Barry Rand Elden, Chief of Appeals (argued), Office of the United States Attorney, Criminal Appellate Division, Chicago, IL, for U.S. in Nos. 92-2794, 92-2797, 92-2799 and 92-2849.

Theodore T. Poulos, Barry Rand Elden, Chief of Appeals (argued), Office of the United States Attorney, Criminal Appellate Division, Chicago, IL, for U.S. in Nos. 92-2796 and 92-2798.

William R. Hogan, Office of the United States Attorney, Criminal Division, Barry Rand Elden, Chief of Appeals (argued), Office of the United States Attorney, Criminal Appellate Division, Chicago, IL, for U.S. in No. 94-1517.

Jerry B. Kurz (argued), Hall & Kurz, Chicago, IL, for Edward S. Williams.

Carl P. Clavelli (argued), Chicago, IL, for Jerome A. Crowder.

Robert G. Clarke (argued), Chicago, IL, for Thomas L. Bates.

Donald Paull, Robert G. Clarke (argued), Chicago, IL, for Louis Hoover.

Richard C. Leng, Leng, Stowell, Friedman & Vernon, Kent R. Carlson (argued), Chicago, IL, for Bernard Green.

Keith Spielfogel, Robert G. Clarke (argued), Chicago, IL, for Roland Lewis.

Before POSNER, Chief Judge, and MANION and DIANE P. WOOD, Circuit Judges.

POSNER, Chief Judge.

The El Rukns were a street gang active on Chicago's south side for two decades ending in the late 1980s. The gang's main activity was the sale of illegal drugs. Its members committed many murders, and engaged in much other violence, in the turf wars that are endemic to the trade in illegal drugs. The trial from which these appeals come to us was the last of five trials in the Northern District of Illinois of different leaders of the El Rukns. The defendants in this trial as in the other ones were convicted and sentenced to long prison terms. But the convictions in all four of the previous trials were later set aside by the trial judges when it became known that the government had knowingly employed perjured testimony in the trials and concealed from the defense the favors that it had showered on the former members of the gang who were the government's key witnesses. Only one of the grants of a new trial was appealed and we affirmed it in United States v. Boyd, 55 F.3d 239 (7th Cir.1995). In the present case, the fifth as we have said, the district judge refused to grant a new trial even though he had before him the identical evidence of the government's misconduct. In addition to appealing the denial of their motion, the defendants appeal a number of rulings made during the trial itself, but we begin with the denial of the motion for a new trial as it is the issue that is pressed hardest.

With four sets of defendants, all accused of participation in the same conspiracy, having been granted new trials because of identical prosecutorial misconduct, denial of the same relief to this fifth set of defendants strikes a discordant note. But the potential for such incongruity is present whenever persons charged with a joint offense are tried separately, as they usually very much desire to be--or for that matter whenever they are tried together, since it is always possible for a jury to exercise lenity and acquit some of the defendants while convicting others who are in fact no more guilty, and when this happens the convicted defendants have no remedy. United States v. Powell, 469 U.S. 57, 105 S. Ct. 471, 83 L. Ed. 2d 461 (1984); United States v. Dotterweich, 320 U.S. 277, 279, 64 S. Ct. 134, 135-36, 88 L. Ed. 48 (1943). Such incongruities are built into the American system of criminal justice and can have no weight in our decision whether to reverse the denial of a new trial to the present defendants.

Another point that is difficult for nonlawyers to understand or accept is that because the question whether to grant a new trial is committed to the discretion of the district judge, as the defendants rightly concede, United States v. Knox, 68 F.3d 990, 1000 (7th Cir.1995); United States v. Maloney, 71 F.3d 645, 654 (7th Cir.1995), it is possible for two judges, confronted with the identical record, to come to opposite conclusions and for the appellate court to affirm both. That possibility is implicit in the concept of a discretionary judgment. Rice v. Nova Biomedical Corp., 38 F.3d 909, 918 (7th Cir.1994). If the judge could decide only one way he would not be able lawfully to exercise discretion; either he would be following a rule, or the circumstances would be so one-sided that deciding the other way would be an abuse of discretion. If the judge can decide either way because he is within the zone in which he has discretion--can decide either for or against the grant of a new trial--this implies that two judges faced with the identical record could come to opposite conclusions yet both be affirmed.

