Opinion by Judge KLEINFELD; Dissent by Judge HALL.
ORDER
The opinion filed on March 7, 2011 is amended as follows:
At 3202 of the slip opinion, strike the following sentence:
This “any reasonable basis” test is no longer good law.
Replace with:
This “any reasonable basis” test is no longer good law when as in this case an administrator operates under a structural conflict of interest.
The amended opinion is filed concurrently with this Order. With this amendment, Judges Kleinfeld and Silverman voted to deny the petition for hearing. Judge Silverman voted to deny the petition for rehearing en banc, and Judge Kleinfeld has recommended the same.
The full court has been advised of the petition for rehearing and no judge of the court has requested a vote on the petition for rehearing en banc. Fed. R.App. P. 35.
[668]*668The petition for rehearing and petition for rehearing en banc is DENIED. No further petitions for rehearing or petitions for rehearing en banc will be entertained.
OPINION
KLEINFELD, Circuit Judge:
We address the standard for overturning an ERISA plan decision, and why the challenger met it.
I. Facts.
Samuel Salomaa worked for American Honda Motor Company, Inc. for more than twenty years. His supervisor described him as “without a doubt the best employee to have worked for me” in her 15 years at Honda. He was never out sick, and never left work early or came in late. At age 47, Salomaa was a dedicated family man to his wife and daughter, and an exercise enthusiast who jogged two miles to and from work every day and enjoyed playing tennis with his wife.
But in October 2003, Salomaa fell ill with what he thought was a stomach flu that made him miss three days of work. He was never the same again. He returned to work, but was tired all of the time, and had difficulty concentrating. His supervisor noted that Salomaa “walked more slowly,” and co-workers asked her about Salomaa’s well-being. Not only did Salomaa no longer jog to work, he did not even walk to work. After work he was completely exhausted, and spent weekends in bed recovering.
Salomaa went to Kaiser Permanente to find out what was wrong with him and get it cured. His complaint was grossly excessive fatigue, beginning when he had his “flu,” along with other symptoms, such as headache, insomnia, and excessive sensitivity to stimuli. His doctors went through a lengthy process of ruling out alternatives to chronic fatigue syndrome.
Over the following months, Salomaa’s Kaiser Permanente physicians worked on a diagnosis. He had reported loss of libido, and a blood test showed low testosterone, but a subsequent blood test was normal, so low testosterone was ruled out as an explanation. An MRI showed no brain abnormalities. The thyroid reading on his blood tests were normal. Heart failure might explain severe fatigue, and an echo-cardiogram showed mild mitral regurgitation, but the examining cardiologist ruled out a heart problem as the cause of the fatigue.
A Kaiser Permanente psychiatrist formed what she called a “working diagnosis” of “atypical depression.” The depression was “atypical” in that Salomaa had no previous psychiatric history, could precisely identify the onset of his fatigue following his October 2003 flu, had no “precipitating stressors” that might have triggered the depression, and denied feeling depressed. The psychiatrist tried treating Salomaa with various anti-depressants and a counseling program. In July 2004, Salomaa took medical leave, based on his doctor’s diagnosis of depression and anxiety. After several months, it was clear that the medication and counseling were not working, so the physician who had made the working diagnosis of depression rejected the diagnosis.
Salomaa’s condition got worse instead of better. Some days, getting up and getting dressed left him too exhausted to drive the two miles to his job, so he stayed home. When he did go to work, he could not do his job as well as he had been before his illness. His supervisor reported that on bad days he seemed confused, and she often insisted that he go home to rest. When he came home from work, he went straight to bed, even eating dinner there. [669]*669In spite of his fatigue, Salomaa also had insomnia.
A physician in the internal medicine department at Kaiser Permanente, Dr. Floyd Anderson, diagnosed chronic fatigue syndrome. He noted the ineffectiveness of various medications that had been tried for other conditions that might explain the symptoms. He wrote on March 4, 2005 that “since beginning our Kaiser Permanente Chronic Fatigue/Fibromyalgia Clinic in 1992, Mr. Salomaa is one of the more severe patients that I have seen in the clinic as far as his energy level. He is probably the most sensitive patient I’ve seen in regard to sensitivity to sound. His memory has also markedly decreased secondary to his illness. Mr. Salomaa is totally disabled and would not be able to work even 30 minutes per day on a daily basis.” The psychiatrist who had tentatively diagnosed depression wrote to the plan administrator concurring in Dr. Anderson’s diagnosis, and stating that Salomaa had “never suffered from Major Depression though that was [her] working diagnosis for several months.”
