SELYA, Circuit Judge.

When a criminal defendant is convicted of a fraud offense, the Sentencing Commission has established amount of loss — generally the higher of actual or intended loss — as a rough proxy for determining the seriousness of the offense and the relative culpability of the offender. Although this concept is easily stated, its application often has vexed sentencing courts. As in so many other instances, the devil is in the details.

This appeal requires us to decide two issues of first impression in this circuit. The first involves the method for calculating the guideline enhancement for amount of loss in an insurance fraud context. The second involves the method for calculating the amount of statutory restitution in that context. Concluding, as we do, that the court below erred in adopting its calculation methods, we vacate the appellant's sentence and remand for resentencing.

I. BACKGROUND

Because this sentencing appeal follows a guilty plea, we distill the pertinent facts from the plea agreement, the change-of-plea colloquy, the undisputed portions of the presentence investigation report (PSI Report), and the transcript of the disposition hearing. See United States v. Almonte-Nuñez, 771 F.3d 84, 86 (1st Cir. 2014).

Defendant-appellant John S. Alphas owns and operates The Alphas Company, a wholesale produce distributor located in Chelsea, Massachusetts. During the relevant time frame, the appellant purchased large quantities of produce, often from distant purveyors. To protect his investment, the appellant routinely obtained insurance on these produce shipments.

In or around March of 2007, the appellant devised a scheme to defraud. Over the next four and one-half years, he submitted at least ten fraudulent claims to his insurers for lost, stolen, or damaged produce.1 The appellant sought reimbursement for the value of allegedly lost, stolen, or damaged produce, together with disposal expenses, shipping fees, and the cost of procuring replacement stock. These claims were largely bogus: in each instance, the appellant either invented fictitious losses or artificially inflated legitimate losses. To make matters worse, he supported his submissions with documents that had been fraudulently altered or, in some cases, constructed out of whole cloth. He compounded his mendacity by making false and misleading statements to insurance adjusters.

The insurance policies at issue contained void-for-fraud clauses. A representative clause stated:

Four of the appellant's claims were never paid: three were withdrawn after suspicions surfaced and the lone claim submitted to Selective Insurance Company was thwarted by early detection of the fraud. The other six claims were paid, but mostly in amounts less than their face value. In sum, the ten claims totaled over $490,000, yet the appellant received payments totaling only $178,568.41.

The appellant's persistent pattern of chicanery sparked a federal criminal investigation. In May of 2014, the government filed an information charging the appellant with a single count of wire fraud. See 18 U.S.C. § 1343. Waiving prosecution by indictment, the appellant pleaded guilty to the charge. In an accompanying plea agreement, the parties stipulated to a base offense level of 7. See USSG § 2B1.1(a)(1).

The parties could not agree, however, as to the amount of loss — a necessary step in determining the guideline sentencing range (GSR). To fill this void, the PSI Report recommended boosting the appellant's offense level by 14 levels based on an intended loss of approximately $480,000. See id. § 2B1.1(b)(1)(H)-(I) (increasing offense level by 14 for losses exceeding $400,000 but not greater than $1,000,000). This calculation derived from the probation officer's conclusion that intended loss should be measured by the face amount of the appellant's claims, less a $1000 per claim deductible.

The appellant objected to the recommendation. He argued that the loss figure should exclude legitimate losses embedded in the fraudulent claims. In conjunction with this argument, he submitted a competing set of calculations that purported to show which component amounts were legitimate. This set of calculations yielded an intended loss amount of roughly $178,000 which corresponded to an increase of 10 (rather than 14) in his adjusted offense level. See id. § 2B1.1(b)(1)(F).

The PSI Report also dealt with restitution. It recommended an award of $178,568.41 (the aggregate amount actually paid out on the claims). The appellant objected to this recommendation, envisioning a finding of actual loss (and, therefore, a restitution amount) of $58,931.36. Once again, the appellant attributed the reduced figure to the fact that a portion of the claims paid corresponded to legitimate losses.

The probation officer stood by her calculations regarding both the guideline enhancement and restitution. Although she conceded that the appellant may have incurred legitimate losses, she maintained that the appellant's fraud rendered the claims altogether illegitimate.

