KLEINFELD, J., delivered the opinion of the Court, which is joined in full by Chief Judge HUG, Judge SNEED, Judge KOZINSKI, and Judge TROTT. Parts I and II are additionally joined by Judge FLETCHER and Judge SILVERMAN. Part I is additionally joined by Judge REINHARDT. Parts II, III, and IV are additionally joined by Judge O’SCANNLAIN.

Opinion by Judge KLEINFELD; Partial Concurrence and Partial Dissent by Judge FLETCHER; Concurrence by Judge REINHARDT; Partial Concurrence and Partial Dissent by Judge O’SCANNLAIN; Partial Concurrence and Partial Dissent by Judge SILVERMAN; Dissent by Judge FERNANDEZ.

This is an ERISA case. We voted to rehear it en banc to reconcile our decisions on the district court standard of review, whether de novo or abuse of discretion, of a plan administrator’s decision. We also consider what record that the district court should consider.

Facts.

Mr. Kearney was a trial lawyer, and the managing partner of his law firm. As part of its benefits package, the firm bought a group disability insurance policy from Standard Insurance Company. The law firm was the policy owner. Under the policy, Standard promised to pay a percentage of an employee’s predisability earnings, if the employee became disabled. For attorneys, the policy definition of “disability” says “[y]ou are only required to be DISABLED from your specialty in the practice of law.”

One day Mr. Kearney was in a judge’s chambers with opposing counsel, when the judge told him he “looked like hell.” When opposing counsel gave him exhibits to look at, he had trouble focusing and felt faint. Mr. Kearney was then 54 years old, with a history of heart trouble, so he went immediately to his physician. He had had a heart attack eleven years earlier, and an angioplasty four years earlier. This time, he was diagnosed with unstable angina pectoris, which basically means chest pain caused by inadequate oxygen supply to the heart. The arteries serving his heart were diseased and partially blocked. He was admitted to the hospital and coronary bypass surgery was performed, grafting in new blood vessels.

The parties disagree about how well Mr. Kearney recovered after surgery. The insurer takes the position that Mr. Kearney has recovered fully enough to practice in his specialty again, but Mr. Kearney’s position is that he has not. Mr. Kearney returned to his law practice for a while, then retired. Mr. Kearney takes the position that since his surgery, fatigue, exhaustion, and memory and concentration problems, prevent him from practicing as a trial lawyer.

Mr. Kearney applied for disability benefits. Standard paid them for two years. Then after obtaining more medical information from Mr. Kearney and from physicians it consulted, the insurer took the position that Mr. Kearney was no longer disabled, and quit paying him.

Mr. Kearney asked Standard to review its denial. It did so and reached the same conclusion. A “Quality Assurance Specialist” in the insurer’s “Group Quality Assurance Unit,” which reviewed the denial, wrote him that the “Quality Assurance Unit” performed “an independent review conducted separately from the individuals who made the original claim determination.” The thorough letter reviews the medical evidence in three-and-a-half single-spaced pages. It concludes that Mr. Kearney’s heart and brain both test out satisfactorily on objective measures, and his lifestyle (playing several sets of tennis every weekend and racing cars at speeds of up to 120 miles per hour) is inconsistent with his claimed inability to perform the functions of a trial lawyer.

Mr. Kearney sued Standard under 29 U.S.C. § 1132 (a)(1)(B) for benefits. The statute cited is the provision of ERISA providing for civil actions to recover benefits under an ERISA plan. The parties filed cross motions for summary judgment. The insurer argued that it was entitled to deferential review for abuse of discretion, and that review should be confined to what it called the “administrative record,” that is, the papers the insurer had when it denied the claim. The court determined that review should be de novo, because the policy was ambiguous about whether discretion was conferred. The court further determined that review should be confined to what the insurer had before it, because Mr. Kearney had had sufficient opportunity to provide evidence to the insurer. On the substantive question of whether Mr. *1087 Kearney was disabled, the dispute boiled down to whether his memory and intelligence, and his ability to work very hard and bear stress, had so deteriorated, that he could not function effectively as a trial lawyer. The district court concluded that Mr. Kearney’s IQ of 130, his playing several sets of tennis every weekend, his car-racing up to 120 miles per hour about ten times a year, and medical opinion that he ought to be able to return to work, left no genuine issue of material fact about whether he was disabled. Though his physical and mental stamina were reduced, the court granted summary judgment to the insurer because they were not so reduced as to disable him from practicing law.