Salomaa applied to Honda’s ERISA plan administrator for long-term medical disability benefits.2 The claim manager denied his claim on April 22, 2005. She wrote that Salomaa had no positive objective physical findings, the lack of objective physical findings apparently forming the basis for the denial. She noted that Salomaa’s “thyroid, calcium, albumin, serum electrolytes, and CBC results were normal.” Contrary to her inference that Salomaa was healthy, the Kaiser Permanente physicians had used these normal results to rule out alternatives to chronic fatigue syndrome. She erred in some respects, suggesting a less than careful examination of Salomaa’s medical record. For example, she wrote that he had “no fevers or weight loss,” but actually he had lost 14% of his body weight in six months according to the medical materials that had been submitted. She misunderstood the Kaiser Permanente evaluation that “you never had major depression” as meaning that “your depression has improved.” In the denial letter, she relied on review by “our medical department,” by which she meant that one physician had read Salomaa’s medical file and written his opinion.
The denial had invited supplementation within thirty days, so Dr. Anderson provided more details. The disability claim manager had provided a form for Salomaa’s physician to use to check off physical abilities, and Dr. Anderson checked “occasionally,” the lowest level allowed on the form, for sitting, standing, walking, grasping, and carrying objects. He wrote a letter as well, stating that Salomaa had severe fatigue, and was “only able to do paperwork for a few minutes and then is very fatigued.” Dr. Anderson explained in his letter that patients with chronic fatigue syndrome have good days and bad days, and that on a good day the patient might be able to perform the activities listed on the form for an hour or two, but then end up in bed for several days due to overexertion. In Salomaa’s case, Dr. Anderson’s letter opined that “Salomaa would not be able to work perhaps 30 minutes to one hour” and that, “[e]ven this, if he happened to overexert, would leave him exhausted.” “Since beginning our clinic here [670]*670in 1992, Mr. Salomaa is one of the more severe cases I have seen” and that Salomaa “definitely could not work.” He pointed out to the disability claim manager that “laboratory tests are always normal and there is no test that is available at the present time for chronic fatigue syndrome.” The plan administrator’s reviewing physician called Dr. Anderson on the phone, and Dr. Anderson reminded him that Salomaa had chronic fatigue syndrome, and that Salomaa’s lack of positive laboratory findings was consistent with that diagnosis.
The disability claim manager sent out a final denial on May 20, 2005. She again recited the absence of positive laboratory results or physical findings, again made the error on weight loss and depression, and noted that Salomaa’s daily activities exceeded Dr. Anderson’s estimations. She pointed out that his daily journals had showed him driving a half-hour to an hour to Home Depot and an hour to pick up his children at school, both taking longer than the half-hour that his doctor said was the most he could work.
Dr. Anderson responded that since Salomaa’s last two visits in May and June, he thought Salomaa’s condition had markedly deteriorated, to where he could no longer work even five minutes per day. Responding to the disability claim manager’s argument that Salomaa had “no physical findings to support chronic fatigue syndrome,” he wrote that in his experience, “most patients’ symptoms and physical findings manifest when they initially develop the viral-type illness,” as Salomaa’s had. He pointed out that usually there were no physical findings for chronic fatigue syndrome except that the patient looked fatigued, just as there were no physical symptoms for migraine headache except that the patient would appear to be in pain.
Rebutting the plan administrator’s contention, Dr. Anderson wrote that on his long trip to Home Depot, Salomaa had gone with his brother-in-law, “and was too fatigued to go into Home Depot and consequently sat in the car the entire time.” He had picked up his children at school only twice, in January, and had been unable to do since. Refuting the administrator’s claim of “no appearance of physical findings,” Dr. Anderson wrote that had been “shocked” at Salomaa’s decline from 163 to 140 pounds, perhaps because he was too fatigued to come to the table for breakfast or lunch and could only sit for five or ten minutes at dinner. Salomaa, Dr. Anderson wrote, was spending most of his time in bed, and at his last visit in June, Salomaa had been unable to sit up on the examining table without assistance and “appeared cachectic [cachexia is severe generalized weakness, malnutrition and emaciation3] and weak.”
