John A. Carrozzella appeals from a sentence of, inter alia, 110 months imprisonment imposed by Judge Thompson after Carrozzella pleaded guilty to one count of conspiracy to commit mail fraud, in violation of 18 U.S.C. § 371, and to three counts of mail fraud, in violation of 18 U.S.C. § 1341. On appeal Carrozzella challenges: (i) a two-level enhancement of offense level for violation of a judicial process under Guidelines § 2F1.1(b)(3)(B) based on his filing of false accounts with a Connecticut probate court; (ii) a four-level enhancement under Guidelines § 3B1.1(a) for being an organizer or leader or organizer of "otherwise extensive" criminal activity; and (iii) a loss calculation of over $10 million, resulting in an enhancement of 15 levels under Guidelines § 2F1.1(b)(1)(P). For the reasons set forth below, we reverse the enhancement under Section 2F1.1(b)(3)(B), remand the enhancement under Section 3B1.1(a) for further findings, and affirm the calculation of loss.

Carrozzella was an attorney in Wallingford, Connecticut, at one time in a partnership with Thomas J. Richardson and later in a solo practice. Carrozzella induced individuals, often clients of his law firm, to entrust money to him for investment. The government suggests that misrepresentations, particularly regarding tax-free income, induced some victims to turn over the funds. Carrozzella maintains that the investors knew that they were putting money into discretionary accounts, although he appears to have promised fixed rates of return.

Carrozzella invested the money in real estate, stocks, and bonds. The investments were in his name or in the name of partnerships in which he was involved. Carrozzella controlled all pertinent checking and savings accounts and all pertinent recordkeeping. It appears that some of the accounts in which the investment funds were kept and disbursed also were used for the law firm's revenues and disbursements.

Beginning at least as early as 1987, the investments began to perform poorly and were lost or significantly diminished. All agree that Carrozzella committed fraud in reports concerning the performance of the investments. Instead of disclosing their perilous state, Carrozzella represented that the investments were doing well in monthly letters to the investors. He also continued to solicit new funds notwithstanding the investments' depressed state because he needed fresh money for the scheme to continue. Some investors had directed that their "interest" not be reinvested. Periodic payments of "interest" had to be made to them lest they demand return of the principal, actions that would likely have caused the scheme to collapse. There were also real estate taxes and mortgage payments to be made. Carrozzella may also have paid off gambling debts. He therefore continued to persuade clients and others to entrust new money to him, right up until days before he resigned from the bar.

Carrozzella also filed false accounts with Connecticut probate courts. For example, in 1989, he represented to the father of a client that he would establish a trust to hold funds from a judgment received by the client for serious injuries suffered in an automobile accident. Carrozzella used the money to purchase real estate in his own name or in the name of partnerships in which he was involved and to make interest payments to prior investors. In June 1995, Carrozzella submitted to a Connecticut probate court an account in which he stated under oath that the trust contained $830,379.72 for distribution. The statement was wrong by some $830,379.72. Similarly, on May 3, 1995, Carrozzella had submitted an account to a probate court for another trust of which he was a fiduciary, stating under oath that a trust contained $345,316.04. That statement was wrong by some $345,316.04.

According to the presentence report as adopted by the district court, Thomas Richardson, Carrozzella's law partner, was aware of the scheme and participated in it. According to the report, Carrozzella also used the "unknowing services" of a personal secretary, three firm secretaries, and a bookkeeper. At the hearing on Carrozzella's objections to the presentence report, the prosecutor suggested that the bookkeeper was a knowing participant who had been given immunity. Apart from the statement itself, there is no factual basis for that assertion in the record. We may infer that the bookkeeper was aware of Carrozzella's investments on behalf of others and of the investments' perilous state. However, the record does not disclose whether she was informed of Carrozzella's fraudulently sanguine reports to investors. If she was so informed, the inference that she had to have known of the falsity of those representations would be overwhelming. The district court found that there were three knowing participants--apparently Carrozzella, Richardson, and the bookkeeper--and that Carrozzella also used the services of employees of the law firm, an accounting firm, stockbrokers, and partners in various real estate ventures in carrying out the fraud.

