OPINION OF THE COURT
BECKER, Circuit Judge.
Defendant Larry Kopp pled guilty to procuring a $13.75 million bank loan by fraudulent misrepresentations, in violation of 18 U.S.C. § 1344 (1988) (subsequently amended). The district court for the District of New Jersey sentenced him to a term of 33 months in prison. In calculating the sentence under United States Sentencing Guideline (“USSG”) § 2F1.1, the district court had to determine, as a specific offense characteristic, the amount of the “loss.” The defendant argued that the “loss” was zero (and therefore that no offense level increase was necessary) on the following grounds: (1) the bank’s actual loss was nil because the bank later sold the security for the loan for more than the loan balance; (2) even if the bank did incur an out-of-pocket loss, such loss resulted from misconduct by the bank and the defendant’s nephew David Kopp, thus the defendant was not responsible for any actual loss; and (3) the defendant did not intend to inflict any loss on the bank. The court rejected the defendant’s position and also declined to accept the probation officer’s calculation that the bank’s actual loss was only $3.4 million (an estimate which also took into account operating expenses, lost interest, and the cost of a low-interest loan to the new purchaser). Instead, the court agreed with the government that the “loss” was the amount that the defendant fraudulently obtained (the full $13.75 million face value of the loan), thereby increasing the offense level by eleven levels and the applicable sentence range to 30-37 months.
[523]*523The defendant’s appeal from the judgment of sentence requires us to decide how to apply the fraud guideline in this context, a question on which other circuits have split. For the reasons that follow, we conclude that the district court erred in fixing the “loss” as the face value of the loan. Rather, we hold that the district court should have calculated the “loss” as actual loss, substituting intended or probable loss if either amount was higher and determinable. In our view, the court’s error stemmed in part from its failure to recognize fundamental differences between theft and fraud crimes, differences which the sentencing guidelines covering the two types of offenses take into account. We further hold that the guideline “loss” should not be reduced simply because the bank or David Kopp may have augmented it. When sentencing, the district court may, however, depart downward if any such augmentation caused the properly calculated “loss” to overstate the seriousness of the offense. Correspondingly, if the court finds that the “loss” understates the seriousness of the offense (which might be the case if actual and intended loss were zero and the risk of loss were significant), it may depart upward, so long as the sentence does not exceed the statutory maximum, in this case five years’ imprisonment.1 Because the district court made no findings on actual or intended loss, we will vacate the judgment of sentence and remand for resentencing. That court remains free not only to recalculate the “loss,” but also to reevaluate other facts pertinent to its decision whether to depart from the guideline range.
The defendant also raises several other issues. He objects to added offense levels for more than minimal planning and for a supervisory role, and he complains that the district court should have granted him an offense level reduction for minor participation. We find these arguments without merit. He also alleges that the district court improperly refused to depart downward from the guideline range based on a misconception of its legal ability to do so. Because the district court is free to revisit the departure issues on remand, we need not reach that issue.
I. FACTS AND PROCEDURAL HISTORY
A. The Offense
In the early 1980s, the defendant and his brother, Marvin Kopp, began a real estate development business, which they operated through various corporations and partnerships. The seed capital came largely from the family of Barbara Kopp, the defendant’s wife. In November 1984, Marvin Kopp died, and, as the business faced ever more severe financial problems, disputes arose between the defendant on one side and Marvin Kopp’s widow Judith Kopp and her son David Kopp on the other.2 By the spring of 1987, David Kopp had become the managing partner, although the defendant remained involved. Financial difficulties, however, continued, even though the Kopps sank still more of their own funds into the partnerships.
In August 1987, the Kopp partnership began to negotiate with Ensign Bank, FSB (“the bank”) for a $14 million loan, $12.3 million of which was necessary to refinance a shopping mall owned by one of the real estate partnerships. The refinancing was apparently required because an earlier [524]*524lender, having discovered that the Kopps had obtained a second mortgage on the property without permission, was about to declare a default. To induce the bank to make the loan, Stuart Sherer, the Kopp partnership’s office manager, prepared and signed false leases and estoppel letters3 that inflated the true amount of rental income the shopping center was then generating. The defendant was aware of the fraudulent inducement of the bank loan and, Sherer claims, personally directed the forgeries. The defendant also personally certified two rent rolls he knew to be false. The misrepresentations suggested that the shopping center was fully occupied and that the rental income stream would be sufficient to cover the mortgage payments on the loan, neither of which was true.
Relying on the misrepresentations, the bank loaned the Kopp partnership $13 million in December 1987 and $750,000 more in January 1988. Apparently the debt-service ratio was insufficient to collect the remaining $250,000 of the $14 million loan commitment. To obtain the first installment, Sherer, with the defendant’s knowledge, submitted five forged leases and twenty-one forged estoppel letters, along with three more leases that the partners knew would be broken by tenants. To obtain the second installment, Sherer further submitted a lease that overstated the rent due, as well as a new forged estoppel letter and an updated fraudulent rent roll.
