dissenting in part and concurring in part:
I join in the majority’s well-reasoned conclusion that the government failed to prove a global conspiracy, thereby creating a variance between Dejuan’s indictment and what the government ultimately proved. But I respectfully disagree with the majority’s holding that the variance did not substantially prejudice Dejuan’s right to a fair trial. In my view, by prosecuting a single global conspiracy, the government subjected Dejuan to largely irrelevant and prejudicial evidence that the jury more likely than not unfairly imputed to him. And although I agree that the relevant evidence, presented in a fair trial and when viewed in the light most favorable to the government, would be sufficient to convict Dejuan of robbing the Arvest Bank, I do not agree that this evidence was strong enough to outweigh a substantial risk of prejudicial spillover. Thus, I respectfully dissent from the decision affirming Dejuan’s convictions. I would instead reverse and remand for a new trial.
Where, as here, the government has charged a defendant with participation in a larger conspiracy but the evidence instead establishes several smaller conspiracies, the defendant is entitled to reversal if the variance prejudices his substantial rights. In turn, a variance prejudices a defendant’s substantial rights “if the evidence adduced against co-conspirators involved in separate conspiracies was more likely than not imputed to the defendant by the jury in its determination of the defendant’s guilt.” United States v. Carnagie, 533 F.3d 1231, 1241 (10th Cir.2008); see Kotteakos v. United States, 328 U.S. 750, 775, 66 S.Ct. 1239, 90 L.Ed. 1557 (1946) (the substantial right we must protect is “the right not to be tried en masse for the conglomeration of distinct and separate offenses committed by others”). Because “[t]he tactic of charging many defendants with a single massive conspiracy is fraught with the potential for abuse[,] ... we must *1275be particularly vigilant when the government seeks to bring many individuals under the umbrella of a single conspiracy.” United States v. Evans, 970 F.2d 668, 674 (10th Cir.1992). “The risk is that a jury will be so overwhelmed with evidence of wrongdoing by other alleged coconspira-tors that it will fail to differentiate among particular defendants.” Id. Accordingly, we must carefully examine .the trial record as a whole, guided but not limited by consideration of three factors. See Kotteakos, 328 U.S. at 762, 66 S.Ct. 1239 (requiring that we engage in a flexible inquiry, considering the proceedings in their entirety, “tempered but not governed in any rigid sense of stare decisis by what has been done in similar situations”). First, we consider “whether the proliferation of separate crimes or conspiracies presented in the case impaired the jury’s ability to segregate each individual conspirator’s actions and the evidence associated with ... his participation.” Carnagie, 533 F.3d at 1241 (brackets omitted). Second, we must assess “whether confusion among members of the jury concerning the legal limitations on the use of certain evidence resulted from the variance.” Id. And third, we are charged with evaluating “the strength or weakness of the evidence underlying the jury’s conviction.” Id. Considering the record in this case, I would conclude the variance substantially prejudiced Dejuan’s right to a fair trial.
I begin, as the majority does, by considering the number of individuals indicted, defendants tried, and conspiracies established at trial. The greater the number of defendants and conspiracies established at trial, the more likely it is that the jury will be unable to segregate each individual co-conspirator’s actions. United States v. Morris, 623 F.2d 145, 150 (10th Cir.1980) (“[t]he possibility of prejudice resulting from a variance increases with the number of defendants tried and the number of conspiracies proven.”). Compare Kotteakos, 328 U.S. at 753, 758, 774, 66 S.Ct. 1239 (concluding that a variance was prejudicial where thirty-two persons were indicted for conspiracy, nineteen defendants were tried together, thirteen names were submitted to the jury, and at least eight separate conspiracies were proven), and United States v. Butler, 494 F.2d 1246, 1251 (10th Cir.1974) (finding a prejudicial variance where twenty-three defendants were indicted and at least three conspiracies were established), with Berger v. United States, 295 U.S. 78, 80, 83-84, 55 S.Ct. 629, 79 L.Ed. 1314 (1985) (holding there was no prejudice when four defendants were tried for a single conspiracy and two separate conspiracies were proven), and Carnagie, 533 F.3d at 1242 (holding the risk of prejudice was minimal where three defendants were tried and at most, three conspiracies were established).
