Full opinion text
MEMORANDUM JAMES ROBERTSON, District Judge. In January 2008,1 found that the government had not succeeded in providing the accounting mandated by the Indian Trust Fund Management Reform Act, and that the record demonstrated the impossibility of rendering such an accounting. Cobell v. Kempthorne, 532 F.Supp.2d 37 (D.D.C.2008) (Cobell XX). On the basis of that ruling, plaintiffs ask for equitable relief in the nature of restitution, seeking the return of funds that have been received into the IIM trust in the years since 1887 but cannot now be proven to have been disbursed or credited to IIM account holders, plus an amount representing the benefit the government has assertedly enjoyed from having the use of those funds. See [Dkt. 3515]. An evidentiary proceeding was convened on June 9, 2008, for the purpose of considering whether such relief was warranted, and, if so, determining its dollar amount. Plaintiffs’ claims for withheld funds and for an amount representing “benefit to the government” raise significant jurisdictional and other legal issues. Those issues were briefed by the parties before the June trial, see Plaintiffs’ Memorandum in Support of Equitable Restitution and Disgorgement, [Dkt. 3515]; Defendants’ Response to Plaintiffs’ Memorandum in Support of Equitable Restitution and Disgorgement, [Dkt. 3519], but I deferred ruling on them in the belief that they would be illuminated by evidence adduced at the trial. [Dkt. 3526], At the trial, the government sought to explain the difference between IIM trust receipts and IIM trust postings noted in my January 2008 opinion, see 532 F.Supp.2d at 85-86, and both sides presented models for estimating the amount of money withheld from, or not disbursed to, IIM account holders over the years. Evaluation of those models, and of the plaintiffs’ legal theories for recovery, is the principal focus of this memorandum. Although the case no longer directly concerns the accounting question that dominated the first twelve years of its existence, the evaluation of the plaintiffs’ legal theories for recovery is necessarily influenced by what we have learned from the government’s failed effort to produce an accounting and from the many round trips the case has taken to the Court of Appeals. It is clear now that this Court has broad equitable authority to deal with a century or more of trustee nonfeasance and to fashion appropriate remedies, see Cobell v. Norton, 240 F.3d 1081, 1108-10 (D.C.Cir.2001) (Cobell VI), but it is also clear that that authority is constrained by traditional doctrinal limits on federal courts that apply in suits against the government, including sovereign immunity and separation of powers. See Cobell v. Norton, 392 F.3d 461, 473 (D.C.Cir.2004) (Cobell XIII). It is clear that the duties of the trustee and the principles of equity that govern failures to account are derived from statutes as informed by common law principles of trust, see Cobell VI, 240 F.3d at 1098-1102, but it is also clear that those statutory and common law principles are tempered by the unique nature of the trust and of the trustee. See Cobell v. Norton, 428 F.3d 1070, 1075-76 (D.C.Cir.2005) (Cobell XVII). Accordingly, methods that might be unacceptable in a typical trust case, such as statistical sampling, are available here, where I am instructed to strike a more forgiving “balance between exactitude and cost.” Id. at 1076; see also id. at 1077-79 (discussing statistical sampling). In these uncharted waters, where the trust is of enormous scope, the trustee of unusual character, and the data affected with such great uncertainty, the law of trusts is a sort of magnetic compass; it cannot be expected to point to due north, or to “map directly” onto this context. Id. at 1078. One useful if not very precise pointer provided by case law is that a trustee may not hide behind obscurity that he himself has created. See, e.g., Rainbolt v. Johnson, 669 F.2d 767, 769 (D.C.Cir.1981) (“Under established principles of trust law, if the former trustee has not kept adequate accounts, the benefit of the doubt is to be given to the beneficiary.”); GeoRge Gleason BogeRt, et al., The Law of Trusts and Trustees § 962 (2007) (“As to a trustee who fails to keep proper records of his trust it is usually stated that, ‘all presumptions are against him’ on his accounting, or that ‘all doubts on the accounting are resolved against him.’ ”). Thus, a consequence of the government’s failure to account is that evidentiary presumptions run in favor of the plaintiffs. But even this principle, like the trust duties themselves, requires compass correction in the context of this suit. The rules that identify and govern a breach of the accounting duty for a simple, 25-year trust with a single beneficiary cannot be applied, unaltered, to a 121-year old perpetual trust, managed by civil servants, with rapidly multiplying beneficiaries and a variety of ever-changing assets. Equity seeks “to do justice to all parties,” Bollinger & Boyd Barge Serv., Inc. v. The Motor Vessel, Captain Claude Bass, 576 F.2d 595, 598 (5th Cir.1978) (emphasis added)— “its orders are adapted to the exigencies of the case,” Taylor v. Sterrett, 499 F.2d 367, 368 (5th Cir.1974), and it seeks to make accurate evaluations of difficult evidence, not to provide “windfalls” for victims or punishment for wrongdoers. See Bollinger & Boyd, 576 F.2d at 598. The application of familiar equitable principles will have to be made fairly to fit the special character of this case and this trust. My conclusions, after attempting to apply a suitably adjusted set of equitable principles to the facts of this case, are that plaintiffs have properly asserted a claim for restitution; that this Court has both the jurisdiction and the power to adjudicate that claim; and that the evidence supports an award in the amount of $455,600,000, a number that is within the range of the government’s own admitted “uncertainty” about the amount necessary to restore the proper balance to the IIM trust. I have rejected the plaintiffs’ claim of entitlement to an additional sum representing “benefit to the government.” This opinion — indeed, this litigation— neither deals with nor resolves any claims that IIM account holders may have for damages against the government. And it leaves for another day the question of how and to whom the award should be distributed. I. Starting Point Two important exhibits received at the October 2007 trial, AR-171 and DX-365, appeared to show that only 77 percent of the dollars collected on behalf of individual Indians over the years had actually been posted to IIM accounts, and that, over those years, the difference amounted to a shortfall of some $3 billion. See Cobell XX, 532 F.Supp.2d at 85-86. I did not believe this to be the intended import of these exhibits, and I said so, noting the government’s mention of the role of lease deposits and other non-individual monies in the system, id. at 86, but also noting that the parties had paid only “desultory” attention to the “throughput” question I had posed at the beginning of the trial. Id. at 82. The government made it clear at the outset of the June 2008 trial that indeed it had not intended to communicate the existence of a $3 billion shortfall. The government’s task of estimating and explaining the actual shortfall was complicated, however, by the fact that it could not produce an individualized accounting and by the paucity of existing