Plaintiff-appellee-cross-appellant Ronald Roganti (“Roganti”) was a successful executive with defendant-appellant-cross-appellee Metropolitan Life Insurance Company (“MetLife”1) until 2005, when he resigned in the face of pay reductions that he claims were levied in retaliation for his opposition to unethical business practices. Roganti *204brought arbitration proceedings against MetLife before the Financial Industry Regulatory Authority (“FINRA”), seeking, among other things, wages that he would have been paid but for the retaliatory pay reductions, as well as compensation for the decreased value of his pension, which was tied to his wages. The FINRA panel awarded Roganti approximately $2.49 million in “compensatory damages,” but its award did not clarify what that sum was compensation for. Roganti then filed a benefits claim with MetLife, arguing that the award represented back pay and that his pension benefits should be adjusted upward as if he had earned the money while he was still employed. MetLife denied the claim because the FINRA award did not say that it was, in fact, back pay. Roganti brought this lawsuit.
The Employee Retirement Income Security Act of 1974 (“ERISA”), 29 U.S.C. § 1001 et seq., creates a private right of action to enforce the terms of a benefit plan. 29 U.S.C. § 1132(a)(1)(B). Roganti’s pension plans vest interpretive discretion in the plan administrator, which means that the plan administrator’s benefits decision is conclusive unless it is “arbitrary and capricious.” Pagan v. NYNEX Pension Plan, 52 F.3d 438, 441 (2d Cir.1995). After a summary bench trial on stipulated facts, the district court (Engelmayer, J.) determined that MetLife’s denial of Roganti’s claim was arbitrary and capricious because it was clear from the arbitral record that the award did represent back pay. We reverse. For the reasons stated below, we conclude that Met-Life’s denial of Roganti’s claim was not arbitrary and capricious, and that MetLife is therefore entitled to judgment in its favor as to Roganti’s benefits claim. In light of this decision, we affirm the district court’s denial of Roganti’s request for attorney’s fees, from which Roganti has cross-appealed.
BACKGROUND
According to his complaint in this action, Roganti began working for MetLife as an account representative in 1971. Over the course of the next three decades, he was promoted multiple times, becoming a Vice President and the Managing Director of a New York business unit called R. Roganti & Associates, as well as the Executive Director of Agencies at another MetLife business unit known as the Tower Agency Group. Roganti’s compensation rose significantly over the course of his career, and particularly between 1994, when he earned $351,000, and 2001, when he earned a high of $2,007 million.
Roganti alleges that beginning in 1999 and continuing until his retirement in 2005, his relationship with MetLife deteriorated as a result of his objections regarding unlawful, inappropriate, and unethical conduct at the company. Among various allegations, Roganti claims that a subordinate of his, Dorian Hansen, came under fire in 1999 for opposing fraudulent business practices employed by some MetLife insurance salespeople. As part of a campaign against Hansen’s efforts to end these practices, Roganti was allegedly told to fire her; when he refused, the Tower Agency Group was dissolved, affecting Ro-ganti’s compensation. Roganti alleges that he thereafter continued to oppose illegal and unethical conduct, and that Met-Life further reduced his compensation in retaliation. As a result, Roganti filed a retaliation complaint with the Occupational Safety and Health Administration (“OSHA”) in 2003, under the Sarbanes-Oxley Act of 2002 (“SOX”), Pub.L. No. 107-204, 116 Stat. 745 (codified in relevant part at 18 U.S.C. § 1514A). The complaint was dismissed, however, when OSHA’s preliminary investigation did not *205validate Roganti’s claims. A second complaint, filed in January 2004 and alleging further acts of retaliation, was dismissed in November of that year.
In July 2004, while the second OSHA complaint was still pending (and some eight months before he retired), Roganti commenced FINRA arbitration proceedings against MetLife. In the arbitration, Roganti advanced three theories of recovery in addition to his claim that MetLife had violated SOX by retaliating against him for opposing its business practices and for filing a SOX complaint: (1) breach of contract, for the alleged breach of Met-Life’s commitment to make certain payments to him for overseeing the Tower Agency Group; (2) quantum meruit, to recover the reasonable value of services Roganti had provided, but for which he had been underpaid; and (3) ERISA violations, on the theory that MetLife had violated the statute by reducing Roganti’s compensation for the purpose of diminishing his pension benefits. Roganti’s statement of claim contained two separate paragraphs describing his request for relief. The first requested “back pay, liquidated damages, compensatory and punitive damages, attorneys fees and an accounting.” J.A. 45. The second — in the statement of claim’s “Wherefore” clause — sought an accounting of R. Roganti & Associates’ revenues and expenses; “appropriate back pay, front pay and reimbursement for lost benefits”; liquidated damages; punitive damages; and attorney’s fees and costs. J.A. 60-61.
The arbitration did not conclude until 2010, after a seventeen-day hearing. Ro-ganti’s counsel made clear throughout the proceedings that Roganti was focused on recovering two categories of damages: damages for “lost comp[ensation]” and “damages for the collateral effect [on] his pension benefits which are directly tied to his comp[ensation].” J.A. 2881; see also, e.g., J.A. 2891 (stating that Roganti “seeks nothing more than the compensation and pension benefits that he worked for and earned” (emphasis added)). Roganti is entitled to pension benefits by virtue of his participation in four MetLife retirement plans (the “Plans”) and, as previously noted, he claimed before the arbitral panel that MetLife had reduced his compensation for the specific purpose of limiting the growth of his sizable pension.2 In his summation to the FINRA panel, Roganti’s counsel explained how these two damages components should be calculated based on the evidence presented during the arbitral hearing.
