John Cappas ran a cocaine ring that operated primarily in Chicago's southwest suburbs. He and several of his associates were charged in a 49-count indictment. Nineteen defendants pled guilty, though Cappas and a few others went to trial. Cappas was convicted on 24 of the 27 counts with which he was charged, including one count of conspiring to possess and distribute cocaine in violation of 21 U.S.C. Sec. 846, one count of running a continuing criminal enterprise (CCE), 21 U.S.C. Sec. 848, and three counts of using a gun in connection with a drug offense, 18 U.S.C. Sec. 924(c). He was sentenced to 45 years in prison: 30 year concurrent terms on the CCE and conspiracy counts (which were to run concurrent with the sentences on each of the remaining counts other than the three "use of a firearm" counts), and--as Sec. 924(c) then required--consecutive five-year terms on each of the three firearm counts.

On appeal, while we otherwise affirmed Cappas' conviction and sentence, we found that the district court may have considered Cappas' guilt on the conspiracy charge in sentencing him on the CCE count. United States v. Alvarez, 860 F.2d 801, 830-31 (7th Cir.1988), cert. denied, 490 U.S. 1051, 109 S. Ct. 1966, 104 L. Ed. 2d 434 (1989), recognized that under Jeffers v. United States, 432 U.S. 137, 97 S. Ct. 2207, 53 L. Ed. 2d 168 (1977), a defendant cannot be punished separately for both CCE and conspiracy. Hence we vacated Cappas' sentence, and remanded the matter to the district court for re-sentencing. United States v. Bafia, 949 F.2d 1465, 1472-75 (7th Cir.1991), cert. denied, --- U.S. ----, 112 S. Ct. 1989, 118 L. Ed. 2d 586 (1992). See generally United States v. Herrera-Rivera, 25 F.3d 491 (7th Cir.1994) (following Bafia ).

On re-sentencing, Cappas pointed out another problem with the government's case. Three of the counts on which Cappas was convicted, counts 12, 28 and 29, were for using or carrying a firearm "during and in relation to" a "crime of violence or drug trafficking crime," 18 U.S.C. Sec. 924(c). Cappas argued that section 924(c) should be interpreted to require multiple predicate offenses before multiple gun enhancements may be imposed. On this theory, a defendant could not be convicted of multiple Sec. 924(c) counts for using multiple guns in a single drug trafficking offense. Rather, each Sec. 924(c) count needs to be tied to a different "crime of violence or drug trafficking crime." Eight courts of appeals have faced this question, seven of them coming down on Cappas' side. See United States v. Lindsay, 985 F.2d 666, 674 (2d Cir.), cert. denied, --- U.S. ----, 114 S. Ct. 103, 126 L. Ed. 2d 70 (1993); United States v. Luskin, 926 F.2d 372, 376-77 (4th Cir.), cert. denied, --- U.S. ----, 112 S. Ct. 68, 116 L. Ed. 2d 43 (1991); United States v. Privette, 947 F.2d 1259, 1262-63 (5th Cir.1991), cert. denied, --- U.S. ----, 112 S. Ct. 1279, 117 L. Ed. 2d 505 (1992); United States v. Pineda-Ortuno, 952 F.2d 98, 104-105 (5th Cir.), cert. denied, --- U.S. ----, 112 S. Ct. 1990, 118 L. Ed. 2d 587 (1992); United States v. Henry, 878 F.2d 937, 942-45 (6th Cir.1989); United States v. Nabors, 901 F.2d 1351, 1357-58 (6th Cir.), cert. denied, 498 U.S. 871, 111 S. Ct. 192, 112 L. Ed. 2d 154 (1990); United States v. Clark, 928 F.2d 733, 737-38 (6th Cir.), cert. denied, --- U.S. ----, 112 S. Ct. 144, 116 L. Ed. 2d 110 (1991); United States v. Sims, 975 F.2d 1225, 1233-37 (6th Cir.1992), cert. denied, --- U.S. ----, 113 S. Ct. 1315, 122 L. Ed. 2d 702 (1993); United States v. Taylor, 13 F.3d 986, 992-94 (6th Cir.1994); United States v. Fontanilla, 849 F.2d 1257, 1258-59 (9th Cir.1988); United States v. Smith, 924 F.2d 889, 894-95 (9th Cir.1991); United States v. Chalan, 812 F.2d 1302, 1315-17 (10th Cir.1987); United States v. Henning, 906 F.2d 1392, 1398-99 (10th Cir.1990), cert. denied, 498 U.S. 1069, 111 S. Ct. 789, 112 L. Ed. 2d 852 (1991); United States v. Ross, 920 F.2d 1530, 1538-39 (10th Cir.1990); United States v. Rogers, 921 F.2d 1089, 1092-93 (10th Cir.1990), cert. denied, 501 U.S. 1211, 111 S. Ct. 2812, 115 L. Ed. 2d 985 (1991); United States v. Moore, 958 F.2d 310, 314 (10th Cir.1992); United States v. Johnson, 977 F.2d 1360, 1376-77 (10th Cir.1992), cert. denied, --- U.S. ----, 113 S. Ct. 1024, 122 L. Ed. 2d 170 (1993); United States v. Parra, 2 F.3d 1058, 1070-71 (10th Cir.), cert. denied, --- U.S. ----, 114 S. Ct. 639, 126 L. Ed. 2d 597 (1993); United States v. Hamilton, 953 F.2d 1344, 1345-46 (11th Cir.), cert. denied, --- U.S. ----, 113 S. Ct. 240, 121 L. Ed. 2d 174 (1992). See also United States v. Casey, 776 F. Supp. 272, 275-78 (E.D.Va.1991), aff'd, 1992 WL 203955, 1992 U.S. App.Lexis 20232 (4th Cir.1992). We agree with our seven sister circuits (and the government does not here argue to the contrary) that have held that the use of multiple guns in a single drug conspiracy will not support multiple convictions under Sec. 924(c).1