aggregate data about the IIM trust. Tr. 500:19-501:16 (Herman) (acknowledging the Court’s request for an explanation of the shortfall between receipts and postings, and attributing the difficulty to the lack of aggregate data). The government’s explanation at the June 2008 trial was, essentially, that receipts recorded in the IIM system include monies not intended for IIM accounts, but the government also conceded that, without transaction-by-transaction accounting, there is essentially no way to distinguish IIM transactions from non-individual transactions. Id. There is some historical data regarding total receipts and disbursements in the early years of the trust, but “for the most part aggregate receipt and disbursement records on IIM weren’t kept.” Tr. 784:3-5 (Angel). For many years during the early period of the trust, there is no receipt and disbursement data at all. See DX-461. Moreover, considerable evidence has been collected over the long life of this litigation, and more was adduced at the June 2008 trial, detailing the various ways in which trust systems purporting to contain receipt and disbursement data have been and still are unreliable, from qualified audits, Tr. 392:10 et seq. (Pallais), to a 73-page compendium of critiques and negative comments by politicians and auditors, each one hyperlinked to an historical document. PX-65. There was also the out-of-balance condition between Interior and Treasury records, see Cobell XX, 532 F.Supp.2d at 74-75, which lent credence to the possibility that substantial funds had gone missing over the life of the trust. II. The “IIM system” The framework for the government’s presentation was its description of how funds move into, out of, around, and through what was described as the “IIM system.” DX-370; Tr. 463:1-492:11 (Herman). Michelle Herman was the government’s principal witness on the flow of IIM funds, at least for the more recent era beginning in approximately 1972. The notion of an “IIM system” was her own construct: a kind of botanical description of the information systems and accounts she has observed and studied over a career of working with the Indian trust. Tr. 571:8-22 (Herman). In fact, many of the labels she used in describing the system are not actually used in the system, nor is the very notion of a system itself. Tr. 647:17-648:7 (Herman). Both parties nevertheless resorted to the term “IIM system” throughout the trial, using it to refer to the interlocking series of accounts, bookkeeping entries, and other data sets, used by the Department of Treasury and the Department of Interior to collect, hold, invest, track, and disburse IIM trust funds. It is a subset of a larger Indian trust “system,” which also includes the tribal trust accounts. Modern information systems, including the Integrated Records Management System (IRMS) and the more recent Trust Funds Accounting System (TFAS), are described at length in Cobell XX. See 532 F.Supp.2d at 43-44. Other components of the system, such as the general and detailed ledgers that were maintained by the Bureau of Indian Affairs and its various agencies and area-level offices before IRMS, have received less attention — except for the repeated finding that the general ledger does not reconcile with either the subsidiary BIA ledgers or the Treasury ledger. See Cobell XX, 532 F.Supp.2d at 74-75. Today, receipts that the government collects in its capacity as trustee for individual Indians are posted to the 14X6039 account at Treasury. See PX-65, hyperlink at ¶ 84, Dep’t of Interior, Audit Report to the Congress of the United States, D084-0005 to 0006 (November 1955) (“Collections made by the Bureau on behalf of these Indians are deposited into the Treasury of the United States in deposit account 14X6039, Individual Indian Money.”). When disbursements are made from the 14X-6039 account in the form of Treasury checks — as opposed to bookkeeping transfers or electronic funds transfers — they are tagged with Agency Location Code (ALC) 4844. Tr. 569:7-11 (Herman) (ALC 4844 relates to the IIM system). The 14X-6039 account exists within the Treasury General Count (TGA), so that, in the modern world, at least, cash posted to 14X-6039 becomes an asset of the government. But the balance of that account represents an offsetting liability of the government, much like a checking account at a bank. Tr. 213:20-214:18 (Miller) (distinguishing cash and funds in the 14X-6039 account). Interior is responsible for the subsidiary accounts (and sometimes just bookkeeping entries) that are recognized as part of the IIM system — from which and to which the funds in the 14X-6039 account flow. These include the IIM accounts themselves; Special Deposit Accounts (SDAs), which are used to temporarily store unallocated funds; so-called Tribal IIM accounts (IIM accounts that actually belong to tribes but are used by those tribes as a convenient checking system); and administrative accounts or account numbers used to record the collection of fees and the like. These SDA, Tribal IIM and administrative accounts and entries might be referred to as non-individual components of the IIM system. They hold or record funds which are collected in the government’s capacity as trustee but may not be owed to or intended for IIM beneficiaries. For example, an SDA might contain a deposit made by a leaseholder or a bidder that will eventually be returned, or money that is destined for a tribal trust account rather than an IIM account. In short, the “IIM system” that the government has been trying to explain and account for includes the IIM accounts. See generally infra, Section IV.A. It receives, holds, transfers, disburses and records money that is associated with the government’s role as trustee for IIM account holders, but which, by design, may never arrive in an Individual Indian Money account. III. Plaintiffs’Model Plaintiffs support their claim for an award of about $47 billion (restitution plus benefit to the government) with a model that rests upon the legal premise that any government data regarding IIM receipts should be treated as an admission, while only negotiated checks or other irrefutable evidence can be probative of disbursements. The plaintiffs’ model thus accepts the government’s estimates of receipts from the October 2007 trial, together with other government receipt data, see PX-41 at n. 1; PX-189-A at n. 1, but uses its own method to calculate year-by-year disbursements from the trust. III.A. Description of plaintiffs’ model Plaintiffs’ model begins with a calculated “disbursement rate” which is arrived at by simple division: the numerator is the value of electronic fund transfers and checks that were sent to (and cashed by) beneficiaries during the years 1988-2002, and the denominator is the total receipts for those years. See Tr. 1481:1-1493:22 (Palmer); PX-189-C (disbursement rate calculation). That disbursement rate is then applied to whatever government receipts data and estimates that can be found for all of the years of the trust dating back to 1887. See PX41; PX-189A. The process yields a spreadsheet which contains, for each year, a dollar amount for receipts, a calculated dollar amount for disbursements (receipts x disbursement rate), and a number that represents the difference between receipts and disbursements. Plaintiffs call these columns “corrected revenues,” PX-189-A at Column E, “disbursements,” id. at Column F, and “nominal benefit to government,” id. at