dissenting.
Realizing that its case against Richard Washam was far from “overwhelming,” (Maj. Op. 573), the government went on a zealous offensive to bolster its prosecution with inadmissible evidence. Having succeeded in securing erroneous evidentiary rulings from the district court, the government then was permitted to disregard the boundaries of narrow evidentiary exceptions. In a case built on three pieces of evidence — two of which were, at best, largely inadmissible — I cannot agree that the jury verdicts were based on a fair trial. Accordingly, I respectfully dissent.
FACTUAL BACKGROUND
On March 26, 2003, an individual walked into a U.S. Bank in Bowling Green, Kentucky, approached a teller, pulled out a gun, and demanded money. Less than a month later, the same bank was robbed again under similar circumstances. In both robberies, the individual was described as an African-American man of slender build, thin facial features, and a calm demeanor, who was dressed in a dark-colored hooded windbreaker and partially disguised by a baseball cap and sunglasses. The crimes went unsolved.
Over one month later and three hours away, authorities apprehended Richard Washam minutes after he robbed a PNC Bank in Florence, Kentucky. Following his arrest, Washam confessed that he committed the Florence robbery to support a cocaine addiction. Believing that Wa-sham’s explanation indicated that this was not likely his first robbery, the FBI culled its records for similar unsolved bank robberies committed by an individual fitting Washam’s description.
Coming across the two unsolved Bowling Green robberies, investigators prepared a photo array containing Washam’s picture and showed it to several of the U.S. Bank employees. That array, however, was un-disputedly suggestive. It did not include individuals who looked similar to Washam, and Washam’s photograph was the only one that fit the description of the Bowling Green robber. Of the five employees shown the suggestive array, three, including two tellers, selected Washam’s photograph as the one that most resembled the robber. After Washam was linked to a vehicle similar to the getaway car used at the second Bowling Green robbery, he was indicted on the present charges. At trial, only three pieces of evidence supported the prosecution’s theory that Washam committed the Bowling Green robberies: the photo array identifications, the similarities to the Florence robbery, and the car. As explained below, neither the photo array identifications nor the evidence of the Florence robbery should have been admitted.
ANALYSIS
I. Photo Array Identifications
A. Suppression Hearing
Prior to trial, Washam moved to suppress the identifications made from the photo array. After an independent examination, the district court agreed that the array was unduly suggestive:
In the present case, the descriptions given by the witnesses to the officers indicated that the bank robber was “thin,” “skinny,” “thin build,” “not heavy,” and “not overweight”. Armed with this de*576scription, the officers presented the witnesses with a photographic array comprised of six African-American males.... However, Washam’s features bear little resemblance to the others in the array. Defendant’s picture clearly contains facial features or characteristics foreign to all of the other pictures. Wa-sham’s face appears long and thin, whereas the other five individuals have rounder, fuller features. Additionally, Washam’s nose and [cheek] bone structure do not in any way resemble the other five individuals.... Clearly, with the witness descriptions of the bank robber as a thin man, the placement of Defendant’s photo in a photo array with clearly heavy set men with round full faces suggests to the witnesses that the Defendant “is more likely to be the culprit.” The Court concludes that the dissimilarities among the participants in the photo array resulted in an identification procedure which was unduly suggestive to Washam.
(RE 75, Op. and Order 4-7) (internal citations omitted).
Having deemed the array suggestive, the court held a second evidentiary hearing allowing the government to prove that the identifications were nevertheless reliable.1 To support that effort, the government elicited the testimony of the three employees who selected Washam’s photograph from the array. For each witness, the government introduced written statements made by the employees when they were first presented with the array. One employee wrote, “I [ ] feel that number 3 looks very similar to the gentleman that robbed our branch. When I saw the lineup he immediately jumped out at me.” (RE 99, Hr’g Tr. 9.) The second employee wrote, “[N]umber 3 looks like the robber.” (Id. at 28.) The third employee wrote, “Possible number 3, same shaped face, cheekbone structure, eyebrows look like the [robber].” (Id. at 40.)
After each employee read his or her written statement out loud, on the next beat, the prosecutor posed the following question:
Now, one might say that when someone says that a photograph looks like someone, that the person identifying the photograph might not be sure. Are you sure that number 3 was the robber?
(Id. at 9, 28-29, 40.)