The majority, citing Camagie, concludes that because the government charged only eight individuals and tried only three defendants (Vernon, Dejuan, and Deandre Hopkins) in this case, the jury was able to compartmentalize the evidence against De-juan from that of his alleged global cocon-spirators. I disagree. Although only three defendants ultimately went to trial, the government introduced evidence regarding seven robberies and six related conspiracies.1 Thus, in terms of the *1276smaller conspiracies established at trial, this case falls somewhere between Berger (two conspiracies) and Kotteakos (eight conspiracies). And the number of smaller conspiracies here is greater than in Cama-gie, where there was evidence of, at most, three smaller conspiracies. In addition, although Dejuan was only alleged to have been involved in the Arvest Bank robbery, the risk the jury would be unable to compartmentalize the evidence against Dejuan and the other alleged coconspirators, despite Dejuan’s discrete involvement, was amplified by the fact that all of the smaller conspiracies to rob banks and pharmacies were alleged to involve shifting combinations from the same group of African-American men alleged to be Hoover Crips gang members or affiliates. This made it difficult to differentiate among the particular participants in each robbery.2 See United States v. Gallegos, 784 F.3d 1356, 1362, No. 13-6236, 2015 WL 1935223, at *5 (10th Cir. Apr. 30, 2015) (holding that in ascertaining whether there was a prejudicial variance, we consider “the complexity of the evidence and the jury’s ability to distinguish the evidence against one defendant from the evidence against his or her co-defendants”); United States v. Coward, 630 F.2d 229, 230-31 (4th Cir.1980) (finding a variance prejudicial where' it was impossible for the jury to consider the appropriate evidence against each defendant because of the subtleties of the evidence and its emanation from common sources). Accordingly, I believe consideration of the number of indicted individuals and conspiracies established at trial made it likely the jury would be unable to segregate each individual coconspirator’s actions.
In any event, although numbers are “vitally important” in examining the risk of prejudicial spillover, other . factors illustrate Dejuan was prejudiced by the variance. Camagie instructs that we should also consider whether the jury was confused concerning the legal limitations on the use of certain evidence that resulted from the variance. 533 F.3d at 1241. Perhaps the most compelling indication of the jury’s confusion is its decision to convict Dejuan and his two codefendants for conspiring to rob six banks under a global conspiracy theory for which the majority concedes there is no evidentiary support.3 *1277Cf. Carnagie, 533 F.3d at 1243 (holding that a defendant was not prejudiced where the jury illustrated that it could compartmentalize the evidence associated with each defendant where it-acquitted some defendants on some charges). But the jury’s confusion is easy to understand in the context of the record as a whole.
Under the guise of prosecuting a global conspiracy orchestrated by the Hoover Crips gang, the government subjected De-juan to a three-week trial, during which it elicited testimony from over sixty witnesses and admitted into evidence almost two-hundred exhibits. Only a small fraction of this evidence related to Dejuan or his alleged participation in the Arvest Bank robbery. See United States v. Dellosantos, 649 F.3d 109, 125 (1st Cir.2011) (“[Tjhere should be little question that the jury’s decision to find the Defendants guilty ... was influenced by the plethora of evidence implicating the other sixteen indicted co-defendants ... in a [separate] conspiracy.... As previously mentioned, this evidence included direct testimony from various co-defendants ... and a selection of hundreds of telephone conversations (from a pool of thousands of calls intercepted by the government) involving hours of intercepted communications.”); United States v. Camiel, 689 F.2d 31, 38 (3d Cir.1982) (finding a prejudicial variance where the “volume and manner of presentation of the evidence created the likelihood of spillover: i.e., that the jury might have been unable to separate offenders and offenses and easily could have transferred the guilt from one alleged co-schemer to another”); cf. United States v. Harrison, 942 F.2d 751, 758 (10th Cir.1991) (concluding there was no prejudicial spillover where the evidence regarding other conspiracies in which the defendant was not involved was “sufficiently narrow and insignificant compared to the overwhelming evidence offered with respect to the conspiracy” in which the defendant was involved).
The risk of prejudice to Dejuan due to the admission of irrelevant evidence related to the global conspiracy was compounded because much of that evidence is inherently prejudicial in nature. For example, the government presented considerable gang evidence of questionable relevance to the Arvest Bank robbery, which, in my view, increased the risk that the jury imputed guilt to Dejuan based on the actions of others. See United States v. Archuleta, 737 F.3d 1287, 1295 (10th Cir.2013), cert. denied, — U.S. —134 S.Ct. 2859, 189 L.Ed.2d 820 (2014) (“Any unfairness from gang-affiliation testimony stems from the implicit connection [drawn] between the crimes committed by gangs and the defendant gang member. In most cases, concern about such innuendo would be proper, because ordinarily the jury should not assume that the defendant gang member committed the [gang’s] crimes.”); United States v. Santiago, 643 F.3d 1007, 1011 (7th Cir.2011) (juries are “likely to associate gangs with criminal activity and deviant behavior, such that the admission of gang evidence raises .the specter of guilt by association or a verdict influenced by emotion.” (internal quotation marks omitted)).
This evidence included testimony from a former Hoover Crips gang member who testified about how the gang operates. Although it seems to have no relevance to the Arvest Bank robbery or the conspiracy *1278to commit that robbery, the witness explained how one becomes a member of the Hoover Crips gang. Specifically, he testified that to become a member of the gang, one can be “jumped in” through a violent gang initiation ritual, or “blessed in” through family members. He then explained that gang members “put work in” or “earn stripes” by committing crimes or retaliating against rival gang members. And although the Arvest Bank robbery did not involve such participants, he also explained that the phrases “little homies” or “baby gangsters” refer to 10-, 11-, or 12-year-old children, whom the gang may use as “crash dummies,” to commit crimes where they are likely to be arrested.