The disposition hearing was held on November 6, 2014. The district court concluded as a matter of law that where, as here, insurance policies contain void-for-fraud clauses, intended loss is equal to the aggregate face value of the claims submitted. Starting with this premise, the court overruled the appellant's objections; enhanced his offense level by 14 levels; deducted 3 levels for acceptance of responsibility, see id. § 3E1.1; placed the appellant in Criminal History Category I; and set the GSR at 27 to 33 months, see id. Ch. 5, Pt. A, sentencing table. The court then varied sharply downward, sentencing the appellant to an incarcerative term of 12 months and 1 day. In addition, the court fined the appellant $60,000; attached a 36-month term of supervised release; and ordered payment of restitution to Zurich North American in the amount of $178,568.41.

This timely appeal ensued. The district court refused to stay the appellant's sentence pending appeal. This court, however, granted a stay. See United States v. Alphas, No. 14-2228 (1st Cir. Dec. 31, 2014) (finding that appeal presented "`substantial question' within the meaning of 18 U.S.C. § 3143(b)(1)(B)").

II. ANALYSIS

The appellant assigns error to the district court's determination of the amount of loss with respect to both the guideline enhancement and the award of restitution. Because these determinations are controlled by different authorities, we bifurcate our analysis.

A. The Guideline Enhancement.

The appellant first contends that the sentencing court used an improper method of calculating intended loss and, thus, erred in enhancing his offense level by 14 levels. This contention is met at the threshold by the government's assertion that we need not reach the merits because any calculation error was harmless. We begin with that assertion and then proceed to the merits.

1. Harmlessness. The government's argument for harmlessness rests on the district court's imposition of a below-the-range sentence coupled with the court's later remark, made while denying the appellant's motion to stay the sentence pending appeal, that a recalculation of the GSR would be unlikely to result in a lower sentence. This argument flies in the teeth of conventional wisdom, which teaches that the improper calculation of a defendant's guideline range comprises a significant procedural error. See Gall v. United States, 552 U.S. 38, 51, 128 S.Ct. 586, 169 L.Ed.2d 445 (2007). Such an error ordinarily requires resentencing. See United States v. Ramos-Paulino, 488 F.3d 459, 463-64 (1st Cir.2007). Indeed, a defendant normally can appeal from an error in calculating his GSR even though the district court imposed a sentence beneath the bottom of the GSR. See United States v. Paneto, 661 F.3d 709, 715 (1st Cir.2011).

To be sure, an appellate court may deem such an error harmless if, after reviewing the entire record, it is sure that the error did not affect the sentence imposed. See Williams v. United States, 503 U.S. 193, 203, 112 S.Ct. 1112, 117 L.Ed.2d 341 (1992). In other words, resentencing is required if the error either affected or arguably affected the sentence. See Ramos-Paulino, 488 F.3d at 463.

The case at hand does not fit within these narrow confines. Although the court below imposed a sentence beneath the bottom of the GSR, there is at least a possibility that the court would have imposed an even more lenient sentence had it started with a lower GSR. See United States v. Foley, ___ F.3d ___, ___ n. 13 (1st Cir. 2015) [Nos. 13-1048, 13-1118, slip op. at 31 n. 13]. While the court stated that a lower GSR was "unlikely" to result in a different sentence, we have recently reaffirmed that the possibility of a lesser sentence is enough to preclude a finding that an error in calculating the GSR is harmless. See id. Such a possibility exists here: saying that an event is unlikely is not the same as a categorical assurance that the event will not come to pass. It follows that if the court below erred on the high side in fashioning the guideline enhancement for amount of loss (and, thus, the GSR), we cannot regard that error as harmless.

2. Calculating Intended Loss. Our rejection of the government's harmlessness argument brings us to the merits of the appellant's claim that the district court committed a calculation error. We review de novo the court's interpretation and application of the sentencing guidelines, including the propriety of its loss-computation method. See United States v. Prange, 771 F.3d 17, 35 (1st Cir.2014); United States v. Walker, 234 F.3d 780, 783 (1st Cir.2000).

In fraud cases like this one, the guidelines direct the sentencing court to augment the defendant's offense level based on the amount of loss. See USSG § 2B1.1(b)(1). For this purpose, loss is defined as "the greater of actual loss or intended loss." Id. § 2B1.1, comment. (n. 3(A)). Here, the district court based its guideline calculation on intended loss. That term means "the pecuniary harm that was intended to result from the offense." Id. at n. 3(A)(ii). Intended loss is not synonymous with probable loss. Rather, the term "includes intended pecuniary harm that would have been impossible or unlikely to occur." Id. Seen in this light, intended loss "is a term of art meaning the loss the defendant reasonably expected to occur at the time he perpetrated the fraud." United States v. Innarelli, 524 F.3d 286, 290 (1st Cir.2008). This standard focuses primarily on the offender's objectively reasonable expectations, see id., though subjective intent may play some role, see id. at 291 n. 6; United States v. McCoy, 508 F.3d 74, 79 (1st Cir.2007).