In response to the prosecutor’s leading question, the first employee changed his identification. As did the second employee. And so with the third employee. In each instance, the employee changed his or her identification to positively identify Wa-sham as the robber, not merely someone who looked like the robber. Based on this testimony, the district court issued a short oral opinion admitting the photo array identifications as reliable. At trial, only one of the three employees was able to successfully identify Washam in person.
B. Legal Framework
“A conviction based on identification testimony violates the defendant’s constitutional right to due process whenever the identification procedure is ‘so impermissibly suggestive as to give rise to a very substantial likelihood of irreparable misidentification.’ ” United States v. Meyer, 359 F.3d 820, 825 (6th Cir.2004) (quoting Thigpen v. Cory, 804 F.2d 893, 895 (6th Cir.1986)). This Court applies a two-step analysis for determining the admissibility of identification evidence. Id. (citing Lea*577better v. Edwards, 35 F.3d 1062, 1071-72 (6th Cir.1994)). First, the defendant must prove that the identification procedure was unduly suggestive. Ledbetter, 35 F.3d at 1071-72. If the defendant proves suggestiveness, the burden then shifts to the government to prove that the totality of the circumstances demonstrates that the identification was “nevertheless reliable.” Id. at 1071.
In reviewing a district court’s ruling on a motion to suppress, we apply the clearly erroneous standard to the district court’s factual findings and the de novo standard to its legal conclusions. Meyer, 359 F.3d at 824 (citing United States v. Dotson, 49 F.3d 227, 229 (6th Cir.1996)). One such question of law is “[w]hether identification evidence was sufficiently reliable so as not to offend [the defendant’s] rights under the due process clause.” Id. (citing Smith v. Perini, 723 F.2d 478, 481 (6th Cir.1983)). In this case, the district court found the array suggestive, and the government does not appeal that ruling. Accordingly, because the court’s factual findings are not in dispute, the only question remaining is the reliability of the identifications, which we review de novo.
Reliability of the eyewitness identification is the “linchpin” of our suppression analysis. See Perry v. New Hampshire, — U.S. -, -, 132 S.Ct. 716, 724-25, 181 L.Ed.2d 694 (2012) (quoting Manson v. Brathwaite, 432 U.S. 98, 114, 97 S.Ct. 2243, 53 L.Ed.2d 140 (1977)). To determine whether an identification made off a suggestive procedure is nevertheless reliable, we apply a totality of the circumstances approach including those factors set forth in Neil v. Biggers, 409 U.S. 188, 93 S.Ct. 375, 34 L.Ed.2d 401 (1972) and reiterated in Brathwaite, 432 U.S. 98, 97 S.Ct. 2243, 53 L.Ed.2d 140 (1977). Perry, 132 S.Ct. at 724-25:
(1) the opportunity of the witness to view the criminal at the time of the crime; (2) the witness’ degree of attention at the time of observation; (3) the accuracy of the witness’ prior description of the criminal; (4) the level of certainty demonstrated by the witness at the confrontation; and (5) the length of time between the crime and the confrontation.
Biggers, 409 U.S. at 199, 93 S.Ct. 375.
C. Analysis
Despite the full suppression hearing, the district court disposed of the Biggers factors in a perfunctory oral opinion. The court’s opinion gave short shrift to the heightened due process concern that exists “when [ ] misidentification is possible because the witness is called upon to identify a stranger whom [the witness] has observed only briefly, under poor conditions, and at a time of extreme emotional stress and excitement.” Ledbetter, 35 F.3d at 1070. The district court’s ruling did not indicate that it considered any of the Big-gers factors that favored Washam: for instance, that the robber was a stranger to the tellers, that the employees only interacted with the robber for a few brief minutes, or that several of the employees stated that the gun distracted their focus from the robber himself.
However, it is the evolution of the employees’ initial identifications into their testimony at the evidentiary hearing that is my primary concern. The changes in their testimony pertain directly to the fourth Biggers factor: “the level of certainty demonstrated by the witness at the confrontation.” Biggers, 409 U.S. at 199, 93 S.Ct. 375. The district court completely mishandled this factor when it allowed the employees’ changed testimony to trump their initial identifications. As with all Biggers suppression, the court’s focus should have remained on the level of cer*578tainty expressed by the employees when they made their initial identifications — in other words, the certainty expressed “at the time of confrontation.” Id. A witness’ level of certainty thereafter, especially after the identified individual has been arrested and indicted, is not part of the Biggers’ analysis. Rather, continued certainty — or in this case, increased certainty — is tested through in-court identification and cross-examination. The district court missed this distinction, and the majority repeats the error by dismissing the convenient change to the employees’ testimony.