At one point, the government had the witness “throw up” the Hoover gang sign for the jury to see and asked him to describe the tattoos pn his own body. One tattoo depicts “NK” and “BK,” which stand for “Neighborhood Killer” and “Blood Killer,” respectively. Another tattoo depicts a devil with the word'“Hoover” on its chest, holding a machine gun and giving the finger. But there was no evidence these tattoos matched Dejuan’s tattoos, or that any gang signs were “thrown” during the Arvest Bank robbery.
In addition, much of the gang evidence presented at trial highlighted the violent nature of the Hoover Crips gang and gang membership generally. Mr. Herron testified that he had to move from Tulsa because his house was “shot up,” allegedly by two individuals, neither of whom are claimed to have anything to .do with the Arvest Bank robbery. And although there was no suggestion that the Arvest Bank robbery was related to any gang rivalry, the government presented testimony regarding the feud between the Bloods and the Crips gangs and the attendant danger of affiliating with a rival gang. A Tulsa Police Department Officer also testified that the Hoover Crips is one of the more powerful African American gangs in Tulsa. He explained that to “gain respect,” members were “required to do criminal activities” such as “shooting at people, doing drives-bys on people’s houses on rival gang members, [or] disrespecting rival gang members in'public to try to start a fight to show your allegiance to the gang.” Another Tulsa Police Department Officer testified that he spent significant time investigating the Hoover Crips because of the “gang problem in Tulsa.” ■ This same officer also testified regarding photographs he obtained from a social media website. These photographs, along with other photographs introduced into evidence, depicted various people throwing up gang signs, holding money in their mouths, and displaying guns. Neither Dejuan, Vernon, nor Stanley, the alleged Arvest Bank robbers, is depicted in any of these photographs.
I do not dispute that the introduction of some gang evidence would be appropriate, even in the absence of a global conspiracy, to prove a smaller conspiracy alleged to involve members of the same gang. See Archuleta, 737 F.3d at 1293. But with respect to the Arvest Bank robbery, the evidence of Dejuan’s gang affiliation has minimal value. Because Dejuan was alleged to have robbed the bank with his two brothers, Vernon and Stanley, “the jury did not need to rely on the gang-affiliation evidence to find knowledge, purpose, or agreement with his coconspirators.” United States v. [Vernon] Hill, 604 Fed.Appx. 759, 778, No. 13-5084, Slip Op. at 41 (10th Cir. May 22, 2015) (unpublished opinion). Indeed, the government did not introduce any evidence that Stanley is a Hoover Crips gang member.
Even if some gang evidence could be relevant to the Arvest Bank robbery, the government’s presentation of evidence *1279here went well beyond anything necessary to establish Dejuan’s conspiratorial relationship with his brothers Vernon and Stanley. In my opinion, most of the extensive gang evidence admitted had no relevance to the Arvest Bank robbery or the smaller conspiracy to commit that robbery. And because of its inherently shocking and inflammatory nature, I am convinced its admission made it highly likely the jury impermissibly transferred guilt to Dejuan. See United States v. Johansen, 56 F.3d 347, 352 (2d Cir.1995) (concluding a variance was prejudicial where the defendant was tried with an alleged coconspirator who was reportedly associated with a mobster and had committed several armed robberies and attempted murder); United States v. Jackson, 696 F.2d 578, 587 (8th Cir.1982) (recognizing the danger of guilt by association “was heightened by the inflammatory nature of much of the evidence presented”); United States v. Bertolotti, 529 F.2d 149, 158 (2d Cir.1975) (holding there was a prejudicial variance where the jury spent “two entire days listening to obviously shocking and inflammatory discussions about assault, kidnapping, guns and narcotics” and concluding the prejudicial nature of this evidence “cannot be underestimated. No defendant ought to have a jury which is considering his guilt or innocence hear evidence of this sort absent proof connecting him with the subject matter discussed”); cf. Archuleta 737 F.3d at 1295 (concluding a jury’s verdict was not unduly swayed by an emotional reaction to gang evidence where the jury parsed the conspiracies charged, finding the defendant guilty of one count in the indictment but not another).
The jury also heard evidence of six robberies that had no relevance to Dejuan in the absence of a global conspiracy. As with the gang evidence, I am convinced the admission of this evidence increased the risk of prejudicial spillover. Unlike the majority, I do not consider the Arvest Bank robbery to be of the “same character” as each of the other robberies. In particular, the Arvest Bank robbery was less violent than the Tulsa Credit Union robbery, in which the robbers tore down security cameras, fired their weapons (narrowly missing one employee), and were physically violent to credit union employees. Indeed, one credit union employee testified a robber grabbed her with enough force to rip her shirt and break her necklace, dragged her through broken glass to the vault, and held a gun to her head while demanding that she give him money. In contrast, although one of the Arvest Bank robbers pointed a semiautomatic pistol at the customers and bank employees and screamed profanities while ordering them to the ground, no shots were fired and no one was physically assaulted. Thus, even if the robberies, including the Arvest Bank robbery, can be characterized as being executed in a violent takeover style, I view the Tulsa Credit Union robbery as involving a greater degree of violence. Cf. Carnagie, 533 F.3d at 1242 (holding that the risk of prejudice was minimized where the transactions not involving defendants were of the “exact same character” as the transactions in which the defendants allegedly participated).
