When this case between an asbestos manufacturer and its insurance companies was first before us five years ago, we decided that insurance coverage for asbes*231tos-related diseases was triggered when a disease “becomes reasonably capable of medical diagnosis,” Eagle-Picher Industries v. Liberty Mutual Insurance Co., 682 F.2d 12, 25 (1st Cir.1982) (“Eagle-Picker I”).1 The intervening years have shown that making a determination of diagnosability is no easy task. The case has returned to us because Eagle-Picher, the manufacturer of products containing asbestos, and one of its insurers, American Motorists Insurance Co. (“AMICO”), continue to urge sharply conflicting methods for determining when an individual’s disease became diagnosable. Each party contends that its method is the most consistent with our earlier judgment, and each insists that the other’s approach is seriously flawed. The district court accepted Eagle-Picher’s method, and AMICO has appealed. In a cross-appeal, Eagle-Picher challenges a certification requirement in the district court’s final order, and also claims that the district court erred in failing to award it damages for bad faith and prejudgment interest.
I.
Eagle-Picher has been named as a defendant in more than 25,000 lawsuits alleging bodily injury or death resulting from exposure to products containing asbestos. Between 1968 and 1979, the company was covered for the losses from these suits by numerous insurance policies provided by several different carriers, including AMI-CO. AMICO provided “excess coverage” of up to $5 million per policy for three separate policies from June 1,1973 to October 10, 1975. As a first-level excess insurer, AMICO is responsible for claims, up to its policy limit, once Eagle-Picher’s primary insurance coverage is exhausted. This case began after the primary insurer, Liberty Mutual Insurance Co., notified EaglePicher in 1977 that its policy limits for 1974 and 1975 were about to be reached. When informed that their policies were about to be triggered, the excess insurers, including AMICO, questioned whether there had, in fact, been exhaustion of the primary policy under the proper theory of insurability. To resolve the conflict, Eagle-Picher brought an action seeking a declaration of the rights and liabilities of its various insurers, which led to our decision five years ago adopting the “manifestation” theory of coverage. Eagle-Picher I, 682 F.2d at 24.
Following the 1982 decision linking insurance coverage to the date of diagnosability, Eagle-Picher asked five pulmonary medicine specialists when an asbestos-related disease was capable of diagnosis. Each opined that such disease was likely to be diagnosable about five years prior to the actual date of diagnosis. At that point, the insurers still declined to settle. Eagle-Picher then commissioned Dr. Hans Weill, an expert on asbestos-related disease, to study the question of diagnosability, and his conclusion was that diagnosability more likely than not preceded diagnosis by about six years.2 After completion of the Weill study, all insurers but AMICO agreed to settle on the basis of what has been called a five-year coverage “rollback,” meaning that the date of diagnosability — and thus the trigger of insurance coverage — is deemed to be the date five years before the actual diagnosis date.3 AMICO has steadfastly maintained that a blanket rollback of *232any number of years is inconsistent with Eagle-Picher I, and argues that a determination of when individual claimants’ diseases were capable of diagnosis can be made only by means of a case-by-case review of the medical data in their files. AMICO did, however, enter into an interim financing agreement with Eagle-Picher under which AMICO agreed, for the limited purpose of the interim agreement, that asbestos-related claims were capable of medical diagnosis one year prior to the actual diagnosis.4 Under this agreement, AMICO has so far paid Eagle-Picher $2.2 million.
The continuing stalemate between EaglePicher and AMICO on the issue of diagnosability led Eagle-Picher to file a motion for further relief in the district court in December 1983. The motion asked the court to compel AMICO to comply with the Weill study’s six-year rollback, and also sought an immediate money judgment, with pre- and post-judgment interest, as well as bad faith and punitive damages, and attorney’s fees. After nine days of hearings in July 1984, supplemented by written testimony and many exhibits, the district court endorsed the six-year rollback as preferable to AMICO’s alternative proposal of an individualized file review.5 In its memorandum of decision filed May 30, 1986, the court also concluded that Eagle-Picher was not entitled to a money judgment or bad faith damages. This appeal resulted.
II.
A.
For anyone to resolve the primary question before us — when should an asbestos-related disease be deemed capable of diagnosis — seems surely to require the wisdom of Solomon. This is so for at least two reasons. First, as the district court noted, the evidence adduced at trial indicates that the question of when any particular claimant’s disease was “reasonably capable of medical diagnosis” is unanswerable. The best that can be hoped for is an informed estimate.6 Second, the two alternative methods proposed by the parties for making this “guestimate” — one a “macro,” the other a “micro” approach — both appear to have significant shortcomings. Eagle-Picher’s proposal of an automatic six-year rollback from the date of diagnosis is based on statistical probabilities and would likely prove accurate in only six percent of the cases to which it is applied. Tr. 3-157. The study on which the proposal was based drew broad conclusions from a small sample and was based entirely on x-ray readings that, from a clinical viewpoint, tell little about whether an individual has an *233asbestos-related disease.7 The study is applicable only to asbestosis, so the six-year rollback result drawn from it has no applicability to the other diseases suffered by Eagle-Picher claimants, including lung cancer and mesothelioma. The Eagle-Picher approach would ignore all medical evidence in the individual claimants’ files despite this court’s prior recognition that “the policies are geared to the injuries of a particular claimant,” 682 F.2d at 25 n. 12.
The AMICO proposal, which is labeled a “Draft Protocol,” would require EaglePicher to examine each claimant’s file to look for evidence of diagnosability, and would include consideration of such factors as the type of disease claimed, the length of exposure to asbestos, the claimant’s symptoms and the results of laboratory tests. The AMICO proposal consists of two parts: (1) a list of the factors on which a diagnosis could be based, and (2) a list of combinations of those factors that would support “an insurance determining date for that disease.” App. 2318. The list of factors includes latency (minimum of 15 years since first exposure) and diagnostic confirmation (laboratory or radiologic evidence). In the case of lung cancer, one combination for establishing an “insurance-determining date” is proof the claimant was exposed to asbestos for a sustained period of time at least fifteen years earlier together with “laboratory confirmation,” which is specifically defined to include certain types of diagnostic findings, such as “sputum cytology.” Given those two elements, the insurance determining date is the “date sample taken, or date of event giving rise to recommendation for testing, including visit to doctor.”8 The shortcomings of this method are inherent both in the analytic approach of the Protocol and in practical application of it. We borrow the district court's language to describe the theoretical problems:
[T]he protocol also relies on a series of assumptions. For example, the protocol assumes that a latency period of at least 15 years is necessary and that to produce the three most prevalent types of lung disease, mesothelioma, lung cancer or asbestosis, different intensities of exposure are required. The protocol defines these intensities as heavy, moderate, and light exposure, assigns to each a numerical value (point) which differs for each disease, and it requires a minimum value, again a different one for each disease, (but the same for each claimant who has the disease) to establish the element of exposure. Moreover, a substantial number of the protocols lead to coverage dates that are in fact dates of diagnosis. Except with respect to claimants suffering from mesothelioma, the protocol ignores earlier symptoms and focuses entirely on “diagnostic confirmation.” For example, the coverage date for lung cancer may be fixed as of the “date sample taken,” or “date x-ray taken,” or “date of onset of effusions.” Since effusions are evident only on x-ray, the last date is also the date of x-rays. Since these are the dates surrounding medical investigation-of symptoms they will most likely be the dates on which diagnosis is made if the results of the investigation are positive. If they are negative they do not count for anything under the protocol.
District Court Opinion at 16-17. The uncertainties surrounding the Protocol’s approach are compounded by the fact that most claimant files apparently contain relatively little medical information, meaning that the Protocol would be useless in determining a date of diagnosability.9 In addi*234tion, many of Eagle-Picher’s cases have been disposed of for relatively small sums of money — an average of about $7,300 as of July, 198410 — and the administrative expense of conducting an individual file review makes it questionable whether EaglePicher would be better off simply absorbing the loss rather than spending resources trying to prove insurance coverage. Finally, the protocol as proposed would limit Eagle-Picher to filing 150 claims per month with the insurance company, or 1,800 per year, which was less than eight percent of the total claims pending against EaglePicher three years ago. In the words of the district court, “[w]hatever may be the merits of the protocol as an abstract proposal, its effect, in practice, would be to leave Eagle-Picher substantially without coverage.”
B.