As to the first component of damages— i.e., for lost compensation (or what Rogan-ti’s attorney referred to as “back pay”)— Roganti retired in 2005, when he was 55 years and six months old, but he testified that he would have continued working at MetLife until age 62 had his compensation not been reduced. Roganti’s counsel therefore argued that the panel should determine what Roganti would have earned not only in the years 2003 to 2005 had the company not reduced his compen*206sation, but also (because he would not have retired) what he would have earned through September 2010.3
The evidence showed that Roganti’s pay rose significantly from 1994 through 2001, and then decreased markedly: he earned $1,168 million in 1998, $1.24 million in 1999, $1,406 million in 2000, and $2,007 million in 2001, before his compensation declined to $1,506 million in 2002, $475,000 in 2003, $383,000. in 2004, and $67,000 for the first quarter of 2005, after which he retired.4 Based on these sums-, Roganti’s counsel proposed three scenarios for determining Roganti’s lost compensation. The first assumed that Roganti’s compensation would have increased four percent annually from an estimate of his expected 2002 earnings contained in MetLife’s pension files, yielding total lost compensation — Roganti’s “but for” earnings from 2003 to 2010 minus the sum of his actual earnings from 2003 to 2005 and the actual pension payments that he had received thereafter — of $14,768 million. The second scenario assumed that Roganti’s compensation would have remained flat at $1,506 million (ie., what he actually received in 2002), yielding a lower total lost compensation figure of $7,350 million. The third scenario effectively split the difference between the first two, yielding total lost compensation of roughly $11 million.
With respect to the second component of damages — ie., for pension benefits to which Roganti would have been entitled but for the retaliatory decrease in his compensation (or what Roganti’s attorney sometimes called “front pay”) — the panel heard extensive testimony regarding how Roganti’s benefits were calculated under the Plans. As Michael Bailey, an Assistant Vice-President and Corporate Actuarial at MetLife, explained (in testimony that was generally consistent with testimony from both Roganti and Robert Benmosche, MetLife’s CEO), Roganti’s retirement benefits were primarily a function of two variables: his pre-retirement compensation and his retirement age. Had Roganti retired at age 62, he would have been entitled to an annual benefit based on his five highest-earning years within his last fifteen years of employment at MetLife. However, this annual benefit was subject to an “early retirement discount” of four percent per year: retiring at 61 would mean receiving 96 percent of the annual benefit, retiring at 60 would mean receiving 92 percent of that benefit, retiring at 59 would mean receiving 88 percent, and so on. J.A. 2894.
Based on an assumption that Roganti’s remaining life expectancy in 2010 was roughly twenty years, Roganti’s counsel advanced a proposal, during his summation, as to how the arbitral panel ought to use the foregoing evidence to award Ro-ganti damages for his decreased pension benefits. Based on MetLife’s own projections of Roganti’s pension had he not retired in 2005 (which assumed four percent annual growth from his actual 2002 compensation), Roganti would have been entitled to about $120,000 a month in pension benefits, or about $830,000 more per year than he had been receiving. Aggregated over twenty years, that difference would amount to lost benefits of about $17 mil*207lion. Alternatively, Roganti’s counsel estimated that by virtue of Roganti’s early retirement alone, even with no increase in compensation above what he had actually earned during his five highest-earning years, he had suffered a loss of about $3.42 million in benefits as a result of the early retirement discount.
Roganti’s counsel acknowledged, however, that the calculation of a pension award involves some measure of speculation. Most notably, if Roganti’s compensation during any year between 2002 and his retirement would, but for MetLife’s unlawful pay reductions, have been higher than the compensation he earned during one of his prior five highest-earning years, that higher figure would properly replace the lower one for calculating his pension. Thus, as an alternative, Roganti’s counsel suggested that the panel could avoid calculating Roganti’s pension by simply deciding what Roganti would have earned while still employed through age 62 (ie., by choosing a back pay award), and then instructing MetLife to determine Roganti’s entitlement to pension benefits based on that increase in compensation. As he put it:
But once again, I don’t want to be unduly speculative. So to the extent this panel has concerns about awarding front pay in specific amounts, I submit that an award should be rendered which says assume Ron had earned such and such through the date of his retirement at 62 and, MetLife, give him the appropriate pension amounts based on that. Met can adjust its pension.
... [T]o completely eliminate the speculation, the answer is to have Met adjust Ron’s pension based on a retirement of 62 and based on earnings of either 2 million or 1.5 million or some other higher or lower figure through 62. This panel can do that. They can order Met to adjust Ron’s pension accordingly or to buy an annuity so that he receives those payments for the rest of his life....
J.A. 2290.
The panel’s chairman, making clear that neither side “should infer anything from [the] question,” then inquired whether the proposed “front pay part of [the] calculation,” representing the loss of pension payments into the future, shouldn’t be discounted to the present value of such payments, “if it is going to be reduced into an award.” J.A. 2291. Roganti’s counsel agreed that this was necessary, that “[t]his should be present valued, this $17 million if you are going to award it that way.” J.A. 2291. He added, “that was one of the reasons I was addressing to the panel the alternative possibility of Met issuing its own annuity, adjusting the figures accordingly according to what Ron’s compensation would be if he worked to 62.” J.A. 2291. The panel chairman then concluded this discussion by noting that “under the scenario that we give you everything you want,” the chairman himself could do the discounting: “I mean, I have a calculator, I could do it.” J.A. 2291.
On August 28, 2010, the FINRA panel issued an award in Roganti’s favor. Consistent with the first above-quoted paragraph in Roganti’s statement of claim, the award noted that Roganti had requested “unspecified compensatory damages, unspecified punitive damages, an accounting of R. Roganti & Associates’ revenues and expenses, appropriate back pay, front pay and reimbursement for lost benefits, attorneys’ fees, costs, disbursements, interest, and such further and additional relief as the Panel may deem just and proper.” J.A. 63. The arbitration panel further noted that Roganti had requested compensatory damages in the range of $11,483,000 *208to $32,764,506. Under the heading “Award,” the award stated simply:
After considering the pleadings, the testimony and evidence presented at the hearing, the Panel has decided in full and final resolution of the issues submitted for determination as follows:
1. [MetLife] is liable for and shall pay to [Roganti] compensatory damages in the amount of $2,492,442.07 above its existing pension and benefit obligation to [Roganti].
2. Any and all relief not specifically addressed herein, including punitive damages, is denied.
J.A. 64: The award did not state the basis on which the panel had found MetLife liable, did not say what the “compensatory damages” were intended to compensate Roganti for, did not allocate the damages to particular years of Roganti’s employment, and did not instruct MetLife to recalculate Roganti’s pension benefits based on the award or to purchase an annuity.
