Midland Career Institute, an accredited trade school with headquarters in Hammond, Indiana, opened a branch in Chicago early in 1992. The branch, and the school itself, closed about a year later. Almost 95% of the Chicago branch's students defaulted on their federally guaranteed loans. That exceptionally high rate led to an investigation and criminal convictions of Alan Frost and Anne Bracken, who purchased the school in 1987 and managed it until the failure in 1993. Both defendants were convicted of conspiracy to defraud the United States, see 18 U.S.C. § 371, and four substantive counts of fraud in connection with federal educational assistance, see 20 U.S.C. § 1097(a). Each was sentenced to 51 months' imprisonment.

According to the prosecutor, Frost and Bracken employed three devices to defraud the United States. First, they instructed the school's employees to lie about students' accomplishments in order to obtain additional Pell Grants (federal scholarships for needy students). Grant funds for each year are disbursed in two parts: half on enrollment, and the second half on certification by the school that the student has successfully completed half of the year's education hours (or will do so within three days). The evidence, taken in the light most favorable to the jury's verdict, showed that the defendants instructed the school's staff to make false certifications in order to obtain Pell Grant funds on behalf of students who were nowhere near completion of the hours required. Second, defendants applied to Indiana for guarantees (ultimately backed by the United States) of bank loans made to students at the Chicago branch. These applications were made using the name, address, and identifier of the main location in Indiana, and the school did nothing to alert Indiana's officials that the education was being provided outside of Indiana to students who lived outside Indiana. Third, the school failed to ensure that all students obtaining federal aid possessed the necessary qualifications: a high school degree, a general education diploma, or a passing grade on an "ability-to-benefit test." The result of these three shortcomings, according to the prosecutor, was that the school received substantial federal funds to which it was not legally entitled. The omission of ability-to-benefit tests for four particular students is the basis of the four substantive convictions.

Frost and Bracken argued at trial that their conduct was appropriate. With respect to Pell Grants, for example, they contended that most of these students were bound to complete the required credit hours eventually, and that the school refunded grant funds received on behalf of students who did not. Evidence that refunds actually occurred did not find its way into the record, a telling omission. But even if refunds had been made, defendants' position is similar to that of a person who obtains bank loans by fraudulently overstating his assets and justifies his conduct with the claim that he intended to repay (and did pay back some loans). The conduct is fraud nonetheless: lies that induce the bank to part with the money are material, and the overstatement of assets exposes the bank to more risk of nonpayment than it willingly assumed. See United States v. Dial, 757 F.2d 163, 169-70 (7th Cir.1985); United States v. Catalfo, 64 F.3d 1070, 1076-78 (7th Cir.1995). So too here. The United States took the risk that someone within three days of completing the required credits might fail to do so (and that the school would go out of business and be unable to repay). It did not agree to assume the larger risk that someone a month or two from completion might drop out of school or flunk courses, or that a school with a larger inventory of "unearned grants" might fold while in debt to the United States, as Midland Career Institute did. Other lines of defense relating to Pell Grants (such as an argument that some of the students actually had completed the required hours, but that the school's records were defective) presented jury questions, which were resolved adversely to the defendants. The record permitted the jury to find as it did. The frauds with respect to Pell Grants are enough by themselves to sustain the conspiracy conviction, even if the evidence on the other two kinds of overt acts were deemed insufficient. See Griffin v. United States, 502 U.S. 46, 112 S. Ct. 466, 116 L. Ed. 2d 371 (1991).

Because it turns out to be important for sentencing, the evidence concerning the guarantees requires some discussion. When it opened a branch in Chicago, the school applied to Illinois for guarantees of loans to its students. The state agency responsible for this task in Illinois told the school that it should apply to Indiana instead, because its main location was in that state. Indiana's rules make students living in counties contiguous to that state eligible, see Ind.Code § 20-12-21.1-1(f)(4), so the school filed applications with Indiana. But it did not tell Indiana's agency that the students lived in Illinois and would attend a branch in Chicago. The documentation submitted in support of the guarantees looked just like the documentation supporting guarantees for Indiana residents studying in Hammond. Defendants say that this was not fraudulent, because residents of Cook County, Illinois, were substantively eligible for guarantees: no harm, no foul. After all, Illinois had told the school to turn to Indiana. But Illinois had not told the school to withhold information from Indiana. Guarantees were not automatic. Indiana might well have wanted to satisfy itself that the Chicago branch was providing a good education to responsible students, so that the loans were likely to be repaid. The Chicago branch had been provisionally accredited when it opened, on the strength of the Hammond school's history, but did not undergo a regular accreditation study until October 1992, months after Indiana began guaranteeing loans — and after this study the provisional accreditation was revoked. Something turned out to be very different between the Chicago and Hammond branches — whether on the school's side or the students' — because the repayment history at the Chicago branch was dramatically inferior to that in Hammond. Candid applications for guarantees might have enabled Indiana's agency to investigate and protect itself (and thus the United States) against the losses that ensued. Perhaps Indiana would have insisted that full accreditation precede guarantees. The information withheld from Indiana — that the students were attending a new branch in a different state, and may have had different willingness or ability to repay — was material to the decisions Indiana had to make. Leaving Indiana in the dark thus was fraudulent.