Faced with an apparent choice between Scylla and Charybdis, our first inclination was to avoid both and to seek some other, better, route. Upon further contemplation, however, we have concluded that it would be inappropriate to send this case back to the district court with a demand that it start from scratch in devising a satisfactory approach to the problem of diagnosability. Playing a prominent role in our decision is the “clearly erroneous” standard of Fed.R.Civ.P. 52(a):11
*235This standard plainly does not entitle a reviewing court to reverse the finding of the trier of fact simply because it is convinced that it would have decided the case differently. The reviewing court oversteps the bounds of its duty under Rule 52(a) if it undertakes to duplicate the role of the lower court. “In applying the clearly erroneous standard to the findings of a district court sitting without a jury, appellate courts must constantly have in mind that their function is not to decide factual issues de novo.” Zenith Radio Corp. v. Hazeltine Research, Inc., 395 U.S. 100, 123 [89 S.Ct. 1562, 1576, 23 L.Ed.2d 129] (1969). If the district court’s account of the evidence is plausible in light of the record viewed in its entirety, the court of appeals may not reverse it even though convinced that had it been sitting as the trier of fact, it would have weighed the evidence differently.
Anderson v. City of Bessemer City, North Carolina, 470 U.S. 564, 573-74, 105 S.Ct. 1504, 1511-12, 84 L.Ed.2d 518 (1985) (emphasis added); Marcoux v. Maine, 797 F.2d 1100, 1106 (1st Cir.1986).
We think this standard is especially appropriate when a district court has received extensive technical information about an obviously complex and elusive subject matter such as medical diagnosability. What the standard specifically means to us in the context of this case is that we may not lightly cast aside the district court’s decision in the absence of a conviction that something better does, in fact, exist. For the most part, this we can not say. While we question in one respect the logic and legality of the district court’s apparent decision to adopt either one or the other of the two proposals in an all-or-nothing fashion — and we accordingly modify its decision — we can not say the district court erred in focusing only on the two proposals presented to it and in generally preferring Eagle-Picher’s approach to AMICO’s. The record indicates that there is a dearth of medical knowledge on the diagnosability of asbestos-related diseases and there is no indication that something more reliable and practical than either party’s proposal is available. Thus, we accept for purposes of our review in this case that the world of possible alternatives is contained within the boundaries of the two proposals offered to the district court.12 We thus turn to consider why we find no clear error in the district court’s preference for Eagle-Picher’s method.
We note first that neither proposal offers a substantial likelihood of accuracy for most cases. The apparent absence of pertinent medical information in claimants’ files means that AMICO’s case-by-case scrutiny is destined to prove fruitless in many, if not most, cases. See supra note 9. The statistical nature of Eagle-Picher’s approach means that it will more often be wrong than right as applied to individual cases. But at least some of the merit in Eagle-Picher’s attempt to standardize the procedure is that the proposal will, in every case, produce an answer to the question of diagnosability. Under AMICO’s proposal, an additional method is necessary to resolve those cases where the medical data is insufficient.13
*236Second, despite its shortcomings, the Weill study was performed by an acknowledged expert in asbestos-related disease, and the evidence indicates that its methodology-while not perfect — was not so suspect that it would be clearly wrong to rely on the study’s conclusions. The fact that the study does not address any particular individual’s disease is simply a byproduct of its nature as an epidemiologic tool. AM-ICO’s expert, Dr. Edward Gaensler, acknowledged that Dr. Weill’s use of x-rays with 1/0 readings may have epidemiologic significance despite their questionable value as diagnostic tools.14 Moreover, the study’s result — a 6.14 year median time to prior diagnosability — is roughly consistent with the five-year figure given independently to Eagle-Picher by five different medical specialists whose opinions were based on their experiences diagnosing and treating asbestos-related diseases.
Third, we think the district court was justified in considering the cost of a case-by-case review in light of the evidence that many Eagle-Picher claims are settled for small amounts. Although we still adhere to our view that administrative convenience should not rule the decision-making process in this context, 682 F.2d at 24, it must likewise be true that an insured should not be effectively denied coverage because of a procedural financial barrier. Although it is unclear to us exactly how costly a case-by-case review would be, the record furnishes some support for assuming that it would impose significant financial and time burdens on Eagle-Picher. See Testimony of Bernard Adams on AMICO review of 1,029 files, Tr. 7-80-82 (AMICO spent about $130,000, at billing rates, and about five months to extract data from 1,029 files).
C.
While these factors are sufficient to convince us that the district court was not clearly erroneous in its preference for Eagle-Picher’s method, we do believe the court clearly erred in rejecting entirely the concept of an individualized review. We think it obvious that, if feasible, a case-by-case review would more closely comport with our expectations in Eagle-Picker I, where we noted that “the policies are geared to the injuries of a particular claimant,” 682 F.2d at 25 n. 12. Moreover, logic dictates that if one can determine with reasonable accuracy when an individual was capable of diagnosis by scrutinizing his medical records, such a scrutiny is preferable to the use of a statistical model that is more likely than not to designate the wrong date for that individual. Thus, if this had been a case of relatively few claims, documented by detailed medical records, we would have had no difficulty mandating an individual review of all claim files. Cf Wilkins v. Grays Harbor Community Hospital, 71 Wash.2d 178, 427 P.2d 716 (1967) (issue of diagnosability of disease in single claimant); Malone v. Continental Life and Accident Co., 89 Idaho 77, 403 P.2d 225 (1965) (same); Broccolo v. Horace Mann Mutual Casualty Co., 37 Ill.App.2d 493, 186 N.E.2d 89 (1962) (same); Dowdall v. Commercial Travelers Mutual Accident Ass’n, 344 Mass. 71, 181 N.E.2d 594 (1962) (same). Obviously, this is not such a case. In addition to the tens of thousands of claims already pending, new lawsuits continue to be filed.15
*237But this is also not a case in which we have no detailed information on any claimant, and for which we have no choice but to devise a statistical approach that could be applied to every claim. At least some medical files involved here are extensive. We are persuaded that it would be unjust to preclude AMICO from demonstrating that, in a given case, the six-year rollback is so clearly wrong that a particular claim does not belong within its policy period. We therefore hold that the six-year period should be given only presumptive impact, and that AMICO may show, by clear and convincing medical evidence, that the date of diagnosability in a particular case falls outside the AMICO policy period.
We believe the burden of showing the inapplicability of the six-year rollback properly belongs on AMICO. The number of cases at issue, the low settlement amounts, and the apparent paucity of medical information in many files contributed to our conclusion that the district court did not err in adopting the six-year statistical model over a case-by-case review. These facts led to a concern that Eagle-Picher would, in effect, be denied coverage if it were forced to establish the date of diagnosability on an individual basis before seeking coverage from AMICO. Placing the burden on AMICO to demonstrate that a given claim does not fall within its policy period avoids stripping Eagle-Picher of the insurance protection for which it contracted while at the same time giving the insurer the option of financing an individual inquiry that might seem worthwhile to it, because of either the size of the claim or the developed nature of the particular case.
Moreover, we think placement of the burden in this fashion is consistent with the AMICO policy language, which obliges the insured to provide “reasonably obtainable information respecting the time, place and circumstances” of the insurance triggering occurrence, namely the asbestos-related disease of the underlying claimant. Policy Provision C, “Notice of Occurrence” (App. 1768). The information provided pursuant to a six-year rollback meets that requirement.
We recognize that our presumption-with-a-challenge holding gives AMICO the incentive to question many individual cases, because it has nothing to lose and everything to gain from a change in the date of diagnosability. We would hope, however, that the same financial impediments that supposedly exist to deter a case-by-case review by Eagle-Picher similarly would discourage AMICO from questioning the vast majority of files designated as being within its policy periods. Moreover, if most files are as incomplete as the record tends to show, an individual review — although potentially raising many questions about the date of diagnosability — will produce no answer more definitive than the six-year presumption. In that case, the six-year presumption will prevail. AMICO may employ a different rollback period only if by clear and convincing evidence it satisfies the district court (or other relevant tribunal, see infra) that the date of diagnosability falls outside the AMICO policy period. In assessing AMICO’s evidence, the court may give to the presumption itself such evidentiary weight as it reasonably determines appropriate.16
*238We further recognize the logistical problem this holding presents in two respects. First, we risk a series of collateral proceedings surrounding AMICO’s challenges to the six-year date of diagnosability. We hope this will not occur, and believe the best way to avoid such a scenario would be for the parties to devise a system short of litigation for resolving any conflicts that do arise. Arbitration seems to be one possibility. Because the district court is in a better position to know whether, rather than a voluntary method such as arbitration, a mandated procedure would better serve the court’s and the parties’ interests,17 we remand this issue to the district court for further consideration and action. Moreover, we believe the district court has the power to sanction and curb the abusive or excessive resort to litigation, much as it has the power to restrict discovery. See Fed.R.Civ.P. 26(b)(1).