On March 24, 2011 (roughly seven months after the panel issued its award), Roganti filed a benefits claim with Met-Life, in its capacity as administrator of the Plans, arguing that the award was compensation for income that MetLife had unlawfully denied him while he was employed there, and that this increase in his pre-retirement earnings justified an increase in his benefits under the Plans. Roganti’s claim was denied, first on June 16, 2011 by Karen Dudas, and then, after Roganti appealed Dudas’s decision, on August 30, 2011 by Andrew Bernstein, MetLife’s Plan Administrator. Dudas and Bernstein informed Roganti that an award of general “compensatory damages” was not benefits-eligible compensation under the Plans, and that nothing in the award indicated that it was intended to compensate him for wages he should have earned during his employment at MetLife, much less for wages attributable to particular years of service.5 As Dudas explained:
In the Relief Requested section of the Award document, “compensatory damages” are mentioned separately from back pay, front pay and reimbursement for lost benefits. The Award document specifically states that any and all relief not specifically addressed is denied. Rather than requiring the Award be included in your compensation earned as an Employee (by requiring treatment as back pay) or included in benefit eligible compensation, the Award document specifically labels the relief as “compensatory damages” (separate and apart from any pension and benefit obligation) and denies any and all other types of requested relief — including categorization of the award as back pay, front pay or reimbursement for lost benefits.
J.A. 830.' Accordingly, MetLife concluded, “[w]e do not see a reasonable basis in the Award document for treating the Award as benefit eligible compensation earned while you were an Employee.” J.A. 832.
Roganti commenced this action in the United States District Court for the Southern District of New York on January 9, 2012, after his appeal at MetLife had been denied. His complaint alleged that in de*209nying his benefits claim, MetLife had violated the terms of the Plans (so Roganti was entitled to relief under ERISA) as well as the anti-retaliation provisions of both SOX and the Dodd-Frank Wall Street Reform and Consumer Protection Act (“Dodd-Frank”), Pub.L. No. 111-203, 124 Stat. 1376 (2010) (codified in relevant part at 15 U.S.C. § 78u-6). The parties stipulated to the dismissal of the Dodd-Frank claim on April 30, 2012. MetLife then moved to dismiss the ERISA and SOX claims for failure to state a claim, and on the ground that they were barred by res judicata and collateral estoppel because Roganti had already sought increased pension benefits from the FINRA panel.
On June 18, 2012, the district court issued an opinion dismissing the SOX claim for failure to state a claim, but denying MetLife’s motion to dismiss the ERISA claim. Roganti v. Metro. Life. Ins. Co., No. 12 Civ. 161(PAE), 2012 WL 2324476 (S.D.N.Y. June 18, 2012) (“Roganti /”). With respect to the latter claim, the district court held that if the FINRA award represented back pay, then it would be benefits-eligible compensation under the Plans. Id. at *6. But because the language of the award itself “d[id] not come close to reliably resolving '.whether or not the award represented back pay,” the district court found it necessary to remand the case to MetLife, as plan administrator, to conduct “a close review of the arbitral record” and determine' based on that review “whether or not the award represented back pay.”6 Id. at *8. (Although the relevant arbitration procedures permitted Roganti to seek clarification of the award from the arbitrators, the three-month period for doing so had passed. Id.) The district court issued an order remanding to MetLife on June 29, 2012. Special App’x 17-19.
On remand, MetLife’s Bernstein adhered to his earlier determination. In a submission dated September 14, 2012, he stated that despite having reviewed the arbitration hearing transcript and exhibits, he had uncovered neither “evidence ... of any mathematical or formulaic calculation that would allow [him] to identify the specific nature of the Award,” nor “evidence of the arbitration] panel’s intent to consider any potential award as eligible for consideration when determining Roganti’s retirement benefits.” J.A. 2897. In the absence of such evidence, Bernstein reasoned that the panel’s failure to indicate that.it was awarding Roganti only back pay — even though it knew he was seeking to recover for his reduced pension benefits — reflected a conscious choice by the panel to “award[ ] an amount in the form of compensatory damages ... to take care of all of Roganti’s claims during the arbitration.” J.A. 2900 (emphasis added).
In light of Bernstein’s submission, the parties agreed to have the district court resolve their dispute by way of a summary bench trial on a stipulated factual record. On September 25, 2013, the district court issued an opinion agreeing with Roganti that the award represented back pay and holding that Bernstein’s decision reaching the opposite conclusion was arbitrary and capricious. Roganti v. Metro. Life Ins. Co., 972 F.Supp.2d 658 (S.D.N.Y.2013) (“Roganti II”).
In concluding that the award was back pay, the district court adopted what it viewed as a “coherent explanation” prof*210fered by Roganti for what the award represented. Id. at 668. Specifically, if the panel had (1) assumed, consistent with Ro-ganti’s second back pay scenario, that his pay would have remained flat at its 2002 level of $1,506 million; (2) declined to credit his testimony that he would not have retired in 2005 but for the retaliatory pay reductions; and (3) awarded him the difference between $1,506 million annually and his actual earnings from 2003 to his retirement in 2005, then the resulting award would have been $2,492,461.53 — just $19.46 more than the panel actually awarded him. Id. at 668-69. Because Bernstein had failed either to rebut this theory or offer his own account of what the award represented, the district court held that he had acted arbitrarily and capriciously. Id. at 670-73. The court therefore ordered MetLife to adjust Roganti’s benefits to incorporate the award as benefits-eligible compensation. Id. at 675. It also denied Roganti’s request for attorney’s fees. Id. at 674-75.
On October 29, 2013, MetLife provided Roganti with a calculation of the judgment due to him as a result of the district court’s September 2013 opinion. On November 14, 2013, the district court issued a short opinion rejecting certain challenges that Roganti had made to this calculation, directing judgment in Roganti’s favor in the amount of $761,051 (plus interest) for previous benefits payments that he had not received, and ordering MetLife to incorporate the arbitration award into Roganti’s 2003-2005 benefits-eligible compensation for purposes of calculating his benefits going forward. Roganti v. Metro. Life Ins. Co., No. 12 Civ. 161(PAE), 2013 WL 6043917 (S.D.N.Y. Nov. 14, 2013) (“Roganti III”). Judgment was entered on November 22, 2013. MetLife filed a timely notice of appeal, and Roganti timely cross-appealed from the district court’s denial of his request for attorney’s fees. We have jurisdiction under 28 U.S.C. § 1291.
DISCUSSION
Our review of a decision following a bench trial on stipulated facts is de novo, at least where, as here, the district court reached legal conclusions based on its review of the record and did not hear witness testimony or make credibility determinations. See Miles v. Principal Life Ins. Co., 720 F.3d 472, 485 (2d Cir.2013).