Mr. Kearney appealed. The insurer argued that under Snow v. Standard Ins. Co., 87 F.3d 327 (9th Cir.1996), the policy vested discretion in itself as the administrator, so district court review was limited to abuse of discretion. The panel decision, rejecting that argument, said, “[w]e have never held that so imprecise and ambiguous a provision as contained in Kearney’s policy vests discretion in the administrator, and we decline to do so now.” Kearney v. Standard Ins. Co., 144 F.3d 597, 605 (9th Cir.), withdrawn, 152 F.3d 1098 (9th Cir.1998). But Standard’s brief had pointed out that the policy “contains exactly the same language” as the policy in Snow . The petition for rehearing pointed out that the disability policy construed in Snow , in which we had reached the opposite result, involved a policy from the same insurance company containing identical language. We rehear this case en banc in order to eliminate the conflict between these two decisions that construed identical policy language. .

Analysis.

The parties, and the panel opinion, have assumed that the insurer was an “administrator” for purposes of ERISA. Because the question whether the insurer is an administrator has not been disputed in district court or in the briefs, we assume for purposes of discussion that it is, although the characterization is not without doubt. 1

I. Standard of review.

The insurer argues that, as administrator, it is entitled to deferential review, limited to whether it abused its discretion based on the materials it had before it. The district court concluded that its review *1088 should be de novo, without deference to Standard’s decision.

The policy says that Standard will pay disability benefits “upon receipt of satisfactory written proof that you have become DISABLED.” Standard argues that the word “satisfactory” implies discretion in Standard to decide whether the claimant really is disabled within the policy definition based on the proof submitted. Therefore, Standard argues, if the insurer reasonably exercised discretion to deny a claim, a court cannot substitute its own judgment, or consider other proof, and grant the claim.

The statute does not say how courts are supposed to review administrators’ claim denials. We therefore begin our analysis with Firestone Tire & Rubber Co. v. Bruch, 489 U.S. 101 , 109 S.Ct. 948 , 103 L.Ed.2d 80 (1989), which does. In Firestone, a question arose whether employees of plants sold by Firestone to Occidental Petroleum, who were rehired by Occidental, were entitled to severance pay under Firestone’s ERISA plan. Firestone interpreted the plan to mean that they were not.

The Supreme Court granted certiorari in part to “address the appropriate standard of judicial review of benefit determinations by fiduciaries or plan administrators under ERISA.” Id. at 105 , 109 S.Ct. 948 . Firestone argued that because ERISA defined a fiduciary as one who “exercises any discretionary authority,” 29 U.S.C. § 1002 (21)(A)(i), it could be inferred that, as a fiduciary, they had discretion and their decisions could be reviewed only for arbitrariness and capriciousness. A number of circuits had adopted the arbitrary and capricious standard. But the Supreme Court rejected it. It held that because there was “no evidence that under Firestone’s termination pay plan the administrator has the power to construe uncertain terms or that eligibility determinations are to be given deference,” judicial review had to proceed without any deference to Firestone’s determination. Firestone, 489 U.S. at 111 , 109 S.Ct. 948 .

Noting that “ERISA abounds with the language and terminology of trust law,” the Court used the Restatement (Second) of Trusts and other trust law authorities to apply “[tjrust principles.” Firestone, 489 U.S. at 110-11 , 109 S.Ct. 948 . Firestone establishes that “[a]s they do with contractual provisions, courts construe terms in trust agreements without deferring to either party’s interpretation.” Id. at 112 , 109 S.Ct. 948 . That means that except where the plan gives the trustee discretion to interpret the terms of its trust, courts review trust provisions de novo, which is to say they decide for themselves what a term of the trust means instead of deciding whether the plan administrator was reasonable in how it construed the term. “ ‘The extent of the duties and powers of a trustee is determined by the rules of law that are applicable to the situation, and not the rules that the trustee or his attorney believes to be applicable, and by the terms of the trust as the court may interpret them, and not as they may be interpreted by the trustee himself or by his attorney.’ ” Firestone, 489 U.S. at 112 , 109 S.Ct. 948 (quoting 3 W. Fratcher, Scott on Trusts § 201, at 221) (emphasis added by Firestone).