The disability claim manager wrote that Dr. Anderson’s letter was unpersuasive because he had not mentioned various matters (many of which had not troubled the disability claim manager in earlier correspondence), and in a telephone conversation with someone he had said that Salomaa’s limitations were based on what Salomaa had said:
Dr. Anderson’s letter dated June 13, 2005 made no specific mention of substantial impairment of cognitive function, sore throat, current infection, tender lymph nodes, specific myalgia or arthralgia, or new onset of headache or post-exertion fatigue. There were no rational specific limitations of physical functional capacity such as the number of hours you are able to stand or the [671]*671maximum amount of weight you are able to lift or carry. Dr. Anderson noted that you looked tired and reported weight loss but the reason behind the weight loss and its relation to specific physical functional capacity is unknown. Furthermore, there was no mention of new positive lab findings that may be supportive of the etiology of Chronic Fatigue Syndrome. There is no indication that you are on an ongoing specific treatment program including pharmacological stimulants or other measures for reported lack of energy. In a telephone conversation with Dr. Anderson on May 19, 2005, he stated that your limitations of physical functional capacity were based entirely upon self report. Under generally accepted medical standards one would not expect to see this constellation of findings in a severe case of Chronic Fatigue Syndrome that is incapacitating in terms of pain or fatigue.
Also during July, Salomaa’s short term disability benefits expired. He could no longer afford his house, so his family sold it and moved from Glendale, California to less expensive Arkansas. He could not help with the move because he was too exhausted. Nor was he strong enough to manage flying, so he lay in the backseat of a car while his brother-in-law drove.
After the May 20, 2005 final denial of disability benefits, Salomaa appealed. His attorney made a written request for Salomaa’s entire file, including correspondence with anyone the plan consulted with regard to the claim. He also asked a number of questions, such as what “positive lab findings” the plan administrator thought might be relevant to the chronic fatigue syndrome diagnosis, it being a diagnosis ordinarily given when there are no positive laboratory findings despite chronic fatigue symptoms. Salomaa’s attorney also offered in his letter to submit Salomaa for an examination by the disability plan’s selected physicians and for such laboratory tests as the plan wished to have performed.
The plan did not respond to Salomaa’s attorney’s letter. Nor did the plan ever have any physician of its choosing examine Salomaa. When Salomaa’s attorney wrote again, the disability claim manager told him that the plan had said all it was going to say, Salomaa’s claim was denied, and that Salomaa had not appealed.
Salomaa’s attorney also provided the plan with additional documentary evidence of Salomaa’s condition. Salomaa’s boss at Honda wrote that he was one of the few people in southern California to walk or jog to work, and had been a superb employee until he got sick, and that those at Honda aware of his difficulty getting his disability benefits were “appalled.” His brother-in-law wrote that he got to know Salomaa when they were students at Harvard and MIT, that Salomaa had been very energetic and intelligent, but that now he had “profound” apparent changes. For example, “any sentence containing a sequence of ideas was too difficult for him to comprehend.” While helping out with chores Salomaa had become unable to do, the brother-in-law observed that Salomaa had lost the “ability to plan” and the stamina for simple household tasks and even for conversation. He could do well for a few minutes, but then collapsed. “I drove Sam and their daughter from Los Angeles to Arkansas. For the entire trip Sam was either fully or partially reclined in the back of their Honda Element.... When stopping to eat, Sam could not manage to eat in the restaurant. I would go in to order the food, and then bring it to Sam to eat in the car.”
Salomaa’s attorney also forwarded a neuropsychological evaluation from New Jersey Medical School. Two neuropsychologists reported on a battery of intelli[672]*672gence and other tests they had administered. They included a standard test for malingering, which showed “a valid profile and that he was putting forth adequate effort.” His intelligence tested as “average,” a full scale IQ of 108 which is fine for many people but shockingly low for a Harvard man with a career in computers. He “performed in the impaired range” when tasks became more difficult and distractors were introduced. The neuropsychologists’ conclusion was that there was no evidence of psychiatric illness, but a “neuropsychological profile consistent with reports in the literature that identify slowed processing speed and decreased mental efficiency (ability to process multiple pieces of information at one time) as the hallmark cognitive symptoms of [chronic fatigue syndrome].”