ERISA creates a private right of action to enforce the provisions of retirement plans. See 29 U.S.C. § 1132(a)(1)(B). The Supreme Court has held that plan administrators’ decisions on benefits eligibility must be evaluated pursuant to principles of trust law, which “make a deferential standard of review appropriate when a trustee exercises discretionary powers.” Firestone Tire & Rubber Co. v. Bruch, 489 U.S. 101, 111, 109 S.Ct. 948, 103 L.Ed.2d 80 (1989); see Restatement (Second) of Trusts § 187 (1959). Accordingly, where (as in this case) the relevant plan vests its administrator with discretionary authority over benefits decisions, see supra note 2, the administrator’s decisions may be overturned only if they are arbitrary and capricious. See Pagan v. NYNEX Pension Plan, 52 F.3d 438, 441 (2d Cir.1995).7 In *211this context, arbitrary and capricious means “without reason, unsupported by substantial evidence or erroneous as a matter of law.” Id. at 442 (quoting Abnathya v. Hoffman-La Roche, Inc., 2 F.3d 40, 45 (3d Cir.1993), abrogated on other grounds by Metro. Life Ins. Co. v. Glenn, 554 U.S. 105, 128 S.Ct. 2343, 171 L.Ed.2d 299 (2008)) (internal quotation marks omitted). This standard is “highly deferential,” and “the scope of judicial review is narrow.” Celardo v. GNY Auto. Dealers Health & Welfare Trust, 318 F.3d 142, 146 (2d Cir.2003).
On appeal, MetLife does not contest the district court’s holding that if the FINRA award represented wages that Roganti would have earned at MetLife but for' the unlawful reduction in his compensation— i.e., if the award represented back pay— then the award was benefits-eligible. See Roganti I, 2012 WL 2324476, at *6 (“Although Roganti was no longer employed at the time of the award, if that award represented back pay to compensate him for services rendered while he was a MetLife employee, such compensation would properly be included in benefits calculations.”). Accordingly, the question before us is whether MetLife was arbitrary and capricious in rejecting Roganti’s contention that the FINRA award did, in fact, represent back pay.
Roganti argues, and the district court held, that MetLife’s denial of his claim was arbitrary and capricious because it is sufficiently evident that the FINRA award represents back pay for the period starting in 2003 and ending upon Roganti’s retirement in 2005. As noted, Roganti’s theory is that the FINRA panel (1) rejected his contention that he would not have retired in 2005 had his compensation not decreased; (2) assumed that, but for the unlawful reduction in his pay, his annual compensation from 2003 to 2005 would have equaled his 2002 compensation of $1,506 million; and (3) awarded him the difference between that “but for” amount and his actual earnings from 2003 to 2005, and nothing else. That theory — under which the award necessarily constitutes back pay — predicts a dollar figure that is $19.46 higher than the amount of the actual award. The district court faulted Met-Life for rejecting Roganti’s “convincing explanation for the Award” without offering its own competing explanation of what the award represented. Roganti II, 972 F.Supp.2d at 672. MetLife argues that this was error, and for the following reasons, we agree.
A.
An ERISA plan administrator such as MetLife owes a fiduciary duty not just to the individual participant or beneficiary whose claim is under review, but to all of the participants and beneficiaries of the plan. See 29 U.S.C. § 1104(a)(1); Morse v. Stanley, 732 F.2d 1139, 1144-45 (2d *212Cir.1984). As a result, “ERISA requires a balance between ‘the obligation to guard the assets of the trust from improper claims [and] the obligation to pay legitimate claims.’ ” Harrison v. Wells Fargo Bank, N.A., 773 F.3d 15, 20 (4th Cir.2014) (quoting LeFebre v. Westinghouse Elec. Corp., 747 F.2d 197, 207 (4th Cir.1984)). In striking this balance with respect to a particular claim, a fiduciary must, among other things, assess whether the claimant has furnished sufficient evidence that he is entitled to the benefits he seeks.
On the one hand, “[p]lan administrators ... may not arbitrarily refuse to credit a claimant’s reliable evidence.” Black & Decker Disability Plan v. Nord, 538 U.S. 822, 834, 123 S.Ct. 1965, 155 L.Ed.2d 1034 (2003). But it is also true that administrators may exercise their discretion in determining whether a claimant’s evidence is sufficient to support his claim. For instance, in cases where the evidence conflicts, an administrator’s conclusion drawn from that evidence that a claim should be denied will be upheld unless the evidence points so decidedly in the claimant’s favor that it would be unreasonable to deny the claim on the basis of the evidence cited by the administrator. Compare Tocker v. Philip Morris Cos., 470 F.3d 481, 490-91 (2d Cir.2006) (upholding administrator’s determination that plaintiff was not a participating employee under the plan where “ample evidence” suggested that he had been terminated, though there was “also evidence that [he] was not terminated”), and Hobson v. Metro. Life Ins. Co., 574 F.3d 75, 89-90 (2d Cir.2009) (rejecting argument that administrator erred in concluding that the claimant was not disabled based on its weighing of competing medical evaluations), with McCauley v. First Unum Life Ins. Co., 551 F.3d 126, 138 (2d Cir.2008) (concluding that administrator’s “reliance on one medical report ... to the detriment of a more detailed contrary report without further investigation was unreasonable”), and Durakovic v. Bldg. Serv. 32 BJ Pension Fund, 609 F.3d 133, 140 (2d Cir.2010) (holding that funds’ reasoning was inappropriately “one-sided” where funds “summarily dismissed” a report by the claimant’s vocational expert, “which was vastly more detailed and particularized than the report on which the Funds relied”). Put differently, if the administrator has cited “substantial evidence” in support of its conclusion, the mere fact of conflicting evidence does not render the administrator’s conclusion arbitrary and capricious. See Durakovic, 609 F.3d at 141 (defining “substantial evidence” as “such evidence that a reasonable mind might accept as adequate to support the conclusion reached by the administrator” (quoting Celardo, 318 F.3d at 146)).