At a hearing on Carrozzella's objections to the presentence report, Richard Finkel, a forensic accountant, estimated the amount of loss due to Carrozzella's scheme. The accountant testified that in his opinion the amount of money initially entrusted to Carrozzella--the "opening balance"--was $8,001,125.34. The accountant determined that subsequent deposits by clients amounted to $5,041,124.01 and that their withdrawals totalled $4,451,139.75. In calculating the loss, the district court relied upon our decision in United States v. Brach, 942 F.2d 141 (2d Cir.1991), holding that sums returned to victims are not to be subtracted from the amount wrongfully taken. Using that approach, the loss would be some $13 million plus. However, the court also noted that an additional $1.7 million was lost in amounts Carrozzella handled for trusts, estates, and conservatorships. The court therefore concluded that a loss between $10 million and $20 million had occurred.

The district court concluded, inter alia, that Carrozzella should receive: (i) a two-level upward adjustment under Guidelines § 2F1.1(b)(3)(B) for "abuse" of the Connecticut probate court's process based on the false accounts to that court, (ii) a four-level upward adjustment under Guidelines § 3B1.1(a) for having been the leader or organizer of a criminal activity that was "otherwise extensive," and (iii) an enhancement of 15 levels under Guidelines § 2F1.1(b)(1)(P) for a loss of over $10 million.

Carrozzella argues that the district court erred in applying a two-level enhancement under Guidelines § 2F1.1(b)(3)(B) for "violation of any judicial or administrative order, injunction, decree, or process not addressed elsewhere in the guidelines." It is conceded that Carrozzella violated no court order, injunction, or decree directed specifically to him in filing false accounts with a Connecticut probate court. Nevertheless, the district court concluded that Carrozzella's conduct fell within Section 2F1.1(b)(3)(B) because it constituted an "abuse" and "violation" of the Connecticut probate "process." We disagree.

Carrozzella's conduct does not fit at all comfortably within Guidelines § 2F1.1(b)(3)(B). We note initially that the term "abuse" of process used by the district court is not found in that Section. The Section adjusts the offense level only for a "violation of any judicial ... process." The term "abuse" of process, which the district court used interchangeably with "violations" appears to have crept into the lexicon of Section 2F1.1(b)(3)(B) through cases in other circuits involving bankruptcy fraud. See United States v. Lloyd, 947 F.2d 339, 340 (8th Cir.1991) (per curiam); see also United States v. Michalek, 54 F.3d 325, 331-32 (7th Cir.1995); United States v. Bellew, 35 F.3d 518, 519 (11th Cir.1994) (per curiam). We discuss these cases below and note at this point only that use of the term "abuse" for purposes of Section 2F1.1(b)(3)(B) is worth reexamining before it is uncritically adopted by us.

"Abuse of judicial process" seems to us a term that encompasses a far greater range of conduct than does the term "violation of any judicial process." "Abuse of process" includes any serious misuse of judicial or administrative proceedings intended to inflict unnecessary costs or delay on an adversary or to confer undeserved advantages on the actor. It would include baseless complaints, motions, or defenses. It might also, but not inexorably, encompass fraudulent filings in, or representations to, courts or agencies. False accounts to probate courts, false filings with the Securities Exchange Commission, and even false tax returns filed with the Internal Revenue Service might qualify as abuses of judicial or administrative processes.

The term "violation of any judicial process"--the actual Guidelines language--seems considerably narrower. "Violation" strongly suggests the existence of a command or warning followed by disobedience. This analysis in turn suggests that the term "process"--the command or warning violated--is used, not in the sense of legal proceedings generally, but in the sense of a command or order to a specific party, such as a summons or execution issued in a particular action. See Bryan A. Garner, A Dictionary of Modern Legal Usage 698 (2d ed.1995). See also Second Circuit Local Rule § 0.19 ("All process of this court....") This narrower reading of Section 2F1.1(b)(3)(B) is also consistent with the general practice--known as ejusdem generis--of construing general language in an enumeration of more specific things in a way that limits the general language to the same class of things enumerated. See Garner at 308. In the present circumstances, the word "process" follows "injunction, order, [or] decree" and is most easily read in the narrower sense of a command or order issued to a specific person or party.

Moreover, the Commentary to Guidelines § 2F1.1(b)(3)(B) strongly supports the narrower reading. It states that "[a] defendant who has been subject to civil or administrative proceedings for the same or similar fraudulent conduct demonstrates aggravated criminal intent and is deserving of additional punishment for not conforming with the requirements of judicial process or orders issued by federal, state, or local administrative agencies." Guidelines § 2F1.1 Background. The examples given of conduct addressed elsewhere in the Guidelines are also limited to orders specifically directed to a defendant. The Commentary thus reinforces the view that the enhancement in question applies to fraud defendants only when they commit their fraud in the face of some type of official warning or order directed specifically to them.