The ability-to-benefit tests played no material role in sentencing, but they do support four separate convictions, each of which carries a $100 special assessment. These convictions are not supported by evidence that shows guilt beyond a reasonable doubt and must be vacated. An amendment to federal law effective in 1991 made a passing score on these tests a condition to federal aid for students who did not have high school or general education diplomas. See § 3005(a) of Pub.L. 101-508, 104 Stat. 1388, 1388-27, 1388-28 (1990). The school engaged Wonderlic, Inc., a reputable firm, to administer these when necessary. Wonderlic hired Regina Biocic to carry out this task. Nothing in the record suggests that Frost or Bracken asked Biocic to provide false test results. But when federal investigators scoured the school's files after its failure, they found 19 students' records that did not note either a diploma or a passing score on an ability-to-benefit test. Four of these 19 were selected as the basis of criminal charges.

Although each of the 19 files lacks proof that the student took and passed an ability-to-benefit test, there are two plausible reasons (other than fraud) for that omission. First, Biocic was not well organized; this much the United States concedes. Likely she failed to ensure that appropriate notations were made in at least some of the records (although she may have indicated in some other way that the applicant passed). Second, the school's files were themselves a mess. After it closed, heaps of miscellaneous records were scattered about. The agent who examined the students' files did not look through these documents. If the evidence showed that the school had not set out to administer proper tests, or that defendants had tried to subvert Wonderlic's process, then a search would not have been necessary. But because the United States concedes that defendants delegated the testing, it was essential to investigate the possibility that the lack of notations in the 19 files reflects a failure at Wonderlic's end, or of the school's filing system, as opposed to culpable fraud on defendants' part. Nineteen students is a small portion of the enrollment, making record-keeping errors more plausible as an explanation (and calling into question whether defendants had much to gain by skipping the test for so few students).

When imposing a sentence, the district court started with an offense level of 6, then added 12 because the loss on the guaranteed loans fell between $1.5 million and $2.5 million. See U.S.S.G. § 2F1.1 (2000 ed.). (Section 2F1.1 was deleted on November 1, 2001, by Amendment 617 and consolidated with § 2B1.1. Enhancements also were increased. Under the revision a loss between $1 million and $2.5 million would increase the base offense level by 16. Our references are to the Guidelines in force when defendants were sentenced.) The judge added 2 more levels because the offenses involved more than minimal planning, see § 2F1.1(b)(2)(A), and 4 more because defendants were the leaders of a large organization. See § 3B1.1(a). The final offense level of 24 combined with criminal history category I to produce a sentencing range of 51-63 months. Each defendant received a sentence at the bottom of that range. Each contends that the judge should have omitted the 4-level enhancement under § 3B1.1(a) and should have set the loss at $0. These changes would decrease the final offense level to 8, with a sentencing range of 0-6 months. We start with the 4-level enhancement.

Section 3B1.1(a) provides: "If the defendant was an organizer or leader of a criminal activity that involved five or more participants or was otherwise extensive, increase by 4 levels." The word "participants" denotes criminally responsible participants, § 3B1.1 Application Note 1, and we shall assume that no one at the school other than Frost and Bracken could have been convicted of conspiring to defraud the United States — that those who aided Frost and Bracken were their dupes rather than knowing participants. Still, "[i]n assessing whether an organization is `otherwise extensive,' all persons involved during the course of the entire offense are to be considered. Thus, a fraud that involved only three participants but used the unknowing services of many outsiders could be considered extensive." Section 3B1.1 Application Note 3. Frost and Bracken used the services of many participants, including (for example) the record-keeping staff who defendants instructed to submit premature requests for Pell Grants and guaranty applications omitting all information that would imply that operations were being conducted in Chicago. The Midland Career Institute was an extensive organization, and defendants conducted its business in such a way that a substantial portion of its income throughout 1992 was obtained by fraud. The enhancement under § 3B1.1(a) is designed for such cases.

As for the loss calculation: the district court started with the $2.8 million in unauthorized Pell Grants plus guarantees on which the United States had to make good when students did not pay their lenders, then deducted the costs of operating the school's Chicago branch in order to estimate net loss. The judge's idea was that the United States agreed to buy education for the students, and the cost of running the school is the best estimate of the value of the education it delivered. The two defendants were sentenced separately. At Frost's sentencing, the district judge calculated this cost at $1.05 million, leaving a loss of about $1.78 million. At Bracken's later sentencing, the judge reckoned the cost of running the branch at $3.21 million, discarding as inaccurate the method he had used to determine the cost when sentencing Frost. But the judge then reduced this figure by 68% because only 32% of the students at the Chicago branch had graduated. This produced an adjusted cost of $1.03 million and a net loss of $1.8 million, so Frost and Bracken ended up in the same sentencing range. Defendants observe that one or the other of these methods must be wrong — and they contend that both are mistaken, because the judge should have used the $3.21 million figure and calculated a net loss to the United States of zero for each of them.