In this case, the sentencing court concluded that the intended loss was equal to the aggregate face value of the claims submitted by the appellant to his insurers, not the aggregate amount by which the appellant fraudulently inflated those claims.2 The court reasoned that had the insurers known of the fraud when they received the claims, they would have invoked the void-for-fraud clauses and paid nothing. The court did not explain what bearing the void-for-fraud clauses may have had on the amount of loss that the appellant intended to cause. It said only that if intended loss were to be viewed solely as the amount of fraudulent inflation, fraudsters would have an incentive to inflate claims because, if caught in the act, they would be punished only for the inflated amount.

The government applauds this reasoning, but the appellant decries it. He notes that the sentencing guidelines treat loss as a proxy for relative culpability. In his view, basing an intended loss computation on the entire amount of an insurance claim rather than on the amount fraudulently claimed irrationally conflates the culpability of fraudsters who commit significantly different offenses.

Fraud has many manifestations, and calculating the loss associated with a particular scheme is sometimes more art than science. As a result, we have eschewed rigid rules and instead taken "a pragmatic, fact-specific approach" to loss calculation. Prange, 771 F.3d at 35 (internal quotation mark omitted). Such a pragmatic view suggests that loss computation should distinguish between an outright swindler who peddles, say, a forged title to a bridge in Brooklyn and a fraudster who contrives to render some value (albeit less than promised) to the victim. See id. at 36; United States v. Blastos, 258 F.3d 25, 30 (1st Cir.2001). By any practical measure, the former seems more culpable than the latter.

Applying the degree-of-culpability approach to the insurance fraud context, it is appropriate for the loss-computation method to distinguish between a fraudster who wholly fabricates a non-existent claim and a fraudster who artificially inflates a legitimate claim. A fraudster who has suffered no loss at all but invents a $100,000 claim out of thin air is not the same as a fraudster who has suffered a legitimate $50,000 loss but artificially inflates his claim to $100,000. See United States v. Smith, 951 F.2d 1164, 1167 (10th Cir.1991).

This gets to the heart of the matter. Under the sentencing guidelines, loss generally does not include sums that a victim would have paid to the defendant absent the fraud. See, e.g., United States v. Evans, 155 F.3d 245, 253 (3d Cir.1998) (holding that actual loss in insurance fraud case does not include value of legitimate claims); United States v. Parsons, 109 F.3d 1002, 1004 (4th Cir.1997) (same, in context of government benefits fraud); see also United States v. Miller, 316 F.3d 495, 498-99 (4th Cir.2003) (applying this principle to intended loss). Given this rule, the question reduces to whether the presence of a void-for-fraud clause makes a dispositive difference when calculating loss under the sentencing guidelines.

In resolving this question, we find instructive the line of cases involving government benefits fraud. When a person fraudulently obtains government benefits, the United States typically is entitled to recover all sums paid (in a civil forfeiture proceeding) even though some benefits would have been paid anyway (that is, absent the fraud). See, e.g., 5 U.S.C. § 8148(a); 28 U.S.C. § 2514. But when the same fraudulent conduct undergirds a criminal conviction, courts have steadfastly refused to equate the amount of loss under the sentencing guidelines with the amount recoverable by the government through civil forfeiture. That is because "[f]orfeiture is a penalty imposed on a criminal independent of any loss to the crime victim." Parsons, 109 F.3d at 1005. Consequently, "[t]he loss itself (whether the actual or intended loss) is limited to the tangible economic loss of the victim." Id. at 1004. The forfeitable amount does not count toward the government's loss, which is measured by the amount of benefits that was (and, in the case of intended loss, would have been) paid as a result of the fraud. See United States v. Harms, 442 F.3d 367, 380 (5th Cir.2006); United States v. Dawkins, 202 F.3d 711, 715 (4th Cir. 2000).

To be sure, government benefits fraud is governed by a special application note, which explains that in such cases loss "shall be considered to be not less than the value of the benefits obtained by unintended recipients or diverted to unintended uses." USSG § 2B1.1, comment. (n. 3(F)(ii)). We believe that this note merely represents a specialized application of the guidelines' general focus on a defendant's relative culpability. See, e.g., Dawkins, 202 F.3d at 714; Parsons, 109 F.3d at 1004-05. Since this is so, it makes good sense to transplant the reasoning of the government benefits cases to the analogous context of insurance fraud.