Second, this holding anticipates challenges to specific claims, and it appears from the record that other insurers’ payments to Eagle-Picher have been made on an aggregate basis. In other words, Eagle-Picher has totaled the number of dollars paid on claims falling within a given policy year and, if the claims amounted to at least the maximum of Eagle-Picher’s various insurance policies, it sought reimbursement from the individual insurers up to the policy limits. Eagle-Picher apparently did not allocate specific claims to these insurers. Thus, although Eagle-Picher consistently has carved out the $5 million per policy it claims from AMICO, it appears that this was not done by setting aside a certain group of claims.
We are uncertain how the district court anticipated that Eagle-Picher would demonstrate which claims properly belonged to AMICO in a manner that would enable AMICO to check to be certain the claims should be assessed against its policy. The uncertainty surrounding this procedure seems to be at the center of the certification issue, discussed infra. In our view, AMICO is entitled to a list of individual claims for which Eagle-Picher contends AMICO is responsible.18 Assuming our understanding of the record is correct, and the settling insurers have not made payments on an individual claim basis,19 we conclude that a proper method for allocating particular claims to AMICO would be to assume that the other insurers paid the claims that chronologically they would have been liable for had each insurer made the reimbursements for which it was responsible as the claims were paid by EaglePicher. Thus, assuming Liberty Mutual, the primary carrier, had a policy limit of $500,000 for 1974, the first $500,000 worth of individual claims and expenses arising from cases diagnosed in 1980 — “first” being determined by the dates of payment by Eagle-Picher — would be allocated to Liberty Mutual. The next $5 million arising from those cases would be allocated to AMICO. Additional claims and expenses would be attributed to insurers whose coverage was layered on top of AMICO’s.
A significant, but we hope not insurmountable, complication is that the settlements by the other insurers were based on a five-year rollback from the date of diagnosis while AMICO’s liability is to be based on a six-year rollback. This would mean, for example, that in the case of claims with diagnosis dates in 1979, the other insurers paid under the policies in effect in 1974— five years before the date of diagnosis. For the purpose of determining AMICO’s liability, however, those 1979 claims, and the payments made against them, would roll back six years to the policies in effect in 1973. Similarly, the 1980 claims and *239payments would have to be re-assigned from the 1975 policies to the 1974 policies.
We note that, under the six-year rollback theory advocated by Eagle-Picher, claims with diagnosis dates in 1978 would not fall within AMICO’s coverage period, although a five-year rollback would have placed at least some of them under AMICO’s policy. We do not know what, if any, reimbursement Eagle-Picher received for those claims from the second-layer excess carriers, the London Market, but any amount carved out on the assumption AMICO would be liable is obviously now not recoverable from AMICO.
Claims with 1981 diagnosis dates also pose a unique problem. Under a five-year rollback, those claims would be outside AMICO’s coverage period, which extended only to October 1975. But going back six years from diagnosis places at least some of the 1981-diagnosed claims within AMI-CO’s 1975 policy. Again, we do not know what, if any, reimbursement Eagle-Picher has received for claims diagnosed in 1981. But because the London Market may have made payments on the assumption that AMICO had no liability for those 1981-di-agnosed claims, some payment readjustments may be necessary.20
Further complicating matters, however, is that the same 1981-diagnosed claims would also be the London Market’s responsibility under the five-year rollback. For example, AMICO’s policy calling for first-layer coverage of occurrences up to October 1975 would embrace a claim with a diagnosis date of September 15,1981 under the six-year rollback approach we endorse in this opinion. But the London Market’s policy calling for first-layer coverage after October 1975 also would embrace a claim with a diagnosis date of September 15, 1981 under the five-year rollback used for purposes of the settlement agreement. It therefore appears that some method of apportioning the payment obligation between the two insurance companies is necessary. Because we have no information about what has actually transpired with regard to payments for claims diagnosed in 1981, we feel ill-equipped to suggest to the district court a method for handling this apparent overlap. We therefore leave to it that determination.21
We believe this system of assigning claims not only is accurate in reflecting what would have happened had the insurers simply paid claims without questioning their liability, but also is equitable because the allocation will be based on the relatively neutral criterion of date rather than on the size of the claim or detail of the file. What is not evident to us, however, is the best procedure to follow when AMICO successfully challenges a claim assigned to it in the fashion we have described, thus opening up some amount of its policy cov*240erage.22 We see no language in the AMI-CO policy excluding Eagle-Picher from seeking reimbursement from AMICO for any claims in excess of Liberty Mutual's underlying limit, even if Eagle-Picher arguably already has received reimbursement for those claims from a tertiary insurer.23 Thus, if AMICO is responsible for any claim (falling within its policy periods) arising after exhaustion of Liberty Mutual’s coverage, up to its $5 million policy limit, one approach would be to allow Eagle-Picher simply to replace the rebutted claim with the next appropriate one in chronological sequence following AMICO’s original $5 million worth of claims.
Even if the policy could be construed to protect AMICO against assessment for a claim already paid by another insurer, it does not necessarily follow that AMICO should be freed from claims originally allocated to other insurers under our system. Although in strict conformance with our fictionalized chronological allocation of claims, a payment by AMICO on such a replacement claim would represent a “double” payment on that particular claim, we recognize that the allocation is, indeed, a fiction. Thus, we could simply modify the fiction to assume that in such a situation, the London Market paid the claim rebutted by AMICO rather than the “replacement claim” that arose later. That later claim, after all, would have been AMICO’s responsibility had the challenge to the rebutted claim been made at the time EaglePicher would have sought payment from AMICO. Indeed, abandoning the strict chronological assignment of claims in this situation probably would more closely reflect the proper allocation.
It may be, however, that the district court would find it appropriate to retain the assumption that the London Market paid the specific claims falling beyond the original $5 million AMICO liability, and further find that it would be inappropriate to allow Eagle-Picher to simply replace a rebutted claim with another claim from the same policy year that had been assigned to a different insurer under our proposed allocation system. The rationale could be that such a replacement would, in effect, constitute a double payment on that claim, and that AMICO is somehow protected against such double payments. In such a circumstance, the district court could still allow Eagle-Picher to seek reimbursement from AMICO for claims other than those originally allocated to AMICO. It could, for example, allow Eagle-Picher to work out a reimbursement procedure with the London Market, eliminating the double payment problem. Or it could allow Eagle-Picher to assign claims to AMICO based on an individual review of the medical files, so long as those claims have been neither paid by another insurer nor deemed paid under our proposed method. These could include claims that Eagle-Picher presently considers to be in excess of all of its insurance *241coverage.24 Because we believe the district court, after consultation with the parties, will be in a better position to determine the specific procedure to follow if AMICO successfully rebuts a claim, we do not recommend any particular procedure at this time.
Moreover, despite our view that the allocation system we have described seems both appropriate and workable, the record leaves us less than certain that the assumptions we have made about aggregate payments are entirely correct. We therefore do not embody any aspect of this procedure in our judgment, but remand the question of claims allocation to the district court, with our proposed system serving as guidance to the extent the district court deems appropriate after a full consideration of the facts surrounding claims allocation.
D.
We are left with one more significant issue to address with regard to the question of diagnosability. Up to this point, we have not distinguished between the various types of asbestos-related disease. AMICO contends that whatever the validity of the six-year rollback to asbestosis, the disease that was the subject of Dr. Weill’s research, it has no relevance whatsoever to the other diseases suffered by Eagle-Picher claimants, notably lung cancer and mesothelioma. The district court declined to treat the diseases separately, in part on the basis of a footnote in EaglePicher I in which we noted “the desirability of similar treatment of all asbestos-related diseases,” 682 F.2d at 19 n. 3. In that case, however, which dealt with the proper theory of insurance coverage, the evidence was that the manifestation theory applied with even more force to the less frequently occurring diseases, such as mesothelioma and lung cancer, than to asbestosis. Id. In such circumstances, there was every reason to treat the diseases similarly, i.e., to apply the manifestation theory to each. In this context, however, the evidence suggests that similar treatment, i.e., identical assumptions in applying the manifestation theory, would be illogical. In the case of mesothelioma, for example, the record indicates that its nature as a quickly progressing disease makes it virtually impossible to diagnose much before death. See Wilson v. Johns-Manville Sales Corp., 684 F.2d 111, 113 (D.C.Cir.1982) (insulation worker diagnosed with asbestosis in 1973; mesothelioma in 1978; death occurred in 1978); Fearson v. Johns-Manville Sales Corp., 525 F.Supp. 671, 673 (D.D.C.1981) (insulation worker diagnosed with asbestosis in 1973; lung cancer in 1979; death occurred in 1979). We think it would therefore be improper to apply the six-year rollback if a claimant suffered only from mesothelioma.25
Although the record suggests two to three years as an appropriate rollback for mesothelioma, A. 2063-64, the district court paid no deliberate attention to the question of separate methods for determining the diagnosability of diseases other than asbestosis. The district court should *242therefore consider this question upon remand. We do not believe a study akin to Dr. Weill’s is necessary, and in any event, it probably is not possible. We do believe, however, that Eagle-Picher’s method of soliciting the opinion of a series of independent medical experts would furnish sufficient information upon which the district court could act in designating either a different rollback period for the other diseases or some other method for calculating the date of diagnosability.
