Before KING, HIGGINSON, and COSTA, Circuit Judges.GREGG COSTA, Circuit Judge:Medicare incentivizes services, not results. Its fee-for-service model risks not only that beneficiaries may receive treatment they do not need, but also that they may miss out on less expensive treatment that might help. See Medicare Payment Advisory Comm’n, Report to the Congress: Reforming the Delivery System 7 (June 2008). To test whether other approaches could be more efficient, Congress authorizes “experiments and demonstration projects” deviating from the ordinary Medicare reimbursement rules. 42 U.S.C. § 1395b-1. This case arises from one such demonstration project, with a twist: Texas Tech Physicians Associates could keep the additional fees it received for implementing the project only if its care management model achieved cost savings. The government says Texas Tech failed, and so demands return of roughly $8 million in fees. Texas Tech resists that demand.I.One of the alternative payment models the Department of Health and Human Services may explore is whether “payments for services other than those for which payment may” ordinarily be made might reduce costs and improve “utilization of services.” 42 U.S.C. § 1395b-1(a)(1)(B). Using that authorization, the Center for Medicare and Medicaid Services (CMS, the agency within HHS that administers Medicare) solicited proposals for a “Care Management for High-Cost Beneficiaries Demonstration.” CMS was looking for projects to test whether “intensive management” of patients and coordination between providers might improve care and reduce costs for Medicare beneficiaries with substantial medical needs. On top of the normal fees for services, participating programs could earn monthly fees contingent on achieving cost savings of 5%. These extra fees would count as costs.Texas Tech Physicians Associates—together with Texas Tech University Health Sciences Center and Trailblazer Health Enterprises—submitted a proposal. The project would employ a “system of telephone contacts, letters, site visits and physician contacts” to help manage and coordinate care for high- risk beneficiaries treated at Texas Tech’s facilities. Interventions would range from home safety assessments to support groups, depending on patients’ risk levels. The application indicated it expected to achieve Medicare cost savings of up to 25%.CMS took Texas Tech up on its proposal, and the parties signed a “demonstration agreement” in January 2006. Texas Tech agreed to implement the project “as proposed in the demonstration application.” In exchange, it would receive a monthly management fee of $117 per participating beneficiary. And as required by the solicitation, it agreed to be responsible for failure to meet the 5% net cost savings figure, up to the amount it would receive in management fees.The difficulty—then and now—was selecting an adequate control group. The control group was important because cost savings would be determined by comparing the costs of patients in the control group with those of Texas Tech’s participating patients, referred to as the intervention group. The demonstration agreement did not itself identify a control group. Instead, CMS would “contract with an independent evaluator” to “assist CMS in designing features for a suitable comparison group.” After the contractor proposed a methodology for matching a control group to the intervention group, Texas Tech would have “the opportunity to review and agree” before beginning the project.Agreeing on a control group proved something of a sticking point. Following an initial proposal from a contractor hired by CMS, RTI International, Texas Tech raised various concerns about the populations from which the control group would be drawn. RTI responded with changes to its initial approach.Following months of back-and-forth, RTI sent an April 2006 memo summarizing a “physician group practice-based” approach. RTI used a “loyalty algorithm” to identify patients whose loyalty to a practice was similar to that of Texas Tech’s patients. Those beneficiaries would be eligible for inclusion in the control group if they were within the 48-county area served by Texas Tech and had high Medicare costs or serious diseases. RTI would then randomly assign eligible beneficiaries to the control group such that, according to historical cost data, it was made up of low-cost, medium-cost, and high-cost patients in percentages nearly identical to Texas Tech’s patient population. Texas Tech reviewed and accepted the April 2006 methodology and began enrolling participants that month.The trouble began less than a year later. In February 2007, Actuarial Research Corporation (another CMS contractor) released