We discern no sound basis for treating a void-for-fraud clause (in the insurance fraud context) differently than a civil forfeiture (in the benefits fraud context). Both the void-for-fraud clause and the civil forfeiture anodyne afford relief after the fact. So, too, such a clause, like the sword of Damocles inherent in a threat of civil forfeiture, serves the salutary purpose of encouraging transactional honesty. And such a clause, like a civil forfeiture, imposes a penalty on the fraudster for acting corruptly: if the insurer discovers the fraud, the insured forfeits everything.

The concept of loss under the sentencing guidelines serves a completely different purpose. See United States v. Hamaker, 455 F.3d 1316, 1337 (11th Cir.2006); Dawkins, 202 F.3d at 715. The guidelines are designed to ensure that the sentence imposed on a defendant "reflect[s] the nature and magnitude of the loss caused or intended by [his] crimes." USSG § 2B1.1, comment. (backg'd.). Consonant with this design, the guidelines use amount of loss as the primary metric by which "the seriousness of the offense and the defendant's relative culpability" are measured. Id. Sums forfeited under a void-for-fraud clause (which is really nothing more than an after-the-fact penalty) do not conform naturally to this metric. Cf. id. § 2B1.1, comment. (n. 3(D)(i)) (providing that "loss" does not include "[i]nterest of any kind, finance charges, late fees, penalties, amounts based on an agreed-upon return or rate of return, [and] other similar costs"). It would therefore be anomalous to read the guidelines to distinguish between two fraudsters who fraudulently inflate their claims by exactly the same amount simply because one is covered by an insurance policy that contains a void-for-fraud clause and the other is covered by an insurance policy that does not contain such a clause.3

The government resists this reasoning, asseverating that the appellant intended to deprive his insurers of the entire amount claimed regardless of whether some portion of that amount represented legitimate losses. To achieve this counterintuitive result, the government asks us to consider what an objectively reasonable fraudster standing in the appellant's shoes would have expected to be paid were the fraud discovered. Because such a fraudster would be aware of the void-for-fraud clauses and know that the misrepresentation would relieve his insurers of their payment obligation, the government's thesis runs, the fraudster would expect his insurers to suffer losses in the full amount of the submitted claims.

We reject the government's thesis, which approaches intended loss from the wrong angle. Fraudsters do not expect to be found out but, rather, expect to reap the benefits of their contrivance. The relevant inquiry, then, is what the fraudster reasonably expected to euchre out of his victim, cf. id. at n. 3(A)(ii) (defining "intended loss" as "the pecuniary harm that was intended to result from the offense" (emphasis supplied)), not what would have slipped through his fingers had he been caught in the act.

That amount necessarily excludes any sums that the fraudster would have been paid absent the fraud. See Burrage v. United States, ___ U.S. ___, 134 S.Ct. 881, 887-88, 187 L.Ed.2d 715 (2014) (explaining that the phrase "results from" implies "a requirement of actual causality," which typically "requires proof that the harm would not have occurred in the absence of — that is, but for — the defendant's conduct" (internal quotation marks omitted)). The sentencing court, then, should have determined whether and to what extent legitimate claims were embedded in the fraud.

Structuring "intended loss" in this way makes good sense. After all, loss is meant to serve as a proxy for the seriousness of the crime and the relative culpability of the offender. The best way to gauge the seriousness of a fraud offense is to determine how much the fraudster set out to swindle — and no fraudster sets out to swindle sums that he would have been paid anyway. That is also the best way to gauge a fraudster's culpability.

The best case for the government's contrary position is United States v. Torlai, 728 F.3d 932 (9th Cir.2013) — but Torlai cannot carry the weight that the government loads upon it.4 Torlai involved a defendant who had obtained crop insurance policies from private insurers through a federal program. See id. at 935-36. Those policies and their application documents included void-for-fraud clauses. See id. at 940 & n. 8. The defendant made a number of misrepresentations both when he procured the policies and when he submitted claims under them. See id. at 936, 941-43.

Appealing his sentence, the defendant argued that the district court erred in calculating loss by failing to determine which portions of his indemnity claims were legitimate. The Ninth Circuit disagreed, noting that the case involved a government benefits program and, therefore, loss constituted "the value of the benefits obtained by unintended recipients." Id. at 938 (emphasis omitted) (quoting USSG § 2B1.1, comment. (n. 3(F)(ii))). The court stated that "[b]y virtue of his fraud, [the defendant] was not eligible for any government benefit under the crop insurance program, and therefore, he was not an `intended beneficiary.'" Id. at 943.