Thus, to summarize our holdings with respect to the proper method for determining when an asbestos-related disease is deemed “reasonably capable of medical diagnosis,” and for assigning claims to AMI-CO based on the proper theory of diagnosability, we first conclude that the district court did not clearly err in adopting generally the theory that diagnosability precedes the diagnosis of asbestosis by six years. The six-year rollback shall serve, however, only as a rebuttable presumption, subject to a showing by AMICO, based on clear and convincing medical evidence, that the asbestosis was first reasonably capable of diagnosis at some time outside of AMICO's policy period. The district court shall, upon remand, determine the presumptive method of determining the date of diagnosability for the other asbestos-related diseases. The court should also determine whether a mandated procedure should be adopted for resolving conflicts between the parties over the diagnosability dates.
We also remand to the district court the question of how to allocate individual claims to AMICO, with the suggestion that it do so by assigning claims to policies chronologically according to their settlement dates.
III.
One matter of unfinished business with regard to the date of diagnosability is AMICO’s separate challenge to the use of diagnosis dates selected by Liberty Mutual as the starting point for the six-year rollback. As part of its function of processing all claims filed against Eagle-Picher, see supra note 4, Liberty Mutual assigns a date of diagnosis to each claim. The district court directed that the six-year rollback be applied to those Liberty Mutual diagnosis dates.
In fixing a date of diagnosis when no such date is specified in a claimant’s file, Liberty Mutual’s official practice is to use at least temporarily either the date of the complaint or a date ten months before the date the complaint was filed (ten months representing an estimated average between the date on which a plaintiff was diagnosed to have an asbestos-related disease and the date on which he instituted suit).26 The date is supposed to be changed as new information is obtained. The procedure actually used for setting diagnosis dates has varied over the years, however, with branch offices often adopting their own practices. App. 226.
AMICO claims the diagnosis dates set by Liberty Mutual are arbitrary and often have no correlation with the actual date of diagnosis. It claims its review of more than 1,000 files demonstrated that 45 percent of the Liberty Mutual-assigned “accident” dates27 were clearly incorrect, and only 19 percent were clearly correct. In addition, it points to a computer printout from Liberty Mutual displaying claims with assigned accident dates from June 1979 through September 1981. The printout shows a disproportionate number of claims with dates of January 1, 1980 and January 1, 1981, and also shows claims assigned to holidays and Sundays, all of which are unlikely diagnosis dates.
Although we acknowledge that the Liberty Mutual diagnosis dates are not flawless, we find no clear error in the district court’s conclusion that those dates on the whole are reasonably accurate and furnish an appropriate starting point for applying the *243six-year rollback. AMICO has convinced us neither that its “correct” dates always are more correct than Liberty Mutual’s nor that any other method would produce overall results significantly different or fairer, relative to AMICO’s coverage, than those produced by Liberty Mutual.
There are two reasons why AMICO’s conclusions with regard to Liberty Mutual’s accuracy are of little value. First, they are based on AMICO’s own criteria for picking a correct date. In one instance, for example, AMICO contends that the Liberty Mutual date is off by more than seven years based on its apparent conclusion that a date of diagnosis should be assigned to the “Date of First Inconclusive X-ray” rather than to the “Date of First Positive Diagnosis.” App. 2407-08.28 We have no reason to second-guess the district court’s apparent conclusion that Liberty Mutual’s choice is reasonable.
Second, Liberty Mutual’s diagnosis dates may reflect information not contained in the Eagle-Picher files — and thus not available for the AMICO review — because Liberty Mutual calculated diagnosis dates based on data contained in the files for all of its insureds, and not just on data contained in the Eagle-Picher files.29
In addition, the disproportionate assignment of claims to January 1 of a year, or the assignment of claims to Sundays and holidays, is logically explained in light of the many files that contain no positive diagnosis dates. In those instances, Liberty Mutual employees used either the ten-months-before-complaint-filed approach or another equally formulaic method for setting dates of diagnosis. For example, if the ten-month method is applied to a complaint filed on May 4, 1981, the result would be a diagnosis date of July 4, 1980. Some employees who failed to follow the ten-month procedure set diagnosis dates of January 1 of the calendar year in which the claim was filed, or January 1 of the preceding year if the suit was received during the first six months of a calendar year. App. 226. Other claims were assigned to the first day of a month or a year if a claimant testified that he was diagnosed “around the first of the year” or “early in September.” Id. Although the dates resulting from these procedures may be termed, in some respects, arbitrary, there is no evidence that more accurate dates could be set based on the available information.
Moreover, AMICO suggests no alternative to the Liberty Mutual dates other than a new case-by-case scrutiny (apparently in accordance with its view of the proper way to decide the date of diagnosis). But even AMICO’s review showed that only 70 percent of the files it examined had sufficient information for assigning a date of diagnosis. Thus, in at least 30 percent of the cases,30 any method of setting dates would be arbitrary, and for those cases we have no reason to challenge the district court’s acceptance of Liberty Mutual’s approximations. For the other cases, as noted above, inaccuracy may well be in the eye of the beholder, and we do not find the vision of the district court to be clearly erroneous.31
*244Finally, even where errors do exist, AMI-CO fails to persuade us that they are significant. More than half of the Liberty Mutual dates are within six months of AM-ICO’s “corrected” dates, and 70 percent are within twelve months. App. 2733. Although we recognize that AMICO provided only a narrow window of coverage to Eagle-Picher between 1973 and 1975, the frequently small discrepancy between AMI-CO’s and Liberty Mutual’s dates suggests to us that “correcting” the date would affect AMICO’s coverage only rarely. This is insufficient for us to conclude the district court erred. Nevertheless, we do not preclude AMICO from offering a clear and convincing showing of prejudicial error in one of those rare cases.
IY.
We now turn to Eagle-Picher’s challenges to the district court’s Final Judgment, addressing first the company’s objection to a “certification clause” requiring that, when submitting asbestos-related claims to AMICO for reimbursement,
Eagle-Picher ... shall certify ... that no payments to Eagle-Picher on account of such loss and expense payments have been made by or claimed or requested from any other source.
This language presents a problem for Eagle-Picher because of its settlements in 1983 with Liberty Mutual and the London Market and its participation in an entity known as the Wellington Asbestos Claims Facility.32
The Claims Facility, named after Dean Wellington of the Yale School of Law, was established pursuant to an agreement between certain asbestos defendants and their insurers in an effort to handle more efficiently the thousands of asbestos claims being filed nationwide. See Agreement Concerning Asbestos-Related Claims (the “Wellington Agreement”) (Supp.App. 372). It acts as a central clearinghouse for asbestos claims, administering, defending, and settling them, and then determining who pays how much of the associated costs.33 Eagle-Picher and certain of its insurers have been participating in the Claims Facility since it was set up in 1985. AMICO has not participated.
Under both the Wellington Agreement and the separate settlements reached in 1983 between Eagle-Picher and some of its insurers, Eagle-Picher has received payments or benefited from payments in ways that preclude it from certifying in the manner specified by the district court some claims that otherwise would fall within AMICO's coverage obligation. For example, the London Market insurers involved in the 1983 settlements received an alphabetical list of claims falling within their coverage period, and determined only that the aggregate losses and expenses during that period exceeded the amount of the underlying insurance provided by other companies, including AMICO. Affidavit of Mary Ann D’Amato (Supp.App. at 443). Thus, the London insurers did not allocate their pay-
*245ments to specific claims; they simply determined that their policies were triggered, and made payments accordingly.
Similarly, under the Wellington Agreement, Eagle-Picher has received payments in a manner that failed to isolate AMICO’s obligation. The process operates so that a participating insured company pays out no money on asbestos claims until the coverage limits of its participating insurers are exhausted. Thus, if the London Market, as a second-level excess insurer, participated in the Claims Facility and AMICO, as a first-level excess insurer, did not, the London Market initially would be assessed for claims falling within AMICO’s coverage period. The idea apparently is to allow the insured company to take advantage of available insurance proceeds at as early a point as possible. Although the London Market insurers would be making payments earlier than they would have if they first waited for the exhaustion of AMICO’S coverage, the London Market’s payments also would end earlier in time because they would reach their coverage limits sooner. When coverage provided by participating insurers is exhausted, the insured company assumes full responsibility for payments owing under the Wellington Agreement.