B. The Default and the Bank’s Actual Loss
The loan went into default in February 1988: no payments were made after the second loan installment was received. The defendant blames the failure to repay in part on David Kopp’s diversion of partnership money and his failure to collect all amounts receivable. The government doubts that the defendant ever intended that the loan be repaid. In any event, the bank demanded and received a deed in lieu of foreclosure and eventually sold the property for $14.5 million, $750,000 more than the face value of the loan. The bank nonetheless calculated that it actually lost approximately $8.4 million overall, due to lost interest ($1.5 million), the bank’s operating expenses when taking over the property ($0.4 million), and the cost of a low-interest loan to the new purchaser ($2.3 million).
The defendant, however, contends that the $3.4 million actual loss estimate was overstated and that any actual loss was a result of misconduct by David Kopp and bank officer Brian Maloney. He called a real estate appraiser, who testified that the value of the foreclosed property was at least $17.5 million, $3 million more than the sale price. He also introduced evidence that the lower sale price and a linked low-interest loan were due to an unethical “sweetheart” deal between the bank (in the person of Maloney) and David Kopp. The bank wanted control over the shopping center more quickly, so sought a deed from both Kopps in lieu of foreclosure. The defendant alleges that to obtain rapid title, Maloney secretly agreed to steer the resale to David Kopp’s friend Clifford Streit, who in turn would give David Kopp back a 25% interest in the property.4 The defendant claims that he was trying to arrange a profitable sale to third parties at that time but consented to giving the bank the deed when Maloney falsely told him there would be no criminal referral if he agreed. Pursuant to the secret exclusive option deal, the bank sold the property to Streit and his hidden partner David Kopp for below-market value. This underhandedness, the defendant continues, also led to an unnecessary low-interest loan, with the result that the bank’s loss, if any, was inflated.
The defendant also claims that the actual loss figure of $3.4 million was improperly calculated, even given the way the resale took place. He argues that Maloney, who made that calculation, had an incentive to distort it because of Maloney’s own miscon[525]*525duct. Moreover, he contends, the bank’s own remaining records5 show that the bank recorded an operating surplus on the property, not a $400,000 deficit, and fail to support the claimed $1.5 million in lost interest on the bad loan. Finally, the defendant asserts that the operating surplus would have been higher but for the bank’s failure to credit rents received in June and July, 1988 and the bank’s payment of $15,-000 to the defendant and his wife for their equity interest. In short, the defendant claims that if the resale had not been tainted by misconduct, the bank would have profited, and that even as events transpired, the bank’s actual loss was far less than $3.4 million.
C. The Criminal Proceedings
The federal authorities discovered the fraud in May 1988, when David Kopp reported the offense, admitted his complicity, and agreed to cooperate in prosecuting Stuart Sherer and his uncle, the defendant. David Kopp began secretly to record his conversations with the other conspirators. Confronted with the recordings, Sherer later agreed to cooperate and record additional conversations with the defendant. On August 8, 1989, the defendant was indicted for, among other things, one count of bank fraud under 18 U.S.C. § 1344 and one count of conspiracy to commit bank fraud. On June 11, 1990, under a plea bargain, the defendant pled guilty to those counts; the remaining ten counts, five of which related to other alleged crimes, were dropped.
The U.S. Probation Office submitted a Presentence Investigation Report on September 10, 1990, which adopted the bank’s estimate of $3.4 million in out-of-pocket loss as the “loss” for purposes of U.S.S.G. § 2F1.1, the applicable sentencing guideline. Both parties objected. The government claimed that because the bank would not have made the loan at all had it known the truth, the defendant fraudulently obtained $13.75 million, which was therefore the proper measure of the “loss.”6 The defendant argued, as discussed above, that the “loss” was zero. The district court held hearings on the defendant’s objections on October 25 and November 2 and 5,1990, but ruled on April 30, 1991, that the full $13.75 million the defendant put at risk was the appropriate measure of loss. Hence an increase of eleven offense levels was required under original U.S.S.G. § 2F1.1 because the “loss” exceeded $5 million.7 On May 17, 1991, the district court sentenced the defendant to 33 months in prison (roughly in the middle of the guideline range) for the reasons stated in its April 30, 1991 opinion. The defendant filed a timely appeal.