Moreover, to the extent Dejuan is alleged to have played a role in the conspiracy, his role was more limited than his alleged global coconspirators or even his codefendants. By way of comparison, although Dejuan is alleged to have participated in only the Arvest Bank robbery, the government alleged Vernon participated in four and Deandre participated in three robberies. Cf. id. at 1243 (holding the risk of prejudice was minimized where the defendants were “charged with conduct of approximately equal culpability.” (internal quotation marks omitted)); Johansen, 56 *1280F.3d at 352 (concluding there was a prejudicial variance where the defendant’s conspiracy involved $7,000-8,000 in credit card fraud but he was tried with unrelated defendants who engaged in over $250,000 worth of fraudulent transactions). In sum, I would conclude the government’s presentation of prejudicial evidence that had little relevance to Dejuan or his participation in the Arvest Bank robbery made it more likely than not the jury impermissibly transferred guilt to Dejuan.4
I turn now to consider “the strength or weakness of the evidence underlying the jury’s conviction.” Carnagie, 533 F.3d at 1241. Contrary to the majority, I would conclude the evidence supporting Dejuan’s participation in the smaller conspiracy to rob the Arvest Bank was not strong enough to alleviate the risk of prejudicial spillover.
I begin by addressing the majority’s analytical framework for examining the strength of the evidence. The majority engaged in a two-part inquiry: first, it determined there is sufficient evidence to convict Dejuan of the smaller conspiracy to rob the Arvest Bank; second, it “considered] whether the evidence of the [smaller] conspiracy in which Dejuan was implicated was so much weaker than the evidence introduced regarding the larger conspiracy that a substantial, prejudicial spillover must have resulted.” I agree with the first premise: the evidence tying Dejuan to the uncharged, smaller conspiracy to rob the Arvest Bank satisfied our traditional sufficiency of the evidence standards when viewed in the light most favorable to the government. But I do not agree with the majority’s second analytical step, which would require the evidence introduced with respect to the Arvest Bank robbery to be “so much weaker” than the evidence of the global conspiracy that prejudicial spillover “must” have occurred. In my view, the strength or weakness of the evidence that would support a conviction for conspiracy to rob .the Arvest Bank is merely one of the three factors we consider to determine “if the evidence adduced against co-conspirators involved in separate conspiracies was more likely than not imputed to the defendant by the jury in its determination of the defendant’s guilt.” Carnagie, 533 F.3d at 1241 (emphasis added) (internal quotation marks omitted). I would conclude the weakness of the evidence supporting Dejuan’s participation in the Arvest Bank robbery weighs in favor of a conclusion that the jury more likely than not imputed guilt to Dejuan from evidence adduced against the alleged global coconspirators.
Under Harrison, we compare the amount or quantity of evidence admitted regarding the irrelevant conspiracies to *1281the amount of evidence related to the conspiracy in which the defendant participated. 942 F.2d at 758. Where the evidence admitted regarding the irrelevant conspiracies is “sufficiently narrow and insignificant” compared to an “overwhelming” amount of evidence offered with respect to the conspiracy in which the defendant is a member, Harrison instructs there is unlikely to be a prejudicial spillover. Compare id. at 758-59, and United States v. Edwards, 69 F.3d 419, 438 (10th Cir.1995) (no prejudice where there was “ample evidence” establishing a single conspiracy), with Dellosantos, 649 F.3d at 125 (determining there was a prejudicial spillover where the jury’s decision was influenced by “the plethora” of evidence related to a separate conspiracy), and Camiel, 689 F.2d at 38 (finding a prejudicial variance where there was a significant amount of irrelevant evidence admitted with respect to the separate conspiracies).
In addition to weighing the quantity of relevant versus irrelevant evidence, we may also consider the quality of that relevant evidence; i.e., whether the relevant evidence presents compelling evidence of the defendant’s guilt. See Gallegos, 784 F.3d at 1362, 2015 WL 1935223, at *5 (holding that we focus on “the possibility of jury confusion and the strength of the evidence proving the defendant’s involvement in the smaller conspiracy”); Harrison, 942 F.2d at 758-59 (holding that in addition to being overwhelming in terms of volume, the evidence of the defendant’s involvement in the smaller conspiracy was consistent with testimony from the defendant’s coconspirators and corroborated by extrinsic evidence). The more compelling the evidence of a defendant’s guilt with respect to the smaller conspiracy, the less likely it is he was prejudiced by the admission of irrelevant and inflammatory information related to the larger conspiracy. See, e.g., Gallegos, 784 F.3d at 1362-63, 2015 WL 1935223, at *5 (holding that a variance was not prejudicial where there was. “overwhelming evidence” establishing defendant’s involvement in the smaller conspiracy); Carnagie, 533 F.3d at 1244 & n. 8 (concluding there was no prejudice where the jury heard “ample evidence” that was “more than sufficient” to convict both defendants of the smaller conspiracies); United States v. Morris, 623 F.2d 145, 151 (10th Cir.1980) (concluding there was no prejudicial spillover where there was a “strong showing of guilt”); United States v. Geibel, 369 F.3d 682, 694 (2d Cir.2004) (holding that defendants were not prejudiced by the introduction of evidence related to a larger conspiracy where there was “overwhelming” evidence of involvement in a smaller conspiracy); Andrews v. United States, 108 F.2d 511, 515 (4th Cir.1939) (no prejudicial variance where evidence ’ of guilt was “so overwhelming that it is not possible that [defendants’] cause could have been prejudiced before the jury by the evidence” of a separate conspiracy); cf. United States v. Rosnow, 977 F.2d 399, 408 (8th Cir.1992) (concluding a variance was prejudicial where, among other factors, there was a “lack of overwhelming evidence of guilt”).