What should have been significant to the district court were the statements the employees made when they were first shown the array. At that stage, each employee made statements indicating that they selected Washam’s picture because his was the one that most resembled or “looked like” the robber. However, as all parties agree, Washam was the only individual pictured that plausibly “looked like” the robber. Therefore, the employees’ certainty on that score provides little to support the reliability of their selections from the photo array.
Tellingly, the employees’ testimony at the evidentiary hearing only reinforces that they noticed Washam’s photograph did not resemble the others. One employee stated that Washam’s photograph “immediately jumped out” from the others. Another noted that Washam’s photograph was the only one with a “shaped face, cheekbone structure, [and] eyebrows” that looked liked the robber — the exact same qualities that led the district court to find the array suggestive. The last employee testified that Washam’s photograph was the only one “even close” to looking like the robber. These statements clearly indicate that the employees’ selections were affected by the suggestiveness of the array and were therefore unreliable. As such, the photo array identifications should have been suppressed.
All too often, investigatory identifications are treated as one in the same with in-court identifications. However, they are different pieces of proof with different evidentiary value, and they ought not be conflated. Excluding a suggestive investigatory identification is not fatal to the prosecution, although of course, doing so may weaken the government’s case. It is, however, the appropriate course of action, especially given the greater evidentiary value of pretrial identifications. (See Maj. Op. 572-73.) Excluding an unreliable pretrial identification only forces the government to rely solely on a witness’ in-court identification. Of course, in this case and with the benefit of hindsight, we know that only one of the three employees was able to repeat his positive identification at trial.2
Finally, I am troubled by the government’s efforts to bolster the photo identifications at the evidentiary hearing. Its efforts suggest that the prosecutor, if no one else, recognized the deficiencies of the identifications and sought to repair them. However, the employees’ willingness to change their testimony at the prosecutor’s behest certainly does not improve their credibility or inspire further confidence about the reliability of the photo array identifications.
II. Florence Bank Robbery
Rule 404(b) of the Federal Rules of Evidence provides that a defendant’s pri- or bad acts may not be used as proof of propensity, though they may be offered for certain other permissible purposes. *579Admission under Rule 404(b) is limited by Rule 403, which requires the balancing of evidence’s probative value against its prejudicial impact. While I agree that Washam’s statement about his cocaine addiction was admissible to show motive, I cannot concur that the Florence robbery was probative of identity under Rule 404(b). Additionally, the district court utterly failed to tailor the admission of the Florence robbery to any permissible purpose, in contravention of Rule 403.
“Prior acts or crimes [may] be admitted to show identity, provided they are of ‘sufficient distinctive similarity’ with the charges in the indictment to ‘create a pattern or modus operandi.’ ” United States v. Allen, 619 F.3d 518, 524 (6th Cir.2010) (citing United States v. Perry, 438 F.3d 642, 648 (6th Cir.2006), U.S. v. Mack, 258 F.3d 548, 554 (6th Cir.2001)). This case falls far below that standard. Here, the commonalities, even if viewed cumulatively, were not sufficiently unique to constitute a signature or modus operandi that would be at all probative of identity. At best, the similarities were vague — a nondescript perpetrator committing a robbery in an unorganized and unsophisticated fashion. This level of generality cannot rise to the level of a common plan, a distinctive pattern, or a signature. See United States v. Clay, 667 F.3d 689, 695-96 (6th Cir.2012); United States v. Phillips, 599 F.2d 134, 136 (6th Cir.1979). Instead, the generalities of the prior crime only emphasized the impermissible inferences that (1) Washam was a “bad man” by virtue of his criminal history; and (2) that because he committed another robbery, be probably committed the ones charged. Phillips, 599 F.2d at 136; United States v. Johnson, 27 F.3d 1186, 1193 (6th Cir.1994). If the parallels claimed here were truly “distinctive,” as the majority contends (Maj. Op. 572), then the exception swallows the rule, and the admission of propensity evidence becomes the standard of the day.
Comparing this case to others where we have applied a totality-of-similarities approach reveals just how threadbare the alleged similarities are in this case. For instance, in Mack we found that Rule 404(b) evidence was properly admitted where, over a series of over nine robberies, the perpetrator wore a ski mask with a hooded sweatshirt, leapt over the teller counter to retrieve the monies, and leapt back over it in leaving. Mack, 258 F.3d at 553-54. In Perry, we found a series of robberies sufficiently unique, where the perpetrator’s signature involved seeking change for a $50 bill and asking to purchase money orders before pulling a gun out of a bookbag and demanding money. Perry, 438 F.3d at 648. Likewise, in United States v. Price, 516 F.3d 597, 603-04 (7th Cir.2008), the robber forced an employee into the closed bank at gunpoint, turned off the alarm, accessed the vault, and forced the employee to lie on the ground until he escaped. By comparison, the alleged commonalities in this case are utterly unremarkable.