Column G. Column G sums to approximately $4 billion in plaintiffs’ rebuttal model and is asserted to be the total dollar amount of funds withheld by the government over a period of 120 years — a number, which, after subtracting the current stated balance of the trust, leaves approximately $3.6 billion that, in plaintiffs’ submission, must be restored to plaintiffs’ accounts. Plaintiffs use the same model to calculate the benefit that they assert the government has enjoyed from not having to borrow and pay the 10-year bond rate on the annual, and accruing, “nominal benefit” amounts. The method is best described by example. In 1887, the first year of the trust, the government had not yet withheld any money from IIM accounts. Thus, the entire benefit to the government in that year is the nominal benefit in Column G. Compare PX-189-A, at Column G, FY 1887, with id. at Column J, FY 1887. The next year, 1888, the government accrues additional nominal benefit based on the difference between receipts and disbursements in that year. See PX-189-A at Column G, FY 1888. But plaintiffs then add to the 1888 nominal benefit an amount representing the value to the government of holding the 1887 benefit — not having to borrow that amount at the 10-year bond rate, see PX-189-A at Column I, FY 1888. This sum, plus the nominal benefit for 1888, plus the balance carried forward from 1887, is called the “accumulated benefit, end of year.” See PX-189-A at Column J, FY 1888. The same operation is carried forward from year to year until 2007, by which time, in plaintiffs’ rebuttal model, the accumulated benefit has grown to approximately $47 billion. The existing balance of the IIM trust accounts is deducted at the end. See generally, Tr. 268:2-272:11 (Cornell) (describing operation of the model). The plaintiffs’ model is excerpted below: As noted, the annual “nominal benefit” number in this model is driven by plaintiffs’ “disbursement rate,” which is not and does not pretend to be actual disbursements, but is rather a calculated factor that is itself driven by the plaintiffs’ legal theory about proper evidence of receipts and disbursements. See PX-189-A at n. 4 (describing disbursement calculations). The only disbursement data whose accuracy plaintiffs were willing to acknowledge were 1988-2002 Electronic Fund Transfers and Automated Clearinghouse payments (EFT/ACH) and numbers from the Check Payment Reconciliation System (CP & R). See Tr. 1481:5-12 (Palmer). Even the CP & R check data were discounted, according to plaintiffs’ own estimate of the value of checks that were never negotiated (because the value of checks that were never cashed was presumed to have remained in the Treasury). See PX-189-C at Column C & n. 5; Tr. 1482:5-22 (Palmer). Plaintiffs first compared the number of cheeks cashed to the number of checks that had been cut, see PX-56 at Column E — the percentage was calculated to be 93.75 percent — and simply assumed that this ratio could be used to calculate the dollar value of negotiated CP & R disbursements, despite the obvious likelihood that it would be the lower value checks that tended not to be deposited. Tr. 310:22-311:11 (Cornell). In their rebuttal model, plaintiffs purported to correct this obvious mistake, now comparing the dollar value of checks negotiated with the dollar value of checks cut. But the resulting ratio of 93.68 percent — remarkably similar to the first one — is derived from only a single year of data, and it disregards government data from the same document that shows that cancelled check values are only about two percent of total check values, see DX-236 at 2 (compare “checks” and “stppay” for 2003). It appears, indeed, that the dollar value of checks returned to Treasury or cancelled because they are not cashed in a timely fashion is less than one percent of the total value of checks cut. See, e.g., DX-236 at 1 (value of cancelled checks was 0.02 percent of total check value for FY 1999); DX-242 at 17 (in a representative study, value of uncashed checks was 0.17 percent of total check values); DX-275 (value of checks cancelled for limited payability were less than 0.2 percent of total check value for same period). III.B. Evaluation of plaintiffs’model Plaintiffs’ model suffers from numerous methodological flaws that were illuminated by the government’s presentation and, in many instances, are obvious to anyone having basic familiarity with the case. The most obvious flaw, perhaps, is that the size of plaintiffs calculated shortfall— $4 billion, out of total receipts of about $14 billion, see PX-189-A at Column B — is uncorroborated by any other event or data. Yes, the United States Government spills billions of dollars a year, loses money to fraud, waste, and abuse, and generally mismanages its affairs. But the Indian trust has been repeatedly audited, and while each of those audits has been qualified, see Cobell XX, 532 F.Supp.2d at 54 (discussing meaning of the many qualified audits), no audit report states or hints at the disappearance of anything close to 30 percent of trust receipts. Plaintiffs even purport to find $315 million in “nominal withholding” for the period from 2003 to 2007, see PX-189-A, despite BIA’s major improvements in accounting and record-keeping practices during that period. Whatever problems have existed in the history of this trust, and however serious the misfeasances and malfeasances of the trustees over 120 years, there has never been any evidence of such prodigious pilfering of assets from within the trust system itself. Methodologically, the two most important flaws in plaintiffs’ model are: (1) that the model includes as collections certain sums that, by definition, will not be reflected as disbursements in the CP & R data, including Osage headright payments; and (2) that the model accepts historical receipt data and disregards historical disbursement data from the same documents and systems, or makes self-serving adjustments to one side of the ledger and not the other. These flaws will be discussed in turn. III.B i CP & R data Even if plaintiffs had treated 100 percent of the checks in the CP & R data as having been negotiated, those data would necessarily under-report disbursements from the IIM system. Transfers of funds from Tribal IIM accounts or SDAs directly to tribal trust funds are effected by intra-bureau bookkeeping transfers, not by checks that would show up in CP & R data. Tr. 569:6-22 (Herman). Michelle Herman provided two examples of such “BB transfers,” see DX-480 at 9; DX-481 at 2, both of which involved millions of dollars. In a government document detailing disbursements for fiscal year 1999, DX-238 at 1, such BB transfers amount to $73.5 million of disbursements from the trust, compared to about $202 million in checks and EFTs. Plaintiffs’ experts agreed that, unless such transfers had been made by check (which they were not), they would necessarily be absent from the CP & R data. Tr. 353:20-25 (Cornell); Tr. 1566:12-1567:10 (Palmer). This basic mistake in plaintiffs’ theory — ignoring or overlooking fund transfers — had the effect of driving up the amount allegedly “withheld” and driving down the disbursement rate upon which plaintiffs relied for their calculations of “withholding” for all the years before 1988. Transfers from the IIM system are not the only transactions overlooked by plaintiffs’ single-minded reliance on CP & R data. The accounting systems relied on by Interior were designed