dissenting in part.
I agree that Martin has not established that trial counsel was ineffective for not requesting a charge pursuant to United States v. Barber, 442 F.2d 517 (3d Cir. 1971), or not calling Polk or El to testify. However, I disagree with the majority’s conclusion that defense counsel was not ineffective for failing to file a pretrial motion to restrict Detective Mary Seifert’s testimony and for inexplicably eliciting Edna Cook’s identification during his cross-examination of Seifert. Accordingly, I respectfully dissent from Part II. A. of the majority’s opinion.
I.
Despite the other evidence in this case, it is exceedingly difficult to review this record without focusing on a single question: “Is Robert Martin the person depicted in the bank surveillance photographs)?”. There is no doubt that Martin looks like the person in those photographs. However, after repeatedly comparing those photographs with Martin’s arrest photo, I continue to wonder if Martin is the person in those photographs. It is obvious from this record that the jury was similarly troubled.
The majority characterizes the evidence of Martin’s guilt as “significant,” “strong,” and “overwhelming.” Maj. Op. at 394-95, 398, and 399-400. However, those characterizations are undermined by the apparent difficulty the jury had in concluding that Martin was the robber. Although there were three eyewitnesses identifications, and numerous bank surveillance photographs, there was no corroborating physical evidence. Moreover, the quality of the photographs and the strength of the eyewitness identifications is a far cry from the “strong,” “significant,” or “overwhelming” evidence of guilt that the majority *402claims. The majority’s characterization of the evidence ignores the poor quality of the bank surveillance photographs, as well as the dangers endemic to eyewitness identifications. As Justice Brennan explained four decades ago:
[^identification evidence is peculiarly riddled with innumerable dangers and variable factors which might seriously, even crucially, derogate from a fair trial. The vagaries of eyewitness identification are well-known; the annals of criminal law are rife with instances of mistaken identification. Mr. Justice Frankfurter once said: ‘What is the worth of identification testimony even when uncontradicted? The identification of strangers is proverbially untrustworthy. The hazards of such testimony are established by a formidable number of instances in the records of English and American trials.’
United States v. Wade, 388 U.S. 218, 228, 87 S.Ct. 1926, 18 L.Ed.2d 1149 (1967) (quoting Felix Frankfurter, The Case of Sacco and Vanzetti 30 (1927)). This robbery lasted less than one minute, and the robber’s face was obscured for much of that brief interval. Moreover, common sense suggests that the tellers were distracted by the presence of a sawed-off shotgun.
My colleagues stress that over eighty surveillance photographs were entered into evidence. See Maj. Op. at 393. However, the vast majority of those photographs are utterly useless. Twenty-five of them either do not show the robber at all, or show him only faintly in the distance with no discernable features except for his sweatshirt and cap. Seven of the remaining photographs show only the security guard running from the bank. Five of the photographs do show the front of the robber from recognizable range, but his face is turned from the camera and/or obscured by his cap. Three clearly show him from the front, but too far from the camera to clearly show the robber’s face. Only two of the photographs show the robber’s face with some clarity, and of those, only one is sharp and clear. I assume that is the single picture my colleagues refer to as “a surveillance photograph of the robber leaving the bank ...”. Maj. Op. at 393 (emphasis added). That single photograph is “clear and definite.” See Maj. Op. at 399. It is also inconclusive.
Thus, the government’s evidence against Martin consisted of his appearance at trial, a single surveillance photograph, the testimony of three bank employees, and the testimony of Detective Seifert, the arresting officer.
On cross-examination, defense counsel highlighted the discrepancies in the bank employees’ descriptions of the robber. One teller described the robber as a Black male with “scraggly” facial hair, “thin”— weighing as little as 130 pounds, a little taller than 5'6", and wearing a blue, zip-front jacket and a green cap; another teller described the robber as a brown-complexioned male with a moustache and scraggly beard, weighing as much as 180 pounds, around 5'8" tall, and wearing a blue or black baseball cap; and the third bank employee described the robber as a Black male, of “medium height” and “medium build,” about thirty to forty years old, and wearing a cap. Martin certainly does fit those general descriptions; so do many other medium-complexioned Black males with some facial hair who are of average build, average weight, and average height. By definition, most people are of “average” proportion.