When we affirmed Judge (now Chief Judge) Aspen's grant of a new trial to the defendants in the Boyd case, we went out of our way to make clear that we were affirming not because we thought he necessarily was right but because we thought he was reasonable, that he had not "abused his discretion." United States v. Boyd, supra, 55 F.3d at 245. Because we found no abuse of discretion in his having granted a new trial we had no occasion to decide whether we would also have affirmed him had he denied a new trial or whether, on the contrary, it was one of those one-sided cases where only one ruling is possible. So the fact that Judge Mills on a record very similar, though as we are about to see not identical, to that before Judge Aspen made the opposite ruling does not necessarily require, as a matter of maintaining consistency with our decision in Boyd, that we reverse Judge Mills.

Rather than speculate on what we would have done had Judge Aspen decided the other way, let us see whether Judge Mills can be said to have been acting unreasonably when he held that the defendants in this case had not been denied a fair trial by the totality of the government's misconduct. The misconduct arose out of the abnormal, and deeply questionable, generosity and solicitude that the government displayed toward its key witnesses, former El Rukns confined in the Metropolitan Correctional Center, the federal jail in Chicago. The government knew that some of these witnesses were lying when they testified that they had stopped using drugs after being arrested in 1988. Partly because the government, as one dimension of its friendly treatment of these witnesses, allowed "contact visits" with members of their families, the witnesses obtained drugs while in jail--even dealt drugs--as the government well knew. Knew, but made no effort to correct the witnesses' lying denials to the jury. And members of the prosecutorial team gave presents to the witnesses, allowed them the free use of telephones to make long-distance calls for themselves and their friends, allowed conjugal visits in the prosecutors' offices, and even threw parties for the witnesses. None of these favors was disclosed to the defense, even though they could have been used to impeach the witnesses' credibility. The government's failure to disclose these things was a violation of the rule of Brady v. Maryland, 373 U.S. 83, 83 S. Ct. 1194, 10 L. Ed. 2d 215 (1963); see Giglio v. United States, 405 U.S. 150, 154, 92 S. Ct. 763, 766, 31 L. Ed. 2d 104 (1972); United States v. Bagley, 473 U.S. 667, 676, 105 S. Ct. 3375, 3380, 87 L. Ed. 2d 481 (1985); Kyles v. Whitley, --- U.S. ----, ----, 115 S. Ct. 1555, 1565, 131 L. Ed. 2d 490 (1995), while its knowing use of false testimony to convict was, of course, also improper. Napue v. Illinois, 360 U.S. 264, 269, 79 S. Ct. 1173, 1177, 3 L. Ed. 2d 1217 (1959); United States v. Agurs, 427 U.S. 97, 103, 96 S. Ct. 2392, 2397, 49 L. Ed. 2d 342 (1976).

But these improprieties justify granting a new trial only if there is a "reasonable likelihood" that they "affected the judgment of the jury," Giglio v. United States, supra, 405 U.S. at 154, 92 S. Ct. at 766, or, in other words, only if they "undermine[d] confidence in the verdict." Kyles v. Whitley, supra, --- U.S. at ----, 115 S. Ct. at 1566; see also United States v. Agurs, supra, 427 U.S at 103, 96 S. Ct. at 2397; United States v. Bagley, supra, 473 U.S. at 678, 105 S. Ct. at 3381-82; United States v. Silva, 71 F.3d 667, 670 (7th Cir.1995); United States v. Boyd, supra, 55 F.3d at 243. The defendants argue that had the jury known that the government had lavished benefits so many and so irregular on the witnesses, it would have refused to believe their testimony--testimony essential to the conviction of the defendants. It would have thought that the government had bought their testimony, or alternatively that the government had doubted their truthfulness and was desperate therefore to keep them in an accommodating mood so that they would testify as the government wanted. If the defendants' argument is correct, confidence that the trial produced a correct verdict would indeed be undermined.