By contrast, the dissimilarities among the robberies Bowling Green and Florence robberies were far more distinctive than were any of their purported similarities. For instance, the means of the robber’s escape was not consistent. In the Bowling Green robberies, the perpetrator wore a baseball cap and sunglasses, but the Florence robber did not take such efforts to conceal his identity. Perhaps the most unusual aspect of the Bowling Green robberies — the ruse of asking for change of smaller bills into larger ones and vice ver-sa — was not used in the Florence robbery. Cf Perry, 438 F.3d at 648. Moreover, the robber’s friendly demeanor was not consistent, as the majority claims. In at least one of the Bowling Green robberies, the *580teller testified that the robber was immediately aggressive with her. (RE 179, Tr. at 112.)
Finally, even if the Florence robbery was relevant for certain limited purposes under Rule 404(b), Rule 403 requires its exclusion “if its probative value is substantially outweighed by the danger of unfair prejudice.” Fed.R.Evid. 403. The district court failed to strike the proper balance here. As the majority admits, the probative value of the Florence robbery to Wa-sham’s guilt of the charged crimes was quite limited. However, despite acknowledging its prejudice, the court made no effort to limit the admission of the Florence robbery to its permissible evidentiary purposes.
Instead, the court allowed the prosecution to introduce the entire body of evidence one would expect if Washam had been on trial for the Florence robbery instead. The government presented three different officers who testified in detail about their investigation of the Florence robbery, Washam’s attempt to evade authorities, his arrest shortly after fleeing the bank, and the incriminating evidence discovered on his person and in his car thereafter. See United States v. Hemphill, 76 Fed.Appx. 6, 15 (6th Cir.2003) (describing similar evidence as extrinsic and prejudicial to the jury’s evaluation of the crimes charged). The jury also heard extensive testimony from the Florence bank teller, who described not only her limited interactions with the robber, but also details that were not probative of the Bowling Green robber’s identity. The most blatant example occurred when the Florence teller was asked to identify Washam in court. Only after the defense objected and the teller was unable to immediately make an in-court identification, did the government abandon its efforts and concede that the Florence teller’s identification was not probative of the Bowling Green robber’s identity.3
Moreover, the court admitted extensive testimony regarding the gun used in the Florence robbery. During opening statements, the prosecutor admitted that the gun was of limited probative value, because none of the Bowling Green employees could identify the gun as the one used against them.4 (RE 178, Tr. 142-43.) Despite the weakness of this link, the court admitted the gun into evidence, allowed the prosecution to prominently display it throughout trial, and permitted the prosecutor’s to encourage multiple witnesses to physically examine the weapon. The lead FBI agent of the Bowling Green crimes also used the gun as a prop and presented a detailed explanation of its capabilities and characteristics. This concerted emphasis went far beyond any limited permissible purpose.
The government argues that it was necessary to introduce the full background of the Florence robbery because Washam’s motive and his link to the Bowling Green getaway car could not have been presented in another manner. These arguments are baseless. Washam’s statement about his cocaine addiction could have been introduced without mentioning the Florence *581robbery at all. Likewise, Washam’s sale of his white sedan the day after the Bowling Green robbery, while assuredly probative of his guilt, was a fact that existed independently to the details of the Florence robbery. The testimony of the car’s purchasers and of the state vehicle registration officials could easily have been presented without introducing the entire Florence case to the jury.
Moreover, even if the purported similarities among the robberies were probative of the robber’s identity, their parallels could have been introduced by a variety of less prejudicial means, including: by defense stipulation, by reading from Wa-sham’s plea agreement in the Florence case, by calling only the lead investigator of the Florence robbery, or by calling only the Florence teller. Clay, 667 F.3d at 697 (citing United States v. Haywood, 280 F.3d 715, 723 (6th Cir.2002); United States v. Merriweather, 78 F.3d 1070, 1077 (6th Cir.1996)) (“One factor in balancing unfair prejudice against probative value under Rule 403 is the availability of other means of proof.”) At the very least, the testimony could have been curtailed substantially to limit its prejudicial effects. Instead, the transcript shows that the prosecution spent roughly forty percent of its case-in-chief presenting evidence to the jury about the Florence robbery. That amount of time, of course, only rises sharply if we consider the amount of time the defense spent combating the improperly admitted evidence.