to track money, not to aggregate throughput data. Accordingly, they register every intrafund transfer as a debit to one account and a credit to another. This (standard bookkeeping) process creates double-counting of receipts if only CP & R data is used to measure disbursements, because, while both the initial collection and the transfer will show up as “receipts,” only checks cut when the money leaves the system will show up as “disbursements.” Government contractors have been working to “map” such transfer transactions and eliminate them from the data, and their receipt and disbursement data has accordingly changed since the October trial. Tr. 481:11-482:15; 626:25-629:21; 648:19-649:15 (Herman). Reliance on CP & R data also results in overstated receipts because beginning IRMS balances registered as new receipts; government contractors have been working to eliminate those transactions as well. Tr. 624:1-626:15 (Herman). Plaintiffs continue to draw their receipts data from government figures produced during the October trial, see PX-189-A n. 1; AR-171, ignoring or refusing to accept the replacement of those figures with updated numbers, see DX-371, that have eliminated phantom receipts. They refuse to accept Ms. Herman’s reductions in receipt data because she did not “show her work.” Plaintiffs are skeptical about a reduction in revenue for the years 1986-1997 of “roughly $243 million” between the AR-171 exhibit in the October trial and the revised DX-371 exhibit produced at these proceedings, see Tr. 628:8-17 (Herman); PX-119, but they have in no way refuted Ms. Herman’s testimony, and I find it to be credible. In any case, Ms. Herman’s treatment of the data is demonstrably even-handed: disbursements have been reduced by roughly as much, if not more, over the same period. Tr. 649:7 (Herman); compare AR-171 at Column G with DX-371 at Column H for FY1986-FY1997. Again, relying on the older numbers has a two-fold skewing effect— driving up the nominal amount withheld in these years and driving down the calculated disbursement rate that controls the other years. III.B.Ü Osage headrights The error of plaintiffs’ total reliance on CP & R data as representative of disbursements is especially palpable when one considers the inclusion in their model of Osage headright monies as trust receipts. Osage headrights are the product of the Osage Allotment Act of 1906, Pub.L. No. 59-321, 34 Stat. 539, which allotted Osage tribal lands to individual members, see id. at § 2, 34 Stat. 540, but preserved the mineral estate of those lands for common management under the direction of the tribe. See id. at § 3, 34 Stat. 543. The proceeds of the mineral estate were to be held in trust for, and distributed per capita to, individual Osage Indians. Id. at § 4, 34 Stat. 544. There is a legal dispute between the parties as to whether Osage headright funds are IIM funds that fall within the class certification order in this case. See Order, [Dkt. 27] at 2-3 (class to consist of “beneficiaries of Individual Indian Money accounts”). What is beyond dispute, however, is that mineral estate proceeds destined for Osage headright holders are collected into a tribal trust fund and disbursed from that fund directly to individual headright owners, without ever passing through the IIM system. Tr. 473:22-474:1 (Herman). Osage headright disbursements are not to be found in CP & R disbursement data, and both of plaintiffs’ experts agreed that Osage mineral estate revenues did not belong in IIM system receipt data. Tr. 1577:8-16 (Palmer); 334:7-23 (Cornell), see, e.g., DX-372 at 2025 et seq. Nevertheless, plaintiffs’ model “corrects” government receipt data by adding in the value of Osage headright payments for every year. See PX-189-A at Column E. The effect of this mistake, once again, is both to add over $800 million in erroneous receipts and to drive down plaintiffs’ calculated disbursement rate, see Tr. 1596:6-17 (Palmer), causing the inaccuracy to metastasize. III.B.iii Data “adjustments” Plaintiffs’ demonstrated willingness to accept data they liked and reject data they disliked did not enhance the credibility of their model. Their initial model would accept government receipt data from a historical document, but reject disbursement data from the same document in favor of their calculated disbursement rate. See Tr. 291:18-292:23 (Cornell). Their rebuttal model continued to make self-serving modifications to the historical data, some driven by their calculated disbursement rate, see PX-189-A at n. 1, and some even more baseless. See generally, Tr. 1578:24-1584:19 (Palmer). Two examples follow. First, plaintiffs’ adjustment of CD & L data: Aggregate government data for the years 1972-1987 were developed by the firm of Chevarria, Dunn, & Lamey (CD & L) from general and detailed ledgers for the various BIA agencies and area offices. See Tr. 604:7-623:12 (Herman). Plaintiffs accept the receipt data, but, asserting (with some basis) that the disbursement data may be flawed, incomplete, or estimated, see id., they modify the disbursement data. See PX-189-A at n. 4. The modification is calculated by dividing the government’s reported disbursement rates from the years in which there is CP & R data (approximately 101 percent) by the plaintiffs’ calculated disbursement rate (74.45 percent). This yields, for those years, a so-called “disbursement adjustment ratio” of 134.1 percent, which appears to represent nothing more than the factor by which the plaintiffs do not trust government disbursement data, but which is applied to make a wholesale downward adjustment of CD & L disbursements data for the years 1972-1987, as well as for the years 2003-2007, when receipt and disbursement data come from qualified audits, and for 1955, where receipt and disbursements come from a historical, audit report. See PX-189-A at n. 4; see also PX-53 (1955 Report). Second, plaintiffs’ adjustment of receipts data located by Morgan Angel: Historical evidence regarding receipts and disbursements from the period 1923-1949 was developed by the government’s historian, Edward Angel of the firm Morgan, Angel and Associates. Tr. 787:23-792:22 (Angel). These reports were produced in response to a 1906 appropriations act which required the heads of departments to prepare reports regarding any receipts and disbursements that were not paid into the Treasury. See Pub.L. No. 59-383, § 5, 34 Stat. 746, 763 (1906). These reports generally bore the title: “Statement of Moneys Received and Expended by Disbursing Agents of the Indian Service During the Fiscal Year [ ] Without Being Paid into the General Treasury of the United States.” See, e.g., DX-426 at 3; DX-27 at 14. Each has an entry for IIM receipts and disbursements. Dr. Angel suspected that these reports would exclude both receipts and disbursements that did not flow into or out of the Treasury. Tr. 790:6-791:10 (Angel). This would make sense, as the statute requires “a detailed account of all payments, made from such funds,” where such funds refers to funds “not paid into the General Treasury.” Thus, if there is any understatement on these reports, it likely encompasses both receipts and disbursements. Nonetheless, the plaintiffs’ model increases the receipts — and only the receipts — by dividing them by 77 percent, the figure that represented the gap between receipts and postings in DX-365. Tr. 1581:7-13 (Palmer). This exhibit from the October 2007 trial was prepared to discuss what