The impact of the challenged ruling is significant. Because the defendant had no prior criminal record, if the “loss” was zero (as he claims), the sentencing range would drop from the 30-37 months that the district court considered to a range of 2-8 months. If the district court also found that no upward departure was warranted, it might have elected to impose probation conditioned on a combination of community confinement, home detention, or intermit[526]*526tent confinement under U.S.S.G. §§ 5B1.1 and 501.1(c), rather than continuous incarceration in prison. For this reason, the defendant sought bail pending appeal from the district court. The district court agreed that the defendant was not dangerous, was unlikely to flee, and that the grounds for appeal were substantial. It stated that if this court did reverse, the probable reduction would be eleven levels, leaving the sentence range at 2-8 months. It therefore ordered the defendant to begin serving an eight-month sentence on July 8, 1991, noting its inclination to impose the maximum in that range if resentencing proved necessary. It further ordered that bail pending appeal would be granted only if no decision .from this court came down in eight months.
The defendant then sought bail pending appeal from this court. A motions panel denied that request without prejudice to reconsideration after hearing the merits. After hearing oral argument and anticipating the probable outcome on the merits, we issued an order granting the defendant bail pending our final decision and eventual re-sentencing.
II. DEFINITION OF “LOSS” IN THE FRAUD GUIDELINE
A. Analysis of U.S.S.G. § 2F1.1 as It Was in Effect at Sentencing
The guideline in effect here is a combination of the fraud guideline in effect at the time of sentencing and the original fraud guideline in effect at the time of the crime. As a general rule, sentencing courts must apply the guidelines in effect at the time of sentencing, not the time of the crime. See 18 U.S.C.A. § 3553(a)(4), (5) (West 1985 & Supp 1991); United States v. Cianscewski, 894 F.2d 74, 77 n. 6 (3d Cir.1990). But where such retroactivity results in harsher penalties, Ex Post Facto Clause problems arise, and courts must apply the earlier version. See Miller v. Florida, 482 U.S. 423, 107 S.Ct. 2446, 96 L.Ed.2d 351 (1987); see also United States v. Underwood, 938 F.2d 1086, 1090 (10th Cir.1991); United States v. Morrow, 925 F.2d 779, 782 (4th Cir.1991). We consequently review the district court’s decision for consistency with the guidelines in effect on May 17, 1991, the date of sentencing, except where amendments since the time of the crime would yield a harsher result.
The version of the fraud guideline, U.S.S.G. § 2F1.1(b)(1), that is in effect here reads:
If the loss exceeded $2,000, increase the offense level as follows:
Loss Increase in Level
(A) $2,000 or less no increase
(K) $2,000,001 — $5,000,000 add 10
(L) over $5,000,000 add 118
The official Commentary provides some further guidance as to the calculation [527]*527of a “loss” and the difference between various types of losses. The version of Application Note 7 in effect at sentencing9 read:
Valuation of loss is discussed in the Commentary to § 2B1.1 (Larceny, Embezzlement, and Other Forms of Theft). In keeping with the Commission’s policy on attempts, if a probable or intended loss that the defendant was attempting to inflict can be determined, that figure would be used if it was larger than the actual loss. For example, if the fraud consisted of attempting to sell $40,000 in worthless securities, or representing that a forged check for $40,000 was genuine, the “loss” would be treated as $40,000 for purposes of this guideline.
Application Note 8 explained the process of estimating a “loss”:
The amount of loss need not be precise. The court is not expected to identify each victim and the loss he suffered to arrive at an exact figure. The court need only make a reasonable estimate of the range of loss, given the available information. The estimate may be based on the approximate number of victims and an estimate of the average loss to each victim, or on more general factors, such as the nature or duration of the fraud and the revenues generated by similar operations .... The offender’s gross gain from committing the fraud is an alternative estimate that ordinarily will understate the loss.
The Commentary also discussed possible upward or downward departures. Application Note 9 provided that where “[d]ollar loss ... does not fully capture the harmfulness and seriousness of the conduct ... an upward departure may be warranted.” In contrast, Application Note 1010 discussed possible downward departures:
In a few instances, the total dollar loss that results from the offense may overstate its seriousness. Such situations typically occur when a misrepresentation is of limited materiality or is not the sole cause of the loss. Examples would include understating debts to a limited degree in order to obtain a substantial loan which the defendant genuinely expected to repay_ In such instances, a downward departure may be warranted.
As have most courts, we begin our interpretation of “loss” under the fraud guideline with Application Note 7’s cross-réfer[528]*528ence to the theft guideline Commentary. At the time of sentencing, Application Note 2 to U.S.S.G. § 2B1.1 (the theft guideline) read:
“Loss” means the value of the property taken, damaged, or destroyed. Ordinarily, when property is taken or destroyed the loss is the fair market value of the particular property at issue. Where the market value is difficult to ascertain or inadequate to measure harm to the victim, the court may measure loss in some other way, such as reasonable replacement cost to the victim.... In cases of partially completed conduct, the loss is to be determined in accordance with the provisions of § 2X1.1 (Attempt, Solicitation, or Conspiracy). E.g., in the case of the theft of a government check or money order, loss refers to the loss that would have occurred if the check or money order had been cashed. Similarly, if a defendant is apprehended in the process of taking a vehicle, the loss refers to the value of the vehicle even if the vehicle is recovered immediately.11
The Background to U.S.S.G. § 2B1.1 continued: “The value of the property taken plays an important role in determining sentences for theft offenses, because it is an indicator of both the harm to the victim and the gain to the defendant.”