The facts of this case bear close similarity to those that caused the Third Circuit to reverse the defendant’s conviction in United States v. Hans, 738 F.2d 88 (3d Cir.1984). In Hans, the trial court admitted the testimony of an investigating FBI agent, who described the manner in which the defendant became the suspect of the charged robbery. Id. at 94. The agent testified that, after learning that one of the other suspects was originally from Michigan, he contacted Detroit FBI agents and presented them with the modus operandi and a description of the robbers and “asked them if they had anyone from the area who might logically fit as a suspect in this matter.” Id. Despite the far more distinctive similarities among the crimes in Hans — all involving three armed robbers wearing Halloween masks, windbreakers, dark gloves, and ranging in height from 5'5" to 5'9" — the Third Circuit held that the agent’s testimony was improperly admitted propensity evidence. Id. at 95. The court explained that “the only reasonable inference that a reasonable juror could draw from [the agent’s] testimony was that [the defendant] was well-known as a bank robber to the Detroit FBI.” Id. Similarly, the court reasoned that, even if admissible under 404(b), the testimony was “highly prejudicial” under Rule 403. The court found it “difficult to imagine testimony more prejudicial than [the FBI agent’s] implication” that the defendant was known to Detroit police to be a professional bank robber. Id. I share the Hans court’s concern that the other acts evidence here, along with the manner by which it was introduced, encouraged the jury to draw the very inferences forbidden by Rule 404(b). Merriweather, 78 F.3d at 1079.
Our review must take into account “what the error meant to [the jury], not singled out and standing alone, but in relation to all else that happened.” United States v. Cowart, 90 F.3d 154, 158 (6th Cir.1996) (citing Kotteakos v. United States, 328 U.S. 750, 764, 66 S.Ct. 1239, 90 L.Ed. 1557 (1946)). “While a limiting instruction can minimize the prejudicial impact of prior criminal acts, it is not a ‘sure-fire panacea for the prejudice resulting from needless admission of such evidence.’ ” Clay, 667 F.3d at 696 (quoting Haywood, 280 F.3d at 724). In a case where well over half the *582evidence presented to the jury concerned a crime for which the defendant was not on trial, I cannot say with a “fair assurance that the jury’s verdict was not substantially swayed” by the improperly admitted evidence, regardless of the court’s brief limiting instructions. Merriweather, 78 F.3d at 1079 (internal quotations omitted).
CONCLUSION
Undoubtedly, the FBI’s instinct that Washam may have committed other robberies prompted its investigation into him for the Bowling Green crimes. The photo array identifications provided evidence to believe that the investigators’ instincts were valid and made Washam a viable suspect for the unsolved robberies. However, evidence that supports probable cause for an arrest is not always admissible at trial. See generally Phillips v. Allen, 668 F.3d 912, 915-16 (7th Cir.2012) (citing Illinois v. Gates, 462 U.S. 213, 103 S.Ct. 2317, 76 L.Ed.2d 527 (1983)). Often, our legal process must shield certain investigatory leads from the jury’s view in order to ensure the defendant receives his constitutionally guaranteed due process. Although the ultimate harm caused by these errors is a perhaps the closer question in this case, the stakes here are high. Wa-sham was 46-years-old at sentencing, and the district court’s 677-month prison term, imposed largely by way of mandatory mín-imums related to his Florence conviction, represents a life-sentence. Accordingly, in a prosecution where the evidence was largely improperly admitted, I cannot simply rubberstamp the verdicts and must respectfully dissent.
. At the defense’s request, Washam did not attend the second evidentiary hearing, in order to prevent taint.
. The majority fails to mention that the only employee to identify Washam at trial was not one of the tellers and was the employee with the worst vantage point to observe the robber.
. The government’s attempt to secure this irrelevant identification also likely prejudiced Washam before the jury, because he refused to stand and smile at the teller's request, giving the jury the impression that he was evasive and uncooperative.
. In fact, the Bowling Green employees never consistently testified that the gun was even similar to the one used against them. All the employees could say was that the gun they saw was a handgun; the employees inconsistently described the gun as silver, black, dark grey, metal, and plastic. (See, e.g. RE 154, 104; RE 179, 8-9.)