Firestone holds that a deferential standard of review for actions by the trustee is “appropriate when the trustee exercises discretionary powers.” Id. at 111 , 109 S.Ct. 948 . Though the Court cites Restatement (Second) of Trusts § 187 for this proposition, the Court puts a slightly different gloss on the proposition than does the Restatement. The Restatement says that “exercise of a power is discretionary except to the extent to which its exercise is required by the terms of the trust or by the principles of law applicable to the duties of trustees.” Restatement (Second) of Trusts § 187 comment a (1959). Thus under the Restatement, in default of anything to the contrary, the trustee has discretion in the exercise of the *1089 powers the trust confers. But the Court holds in Firestone that “denial of benefits challenged under [29 U.S.C.] § 1132(a)(1)(B) is to be reviewed under a de novo standard unless the benefit plan gives the administrator or fiduciary discretionary authority to determine eligibility for benefits or to construe the terms of the plan.” Firestone, 489 U.S. at 115 , 109 S.Ct. 948 . That means the default is that the administrator has no discretion, and the administrator has to show that the plan gives it discretionary authority in order to get any judicial deference to its decision.

Thus our task now is to examine the instrument to determine whether it confers discretion on Standard to decide whether a claimant is disabled. The language Standard relies on for its argument that the instrument confers discretion is the word “satisfactory” in the requirement in the insuring clause that a claimant provide “satisfactory written proof’ of disability:

A majority of us have concluded that the phrase “will pay the LTD BENEFIT described in Part 8 upon receipt of satisfactory written proof that you have become DISABLED” is ambiguous. At least three interpretations are reasonable.

One reading natural to a lawyer experienced in insurance litigation would be that “satisfactory written proof’ is a variant of the very old phrase “satisfactory proof of loss.” Such phrases have been used in insurance policies for at least a century. The phrase traditionally confers discretion on the insurance company to decide whether the quantum of proof is sufficient, as opposed to whether.the loss is covered. The word “satisfactory” is traditionally limited by an objective standard, so that the insurance company is not permitted to reject proof that would be satisfactory to a reasonable person. See Charles R. Elliott, A Treatise on the Law of Insurance 319 (1907); William R. Vance, Handbook on the Law of Insurance 897 (Buist M. Anderson ed., 3d ed.,1951) (1904); 13A Couch on Insurance § 49A:27 (2d ed. rev. 1982); 3 Insurance Law and Practice § 1443 (John Alan Appleman & Jean Ap-pleman, rev. ed.1967).

A second reading of “will pay ... upon receipt of satisfactory written proof that you have become DISABLED” applies “satisfactory” to whether the proof establishes a covered disability as well as whether there is enough of it, but construes the word “satisfactory” as objective rather than subjective. Thus “satisfactory written proof that you have become disabled” means “proof that would be satisfactory to a reasonable person that you have become disabled.” Though the Sixth Circuit rejected this reading, Judge Boggs in dissent urged it persuasively. See Perez v. Aetna Life Ins. Co., 150 F.3d 550, 559 (6th Cir.1998) (en banc) (Boggs, J., dissenting). This reading is consistent with well established common law principles for reading contracts, which are the principles that we are required, under Firestone , to apply. As a matter of common law, where a contract contains a condition that the obligor be “satisfied,” “an interpretation is preferred under which the condition occurs if ... a reasonable person in the position of the obligor would be satisfied.” Restatement (Second) of Contracts § 228 (1981). There is an exception to this objective construction, where the subject matter can only be subjectively and not objectively satisfactory (e.g., “if you paint a portrait of my daughter with which I am entirely satisfied, I’ll pay you $5,000 for it”), id. at illustration 4, but satisfactory proof of disability is not- inherently subjective, so the exception does not apply. This objective reading subjects the obligor to a *1090 more stringent test than that his dissatisfaction must be genuine and in accord with the duty of good faith and fair dealing. Even honest dissatisfaction will not avoid his obligation, if a reasonable person in his position would be satisfied. Id. at comments a & b.

A third reasonable construction is the one articulated in the dissent, that “will pay ... upon receipt of satisfactory written proof that you have become DISABLED” means that, subject to its fiduciary duty, 2 if the administrator is not satisfied (and has not abused its discretion in so deciding), then it does not have to pay. The dissent’s position is in accord with Perez, 150 F.3d 550 , although the phrasing of the clause there was slightly different. We concede the syntactical reasonableness of the dissent’s reading. But the dissent goes too far, by suggesting that if anything is committed to the administrator’s discretion, then everything is. And the dissent’s reading runs contrary to the common law construction of conditions of “satisfaction.” We reject the proposition that the dissent’s is the only reasonable reading.