percentage of dollars would be covered by the government’s accounting procedures, and had nothing to do with the percentage of funds that might have been collected and not paid into the Treasury, a detail about which the plaintiffs’ witness (a last-minute surrogate) was evidently unprepared to testify. Tr. 1578:24-1584:19 (Palmer). Dr. Palmer’s ultimate defense of the “adjustment” he had made was that it was similar to the calculated disbursement rate, Tr. 1586:16-32 (Palmer), which — as has already been discussed— plaintiffs’ model has artificially deflated. My overall conclusion about plaintiffs’ model is that it cannot be used as a representation or even an estimate of the amount of trust funds that the government has failed to disburse or post to IIM accounts. Instead of providing unbiased opinions, plaintiffs’ expert witnesses essentially provided plaintiffs with a way to put a dollar value on their argument that all data that favors the plaintiffs may be treated as admitted, and all data that disfavors them must be proven by the government with discrete, transactional evidence. IV. The Government’s Case The government’s presentation had two parts. First, the government endeavored to explain how DX-365, the chart it introduced in the October 2007 trial, could state that postings to IIM accounts were 77 percent of collections without admitting that the other 23 percent of the funds were unaccounted for. Second, the government developed its own model for evaluating the amount of funds that might reasonably be considered missing from IIM accounts. IV.A Collections and postings to IIM accounts Ms. Herman’s “IIM system” testimony has been discussed in previous sections of this memorandum. Two of her explanations for the difference between collections and postings — the double-counting of opening balances and intrafund transfers — have already been covered. The rest of the explanation is provided by Ms. Herman’s testimony that a significant amount of the funds that are considered “receipts” of the IIM system are never intended for an IIM account. Tr. 487:23-488:10 (Herman). As described above, the IIM system includes many non-individual accounts, including SDAs and administrative accounts. See supra, Part II. Using record examples, Ms. Herman described three types of disbursements that might be made from these non-individual accounts without money passing through any IIM trust. See generally, DX-370 (graphically displaying the flow of funds). First, funds might be transferred directly from an SDA or an administrative account to third parties. Tr. 517:25-518:14 (Herman). Such funds might include bid deposits associated with contracts, earnest money associated with land sales, deposits on leases, and administrative fee payments including fees to the government, among other things. Id. One example, from the modern, electronic era, was a $5.2 million disbursement from an SDA to an insurance agency on behalf of a tribe. See DX-474; Tr. 505:18-506:23 (Herman). Another example, one that straddled the paper and electronic eras, was a $3000 performance bond, deposited in 1978 and disbursed in 2006, with interest, as $16,767.16. DX-491; Tr. 534:20-536:15 (Herman). A 1941 audit report showed special accounts for a cemetery fund and for school activities. See DX-486 at 2. And a Commissioner’s report from 1910 showed that $2.7 million out of $7.6 million in disbursements represented returns of funds to unsuccessful bidders. DX-33 at 2; see also DX-32 at 7 (1909 Report). The government’s conclusion that significant funds moved from special accounts or administrative accounts direct to third parties without posting to IIM accounts appears well-founded. A second instance in which money received into the IIM system would not be posted to IIM accounts would be transfers from SDAs and administrative accounts into tribal trust accounts. Tr. 485:25-486:6 (Herman) (transfers to tribal trust accounts were a “significant amount of money”). One example was a timber transaction that moved through an SDA, with $4.46 million of the total transferred to tribal trust accounts (a non-check, “BB” transfer), $563,000 used for a reforestation payment (a third party disbursement), and only $609,000 posted to IIM accounts. See DX-480; Tr. 521:21-525:19 (Herman). (Significant amounts of so-called “Tribal IIM” have passed through the system, although these amounts were not included in DX-365.) The government’s conclusion that there was money moving through the system that was intended for the tribes and not for IIM accounts is supported by the evidence. Finally, there are payments to what Ms. Herman described as “stakeholders.” Tr. 486:7-487:4 (Herman). Stakeholder payments are payments made to individual Indians, often — but not always- — IIM account holders, that do not move through IIM accounts. For example, a tribe may have a judgment account or an account for per capita distribution, which amount is deposited in the IIM system is the form of a non-individual account. A large portion of that money then goes directly to stakeholders by check, without touching IIM accounts, while some of the funds — typically, for minors and the like — do go to IIM accounts. The government presented just such an example — a Per Capita account of $43.5 million where $7.8 million was transferred into “land-based” IIM accounts while $27.5 million was paid direct without touching an individual account. See DX-475; Tr. 507:8-510:3 (Herman). Judgment and Per Capita funds go to tribe members as such, not to “land-based” IIM account holders. Judgment and Per Cap-ita accounts themselves only exist inside the IIM system for bookkeeping and check-writing convenience. Amounts that are transferred into IIM accounts are typically for beneficiaries who are minor, incompetent, or impossible to locate. Again, the explanation that stakeholder payments were reflected in IIM system receipts, but not in individual account postings, appears to be well supported. Ms. Herman’s flowchart, and her testimony and documentary examples of non-IIM payments from non-individual accounts, together with the evidence that receipts can be double-counted, went a long way towards explaining the nature of the discrepancy between receipts and postings in DX-365. Her explanation did not quantify the discrepancy, however: Ms. Herman admitted that there was no ready way to determine what percentage of receipts were for stakeholders, for tribal trusts, and for third parties. Such a determination is not impossible, but would be very time consuming, would require finding and reviewing financial documents, and has not been done. See Tr. 641:23-642:16 (Herman); Tr. 636:15-24 (Herman) (same regarding intrafund transfers). IY.B. The government’s model After the government’s admission in the October 2007 trial that perhaps $3 billion of IIM system receipts had not been posted to IIM accounts, my finding that the government had not and could not provide an adequate accounting of its IIM trusteeship, and plaintiffs presentation of at least a theory that some $4 billion of IIM funds had never been disbursed to IIM account holders, the burden of quantifying and explaining the shortfall shifted to the government. Because Ms. Herman’s explanations of the various ways that IIM system receipts might be transferred or disbursed without passing through IIM accounts was neither quantified nor reasonably quantifiable, and because there is no aggregate IIM receipt and disbursement