In other cases, the record evidence may not conflict — the only available evidence may, in fact, point in the claimant’s favor — but it may nonetheless be permissible for the plan administrator to conclude that the evidence is insufficient to support the claim. See Juliano v. Health Maint. Org., 221 F.3d 279, 287-88 (2d Cir.2000) (explaining that an ERISA claimant bears the burden of establishing his entitlement to benefits).8 For instance, in Jiras v. *213Pension Plan of Make-Up Artist & Hairstylists Local 798, the plaintiffs claim for pension benefits turned on whether he had worked in 1964 or 1965 for a union employer that contributed to the fund. 170 F.3d 162, 163-64 (2d Cir.1999). We held that it was not arbitrary and capricious for the plan administrator to deny the plaintiffs claim, given that the only evidence of his union employment over that period was an unsubstantiated affidavit from an unreliable witness. See id. at 166.
Under certain circumstances, it may be arbitrary and capricious for the administrator to reject a claimant’s evidence as inadequate without making a reasonable effort to develop the record further. See Gaither v. Aetna Life Ins. Co., 394 F.3d 792, 808 (10th Cir.2004) (“An ERISA fiduciary presented with a claim that a little more evidence may prove valid should seek to get to the truth of the matter.”); Miller v. United Welfare Fund, 72 F.3d 1066, 1073 (2d Cir.1995) (faulting fund for not seeking evidence to confirm whether its “speculation” as to plaintiffs need for a private nurse was correct). Ultimately, however, “[t]he rule is one of reason,” and “[njothing ... requires plan administrators to scour the countryside in search of evidence to bolster a petitioner’s ease.” Harrison, 773 F.3d at 22; see also O’Reilly v. Hartford Life & Accident Ins. Co., 272 F.3d 955, 961 (7th Cir.2001) (explaining that ERISA requires a “reasonable inquiry,” not a “full-blown investigation” (internal quotation marks omitted)). Thus, a claimant’s evidence may simply be insufficient to establish his entitlement to benefits, even in the absence of evidence tending to refute his theory of entitlement. See, e.g., Jiras, 170 F.3d at 166.
B.
In this case, it is undisputed that Roganti offered no direct evidence that the FINRA award represents back pay, and that the time for seeking clarification of the award from the arbitral panel was allowed to pass. Accordingly, the question is whether it was arbitrary and capricious for MetLife to determine that Roganti failed to establish his entitlement to additional pension benefits based (as Roganti’s claim was) solely on inferences that Ro-ganti urged should be drawn from the materials that MetLife reviewed in assessing his claim. In light of the principles described above, we conclude that Met-Life’s determination was not arbitrary and capricious.
With his initial claim to MetLife, Rogan-ti submitted only the award itself, a news article describing the award, the statement of claim that he had submitted in the FINRA arbitration, and proof that Met-Life had, in fact, paid him $2,492,442.07. Based on these materials alone, it was not arbitrary and capricious for MetLife initially to reject Roganti’s claim on the ground that the award did not “clearly fall within the definition” of benefits-eligible compensation under the Plans. J.A. 831. As Dudas explained in her letter denying Roganti’s claim, and as Bernstein likewise emphasized in adhering to Dudas’s determination on appeal, the award itself — consistent with Roganti’s statement of claim— indicated that Roganti had asked for “unspecified compensatory damages” in addition to punitive damages, an accounting of R. Roganti & Associates’ revenues and expenses, back pay, front pay, and reimbursement for lost benefits, plus fees, expenses, and interest. J.A. 803. The award specifically stated both that it represented “compensatory damages” and that “[a]ny and all relief not specifically addressed herein” was denied. J.A. 804. *214The award itself therefore undermined Ro-ganti’s argument that it was back pay, which both Roganti and the panel apparently viewed as a separate damages category that was not specifically awarded and thus denied.
As the district court recognized, it arguably begs the question somewhat to conclude that the award was not benefits-eligible simply because it was labeled “compensatory damages.” If it were clear, for example, that the award were “compensating” Roganti only for wages he should have earned while still employed at MetLife, then the award might appropriately be considered benefits-eligible back pay in substance, regardless of what the FINRA panel labeled it. Judging only by the materials Roganti initially submitted to MetLife, however, the award easily could have been intended to “compensate” Ro-ganti not only for his -reduced pre-retirement wages, but also for wages that he would have earned had he not retired in 2005 and/or for the reduction in pension benefits traceable to his deflated pay— each of which .was a category of relief requested in Roganti’s statement of claim. Accordingly, even assuming that some of the FINRA award might have constituted back pay, it was impossible to tell how much based only on the materials Roganti submitted. It was reasonable for MetLife to determine that Roganti was not entitled to any increase in benefits at all in light of the dearth of probative evidence regarding the extent to which his benefits-eligible compensation might have increased.
Of course, MetLife was later presented with the entire arbitral record following remand from the district court.9 As noted, Bernstein reviewed this record and still found it impossible to determine the extent, if any, to which the award represented back pay. In particular, Bernstein opined that the extensive discussion during the arbitration of how Roganti’s pension benefits were calculated, combined with the fact that Roganti had asked the panel to award him damages for the decrease in his pension, suggested that “any additional pension benefits Roganti sought during the arbitration were already included in and made part of’ the award. J.A. 927. Bernstein also highlighted the fact that the award was not divided among multiple earnings years, notwithstanding the FINRA panel’s evident understanding that calculating Roganti’s pension required determining his benefits-eligible compensation on a yearly basis. The panel’s failure to divide the award among multiple years implied that the award already included Roganti’s decreased pension benefits because it suggested that the FINRA panel was not leaving it to MetLife itself to adjust Roganti’s pension.10
*215The district court faulted' Bernstein’s analysis as a “naysaying submission” that failed to provide “a rational accounting for the Award as issued.” Roganti II, 972 F.Supp.2d at 671, 672. In contrast, it found Roganti’s argument as to what the panel had done to be “coherent and logical.” Id. at 670. For the following reasons, however, we disagree with the district court’s characterization, and with its conclusion that Bernstein’s explanation of the denial of Roganti’s claim constituted an abuse of discretion.