Only by excluding alternative readings as unreasonable could we conclude that the conferral of discretion is unambiguous. There are at least three fair readings of the phrase with quite different consequences. Thus we are compelled to conclude that, to the extent discretion is conferred, the conferral and its scope are at best ambiguous, and at worst (from Standard’s viewpoint) the phrase does not confer discretion at all because it does not say that it does.

In Bogue v. Ampex Corp., 976 F.2d 1319, 1325 (9th Cir.1992), we held that an administrator had discretion only where discretion was “unambiguously retained” by the administrator. This is consistent with the established principles that ambiguities are construed contra proferentem, and that ambiguities are construed in favor of the insured. Mongeluzo v. Baxter Travenol Disability Benefit Plan, 46 F.3d 938 , 942 (9th Cir.1995). We cannot conclude that Standard “unambiguously retained” discretion by means of the phrase “satisfactory written proof that you have become disabled,” because the phrase is subject to at least two reasonable constructions to the contrary. Thus we conclude that the district court was correct in its determination that Mr. Kearney’s claim should be reviewed de novo.

II. The Record to be Reviewed.

If a court reviews the administrator’s decision, whether de novo as here, or for abuse of discretion, the record that was before the administrator furnishes the primary basis for review. Should the district judge review anything else? Standard moved for an order that the district court review only the materials Kearney had submitted to Standard. The district judge granted the order, following our decision in Mongeluzo, 46 F.3d at 943.

In Mongeluzo, we held, following the Fourth Circuit in Quesinberry v. Life Insurance Company, 987 F.2d 1017 , 1025 (4th Cir.1993) (en banc), that the district court had discretion to allow evidence that was not before the plan administrator “only when circumstances clearly establish that additional evidence is necessary to conduct an adequate de novo review.” Mongeluzo, 46 F.3d at 944 (quoting Quesinberry, 987 F.2d at 1025) (internal quota *1091 tion marks omitted). Though we allowed consideration of additional evidence because of circumstances peculiar to that case, we emphasized that “a district court should not take additional evidence merely because someone at a later time comes up with new evidence” and “[i]n most cases” only the evidence that was before the plan administrator should be considered. Id.

On appeal, Mr. Kearney argues that the district court ought to have taken additional evidence regarding the medical relationship between his cardiac condition and his cognitive ability. But the brief does not say what new evidence, nor was any specific new evidence proposed to the district court. The argument is merely a suggestion that the door be opened to whatever new evidence might be developed. But Kearney does not argue that we should reexamine Mongeluzo. Starting the evidence over from scratch would be contrary to Mongeluzo.

In his opposition to Standard’s motion to limit the record, Mr. Kearney submitted pages of an “Occupational Outlook Handbook” published by the U.S. Department of Labor, which said that trial lawyers “need an exceptional ability to think quickly” and regularly work “under particularly heavy pressure” when a case is being tried. The district judge exercised his discretion to limit review to the evidence that was before the administrator for two reasons, that Mr. Kearney could as easily have submitted this material to Standard, and that the court did not need it to conduct an adequate de novo review.

The district judge did not abuse his discretion in so ruling. Both reasons he gave for confining the record to what was before Standard are good ones under Mongeluzo and are sensible in the circumstances of this case. No reason has been suggested why Mr. Kearney could not have submitted the pages of the Occupational Outlook Handbook to Standard. Insurance executives would likely be less familiar with what trial lawyers do, and more in need of such a source, than an experienced trial judge. A district judge would likely know, as this judge’s remarks indicated he did, what trial lawyers do, without the assistance of a secondary source such as the Occupational Outlook Handbook. The handbook pages are an apt example of “additional evidence ... not necessary for adequate review of the benefits decision,” as well as evidence that could as easily have been submitted to the administrator, which under Mongeluzo are disfavored.

Kearney argues that Standard did not provide the “full and fair review” required by 29 U.S.C. § 1133 (2) because it did not obtain the heart films Dr. Weinreb referred to in his report, and did not obtain Dr. Fulton’s test data referred to by Dr. Bittle. But Dr. Bittle wrote that he got Dr. Fulton’s data from Mr. Kearney’s own lawyer, so if Mr. Kearney thought Standard should have reviewed it, he should have sent it to them. As for Dr. Weinreb’s films, Kearney argues that Standard’s in-house medical advisor suggested that they be obtained, but the context of the suggestion was that he wanted them so he could better criticize Dr. Weinreb’s report. Mr. Kearney also argues that Standard failed to comply with the notice provision of 29 C.F.R. § 2560.503 — 1(f)(3), because it did not tell him what additional material he needed to submit in order to perfect his claim. • The argument is without force, because Kearney’s claim did not fail because he failed to submit needed evidence. It failed because Standard, having considered all the evidence, concluded that it needed no more and that Mr. Kearney was not disabled.