In light of this reasoning, Cappas pointed out to the district court that the indictments in counts 28 and 29 both alleged that guns were used in relation to the same predicate offense: conspiracy to possess and distribute narcotics as charged in count 2. And the third gun charge, count 12, alleged that Cappas used a gun in relation to two predicate acts, collection of a debt by extortionate means (as charged in count 11) and the general conspiracy charged in count 2. Reasoning that the jury might have convicted him three times of using a gun in connection with the same drug offense, Cappas moved the district court, pursuant to 28 U.S.C. Sec. 2255, to vacate his convictions on two of the three counts.

The government responded before the district court by advancing three arguments. First, it contended that because Cappas did not make this argument earlier, it could not be raised for the first time in a Sec. 2255 motion. Second, it argued that a single drug trafficking offense could support multiple section 924(c) convictions. And third, it insisted that the Sec. 924(c) conviction under count 12 was supportable by a separate crime of violence--the extortion charged in count 11--such that, even if the court did not accept either of its first two arguments, two of the three Sec. 924(c) counts could survive.

The district court rejected each of these arguments, and found that Cappas could be sentenced on only one of the three Sec. 924(c) counts. Accordingly, following our instructions on remand (regarding sentencing on CCE and conspiracy), the court sentenced Cappas to 19 years: concurrent 14-year terms on the conspiracy and CCE counts (with equal or lesser concurrent sentences on the other counts, other than the gun charge), and a single consecutive five-year term on count 12. The court dismissed counts 28 and 29. The government here appeals.

* On appeal, the government has abandoned the first two arguments it presented before the district court. It no longer contends either that Cappas has procedurally defaulted on the argument he makes here (the district court having recognized that a defendant may advance--in a resentencing following the vacation of a previous sentence--any arguments that could have been brought the first time, see United States v. Atkinson, 979 F.2d 1219, 1223 (7th Cir.1992)), or that multiple Sec. 924(c) convictions for the use of multiple guns in connection with the same conspiracy are permissible.