a report on the first six months of Texas Tech’s project. After adjusting for baseline cost differences, the intervention group’s monthly per-member costs were 7% higher than the control group’s. Texas Tech chalked the cost differences up to inconsistencies between the groups. It pointed, for instance, to two features of the intervention group that it believed explained its higher costs: more deaths and more patients residing in nursing homes. In light of those differences, Texas Tech asked that the control group be adjusted. But CMS declined; it said doing so was “not feasible” and its investigation of disparities between the groups did not reveal “any reasonable adjustments that could produce a match . . . that mitigated disparities.”Texas Tech then terminated the program effective July 31, 2007. The demonstration agreement permitted early termination, in which event it required a prompt “final reconciliation.”Actuarial Research Corporation issued a reconciliation report a year later. It concluded that (1) Texas Tech had received $7.99 million in administrative fees; (2) Texas Tech’s savings guarantee required just over $10 million in net savings—$2.05 million in cost reductions[1] plus the $7.99 million in fees; and (3) the intervention group’s Medicare costs were $6.79 million higher than the control group’s. This meant a “savings shortfall” of $16.79 million, exceeding the contingent monthly fees Texas Tech had received and requiring their return in full.Unhappy with that result, Texas Tech hired its own consultant to investigate disparities between the intervention and control groups. Its report raised concerns about whether the control group’s historical cost trends matched the intervention group’s, and whether differences in death rates made the groups a poor match. CMS asked RTI to look into those concerns, but RTI concluded that the differences between the groups “cannot account for the estimated Texas Tech intervention effect of higher, not lower, cost growth.”CMS sent a letter to Texas Tech in March 2009 requesting return of the $7.99 million. Texas Tech demurred. Four years passed before CMS sent Texas Tech another letter in September 2013, renewing its demand that Texas Tech return the money. That letter said it would represent CMS’s “final decision” unless Texas Tech appealed to HHS’s Departmental Appeals Board.Texas Tech did appeal to the Board, only to promptly ask for dismissal of its appeal. It argued that the demonstration agreement was a procurement contract and thus governed by the Contract Disputes Act, 41 U.S.C. §§ 710109, which would mean the Board was not the proper forum for resolving this dispute. It also argued that the Act’s six-year statute of limitations barred CMS’s delayed attempt to collect the debt. The Board denied Texas Tech’s motion to dismiss, holding that the demonstration agreement was a grant agreement rather than a procurement contract.In the merits appeal that followed, Texas Tech raised a variety of contract theories to avoid CMS’s demand for repayment, including three alleged breaches of the agreement as well as common law defenses like mistake and impossibility. In a thorough opinion, the Board rejected those arguments in part because it determined that “[c]ommon law contract theories are inapplicable” to grant agreements. In light of this view, whether CMS breached the agreement mattered only insofar as the demonstration agreement conditioned Texas Tech’s repayment obligation on CMS’s performance. And in addition to finding that none of Texas Tech’s allegations would have affected its duty to meet the cost savings target, the Board determined that CMS complied with the agreement.Texas Tech sought further review by suing HHS under the Administrative Procedure Act, asking the district court to set aside the Board’s decision and enjoin CMS from enforcing its demand.[2] On competing summary judgment motions, the district court ruled that Texas Tech had not met its burden under the APA and granted HHS’s motion. Texas Tech appeals.II.Texas Tech first argues that the HHS Departmental Appeals Board should not have heard this case. That is because, in its view, the demonstration agreement was a procurement contract governed by the Contract Disputes Act. See 41 U.S.C. § 7102(a)(2). If that statute applies, it affects where and when this dispute should have been resolved. Disputes over government procurement contracts start with presenting a claim to the appropriate contracting officer. Id. § 7103(a). Any such claim—whether by the government or a private party—must be presented within six years. Id. § 7103(a)(4)(A). If a party to the contract disagrees with the contracting officer’s decision, it