data for considerable portions of the history of the Indian trust, the government resorted to a statistical modeling approach. IV.B.i Explanation of the government’s model The government’s model was developed and presented by an expert statistician, Frederick Scheuren. It uses available receipt and disbursement information and then employs a statistical technique known as “multiple imputation” to fill in the blanks, which are many. Tr. 929:12-25 (Scheuren). Multiple imputation is an established technique that uses available data to impute missing data, but instead of yielding a single value that best fits the data as modeled, it uses modern computing power to impute multiple values for each missing data point. Tr. 932:9-23 (Scheu-ren). Its chief value in this regard is that it allows the statistician to “see the uncertainty” that exists because of the missing data. Id. In the words of the Dr. Scheu-ren, “missing data is not free,” Tr. 934:1-3, meaning that, the more data that is missing, the lower the confidence that can be placed in the values the model generates. Dr. Scheuren used the following procedure to create his multiple imputation model. See generally DX-460 (outlining steps in government’s approach). First, his firm (NORC) assessed the available data. This involved four kinds of data: historic reports that Dr. Angel had uncovered for 1909-1911, 1922-1949, and 1955; data from the general and detailed agency ledgers identified by CD & L for 1972-1985; IRMS data from 1986-1995; and audit report data from 1996 to the present. See DX-372; Tr. 939:14-940:7 (Scheuren). This left a significant number of years for which there was no data. See DX-471 (NORC’s data set). NORC also assessed the data set for data points that did not appear to make sense, and treated “outliers” as missing values. See DX-461; Tr. 942:22-943:17 (Scheuren). The resulting data set represented the information that NORC would use to try to assess the likely values of — and resulting uncertainty from — the missing receipt and disbursement data. The next step involved using scatter plots and statistical techniques to assess the relationship between available variables. This is an exploratory step where different variables are compared using familiar statistical techniques, such as regression analysis, to assess their correlation with each other and the explanatory force they might have in a statistical model. Tr. 947:6-948:16 (Scheuren). NORC’s model employed five variables: (1) fiscal year; (2) receipts; (3) disbursements; (4) reported trust balance; and (5) the value of one Osage headright. Receipts and disbursements were chosen because (obviously) they were the values that NORC was attempting to study; balance and head-right value were chosen because exploratory statistical testing showed them to be correlated to the other variables. Neither correlation is surprising: balance information is quite likely to be related to the amount that enters and leaves the trust each year, and the value of an Osage head-right, while not strictly related to IIM, is something of a surrogate for the overall condition of the economy, especially useful as a metric for the value of certain oil and gas assets that might powerfully affect Indian trust data. Tr. 998:24-999:7 (Scheuren). Osage headright value was also used because its value was known back to the first year of the trust. Tr. 948:23-949:9 (Scheuren). It was thus an important variable in the model. Tr. 948:15-16 (Scheuren) (“[T]he Osage variable was really the main variable we had.”) Once the modeled relationships between the variables were established, NORC performed its multiple imputation step using a statistical application called SAS. Tr. 949:22-950:9 (Scheuren). Ten thousand values were imputed for each missing data point. Tr. 952:14-24 (Scheuren). NORC produced a data set reflecting the average or “point estimate” for each data point, see DX-492, but hiding behind each of these values are all 10,000 imputations, which are carried forward and affect the uncertainty in the data that is calculated by the computer program. Tr. 937:18-22 (Scheu-ren) (data set is completed “in a way that would allow us to measure its uncertainty.”). Next, NORC used a second model, a “time series model,” to assess the fit between all the imputed and fixed values in the newly completed data set. See generally, DX-460; Tr. 953:17-961:24. This “remodeling” was done in large part because NORC recognized that even much of the data treated as fixed “had its own problems,” and that “that uncertainty needs to be incorporated into this calculation.” Tr. 954:4-7 (Scheuren). The time series model assessed the strength of the relationship between data in a given year and data for the previous seven years — a kind of rolling average. Tr. 957:9-959:10 (Scheuren). This was done all the way up until 1996, when IIM data began to be regularly audited, and so it helped to assess the uncertainty underlying the CD & L data from the general and detailed ledger, and the data from the IRMS system. See DX-462 at 3 (model adjustments end in 1996). This time series modeling did not affect point estimates very much, Tr. 961:23-24 (Scheuren); it was essentially a search for irregularity over time, which was scored as uncertainty and which caused the uncertainty in the model to increase by about one-third. Tr. 961:9-22 (Scheuren). This uncertainty manifested itself in the final step of the model, which involved calculating an average difference between overall receipts and disbursements, and then creating a distribution of values around that mean based on the amount of variance and uncertainty that was detected through the previous steps. Tr. 964:11-971:19 (Scheuren) (describing final step). The resulting histogram, DX-463, shows that the calculated mean would be $583.6 million, which is $159.9 million more than the current stated balance of the trust. See DX-464; Tr. 967:22-970:13 (Scheu-ren). Because there is so much uncertainty in the data, however, the histogram displays a wide variance. Tr. 970:21-24 (Scheuren). The 95 percent confidence interval is wide enough that it encompasses a zero difference between the calculated and stated balance of the trust, meaning that the current, stated balance could very well be exactly correct. Id. (“[T]he values are spread so widely that you cannot reject [the] hypothesis that the data could be ... consistent with the $423 million [stated balance].”). But that variance cuts both ways: At the 95 percent confidence interval, the histogram encompasses a shortfall between IIM receipts and IIM disbursements of up to $365.7 million. Tr. 970:25-971:6 (Scheuren); DX-464. The 99 percent confidence interval would accommodate a finding of a shortfall of up to $455.6 million. Tr. 972:21-973:9 (Scheuren). IV.B.