For its part, the United States contends that the district judge should have stopped with the $2.8 million gross loss, making no deduction. (This implies that the sentence is too low, but a prevailing party is entitled to defend its judgment on any properly preserved ground. Compare Massachusetts Mutual Life Insurance Co. v. Ludwig, 426 U.S. 479, 96 S. Ct. 2158, 48 L. Ed. 2d 784 (1976), with El Paso Natural Gas Co. v. Neztsosie, 526 U.S. 473, 479-81, 119 S. Ct. 1430, 143 L. Ed. 2d 635 (1999).) One argument offered in support of this position — that the benefit of higher education does not count because it was received by the students rather than the United States — is unconvincing. The Guidelines call for the use of net rather than gross loss (for example, a defendant receives credit for the value of goods delivered, if the fraud entails overcharging). Section 2F1.1 Application Note 8. When one party sets out to benefit another, then the net loss is the cost to the payor less the benefit delivered to the intended recipient. So in a food stamp case, where the grocer accepts stamps for both bread (allowed) and liquor (prohibited), the net loss to the United States is the amount of stamps redeemed, less the value of eligible items delivered to the beneficiaries. See United States v. Hassan, 211 F.3d 380 (7th Cir.2000); United States v. Barnes, 117 F.3d 328 (7th Cir.1997). The same principle allows educators to deduct from the gross loss the value of any education for which the United States was willing to pay.

That qualifier — "willing to pay" — is vital. Suppose Donald Trump were to hand a grocery store food stamps to pay for his purchase of milk and bread. The cashier knows that Trump's income is too high to make him eligible for food stamps but redeems them anyway. The grocery could not offset the value of the milk and bread delivered to Trump against the amount of its food stamp redemptions when calculating net loss to the United States, because Trump is not among the lawful participants in the program. Right items, wrong beneficiary. That's the big problem defendants face here. Education is something for which the United States is willing to pay — but was it delivered to those persons on whose behalf the United States is willing to cover the costs? By concealing the location of the branch and its students from Indiana's agency, Frost and Bracken prevented it from making an informed decision about this subject. Perhaps with full knowledge Indiana would have approved the guarantees, but we'll never know. When imposing sentence a judge may assume that funds that were furnished only because of material deceit would not otherwise have been forthcoming. That's the normal approach in a bank-fraud case. The borrower who overstated his assets, received a loan, and did not repay could not argue that the loss really is zero because the bank would have extended credit anyway (though perhaps at a higher interest rate).

There is a second reason why it is inappropriate to net out the costs of running the branch: Costs differ from benefit delivered. Suppose the United States contracted for construction of a highway, and the winning bidder obtained the contract by fraud. The loss would be the contract price less the value of the highway actually built, not price less the contractor's costs. This would be easy to see if the contractor used shoddy materials and the road had to be torn up and rebuilt. Its value would be zero at best and could be negative (because of the expense of demolition). Suppose instead that the highway were well built; by hypothesis the only problem is that the Treasury overpaid. If costs are deducted from price, then a contractor that worked efficiently, and thus had low expenses, would receive a higher sentence than an inefficient builder. What sense would that make? Similarly with education. A trade school that owns a private jet and rents space in fancy office towers will have higher costs than a school located in a suburban strip mall, but these differences would be poorly correlated with value delivered; they might instead reflect the consumption value that the school's managers extract (taking trips on the corporate jet and enjoying the lofty views as perks). The relation between costs and educational value may have been especially poor for the Chicago branch, which closed less than 15 months after it opened and so threw out many students without providing the full course of instruction. They may have had to start from scratch elsewhere. Frost and Bracken did not attempt to make more direct estimates of the school's educational value. They did not, for example, obtain before-and-after incomes for their students. Nothing in this record enabled the district judge to estimate the value added provided by the Chicago branch, so there was no reliable basis for deducting anything from the gross loss.

Some deduction might have been justified on a different ground. The district judge counted as loss all of the loans that the United States had to make good as guarantor. This implicitly assumes that, if the United States had used its funds to guarantee loans to the same students, at other schools, it would not have suffered any losses. Yet all educational institutions experience some defaults. According to the Department of Education, see <http://www.ed.gov/offices/OSFAP/defaultmanagement/1999institu—rate.html>, the rate for 1999 ran from a low of 3.7% (students at private, four-year colleges) to a high of 10.9% (students at proprietary trade schools offering less than two years of instruction). In 1997 the trade school rate was 18.5%, and in 1998 14.2%. Deducting the normal default rate in 1992 and 1993 from the 95% rate experienced at the Chicago branch would have produced a figure more closely approximating the loss caused by the defendants' fraud. Such a deduction would have put the loss in the $1.5 million to $2.5 million range — exactly the range the district judge actually used, though he got there by deducting estimates of the school's costs. There is accordingly no reason to disturb the 51-month sentences.

The substantive convictions are reversed. The conspiracy convictions and prison sentences are affirmed. The restitution order is vacated and must be recalculated on remand in light of this opinion.