First, Roganti’s contention that the award exclusively represents back pay requires the assumption that the arbitral panel did not include increased pension benefits in the award, even though Roganti (sometimes) specifically asked it to do so. As Bernstein explained, however, the panel heard extensive testimony about how Roganti’s pension was calculated. It is true that Roganti’s counsel, in summati.on, presented the panel with the option of simply awarding Roganti back pay and leaving the benefits calculation up to Met-Life. But had the panel taken that route (and there is nothing in the arbitral transcript to suggest that it did), one would reasonably expect the award to have made the panel’s intentions clear to MetLife by using explicit language, breaking the award out on a year-by-year basis, or both. The district court’s indication that “the Award’s inexactitude” was an impermissible basis for denying Roganti’s claim because that inexactitude was what “prompt[ed] these proceedings,” id. at 672, was misplaced. As Bernstein explained, the award’s opacity was not merely a source of uncertainty; the omission of particulars that would have been necessary to implement Roganti’s version of the award’s meaning constituted evidence against Ro-ganti’s claim.
Although Bernstein did not advance his own mathematical breakdown of the award, he was not required to do so in order to deny Roganti’s claim. The district court faulted Bernstein for “passing] on the Court’s request that [he] posit, based on the record, a concrete alternative explanation to Roganti’s for the Award.” Id. at 671. Initially, however, we have doubts as to whether the district court clearly communicated this request. Its opinion determining that remand was necessary stated that MetLife should review the record “to permit a determination to be made, in the context of the evidence offered and arguments made by both sides at the arbitration, whether or not the aviard represented back pay.” Roganti I, 2012 WL 2324476, at *8. The court suggested that MetLife might then be able to “posit an alternative explanation” for the award, id., but it did not say that doing so was the only way for MetLife permissibly to determine that Roganti had failed to establish his entitlement to benefits. Similarly, the remand order merely instructed MetLife to “conduct a close review” of the district court’s decision and the arbitral record “to determine whether or not ... the arbitrator’s award ... constituted benefits eligible compensation,” and to explain its conclusion “with clarity and in detail.” Special App’x 18.
■ Regardless, while we appreciate the able district judge’s impulse to get to the bottom of what the award represented, we must measure Bernstein’s explanation against the arbitrary-and-capricious stan*216dard, not against any and all obligations imposed by the district court. The district court relied on cases indicating that a plan administrator acts arbitrarily and capriciously when it denies a claim on the basis of evidence that is significantly weaker than the evidence supporting the claim. See Roganti II, 972 F.Supp.2d at 671 (citing McCauley, 551 F.3d at 138, and Durakovic, 609 F.3d at 140). But we think this analogy is inapt. If Roganti’s evidence taken alone was inadequate to support his claim (which MetLife rationally found that it was), then MetLife could appropriately deny his claim regardless of the strength of the evidence pointing the other way. See Jiras, 170 F.3d at 166.
For similar reasons, we place little significance on Bernstein’s failure specifically to refute Roganti’s explanation for the award. Again, it is unclear from the réc- „ ord whether Bernstein was even aware of that explanation. The district court’s first opinion mentioned that Roganti had a theory but did not say what it was. See Roganti I, 2012 WL 2324476, at *8. In reviewing Bernstein’s decision, the district court indicated that Roganti had “first articulated” his theory “during the motion to dismiss litigation,” Roganti II, 972 F.Supp.2d at 671, but as noted, the remand order instructed Bernstein only to look at the. district court’s decision and the arbi-tral record, and not at briefing or oral argument transcripts.
In any event, Roganti’s theory is undermined by the very facts that Bernstein identified. Roganti’s theory depends on the assumption that the arbitrators awarded him the difference between $1,506 million annually and his actual earnings from 2003 to 2005, and then intended for Met-Life to calculate his pension. But despite the fact that Roganti’s counsel specifically suggested that the arbitrators direct Met-Life to recalculate the pension, the award does not manifest any such intention. The silence is significant. The arbitrators’ authority to craft a remedy was not limited to Roganti’s proposed damages calculations, and Roganti’s counsel acknowledged that the arbitrators were capable of reducing pension adjustments to a lump sum themselves. As one arbitrator observed, “I have a calculator, I could do it.” J.A. 2291. As a result, the arbitrators could have arrived at a back pay figure for 2003 to 2005 that was lower than what Roganti requested and added their own pension adjustment in lump-sum form.
We also find the $19.46 discrepancy between the actual award and Roganti’s calculation more notable than the district court did. The appeal of Roganti’s theory lies solely in its power to account for the very precise amount of the panel’s award. But that power is undermined if the theory does not, in fact, predict the exact amount of the award. The district court attributed this discrepancy to the necessity for “some degree of arithmetic extrapolation” to account for “the 2005 stub year,” id. at 672 n. 8 — to the fact that Bernstein resigned in the first quarter of 2005 — but neither the court nor Roganti explained how the panel might have arrived at a number that was off by just $19.46. Given these shortcomings, we cannot say that MetLife was required to accept Roganti’s theory as adequate proof of his entitlement to increased benefits.
Finally, it warrants emphasis that Roganti failed to obtain clarification of the award from the FINRA panel within the time limit prescribed for doing so. Once that opportunity was no longer available, it effectively became impossible for anyone— ■Roganti, MetLife, or the courts — to determine definitively what the award was compensation for. Permitting Roganti to recover under these circumstances, despite the uncertainty that he could have helped *217to prevent, would provide poor incentives for future claimants in his position. Only a claimant, and not his plan administrator, will know after he receives an arbitral award whether he will claim benefits on the ground that the award increases his benefits-eligible compensation. And the plan administrator, of course, will have no reason to seek clarification of the award until a claim is actually made. If the claim is made after the deadline for seeking such clarification has passed, the administrator will be powerless to shed light on what the award represents, except by recourse to the kind of unsatisfactory hypothesizing that the district court prescribed here. From this perspective, the better course is to place the burden on the claimant to ensure that the basis for his claim is as clear as possible before it is made, and to seek clarity during the arbitration or timely afterward.
The district court recognized the peculiarity of forcing a plan administrator to comb through thousands of pages of arbitration testimony in order to reverse-engineer an explanation for an arbitral award. In the district court’s view, this was a reason to grant less deference to MetLife’s determination on remand. See Roganti II, 972 F.Supp.2d at 673-74. But that conclusion does not follow. Where an administrator has discretionary powers, courts reviewing its decisions are assessing the reasonableness of those decisions, and not “considering the issue of eligibility anew.” Pagan, 52 F.3d at 442. Among other goals, “[d]eferenee promotes efficiency by encouraging resolution of benefits disputes through internal administrative proceedings rather than costly litigation.” Conkright v. Frommert, 559 U.S. 506, 517, 130 S.Ct. 1640, 176 L.Ed.2d 469 (2010). Accordingly, a plan administrator’s decision is intended to be final — within the bounds of the highly deferential arbitrary- and-capricious standard — and not merely an input with “the potential to assist the Court” in making the ultimate determination.11 Roganti II, 972 F.Supp.2d at 673. If MetLife could not reasonably be expected to have the institutional capacity required to reconstruct the arbitrators’ intentions on remand, then its failure to perform that task to the district court’s satisfaction was not a sound basis for overturning its decision.