However, the standard that we have articulated requires first the district judge, and then this court in reviewing his ruling for possible abuse of discretion, to examine the trial record as a whole, considering not only the parts of the record infected by the government's improprieties but also the untainted evidence that the jury heard. Although it is not enough that the untainted evidence be sufficient for conviction, Kyles v. Whitley, supra, --- U.S. at ---- n. 8, 115 S. Ct. at 1566 n. 8, that evidence must not be ignored in the making of the determination whether there is a reasonable likelihood that the outcome would have been different had the government not misbehaved.

When we do what the cases require--take all the evidence into account, and not only the tainted parts--the blandishments showered on the witnesses and not disclosed to the defendants pale in comparison with the incentives that the witnesses had wholly apart from those blandishments, and that the jury knew they had, to lie to save their skins. All the witnesses were under heavy sentence--one was on death row--or threat of heavy sentence. Their best hope for mercy was that their testimony would convict the defendants and so earn them the government's gratitude and assistance. The witnesses' plea agreements, which conditioned punishment discounts (or promises not to prosecute further--and in one case not to seek the death penalty) on their cooperation in the prosecution of the El Rukns defendants, were before the jury along with evidence that the government had given witness Evans money for the support of his family. Knowing all this, the jury probably discounted the witnesses' testimony heavily. The defendants had information about money given by the government to other prisoner witnesses and could have used the information to undermine their testimony further. They chose not to do so, apparently believing that additional impeachment materials, when so much was already before the jury, would not help.

We do not think that the district judge can be said to have abused his discretion when he concluded that the withholding of evidence about the favors bestowed on the government witnesses had not resulted in a verdict unworthy of confidence. The jury had a number of reasons for convicting the defendants notwithstanding the characters and incentives of the key witnesses. Some of these reasons were common to the other El Rukn trials, but some were unique to this one.

First, the testimony of the favored prisoner witnesses was corroborated by the testimony of law enforcement personnel, and they neither were in quest of a reward from the prosecutors nor were recipients of largesse from the prosecutors.

Second, the offenses of which these defendants were convicted were drug offenses. The evidence of the defendants' violent acts was introduced mainly in order to show how the gang operated. Even if that evidence had been disbelieved, a rational jury would have had to convict the defendants of participation in a far-flung, protracted conspiracy to sell illegal drugs. The violent acts did not assume independent significance until the sentencing phase, and it is not contended that Judge Mills would have given the defendants shorter sentences had he known about the government's shenanigans. The evidence of the defendants' drug dealing, as distinct from the evidence (which came mainly from the tainted witnesses) of the violent acts committed by the gang, was overwhelming. Much of it came from law enforcement officers who had conducted controlled drug buys and raids on buildings from which the defendants sold drugs.

In saying that the violent acts were not essential to conviction, we do not mean that they were irrelevant to the charges against the defendants; if they had been, the evidence about them would have been vulnerable to challenge as excessively prejudicial in relation to the value of the evidence as proof of the defendants' guilt. Fed.R.Evid. 403. Any act in furtherance of a conspiracy is part of the criminal conduct encompassed by the charge of conspiracy, including killing rivals in order to maximize the conspiracy's profits. Recognizing this point is different from saying that particular such acts are indispensable evidence of the conspiracy.

Third, one of the defendants, Crowder, made the mistake of testifying. In the course of a futile effort to convince the jury that he had believed the El Rukns to be a purely religious organization, he fingered each of the other defendants, and himself, as "generals" in the El Rukns organization. The question whether particular German and Japanese generals in World War II knew about and could properly be held criminally responsible for the violations of the laws of war committed by soldiers under their command is still being debated. The El Rukns are a formidable organization by American urban gang standards, but they are not of military scale and the likelihood that a "general" in the organization would not be sufficiently involved in their drug dealing to be culpable is slight.