Turning to the present case, Carrozzella's filing of false accounts with a probate court might well fall within the broader "abuse of process" view but would not seem to fall within the narrower view of a "violation of process." To be sure, Carrozzella violated a command not to file false accounts, but the command was a rule applicable to all trustees and not specifically directed to him.

Our doubt on this matter need not be fully resolved, however--no matter how great our skepticism--because the failure of Carrozzella's conduct to fit comfortably within Section 2F1.1(b)(3)(B) reveals a fatal problem with imposing that Section's two-level adjustment. That Section applies only to a "violation of any judicial or administrative order, injunction, decree, or process not addressed elsewhere in the guidelines." Guidelines § 2F1.1(b)(3)(B) (emphasis added).

In our view, Carrozzella's conduct in filing false probate accounts is "addressed elsewhere" because it fits quite comfortably within Guidelines § 3B1.3. Section 3B1.3 provides that "[i]f the defendant abused a position of public or private trust, or used a special skill, in a manner that significantly facilitated the commission or concealment of the offense, increase by 2 levels." Carrozzella's fraud as a trustee in submitting false accounts to the probate court seems to fit squarely within this provision. His position as a trustee in the probate court was characterized by precisely the type of discretion and consequent lack of supervision that the commentary to the guideline sets out as the key feature of a position of trust. Guidelines § 3B1.3 Application Note 1. Compare United States v. Jolly, 102 F.3d 46 (2d Cir.1996); United States v. Broderson, 67 F.3d 452 (2d Cir.1995). The accounts he filed were a standard part of a fiduciary's responsibility. Whether his relationship to the probate court and trust beneficiaries was one of public or private trust--or a hybrid--is not relevant because the Section covers both. Because Carrozzella's conduct is well within the heartland of the conduct addressed by Section 3B1.3 and fits only very doubtfully within Section 2F1.1(b)(3)(B), we hold that it is "addressed elsewhere." Although it does not affect our conclusion that the false probate accounts were not a violation of a judicial process, we do note that Carrozzella received--and does not challenge--an enhancement under Section 3B1.3.1

We find nothing in our caselaw that undermines our conclusion that Section 2F1.1(b)(3)(B) does not call for an enhancement based on Carrozzella's conduct. See United States v. Deutsch, 987 F.2d 878, 885-86 (2d Cir.1993) (violation of judicial order forbidding defendant from holding himself out as attorney); United States v. Probber, No. 94 Civ. 2027(RPP), 1994 WL 376083 * 1 (S.D.N.Y.1994) (violation of judicial order that a defendant cease fraudulent activity).

Our only concern regarding contrary decisions are the decisions in other circuits that have held that a Section 2F1.1(b)(3)(B) enhancement may be based upon bankruptcy fraud involving the concealing of assets. See United States v. Michalek, 54 F.3d 325, 331-32 (7th Cir.1995); United States v. Bellew, 35 F.3d 518, 519 (11th Cir.1994) (per curiam); United States v. Lloyd, 947 F.2d 339, 340 (8th Cir.1991) (per curiam); see also United States v. Gunderson, 55 F.3d 1328, 1332-33 (7th Cir.1995). As noted above, the theory of these cases is that bankruptcy fraud is an "abuse of process." Although we agree that such fraud may fall within that term in light of the automatic stay that freezes creditors from pursuing assets, it is, as we have noted, quite difficult to fit such conduct within the actual language of Section 2F1.1(b)(3)(B) or to match it with the accompanying commentary and examples. We do not reach the issue, however, because, unlike Carrozzella's conduct, bankruptcy fraud is not addressed by the Guidelines provisions relating to abuse of a position of trust. Trust and discretion are not accorded debtors in bankruptcy by their creditors, and bankruptcy proceedings are indeed adversarial. The bankruptcy fraud cases are, therefore, distinguishable, and we do not reach the questions of whether they should be adopted in this circuit.

B. Leadership of an "Otherwise Extensive" Criminal Activity

Carrozzella also argues that the district court incorrectly imposed a four-level enhancement under Guidelines § 3B1.1(a) for his being an organizer or leader of an extensive criminal activity. Although he does not challenge his organizer or leadership role, he argues that his criminal activity was not "otherwise extensive" within the meaning of that Section. We agree that the district court relied upon too broad a test for extensiveness.