C.
Roganti’s argument that MetLife was operating under a conflict of interest does not affect our analysis. In Metropolitan Life Insurance Co. v. Glenn, the Supreme Court held that when a claimant demonstrates that an ERISA plan administrator with discretionary authority is subject to a conflict of interest, courts should not apply a more exacting standard 'of review, but should instead weigh the conflict “as a factor in determining whether there is an abuse of discretion.” 554 U.S. at 115, 128 S.Ct. 2343 (quoting Firestone, 489 U.S. at 115, 109 S.Ct. 948) (internal quotation marks omitted). Once a conflict of interest is identified, determining the weight that it should receive relative to other relevant factors requires a totality-of-the-circumstances approach:
[A]ny one factor will act as a tiebreaker when the other factors are closely balanced, the degree of closeness necessary *218depending upon the tiebreaking factor’s inherent or case-specific importance. The conflict of interest at issue ... should prove more important (perhaps of great importance) where circumstances suggest a higher likelihood that it affected the benefits decision, including, but not limited to, cases where an insurance company administrator has a history of biased claims administration. It should prove less important (perhaps to the vanishing point) where the administrator has taken active steps to reduce potential bias and to promote accuracy, for example, by walling off claims administrators from those interested in firm finances, or by imposing management checks that penalize inaccurate de-cisionmaking irrespective of whom the inaccuracy benefits.
Id. at 117, 128 S.Ct. 2343 (citation omitted); see also, e.g., Durakovic, 609 F.3d at 138-41; Hobson, 574 F.3d at 82-83; McCauley, 551 F.3d at 131-33.
In this case, it is undisputed that MetLife has a “categorical” conflict of interest because it both evaluates and pays claims. Durakovic, 609 F.3d at 138. We conclude, however, that this conflict should receive no weight in our assessment of whether MetLife’s denial of Roganti’s claim was arbitrary and capricious. In the district court, MetLife submitted an unrebutted affidavit from Bernstein, who averred, among other things, that Met-Life’s business and finance departments “are kept completely separate from the administration of the Plans,” that he “did not consult or consider MetLife’s or the Plans’ finances in connection with [his] determinations” in this case or discuss Ro-ganti’s claim with the business or finance departments, and that his compensation is not tied to whether he upholds or denies benefits claims. J.A. 2905-06. These are the kinds of “active steps” that, the Supreme Court suggested in Glenn, may reduce a conflict’s importance “to the vanishing point.” 554 U.S. at 117, 128 S.Ct. 2343. Nor has Roganti established that MetLife has a history of biased claims administration or provided any other reason to assign weight to MetLife’s conflict.
Moreover, we have in the past declined to assign any weight to a conflict of interest “in the absence of any evidence that the conflict actually affected the administrator’s decision.” Durakovic, 609 F.3d at 140; see Hobson, 574 F.3d at 82-83. Roganti has identified no such evidence here. It is true that a smoking gun is not always required; under certain circumstances, an irrational decision or a one-sided decisionmaking process can alone constitute sufficient evidence that the administrator’s conflict of interest actually affected the challenged decision. See, e.g., Durakovic, 609 F.3d at 140 (assigning weight to a conflict because of the “decisionmaking deficiencies” evident in the defendants’ denial of the plaintiffs claim); McCauley, 551 F.3d at 136 (inferring that a conflict had affected the defendant’s decisionmaking because nothing else explained its failure to allow the plaintiff to cure certain defects in his claim). But as discussed above, we think that MetLife’s rationale for denying Roganti’s claim — ie., that it was impossible to determine whether, or the extent to which, the FINRA award represented back pay — was not, in fact, unreasonable.12
*219Because we conclude that MetLife’s denial of Roganti’s benefits claim was not arbitrary and capricious, we do not reach MetLife’s alternative argument that the district court' erred in refusing to dismiss this action on the basis of res judicata or collateral estoppel. Additionally, in light of our conclusion that the district court’s judgment must be reversed, we reject Ro-ganti’s cross-appeal from the district court’s refusal to award him attorney’s fees under ERISA’s fee-shifting provision, 29 U.S.C. § 1132(g)(1). Assuming (purely arguendo) that a party who wins at trial but loses on appeal is eligible for a fee award under § 1132(g)(1) by virtue of achieving “some degree of success on the merits,” Hardt v. Reliance Standard Life Ins. Co., 560 U.S. 242, 245, 130 S.Ct. 2149, 176 L.Ed.2d 998 (2010) (quoting Ruckelshaus v. Sierra Club, 463 U.S. 680, 694, 103 S.Ct. 3274, 77 L.Ed.2d 938 (1983)) (internal quotation marks omitted), we detect no abuse of discretion in the district court’s conclusion, after weighing the relevant factors, see Donachie v. Liberty Life Assurance Co., 745 F.3d 41, 46-47 (2d Cir.2014), that Roganti was not entitled to such an award in this case.
CONCLUSION
We have considered Roganti’s remaining arguments and find them to be without merit. For the reasons set forth above, the portion of the district court’s judgment granting Roganti relief -under ERISA is REVERSED, and the portion of that judgment denying Roganti’s request for attorney’s fees is AFFIRMED.
. Roganti also named as defendants four Met-Life retirement plans in which he is a participant. See infra note 2 and accompanying text. In the remáinder of this opinion, we will refer to MetLife alone as the defendant for simplicity’s sake.
. The four Plans, which Roganti named as defendants in this lawsuit along with MetLife itself, are (1) the Metropolitan Life Retirement Plan for United States Employees, (2) the Savings and Investment Plan for Employees of Metropolitan Life and Participating Affiliates, (3) the MetLife Auxiliary Pension Plan, and (4) the Metropolitan Life Supplemental Auxiliary Savings and Investment Plan. The differences in the Plans’ terms are not relevant to the issues presented in this case, so we will refer to the Plans collectively in the remainder of this opinion. As noted earlier, all four of the Plans give MetLife, as plan administrator, discretionary authority over benefits decisions. See, e.g., J.A. 2856 (“Benefits will be paid under the Plan only if the Administrator, or its delegate, determines in its discretion that the applicant is entitled to them.”).