Crowder did something else in his testimony to damage his side. He corroborated witness Harris's translation of tapes of the defendants' discussing their illegal activities. The tapes were important evidence and on them the defendants spoke in a code to thwart electronic surveillance. Harris, one of the tainted witnesses, was the sole translator, so that if the jury disbelieved him the prosecution would be in trouble. Crowder corroborated Harris's accuracy as translator by translating some of the key words in the code identically and by agreeing with Harris's translations of several passages.

Crowder's testimony probably doomed any chance that any of the defendants would be acquitted.

Judge Mills had to decide whether, given this record which he had seen created as the trial progressed, there was, in the language of Kyles v. Whitley, supra, --- U.S. at ----, 115 S. Ct. at 1566, a "reasonable probability of a different result," or, in other words, whether the jury might (with some nontrivial degree of probability) have acquitted the defendants had it known what we now know about the government's misconduct. The defendants did not have to show that they would in fact have been acquitted, only that such an outcome would have been more than remotely likely. The judge thought not and we cannot say that he was unreasonable, and thus abused his discretion, in coming to this conclusion. While the evidence of the government's misconduct was identical to that in the previous El Rukns trials--in fact Judge Mills adopted the lengthy summary of it in Judge Aspen's opinion in Boyd; see United States v. Bates, 843 F. Supp. 437 (N.D.Ill.1994)--the evidence of the defendants' guilt was stronger. And while the same former El Rukns testified and gave testimony very similar to what they had given in the other four trials, there was this difference: In the trial before Judge Mills, witness Evans did not testify that he had seen the light when he was arrested. It is true that he and the other prisoner witnesses insinuated in other terms that they had seen the error of their ways after being arrested and could now be trusted--though at the same time some of them freely acknowledged that they would do anything to save themselves, including lying under oath. But no one used that vivid phrase, "seen the light," which Evans had used repeatedly in the trial of the defendants in the Boyd case and which we had singled out in our opinion in Boyd as a circumstance that supported Judge Aspen's ruling. 55 F.3d at 246. The difference is not momentous in itself. But when combined with the other differences that we have discussed it shows that the results in the previous El Rukns cases would not have dictated the result here even if the standard of appellate review were not deferential, deference implying as we have said that Chief Judge Aspen and Judge Mills might reach opposite results on the same record yet both be operating within the outer limits of their lawful discretion.

The principle of deferential review makes a lot of sense in a case such as this. Having watched the jury as they listened to the testimony, having listened to the testimony and the arguments himself, having his finger as it were on the pulse of the trial--a trial that occupied 28 days scattered over four months--the district judge was in a better position than we to weigh the imponderables involved in a judgment of prejudice. Id. at 242. The chief imponderable was whether additional impeachment of the inmate witnesses, by means of evidence showing their continued drug use and drug dealing while jailed awaiting the trial, and the presents and other favors that the prosecution lavished on them, together with the awareness that this evidence would have produced in the minds of jurors that the government had engaged in misconduct, would have persuaded the jury to acquit. If a person under sentence of death can lie his way out of it under penalty of perjury he will do so, and throwing a conjugal visit into the pot is unlikely to make a difference to the jury. Or so at least the district judge might reasonably conclude. But if it would not make a difference, why did the prosecution do these things? Thinking about this question might lead a jury to wonder whether the prosecution doubted the strength of its case, perhaps because it was not so strong a case as it seemed. But the fact that the prosecution was trying to bolster its case does not prove that without that bolstering the case would have been weak. Otherwise the case law would not require, as it does, a separate inquiry into the prejudicial effect of prosecutorial misconduct. Nor is bolstering the only possible motive for such misconduct. Lawyers are cautious people by and large, and even a lawyer who has a very strong case will want to make it stronger.