A significant aspect of this process is that a participating insured company is obligated to seek recovery from a wo«-participating insurer. Any money recovered replenishes the participating company’s available coverage in the Claims Facility. If such recovery is not obtained within specified time limits, the insured company must pay interest to the participating insurers on the amounts “advanced” by those insurers to fulfill a non-participating insurer’s obligation. Also significant for our discussion is the way in which Eagle-Picher has modified the Claims Facility procedure. EaglePicher has agreed to reimburse its participating insurers for those costs advanced on behalf of non-participating insurers.
Thus, the Claims Facility procedure is even more problematic in relation to the certification language than the aggregate payments made through the 1983 settlements. Under the Claims Facility procedure, participating insurance companies expressly assume the liabilities of non-participating companies. The certification language, however, prohibits Eagle-Picher from submitting claims on which it has received, requested or claimed payment from another source. Technically, this means that many claims properly attributable to AMICO as the first-layer excess insurer may not be submitted to AMICO for payment because they have been covered (although reimbursed) or are coverable under the Wellington Agreement by other insurers.
This background persuades us that the certification question warrants further consideration by the district court. The district court denied Eagle-Picher’s motion to amend the Final Judgment by deleting the certification language for two reasons. First, it believed the judgment was consistent with the requirements of the AMICO policy, and second, because it had “no evidence as to the manner in which the Claims Facility assigns claims to particular periods or policies and makes payment, the court cannot evaluate plaintiff's objections to the judgment.” We believe the certification clause requires more of Eagle-Picher than does the AMICO policy, and we also believe that any certification obligation on the part of Eagle-Picher must take into account the manner in which the company has received payments from other insurance companies.
As for policy language, the parties have pointed to two provisions relevant to the question before us. First, the policy states that AMICO shall be liable for amounts “in excess of the underlying limit,” and “underlying limit” is defined as:
the amount of the applicable limits of liability of the underlying insurance ... and ... the amount of any other valid and collectible insurance available to the insured, whether such other insurance is stated to be primary, contributing, excess or contingent (except insurance purchased to apply in excess of the sum of the underlying limits ... and the limit of liability hereunder); ...
AMICO Policy, Sections II and IV (emphasis added) (App.1767). This provision clear*246ly states that payments available from excess insurers whose policies are triggered after AMICO’s coverage is exhausted are not to be considered in determining AMI-CO’s liability. The payments at issue here fall into that category.
Second, as noted above, see supra note 23, the policy also states that AMI-CO’s obligation is for “ultimate net loss,” a term whose definition excludes any expense “included in other valid and collectible insurance.” Id. at Section IV. This provision may refer only to expenses as distinguished from claims payments. Even if it covers both types of payments, the section does not bar Eagle-Picher from seeking coverage for claims already presented to other insurers pursuant to the 1983 settlement agreements and through the Claims Facility. For the 1983 settlements, the other insurers simply paid a chunk of the claims falling within both their and AMICO’s coverage period, explicitly excluding the $5 million AMICO coverage amount. In the Claims Facility context, the participating insurers temporarily paid AMICO’s obligation, but their own coverage obligation was correspondingly reduced. In neither instance was the AMI-CO liability “included in other valid and collectible insurance.”
Thus, Eagle-Picher’s actions in obtaining insurance payments have not conflicted with either of these provisions. Yet the district court’s certification requirement— by focusing on individual claims and on whether payments for them were made by or claimed or requested from another source — arguably could exclude EaglePicher from submitting to AMICO all claims for which the company had insurance coverage in excess of AMICO’s. This is so because, in the context of the 1983 settlements, Eagle-Picher submitted all claims falling within a given time period to the other insurers to substantiate its request for payment on their policies. And in the Claims Facility context, the facility’s methodology requires Eagle-Picher to submit all claims for payment regardless of which layer of insurance is applicable. In effect, in light of Eagle-Picher’s arrangements with other insurers, the district court’s certification clause potentially could deprive the company of payments to which it is entitled.
We are not at this time equipped to propose substitute certification language. It may be that upon remand, the district court will find a certification clause to be unnecessary. Or it may conclude that, in light of whatever allocation system it adopts, Eagle-Picher should be required to make entirely different types of assurances. At this time, we simply vacate the present language with directions that the district court at least revise its certification language in light of the Wellington Agreement and the principles of claims reimbursement expressed in this opinion.34 We do make two final observations, however: 35 first, AMICO should not be able to evade its proper responsibility for insurance coverage simply because events have occurred during the lengthy tenure of this case that preclude Eagle-Picher from making a straightforward submission of unpaid claims to it. Second, Eagle-Picher should not be able to evade its responsibility to present AMICO with individual claims simply because it chose to reach settlements with other insurers in ways inconsistent with AMICO’s policy requirements. Thus, any certification requirement imposed by *247the district court should be designed to reflect both of these concerns.36
V.
We see no need for a prolonged discussion on Eagle-Picher’s claim that it is entitled to bad faith damages and attorney’s fees from AMICO. Eagle-Picher claims that our 1982 ruling establishing when insurance coverage for asbestos-related disease is triggered made it absolutely “clear-cut” that AMICO’s coverage would be triggered, no matter how our mandate was implemented. As a result, Eagle-Picher argues, AMICO from that point on had “an affirmative obligation to settle claims promptly and in good faith.” Eagle-Picher Brief at 30.
Our discussion thus far demonstrates that the issues surrounding AMICO’s obligations to Eagle-Picher are extremely complex and were by no means resolved by our 1982 ruling. Although it may well be that AMICO’s conduct in this litigation has at least in part been motivated by “a strategy of ‘delay and more delay,’ ” District Court Opinion at 24, we fail to see how, under any state law, its partially successful effort to secure the right to consider claims individually could be classified as bad faith. Moreover, AMICO has made all payments requested by Eagle-Picher under an interim financing agreement based on a one-year rollback from the date of diagnosis. Thus, AMICO has made timely payments in instances in which the basis for Eagle-Picher’s request for funds was not disputed. We find none of AMICO’s other actions arising to the level of bad faith. Accordingly, we affirm the district court’s judgment denying Eagle-Picher damages and attorney’s fees on account of bad faith.
VI.
Eagle-Picher claims that the district court erred in denying its request for prejudgment interest. Although the district court did not explain why it did so, it is evident to us that the court denied interest because it found that Eagle-Picher was not entitled to a judgment for money damages under any of the three theories the court was asked to consider. District Court Opinion at 19-26. The court stated that Eagle-Picher was not entitled to damages “or [to] any interest on the amounts claimed in damages.” Id. at 26.
Eagle-Picher appealed the district court’s damages decision only on the theory of bad faith and, in Section V above, we affirm the court’s conclusion on that point. Thus, Eagle-Picher has not received a money judgment in this proceeding to which an award of prejudgment interest could be attached. Although Eagle-Picher obviously hoped that we would reverse the damages decision, thus providing a foundation on which it sought to add interest, it appears that the company is also asking us to decide on the question of interest even in the absence of a damages award. In effect, Eagle-Picher seems to want us to declare that it is entitled to interest on the payments due as a result of the present judgment. It is unclear to us whether Eagle-Picher made this particular argument to the district court. Nevertheless, in light of the possibility that this argument was raised below, the already lengthy history of this case, and the virtual certainty that any decision on Eagle-Picher’s entitlement to prejudgment interest would be appealed, we believe it is appropriate to address the question at the present time.
We consider first the proper state law to apply. Massachusetts choice of law rules govern this diversity action. Klaxon Co. v. Stentor Electric Manufacturing Co., 313 U.S. 487, 496, 61 S.Ct. 1020, 1021, 85 L.Ed. 1477 (1941). See also Furtado v. Bishop, 604 F.2d 80, 97 (1st Cir.1979) (“[S]tate law governs the imposition of prejudgment interest in diversity cases____”). The Massachusetts Supreme Judicial Court *248traditionally has held that “material issues arising in actions to enforce contracts.are to be answered according to the law of the place where the contract was made____” Choate, Hall & Stewart v. SCA Services, Inc., 378 Mass. 535, 540, 392 N.E.2d 1045, 1048 (1979). In SCA Services, however, the SJC observed that most states have replaced “place-of-making or other one-factor tests with a more functional approach.” 378 Mass, at 541, 392 N.E.2d at 1049. Indeed, Massachusetts courts have adopted this more flexible approach for tort actions. See Cohen v. McDonnell Douglas Corp., 389 Mass. 327, 333, 450 N.E.2d 581, 585 (1983) (although law governing tort action usually is that of place of injury, “ ‘on the particular facts of a case another jurisdiction may sometimes be more concerned and more involved with certain issues’ ”) (quoting Pevoski v. Pevoski, 371 Mass. 358, 360, 358 N.E.2d 416, 417 (1976)).