Ü Evaluation of the government’s model The greatest strength of the government’s model is that its outcome comports with what is actually known about the IIM trust. First, the $159.9 million gap between the calculated mean and current stated balance of the IIM trust may be taken as a sort of admission that the correct balance of the IIM trust is greater than its stated balance. Second, the wide variance of possible outcomes around the mean is consistent with the one thing that 12 years of this litigation has truly settled, which is that there is still much that is unknown about the trust. Third, the fact that the exact current stated value of the trust lies within the confidence interval reflects the reality that, in the absence of some kind of equitable evidentiary presumption in favor of the plaintiffs, one permissible conclusion from the record would be that the government has not withheld any funds from plaintiffs’ accounts. Indeed, despite a profusion of evidence and opinion about the unreliability of IIM records, there has been essentially no direct evidence of funds in the government’s coffers that belonged in plaintiffs’ accounts. And finally, the very large value that represents the high end of the 99 percent confidence interval reflects the government’s frank acknowledgment that uncertainty is not free. That number is best understood as an admission that the range of uncertainty about aggregate IIM trust date includes the possibility that the return of $455.6 million is necessary to correct the trust’s balance. Crediting all the uncertainty to the plaintiffs — as I do^ — • results in the conclusion that the balance of the trust should be roughly double its current stated amount. Not only do I find the government’s model plausible, but I find that it is based on the sound and principled use of available data. Dr. Scheuren was the only statistician to testify about these matters. I found his testimony to be extremely knowledgeable and credible — he was forthcoming and cooperative, and willing to acknowledge any mistakes and weaknesses in the model or underlying data. I was also impressed by the extent to which Dr. Scheuren was responsive to concerns expressed by the plaintiffs. He appears to have exercised his expertise free from direction by defense counsel, and — in contrast to plaintiffs’ recently retained experts — he brought a considerable amount of experience within the case to bear on his modeling. I credit his testimony that multiple imputation was a sound method of evaluating the impact of missing data on the uncertainty regarding trust balances. I also found that the model was methodologically fair and carefully employed. Many of the outliers that NORC excluded seemed to favor the government. See DX-461 (low throughput in FY 1922 excluded as outlier; low trust balance in 1926 excluded as outlier). And any data treated as missing favored the plaintiffs, because it increased the uncertainty in the model. The model used essentially all the available historical data without regard to whether it favored either side, and did not make self-serving adjustments. Unlike plaintiffs’ model, it incorporated the latest data mapping, which reduced both receipts and disbursements in the modern period. Although the government could have argued that the data include monies not intended for individual Indians and attempted to exclude those values, see supra, Section III.B., it recognized that it was difficult to disentangle those transactions, and instead relied on overall receipt and disbursement data. Tr. 641:23-642:16 (Herman). Above all, I thought the model’s use of available data was evenhanded — designed to illuminate what little is known about the trust, not to argue a legal theory about certain monies that should or should not be counted. This is not to say that the government’s model was perfect. Although it was designed to evaluate uncertainty even in the existing data, it may not have adequately accounted for just how unreliable the underlying data was. NORC treated Dr. Angel’s data from 1928-1949 as complete, even though Dr. Angel himself believed that both receipts and disbursements might be understated. See supra, Section IILB.iii. The ledger data from CD & L that was treated as fixed may have been incomplete, and it certainly reflected estimation techniques, because some data fields for some agencies in some years were filled in using data from previous years. See Tr. 609:4-623:9 (Herman). Dr. Scheuren was impressed by the weaknesses demonstrated in the CD & L data by plaintiffs’ counsel, and although his model already attempted to evaluate the uncertainty in that data, he suggested that it might be appropriate to increase the confidence interval from 95 percent to 97.5 percent in response to these revelations. Tr. 974:15-976:3 (Scheuren). The record in this case is replete with indictments of the reliability of IRMS data, which NORC’s model employed from 1986-1995. Also, the model does not produce point estimates that “foot” — if you take the balance in year T and add estimated receipts and subtract estimated disbursements for year T + l, you typically do not get the imputed balance for year T + l. The fact that the balance data is used to model the imputed values, but does not fit perfectly, indicates that the model itself has considerable built-in uncertainty. Finally, the audit data from 1996 forward was not even subjected to the time series remodeling step that added uncertainty to the model, even though these numbers reflected only qualified audits that continued to express concerns about system reliability. But whatever criticisms plaintiffs had about the government’s model, they were not able to prove any bias that would render it fundamentally unsound. Thus, for example, plaintiffs showed that about $580 million of the disbursements in the CD & L ledger data reflected substitutions, where the agency data for that year were filled in using the data from the previous year, see Tr. 621:9-12 (Herman), but review of the ledgers shows that in years where disbursement data was substituted, receipt data was also substituted at those same agencies. See DX-372 at tabs GLDL-R and GLDL-D. And if Dr. Angel’s data for 1923-1949 under-reported receipts, they likely under-reported disbursements as well. Plaintiffs argue that the fit between the Osage variable and the other variables is suspect because Osage revenues are oil and gas driven and have nothing to do with timber and other revenue streams. But Osage headright values reflect' a variable that is internal to the Indian trust system and that incorporates information about economic well-being. Moreover, the existence of a fit between two variables is simply a matter of statistical correlation, and if the fit is not very strong, the result is to benefit the plaintiffs by adding uncertainty to the model. None of these arguments identifies systematic bias in the data, and whatever omissions the plaintiffs were able to identify are minor when compared with plaintiffs’ model, which is systematically designed to misrepresent the data in the plaintiffs’ favor. In sum, the government’s model is imperfect, but it presents a plausible estimate of funds withheld, and it is particularly useful in evaluating the uncertainty in existing trust data. Dr. Scheuren credibly testified that the strength of his model lies, not in establishing point estimates for certain values in certain years, but in evaluating “overall uncertainty at the balance level.” Tr. 1027:11-12 (Scheuren). Y. “Benefit to the Government” In plaintiffs’ submission, every dollar of Treasury receipts that was intended for an IIM trust beneficiary but neither disbursed nor posted to an IIM account was a dollar the government did not need to borrow. Plaintiffs’ calculated “benefit to the government” thus grows their calculated “nominal benefit” number by adding interest at the 10-year bond rate, compounded annually. In their rebuttal model, the effects of compound interest transform their $3.5 billion nominal withholding number into some $46.8 billion of benefit conferred. That submission is rejected, for