Certain facts underlying the court's "unintended beneficiary" rationale help to distinguish Torlai from this case. First, the Ninth Circuit found sufficient evidence that the defendant had suffered no reimbursable losses, see id. at 941-43, so he would not have received anything absent his fraudulent claims. Second, the defendant made material misrepresentations to procure the crop insurance policies, see id., making it unlikely that the insurer would have issued the policies had it been told the truth. This latter fact indicates that, regardless of the legitimacy vel non of the claimed losses, the insurer would not have been obliged to pay. Cases in which an insurer would have paid nothing absent the fraud are materially different from those in which the insurer would have paid something (but less than the full amount claimed) absent the fraud. See United States v. Sharma, 703 F.3d 318, 324-25 (5th Cir.2012).5

To recapitulate, we conclude that the district court erred in calculating the amount of intended loss attributable to the fraud and, thus, in fashioning the guideline enhancement. We are, therefore, constrained to remand for resentencing. On remand, the district court should compare what the appellant sought to bamboozle his insurers into paying with what they would have paid had the appellant submitted only bona fide claims.

We add a coda. It is apodictic that the government bears the burden of proving the applicability of a sentencing enhancement by preponderant evidence. See Paneto, 661 F.3d at 715. But in a case such as this — where a defendant's claims were demonstrably rife with fraud — a sentencing court may use the face value of the claims as a starting point in computing loss. See United States v. Campbell, 765 F.3d 1291, 1304-05 & n. 13 (11th Cir.2014); United States v. Hebron, 684 F.3d 554, 562-63 (5th Cir.2012). The burden of production will then shift to the defendant, who must offer evidence to show (if possible) what amounts represent legitimate claims. See Hebron, 684 F.3d at 563; United States v. Jimenez, 513 F.3d 62, 86 (3d Cir.2008). After the record is fully formed, the sentencing court must determine the amount of loss that the government (which retains the burden of proof) is able to establish. The court, however, "need only make a reasonable estimate of the loss." USSG § 2B1.1, comment. (n. 3(C)); see Blastos, 258 F.3d at 30. Depending on the defendant's offer of proof, the court might well conclude that the amount of loss is equal to the face value of the submitted claims.

B. The Restitution Order.

The appellant further contends that the district court erred in calculating restitution. The government counters both that this claim is waived and that it is impuissant.

1. Purported Waiver. Waiver typically occurs when a party intentionally relinquishes a known right. See United States v. Rodriguez, 311 F.3d 435, 437 (1st Cir.2002). In the sentencing context, an appellant may waive an issue when he initially raises it as an objection to the PSI Report but later explicitly withdraws the objection. See United States v. Eisom, 585 F.3d 552, 556 (1st Cir.2009); Rodriguez, 311 F.3d at 437. Once an issue is waived, there is nothing for an appellate court to review. See Rodriguez, 311 F.3d at 437.

Here, the appellant challenged the loss computation both in objections to the PSI Report and in his sentencing memorandum. His objection was directed to the calculation of both the guideline enhancement and the restitution award. A common thread ran through both halves of his argument: loss simply does not include amounts that the victim would have paid absent the fraud. At the disposition hearing, the appellant's trial counsel forcefully presented this argument in connection with the calculation of the guideline enhancement. The court rejected it. Later on, the appellant's counsel was asked whether he wanted the court to consider any other modifications to the PSI Report. Counsel responded in the negative.

The sentencing proceeding continued. At one point, the court remarked that it did not understand the amount of restitution to be disputed and asked defense counsel whether he was looking at the same set of numbers. Counsel responded that he was.

The government urges us to find that these (and similar) statements amounted to an intentional relinquishment of the right to contest the amount of restitution. This exhortation has some superficial appeal: taken in a vacuum, the sentencing colloquy can be read as an acknowledgment that the appellant was not pressing any arguments about the amount of restitution. It surely would have been better practice for counsel to have clarified that he had no objections to the restitution amount save for the method of calculation (that is, the court's announced unwillingness to deduct any legitimate portions of the claims from the loss calculation). But reading the sentencing transcript as a whole, we do not think that defense counsel's statements clear the high bar needed for a finding of waiver.