All this is speculation. But picking one's way through the labyrinth of speculation to the ultimate judgment of whether the prosecutors' misconduct is likely to have changed the outcome of the trial is the task of the district judge in the first instance. If the judge comes to a reasonable conclusion, we are bound. Because the defendants do not argue that had it not been for the government's misconduct they would have been acquitted of particular counts, as opposed to being acquitted of all the charges against them, we have proceeded on the assumption that their arguments apply to all aspects of the case equally. It was not an abuse of discretion to repel this broad attack. We need not decide whether an attack targeted on particular counts, perhaps ones where the evidence of guilt was weaker than it was on other counts, would have fared better. We can therefore move on to the rulings at trial that the defendants challenge.

They argue that the judge should have excluded evidence of the violent acts committed by the El Rukns because the defendants were not proved to have known about them. The argument misconceives the nature of criminal liability for participation in a conspiracy. A member of a conspiracy is liable for the acts of other members, provided those acts are foreseeable as being within the scope and contemplation of the conspiracy, even if he does not know about them. E.g., United States v. Vega, 72 F.3d 507, 512 (7th Cir.1995); United States v. Manzella, 791 F.2d 1263, 1268 (7th Cir.1986). Having agreed to a conspiracy of which those acts were an expectable incident, he cannot by shutting his eyes avoid liability for his contribution to setting in motion the chain of events that led to them. The El Rukns were a conspiracy to sell drugs and do whatever was necessary, whether in the way of intimidating witnesses or murdering rivals, to maximize the profits, and minimize the expense, including expected punishment costs, of their sales activities. As "generals" in the El Rukns these defendants knew the aims of the conspiracy, knew therefore that they included violence where necessary to protect and promote the welfare of the conspiracy, and by knowing these things became responsible for the acts of violence that anyone who knew what they knew would have foreseen as likely. E.g., United States v. Gutierrez, 978 F.2d 1463, 1468 (7th Cir.1992); United States v. Romero, 897 F.2d 47, 51-52 (2d Cir.1990); United States v. Alvarez, 755 F.2d 830, 848-51 (11th Cir.1985). The judge also had an adequate factual basis for rejecting the argument that the defendants did not know about the heroin sold by their "gorillas" (rank and file members of the gang).

We mentioned earlier the role of witness Harris as the translator of the El Rukns code. The district judge designated Harris an "expert witness." The defendants complain that this gave Harris too inflated an image in the eyes of the juror. Not likely. There was no pretense that he was impartial, or a member of a learned profession. Neither condition is required to qualify a person as an expert witness under the current rules of evidence. Fed.R.Evid. 702; Tagatz v. Marquette University, 861 F.2d 1040, 1042 (7th Cir.1988); Wilson v. City of Chicago, 6 F.3d 1233, 1238 (7th Cir.1993). All you need to be an expert witness is a body of specialized knowledge that can be helpful to the jury. United States v. Rollins, 862 F.2d 1282, 1292 (7th Cir.1988); United States v. Penny, 60 F.3d 1257, 1265 (7th Cir.1995). Knowledge of the El Rukns' code was specialized, all right; it was possessed by Harris; it was helpful to the jury. There is not even a paradox in the suggestion that the biggest experts on crime are, often, criminals. See United States v. Johnson, 28 F.3d 1487, 1496-97 (8th Cir.1994), another case in which a criminal was allowed to testify as an expert on the drug trade. This is not a case in which scientific evidence is tendered, and the question is whether it is real science or junk (courtroom) science. Daubert v. Merrell Dow Pharmaceuticals, Inc., 509 U.S. 579, 113 S. Ct. 2786, 125 L. Ed. 2d 469 (1993); Rosen v. Ciba-Geigy Corp., 78 F.3d 316 (7th Cir.1996). You don't have to be a scientist or use the methodology of science, or even be an honest, decent, law-abiding citizen, in order to possess specialized knowledge about a criminal activity.