The director of the New Jersey Medical School Chronic Fatigue Syndrome/Fibromyalgia Center wrote after personally examining Salomaa that Salomaa “has underlying chronic fatigue syndrome superimposed on an extreme stress sensitivity.... In fact, I think he is one of the most disabled individuals I have seen in over 15 years of practice. I do not believe this patient can work, even on a part-time basis. Simple cognitive tasks produce a dramatic worsening of his entire symptom complex.” The dramatic symptom worsening from even minimal exertion was characteristic of chronic fatigue syndrome and was the thing disabling Salomaa.
Meanwhile, Salomaa had applied to the Social Security Administration for disability benefits. The Social Security Administration found that Salomaa was completely disabled and unable to perform any occupation in the national economy, and awarded benefits. In January 2006, Salomaa’s lawyer forwarded a copy of the SSI award to the plan.
But on February 14, 2006, the appeals claim manager for CIGNA Group Insurance, the plan administrator, wrote Salomaa’s lawyer that on appeal CIGNA affirmed its previous denial. The plan administrator’s letter stated that the medical evidence had to show that Salomaa was unable to perform his occupation from July 24, 2004 to February 20, 2005, and it did not. The plan defines “disability” as inability, because of injury or sickness, to perform, or earn 80% as much from, his regular occupation.4 The “elimination period” screens out temporary disabilities by requiring a continuous specified period of disability.5 CIGNA’s denial letter quoted from a file review it had obtained from its own consulting physician. This consulting [673]*673physician’s report, like the previous one, was not provided to Salomaa’s lawyer.
Salomaa sued the plan in district court, claiming wrongful denial of benefits.6 The district court applied an abuse of discretion standard, taking into consideration the plan’s conflict of interest, and upheld the plan administrator’s denial of benefits. The district court emphasized the degree to which Salomaa’s diagnosis depended on his own symptom reports because of the lack of objective laboratory or other findings, that the plan had taken the required steps of considering Salomaa’s evidence and consulting qualified experts, and even with the abuse of discretion standard modified on account of the insurer’s conflict of interests, that “ ‘abuse’ is still a powerful word.”
Salomaa appeals.
II. Analysis.
A. Standard of Review.
We held in Kearney v. Standard Insurance Co.7 that by default, review of denial of ERISA benefits is de novo, and that to obtain the more lenient abuse of discretion standard of review, a plan must unambiguously so provide. The plan in this case does so. It expressly and unambiguously gives the administrator discretion to determine eligibility.8 Thus, under Firestone Tire & Rubber Co. v. Bruch,9 we review the administrator’s decision for abuse of discretion, rather than de novo.
We have gradually refined and restated our standard of review. In Horan v. Kaiser Steel Retirement Plan,10 applied in Jordan v. Northrop Grumman Corp. Welfare Benefit Plan11 and our more recent decision in Sznewajs v. U.S. Bancorp,12 we held that “[a] decision is not arbitrary unless it is ‘not grounded on any reasonable basis.’ ”13 This “any reasonable ba[674]*674sis” test is no longer good law when as in this case an administrator operates under a structural conflict of interest.