III. Genuine Issue of Fact.

Mr. Kearney claimed disability based on fatigue, exhaustion, and mental disabilities following his coronary bypass surgery. He wrote that in tennis, “I am totally wiped out after one set when, before, I played a minimum of six sets on a weekend, sometimes as many as ten or twelve.” Stress now caused chest discom *1092 fort and exhaustion. “I am unable to focus or concentrate as I did before, and my memory is definitely gone.”

Mr. Kearney submitted medical authorizations and other materials so that Standard could develop a claims file enabling it to make a decision. His cardiologist, Stephen L. Morrison, M.D., reported to his internist, David Lehman, M.D., in December 1992, shortly after the surgery, that “I am also concerned about his memory and I am [sic] told him that I am impressed [sic] that this occurs frequently after bypass surgery, and I am hopeful that this will also significantly improve.” A few months later, in April 1993, Dr. Morrison wrote that Mr. Kearney “has a tremendous complaint relative to memory loss that he associates with the coronary bypass surgery.” The letter says “from a cardiac standpoint he appears to be doing very well,” but “he seems very distraught and depressed” and recommended a tranquilizer. The next month, Mr. Kearney wrote to Dr. Lehman that his memory problem had grown worse.

In February of 1993, Mr. Kearney visited a neurologist, Peter C. Heublein, M.D. Dr. Heublein noted Mr. Kearney’s complaints of memory problems and history of heart surgery and of several concussions from auto accidents when Mr. Kearney was a teenager. Dr. Heublein’s impression was uncertain, because “my mental status examination now is normal,” metabolic disturbance and early Alzheimer’s disease were doubtful, and an episode of ischemia or. embolus to the brain was possible, as was memory disturbance from anxiety and depression. He recommended a brain scan and psychometric testing to clarify a diagnosis.

An investigator interviewed Mr. Kear-ney in January 1994 for Standard. Mr. Kearney told him that he was subject to exhaustion and lacking the mental sharpness and memory he once had. He was no longer practicing law actively, and had farmed out his cases. Although his psychiatrist had conducted a battery of tests and told him his memory was normal for a person his age, Mr. Kearney felt it was not up to the standards necessary to conduct trial work. Mr. Kearney was then living on $8,000 a month from Standard, $2,000 a month from other disability insurance with another insurance company, $3,000 a month from his stock portfolio and limited partnerships, interest on $200,000 in savings, state disability income of $300 a week, and a $500,000 fee he had earned as of December 1992.

In the fall of 1994, Mr. Kearney submitted to Standard a lengthy report from Irwin Weinreb, M.D., a cardiologist hired by a workers’ compensation insurer (not Standard) to evaluate Mr. Kearney’s workers’ compensation claim. Dr. Weinreb’s report focuses on whether Mr. Kearney’s medical condition was “non-industrial.” Mr. Kearney told Dr. Weinreb that he had not seen any doctor since seeing Dr. Lehman and Dr. Morrison, apparently in early 1993, a year and a half before. He told Dr. Weinreb that he had had a brain scan, which had come out normal, as did his MMPI, 3 and his psychiatrist had told him that his memory was normal for his age, but he felt his memory was now inadequate to do trial work. Dr. Weinreb opined that Mr. Kearney’s heart condition precluded him from “very heavy work and very severe emotional stress” or a “very tight time schedule,” though he “certainly could do legal work outside of trial work.” Mr. Kearney’s claimed intellectual deficit is mentioned only as history he gave to Dr. Weinreb, not as a finding by Dr. Weinreb.