The government seizes upon the first sentence of Application Note 2 to the theft guideline, which equates the “loss” with the value “taken.” The government claims that the defendant “took” $13.75 million that was not rightfully his, because, as proved at the hearing, the bank would have lent him nothing if it had known the true monetary condition of the shopping center.12 Furthermore, the government correctly notes, under Application Note 2 to the theft guideline, theft “loss” is not affected by recovery of any of the property stolen. See, for example, United States v. Chiarelli, 898 F.2d 373, 384 (3d Cir.1990); United States v. Cianscewski, 894 F.2d 74, 80-81 (3d Cir.1990). The government therefore concludes that only the defendant’s conduct matters to “loss” calculation, and that the fortuity of recovery and other aspects affecting actual loss to the victim are irrelevant.
But the analysis is not so simple. Even the theft guideline is not entirely perpetrator-oriented. The Commentary to U.S.S.G. § 2B1.1 focuses significantly on the victim’s loss, as well as the perpetrator’s gain. Replacement cost to the victim may be used to measure theft “loss” where the market value of the stolen property is difficult to ascertain or inadequate to measure harm to the victim. And the Background to U.S.S.G. § 2B1.1 emphasizes that “loss” measures both harm to the victim and gain to the defendant.
More basically, however, U.S.S.G. § 2B1.1 is, by definition, a theft guideline, and fraud differs from theft. In this case, [529]*529for example, the defendant did fraudulently obtain $13.75 million, but in so doing he simultaneously gave a mortgage on the property — an obligation that eventually forced him to forfeit the collateral to the bank. He did not “take” $13.75 million for nothing, as a thief would. Furthermore, all thefts involve an intent to deprive the victim of the value of the property taken. As the Seventh Circuit has noted, the same is not always true for fraud: some fraud involves an intent to walk away with the full amount fraudulently obtained, while other fraud is committed to obtain a contract the fraud perpetrator intends to perform. See United States v. Schneider, 930 F.2d 555, 558 (7th Cir.1991) (noting that a bilking con artist differs greatly from a contract bidder who misstates its qualifications but intends to perform).
Mechanical application of the theft guideline in fraud cases would frustrate the legislative purpose of the guidelines and contravene the specific language of the Commission. The sentencing guideline system was designed to sentence similarly situated defendants similarly; basing all fraud sentences on a simple “amount taken” rule without regard to actual or intended harm would contravene that purpose. We think it plain that actual harm is generally relevant to the proper sentence. More importantly, Congress has explicitly said so in 28 U.S.C.A. § 994(c)(3) (West Supp.1991), which directs the Commission to take “the nature and degree of the harm caused by the offense” into account in its guidelines. See also U.S.S.G. § lB1.3(a)(3) (factors that determine guideline ranges include actual and intended harm).
Indeed, the fraud guideline does reflect the differences between theft and fraud and between the types of fraud. Even after the June 15, 1988 amendment, fraud guideline analysis only begins and does not end with the discussion of “loss” in the theft guideline. The revised Application Note 7 to U.S.S.G. § 2F1.1 does not say that the definitions of “loss” for theft and fraud crimes are identical, just that “[valuation of loss is discussed in the Commentary to § 2B1.1 ...” (emphasis added). While the calculations under the two guidelines are essentially consistent, the fraud analysis is slightly (and crucially, here) more complicated, as a detailed analysis of the entire fraud guideline Commentary reveals. We part company with several other circuits and join ranks with the Seventh Circuit largely because we decline to impose an identical analysis for theft and fraud crimes in all cases. Compare United States v. Brack, 942 F.2d 141, 143 (2d Cir.1991) (applying theft analysis to fraud crime with no qualification), and United States v. Johnson 941 F.2d 1102, 1113-15 (10th Cir.1991) (same), with United States v. Schneider, 930 F.2d 555, 558-59 (7th Cir.1991) (“simple” but “irrational” to treat all frauds the same, whether or not the defendant intended to walk away with the face value). See generally Part II.B at 532-534.