The administrator of the plan before us has a conflict of interest, as the term is used in ERISA cases, because the insurer acts as both funding source and administrator.14 In our en banc decision in Abatie v. Alta Health, we held that if a plan gives discretion to an administrator operating under a conflict of interest, the “conflict must be weighed as a factor in determining whether there is an abuse of discretion.” 15 Procedural errors by the administrator are also “weighed in deciding whether the administrator’s decision was an abuse of discretion.”16 We held in Saffon v. Wells Fargo & Company Long Term Disability Plan17 that we apply different levels of skepticism on account of conflicts of interest, depending on various factors such as inconsistent reasons for denial or evidence of malice. We held that “when reviewing a discretionary denial of benefits by a plan administrator who is subject to a conflict of interest, we must determine the extent to which the conflict influenced the administrator’s decision and discount to that extent the deference we accord the administrator’s decision.”18
Subsequently, the Supreme Court issued its own refinement, superseding ours to the extent that there is any difference. Metropolitan Life Insurance Co. v. Glenn holds that where “the entity that administers the plan, such as an employer or an insurance company, both determines whether an employee is eligible for benefits and pays benefits out of its own pocket ... this dual role creates a conflict of interest; that a reviewing court should consider that conflict as a factor in determining whether the plan administrator has abused its discretion in denying benefits; and that the significance of the factor will depend upon the circumstances of the particular case.”19 Under Glenn, the conflict of interest must be “weighed as a factor” but does not convert abuse of discretion review into de novo review. The weight given the factor varies.20 The Court emphasized that its “elucidation of Firestone ’s standard does not consist of a detailed set of instructions” and, importing language from the standard of review of administrative agency decisions, “there ‘are no talismanic words that can avoid the process of judgment.’ ”21
The Supreme Court further refined the standard of review in its decision this year in Conkright v. Frommert, holding that “a single honest mistake in plan interpretation” administration does not deprive the plan of the abuse of discretion standard or justify de novo review for subsequent related interpretations.22 The Court emphasized that under Glenn, “a deferential standard of review remains appropriate even in the face of a conflict.”23 Conk-right noted, though, that “[a]pplying a deferential standard of review does not mean that the plan administrator will prevail on [675]*675the merits.”24 What deference means is that the plan administrator’s interpretation of the plan “ ‘will not be disturbed if reasonable.’ ”25
It is much easier to state the words of the formula for the standard of review than to say what the formula means in practice. We now know that the administrator’s decision cannot be disturbed if it is reasonable. And we know that even an unqualified abuse of discretion standard of review does not mean that the administrator necessarily prevails on the merits, because “no talismanic words ... can avoid the process of judgment.” We know that we are supposed to “weigh” a conflict of interest in deciding how skeptical to be of the administrator’s decision, according varying weight to it depending on other factors, but that is a hard standard to apply. “Weighing” is a metaphor. Real weighing is done with a scale. For fine work one may gradually add two gram brass weights on one side of the scale, or use the one gram slider, until the trays on both sides are level. Because this connotes careful, precise, scientifically accurate results, it is a comforting metaphor for judicial work. But unlike weighing potassium bromide and potassium ferricyanide in a traditional darkroom, our “weighing” is done without a scale, without the little brass weights, and without a substance to weigh that has any weighable mass.
Nor is it easy to decide how many metaphorical grams should go on the metaphorical scale when we pretend to weigh conflicts of interest. The misleading precision of the metaphor is indeed a serious concern, because of the special protection the statute gives to insurance companies against claims. An insurance company that approaches claims-handling unfairly in an ERISA plan may have an incentive to be more unfair than, say, a life insurer or auto-liability insurer, because it cannot be subjected to the punitive damages for bad faith that are the bogeymen of insurance companies in those fields.26 Usually the record does not disclose an insurance company’s claims-handling history in other cases or its internal directives to claims managers about how to evaluate claims. Thus we are ordinarily ignorant of much of what we are supposed to “weigh.” For all we know, the claims administrator evaluating Salomaa’s claim would be voted for promotion based on the percentage of claims rejected, or had formed a personal opinion (or her boss had) that all chronic fatigue syndrome claims were fraudulent.
Where, as in this case, the plan gives the administrator discretion, and the administrator has a conflict of interest, we are to judge its decision to deny benefits to evaluate whether it is reasonable. Reasonableness does not mean that we would make the same decision. We must judge the reasonableness of the plan administrator skeptically where, as here, the administrator has a conflict of interests. Even without the special skepticism we are to apply in cases of conflict of interest, deference to the plan administrator’s judgment does not mean that the plan prevails. “Deference” is not a “talismanic word[] that can avoid the process of judgment.”27 The conflict of interest requires additional skepticism because the plan acts as judge in its own cause.
[676]*676The meaning of “abuse of discretion” is elucidated in our en banc decision in United States v. Hinkson.28 There we held that the test for abuse of discretion in a factual determination (as opposed to legal error) is whether “we are left with a definite and firm conviction that a mistake has been committed,” and we may not merely substitute our view for that of the fact finder.29 To do so, we consider whether application of a correct legal standard was “(1) illogical, (2) implausible, or (3) without support in inferences that may be drawn from the facts in the record.”30 That standard makes sense in the ERISA context, so we apply it, with the qualification that a higher degree of skepticism is appropriate where the administrator has a conflict of interest.