*1093 Standard had Mr. Kearney tested by a psychologist, Randall B. Smith, Ph.D., in May 1994 for evidence of the intellectual and memory deficits he claimed. Mr. Kearney reported that he worked in his office perhaps half an hour per day, raced cars at 110 to 120 mph, played tennis, and exercised. On examination, there was “no evidence of any word-finding pauses, aphasia, paraphasias, or articulation problems.” Dr. Smith’s clinical impression was that Mr. Kearney was in the “superior” intelligence range. When he gave Mr. Kearney pencil and paper tests, his intelligence scored as average on at least one test, extremely high on another, and his memory as “superior” and “unimpaired.” His MMPI was normal. Dr. Smith noticed in Dr. Lehman’s records that Mr. Kearney had complained of memory problems at least once in the early 1980’s, while he was still active as a trial lawyer and long prior to his claim, and again in mid-1992 before his bypass surgery. “In essence,” Dr. Smith concluded, “on formal testing I could find no evidence of cognitive deficiencies.”

Mr. Kearney challenged Dr. Smith’s conclusion that there was nothing wrong with his intellect or memory with a letter by Robert M. Bittle, M.D., a psychiatrist and neurologist. Dr. Bittle had examined Mr. Kearney for two hours, and had looked at Dr. Smith’s report and at another psychologist’s report Mr. Kearney’s attorney gave him from a Dr. Fulton, which is not in the record. Dr. Bittle opined that Dr. Smith’s foundation for his conclusions was inadequate, because there were some additional intelligence tests he should have given. Dr. Bittle says that the “raw data” from Dr. Fulton’s tests in Dr. Bittle’s opinion showed cognitive and memory deficits. Dr. Fulton’s test data are not in the record. Dr. Bittle’s own mental-status examination showed that Mr. Kearney’s “estimated intelligence overall is superior,” but he also observed “some immediate memory and concentration problems.” Dr. Bit-tie wrote that “post-operative cognitive dysfunction and deficiencies with memory and concentration problems” are common complications from open-heart surgery, and while they usually clear up, sometimes they do not. Dr. Bittle opined that “it is highly medically probable that Mr. Kear-ney’s memory/concentration and cognitive deficits do interfere significantly with his ability to function in the highly complex arena as a trial attorney.”

Standard asked Dr. Smith to review Dr. Weinreb’s and Dr. Bittle’s letters and reassess his own opinion in light of them. Dr. Smith said he did the tests appropriate to Mr. Kearney’s description of his problems, as opposed to those he might do after a head injury, and that doing the additional tests suggested by Dr. Bittle would be like doing a leg x-ray on someone with no symptoms in his leg. Mr. Kearney scored in the 98th percentile on analytic tasks, and showed minimal memory deficit which could result from psychological factors as well as hypothesized edema or hypoxia during heart surgery. He maintained his opinion that Mr. Kearney could resume work as an attorney “even in a trial setting.” Dr. Smith noted, however, that he had not been furnished with the “raw test data” from Dr. Fulton that Dr. Bittle reported receiving from Kearney’s attorney.

Based on'this record, we conclude that there was a genuine issue of fact as to whether Kearney was disabled in the sense defined by the policy. Such evidence as Dr. Smith’s and Dr. Heublein’s could justify a reasonable trier of fact in concluding that Mr. Kearney’s memory and intellect were unimpaired, and his car racing and intense tennis playing could justify the conclusion that he could handle the mental and physical demands of trial work. On the other hand, such evidence as Dr. Bittle’s and Mr. Kearney’s could justify the conclusion that Mr. Kearney’s memory and intellect disabled him from trial work, and Dr. Weinreb’s report could justify the conclusion that Mr. Kearney *1094 could no longer handle the stress of trial work.

Because the record establishes a genuine issue of fact as to whether Mr. Kear-ney was disabled under the terms of the policy, we must reverse the summary judgment.

Mr. Kearney also argues that the district court should not have used only the “unable to perform your specialty” definition of disability, and that it should have also considered the “unable to earn more than 80%” of predisability earnings definition. Because Mr. Kearney did not challenge Standard’s use of the “unable to perform” subsection in district court until his motion to vacate the judgment, he cannot challenge it for the first time on appeal.

Mr. Kearney also challenges the attorneys’ fees award made against him pursuant to 29 U.S.C. § 1132 (g)(1). See Estate of Shockley v. Alyeska Pipeline Serv. Co., 130 F.3d 403 (9th Cir.1997). Because the district court judgment is vacated anyway, on account of the genuine issue of fact established by the record, the district court will necessarily revisit the issue of attorneys’ fees after the case is concluded and exercise its discretion in the different circumstances then existing. We accordingly need not reach the issue.

TV. Remand.