Examining the strength of the evidence under this lens leaves me with the conviction that the variance was prejudicial in this case. First, as previously discussed, in an attempt to prove a global conspiracy, the government introduced an overwhelming amount of evidence that was largely irrelevant to the Arvest Bank robbery. Second, there was a relatively small amount of evidence linking Dejuan to either the Hoover Crips gang or to the Arvest Bank robbery. This significant disparity created a great risk of prejudicial *1282spillover.5 And, for the reasons explained below, I am not convinced the relevant evidence linking Dejuan to the conspiracy to rob the Arvest Bank was strong enough to make the risk of prejudice unlikely.
The majority has thoroughly and thoughtfully considered the sufficiency of the evidence related to Dejuan’s participation in the Arvest Bank robbery, -and I see no need to reexamine it in full. As explained, I concur in the majority’s conclusion that the evidence, when viewed in the light most favorable to the government, would be sufficient to support a verdict, entered after a fair trial, finding Dejuan guilty of robbing the Arvest Bank. It would also be sufficient to establish that he entered into a smaller conspiracy to rob the Arvest Bank. But I take this opportunity to highlight some of the weaknesses and inconsistencies in the evidence connecting Dejuan to the Arvest Bank robbery to illustrate why I do not consider the evidence strong enough to render the risk of prejudicial spillover unlikely. See Kotteakos, 328 U.S. at 765, 66 S.Ct. 1239 (“The inquiry cannot be merely whether there was enough to support the result, apart from the phase affected by the error. It is rather, even so, whether the error itself had substantial influence.”).
As the majority recognizes, there is no direct evidence implicating Dejuan in the Arvest Bank robbery. Thus, connecting him to the robbery requires several inferential steps taken from circumstantial evidence. Although a conviction can be based on reasonable inferences drawn from circumstantial evidence, the inferences here do little more than meet the minimum requirements of sufficiency. Cf. United States v. Summers, 414 F.3d 1287, 1296 (10th Cir.2005) (holding that the multiple inferences required to support the.defendant’s guilt were too attenuated from the evidence to support his conviction for conspiracy to commit bank robbery).
■ For example, the government’s theory of the case relies on the inference that three individuals were involved in the Arvest Bank robbery: Vernon and Dejuan, who physically entered the bank, and Stanley, who acted as the getaway driver. But none of the witnesses observed more than two robbers. Thus, the key piece of evidence supporting this theory is Officer Johnson’s identification of Dejuan driving on an alley leading away from 1107 E. Pine Street (the Hill residence) shortly after the robbery. The government argues the evidence of Dejuan’s presence in the alley compels the conclusion that Dejuan robbed the Arvest Bank with Vernon and Stanley and the three of them entered the house after the robbery. Even if we assume Officer Johnson’s identification of De-juan — which occurred three-and-a-half months after Officer Johnson’s two-second observation of the individual in the Nissan — is reliable, the strength of this evidence is limited by the timeline established *1283at trial. The Arvest Bank robbery occurred at approximately 8:30 a.m., and the money came to rest in the vicinity of the Hill residence at approximately 8:44 a.m. Officer Johnson testified he arrived at the Hill residence one minute later, at approximately 8:45 a.m., and that other law enforcement officers were present at that time establishing a perimeter around the house. Officer Johnson then observed De-juan driving out of the alley at around 9:00 a.m. But despite the fact that law enforcement officers were present and observing the house for approximately fifteen minutes before Dejuan drove out of the alley, no one saw Dejuan leave the house. This leads to either of two conclusions: Dejuan was not inside the house in the first place, or it was possible to leave the house without notice, “despite the presence of numerous police officers in the vicinity.”