B. Reasonableness of the denial.
In this case, the plan abused its discretion. Its decision was illogical, implausible, and without support in inferences that could reasonably be drawn from facts in the record, because: (1) every doctor who personally examined Salomaa concluded that he was disabled; (2) the plan administrator demanded objective tests to establish the existence of a condition for which there are no objective tests; (3) the administrator failed to consider the Social Security disability award; (4) the reasons for denial shifted as they were refuted, were largely unsupported by the medical file, and only the denial stayed constant; and (5) the plan administrator failed to engage in the required “meaningful dialogue”31 with Salomaa.
At least four physicians examined Salomaa personally, as well as two psychologists who personally administered tests of Salomaa’s cognitive processing and a test to rule out malingering. Every one of them concluded, often in dramatic language, that Salomaa was totally disabled by his physical condition. Not a single physician who actually examined Salomaa concluded otherwise. The only documents with an “M.D.” on the signature line concluding that he was not disabled were by the physicians the insurance company paid to review his file. They never saw Salomaa. Salomaa’s lawyer wrote to the plan, offering to make Salomaa available for examination by its physicians. The administrator did not even respond to this offer. Thus the plan not only did not have its physicians examine Salomaa, but also rejected the opportunity to do so. An insurance company may choose to avoid an independent medical examination because of the risk that the physicians it employs may conclude that the claimant is entitled to benefits. The skepticism we are required to apply because of the plan’s conflict of interests requires us to consider this possibility in this case. The medical record by physicians who actually examined Salomaa is entirely one sided in favor of Salomaa’s claim. The plan rejected its opportunity to see if there was another side.
The plan’s reasons for denial shifted, as old reasons, proved erroneous, were replaced by new ones. The initial denial emphasized that there were “no positive physical findings.” It mistakenly said that there was no weight loss, even though Salomaa had lost 14% of his body weight over a six-month period. Expanding upon the “no physical findings” reason, the denial said that “thyroid, calcium, albumin, [677]*677serum electrolytes and CBC results are normal.” Likewise the final denial emphasized Salomaa’s normal objective findings, and that there was “no underlying condition, such as cancer or HIV disease” to explain his fatigue or weight loss. These reasons were illogical, because such objective measures as blood tests are used to rule out alternative diseases, not to establish the existence of chronic fatigue syndrome.
There is no blood test or other objective laboratory test for chronic fatigue syndrome. As we said in Friedrich v. Intel Corp., the condition “does not have a generally accepted ‘dipstick’ test” and “[t]he standard diagnosis technique for [chronic fatigue syndrome] includes testing, comparing symptoms to a detailed Centers for Disease Control list of symptoms, excluding other possible disorders, and reviewing thoroughly the patient’s medical history.” 32 Salomaa’s physicians explained this to the plan administrator, but were evidently ignored, as was the Center for Disease Control definition provided to the administrator:
The chronic fatigue syndrome is a clinically defined condition (1^4) characterized by severe disabling fatigue and a combination of symptoms that prominently features self-reported impairments in concentration and short-term memory, sleep disturbances, and musculoskeletal pain. Diagnosis of the chronic fatigue syndrome can be made only after alternative medical and psychiatric causes of chronic fatiguing illness have been excluded. No pathognomonic signs or diagnostic tests for this condition have been validated in scientific studies (5-7); moreover no definitive treatments for it exist (8).
The plan administrator issued a “Coverage Position” paper for chronic fatigue syndrome acknowledging that there are no objective tests for it, and adopting the Center for Disease Control criteria:
The multiple symptoms of [chronic fatigue syndrome], which are seen in numerous other conditions, make it a difficult condition to diagnose. Therefore, diagnosis is made by exclusion of other conditions.
Despite extensive research, the etiology of [chronic fatigue syndrome] is unknown.