Because the summary judgment is reversed because of a genuine issue of fact, the genuine issue of fact must be resolved by trial. But there is a complexity, because this is an ERISA case. If the trial starts from scratch, and any evidence is admissible whether it was furnished to the administrator or not, then the effect of a genuine issue of fact is to change the question. Instead of de novo review testing whether the individual was entitled to benefits based on the evidence before the administrator and such other evidence as might be admissible under the restrictive rule of Mongeluzo, “review” would be converted into a trial de novo based on evidence entirely unrestricted by what had been presented to the administrator.

The statute adopts a policy “to increase the likelihood” that beneficiaries “will receive their full benefits” and “to maintain the premium costs of such system at a reasonable level.” 29 U.S.C. § 1001b(c)(3),(5). The procedure the statute requires for disputed claims includes “a reasonable opportunity to any participant whose claim for benefits has been denied for a full and fair review by the appropriate named fiduciary of the decision denying the claim.” 29 U.S.C. § 1133 (2). The Supreme Court has reminded us of “the public interest in encouraging the formation of employee benefit plans” and also “the need for prompt and fair claims settlement procedures.” Pilot Life Ins. Co. v. Dedeaux, 481 U.S. 41, 54 , 107 S.Ct. 1549 , 95 L.Ed.2d 39 (1987).

A full trial de novo in any ERISA dispute where there was a genuine issue of fact as to whether the individual qualified for a benefit would undermine these policies. Trial de novo on new evidence would be inconsistent with reviewing the administrator’s decision about whether to grant the benefit. The means that suggests itself for accomplishing trial of disputed facts, while preserving the value of the fiduciary review procedure, keeping costs and premiums down, and minimizing diversion of benefit money to litigation expense, is trial on the administrative record, in cases where the trial court does not find it necessary under Mongeluzo to consider additional evidence.

Although Rule 43(a) requires that “testimony” be taken in open court, the record should be regarded as being in the nature of exhibits, in the nature of documents, which are routinely a basis for findings of fact even though no one reads them out loud. We have affirmed bench trials on records in other cases. See Adair v. Sunwest Bank (In re Adair), 965 F.2d 777, 779 (9th Cir.1992). A majority of us conclude that, in its discretion guided by Mon- *1095 geluzo, the district court may try the case on the record that the administrator had before it. 4 This is vastly less expensive to all parties, accomplishes the policies enacted as part of the statute, and also gives significance, which would otherwise largely evaporate, to the administrator’s internal review procedure required by the statute.

Is there any point in remanding for such a trial, considering that the district judge may read exactly what he has already read, which did not persuade him that Mr. Kearney could establish even a genuine issue of fact as to whether he was disabled? A majority of us concludes that there is. 5 The district judge will be asking a different question as he reads the evidence, not whether there is a genuine issue of material fact, but instead whether Mr. Kearney is disabled within the terms of the policy. In a trial on the record, but not on summary judgment, the judge can evaluate the persuasiveness of conflicting testimony and decide which is more likely true. The difference in the questions he is asking of the material may lead the judge to read it differently. The district judge will retain discretion to decide, subject to Mongeluzo, whether, in order to answer this different question, he should take additional evidence. “[Tjhere is no such thing as ... findings of fact, on a summary judgment motion.” Thompson v. Mahre, 110 F.3d 716, 719 (9th Cir.1997). But in a bench trial on the record, the judge will have to make findings of fact under Federal Rule of Civil Procedure 52(a). The process of finding the facts “specially,” as that rule requires, sometimes leads a judge to a different conclusion from the one he would reach on a more holistic approach. Also, it completely changes our authority on review, from de novo review of summary judgment, to clearly erroneous review of findings of fact. 6 That change could be outcome determinative. Thus trial on the record, even if it consists of no more than the trial judge rereading what he has already read, and making findings of fact and conclusions of law instead of a summary judgment decision, may have real significance.

Conclusion.

Though we conclude that the district court was correct as to its scope of review, *1096 de novo, and within its discretion in limiting the scope of the evidence to what was before the administrator, we nevertheless must REVERSE because on de novo review of the summary judgment, we conclude that there is a genuine issue of material fact. Accordingly, we REMAND for trial consistent with this opinion.

No costs.