Rather, it argues only that the jury in this case rested its conviction on count 12 on the use of a gun in connection with the extortion charged in count 11, not on the general conspiracy charged in count 2. Therefore, the government contends, count 12, as well as one of the two charges (counts 28 or 29) for using a gun in connection with the general conspiracy charged in count 2 may stand.

We agree with this general approach to the statute. While a defendant cannot be convicted twice under Sec. 924(c) for using two guns in connection with the same drug trafficking or violent offense, separate convictions are permissible so long as the court's instructions require the jury to connect each gun use to a separate predicate offense. See Henry, 878 F.2d at 945; Johnson, 977 F.2d at 1376-77; Chalan, 812 F.2d at 1315-17; Privette, 947 F.2d at 1262-63; Lindsay, 985 F.2d at 676-77. And by "separate offense," we mean no more than that the two cannot be the same offense for double jeopardy purposes. Therefore, if the jury finds that a defendant used one gun in connection with a narcotics distribution count, and another gun in connection with a general conspiracy (of which that distribution was a part), he may be convicted on two Sec. 924(c) charges. E.g. Pinkerton v. United States, 328 U.S. 640, 643-44, 66 S. Ct. 1180, 1181-82, 90 L. Ed. 1489 (1946) (commission of the substantive offense and a conspiracy to commit it are separate and distinct offenses).

The government thus argues that the gun use charged in count 12 is linked to a separate crime of violence--the extortion charged in count 11. Had the jury been required by its instructions to make this connection, this would be an easy case. But the jury did not receive any such instruction, so the government is left to argue that we can infer that the jury made this connection by examining the indictment and the evidence introduced at trial.

The indictment in count 12 charged that on March 3, 1987, Cappas used a particular gun--a Dan Wesson .357 revolver bearing serial number 95605--"during and in relation to" an attempt to collect on a debt by extortionate means and in relation to the broad drug conspiracy. Further, the government argues that the evidence introduced at trial made clear that Cappas shot up James Renda's house in an attempt to collect on Renda's cocaine debt. Since the jury convicted Cappas on both the extortion count and the gun count, the government figures, it must have concluded that he used the gun in connection with the extortion.

Cappas responds by pointing to the jury instructions, which told the jury to return a guilty verdict on count 12 so long as it found that Cappas used a gun during and in relation to "the offense charged in either Count [2 (the general conspiracy) ] or Count [11 (the extortionate collection of debt) ]" (emphasis added). Cappas therefore argues that the jury might have, following these instructions, convicted him on count 12 because he used the gun in connection with the overall drug conspiracy in count 2, without relying on the extortion charged in count 11. Because count 2 is the same predicate drug offense charged in counts 28 and 29, Cappas contends that he may well have been convicted three times for using three firearms in connection with the same predicate offense--the drug conspiracy.

If this is the case, the double jeopardy clause would require the district court to have taken the action that it did--dismiss counts 28 and 29. Note that this is not to say that it would necessarily violate double jeopardy for Congress to impose multiple punishment for the use of multiple guns in connection with a single drug offense. If Sec. 924 provided that each use of a gun were separately punishable, we do not think such a statute would run into constitutional difficulty; the use of separate guns would represent separate elements, so double jeopardy would not be violated. See generally Blockburger v. United States, 284 U.S. 299, 52 S. Ct. 180, 76 L. Ed. 306 (1932). But because we have concluded, like seven other courts of appeals, that Congress intended the use of several guns in connection with a single drug offense to amount to only a single violation of Sec. 924(c), Cappas' conviction on multiple Sec. 924(c) counts for using several guns in connection with the same drug offense would therefore of course violate double jeopardy. See United States v. Felix, --- U.S. ----, ----, 112 S. Ct. 1377, 1382, 118 L. Ed. 2d 25 ("At its root, the Double Jeopardy Clause forbids the duplicative prosecution of a defendant for the 'same offence.' ").

This appeal therefore turns on the government's contention that the jury convicted on count 12 because it found that Cappas used the gun in connection with the extortion charged in count 11--not the general conspiracy charged in count 2.