the reasons set forth below. First, as explained earlier in this memorandum, plaintiffs’ model contains plain inaccuracies, and the compound interest factor only compounds them. For example, plaintiffs’ incorrect inclusion of about $660,000 in Osage funds for the years 1887-1889 results in more than $44 million of (erroneous) calculated benefit to the government (to say nothing about the way that plaintiffs’ misuse of Osage data also hypes the model by driving down its calculated disbursement rate.) Second, plaintiffs presented no evidence that their theoretical benefit to the government was anything other than a theory. Plaintiffs’ only witness with knowledge of the borrowing practices of Treasury was Dr. James C. Miller, former director of the Office of Management and Budget. All he could say of plaintiffs’ theory regarding saved borrowing costs was that it was sound as a matter of economics and common sense; he said so on the basis of no personal knowledge of the day-to-day mechanics of borrowing decisions at Treasury. Tr. 253:21-254:2 (Miller). Plaintiffs’ theory assumes that any cash in the government’s hands benefits the government regardless of where it is held, but a government witness with knowledge of Treasury borrowing practices testified that cash and investments held outside the Treasury General Account are not considered in Treasury borrowing decisions, Tr. 1245:21-1246:13 (Grippo); plaintiffs believed otherwise, but they did not produce a witness with personal knowledge who could dispute this testimony. Historian Terence Kehoe testified for the government that, especially in early periods, significant amounts of IIM were held outside the Treasury. See generally Tr. 1077:4-8, DX-497. (How much weight to accord that testimony is unclear, because it was evident that Dr. Kehoe could not testify confidently about precisely how and where IIM monies were historically held.) Suffice it to say that the evidence produced at trial did not illuminate whether and how any benefit would have accrued to the government from the withholding of IIM monies, especially during the early periods of the trust. Plaintiffs adduced no evidence that the Treasury ever decided to make an actual investment of IIM funds to the government’s own benefit and produced no evidence of any decision not to borrow money on account of IIM funds. All that has been put forward is a general theory that, by inexorable operation of the government’s cash management system and status as a perpetual debtor, every undis-bursed IIM trust dollar that the government holds is a dollar the government does not have to borrow. One need not have the mind-set of a Creationist to reject that theory as unproven. VI. Jurisdiction The accounting that Congress mandated and that plaintiffs demanded in this suit would have documented the management and disposition of allotted assets over the life of the trust. Such an accounting would have allowed individual plaintiffs to determine whether they had viable claims for damages — for failure to collect income, enforce lease terms, or make payments on direct pay contracts; for misfeasance in negotiating sales or leases at below-market prices; or for negligence in allowing funds to be lost or stolen. I found, however, that Interior’s 2007 accounting plan did not provide such an accounting and, that providing plaintiffs with the “instrumental right” contemplated by Congress, see Cobell XIII, 392 F.3d at 468, would be prohibitively expensive and, as a practical matter, impossible. The trustee’s irremediable breach of its accounting duty has unquestionably harmed individual plaintiffs (if not necessarily the plaintiff class): their putative damages claims have been prejudiced by the impossibility of assembling accurate data about the disposition of their assets. That harm cannot be remedied by a monetary award in this Court, however, because any such relief would be a substitute for the accounting itself and beyond this Court’s jurisdiction to award. Alternative remedies that involve monetary awards against the government are difficult to come by in the district courts. Our jurisdiction over non-tort claims for damages against the government is conferred by 28 U.S.C. § 1346(a)(2). It is concurrent with the jurisdiction of Court of Federal Claims for disputes up to $10,000, but, for claims exceeding that amount, the Court of Federal Claims has exclusive jurisdiction. See 28 U.S.C. § 1491(a)(1) (providing same jurisdiction as § 1346(a)(2) without $10,000 limit). The only other applicable statute that can provide the necessary waiver of sovereign immunity is 5 U.S.C. § 702, within the Administrative Procedure Act, which allows for an “action in a court of the United States seeking relief other than money damages.” Id. (emphasis added). If they are to recover a monetary award in this case, the plaintiffs must fit their claim within that provision. An award of money is not always “money damages,” see Bowen v. Massachusetts, 487 U.S. 879, 108 S.Ct. 2722, 101 L.Ed.2d 749 (1988), but the Supreme Court has recently recognized, and with increasing emphasis, that “[a]lmost invariably ... suits seeking (whether by judgment, injunction, or declaration) to compel the defendant to pay a sum of money to the plaintiff are suits for ‘money damages,’ as that phrase has traditionally been applied.” Great-West Life & Annuity Ins. Co. v. Knudson, 534 U.S. 204, 210, 122 S.Ct. 708, 151 L.Ed.2d 635 (2002) (citing Justice Scalia’s dissent in Bowen, 487 U.S. at 918-919, 108 S.Ct. 2722). Even under Bowen and its progeny, § 702 allows for a monetary award against the government only insofar as money may be “the very thing to which [the plaintiff] was entitled” — that is to say, the non-payment of money must be the very unlawful act of which the agency is accused. See, e.g., Dep’t of the Army v. Blue Fox, Inc., 525 U.S. 255, 262, 119 S.Ct. 687, 142 L.Ed.2d 718 (1999). If the money is sought as compensation for another harm, and not as “specific relief,” then it is money damages, and it is not within the waiver. Id. Plaintiffs are aware of the jurisdictional wire they walk, and they carefully do not seek compensation for the accounting they have not received. Instead, they demand the restitution of monies collected for their benefit that the trustee has failed to distribute to them or to post to their IIM accounts. They also demand disgorgement of an amount of money that represents the “benefit to the government” of the government’s having the use of their undisbursed money. The jurisdictional question is whether the claim for money withheld, or the claim for “benefits to the government,” or both, are claims for “specific relief,” within the waiver of § 702 and within the jurisdiction of this Court. VI.A Money withheld The duty of a trustee to collect trust revenues and allocate them to the beneficiary and not to himself is as plain as any trust duty, see Restatement (ThiRd) of Trusts § 84 (“The trustee has a duty to see that trust property is designated or identifiable as property of the trust.”); bo-gert, § 541 (“The trustee has the duty to collect and preserve the property made subject to the trust”). The failure faithfully to perform that duty is a breach of trust, the remedy for which is the restitution of withheld or misplaced money to the beneficiary’s account. That money is not a substitute remedy for a failed accounting: it is the trust property itself. It is “the very thing to which [the b