Although the SJC has not yet had the opportunity to expressly adopt the more modern functional approach for contract cases, see SCA Services, 378 Mass, at 541, 392 N.E.2d at 1049, we read its opinions as indicating that that court no longer subscribes to the rigid application of a one-factor test. We therefore are confident that it would conclude that Ohio law governs issues concerning the insurance contract between Eagle-Picher and AMICO because that state has the “most significant relationship to the transaction and the parties.” Restatement (Second) of Conflict of Laws § 188 (1971). Eagle-Picher is an Ohio corporation, with its principal offices in Cincinnati. Although AMICO is based in Illinois and “will perform appropriate obligations ... under the contract when they come due by writing and posting a check in Illinois,” AMICO brief at 12, we believe this link with the contract is far surpassed by the fact that the contract was designed to cover risks arising in Ohio. Whether Eagle-Picher is entitled to reimbursement, and to what extent, for the many millions of dollars in claims paid as a result of injuries caused by its products is of major concern to the state in which the business is domiciled. Moreover, to the extent Massachusetts still would consider the place of making of the contract to be an important factor in the choice of law calculus, we note that the policy was countersigned by AMI-CO’s agent in Ohio. App. 1771. In addition, the SJC at times has referred to the Restatement (Second) of Conflict of Laws in resolving choice-of-law questions. See, e.g., Cohen, 389 Mass, at 336, 450 N.E.2d at 586. Under the Restatement, the rights of the parties to a casualty insurance policy “are determined by the local law of the state which the parties understood was to be the principal location of the insured risk during the term of the policy” unless some other state has a more significant relationship with the parties concerning a particular issue. Restatement (Second) of Conflict of Laws § 193 (1971). See Diamond Int’l Corp. v. Allstate Insurance Co., 712 F.2d 1498, 1501 (1st Cir.1983) (“The most significant relationship in insurance cases” ... is “ ‘ “[t]he principal location of the insured risk.” ’ ”) On the specific question of interest, the SJC has indicated that Massachusetts would “turn to the law governing rights and duties under the contract to determine the interest payable for breach of contract.” Morris v. Watsco, Inc., 385 Mass. 672, 678, 433 N.E.2d 886, 890 (1982).
Under Ohio common law, an insured is entitled to recover interest on money owed to him under an insurance policy. Hoskins v. Aetna Life Ins. Co., 6 Ohio St.3d 272, 452 N.E.2d 1315, 1320 (1983). In addition, Ohio Rev.Code § 1343.03(A) provides that “when money becomes due and payable upon any ... instrument of writing ... the creditor is entitled to interest at the rate of ten percent per annum, and no more, unless a written contract provides a different rate____” An insurance policy is a written instrument under this statutory provision. Florence v. New York Life Insurance Co., 48 Ohio St.2d 59, 357 N.E.2d 35, 37 (1976).
The Ohio statute expressly entitles Eagle-Picher to interest on money owed to it under the AMICO policy. See Jeppe v. Blue Cross of Northeast Ohio, 67 Ohio App.2d 87, 425 N.E.2d 947, 952 (1980) (appellant need not request interest when *249money judgment rendered in action for declaratory judgment because “interest would have been included thereon by operation of law under R.C. 1343.03”). Thus, the issue before us is not whether Eagle-Picher is entitled to interest but when its entitlement arose, i.e. when did money become “due and payable” under the AMICO policy.
Ohio law provides that even when a sum due is unliquidated, “ ‘where the amount is capable of ascertainment by mere computation, or is subject to reasonably certain calculations by reference to existing market values,’ ” prejudgment interest will not be denied. Shaker Savings Ass’n v. Greenwood Village, Inc., 7 Ohio App.3d 141, 454 N.E.2d 984, 986 (1982) (quoting McKinney v. White Sewing Machine Cory., 200 N.E.2d 596 (Ohio App.1964)). See also Nursing Staff of Cincinnati, Inc. v. Sherman, 13 Ohio App.3d 328, 469 N.E.2d 1031, 1034 (1984); Royal Crown Plastics & Sales v. Motorists Mutual Ins. Co., 51 Ohio App.2d 79, 366 N.E.2d 294, 296 (1976). Eagle-Picher argues that this principle applies here because it has proved “the exact losses and expenses chargeable to AMICO as of April 1, 1984, under the coverage implementation method the district court endorsed.” Eagle-Picher Brief at 48.
It is true that there is nothing uncertain about the amounts of Eagle-Picher’s claims payments. The accuracy of the lengthy computerized lists of figures representing the company’s resolution of individual lawsuits alleging asbestos-related disease is undisputed. What has not been settled until our decision today, however, is which of those claims properly belong within AM-ICO’s policy coverage. It can not reasonably be argued that AMICO’s obligation could be determined by “mere computation” or “reasonably certain calculations” until we ruled on how one is to determine when a disease is reasonably capable of medical diagnosis. Until now, it was unclear which claims would become AMICO’s responsibility. The uncertainty that existed until today is underscored by our decision to modify the district court’s approach.
Indeed, it could be argued that even after our judgment today, the amount owed by AMICO to Eagle-Picher is not sufficiently certain to entitle Eagle-Picher to interest because AMICO may challenge claims that mechanically fall within the six-year rollback period on diagnosability or other grounds. Thus, while the method for assigning claims is now fixed, the amount of AMICO’s debt to Eagle-Picher is still dependent upon the resolution of those potential challenges. Nevertheless, several factors persuade us that interest should begin to run from the time our judgment becomes final. First, we anticipate that most claims will be assigned to AMICO by means of the mechanical application of the six-year rollback — which will allow a determination of most of the debt by a method we deem equivalent to “mere computation.” Second, the running of the interest clock should discourage AMICO from further prolonging resolution of claims that properly are its responsibility. Finally, and most importantly, we believe an award of interest from this point on serves the policy of full compensation that underlies interest awards. See Hardiman v. Zep Mfg. Co., 14 Ohio App.3d 222, 470 N.E.2d 941, 948 n. 8 (Ohio App.1984) (“[t]he policy behind prejudgment interest is to make the plaintiff whole”). Eagle-Picher long ago paid most of the claims for which it now seeks reimbursement from AMICO. Although the past uncertainty of the amount owed by AMICO precludes an award of interest from the time Eagle-Picher made those payments, the relative certainty at this time persuades us that it would be unjust to ignore the economic reality that AMICO has for many years had the use of substantial sums of money that belonged to EaglePicher. We therefore hold that AMICO’s debt to Eagle-Picher is now “due and payable,” and that AMICO must pay interest to Eagle-Picher at the Ohio statutory rate from the date our judgment becomes final.
VII.
We summarize our major holdings as follows:
(1) To determine when asbestosis was “reasonably capable of medical diagnosis,” the *250parties shall presume that diagnosability occurred six years before actual diagnosis unless AMICO can show, based on clear and convincing medical evidence, that the asbestosis was first reasonably capable of diagnosis at some time outside of AMICO’s policy period;
(2) The district court must develop separate methods for determining the date of diagnosis for other asbestos-related diseases;
(3) The district court did not err in adopting the diagnosis dates selected by Liberty Mutual as the starting point for application of the six-year rollback;
(4) The district court must at least revise its certification language in light of the Wellington Agreement and the principles of claims reimbursement expressed in Section IV of this opinion;
(5) Eagle-Picher is not entitled to bad faith damages and attorney’s fees from AMICO;
(6) Eagle-Picher is entitled to interest on the amounts due from AMICO starting from the time our judgment becomes final.37
Affirmed in part, modified in part, vacated in part, and remanded for further proceedings consistent with this opinion.
. In so doing, we adopted what is called the "manifestation" theory of insurance coverage: each insurer whose policy was in effect at the time the asbestos-related disease first manifested itself by way of medically diagnosable symptoms must provide coverage. Most of the defendant insurers had argued for an "exposure” theory: each insurer whose policy was in effect during the time of exposure to asbestos must indemnify the manufacturer for a pro-rata share of its liability, the proportion to be determined by the ratio of the number of years the insurer was on the risk to the total number of years of exposure.
. Dr. Weill’s research determined that the median interval between an actual diagnosis of asbestos-related disease and the date when the disease reasonably could have been diagnosed is 6.14 years.
. The settlements were reached in 1983 with Liberty Mutual and more than twenty-five underwriting syndicates and insurance companies collectively known as the "London Market."