may appeal to the Civilian Board of Contract Appeals or the Court of Federal Claims, id. §§ 7104, 7105(b), but not to “any [other] forum, tribunal, or Federal Government agency,” id. § 7103(g). So if the demonstration agreement was a procurement contract, then the government may have been late in seeking return of the fees, and the Board lacked jurisdiction.But none of the Contract Disputes Act’s requirements matters if the Board correctly determined that the demonstration agreement was not a procurement contract but instead a grant agreement. See id. § 7102(a)(2) (specifying the types of government contracts to which the CDA applies, including those for the “procurement of services”); Coastal Corp. v. United States, 713 F.2d 728, 730 (Fed. Cir. 1983) (“[The CDA] does not cover all government contracts.”). The Federal Circuit, well versed in government contract disputes, looks to the Federal Grant and Cooperative Agreement Act when characterizing government agreements. See, e.g., Hymas v. United States, 810 F.3d 1312, 1325-26 (Fed. Cir. 2016). So do the parties.That statute says executive agencies must use procurement contracts when:the principle purpose of the instrument is to acquire (by purchase, lease, or barter) property or services for the direct benefit or use of the United States government.31 U.S.C. § 6303(1). An example is a Navy contract for purchase of a ship. The government gets to use what it purchased.In contrast, agencies must use a grant agreement when:the principle purpose of the relationship is to transfer a thing of value to the State or local government or other recipient to carry out a public purpose of support or stimulation . . . instead of acquiring (by purchase, lease, or barter) property or services for the direct benefit or use of the United States Government.Id. § 6304(1). An example is an agreement with a school district to provide free school lunches. The federal government pays for the lunches, but students receive the benefit.We agree with the Board that the demonstration agreement was not a procurement contract. The “direct” benefits of the agreement ran to the Medicare beneficiaries participating in Texas Tech’s program, not to CMS. See United States v. Univ. Hosp. of State Univ. of N.Y. at Stony Brook, 575 F. Supp. 607, 612-13 (E.D.N.Y. 1983) (distinguishing procurement contracts from agreements to reimburse hospitals because “a hospital receiving reimbursements under Medicaid or Medicare provides services directly to the public”). CMS hoped to gain insight into how best to operate Medicare for high- cost beneficiaries, but advancing an “agency’s overall mission” (something that one would hope all grant agreements do) is not a direct benefit. Hymas, 810 F.3d at 1328. And CMS would not achieve any cost savings directly. The savings would have resulted—had the project gone as hoped—from Texas Tech’s making participating beneficiaries healthier, thereby decreasing their need for Medicare services.[3]The Board had jurisdiction.III.Under the APA, the Board’s action may be set aside if “arbitrary, capricious, an abuse of discretion, or otherwise not in accordance with law.” 5 U.S.C. § 706(2)(A). An agency’s decision is presumptively valid; the plaintiff bears the burden of showing otherwise. Texas Clinical Labs, Inc. v. Sebelius, 612 F.3d 771, 775 (5th Cir. 2010). We defer to the agency’s findings of fact if they are supported by substantial evidence. Buffalo Marine Servs. Inc. v. United States, 663 F.3d 750, 753-54 (5th Cir. 2011). On questions of law, our degree of deference varies according to the source of law. Although we give Chevron deference to an agency’s interpretation of an ambiguous statute it administers, Memorial Hermann Hosp. v. Sebelius, 728 F.3d 400, 405 (5th Cir. 2013), when an agency makes other conclusions of law—including interpreting “contractual agreements”—our review is “effectively de novo,” Inst. for Tech Dev. v. Brown, 63 F.3d 445, 450 (5th Cir. 1995).A.Underlying all of its merits arguments is Texas Tech’s objection to the Board’s characterization of the demonstration agreement as something other than a contract. Although the Board acknowledged that “the terms of a grant agreement are binding on both the grantee and the grantor,” Brown, 63 F.3d at 449, it concluded that “contract law doctrines do not dictate its analysis or resolution of disputes involving federal grants.” Several cases support that proposition. See, e.g., Bennett v. Ky. Dep’t of Educ., 470 U.S. 656, 669 (1985) (“Although we agree with the State that Title I grant agreements had a contractual aspect, the program cannot be viewed in the same manner as a bilateral contract governing a discrete transaction.” (citation omitted)); United States v. Melendez, 944 F.2d 216, 219 (5th Cir. 1991) (holding that “contract principles do not apply” in dispute over federal scholarship agreement).Texas Tech attempts to distinguish those decisions as limited to federal grants in which the parties’ obligations are dictated by statute or regulation, as opposed to the terms of a separately negotiated grant agreement. But we ultimately need not resolve whether the Board erred in suggesting that the common law of contracts never informs grant disputes. Even if it did, the Board made valid findings justifying the rejection of Texas Tech’s various contract theories.