Waivers are strong medicine, see Pike v. Guarino, 492 F.3d 61, 72 (1st Cir.2007), and that medicine should not be dispensed in criminal cases where ambiguity lurks, see United States v. Bradstreet, 207 F.3d 76, 79-80 (1st Cir.2000). On this scumbled record, defense counsel's statements, taken in the context of the sentencing hearing as a whole, were ambiguous. The main event at sentencing was the battle over the method of calculating loss.6 Having lost that battle after a full airing, counsel's statements can fairly be read as an acknowledgment that the court's arithmetic (though not its method of calculation) was correct with respect to restitution. Any other reading would be illogical: it would have been odd for the appellant to press his computational argument to the bitter end with respect to the guideline enhancement and then impliedly abandon the very same argument with respect to restitution.

The short of it is that there was no waiver. Defense counsel's statements are most sensibly read as a recognition that the critical issue already had been decided against his client, not as a concession of that issue.

As a fallback, the government suggests that the appellant forfeited the restitution issue by not raising it at the disposition hearing, thus engendering review for plain error. See Rodriguez, 311 F.3d at 437. This argument is fruitless: the appellant raised the issue in written objections to the PSI Report and in his sentencing memorandum, and his attorney presented a legal basis for the argument to the sentencing court. No more was exigible to avoid a finding of forfeiture. See United States v. Theodore, 468 F.3d 52, 58 (1st Cir.2006).

2. Amount. The district court ordered restitution pursuant to the Mandatory Victims Restitution Act (MVRA), 18 U.S.C. § 3663A. Since the appellant's challenge to the restitution order is based on an alleged error of law, our review is plenary. See United States v. Cutter, 313 F.3d 1, 6 (1st Cir.2002); see also Walker, 234 F.3d at 783 (explaining that proper method of loss computation is "a prototypical question of legal interpretation").

The MVRA authorizes a court to award restitution only in the amount of a victim's actual loss. See Innarelli, 524 F.3d at 294-95. Thus, the question boils down to whether the presence of the void-for-fraud clauses in the insurance policies converts the entire amount paid in response to the appellant's claims into an actual "loss."

The answer depends, in relevant part, on the MVRA's definition of a victim as a "person directly and proximately harmed as a result of the commission of an offense for which restitution may be ordered."7 18 U.S.C. § 3663A(a)(2). Mindful of this definition, we have held that, under the MVRA, a court may only order restitution for losses that have an adequate causal link to the defendant's criminal conduct. See Cutter, 313 F.3d at 7.

Importantly, we have made it pellucid that "restitution should not be ordered if the loss would have occurred regardless of the defendant's misconduct." Id. This means that the government must establish a but-for connection between the defendant's fraud and the victim's loss. See United States v. Vaknin, 112 F.3d 579, 590 (1st Cir.1997) (interpreting parallel provision in Victim and Witness Protection Act, 18 U.S.C. § 3663). Our case law on this point is consistent with the weight of authority under the MVRA. Actual loss is widely (and correctly) thought to be limited to pecuniary harm that would not have occurred but for the defendant's criminal activity. See, e.g., Sharma, 703 F.3d at 324; United States v. Petruk, 484 F.3d 1035, 1038 (8th Cir.2007); United States v. Feldman, 338 F.3d 212, 220-21 (3d Cir. 2003); Dawkins, 202 F.3d at 715.

Applying the MVRA and the case law interpreting it, we think it evident that the district court erred in ordering restitution in the full amount paid to the appellant simply because the insurance policies included void-for-fraud clauses. While such clauses may suffice to ground claims for disgorgement in civil proceedings, an insurer's recoverable loss for MVRA purposes is confined to the amount the insurer would not have paid but for the fraud. See United States v. Chalupnik, 514 F.3d 748, 754 (8th Cir.2008); Petruk, 484 F.3d at 1038-39.

This gets the grease from the goose. The appellant has contended all along that some portion of his claims represented legitimate losses. This remains to be proven. But if it is true, Zurich North American presumably would have reimbursed him for those losses had he presented them without embellishment. The district court must, therefore, reconsider its restitution order, taking into account the extent (if at all) to which the appellant's claims encompassed legitimate losses.8 The district court need only make "a reasonable determination of appropriate restitution by resolving uncertainties with a view towards achieving fairness to the victim." United States v. Burdi, 414 F.3d 216, 221 (1st Cir.2005) (internal quotation marks omitted).

III. CONCLUSION

We need go no further. For the reasons elucidated above, we vacate the appellant's sentence (including the restitution order) and remand for resentencing consistent with this opinion. On remand, the only issues open to consideration shall be the appropriate amount of intended loss for purposes of the guideline enhancement, the appropriate amount of actual loss for purposes of restitution, and, of course, the dimensions of the sentence to be imposed.

Vacated and Remanded.