. The date of diagnosis used so far for all purposes in this case is a date set by Liberty Mutual Insurance Co., Eagle-Picher’s primary insurance carrier. Eagle-Picher has delegated the responsibility for defending all of the complaints against it to Liberty Mutual. Cases that fall outside Liberty Mutual's coverage responsibility are processed under a claims-handling agreement under which Liberty Mutual evaluates the claims, gathers medical data, retains and supervises outside counsel, and decides whether and when to settle.
In setting a date of diagnosis when a claimant’s file failed to specify it, Liberty Mutual decided to use at least temporarily the date of the complaint or a date ten months before the filing date (ten months being the average between the date on which a plaintiff was diagnosed to have an asbestos-related disease and the date on which he instituted suit). The date would be changed as new information was obtained. AMICO challenges the reliability of Liberty Mutual's diagnosis dates, an issue we shall address infra.
. The court held that the insurance-determining date "is the date six (6) years prior to the date of actual diagnosis of the asbestos-related disease or, with respect to those cases in which no diagnosis was made prior to death, six (6) years prior to the date of death."
. Our assumption throughout this opinion is that the question is whether "there is a distinct symptom or condition from which one learned in medicine can [today] with reasonable accuracy diagnose the disease,” Eagle-Picher I, 682 F.2d 24, because that seems to be the assumption of the parties. Neither AMICO’s proposal for resolving the diagnosability question, see AMICO Protocol, A. 2319, nor the study on which Eagle-Picher bases its approach, A. 1829, attempted to evaluate claims based on the "state of the art" of diagnosis techniques at various times. We see no reason to complicate the analysis by requiring use of a changing standard of diagnosability, but the district court should feel free to adopt whatever approach it deems best.
. For a detailed description of the study conducted by Dr. Hans Weill, professor of medicine at Tulane University School of Medicine and Chief of its Pulmonary Diseases Section, see the district court’s opinion at pages 11-12.
. AMICO’s approach is a clinical one, linking the decision on each claim to the specific clinical evidence in the individual’s file. In contrast, Eagle-Picher advocates an epidemiological approach, which relies on statistics to derive a conclusion that would be applicable generally to a given group.
. Despite AMICO’s assertions that the individual medical files frequently would provide enough information to determine the date of diagnosability, we find no clear error in the district court’s conclusion that most claimant files contain insufficient information to make individual judgments. None of the evidence presented by AMICO and cited in its briefs in support of its *234claim that a case-by-case review is feasible convinces us that the district court wrongly minimized the nature of the information in the files. For example, the testimony of Dr. Edward Gaensler that "[i]n most cases where real disease exists there are substantial medical records in the file," App. 2302, was based on his own clinical work and not on a review of Eagle-Picher claims files. Dr. John McDonald, a physician and epidemiologist with extensive work in the area of asbestos-related disease, testified that the AMICO protocol could he applied to all claims “with a reasonable amount of information in the file” — but he did not state that the files at issue here contained the requisite amount of information. App. 2839.
Andrew Meyers, a systems analyst who supervised the computer analysis of information gathered by AMICO in a review of 1,029 EaglePicher claims files, produced statistics showing that only about 70% of the files examined had sufficient information to determine the date of diagnosis, let alone the more problematic date of diagnosability. App. 2532.
In addition to AMICO’s failure to show affirmatively that the majority of claims files would be useful in a case-by-case review, Eagle-Picher presented testimony indicating that the files would not be helpful in such a review. James Ralston, Eagle-Picher’s vice president and legal counsel, testified that the files were not geared toward prior diagnosability but were aimed at showing a current diagnosis that would establish the claimant’s right to recover a payment from Eagle-Picher. Tr. 2-50. He testified that the files usually contained summaries of doctor visits and workers compensation histories, but generally did not contain evidence of serial x-rays that could have suggested the development of disease over time. Tr. 2-49, 2-51. Leon Prickett, supervising examiner of the asbestos unit at Liberty Mutual, who stated that he had reviewed about 2,500 files, testified that a claims file will contain historical medical data with "[vjery limited frequency." Tr. 4-52-53. We note also that AMICO’s case review included all closed claims prior to 1977, all closed claims with a paid loss in excess of $10,000, and one-sixth of the remainder. Prickett testified that the oldest claims tended to be more fully documented because there were fewer claims to handle. Tr. 4-85-87. The larger claims were better documented because of the greater potential exposure. Tr. 4-73-74.
. James Ralston, Eagle-Picher’s vice president and legal counsel, testified that the average cost of disposing of cases fluctuates and that the figure is calculated periodically. As of the trial in July 1984, the average loss per claim — which is the amount paid to the claimant — was $5,000, and the remainder of the $7,300 amount was the cost of counsel and administrative expenses. Tr. 1-72-73. Of 6,763 cases closed as of May 10, 1984, 2,375 involved a loss up to $1,000; 1,072 involved a loss between $1,000 and $2,000; and 1,096 involved a loss between $2,000 and $3,000. Of these closed cases, 35 percent were under $1,000 and 67 percent were under $3,000. App. 1822.
. AMICO is correct that, to the extent the district court’s adoption of a standard rollback of any duration violates our prior judgment, it would be an error of law not governed by the clearly erroneous standard. The district court found, however, that it is, in effect, impossible to obtain what we seemingly requested in EaglePicher I, namely an individualized determination of the date of diagnosability. We obviously can not require the impossible, and so if the district court’s finding regarding the feasibility of individualizing diagnosability is not clearly erroneous, there would then be no error of law in its decision to adopt some other method of determining when a disease becomes reasonably capable of medical diagnosis.
We find no error in the conclusion that an individual date of diagnosability is impossible to *235discern. Witnesses for both sides agreed on that much. The question then became how best to approximate the date of diagnosability in light of the circumstances of this case. That is the issue that we address in the text and that we believe is governed by the clearly erroneous standard.
. We emphasize, however, that we do not intend this decision to establish for all time the proper method for determining the date of diagnosability of asbestos-related diseases. Indeed, we would hope that a continuing increase in medical knowledge and diagnostic sophistication will make this an easier question to answer in the future.
. We reject AMICO’s suggestion that the date of diagnosis could be used when the earlier date of diagnosability is not obtainable. There is no dispute that asbestosis and the other relevant lung diseases are progressive, and that in most cases the claimants’ diseases would have been capable of diagnosis at some point considerably before the actual diagnosis was made. So long as a rational method exists for setting that earlier date we decline to undo our decision in Eagle-Picher I that insurance coverage is triggered by the date of diagnosability and not by the date of diagnosis.
. The International Labor Office has developed, for epidemiologic purposes, a set of standards for classifying opacities in lung x-rays. They range from 0 (normal) to 3 (severely abnormal). Each classification is subdivided so that the overall classification indicates the strength of the reader’s conviction that an x-ray fits into a particular category. One category, for example, would consist of x-rays that the reader considers to be slightly abnormal, and warranting an initial reading of 1. Those x-rays would be further subdivided into readings of 1/0,1/1 and 1/2. A reading of 1/0 means that the reader considers the x-ray to be slightly abnormal, but it may warrant consideration as possibly normal. A reading of 1/2 means that the reader considers the x-ray to be slightly abnormal, but it could be read to suggest a greater abnormality. A 1/1 would indicate a strong conviction that this is a slightly abnormal x-ray.
Although a reading of 1/0 is too uncertain to be used to diagnose a particular case. Dr. Weill used the 1/0 as diagnostic of asbestosis for purposes of his study.
. The district court observed that from May 1984 to September 1985, the number of pending asbestos cases in the District of Massachusetts had increased from 1,723 to 3,236. Eagle-Pich*237er is a defendant in most of these. District Court Opinion at 2.
. We do not mean by allowing AMICO to challenge individual cases that it should apply its Draft Protocol, or any variation on the protocol, to establish rules-of-thumb about when certain cases would have been reasonably capable of diagnosis. Rather, what we envision is AMI-CO’s possibly challenging individual cases in which the clinical medical evidence strongly demonstrates that a rollback into AMICO's coverage period would be incorrect. For example, it could be that a particular file will contain a series of x-rays over the six years preceding diagnosis, that nothing suspicious appeared in the first three years and that signs of asbestos-related disease were noted for the first time three years before diagnosis. If the diagnosis date were 1980, such a claim would fall outside of AMICO’s 1973-1975 coverage period. Another file might show that the claimant had no symptoms and was diagnosed with a particularly mild form of asbestosis in 1981 as the result of a mass shipyard screening that included x-rays. If the claimant’s x-ray were at the lowest level allowing a positive diagnosis, it seems that the date of diagnosability in such a case should be no more than a year or two before the date of the screening, which would also have been *238the date of diagnosis. Again, such evidence would show that the claim was not AMICO’s responsibility.