What is correct, and relates to the directly preceding observation, is that Harris didn't have to be designated an expert witness in order to be permitted to testify to the meaning of the El Rukns' code. The difference between an expert witness and an ordinary witness is that the former is allowed to offer an opinion, while the latter is confined to testifying from personal knowledge. Fed.R.Evid. 701, 702; Richardson v. Consolidated Rail Corp., 17 F.3d 213, 218 (7th Cir.1994); United States v. Fidelity & Deposit Co., 986 F.2d 1110, 1117-18 and n. 8 (7th Cir.1993); Kaczmarek v. Allied Chemical Corp., 836 F.2d 1055, 1060-61 (7th Cir.1987). An economist, for example, is allowed to testify that a particular pattern of pricing indicates that the defendant participated in a price-fixing conspiracy, whereas the lay witness could testify only to what the prices were. A linguist or professional code-breaker might testify to the probable meaning of the El Rukns' code words, but Harris could testify from his personal knowledge about their meaning, as in United States v. Estrada, 39 F.3d 772, 773 (7th Cir.1994) (per curiam).

So the question is not the admissibility of Harris's testimony but whether, as we intimated at the beginning of this discussion, giving Harris the unnecessary designation of "expert" puffed him up too much in the eyes of the jury. Maybe; but the defendants could have requested an instruction explaining to the jury that an "expert witness" is not necessarily a person of learning and probity. That would have taken care of the matter, as in United States v. DeSoto, 885 F.2d 354, 361 (7th Cir.1989). No such instruction was given. We do not know whether one was requested. It is enough that the defendants do not complain about its absence.

The defendants were sentenced under the sentencing guidelines because the conspiracy straddled the promulgation of the guidelines. They argue that the conspiracy ended in 1986, before the guidelines were promulgated, but the judge found against them on sufficient evidence. Defendant Williams makes the distinct argument that he withdrew from the conspiracy before the guidelines were promulgated. United States v. Masters, 924 F.2d 1362, 1369 (7th Cir.1991); United States v. Morgano, 39 F.3d 1358, 1369 (7th Cir.1994). It will help in evaluating his argument to distinguish between the fact of withdrawal and liability for acts of the conspiracy after withdrawal. To withdraw from a conspiracy means to quit it, whether by resignation, death, or going to the police. United States v. Maloney, supra, 71 F.3d at 654-55; United States v. Sax, 39 F.3d 1380, 1386 (7th Cir.1994); United States v. Patel, 879 F.2d 292, 294 (7th Cir.1989). Mere cessation of active participation, even when combined with a subjective determination not to resume, will not do. You must tell either another conspirator or the authorities that you're quitting. United States v. Greenfield, 44 F.3d 1141, 1149-50 (2d Cir.1995). Otherwise you could sit back, wait to see whether the conspiracy succeeded or failed, and then announce that you were a member--or not, depending on which position was advantageous in light of how things had turned out. That sort of opportunism is not permitted. There is no private withdrawal from a conspiracy.

Even when withdrawal is shown, liability may continue. X and Y build a bomb. His work finished, X "resigns." Y takes the bomb and blows somebody up with it. X may have withdrawn from the conspiracy but he remains liable for its consequences, which he set in train. United States v. Patel, supra, 879 F.2d at 294 (dictum). This is not such a case. The government concedes that if Williams withdrew from the conspiracy before the guidelines went into effect, he cannot be sentenced under them. The evidence was mixed. But there was plenty to support the district judge's finding that Williams did not withdraw before the guidelines went into effect. Since the issue went to sentencing rather than to guilt, the standard of proof was merely whether there was a preponderance of evidence against withdrawal. E.g., United States v. Morgano, supra, 39 F.3d at 1369.

Having mentioned the standard of proof, we add for completeness that when withdrawal is an issue at the guilt phase, the defendant bears the burden of production but not of persuasion. United States v. Schweihs, 971 F.2d 1302, 1322-23 (7th Cir.1992); United States v. Starnes, 14 F.3d 1207, 1210 (7th Cir.1994). Language in United States v. DePriest, 6 F.3d 1201, 1206-07 (7th Cir.1993), could be thought to place the burden of persuasion as well on the defendant. But when that language is read in context it is apparent that all the court was saying was that there must be proof of an act constituting withdrawal, not just the defendant's statement that he quit without bothering to tell anyone.