Against this backdrop, Detective Seifert testified on direct examination as follows:
Q: Now, after the robbery itself, did you receive anything connected with the robbery?
A: Yes, I did.
Q: What did you receive?
*403A: I received a photograph, several photographs from the—from Detective Joseph Passio of our major crimes unit, who works with the FBI, and wanted a poster.
Q: Now, I’m showing you what’s been marked as Government Exhibit 6, and ask you—I’ll ask you whether this is similar to the photographs that you received from Detective Passio.
A: Yes, it is....
Q: Now, directing your attention to approximately March 25, 1998, were you on duty that day?
A: Yes, I was.
Q: By that time, had you seen this photograph once or more than once?
A: More than once....
Q: And were you familiar with that photograph by then?
A: Yes, I was.
Q: On that day, did you arrest anybody?
A: Yes, I did.
Q: Who did you arrest?
A: I arrested the defendant, Mr. Martin.
Q: Where did you arrest him?
A: Inside the barbershop at 2125 Ridge Avenue in Philadelphia.
Q: And had you gone there hoping that he would be there?
A: Yes.
Q: Why were you looking for a person matching this photograph?
A: To arrest him for bank robbery....
Q: When you went into the barbershop, was there more than one person inside or only one person?
A: There was more than one.
Q: And you arrested this defendant?
A: Yes, I did.
Q: Again, what did you arrest him for, as opposed to somebody else in the barbershop?
A: I recognized him as being the person in this photograph, which is a surveillance photo taken the day of the robbery of the bank robber.
Supp. app. at 157-58. On cross-examination, Martin’s counsel elicited the following from Detective Seifert:
Q: Okay. And you received information that day, as you indicated by indicating you had the name Rob from a woman named Edna Cook, is that correct?
A: That is correct, sir.
Q: Okay. And Edna Cook had seen that photograph that you have or a copy of that photograph that you have, and had indicated to you that she thought that was Rob, is that fair to say?
A: That’s correct.
Q: And so armed with that information, you went to a location where Edna Cook said that this person Rob was, my client, and went in there and you found him inside, and today you’re indicating that you think that the individual pictured in that photograph is my client, Rob Martin, is that fair to say?
A: It is him.
Q: Well, it is him? How do you—did you—do you have fingerprints—
A: Well, it looks like him.
Q: —from the bank?
A: It looks like him.
Q: It looks like him. Thank you.
Supp. app. at 160-61. Thus, defense counsel asked questions that not only allowed reinforced Seifert’s identification of Martin, it also elicited the out-of-court identification of Edna Cook.
*404In his case, Martin called Richard Vorder Bruegge, an expert in the field of photographic identification. Bruegge testified that he received copies of the bank surveillance photos, video footage from the bank’s surveillance tape, and pictures of Martin in order to compare Martin’s face to the robber. Bruegge explained that, when attempting to make an identification from photographs he compares scars, freckle patterns, the details of a person’s ear pattern, shape of the eyes, shape of the nose, creases and lines on the person’s face, blemishes on the face, and overall ear shape. He can reach any one of three conclusions: he can positively identify the suspect, eliminate the suspect, or not be able to reach any conclusion. After comparing photographs of Martin to the bank surveillance photographs, he was not able to tell if Martin is the person depicted in the surveillance photographs.7
During trial, the district court granted the government’s request to have Martin remove his eyeglasses, put on a cap, walk to the lectern in front of the jury and turn around. The baseball cap he put on was embossed with a Philadelphia Eagles logo, as was the hat worn by the robber.
The court told the jury that the cap “ha[d] been purchased as a prop by the [gjovernment for purposes of this case.” Supp. app. vol. II, at 254. The government had two caps available at trial: one bearing a logo of the Philadelphia Eagles, the other with no logo. It is significant that none of the bank employees who testified said that the robber’s cap had any logo on it. The government nevertheless pressed the court to require Martin to wear the cap with the Eagles logo, arguing: “it will be beneficial to the jury to see [Martin] in something that is as close as possible to what the bank robber wore.” Id. at 252. However, if the jurors were actually comparing Martin’s face to the face in the surveillance photos, having Martin wear the Eagles cap as opposed to a similar cap with no logo was certainly not “beneficial to the jury.” It was, however, very beneficial to the government’s chances of having the jurors conclude that Martin was the person in those photographs.