The remainder of Application Note 7 to U.S.S.G. § 2F1.1 (which was part of the original Commentary) made a critical distinction between actual, probable, and intended loss. Under Application Note 7 as in effect here, “if a probable loss or intended loss ... can be determined, that figure would be used if it was larger than the actual loss.” Consistent with the ordinary meaning of “loss,” sentencing courts were to use the victim’s actual loss as the basic “loss” under the fraud guideline. But, consistent with the guidelines’ overall theory of culpability for attempts, see U.S.S.G. § 2X1.1, courts were to switch to either “probable” or “intended” loss if either was both reasonably calculable (as discussed in Application Note 8) and higher. The fraud guideline thus has never endorsed sentencing based on the worst-case scenario potential loss (here, the face value of the loan). See also U.S.S.G. § 1B1.3 appl. note 4 (fraud guideline only refers to actual, attempted, or intended harm, hence risk of harm enters only into the base offense level and the possibility of an upward departure).
This plain reading of Application Note 7 is in fact consistent with the cross-reference to the theft guideline. In both theft and fraud cases, the guideline “loss” turns out to be the higher of the actual loss and the intended loss. Admittedly, the theft guideline does not state this approach in terms. But in a theft case, the [530]*530thief intends to steal whatever he or she takes; the amount taken is the loss the defendant intended to inflict (even though the defendant may not have known exactly the value on the market or to the victim of the things taken). Although the theft guideline does not tell the sentencing court to compare actual and intended loss, as the fraud guideline Commentary explictly does, no comparison is necessary. In a theft case, unlike a fraud case, the amount taken (the intended loss) is always as high or higher than the amount the victim actually lost (which may be reduced due to fortuitous recovery of the stolen property).13 Our interpretation of the fraud guideline therefore harmonizes with the approach of the theft guideline.14
Aside from the cross-reference to the theft guideline, the only support in the fraud guideline for the government’s position comes from Application Note 8. That Note primarily emphasizes that the “loss” need not be estimated with precision, but the version in effect at the time of sentencing did end with the statement that “[t]he offender’s gross gain from committing the fraud is an alternative estimate that ordinarily will understate the loss” (emphasis added).15 The government suggests that the “gross gain” here was the $13.75 million that the defendant obtained but would not have received had he not committed fraud, and that $13.75 million is an appropriate alternative estimate of the “loss.” Although that interpretation seems sensible, we reject the use of an alternative estimate when, as here, the true “loss” is measurable. Application Note 8 primarily addresses “loss” estimation, not the proper definition of “loss,” which is discussed in Application Note 7.
The government also relies on original Application Note 11, which noted that “a downward departure may be warranted” [531]*531when “a misrepresentation is of limited materiality or is not the sole cause of the loss.” Indeed, one example given is comparable to this case: where a defendant “understates] debts to a limited degree in order to obtain a substantial loan which the defendant genuinely expeet[s] to repay,” the court may decide to depart downward. The district court concluded that the use of “substantia] loan” in the Application Note suggests that the drafters had the full amount of the loan in mind as the “loss,” and to the extent that this “loss” measure was overstated, the court could depart downward (although it chose not to). We disagree. Application Note 11 discussed downward departures, not the calculation of “loss” in the first instance. We think that the Commission was emphasizing the defendant’s lack of culpability — his or her “limited degree” of understatement and his or her intent to repay — as a basis for downward departure, rather than suggesting an alternative definition of “loss.”
The government is correct on one point, however: Application Note 11 definitively rejected adjusting the “loss” itself downward to reflect other causes beyond the defendant’s control. As an example of when the dollar loss may overstate the seriousness of the offense and hence a downward departure may be appropriate, Application Note 11 included situations where the “misrepresentation ... is not the sole cause of the loss.” Application Note 11 made it clear that actual loss was how much better off the victim would be but for the defendant’s fraud. To the extent actual loss had other, more proximate causes, a discretionary downward departure — but not a mandatory “loss” adjustment — might be appropriate. Here, then, the district court might conclude that the defendant deserves a downward departure due to misconduct by David Kopp and Brian Maloney, but the level of the “loss” itself would remain unaffected.
The foregoing analysis leads to the conclusion that the fraud guideline defines “loss” primarily as the amount of money the victim has actually ended up losing at the time of sentencing, not what it could have lost. Under the guideline in effect at sentencing, the “loss” should have been revised upward to the amount of loss that the defendant intended to inflict on the victim, or to the amount of probable loss,16 if either intended or probable loss was esti-matable and higher. Before finally adopting this view, however, we will survey the case law interpreting “loss” calculation under the fraud guideline to see if it affects our tentative conclusion.