It is apparent the government did not have confidence in the second possible conclusion — that an individual involved in the Arvest Bank robbery left the house undetected — because it initially tried only Vernon and Stanley for the robbery. And during that trial, the prosecutor told the jury: “I submit to you the testimony has been unrefuted that no one came or went to [the house] other than these two defendants.... No one else. The testimony was unrefuted.” United States v. [Vernon] Hill, 737 F.3d 683, 685 (10th Cir.2013), cert. denied, — U.S. —, 134 S.Ct. 1905, 188 L.Ed.2d 934 (2014). She elaborated, “[y]ou might listen for argument regarding this mysterious car that was seen in the area of [the house] ... Officer Johnson testified that, one, he wasn’t the first person' on the scene. So think about the point in time when he arrived. Certainly a perimeter had already been established. You heard testimony from officers who said that they had it on-point, they were watching that back door. Didn’t see anyone come or go until [Vernon and Stanley] came out.” Id. Thus, even the government has recognized that Dejuan’s presence in the Nissan on an alley by the Hill residence shortly after the robbery is weak evidence of his presence inside the house. And, even if we assume that Dejuan was present inside, his mere presence in the house where his family members' lived is insufficient to establish his guilt for robbing the Arvest Bank. See Summers, 414 F.3d at 1296.
The majority concludes the lack of direct evidence that Dejuan was involved in the robbery is overcome by his presence in the alley, coupled with the cell phone records for Landrum 1 and Landrum 2. According to the majority, these cell phone records establish that there were three people involved in the robbery and that one of those people was Dejuan. The theory works like this: Stanley and Dejuan both had access to Landrum 1 and Landrum 2. These two phones, along with Vernon’s phone, were all in the same location (the government suggests at Ms. Landrum’s house) shortly before the robbery, indicating that the users of these phones met up before the robbery. Then, immediately before the robbery, Landrum 2 was turned off, which the government implies indicates that one of the robbers (presumably Dejuan) turned the phone off during the commission of the robbery. In addition, Vernon’s cell phone called Landrum 1 immediately before the robbery, from which the government infers Vernon called the getaway driver (presumed to be Stanley), who was then in possession of Landrum 1. The government offered no direct evidence that either Stanley or Dejuan used Landrum 1 or 2 on the morning of the Arvest Bank robbery. But the majority concludes Dejuan can be linked to Landrum 1 because Landrum 1 called Vernon’s cell phone at about the time Officer Johnson saw Dejuan turn out of the alley and proceed in a southern *1284direction and the cell phone records indicate Landrum 1 was also moving south at that time. I am not persuaded this evidence so strongly implicates Dejuan in the Arvest Bank robbery that it made the possibility of prejudicial spillover unlikely.
First, the location of the cell phones was imprecise. Although the involvement of a particular cell tower indicates a general location of the cell phone user, the user could be anywhere within a multiple-block area. And the cell phone records do not reflect who used a particular phone or what was said in any particular conversation. Thus, it is possible that someone other than Dejuan or Stanley could have been using either cell phone at any given time, and for purposes related or unrelated to the robbery. Indeed, even the government seemed confused about what can be inferred from the cell phone records.
At trial, the government first argued that one of the robbers who entered the bank (presumably Dejuan) must have used Landrum 2 because it was turned off during the robbery. The government next asserted Vernon used his own cell phone and that his call prior to the robbery was to contact a getaway driver (presumably Stanley), who used Landrum 1. But if De-juan was the second robber who actually entered the Arvest Bank as the government argued, that would put him in possession of Landrum 2 — the phone turned off immediately before the robbery — not Landrum 1, the phone used by the getaway driver. However, the cell phone that called Vernon’s cell phone while traveling south at about the time Officer. Johnson saw Dejuan in the Nissan is Landrum 1. Thus, in addition to inferring all of the other facts linking Dejuan to the robbery, we must also infer that he entered the house briefly, obtained Landrum 1 from Stanley, exited the home without being observed by any of the police officers surrounding it, and then called Vernon on Landrum 1 to warn him of the police presence. While this may be true, it requires considerable effort by the jury to fill in the significant gaps in the evidence. In my opinion, this rendered the evidence weak enough that the glut of irrelevant and prejudicial evidence about the nonexistent global conspiracy more likely than not improperly influenced the jury’s verdict.
The government and the majority also rely on evidence of the physical description of the two robbers who entered the bank to place Dejuan in the bank and to exclude Stanley as the second robber. In particular, the government contends the testimony concerning the relative- skin tone and height of the suspects is compelling evidence supporting the conclusion that De-juan robbed the Arvest Bank. I am not convinced. First, if the eyewitness, testimony and surveillance footage so clearly eliminated Stanley as a suspect, it seems odd that the government first prosecuted, albeit unsuccessfully, Stanley as the second robber who entered the bank with Vernon. See United States v. [Stanley ] Hill, 749 F.3d 1250, 1252 (10th Cir.2014) (recognizing that the government initially tried Stanley for bank robbery under the theory that Vernon and Stanley physically entered the bank). Second, a careful examination of the evidence related to the robbers’ skin tone and height belies the suggestion that it was sufficiently strong to negate the risk of prejudicial spillover. Because the robbers wore masks, long sleeves, and gloves, only small parts of their skin were visible. Although one eyewitness testified that in his subjective opinion one of the robbers had a darker skin tone and the other had a lighter skin tone, three other eye witnesses were unable to distinguish between the robbers’ skin tone. In fact, one eyewitness testified that the masks made the robbers’ skin appear darker than it likely was, which *1285calls into question the reasonableness of the inference that Stanley’s “medium” skin tone is not dark enough to match the second robber. Another witness testified she could not see any skin showing — not even enough to identify the robbers’ race, while a third testified she didn’t notice the robbers’ skin color. Accordingly, the evidence was less than compelling concerning the relative skin tones of the robbers.