Sentencing Guidelines § 3B1.1(a) provides a four-level increase in offense level "[i]f the defendant was an organizer or leader of a criminal activity that involved five or more participants or was otherwise extensive." Organizers or leaders of non-extensive criminal activities are subject only to the two-level enhancement of Guidelines § 3B1.1(c). (Carrozzella concedes that at least this two-level enhancement is appropriate.) Application Note 1 explains that "[a] 'participant' is a person who is criminally responsible for the commission of the offense, but need not have been convicted." Guidelines § 3B1.1. Because the government does not dispute that the number of criminal participants in this case was under five, any application of Guidelines § 3B1.1(a) must rest on Carrozzella's criminal activity having been "otherwise extensive."

As to what is "otherwise extensive," Application Note 3 states that "[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." Id. In United States v. Liebman, 40 F.3d 544 (2d Cir.1994), we concluded that "in using the term 'organization' in the commentary, the drafters did not intend to restrict the scope of the guideline to structured enterprises. It is more likely that they chose the word as a term of convenience, to denote any association of people involved in criminal activity." 40 F.3d at 549. Consonant with this view, the background commentary states that the adjustments in Guidelines § 3B1.1 are "based upon the size of a criminal organization (i.e., the number of participants in the offense) and the degree to which the defendant was responsible for committing the offense." Guidelines § 3B1.1 Background. This commentary and our decision in Liebman indicate that an adjustment under Guidelines § 3B1.1 is based primarily on the number of people involved, criminally and noncriminally, rather than on other possible indices of the extensiveness of the activity. See also United States v. Boula, 932 F.2d 651, 654 (7th Cir.1991); cf. United States v. Tai, 41 F.3d 1170, 1175 (7th Cir.1994) (district courts "still free to examine factors in addition to headcount"); United States v. McKenzie, 922 F.2d 1323, 1329 (7th Cir.) (cross-country trips and numerosity of transactions considered in extensiveness calculus) cert. denied, 502 U.S. 854, 112 S. Ct. 163, 116 L. Ed. 2d 127 (1991).

In limiting Section 3B1.1 primarily to head counting or analysis analogous to that, we carry out the intent of the Commission. So narrowed, of course, the term "extensive" is a not entirely accurate description of the characteristics of a defendant's criminal conduct that leads to a Section 3B1.1(a) enhancement. However, many characteristics that might ordinarily be considered evidence of "extensive" activity are dealt with elsewhere in the Guidelines. For example, in fraud cases, the base offense level can be raised according to the amount of loss, the extent of planning, and the number of victims. Guidelines § 2F1.1. Further adjustments can be made according to the vulnerability of the victim, § 3A1.1, the defendant's role, §§ 3B1.1, 3B1.2, and abuse of a position of trust, § 3B1.3. These factors all arguably relate to the "extensiveness" of a defendant's criminal activity, and, excluding the two-level enhancement for violation of judicial process discussed above and the net addition of two levels discussed here, these factors led to a 23 level enhancement in Carrozzella's case. Hypothetically, at least, there are factors other than head counting that are not weighed elsewhere in the Guidelines (e.g., the use of multiple identities by the defendant) which may be properly considered in the "otherwise extensive" determination. Moreover, it is possible that some factors considered elsewhere in the Guidelines might still be properly counted towards "extensiveness" in cases where the defendant's conduct so far exceeds the contemplation of the otherwise applicable Guideline (e.g., a crime resulting in a $10 billion loss that is enhanced under Guidelines § 2F1.1 as being "[m]ore than $80,000,000").

Section 3B1.1 is, therefore, not so much about extensiveness in a colloquial sense as about the size of the organization in terms of persons involved that a defendant "organize[d]" or "le[d]." It is one of two provisions--Section 3B1.2 being the other--that are designed in part to allocate culpability among defendants according to the size of the criminal organization and the relative significance or insignificance of the role played by each in it. See Guidelines § 3B1.1 Background. These Sections also reflect the view that supervisors and managers in large organizations may be more dangerous and more likely to recidivate, id., whereas the distinction between leaders and underlings is of less importance in smaller, less-structured organizations, id.

In the instant matter, Carrozzella does not dispute that he was "an organizer or leader" in his fraudulent scheme and therefore subject to a two-level enhancement under Section 3B1.1(c). Rather, he objects to the four-level enhancement for being "an organizer or leader" of a criminal activity that involved five or more participants or was otherwise extensive under Section 3B1.1(a).