Only one other issue need be discussed, but it turns out to be critical. It is the issue of the "open unit" and it ties back to the motion for a new trial based on prosecutorial misconduct. Some of the prisoner witnesses testified that they had had no opportunity to coordinate their testimony and in fact had been separated in the jail. In both opening and closing arguments the government emphasized this lack of opportunity as a consideration bolstering the credibility of these rather dubious characters. During the trial, however, the defendants got wind of the fact that for a period of a year ending four months before the trial began the witnesses had actually been assigned to an "open unit" of the jail, where they could mingle for 16 hours of the day, discussing their forthcoming testimony to their heart's content. The district judge refused to allow the evidence of the witnesses' assignment to the open unit to be introduced at trial, where it would have been used in an effort to impeach their testimony. His ground was that the evidence would be merely cumulative, since the jury had been informed that the witnesses were let out of their cells for an hour each day and during that time could mingle freely with each other. There was also evidence that the witnesses had actually been housed together for a time and had discussed their trial and grand jury testimony, and that at times some of the witnesses had been used by the government to carry transcripts of other witnesses' testimony to those other witnesses. All this evidence too Judge Mills excluded, presumably because he thought it also cumulative, although he did not state a reason.

Evidence is "cumulative" when it adds very little to the probative force of the other evidence in the case, so that if it were admitted its contribution to the determination of truth would be outweighed by its contribution to the length of the trial, with all the potential for confusion, as well as prejudice to other litigants, who must wait longer for their trial, that a long trial creates. Fed.R.Evid. 403; United States v. Boyd, supra, 55 F.3d at 246; United States v. Pulido, 69 F.3d 192, 204 (7th Cir.1995). The evidence about the open unit (not to mention the related evidence about communication between witnesses) would not have been merely cumulative. The mingling potential of being out of your cell only one hour out of 24 is limited. If one has so little precious time of semi-freedom one may not wish to devote all or even any of it to discussing one's forthcoming testimony with other prospective witnesses. But if one mingles with these people for 16 hours of every day for a year, as the excluded evidence would have allowed the jury to infer, there is abundant time for close coordination of testimony.

The evidence should have been admitted. Had this been the only error committed at the trial, we could not say that it required reversal and a new trial. Given the totality of the evidence in the case, including some that was admitted concerning coordination of testimony by the prisoner witnesses, the exclusion of the evidence about the open unit, although unquestionably error in our judgment, may have been harmless in itself. But its prejudicial impact cannot be assessed in isolation from the prejudicial impact of the government's knowingly using perjured testimony and concealing evidence of favoritism which the defendants could have used to attack the credibility of the prisoner witnesses. It is the total impact of all the irregularities at trial, rather than the impact of each one examined in isolation, that determines whether a defendant is entitled to a new trial. United States v. Boyd, supra, 55 F.3d at 243. The determination of prejudice is to be made in the first instance by the district judge and in making this determination Judge Mills did not take into account the impact of the exclusion of the evidence about the witnesses' placement in the open unit because he did not think the exclusion erroneous. We know that he did not think the prosecutorial misconduct alone could have changed the result of the trial or undermined confidence in the verdict. But we do not know what he might think of the combined impact of that misconduct and the exclusion of the evidence about the open unit and--a point we haven't mentioned before--the evidence of prior bad acts of two of the defendants that was improperly presented to the jury and that Judge Mills instructed the jury to ignore; he may not have thought that the curative instruction reduced the prejudicial impact of the bad acts evidence to zero. We must remand the case to Judge Mills for a determination whether there is a reasonable likelihood that the totality of the irregularities that occurred at the trial affected the verdicts.

The judgments are vacated and matter is remanded to the district court for further proceedings limited to determining whether exclusion of the evidence about the open unit, when added to the instances of prosecutorial misconduct (including eliciting testimony as to prior bad acts by two of the defendants), constituted a prejudicial error and so entitles the defendants to a new trial.

VACATED AND REMANDED.