The evidence that Dejuan matches the height of the second robber while Stanley does not is also problematic. It is based exclusively on an FBI agent’s comparison of the height of the teller counter with the agent’s assumptions about the height of the second robber, based on the.surveillance video’s recording of the second robber jumping over the counter. Based on this comparison, the agent estimated the second robber’s height between 5'8" and 5'11" tall. The same agent opined that 'Stanley is 5'6" to 5'7" tall. Even if we accept the accuracy of these estimates, Stanley is within one inch of the height range estimated for the robber. In my view, this is not a great enough discrepancy that he can be excluded as the second robber based on his height. Nor can De-juan be strongly implicated as the second robber based on the fact that he is taller than Stanley.6 The agent’s estimate of the second robber’s height, which spans three inches, is hardly conclusive evidence that Dejuan was involved. And it was not the only evidence presented to the jury about the height of the Arvest Bank robbers. A witness who observed the robbers fleeing on foot after the robbery reported to the FBI that they were 6'0" and 6'2" tall, respectively. This would make both robbers taller than Dejuan (and much taller than the agent estimated the second robber to be).
But even if Stanley could be eliminated as one of the robbers who entered the bank based on his height or skin tone, it does not identify the second robber as Dejuan. Many African-American men undoubtedly have “darker” skin and stand between 5'8" and 5'11" tall. And some of the men who meet both of those criteria may also be members of the Hoover Crips gang or have other connections to Vernon and/or Stanley. Indeed, Vernon was accused of conspiring to rob the Metro and CVS pharmacies with multiple people who may fit the general description. Therefore, even if it can be reasonably inferred that a third person participated in ’ the Arvest Bank robbery, the height or skin tone evidence does not strongly identify Dejuan as that person.
Finally, the strength of the evidence related to the robbers’ respective identity based on their skin tone and height is debatable because the government posited inconsistent theories at trial regarding the roles that the lighter-skinned or darker-skinned robber played during the robbery. During the Arvest Bank robbery, one of the robbers stayed in the lobby holding the victims at gunpoint while the other robber jumped over the teller counter to collect the cash. The one eyewitness who claimed he could distinguish between the robbers based on their relative skin tones testified the darker skin-toned robber (allegedly Dejuan) jumped behind the teller counter while the lighter skinned-toned individual (allegedly Vernon) stayed on the other side controlling the victims. The government made much of this fact in opening arguments, stating the evidence would show that Vernon held the victims at gunpoint while “Dejuan Hill mounts the- teller coun*1286ter, jumps over the counter, and begins collecting money.” Later, however, the government contradicted its own theory. It showed the jury a video of a masked robber (alleged to be Vernon) leaping over the counter of the Metro Pharmacy, and then argued this person was the exact same person who jumped over the counter during the Arvest Bank robbery. The prosecutor explained in closing argument, “If you’re a football fan and you have a favorite running back, you can tell their style. Look at Vernon Hill’s style of running. You look at that.... And when you see him leap the counter in the Metro Pharmacy robbery, it is exactly the same as the person who leaps the counter in the Arvest robbery.” But if the skin tone and height of the second robber is compelling evidence of Dejuan’s involvement, it is compelling evidence that he, not Vernon, jumped over the counter during the Arvest Bank robbery. Where even the government exhibited confusion regarding which brother played which role in the robbery, I am convinced the jury more likely than not confused the evidence as well. For these reasons, I would conclude that the skin tone and height identification evidence related to the second robber provides only weak evidence of Dejuan’s guilt.
Finally, I address the fact that law enforcement officers searching the Hill residence located only one set of clothing worn by the robbers. The majority adopts the government’s current theory that the absence of a second set of clothing in the Hill residence strongly indicates Dejuan was wearing the second set when he drove away. But here too, the evidence is weaker than the government acknowledges. Although law enforcement officers found one hoodie, one ski mask, and one pair of black pants- in the house, officers also found two pairs of black gloves and numerous generic gray and black tee shirts matching the description of the clothes worn by the second robber. Thus, the government’s characterization of this evidence as establishing that the officers executing the search warrant found only one set of robber’s clothing is generous.7 Even if the government is correct and only one set of the clothes worn by the robbers was found at the Hill residence, the inference that Dejuan was wearing the clothes when he drove past Officer Johnson is merely one reasonable possibility. It could just as easily mean the second robber never went to the house in the first instance. Again, the jury was required to make numerous inferences to convict De-juan. In my view, this created a high likelihood that the jury would be improperly influenced by the overwhelming and prejudicial evidence unrelated to the Ar-vest Bank robbery.
In summary, I would conclude the prejudicial evidence admitted under the guise of prosecuting a global conspiracy and the weakness of the evidence connecting De-juan to the Arvest Bank robbery indicate the variance substantially prejudiced De-juan’s right to a fair trial. As a result, I would remand to the district court for a new trial.