In order to receive a diagnosis of [chronic fatigue syndrome], a patient must meet the following two criteria:
• The patient must have clinically evaluated, unexplained persistent or relapsing chronic fatigue that is of new or definite onset (i.e., not lifelong), is not the result of ongoing exertion, is not substantially alleviated by rest, and results in substantial reduction in previous levels of occupational, educational, social, or personal activities.
• The patient must have concurrent occurrence of four or more of the following symptoms: substantial impairment in short-term memory or concentration; sore throat; tender lymph nodes; muscle pain; multi-joint pain without swelling or redness; headaches of a new type, pattern, or severity; unrefreshing sleep; and post-exertional malaise lasting more than 24 hours. These symptoms must have persisted or recurred during six or more consecutive months of illness and must not have predated the fatigue.
There are no specific diagnostic studies (ie., laboratory, radiography, psychosomatic or other testing) or physical find[678]*678ings that are specific to the diagnosis of [chronic fatigue syndrome]. Diagnosis of this syndrome is by exclusion of other underlying diseases, (emphasis added)
Salomaa’s medical documentation established that he met the criteria specified by the Center for Disease Control and CIGNA’s own position paper. The plan administrator never claimed that he did not. In its brief, the plan argues not that Salomaa failed to meet CIGNA’s diagnostic criteria, but that the criteria should be ignored, because the “Coverage Position” was drafted by CIGNA Health Care, a “wholly different entity” from Life Insurance Company of North America, and was issued after the final denial. The first reason is frivolous, because a different part of the brief concedes that Life Insurance Company of North America is a wholly owned subsidiary of CIGNA, and the plan administrator’s denials are all written on CIGNA stationary, stating that they come from “CIGNA Disability Management Solutions.” As for timing, the parties were still engaged in a dialogue after the final denial, and the CIGNA position is nothing new, just a restatement of the previously issued Center for Disease Control diagnostic criteria. As we said in dicta in a fibromyalgia case, “if the administrator had said, “we will not accept fibromyalgia as a diagnosis unless you present objective evidence of it such as positive findings on x-rays,’ she would have been demanding what cannot exist....”33 We now establish as holding what was then dicta, that conditioning an award on the existence of evidence that cannot exist is arbitrary and capricious.
The plan’s reasons for denial were shifting and inconsistent as well as illogical. The initial denial says that there were “no specific serial descriptions of appearance or physical signs consistent with chronic fatigue syndrome,” but the final denial omits any mention of physicians’ observations, because the physicians’ letters to CIGNA are replete with dramatic descriptions of their observations of Salomaa’s appearance and physical condition. About the only thing that stays the same from the initial denial to the final denial is the irrelevant emphasis on absence of objective evidence such as blood tests.
One can understand the frustration of disability plan administrators with claims based on such diseases as chronic fatigue syndrome and fibromyalgia. Absence of objective proof through x-rays or blood tests of the existence or nonexistence of the disease creates a risk of false claims. Claimants have an incentive to claim symptoms of a disease they do not have in order to obtain undeserved disability benefits. But the claimants are not the only ones with an incentive to cheat. The plan with a conflict of interests also has a financial incentive to cheat. Failing to pay out money owed based on a false statement of reasons for denying is cheating, every bit as much as making a false claim. The plan has no exception to coverage for chronic fatigue syndrome, so CIGNA has taken on the risk of false claims for this difficult to diagnose condition. Many medical conditions depend for their diagnosis on patient reports of pain or other symptoms, and some cannot be objectively established until autopsy. In neither case can a disability insurer condition coverage on proof by objective indicators such as blood tests where the condition is recognized yet no such proof is possible.
[679]*679The Social Security Administration was persuaded that Salomaa was indeed unable to work at any job in the national economy, and awarded disability benefits to him. Its award was provided to the plan administrator. Amazingly, the plan’s initial and final denial letters do not even mention the Social Security award. Social Security disability awards do not bind plan administrators,34 but they are evidence of disability. Evidence of a Social Security award of disability benefits is of sufficient significance that failure to address it offers support that the plan administrator’s denial was arbitrary, an abuse of discretion.35 Weighty evidence may ultimately be unpersuasive, but it cannot be ignored.