Determining whether there were five or more knowing participants is the relatively easy task of determining how many individuals were criminally responsible for the offense. See Section 3B1.1 Application Note 1. Determining whether an organization is "otherwise extensive" in headcount terms is more difficult. The term is not well defined in the Guidelines. We are told only that the sentencing court must consider "all persons involved" and that "a fraud that involved only three participants but used the unknowing services of many outsiders could be considered extensive." Id. at Application Note 3. This guidance is of limited aid because "us[ing] the services of ... outsiders" has no evident limit save the absurdity of counting a gas station attendant who fuels the family car that days later carries the defendant to the scene of criminal acts. Even short of the absurd, the difficulties are exacerbated where, as here, the organization in question was a firm that engaged in legitimate activities while supporting a fraudulent scheme.

From caselaw we know that not every otherwise lawful activity directed by a defendant to further criminal activity, no matter how minimally, leads inexorably to counting the persons involved as unknowing participants for purposes of Section 3B1.1(a). Rather, the "otherwise extensive" language is deemed an alternative to finding five criminally responsible participants. In rejecting an argument that "otherwise extensive" is a lesser requirement than five knowing participants, the court in United States v. Tai, 41 F.3d 1170 (7th Cir.1994), noted that, "[g]iven the Sections's five participant prong, it would be anomalous to conclude that the presence of five individuals--not all of whom are participants--warranted an increase. At the very least, Section 3B1.1's 'otherwise extensive' prong demands a showing that an activity is the functional equivalent of an activity involving five or more participants." 41 F.3d at 1174; see also United States v. Mohammad, 53 F.3d 1426, 1436-37 (7th Cir.1995) (quoting Tai ). We follow these decisions.

In determining whether a criminal activity is "otherwise extensive" as the functional equivalent of one involving five or more knowing participants, we believe that the following must be determined by the sentencing court:

(i) the number of knowing participants;(ii) the number of unknowing participants whose activities were organized or led by the defendant with specific criminal intent;

(iii) the extent to which the services of the unknowing participants were peculiar and necessary to the criminal scheme.

The number of knowing participants, (i), is clearly relevant because a criminal scheme with four knowing participants that is aided by unknowing participants is more likely to be "otherwise extensive" than a scheme with a single knowing participant.

The number of unknowing participants who were organized or led with specific criminal intent, (ii), is necessary to separate out service providers who facilitate a particular defendant's criminal activities but are not the functional equivalent of knowing participants. Salespeople who unknowingly conveyed fraudulent misrepresentations at a defendant's request are on an entirely different footing from the taxi driver who brought a leader of the fraudulent scheme to work on a single occasion. The organization of, and direction given to, the salespeople is done with the specific intent of causing victims to be deceived into parting with their money. The direction to the taxi driver will usually be given without any such specific intent. For another example, the services of bank employees who facilitate the activities of a money launderer are on a different footing from those who establish accounts and accept deposits from persons who acquired the money illegally. See United States v. Allibhai, 939 F.2d 244, 252-53 (5th Cir.1991), cert. denied, 502 U.S. 1072, 112 S. Ct. 967, 117 L. Ed. 2d 133 (1992); see also United States v. Patasnik, 89 F.3d 63, 68-69 (use of lawyer, accountant, and loan brokers to carry out advance fee loan scam).

Finally, the nature of the services provided, (iii), will separate out the residue of incidental services by unknowing participants that are not the functional equivalent of knowing participation. A perpetrator anxious to close a fraudulent deal may occasionally have the crime in mind when hailing a cab to hurry to a meeting with victims. Even so, the taxi driver should not be counted. Lawful services that are not peculiarly tailored and necessary to a particular crime but are fungible with others generally available to the public are not the functional equivalent of knowing participation. To hold otherwise would sweep within Section 3B1.1(a) activity that is far less culpable than organizing and leading five knowing participants.

A final word. Section 3B1.1(a) does not specify the number of unknowing participants that would cause criminal activity to be deemed "otherwise extensive." Given that failure, we believe that more than a head-count of unknowing participants is needed to determine that an activity was "otherwise extensive" under the functional equivalence test. The use of unknowing participants to carry out a criminal activity may be more inefficient than the use of knowing participants. The need to conceal the true nature of the scheme from the former, for example, may limit the tasks for which they are suitable. Moreover, even if their participation falls within factors (ii) and (iii), it may still only minimally further the criminal activity. The extent of harm and degree of culpability in organizing or leading five unknowing participants may thus be less than in the case of organizing or leading five knowing participants. In no way, therefore, do we preclude a sentencing court from deciding in a particular case that organizing and leading five unknowing participants was not the functional equivalent of organizing or leading five knowing participants. Rather, in determining whether an activity was "otherwise extensive," a sentencing court may take into account the role and performance as well as the number of unknowing participants.