. The record makes clear that there was evidence of six smaller conspiracies, not five as the majority suggests. As the majority notes, the government established evidence of seven robberies, including the (1) IBC Bank; (2) Dooley Pharmacy; (3) Barnes Pharmacy; (4) Metro Pharmacy; (5) Tulsa Credit Union; (6) Arvest Bank; and (7) CVS Pharmacy (neither charged nor identified as an overt act furthering the global conspiracy but offered as prior acts evidence). However, all of these robberies — with the exception of the IBC Bank— *1276involved multiple parties. Unlike the majority, I include the CVS Pharmacy as an established conspiracy because the government presented evidence that showed three people robbed the CVS Pharmacy. But even if I were to exclude the CVS Pharmacy as a smaller conspiracy, it would not change my conclusion that the government’s presentation of the case, taken as a whole, prejudiced Dejuan. See Kotteakos v. United States, 328 U.S. 750, 762, 66 S.Ct. 1239, 90 L.Ed. 1557 (1946) (requiring that we' consider the proceedings in their entirety, "tempered but not governed in any rigid sense of stare decisis by what has been done in similar situations”).
. Specifically, the government alleged that Marquis Devers and Kenneth Hopkins robbed the Dooley Pharmacy; Deandre Hopkins and two unknown others robbed the Barnes Pharmacy; Vernon, Mr. Devers, Deandre, Chris Lewis, Dontáyne Tiger, and Duncan Herron robbed the Metro Pharmacy; Vernon, Mr. Devers, and Mr. Lewis robbed the CVS Pharmacy;. Mr. Lewis, Deandre, Mr. Devers, Mr. Tiger, James Miller, and Duncan' Herron robbed the Tulsa Credit Union; and Vernon, Dejuan, and Stanley robbed the Arvest Bank.
. The confusion of the jury is highlighted by the fact that the government’s evidence actually refuted the existence of a global conspiracy. For example, the evidence established that the alleged global coconspirators are all members of different sets of the Hoover Crips gang. Yet the government’s witness, a former Hoover Crips gang member, testified that each set operates within itself, and there is no individual at the top who has control over multiple sets. He also explained that if the members of a particular set committed a robbery, they would be unlikely to share that money with anyone outside their set or to give it to someone who might be higher in the *1277organization. Cf. United States v. Caldwell, 589 F.3d 1323, 1329 (10th Cir.2009) (recognizing that interdependence is a required element in conspiracy and that "[i]nterdepen-dence exists where coconspirators intend to act together for their shared mutual benefit within the scope of the conspiracy charged”).
. The majority concludes that the risk of prejudice from this largely irrelevant evidence, must not have been great because "Dejuan did not object to the introduction of much of the evidence introduced to support the global conspiracy." But Dejuan did object to the use of evidence related to the global conspiracy in which he played no role. Before trial, he argued that he should not be joined with the other defendants under Federal Rule of Criminal Procedure 8 because he was not a part of the global conspiracy (a point on which Dejuan was ultimately correct). He also argued that even if joinder was proper, the court should sever his trial because the evidence related to the global conspiracy would be unfairly prejudicial. The court rejected his. arguments. Once it did so, all evidence related to the global conspiracy was relevant and admissible in the joint trial, making further objection futile. It is for precisely this reason that we must be "vigilant” in determining whether the government’s decision to charge a global conspiracy, which it fails to prove at trial, has prejudiced a defendant. See United States v. Evans, 970 F.2d 663, 674 (10th Cir.1992).
. The majority concedes there was an overwhelming amount of evidence admitted that had little relevance to Dejuan, but contends the admission of this evidence actually bene-fitted Dejuan by allowing him to compare his behavior to that of more culpable defendants. Specifically, the majority claims "a reasonable jury might wonder whether Dejuan was really involved in the Arvest Bank robbery considering the dearth of evidence that he was involved in the other criminal activity. ” It is true that Dejuan tried to minimize the impact of the prejudicial evidence by accurately arguing in closing that the government had put on very little evidence tying him to the global conspiracy or the Arvest Bank robbery. But I do not agree this weighs against a finding of prejudice. Despite the dearth of evidence admitted to prove a global conspiracy, the jury found one existed and that Dejuan participated in it. Thus, notwithstanding counsel’s best efforts, the jury was obviously ' confused in a manner that prejudiced Dejuan.
. There was no evidence introduced at trial regarding Dejuan’s actual height, although he was present and could be observed by the jury. Information in the record indicates that Dejuan is 5'10" tall.
. The compelling nature of this evidence is also undermined by the fact that when prosecuting Stanley for robbing the Arvest Bank, the government considered the clothing found in the home to be consistent with their theory that only Vernon and Stanley had participated in the Arvest Bank robbery. See United States v. [Vernon] Hill, 737 F.3d 683, 685 (10th Cir.2013) cert. denied, — U.S. —, 134 S.Ct. 1905, 188 L.Ed.2d 934 (2014) (the prosecutor reminded the jury that Vernon and Stanley must have been the same two men who robbed the bank because officers found items in the house that were consistent with items worn by the robbers).