. The statute provides that the administrator is the person so designated in the instrument. 29 U.S.C. § 1002 (16)(A)(i)-(ii). The instrument (in this case the insurance policy) says in Part 6 that the policy "and the application” constitute the contract. The application designates the law firm, not Standard Insurance Company, as "plan administrator.” This is consistent with the common arrangement, whereby the employer, union, or some employer-union trust, is the administrator of an ERISA plan. The statutory default, in the absence of designation of an administrator, is that the "plan sponsor,” which is the employer in a case like this, is the administrator. 29 U.S.C. § 1002 (16) (A)(ii), 29 U.S.C. § 1002 (16)(B)(i). The ERISA plan, according to the "ERISA information and ERISA notice of your rights” published by Standard for employees of the firm said that they had the right to "have the PLAN ADMINISTRATOR review and consider any denial of your claim.” The designated administrator, Mr. Kearney’s law firm, had ceased its practice and had no employees by the time Standard issued its final denial.

A Standard house counsel's affidavit in the record says that the ERISA notice of rights quoted above contained a scrivener’s error, and "should have read” "Standard” where it said "the plan administrator.” Mr. Kearney disputes whether Standard is entitled to the benefit of what it meant to say as opposed to what it did say, and argues in his reply brief (but not his opening brief) that the law firm was the administrator, but does not argue that this bears on the standard of review issue.

We need not decide, because the briefs do not put the issue before us, how the case might proceed were the insurer not deemed to be the administrator. Compare Moran v. Aetna Life Ins. Co., 872 F.2d 296 (9th Cir.1989) with Vega v. National Life Ins. Services, Inc., 145 F.3d 673 , 677 n. 24 (5th Cir.1998). We proceed on the same basis that the parties did, that Standard was the administrator.

. Kearney also urges us to apply less deferential review because of Standard’s potential conflict of interest (it is deciding whether it owes its own money to Kearney). See Atwood v. Newmont Gold Co., 45 F.3d 1317 (9th Cir.1995); Lang v. Long-Term Disability Plan, 125 F.3d 794 (9th Cir.1997). Because we conclude that Kearney is entitled to de novo review, which gives no deference at all to Standard's decision, we do not reach the question whether he would be entitled to less deferential review were he entitled only to review for abuse of discretion. Thus we have no occasion to reconsider the validity of Atwood and Lang in light of Firestone Tire & Rubber v. Bruch, 489 U.S. 101 , 109 S.Ct. 948 , 103 L.Ed.2d 80 (1989).

. The MMPI is the abbreviation for the "Minnesota multiphasic personality disorder inventory,” defined as "[a]n empirical scale of an individual's personality based mainly on his own yes-or-no responses to a questionnaire of 550 items; designed to provide scores on all the more important personality traits and adaptations, and including special validating scales which measure the individual's test-taking aptitude and degree of frankness.” Blakiston's Gould Medical Dictionary (3d ed.1972).

. The author of this opinion writes with some trepidation about this novel form of trial. But it is more important that the law on a matter involving large numbers of transactions be clear and settled, than that each judge's personal views be articulated separately. As a practical matter, we have a majority for the view expressed in this section, and it is essential to claimants, administrators, and employers who fund ERISA plans, as well as district judges, that they have some authoritative statement of the law that they can safely apply.

. Judge Silverman’s concurrence and Judge Fernandez’s dissent suggest that a remand is a waste of time, because the district judge has already evaluated the evidence and would merely be engaging in an empty formality to reach the same conclusion. That is a reasonable concern. But we currently are divided about whether the district judge’s decision was correct on the evidence before him. Some of us think that the evidence before the administrator did not amount to satisfactory proof of disability. Some of us think that it did. This difference of opinion about whether there was satisfactory proof of disability establishes the desirability of having a trier of fact decide what the evidence proved. This kind of question, what the evidence proved as opposed to whether there was a genuine issue of fact, is ordinarily committed for decision to trial courts, not appellate courts.

. Judge Reinhardt argues that this discussion of standard of review is dicta, and suggests that the standard of review may be de novo. The reason why standard of review is discussed is that it is among the reasons why remanding for findings of fact is more than an empty formality. As for what it is, we have in numerous prior cases held that when a district judge tries a case on a written record without live testimony and makes findings of fact, "we apply the 'clearly erroneous’ rule, Fed.R.Civ.P. 52(a) in reviewing the judge’s findings.” See, e.g., Starsky v. Williams, 512 F.2d 109, 111 (9th Cir.1975); Wolfe v. United States, 798 F.2d 1241 , 1243 n. 2 (9th Cir.1986); EEOC v. Maricopa County Community College District, 736 F.2d 510, 513 (9th Cir.1984).