. The district court could, for example, assign a master to oversee the allocation of claims.
. Eric Thiemann, supervisor of insurance in the Corporate Insurance Department of EaglePicher, testified that Eagle-Picher would be able to provide names and numbers for claims attributable to AMICO. Tr. 4-141-42.
. See testimony of Roger Prickett, Tr. 6-6-10, describing billing process in which Eagle-Picher would bill on the basis of the gross amount of claims assigned to a particular time period.
. For example, if Eagle-Picher has paid out $6 million on claims arising from asbestos-related disease diagnosed in 1981, the London Market may have reimbursed the company for all but the amount due under Liberty Mutual’s primary policy — not carving out any amount as AMICO’s responsibility. This could be so because the London Market provided both first-layer and second-layer coverage after AMICO’s coverage period ended in October 1975. Thus, in effect, the London Market would have paid claims that fall within AMICO’s policy period.
. We make three final observations with regard to the reallocation of claims based on the six-year rather than five-year rollback approach. First, because AMICO’s policies covered only portions of the years 1973 and 1975, the shifting of claims will have to take into account not only the year but the specific date of the diagnosis and resulting rollback. Second, we realize that the London Market obligations under the 1983 settlements were calculated on a calendar year basis even though the policies actually covered portions of different years. We do not expect that this will affect the feasibility of the reallocation of claims. Finally, while we acknowledge that this process of reallocation seems burdensome, we anticipate that most of the rearranging will be accomplished by means of the computer programs already in existence. We also expect that AMICO will assist in the computer work. The insurer previously has offered to process the Eagle-Picher claims information at its own expense if provided with the computer tape containing the information in the Liberty Mutual data base for all claims against EaglePicher. Supp.App. 287. Thus, assuming the district court adopts our specific approach, we expect the parties to cooperate in producing a product for the district court that follows our guidelines and clearly sets out what we hope will be few areas of dispute.
. For example, suppose AMICO is able to show that a $10,000 claim with a diagnosis date of May 1980 that was assigned to the AMICO policy in effect in May 1974 (on a six-year rollback theory) in fact should have been assigned to a policy in effect in September 1978, when AMI-CO was not providing coverage. The question is what to do about the newly available $10,000 of coverage under AMICO's 1974 policy.
. The policy states that AMICO’s obligation is for "ultimate net loss,” a term whose definition excludes any expense "included in other valid and collectible insurance.” AMICO Policy, Section IV (App.1767). In context, this phrase seems to refer only to the “expenses" incident to the claims payments rather than to the damages payments themselves. Eagle-Picher, however, apparently construes this section to prevent an insured from recovering more than once on damages covered by overlapping policies. See Eagle-Picher Brief at 39. See also R.E. Keeton, Insurance Law § 3.11(a), at 168 (1971) (Insurance policies sometimes contain clauses declaring that the insurer shall have no liability if the policyholder has other insurance applicable to the loss; these clauses are more common in past than currently.).
In any case, double recovery because of payments made by secondary and tertiary insurers probably should be primarily the concern of the London Market, rather than AMICO, since the London Market insurance policies undoubtedly specify that they are responsible only for those claims beyond AMICO's $5 million obligation. Of course, the London Market settled without regard to individual claims, and so those insurers’ only concern probably is to be certain that Eagle-Picher’s total exposure is high enough to exhaust both AMICO’s $5 million and the amount of the 1983 settlements.
. We note that Eagle-Picher emphasizes that its total exposure “dwarf[s]” its available coverage. Eagle-Picher Brief at 40.
. When a claimant suffers a combination of diseases, we see no reason not to focus on the one occurring first since it appears that EaglePicher’s practice has been to settle cases based simply on the presence of any asbestos-related disease. In a combination case in which the differences in the diseases have been relevant for other reasons, however, it may be appropriate to consider the differences for the issue of diagnosability and insurance coverage. For example, if an asbestosis claim has been barred by the statute of limitations, and only a mesothelioma claim remains, it obviously would be appropriate to look at only the diagnosability of the mesothelioma even if the asbestosis occurred
first. Our assumption is that the second occurring disease would not have a rollback beyond the primary disease, so that we are in essence suggesting use of the earliest date of diagnosability.
One additional point needs to be noted. Andrew Meyers, the systems analyst who assisted AMICO in its file review, testified that 21 percent of the files in AMICO’s study lacked sufficient information to determine a primary diagnosis. App. 2529. In the absence of any other indication, we recommend assuming that asbestosis occurred first (assuming the claimant suffers from asbestosis) because it is the most frequently occurring disease and its date of diagnosability apparently is earlier than those of the other major asbestos-related diseases.
. The ten-month period resulted from a study performed by Liberty Mutual for Eagle-Picher and other insureds seeking to determine the eventual diagnosis dates for claims that originally came in without such dates, but for which more information was anticipated. Tr. 4-102.
. Liberty Mutual uses the term "accident date” for the date of diagnosis.
. Prickett testified that Liberty Mutual employees were told to look for a specific date when a doctor stated that an individual had a disease, Tr. 6-48. AMICO’s file reviewers would in some instances assign diagnosis dates based on “signs or symptoms" noted in the claimants’ files. Testimony of Bernard Adams, Tr. 8-61.
. Liberty Mutual insures a number of companies involved in asbestos litigation and maintains separate files for each of them. It also keeps separate files in its branch offices and in its home office, the latter being computerized. Information from the branch office files is shared electronically at the home office, Tr. 4-62, but AMICO reviewed only branch office files. Because each insured's individual defense counsel decides what medical information to submit to Liberty Mutual, Tr. 4-52, it apparently is possible that some branch office files for the same claimant would be less complete than, or contain different information from, others.
We note, however, that we have no reason to believe that even if all files were considered, a case-by-case search for dates of diagnosability would be made any less difficult than the present record suggests it is. See supra note 9.
. We note again that many of the files AMICO reviewed are likely to be the more fully developed. See supra note 9.
. We recognize that one problem with the Liberty Mutual dates that could be remedied is the inconsistency in the methods used for setting a date when the file contains no clear indication *244of the time of diagnosis. Any date in those circumstances would be relatively arbitrary, however, and we have no reason to believe AMICO is disadvantaged by the inconsistency.
. Our information regarding the Claims Facility and its operation is drawn from Eagle-Picher’s Memorandum in Support of its Motion for Relief From Judgment Under Fed.R.Civ.P. 60(b). We see no problem in basing our conclusion at least in part on the information provided in the Memorandum because AMICO does not challenge the accuracy of that information. We realize, however, that upon remand, if the parties fail to agree on the details concerning the Claims Facility, the district court may not rely on our opinion as establishing any facts but would need to receive evidence about the facility-
. Under the Wellington Agreement, participating insurance companies agreed to cover any claim if some portion of the period from the date of a claimant’s first exposure through the date of diagnosis of an asbestos-related disease fell within their policy periods. This type of insurance coverage is often called Keene type coverage, based on the decision by the District of Columbia Circuit in Keene Corp. v. Insurance Co. of North America, 667 F.2d 1034 (D.C.Cir. 1981), cert, denied, 455 U.S. 1007, 102 S.Ct. 1644, 71 L.Ed.2d 875 (1982), which adopted this method. In return for broadening the insurance companies' period of vulnerability, the insured companies agreed to dismiss with prejudice all claims for punitive and bad faith damages filed against its participating insurers.
. We note that one of the district court’s apparent goals for the certification provision was to ensure that Eagle-Picher receive no double payments on specific claims. We by no means reject this goal, although we are uncertain whether the policy requires such an assurance. In any case, because at least pursuant to the 1983 settlement agreements, payments were not made on an individual basis, it may be that any concerns about Eagle-Picher’s receiving a windfall could be resolved by means of a certification provision that does not require the company to prove that a given claim had not been paid previously but by a provision that ensures that the company has received no more money that it would be entitled to under the allocation system adopted by the district court.
. Our disposition on the merits of the Final Judgment makes it unnecessary to consider Eagle-Picher’s separate appeal of the district court’s denial of its motion under Fed.R.Civ.P. 60(b).
. On remand, the district court also should consider expressly extending its Final Judgment to asbestos claims administered by the Wellington Asbestos Claims Facility. The judgment now adopts Liberty Mutual’s diagnosis date as the date of actual diagnosis for a given claim. Because the Claims Facility apparently sets its own dates of diagnosis, the judgment as written apparently does not cover claims handled by the facility.
. We also remand for consideration by the district court AMICO’s request for a protective order requiring Eagle-Picher to refrain from destroying certain claims files.