Displaying the defendant to the jury in this manner is, of course, normally permissible. See Holt v. United States, 218 U.S. 245, 252, 31 S.Ct. 2, 54 L.Ed. 1021 (1910). However, the procedure was unduly suggestive under the circumstances here. Moreover, if the evidence from the surveillance photographs and identifying witnesses was as strong as the government and my colleagues suggest, this demonstration would not have been necessary. Yet, it was not only necessary, it had to be repeated.
During deliberations, the jury asked to see Martin once again wearing the cap. The court granted the jury’s request and instructed Martin to put the cap on, and pull it down close to his eyes so that he would appear before the jury wearing the cap the same way the robber is depicted wearing an identical cap in the surveillance photographs.
The jurors had two days during the course of this trial to observe Martin. They had every opportunity to compare his face with that of the purportedly “clear” and “definite” images, in the numerous bank surveillance photographs. Yet, at perhaps the most important moment of the trial, Martin had to put on a *405cap identical to the robber’s and pull it down close to his eyes before the jury could agree on an identification. Not only did this increase the likelihood of a misidentification based upon the similarity of the cap and general appearance, it obscured Martin’s face.
II.
In order to establish ineffective assistance of counsel, Martin must establish that counsel’s efforts fell below an objective standard of reasonableness, and that the dereliction prejudiced him. Strickland v. Washington, 466 U.S. 668, 687, 104 S.Ct. 2052, 80 L.Ed.2d 674 (1984). We presume, however, that “counsel’s conduct might have been part of a sound strategy.” Thomas v. Varner, 428 F.3d 491, 500 (3d Cir.2005). The presumption is rebutted by “showing either that the conduct was not, in fact, part of a strategy or by showing that the strategy employed was not sound.” Id. Where, as here, “the record does not explicitly disclose trial counsel’s actual strategy or lack thereof ... the presumption [of sound strategy] may only be rebutted through a showing that no sound strategy posited by the [government] could have supported the conduct.”8 Id.
A.
Martin first argues that trial counsel was ineffective for failing to file a motion in, limine to preclude Detective Seifert from offering a lay opinion pursuant to Fed.R.Evid. 701. The majority agrees with the government’s claim that Detective Seifert only testified to show the circumstances leading up to Maxiin’s arrest, and they therefore reject Martin’s allegation of ineffectiveness because background evidence is usually admissible. According to the government, absent Seifert’s explanation of why Martin was arrested, the jury might have concluded that it had not been told the whole story. Although I would normally agree with this contention, I can not agree here because the disputed testimony was not offered for background. It was offered to show that Detective Seifert identified Martin as the robber. In fact, the government undermines its own argument by its subtle concession that Detective Seifex't’s testimony “was not the most probative aspect of [its] proof.” Gov’t Br. 30 n. 3 (emphasis added). Evidence offered only for background would hardly constitute a “probative aspect of [the government’s] proof.”
Moreover, even if the government only intended the testimony as background, the jury had no way of knowing of that limitation. Rather than giving an appropriate limiting instx-uction, the court told the juiy: “And you can consider that also, there was other testimony'—other identification testimony other than the eye witnesses, of course. There was the—particularly, the testimony of the arresting offieer[ ].” Supp. app. at 338 (emphasis added). Thus, any attempt to dismiss Seifert’s testimony as mere “backgx'ound” ignores the record.
Furthermore, insofar as background evidence was appropriate at all, Detective Seifert merely had to explain that she went to the barbershop where Martin worked because she had information that someone who worked there resembled the person depicted in the surveillance photos. That would not only have addressed the government’s concerns about apparent gaps in its case, it would have had the additional virtue of being a more accurate account of what happened. It would have done that without allowing Seifert to offer her opinion that Marlin is the person in *406the surveillance photos. This would also have avoided the taint of allowing Detective Seifert to improperly offer her lay opinion that Martin was the person in the surveillance photographs.9
B.