B. The Case Law
One other circuit, the Seventh, is on record with a similar interpretation of [532]*532“loss” in fraud cases. United States v. Schneider, 930 F.2d 555 (7th Cir.1991), involved defendants who procured $142,400 of government contracts, but submitted fraudulent payment and performance bonds in the process. Apparently, however, they were the low bidders and would in fact have performed the contracts, as they had many others. The government claimed that the “loss” was the $142,400 face amount of the contracts obtained, but the court disagreed. The court rejected as “irrational” a holding that would apply the same sentence against a performing contractor who lied on its application as against “a con artist who intended to winkle $142,400 ... from a senile old lady.” Id. at 559. Instead, the court held that because the Schneiders were the low bidders the government’s only financial “loss” was its expenses in recontracting, which the government there had failed to prove. We agree with the Seventh Circuit’s analysis and conclusion that the government’s theory was simple, but irrational. The result also has underpinnings in the guideline text and Commentary themselves, as we have explained above.
To support its conclusion, the Schneider court cited United States v. Whitehead, 912 F.2d 448 (10th Cir.1990), a case that also supports our earlier reasoning. There the defendant presented fraudulent documents when renting a home and obtaining an option to purchase. The home was worth $168,000, and the government claimed that the whole amount was the “loss.” The court disagreed, reasoning that Whitehead had not yet attempted to purchase the house itself, and it was uncertain that he would ever succeed in doing so, or that the full value of the home would be lost if he did. Therefore only the value of the option ($2,000) counted as the “loss.” Id. at 451-52. We certainly agree that the “loss” was not the $168,000, although we are not certain that even the entire $2,000 option value was properly considered as “loss.” 17
We also find persuasive United States v. Hughes, 775 F.Supp. 348 (E.D.Cal.1991). There the defendant conspired to present false loan applications to buy three homes. Hughes, however, clearly had no intention of defaulting on the loans. Indeed, Hughes, a male friend, and their female companions lived in the homes; the payments on two loans were in good standing, and although the third house was sold in bankruptcy, it was sold at a profit. The court also found that no actual economic loss had been realized, and the government could not prove that default was probable or that any future loss was expected. Discussing U.S.S.G. § 2F1.1 and its Commentary in detail, and rejecting blind application of the theft guideline, the court found that the “loss” was zero and that no sentence enhancement was required.
The government cites a number of fraud guideline cases that appear to support its view because they upheld “loss” calculations of the full amount fraudulently obtained. On closer analysis, however, these cases do not conflict with our reasoning. For example, in United States v. Wills, 881 F.2d 823 (9th Cir.1989), the defendant was guilty of credit card fraud. He masterminded a scheme to take $52,000, but $25,-000 from one of the transactions was recovered. The court summarily (and properly) upheld the “loss” calculation of $52,000, specifically because the defendant intended to cause a $52,000 loss. Id at 827. Similarly, in United States v. Davis, 922 F.2d 1385 (9th Cir.1991), Davis would call a business with an order and promise to pay by wire transfer. He would then have someone call the business and pose as a bank officer, saying that the wire transfer had arrived. He would then receive the goods. In the particular instance in the case, the actual loss was zero because jewels were never shipped. The court, however, properly held that the loss was the market value of the jewels — again specifically because the probable and intended losses were higher than actual loss. Id. at 1391-92.
The government relies most heavily on United States v. Johnson, 908 F.2d 396 [533]*533(8th Cir.1990). There Johnson had a poor credit history, so she obtained false new identification and fraudulently obtained $22,000 in car loans under that new name. The court held that the full $22,000 was the proper measure of the “loss,” even though the two banks involved eventually recovered most of the money by repossessing Johnson’s cars, saying that “the focus for sentencing purposes should be on the amount of the possible loss which Johnson attempted to inflict on the banks.” Id. at 398.
The Eighth Circuit’s opinion is ambiguous, but it may in fact be consistent with our approach. Although the court did mention “possible loss,” it did so only in referring to the loss that Johnson attempted (and therefore intended) to inflict. And in the next sentence, the court referred to Application Note 7’s discussion of “probable or intended loss” as an alternative measure to actual loss. Id. On a reasonable reading of the case, therefore, the court held that Johnson intended to commit actions she knew would inflict the full $22,-000 loss. See Schneider, 930 F.2d at 559 (reading Johnson as a case where the defendant had no intention to repay). If so, we agree with the holding, but if Johnson did legally equate “possible loss” with “probable or intended” loss, we must reject that linguistic stretch.