. Roganti’s counsel chose to calculate compensation through September 2010 because the date coincided with the expected end of the FINRA proceeding and with Roganti’s 61st birthday. Because Roganti argued that he would not have retired until he turned 62, he also asked for additional compensation ''during that final year of employment.
. For the sake of completeness, Roganti earned $351,000, $412,000, $549,000 and $756,000 in, respectively, 1994, 1995, 1996, and 1997.
. Attribution of benefits-eligible compensation to a particular year of employment is pertinent principally because, as previously noted, Roganti's pension (approximately $610,000 per year prior to this litigation) is based on the average of his highest five years of earnings during the fifteen years preceding his retirement. Thus, as Dudas stated, "even assuming that the Award or any part thereof would be considered Annual Compensation under the Retirement Plans in order to calculate the appropriate benefits, ... [i]n the absence of an allocation to a particular year or. over particular years it is not clear what, if any, impact on benefit accruals the Award would have.” J.A. 831.
. The district court rejected MetLife's res ju-dicata and collateral estoppel arguments on the ground that Roganti’s claim in this federal action arose after the arbitration and, therefore, that Roganti could not have had a “full and fair opportunity to litigate that issue in the prior arbitral forum.” Roganti I, 2012 WL 2324476, at *4-5.
. Some authority arguably suggests that an administrator’s interpretation of an arbitral award is more appropriately reviewed de novo. See Hogan v. Raytheon Co., 302 F.3d 854, 856 (8th Cir.2002) (holding that de novo review was appropriate where administrator’s decision turned on interpretation of divorce decree); Dial v. NFL Player Supplemental Disability Plan, 174 F.3d 606, 611 (5th Cir.1999) (prescribing de novo review for plan administrators’ interpretation of a con*211tract and a domestic relations order); see also Weil v. Ret. Plan Admin. Comm. of Terson Co., 913 F.2d 1045, 1049 (2d Cir.1990) (applying de novo review where administrator’s decision turned on a "question of law”), vacated in part on other grounds on reh’g, 933 F.2d 106 (2d Cir.1991). Here, we discern no clear reason to apply de novo review from the text of the Plans, from principles of trust law, or from ERISA’s purposes. See Conkright v. Frommert, 559 U.S. 506, 512, 130 S.Ct. 1640, 176 L.Ed.2d 469 (2010) (indicating that the appropriate standard for reviewing an administrator’s benefits determination depends on these three variables). Regardless, apart from his argument that MetLife operated under a conflict of interest, see infra pp. 217-18, Roganti does not dispute that we should review MetLife’s decision under the arbitrary- and-capricious standard. Any such argument is therefore forfeited. See Presidential Gardens Assocs. v. United States ex rel. Sec’y of *212Hous. & Urban Dev., 175 F.3d 132, 141 (2d Cir.1999).
. The nature of a claimant’s burden may depend in part on the specific terms of the plan at issue. See Gaither v. Aetna Life Ins. Co., 394 F.3d 792, 804 (10th Cir.2004); see also Hobson, 574 F.3d at 88 (holding that plan administrator could appropriately require objective medical evidence supporting disability claim where ''[s]uch a requirement is not contradicted by any provision of [the administrator's] own policy”). In this case, the parties have not called our attention to any provision in the Plans that might affect Roganti’s evidentiary burden, so we will confine our *213discussion to "the minimum standards of ERISA.” Gaither, 394 F.3d at 804.
. We express no view on whether it was appropriate for the district court to remand to MetLife with instructions to review the arbi-tral record, given that Roganti had not initially submitted the record with his claim. As noted, it may sometimes be unlawful for a plan administrator to deny a claim while ignoring evidence that the claimant failed to submit but which the administrator could readily obtain. E.g., Harrison, 773 F.3d at 22-23. MetLife was a party to the FINRA arbitration, but it is not entirely clear whether the arbitral record was readily available to Dudas and Bernstein. Because the district court's decision to remand following Met-Life’s motion to dismiss is not before us, we need not explore this issue further. We observe, however, that the district court (incorrectly) framed its inquiry on MetLife's motion to dismiss as “whether the arbitral award represented back pay,” Roganti I, 2012 WL 2324476, at *7—not whether MetLife's denial of Roganti’s claim was arbitrary and capricious, which is a distinct question.
. Roganti suggests that the FINRA panel did not have the authority to award pension benefits because the Plans were not parties to the arbitration. This argument is unavailing. An ERISA plan participant or beneficiary need *215not sue the plan itself to recover benefits, but may sue the administrator instead, see, e.g., Chapman v. ChoiceCare Long Island Term Disability Plan, 288 F.3d 506, 510 (2d Cir.2002)—which in this case was MetLife, a party to the arbitration. It is also clear from the arbitral record that no one (including Rogan-ti, who specifically sought increased pension benefits from the panel) disputed the arbitrators' authority to award such relief.
. It is for this same reason that we held, in Miller v. United Welfare Fund, that "a district court’s review under the arbitrary and capricious standard is limited to the administrative record.” 72 F.3d at 1071. As we explained in that case, this rule is consistent with the absence of evidence that Congress " 'intended that federal district courts would function as substitute plan administrators’ and with the ERISA 'goal of prompt resolution of claims by the fiduciary.’ ” Id. (quoting Perry v. Simplicity Eng’g, 900 F.2d 963, 966 (6th Cir.1990)).
. Roganti also suggests that decreased deference is appropriate because MetLife denied his claim three times, not just once. He cites Conkright v. Frommert, in which the Supreme Court held that arbitrary-and-capricious review does not become de novo even when a Court of Appeals has already determined that the plan administrator's claim denial was erroneous and reviews the administrator’s decision again following remand. 559 U.S. at *219513, 130 S.Ct. 1640. Conkright would not support Roganti’s position even if we had previously found MetLife’s decision erroneous. In any event, the correctness of Met-Life’s decision is the very question before us in this appeal. It would make no sense to afford that decision decreased deference simply because the district court held, in the order under review, that MetLife’s decision was erroneous.