Fed.R.Evid. 701 provides:
If the witness is not testifying as an expert, the witness’ testimony in the form of opinions or inferences is limited to those opinions or inferences which are (a) rationally based on the perception of the witness, (b) helpful to a clear understanding of the witness’ testimony or the determination of a fact in issue, and (c) not based on scientific, technical, or other specialized knowledge within the scope of Rule 702.
Fed.R.Evid. 701.
Although Detective Seifert’s opinion that the surveillance photographs are pictures of Martin was rationally based on her perceptions, it was anything but “helpful.”10 Detective Seifert was no more qualified to compare Martin to the individual in the surveillance photographs than the jurors. “[Wjherever inferences and conclusions can be drawn by the jury as well as the witness, the witness is superfluous; ... a lay opinion is received because and whenever facts cannot be so told as to make the jury as able as [the witness] to draw the inference.” 7 Wigmore on Evidence § 1917.8, at 10 (Chadbourne rev. 1978).
In United States v. LaPierre, 998 F.2d 1460 (9th Cir.1993), an investigating officer informed the jury of his opinion that La-Pierre was the person depicted in several bank surveillance photographs. On appeal, the Court of Appeals reversed the conviction in part because the investigating officer’s testimony was not helpful and therefore inadmissible under Rule 701. “The jury ... was able to view the surveillance photos ... and make an independent determination whether it believed that the individual pictured in the photos was in fact [the defendant].” Id. at 1465. Thus, the officer’s opinion testimony “ran the risk of invading the province of the jury and unfairly prejudicing [the defendant].” Id.
The court concluded that Rule 701 limits a lay witness’s opinion regarding identification to situations where “the witness has had substantial and sustained contact with the person in the photograph” or “the defendant’s appearance in the photograph is different from his appearance before the jury and the witness is familiar with the defendant as he appears in the photograph.” Id. In “both [situations] there is reason to believe that the witness is more likely to correctly identify the person than is the jury.” Id. Thus, absent circumstances not present here, “[l]ay opinion testimony of the type given by [Seifert] is of dubious value.” Id.
My colleagues attempt to dismiss Martin’s reliance on LaPierre, by suggesting that “the context was different,” and stressing that there is no “ironclad rule against identification testimony from an arresting officer.” Maj. Op. at 397. I agree that Rule 701 does not create an ironclad rule against a police officer offering identification testimony. However, dismissing LaPierre on that basis ignores the court’s analysis.11 The essence of La-*407Pierre’s holding is that no lay witness, may offer an opinion identifying a defendant absent prior contact with the defendant, or other circumstances that suggest that the witness has a better basis to identify the defendant than the jury. There is no reason to create an exception for the lay opinion of police officers, and my colleagues offer none.
Indeed, given the danger that jurors may give a police officer’s uninformed identification more weight than one based “only” on their own untrained eyes, it is particularly important to adhere to the limitations of Rule 701 when the lay opinion of identification is offered by a police officer. Lay opinion testimony identifying a defendant from surveillance photos is thus only admissible “if there is some basis for concluding that the witness is more likely to correctly identify the defendant from the photograph than is the jury.” United States v. Towns, 913 F.2d 434, 445 (7th Cir.1990) (quoting United States v. Farnsworth, 729 F.2d 1158, 1160 (8th Cir. 1984)); see also United States v. Jackman, 48 F.3d 1, 4-5 (1st Cir.1995) (collecting cases and finding such testimony admissible “at least when the witness possesses sufficiently relevant familiarity with the defendant that the jury cannot also possess, and when the photographs are not either so unmistakably clear or so hopelessly obscure that the witness is no better-suited than the jury to make the identification.”). Here, Bruegge was the only expert qualified to give an opinion about Martin’s identification. He was not able to identify Martin, despite his empirical approach and scientific methodology.
C.
The majority also concludes that defense counsel was not ineffective for eliciting testimony about the role Cook played in Martin’s arrest because eliciting this information could have been part of counsel’s trial strategy. The government contends that defense counsel asked Detective Seifert about Cook to establish that the detective was not able to identify Martin without Cook’s help and thereby undermine Seifert’s identification of Martin. That post hoc explanation strains credulity. Had Seifert never been allowed to say that she thought Martin was depicted in the surveillance photographs, it would not have been “necessary” to use Cook to try to limit the impact of Seifert’s identification.