But we must first confront how we should even go about answering this question. First--having no evidence at all of what the jury actually thought--we must decide how to reconstruct its deliberations. And second, because we are necessarily working off of a reconstruction, we need to decide how certain we must be that the jury did not link the gun charge in count 12 to the general conspiracy in count 2 before we allow another gun charge (that clearly is linked to the conspiracy in count 2) to stand. The answers to these questions--which do not seem to us to be obvious--are not discussed at much length in the various cases we have found addressing our general problem.

* The first problem is deciding how to figure out what the jury was thinking. One logical starting point in this endeavor is the analogous harmless error doctrine, where the court engages in a similar analysis. In a harmless error case, an appeals court knows that a constitutional error has been committed (for example, illegally obtained evidence has been erroneously admitted) and the court is asked to determine whether the error was "prejudicial" (which is to say, whether the jury relied on this evidence in forming its verdict). Our situation is similar, the only difference being that we mine the jury's verdict not to determine what effect an acknowledged error had on the outcome, but rather to determine whether an error was committed at all.

And just as in the harmless error context, there are two obstacles in our way. First, we have no actual evidence of what the jurors were thinking. "How can anyone determine what went on in the mind of another or of twelve others who served as triers of fact? The only source of direct evidence would be their testimony. If the facts had been tried by a juror, such testimony would be precluded by the rule forbidding affidavits or evidence of any sort that tends to contradict, impeach, or defeat the jury's verdict. Thus, there is no possibility of tapping the only source of direct evidence on the effect, if any, of an error upon the jury's verdict." Roger Traynor, The Riddle of Harmless Error 23 (1970).

And second, even if we had such evidence with regard to particular jurors, it isn't clear how useful it would be. Even where individual jurors have full comprehension of what they are asked to do (and in Gacy v. Welborn, 994 F.2d 305 (7th Cir.), cert. denied, --- U.S. ----, 114 S. Ct. 269, 126 L. Ed. 2d 220 (1993), and Free v. Peters, 12 F.3d 700 (7th Cir.1993), we rejected an empirical study that purported to show that they often do not), juries--like legislatures--are collective bodies. And when collective bodies are required to speak with one voice (in federal courts a unanimous voice), compromises and logrolling are perhaps inevitable. See generally Frank H. Easterbrook, Statutes' Domains, 50 U.Chi.L.Rev. 533, 547 (1983) (legislatures); Harry Kalven and Hans Zeisel, The American Jury (1966) (juries). The use of such devices renders problematic our efforts to divine how the body would have answered a question that it did not, in fact, address. "Was the gun charge in count 12 tied to the extortion in count 11?," just like "would you have convicted had the defendant's confession been excluded?," is an example of such an unanswerable question.

The harmless error doctrine avoids attempting the impossible by attributing to the jury the attributes of a single finder of fact behaving rationally. The process therefore requires us to hypothesize, rather than to determine, what the jury was thinking. And an hypothesis--by definition--is uncertain. We cannot say with mathematical certainty that an error had no prejudicial effect, but instead--looking at the evidence the (presumably rational) jury had before it, and the (presumably obeyed) instructions they were given on how to evaluate it--rely on formulations like the error needs to be harmless "beyond a reasonable doubt," Chapman v. California, 386 U.S. 18, 87 S. Ct. 824, 17 L. Ed. 2d 705 (1967) (constitutional error on direct appeal), or the error cannot have had a "substantial and injurious effect" on the jury's verdict, Kotteakos v. United States, 328 U.S. 750, 66 S. Ct. 1239, 90 L. Ed. 1557 (1946) (all other errors--including, after Brecht v. Abrahamson, --- U.S. ----, 113 S. Ct. 1710, 123 L. Ed. 2d 353 (1993), constitutional error on habeas review).

This leads us to our second question, what standard to employ. In the harmless error context, Chapman v. California, 386 U.S. 18, 87 S. Ct. 824, tells us that where we examine constitutional error on direct appeal, a court of appeals is to affirm the conviction so long as the government proves--beyond a reasonable doubt--that the error did not affect the verdict.