The continual shifting of the plan’s grounds for denial also suggest abuse of discretion. The initial denial gave absence of weight loss as a reason. The final denial gave absence of cancer or HIY disease as a reason to explain the previously ignored 14% weight loss. The initial denial says “no major depression” and “your depression has improved,” the final denial does not mention depression (perhaps because ruling out depression by administering serotonin reuptake inhibitors without effect was among the grounds for the chronic fatigue syndrome diagnosis). The initial denial notes Salomaa’s physician’s report of “decreased memory” but absence of “formal mental status tests to quantify any specific abnormalities of cognitive functioning,” the final denial says that even though the specific cognitive testing “shows some abnormalities, it does not support that those deficits were present during the elimination period” (even though, in combination with the consistent symptom reports and observations of others during the elimination period, the tests did indeed tend to show that the deficits were present throughout the elimination period).
The plan also failed to conform to the claims procedure required by statute and regulation. The statute entitles the claimant to “full and fair” review of a denial.36 The regulations contain many requirements that the plan failed to meet, among them that it explain, upon denial, any “additional information needed,”37 and that it give the claimant “reasonable access to, and copies of all, documents, records, and other information relevant to the claimant’s claim for benefits.”38 The review was not “fair,” as the statute requires, because the plan did not give Salomaa and his attorney and physicians access to the two medical reports of its own physicians upon which it relied, among other reasons. In addition, the plan administrator denied the claim largely on account of absence of objective medical evidence, yet failed to tell Salomaa what medical evidence it wanted. Where it did tell him, such as “no formal clinical mental status tests,” Salomaa provided the evidence. The initial denial said he should provide “x-rays, CT, MRI reports, etc. that support your physician’s assessment,” but did not tell him [680]*680what x-rays etc. it wanted. The request was of course absurd, since x-rays, computerized tomography, and magnetic resonance imaging are not used to diagnose chronic fatigue syndrome. A layman might be fooled by this statement of reasons into thinking he left something relevant out of his claim package, but fooling someone unfamiliar with the medical terms with irrelevant medical mumbo jumbo violates the statutory duty to write a denial “in a manner calculated to be understood by the claimant.”39
The plan evidently based its denial in large part on review of Salomaa’s file by two physicians, one for the first denial, another for the final denial. They both wrote their appraisals for the plan administrator. Yet the plan failed to furnish their letters to Salomaa or his lawyer. The regulation, quoted above, requires an ERISA plan to furnish “all documents, records, and other information relevant for benefits to the claimant.”40 A physician’s evaluation provided to the plan administrator falls squarely within this disclosure requirement.41 The disclosure requirement serves the purpose of facilitating what the regulation also requires, providing claimants “the opportunity to submit written comments, documents, records, and other information relating to the claim for benefits.”42 Had the plan met its duty of providing copies of its physicians’ evaluations, then Salomaa’s treating physicians could have provided such comments and performed such additional examinations and tests as might be appropriate. By denying Salomaa the disclosure and fair opportunity for comment, the plan denied him the statutory obligation of a fair review procedure.
The administrator’s procedural violations are similar to those in Saffon v. Wells Fargo & Company Long Term Disability Plan43 and Booton v. Lockheed, Medical Benefit Plan.44 There, as here, the administrator did not provide material sufficient to meet the requirement of “meaningful dialogue.” We held in those cases, where the denials were based on absence of some sort of medical evidence or explanation, that the administrator was obligated to say in plain language what additional evidence it needed and what questions it needed answered in time so that the additional material could be provided. An administrator does not do its duty under the statute and regulations by saying merely “we are not persuaded” or “your evidence is insufficient.” Nor does it do its duty by elaborating upon its negative answer with meaningless medical mumbo jumbo. In this case, the skeptical look required by us in a case of a conflicted administrator requires us to conclude that the administrator acted arbitrarily and capriciously, both procedurally and substantively, thereby abusing its discretion in the denial of Salomaa’s claim.
III. Conclusion.
The plan violated its procedural obligations and violated its substantive obligation by abusing its discretion and judg[681]*681ing the disability claim arbitrarily and capriciously. Our skepticism of its approach is heightened because of its conflict of interests. Where the plan administrator has a conflict of interests, review for abuse of discretion is not as deferential as abuse of discretion review of district court or administrative agency decisions. On the record it had, the plan was obligated to award benefits. Accordingly, we REVERSE and REMAND with instructions to direct an award of benefits.