In another case also styled United States v. Johnson, 941 F.2d 1102 (10th Cir.1991), the Tenth Circuit unhesitatingly applied the “value of the property taken” language from U.S.S.G. § 2B1.1 Application Note 2 in remanding a sentence for mail fraud under U.S.S.G. § 2F1.1. Johnson somehow bought eighteen homes worth over $400,000 from a realty company with no money down, even though he was only making $1200 each month in salary. He never made a payment even though he collected $16,000 in rent. The lender foreclosed, and Johnson was convicted of mail fraud and equity skimming. The court of appeals agreed with the district court that the “loss” covered both the value of the real estate and the cash rents, but its entire discussion of “loss” centered on the Commentary to U.S.S.G. § 2B1.1: the real estate was “taken” and the reacquisition by foreclosure was irrelevant. Id. at 1113-14. The court remanded only because it was unsure of the fair market value of the real estate under USSG § 2B1.1. Id. at 1114-15. The same result could easily (and more properly) have been reached under the fraud guideline by determining that Johnson intended to cause such a loss. To the extent the Tenth Circuit believes that fraud “loss” must be calculated under the theft guideline, we disagree for the reasons outlined above.
The Second Circuit has also rejected the holding advanced by the defendant here. In United States v. Brach, 942 F.2d 141 (2d Cir.1991), Brach fraudulently procured a loan of $250,000 from a town. Upon discovering the fraud, the town demanded the money back. Brach refused, but after the FBI came, he returned the money. The Second Circuit upheld the use of the $250,-000 face value of the loan as the “loss,” rather than the few days’ interest the victim actually lost. The court relied on the cross-reference to U.S.S.G. § 2B1.1 and specifically declined to treat fraud cases any differently from theft cases. Id. at 143. According to the court, the “loss” was therefore the amount taken and put at risk, and Brach’s claimed intent to repay the loan and the trivial ultimate harm to the victim were irrelevant. Id. (also equating “probable” loss with the amount put at risk). Although the result in Brach may have been correct, we respectfully decline to follow its reasoning. If the court believed that Brach intended to steal the whole $250,000 (not simply to borrow it), then the intended loss was $250,000 and the result was proper. But if Brach intended to repay the borrowed money — something the opinion declined to address — the case is more like the present one, and we disagree with the Second Circuit. In our view, Application Note 7 to U.S.S.G. § 2F1.1 straightforwardly requires the sentencing court to use actual loss in the first instance. Indeed, actual and intended loss, the two factors that the Second Circuit insisted were irrelevant, 942 F.2d at 143, are the primary proper factors in determining a “loss” under the fraud guideline.18
[534]*534C. The Post-Sentencing Amendments to U.S.S.G. § 2F1.1
The Sentencing Commission’s recent (postsentencing) elucidation of proper “loss” calculation in the fraudulent loan procurement context buttresses our interpretation and that of the case law on which we rely, although we emphasize that we need not and do not rely upon the postsen-tencing amendments as “clarifications” of the fraud guideline in holding as we do. See also United States v. Ofchinick, 877 F.2d 251, 257 n. 9 (3d Cir.1989) (proposed amendment purporting to clarify sentencing guideline noted for its support, but discussion explicitly labeled as not necessary to the result). Technically, the post-sentencing amendments are subsequent legislative history, always a controversial interpretative tool. See note 9. As subsequent legislative history, the amendments do not directly apply retrospectively to earlier sentencings, but they may still have limited relevance as indications of what the guidelines in effect here meant. We also feel it necessary to discuss the amendments in light of our ultimate decision to remand: those amendments not posing ex post facto problems will be in effect at resentencing.
The most recent round of guideline amendments went into effect on November 1, 1991. Of the recent amendments, two address “loss” calculation. The first responded to section 2507 of the Crime Control Act of 1990, Pub.L. 101-647, 104 Stat. 4789, 4862, reprinted in note following 28 U.S.C.A. § 994 (West Supp.1991), which instructed the Commission to promulgate or amend guidelines to require an offense level of at least 24 for defendants deriving over $1 million in “gross receipts” from crimes affecting financial institutions.19 The Commission responded by amending several guidelines, including U.S.S.G. § 2F1.1, to increase offense levels by four levels (but at least to level 24) for defendants deriving over $1 million in gross receipts from their bank crimes. See 1991 Guidelines Manual Appendix C at 184-85 (amendment 364). Because the new amendment’s effective date postdates the defendant’s crime, and it works a substantive change that could only be detrimental to the defendant, it obviously cannot apply to him, whether or not he received $1 million in “gross receipts.” But the Commission’s use of both “gross receipts” and “loss” in the same Commentary does suggest that the two terms do not mean the same thing (to today’s Commission, at least), which renders the government’s position here even more dubious. See also United States v. Hughes, 775 F.Supp. at 352 (same reasoning).