Turning to the present matter, the district court found Carrozzella's scheme "otherwise extensive"

because of a combination of factors, namely: One, that in addition to the three participants who were criminally responsible, the services of other employees of the law firm, an accounting firm and stockbrokers were used to carry out the scheme, and eight individuals who were not participants were involved as partners in real estate investments; two, that the fraud continued over a period of many years, from at least 1987 to June of 1995; three, that it involved a large number of investors; and four, that the scheme to defraud the investors or clients was sophisticated, involving the mailing of monthly letters to some of the investors misrepresenting the status of their investment.

We believe that these findings do not suffice to support the enhancement under Section 3B1.1(a) and that a remand for further findings and resentencing is necessary. First, the court took into account impermissible factors. The number of clients defrauded and the sophistication of the scheme are the subject of Section 2F1.1 and not permissibly considered under Section 3B1.1(a), for reasons discussed above. These matters should, therefore, have been excluded from consideration. Moreover, while the duration of a scheme may have some relevance to its extensiveness, it also substantially overlaps with the amount of loss under Section 2F1.1(b)(1). On remand, the district court should, therefore, reexamine the weight to be given to the duration of the scheme.

With regard to the quantity and quality of the services of unknowing participants in Carrozzella's scheme, we must remand for reconsideration by the district court, in light of the present discussion. It is not disputed that the one-time law partner, Richardson, was a knowing participant. The bookkeeper was described by the presentence report as an innocent participant but by the prosecutor as having been granted immunity because she was knowingly involved. The role of other firm employees, the accounting firm, and the stockbrokers is even more obscure. Finally, we have no idea of the roles played by the partners in the real estate ventures.

We therefore remand to the district court for more specific findings in accord with this opinion.

Finally, Carrozzella challenges the calculation of the loss as being over $10,000,000, resulting in a 15-level increase, rather than a 14-level increase, in his offense level under Guidelines § 2F1.1(b)(1). Carrozzella argues that the loss should be calculated as no greater than the "ending balance amount," i.e., $8,591,109.60. He derives that amount by subtracting from the total known deposits--$13,042,249.35 ($8,001,125.34, the opening balance, plus $5,041,124.01, later deposits)--repayments in the amount of $4,451,139.75. We disagree.

We have held that loss in fraud cases includes the amount of property taken, even if all or part has been returned. See United States v. Mucciante, 21 F.3d 1228, 1238 (2d Cir.), cert. denied, 513 U.S. 949, 115 S. Ct. 361, 130 L. Ed. 2d 315 (1994); United States v. Arjoon, 964 F.2d 167, 172 (2d Cir.1992); United States v. Brach, 942 F.2d 141, 143 (2d Cir.1991); see also United States v. Shaw, 3 F.3d 311, 313 (9th Cir.1993); United States v. Rayborn, 957 F.2d 841, 844 (11th Cir.1992). But see, e.g., United States v. Bailey, 975 F.2d 1028, 1031 (4th Cir.1992) (loss is amount of "actual" harm to victim); United States v. Kopp, 951 F.2d 521, 535-36 (3d Cir.1991) (same). One reason for this rule is that, as in Carrozzella's case, the return of money as interest or other income is often necessary for the scheme to continue. A finding of loss in excess of $13 million is, therefore, not clearly erroneous. Moreover, the district court also found--and Carrozzella is silent about this finding--that $1.7 million entrusted to Carrozzella for estates, trusts, or conservatorships was lost. Even if the amount returned should have been subtracted from the $13 million, therefore, the loss would still exceed $10 million. In fact a finding of a loss in excess of $14.7 million would not have been clearly erroneous.

For the foregoing reasons, we reverse the two-level adjustment for violation of the probate process under Guidelines § 2F1.1(b)(3)(B). We remand for further findings the four-level enhancement for organization and leadership of an "otherwise extensive" criminal activity under Guidelines § 3B1.1(a). After making such findings the district court will decide whether to apply the four-level enhancement of Guidelines § 3B1.1(a) or the two-level enhancement of Guidelines § 3B1.1(c). We affirm the 15level upward adjustment under Guidelines § 2F1.1(b) for a loss of over $10,000,000.