[4]B.Texas Tech’s most forceful complaint is that patients in the control group were too dissimilar from patients in the intervention group—that is, that CMS breached the demonstration agreement by failing to provide an appropriately matched control group.This is what the agreement said about selection of the control group:For a matched control cohort, the evaluator will specify the approach to be used in identifying such cohort. For each site that is not randomized, attention will be given to ensuring that the matched control cohort represents an appropriate comparison group that is specific to the intervention mode. [Texas Tech] will have the opportunity to review and agree to the approach that is proposed by the evaluator and CMS. CMS and [Texas Tech] will mutually agree upon the criteria and methodology used to select the relevant comparison group pursuant to a memorandum from the evaluation contractor documenting these methods and written documentation from [Texas Tech] stating its agreement with such memorandum.The agreement thus required that “attention” be given to “ensuring” the control group was sufficiently matched to the intervention group. It further explains how that would be done: CMS’s obligation was to select a third-party evaluator, which would propose an “approach” for choosing a control group. Texas Tech had to review and agree to that approach before starting the project.So agreement by the parties on “criteria and methodology” was the mechanism for ensuring adequacy of the control group, not after-the-fact assessment of whether the control group had an unspecified degree of similarity with the intervention group. Texas Tech invoked its right to influence the selection criteria, engaging in months of back-and-forth to flesh out the details. The agreement did not require more than that for selection of the control group.C.Texas Tech next argues that CMS breached the agreement by refusing to allow it to access relevant Medicare claims data. In apparent response to Texas Tech’s request to obtain this information from Trailblazer—a partner in the project—CMS told Texas Tech (1) that it could not obtain Trailblazer data in order to create its own “pseudo” control group; and (2) that it could not obtain Trailblazer data on patients in the intervention group without submitting a data use agreement and getting it approved by CMS. CMS did eventually approve a data use agreement between Texas Tech and Trailblazer, but not until May 2007, two months before Texas Tech terminated the project. Texas Tech’s idea is that had it been armed with Trailblazer data sooner, it could have better monitored its own performance as well as the accuracy of the control group.Attempting to locate its right to Trailblazer data in the demonstration agreement, Texas Tech points to its project proposal. In it, Texas Tech suggested that claims data would be available to it from the “Fiscal Intermediary Standard System,” one of several sources of data it might use to target its care management interventions. Because the agreement required Texas Tech to “implement the care management program as proposed,” Texas Tech claims a right to access and use data from fiscal intermediaries, including Trailblazer.We agree with the Board that CMS’s declining or delaying Texas Tech’s access to Trailblazer data did not violate the demonstration agreement. Even if the proposal indicated that Texas Tech intended to use Trailblazer’s claims data, the agreement incorporated the proposal to specify the details of the project Texas Tech agreed to implement. The proposal thus became a Texas Tech obligation, not a CMS obligation, so CMS could not have violated it. To be clear, the agreement (as opposed to the proposal) did impose data-related obligations on CMS. CMS had to give Texas Tech demographic information and claims data on the intervention group going back 12 months, but Texas Tech does not claim CMS failed to do so. Nothing in the agreement obligated CMS to give Texas Tech access to additional data from fiscal intermediaries. To the