Our first question is therefore to ask whether the Court's holding in Chapman extends to this case. There is at least one theory of the harmless error doctrine under which it might. That is, it is sometimes argued that the harmless error doctrine is best seen as a gloss on constitutional rights themselves. On this theory, a defendant's right not to have a coerced confession introduced against her is to be understood as the right not to be convicted if an appeals court cannot determine--beyond a reasonable doubt--that a coerced confession did not affect the jury's decision. See Phillip J. Mause, Harmless Constitutional Error: The Implications of Chapman v. California, 53 Minn.L.Rev. 519, 532 (1969). If this is the correct understanding of the harmless error doctrine, one might think that its conception of the underlying constitutional rights would extend to our circumstance as well. We would therefore understand the double jeopardy guaranty essentially to require the government to prove beyond a reasonable doubt that the jury did not convict for the same offense twice.

But as we have earlier commented, the Supreme Court's decision in Brecht v. Abrahamson, --- U.S. ----, 113 S. Ct. 1710, 123 L. Ed. 2d 353, which abolishes the Chapman standard in habeas cases, appears to spell the end of this theory. See Sokol Crystal Products v. DSC Communications Corp., 15 F.3d 1427, 1435 n. 5 (7th Cir.1994). After Brecht, it is no longer plausible to argue that the harmless error doctrine provides a gloss on the underlying constitutional rights, or is even itself a rule derived from the constitution. Instead, the emerging view appears to be that the Chapman standard is a pronouncement of "constitutional common law," see Henry P. Monaghan, Harmless Error and the Valid Rule Requirement, 1989 S.Ct.Rev. 195, 200 n. 30. The "beyond a reasonable doubt" standard is not "real" constitutional law, but--like the Bivens remedy and arguably the exclusionary rule--a prophylactic remedial standard that fills a vacuum created by constitutional and legislative silence. See Daniel J. Meltzer, Harmless Error and Constitutional Common Law, 61 U.Chi.L.Rev. 1 (1994). So understood, the standard does not define the scope of underlying substantive constitutional rights (such as the double jeopardy clause), but is rather about what remedy is required where it is agreed that constitutional rights have been violated.

We therefore do not believe that Chapman and the harmless constitutional error rule govern the disposition of this case, which is not about the remedy for a constitutional violation but instead about whether a violation occurred in the first instance. We must instead look to more general principles governing convictions that may rest on impermissible grounds. To that end, Stromberg v. California, 283 U.S. 359, 367-68, 51 S. Ct. 532, 535, 75 L. Ed. 1117 (1931), and Yates v. United States, 354 U.S. 298, 77 S. Ct. 1064, 1 L. Ed. 2d 1356 (1957), both hold that a verdict must be set aside where a verdict is supportable on one ground, but not on another, and "it is impossible to tell which ground the jury selected." Id. at 312, 77 S.Ct. at 1073. Particularly compelling is the Supreme Court's decision in Griffin v. United States, 502 U.S. 46, 112 S. Ct. 466, 116 L. Ed. 2d 371 (1991), which seems to suggest that this rule--unlike the harmless error rule--is a direct interpretation of the underlying constitutional guarantee. "[W]here a provision of the Constitution forbids conviction on a particular ground, the constitutional guarantee is violated by a general verdict that may have rested on that ground." Id. 502 U.S. at --, 112 S. Ct. at 471. And it appears to us that the Court's language--a conviction cannot stand where it "may have rested" on an impermissible ground--requires the dismissal of one of the counts whenever a jury, following its instructions, could reasonably have convicted a defendant twice for committing a single offense.2

That leaves us with the task of applying this standard to Cappas' case.3 Count 12 charged that Cappas used a gun in connection with two predicate offenses, the extortion in count 11 and the conspiracy in count 2. Our role is essentially to figure out which of those two predicate offenses the jury was thinking about when it returned a guilty verdict on count 12. This requires us to examine the facts surrounding the events in question in somewhat greater detail.