Even better evidence that the current Commission does not define “loss” as the amount fraudulently obtained comes from the other recent amendment to U.S.S.G. § 2F1.1 See 1991 Guidelines Manual Appendix C at 221-24 (amendment 393). After the changes, the relevant portions of Application Note 7 to U.S.S.G. § 2F1.1 now read:
Valuation of loss is discussed in the Commentary to § 2B1.1 (Larceny, Embezzlement, and Other Forms of Theft). Consistent with the Provisions of § 2X1.1 (Attempt, Solicitation, or Conspiracy), if an intended loss that the defendant was attempting to inflict can be determined, this figure will be used if it is greater than the actual loss. Frequently, loss in a fraud case will be the same as in a theft case. For example, if the fraud consisted of selling or attempting to sell $40,000 in worthless securities, or representing that a forged check for $40,000 was genuine, the loss would be $40,000. There are, however, instances where ad[535]*535ditional factors are to be considered in determining the loss or intended loss:
(b) Fraudulent Loan Application and Contract Procurement Cases
In fraudulent loan application cases and contract procurement cases where the defendant’s capabilities are fraudulently represented, the loss is the actual loss to the victim (or if the loss has not yet come about, the expected loss). For example, if a defendant fraudulently obtains a loan by misrepresenting the value of his assets, the loss is the amount of the loan not repaid at the time the offense is discovered, reduced by the amount the lending institution has recovered, or can expect to recover, from any assets pledged to secure the loan.
In some cases, the loss determined above may significantly understate or overstate the seriousness of the defendant’s conduct. For example, where the defendant substantially understated his debts to obtain a loan, which he nevertheless repaid, the loss determined above (zero loss) will tend not to reflect adequately the risk of loss created by the defendant’s conduct. Conversely, a defendant may understate his debts to a limited degree to obtain a loan {e.g., to expand a grain export business), which he genuinely expected to repay and for which he would have qualified at a higher interest rate had he made truthful disclosure, but he is unable to repay the loan because of some unforeseen event {e.g., an embargo imposed on grain exports) which would have caused a default in any event. In such a case the loss determined above may overstate the seriousness of the defendant’s conduct.
Application Note 7 still cross-references the theft guideline Commentary, but notes that “[frequently, loss in a fraud case will be the same as in a theft case” (emphasis added). Thus the Commission implicitly recognized that occasionally the definitions will differ. Moreover, although Application Note 7 no longer discusses actual, probable, and intended loss in precisely the same way, a “loss” is generally still to be calculated as we concluded above: actual loss incurred at the time of sentencing remains the basic “loss,” and intended (attempted) loss is substituted if greater.20 More tellingly, new subheading (b) in Appli- ■ cation Note 7 specifically addresses fraudulent loan application cases and supports our earlier analysis entirely. It plainly states that where, as here, the defendant fraudulently misstates its assets, the “loss” is the victim’s actual loss — unpaid principal and interest less the amount the lender has recovered (or can expect to recover) from the loan collateral. As the revised Application Note suggests, that “loss” calculation may well over- or understate the seriousness of the offense, but the proper solution under former Application Notes 9 and 10 and current Application Note 10 is to depart upward or downward from the sentencing range.
We accordingly find nothing in the subsequent legislative history that causes us to doubt our analysis under the original guidelines themselves. Indeed, to the limited extent we consider the subsequent amendments to U.S.S.G. § 2F1.1 at all, our earlier conclusions are only confirmed.
D. Summary
In sum, a close examination of the fraud guideline and its entire official Commentary requires the conclusion that the district court erred here by equating the “loss” with the full amount of the loan. Although the courts have split on how to define fraud “loss,” we find the logic in Schneider and Hughes compelling, and we believe that the contrary case law relies on a flawed equation of fraud and theft crimes. From the recent amendments to U.S.S.G. § 2F1.1, it appears that the Sen-[536]*536tenting Commission also agrees. We therefore confirm our tentative view above and hold that fraud “loss” is, in the first instance, the amount of money the victim has actually lost (estimated at the time of sentencing), not the potential loss as measured at the time of the crime. However, the “loss” should be revised upward to the loss that the defendant intended to inflict, if that amount is higher than actual loss.
The record in this case requires vacatur of the judgment of sentence and remand to the district court for resentencing. The district court made no findings on actual or intended loss, and the parties contest both amounts. As discussed above, the government claims that actual loss was at least $3.4 million, while the defendant claims that the bank’s records cannot support an actual loss of anywhere near that amount. Also, the defendant claims he intended to repay the loan and so intended no loss, while the government (somewhat belatedly) contends that the defendant’s failure to make any payments after receiving the second loan installment belies his claim. It is the district court’s province to resolve these questions, and we leave it to that court on remand to decide whether further hearings on these issues are necessary.21 We must also remand because only the district court is authorized to determine whether the properly calculated “loss” significantly over- or understates the gravity of the crime, and therefore whether departure from the normal sentencing range is appropriate.22
III. CONCLUSION
The district court erred in concluding that a “loss” under the fraud sentencing guideline, U.S.S.G. § 2F1.1, is always the amount fraudulently obtained, regardless of intended or actual loss. We will therefore vacate the judgment of sentence and remand for resentencing consistent with this opinion.