Rather than appropriately chipping away at the layers of identification, counsel added two additional layers himself by eliciting Seifert’s opinion and Cook’s out-of-court identification.
I am at a loss to understand the soundness of any defense strategy that would allow a police detective to impermissibly identify the defendant, and then elicit an out-of-court identification from the only person who knew Martin personally to undermine the detective’s inadmissible identification.12
The bank employees’ identifications were based on observations during a highly stressful robbery that lasted less than one minute, and afforded an opportunity to observe the robber’s face that was substantially less than one minute. Moreover, common sense suggests that fear and the *408menacing presence of a sawed-off shotgun were formidable obstacles to the tellers’ ability to focus on the robber’s face during the few seconds that it was not obscured. That distraction and the minimal opportunity to observe explain why none of the tellers noticed the large contrasting eagle prominently embroidered on the front of the robber’s cap; a logo that is exceedingly difficult to miss even if one does not follow professional football. Thus, it is not surprising that the jurors insisted on comparing Martin to the surveillance photographs themselves rather than rely on the testimony of the bank tellers.
Cook is different. Cook knew Martin, and her out-of-court identification was based on her casual observation of the surveillance photos. Thus, by allowing Detective Seifert to tell the jury that Cook had identified Martin from those photos, counsel’s cross-examination of Seifert fortified the inherently unreliable identifications of the bank tellers. Accordingly, defense counsel’s handling of Detective Seifert was not only ineffective, it was foolhardy.
D.
Of course, counsel’s mistakes do not automatically result in relief under Strickland. Rather, Martin must show “that there is a reasonable probability that, but for counsel’s unprofessional errors, the result of the proceeding would have been different.” Strickland, 466 U.S. at 694, 104 S.Ct. 2052. However, Martin “need not show that counsel’s deficient performance ‘more likely than not altered the outcome in the case’—rather, he must show only ‘a probability sufficient to undermine confidence in the outcome.’ ” Jacobs v. Horn, 395 F.3d 92, 105 (3d Cir.2005) (quoting Strickland, 466 U.S. at 693-94, 104 S.Ct. 2052). That standard “is not a stringent one.” Jacobs, 395 F.3d at 105.
My colleagues conclude that, given the evidence here, Martin would have been convicted even without Detective Seifert’s testimony on direct and cross-examination. See Maj. Op. at 398. However, the remaining evidence was not sufficiently “clear and definite,” to convince the jury of his guilt beyond a reasonable doubt after observing him during two days of testimony. Rather, despite the purportedly clear and definite surveillance photographs, the tellers’ identifications, and two days of observation, the jury still had to interrupt deliberations to have Martin don the baseball cap in the manner of the robber for a second time. The dangers inherent in this procedure here should be obvious.
We clearly do not know what motivated the jurors to ask to see Martin with the cap a second time. Perhaps they concluded that Martin was the robber after observing him during the trial and comparing him to the surveillance photographs and merely wanted to assuage a lingering doubt of one of the jurors. Perhaps seeing Martin the second time only overcame substantial (and “reasonable”) doubts when combined with Detective Seifert’s lay opinion and Cook’s out-of-court identification. We are left wondering.
Strickland does not require Martin to show that “but for” his attorney’s error, he would not have been convicted. Given the sanctity of jury deliberations, that test could never be satisfied. Rather, Strickland only requires that Martin establish a reasonable likelihood that he would not have been convicted without his attorney’s deficient performance. I think he has satisfied that burden.
III.
Despite my colleagues’ characterization of the evidence, this remains a very difficult case. As I said at the outset, I have stared at the only useful surveillance photo *409and compared it to Martin’s arrest photo over and over and over again. I can understand the jury’s difficulty. As I noted earlier, there is clearly a resemblance. That resemblance has caused me alternating degrees of doubt and certainty as I have compared the only useful surveillance photograph to Martin’s arrest photograph. I have not been able to consistently conclude that the surveillance photograph is a picture of Martin, nor have I been able to consistently conclude that it is not a picture of him. Of course, I did not have the advantage of actually seeing Martin in person during two days of trial. However, that distinction would have been more convincing if the trial court had not made Martin put on a cap identical to the robber’s and pull it down over his face during deliberations, and if the jury had not found it necessary to repeat the procedure.