Cappas was apparently "fronting" the drugs to Renda. Cappas would supply the drugs to Renda, who would resell them, and return some portion of the proceeds to Cappas. The government's theory is that Renda fell behind on his payments, and that Cappas responded by sending Ray Bonemma (a co-conspirator, who was accompanied by Brian Baffia, another co-conspirator) to shoot up Renda's house.

Cappas tells another story, denying any involvement in the shooting, and insisting that Bonemma--who had apparently also sold drugs to Renda--was acting on his own behalf, shooting up Renda's house to collect on a debt that Cappas owed him. Cappas relies on Baffia's testimony, according to which Bonemma was acting of his own accord--neither Baffia nor Cappas had any advance knowledge of this shooting. Bonemma, on the other hand, testified (consistent with the government's theory) that he and Baffia were both following Cappas' orders.

The point of Cappas' theory is to suggest that the jury may never have found that he was at all involved in the shooting at Renda's house. Instead, on his theory, he could have been convicted on count 11 (the extortion count) simply for making threatening calls to Renda. And further, he says that he could have been convicted on count 12 based on the evidence that he had a virtual arsenal of guns that were used in connection with the drug conspiracy.

The government responds to all of this by pointing out that Cappas' theory relies on the possibility that the jury might have accepted Baffia's testimony that Bonemma was acting alone in shooting up Renda's house. But, the government contends, since the jury also convicted Baffia on that same count, it must have rejected Baffia's testimony.

A possible response to this might be to search for alternative bases in the record that might have supported convicting Baffia of extortion. Or, alternatively, to argue that the jury might have believed that part of Baffia's testimony in which he exculpated Cappas, but not that part in which he exculpated himself. But it rather seems to us that the entire endeavor of picking apart the record and undertaking to discern what parts of whose testimony the jury was likely to have believed is fraught with peril. Instead, it seems to us that the more sensible approach is to focus on the jury instructions.4

On this view, the question is whether the jury--following its instructions--might have convicted Cappas on Count 12 because he used a gun in relation to the general conspiracy (charged in count 2), rather than the act of extortion charged in count 11. We of course assume that jurors understand and follow their instructions. Delli Paoli v. United States, 352 U.S. 232, 242, 77 S. Ct. 294, 300, 1 L. Ed. 2d 278 (1957); Bruton v. United States, 391 U.S. 123, 135-36, 88 S. Ct. 1620, 1627-28, 20 L. Ed. 2d 476 (1968); Richardson v. Marsh, 481 U.S. 200, 211, 107 S. Ct. 1702, 1709, 95 L. Ed. 2d 176 (1987). On the other hand, the testimony of witnesses and arguments advanced by counsel may well be ignored by the jury. Taylor v. Kentucky, 436 U.S. 478, 488-89, 98 S. Ct. 1930, 1936-37, 56 L. Ed. 2d 468 (1978). "Amid a sea of facts and inferences, instructions are the jury's only compass." United States v. Walters, 913 F.2d 388 (7th Cir.1990).

Here, as we have noted, the jury was instructed to convict on count 12 if it found that Cappas "used or caused the use of a firearm on or about March 3rd, 1987, during and in relation to the offense charged in either Count II or Count XI." Because the shooting of Renda's house allegedly took place on March 3, 1987, it seems likely that the jury had this act of extortion in mind when it convicted on count 12. Perhaps, if the evidence firmly excluded the possibility that the jury relied on count 2, it might be permissible to allow two of the Sec. 924(c) convictions to stand despite the instruction's wording. But that question is not before us today. Instead, ascertaining what the jury was thinking in this case turns on what inferences one draws from the conflicting evidence. The government all but admits this in asking us (in vain) to read the evidence in its favor. See supra n. 4. Therefore, because the instructions would have permitted the jury to convict Cappas three times for committing what Congress intended to be one offense, and because the record does not firmly exclude the possibility that the jury did exactly that, the decision of the district court, dismissing counts 28 and 29, is AFFIRMED.