that product by 1 minus the appropriate reduction factor under 4.05(b)(5), the lump-sum payment goes up, not down; dividing a number by a number that is less than 1 increases the number. This is not a drafting mistake. The Form of Benefit section makes an adjustment for early-retirement reductions that were at one time taken upstream of the 4.02(b) calculations. All of the 4.02(b) alternatives intentionally restore the early-retirement MRIs to their values before the early-retirement reduction by dividing by one minus the applicable reduction. And Norton's proposed calculations do so as well. (See infra note 2.)

Thus, reading section 4.05 on Early Retirement and section 4.02(b)(6) on lump-sum payments together does, indeed, reveal an ambiguity, because 4.02(b)(6) continues to refer to 4.05(b)(5) and the other reducers, and reverses early-retirement reductions that are no longer applicable. But that ambiguity is not whether the 4.05(b)(5) reduction should be taken; it is whether the 4.02(b)(6) formula (which increases the payment) should still apply in light of the elimination of 4.05(b)(5)'s upstream reductions. The majority observes that "for the post-2004 early Retirees, the Plan commands us to apply a section that, by its terms, does not apply. This is a classic example of a patent ambiguity. ... But since we cannot discern the contours of the parties' intent from the text of the document, Firestone requires the courts to defer to Norton's interpretation unless it is arbitrary and capricious." (Maj. Op. at 277.) I agree. But this does not lead to the conclusion the majority reaches.

Any reading of the Plan that fails to recognize that 4.05(d) rendered all the provisions of 4.05(b), including 4.05(b)(5), inapplicable would be arbitrary and capricious. As explained, the ambiguity introduced by 4.02(b)(6) is not whether to apply the reduction found in 4.05(b)(5) itself, because it clearly no longer applies; the ambiguity is whether 4.02(b)(6) should be applied in accordance with its explicit terms, or whether the final adjustment (dividing by 1 minus the reduction factor) should be ignored because the reduction was never taken. This is the issue regarding which Norton is entitled to Firestone deference. But because the Retirees do not argue that they are entitled to have their MRIs increased under 4.02(b)(6), and argue only that Norton wrongly decreased their benefits under 4.05(b)(5), we do not have to address the real ambiguity. Because Norton's contention that 4.02(b)(6) somehow entitles it to reduce benefits under 4.05(b)(5) is entirely unsupported and contrary to the plain language of the Plan, it is arbitrary and capricious. I would therefore affirm the district court's ultimate conclusion that 4.05(b)(5) cannot be used to reduce the MRI based on the unambiguous language of the Plan, and would not remand for further consideration of the issue.

This leads me to the contra proferentum issue. I would accept the district court's statement that it did not apply the doctrine, and find no need to address it because, as in our prior decisions, the language at issue is unambiguous. I acknowledge that the majority reaches the contra proferentum issue because it finds the early-retirement provisions ambiguous. However, given our many references to the doctrine in prior cases, we should refrain from foreclosing application of contra proferentum in conjunction with Firestone deference in all cases until presented with a case in which that doctrine is actually applied. To be sure, the two approaches are in inherent tension, and in light of Firestone , there can be no binding "rule of interpretation that would completely contradict the deference paid to an administrator's decision" in an ERISA case. Mitchell v. Dialysis Clinic, Inc. , 18 Fed.Appx. 349, 353 (6th Cir. 2001). However, we have said that "even an arbitrary and capricious standard of review can be tempered by ... construing ambiguities against a plan drafter." Copeland Oaks v. Haupt , 209 F.3d 811, 813 (6th Cir. 2000) (citing University Hosps. of Cleveland v. Emerson Elec. Co. , 202 F.3d 839, 846–47 (6th Cir. 2000) ). What that application might look like is an open question. However, acknowledging the tension between contra proferentum and Firestone deference, I find the analogy to the conflict-of-interest situation apt. Just as we are instructed in appropriate cases to temper the arbitrary and capricious standard by taking into account the fiduciary's conflict of interest, Metro. Life Ins. Co. v. Glenn , 554 U.S. 105, 117, 128 S.Ct. 2343, 171 L.Ed.2d 299 (2008), there may be situations in the gray area between arbitrary and capricious and Firestone deference where the contra proferentum doctrine may have a role. Because we have not been presented with such a case, we have not been required to resolve the issue. And because the district court did not apply the doctrine, I would not do so now.

I join in the majority opinion in all other respects.