contrary, the only relevant provision in the agreement said that if Texas Tech “wishes to receive additional data . . . from an FI,” it must notify CMS, which would decide only “if such information can be provided.”Texas Tech’s argument faces another problem. Even if we were to read the agreement to require CMS to allow Texas Tech access to Trailblazer data, that requirement would extend only to data on the intervention group—the proposal said nothing about using fiscal intermediary data for purposes other than honing Texas Tech’s care management interventions. And the Board found that under HHS privacy regulations, CMS could not have allowed Texas Tech to access Trailblazer’s claims data on the intervention group without a data use agreement. See 45 C.F.R. § 164.514(e)(3)(i), (4)(i) (permitting a “covered entity” to disclose medical data “only if the covered entity obtains satisfactory assurance, in the form of a data use agreement” that the recipient will use the data “for limited purposes”). Texas Tech does not dispute that. It has not met its burden of showing that the Board’s ruling on this issue should be set aside.D.Texas Tech also contends that CMS breached the demonstration agreement by engaging RTI to evaluate whether differences between the intervention and control groups may have accounted for the apparent lack of cost savings. Because RTI designed the parameters for the control group, Texas Tech contends it had a conflict of interest in checking its work.But the only time RTI analyzed the alleged disparities in the groups was after Texas Tech had terminated the project, Actuarial Research Corporation had conducted the final reconciliation, and Texas Tech had alleged that the study was flawed. The demonstration agreement imposes no post- reconciliation obligations on CMS, so CMS cannot have breached the agreement by asking RTI to look into Texas Tech’s concerns.[5]E.In addition to its breach arguments, Texas Tech asserted a variety of common law contract doctrines before the Board. The Board did not expressly address two of them—mistake and impracticability[6]—because of its view that those common law principles do not apply to grant agreements.But the Board did make findings that doom those two defenses. It found, based on the months of negotiation over criteria for the control group, that Texas Tech “proceeded despite its concerns” and “could not reasonably have failed to understand . . . that it had assumed the financial risks associated with shortcomings in study design and implementation.” That undermines Texas Tech’s mistake theory; Texas Tech assumed the risk by agreeing to perform the demonstration, knowing that the control group might have problems. Restatement (Second) of Contracts § 154(b). It also undermines Texas Tech’s impracticability theory because it shows that the accuracy of the control group was not a “basic assumption” of the parties’ agreement. Id. § 261.Although the Board did not make these findings in the context of discussing mistake and impracticability, they would defeat those defenses had the Board directly addressed them. As a result, and because substantial evidence supports these findings, any error was harmless. See Shinseki v. Sanders, 556 U.S. 396, 411 (2009) (clarifying that administrative harmless error analysis turns on, among other things, the “likelihood that the result would have been different” but for the error). The administrative harmless error rule can sit uneasily alongside the Chenery doctrine, which typically confines judicial review to the agency’s reasoning. See Nicholas Bagley, Remedial Restraint in Administrative Law, 117 Colum. L. Rev. 253, 302-07 (2017). But that tension arises when a reviewing court makes findings the agency should have made in the first instance. See id. at 305 (observing “controversial[]” cases upholding agency decisions when “ the evidence in the record so strongly supports the result that the court is confident the agency would reach the same decision on remand”). Chenery cautions against intruding onto an agency’s turf, not against giving an agency’s own findings their obvious effect. See Henry J. Friendly, Chenery Revisited: Reflections on Reversal and Remand of Administrative Orders, 1969 Duke L.J. 199, 211 (explaining that remand to an agency is “necessary only when the reviewing court concludes that there is a significant chance that but for the error the agency might have reached a different result” because “in the absence of such a possibility, affirmance entails neither an improper judicial invasion of the administrative province nor a dispensation of the agency from its normal responsibility”).***The district court’s judgment is AFFIRMED.