I realize this precise point isn’t argued here, but given the issues that are argued on appeal and the dangers inherent in identification testimony, the jury’s apparent expression of doubt despite their opportunity to compare Martin to the surveillance photos is relevant to the prejudice prong of Martin’s Strickland claim.
IV.
The majority emphasizes that the security guard’s identification of Martin “was itself a basis for the jury’s verdict.” Maj. Op at 399. It clearly could have been. However, as I have explained, it is obvious from this record that the jury did not find that testimony as convincing as my colleagues do. My colleagues suggest that my concerns about the jury’s apparent hesitancy and doubts are “speculative.” Maj. Op. at 399. I admit that my analysis involves some degree of speculation. However, I submit that, since jury deliberations are necessarily shrouded in secrecy, some speculation is endemic in any inquiry under the prejudice prong of Strickland. Given this record and counsel’s deficient performance, the speculation is not only warranted, it is compelled. Moreover, for reasons I have already explained, I fear that the verdict is also speculative and that the jury’s speculation was resolved by the out-of-court identification of Edna Cook, and the opinion of Detective Seifert that defense counsel should have excluded under Rule 701.13
I hope that I am wrong and that my colleagues are correct. I hope that Martin was convicted because he looks so much like the robber that the jurors were convinced beyond a reasonable doubt that he is the robber; and that he was not convicted because he resembles the robber when weai’ing an identical cap pulled down to obscure his face. However, because the jury no doubt considered Edna Cook’s out-of-court identification, as well as Seifert’s opinion, this verdict inspires more speculation than confidence. Since that evidence was only admitted because of defense counsel’s ineffectiveness, I must respectfully dissent.
. Bruegge stated on cross-examination that there must be overwhelming similarity to make a positive identification. Here, he said that after comparing the pictures of Martin to the surveillance-video footage and photos taken from the bank he was unable to conclude that Martin was the bank robber depicted in the surveillance images. I concede, however, that Bruegge also stated on cross-examination that he was close to reaching that conclusion.
. During the hearing of Martin's habeas petition, Martin did not ask counsel why he failed to file a motion in limine to limit Detective Seifert's testimony or why he elicited the testimony regarding Cook's out-of-court identification.
. Indeed, the type of background provided by Detective Seifert’s testimony is commonly stipulated to precisely because there is no dispute about how the defendant came to be arrested.
. The third prong of the test for admissibility under Rule 701, that the testimony not be "based on scientific, technical, or other specialized knowledge within the scope of Rule 702,” is clearly satisfied in this case. Fed. R.Evid. 701.
. The majority states that "[pjolice officers may complete the story of the charged offense, absent prejudice or inadmissibility.” *407Maj. Op. at 407. That testimony is certainly relevant during a suppression hearing for purposes of establishing probable cause to arrest, and that is the context the majority refers to. However, it is not permissible under Rule 701 for police officers to offer identification testimony by stuffing it into the Trojan horse of background testimony.
. The soundness of this strategy must also have escaped defense counsel because he never argued that Detective Seifert’s testimony was undermined because she could not identify Martin without Edna Cook’s help.
. See United States v. Brownlee, 454 F.3d 131, 142 (3d Cir.2006) ("|M]istaken eyewitness identifications are responsible for more wrongful convictions than all other causes combined.”) (Quoting A. Daniel Yarmey, Expert Testimony: Does Eyewitness Memoty Research Have Probative Value for the Courts?, 42 Canadian Psychology 92, 93 (May 2001)). That conclusion is based on a study of 328 defendants who were ultimately exonerated based on subsequently available DNA testing. The study concluded that " '[E]yewitness evidence presented from well meaning and confident citizens is highly persuasive but, at the same time, is among the least reliable fomts of evidence.’" Brownlee, 454 F.3d at 142 (brackets and italics in original).