In 1982, an Illinois jury convicted eighteen-year-old Patrick Hampton of deviate sexual assault, attempted rape, robbery, and aggravated battery, and the trial judge ordered him to serve an extended prison term of sixty years. Hampton filed a petition for a writ of habeas corpus pursuant to 28 U.S.C. § 2254, alleging (among other things) that his trial counsel was constitutionally ineffective for failing to investigate and interview exculpatory eyewitnesses to the crimes of which he was convicted and for making promises in his opening statement to the jury that he did not keep. Following an evidentiary hearing, the district court granted the writ on these grounds. Hampton v. Leibach, 2001 WL 1518533 (N.D.Ill. Nov.29, 2001). The State has appealed. By order of the United States Supreme Court, Hampton has been released from prison during the pen-dency of this appeal. We affirm the district court’s judgment.
I.
The offenses of which Hampton was convicted took place at a rhythm and blues concert held at the Chicago International Amphitheatre on the evening of December 29, 1981. Four bands performed at the concert: Chocolate Milk, Slave, Michael Henderson, and Zapp. Shortly after midnight, while the last band was still playing, a group of up to forty individuals marched up the aisle toward the stage, chanting “Black Gangster Disciples” and “Third World Disciple Nation,” pounding their *222fists together, and making gang signals with their hands. Three Latino concertgoers — Hugo N., Martha N., and Denise M.1 — were seated in the fifth row of the theater. As they arose from their seats and attempted to leave, the group in the aisle attacked them, removing their clothes, taking their wallets and jewelry, beating them, and sexually assaulting the two women. Security guards eventually intervened and rescued the three victims of the assault. None of the perpetrators was detained at the scene.
Fourteen year-old Keith Powell attended the concert and witnessed the attack. He later identified a number of former acquaintances from the Robert Taylor Homes (a public housing project) as having been in the group of people who had marched toward the stage of the theater. Hampton was among the individuals that Powell identified, although Powell would later testify that he did not actually see Hampton participate in the attack on the three Latinos.
Hampton was arrested on December 31, 1981. He was eighteen years old at that time and had never before been arrested.
Ultimately, nine individuals, including Hampton, were charged with the attacks. Six of them pleaded guilty and were sentenced to the short periods of time they had already spent in jail awaiting trial. Three defendants — Hampton, Ronald Mallory, and Ricky Knight — pleaded not guilty. They were tried jointly before three separate juries.
Attorney Jack Rodgon represented Hampton at the trial. Hampton’s family had retained Rodgon, who previously had represented Hampton’s brother. In advance of trial, Rodgon sought to withdraw as Hampton’s counsel, asserting that his fees were not being paid and that Hampton and his family were not cooperating with him in preparation of the case. The trial judge, Hon. Earl E. Strayhorn, proposed to solve the problem by appointing Rodgon. Rodgon demurred, indicating there were “some problems” with representing Hampton. R. 48-1 at 129. The judge was unmoved and refused to release Rodgon from the engagement; he subsequently granted Rodgon’s motion to continue as Hampton’s counsel by appointment.
At trial, Powell testified that near the end of the concert, a group of men approached the Amphitheatre stage, making gang-related signs with their hands and chanting gang slogans. The three defendants were members of that group. Powell had known the defendants for two to three years; he had once lived in the Robert Taylor Homes where the defendants resided. Powell saw a disturbance break out near the stage. At some point during the melee, a naked woman ran up the aisle; he also saw Knight throw a pair of pants in the air. Although the stage lights were lit during this incident, the rest of the lights in the Amphitheatre were darkened. Following the concert, Powell took the 43rd Street bus back to the Robert Taylor Homes near 43rd and State Streets, where he was staying with his aunt. He saw each of the three defendants (among other individuals that he knew) on that bus. He also overheard a conversation in which someone — he could not say who — bragged about having “stuffed his fingers” into the vagina of a woman, R. 48-2 at 563, and having taken jewelry. Powell reported what he had seen and heard to the police on the after*223noon of December 31. Powell testified that although Hampton was in the group that approached the stage, he had not seen Hampton attack anyone, nor had he heard Hampton say anything on the bus. Powell also testified that he had picked Hampton out of a line-up, and he was positive on this point. R. 48-2 at 624. However, the trial evidence would subsequently reveal that he had never picked Hampton out of a line-up. R. 48-3 at 1087.
Hugo N., one of the three victims of the assault, testified that he had attended the concert with his girlfriend, Denise M., his sister, Martha N., and Martha’s boyfriend, Scott S. They sat in the fifth row on the main floor of the Amphitheatre. A number of disturbances preceded the assault in which he, Denise, and Martha were injured: In the intermission following the second act, Martha’s boyfriend was struck in the head with a crowbar; during the third act, Hugo saw security personnel chasing someone through the theater; and during the intermission between the third and final acts, he saw another fight break out among eoneertgoers. During the last act, when Hugo, Denise, and Martha saw the group of men moving down the aisle toward the stage chanting “Third World Disciple Nation” and making hand signals, they decided to leave. (Scott S. had already left due to his injury.) As they attempted to do so, however, the group blocked their exit and attacked them. Hugo lost sight of Martha. As Hugo attempted to shield Denise from the group, they were both kicked and punched. His shirt was torn off, he felt people going through his pockets (his wallet, keys, and jewelry all were taken) and then the rest of his clothes were torn from his body. Denise was also being stripped of her clothes. Knight twice struck him and Denise with a chair, and on the second blow they both fell to the floor. At this point, a number of attackers had exposed their penises, and Hugo saw Knight put his in Denise’s mouth, telling her “something like ‘here, take it.’ ” R. 48-2 at 766. A security guard eventually came to his aid; another helped Denise. They were taken to a first aid station and subsequently by ambulance to a hospital. Hugo later identified Knight from photographs and from a line-up. He did not identify Hampton as one of his attackers. Nor did he give the police a description of his assailants prior to the first line-up that he viewed. Hugo agreed that at the time of his assault, the last band was still playing, the air in the Amphitheatre was smoke-filled, and the only light was coming from the multi-colored stage lights.
Denise M. testified that when she, Hugo, and Martha attempted to leave the theater, their path was blocked by a group of fifteen to twenty men who were chanting and making hand signals. She and Hugo were pushed and shoved, and she felt hands grabbing her. Her clothing and jewelry were ripped from her person, and she felt fingers being pushed into her vagina. After someone twice struck them with a chair, she fell dazed to the ground along with Hugo. A number of the men had their penises out, and two of them approached her. Knight attempted to put his in her mouth (saying “here, take it,” R. 48-3 at 1017) as did Mallory. Several men sat on her legs and one tried to pull them apart. Denise saw one of the attackers, whom she identified as Hampton, move his hands and she felt him try to force a cold, hard object into her vagina. Eventually, a security guard arrived and managed to break up the assault. Denise suffered a number of injuries from the attack, including a tear in her vagina that required surgical repair, a bladder infection, and scarring on her breast. From photographs and from a line-up, she subsequently identified Hampton as the individual who had attempted to *224force an object into her vagina. On cross-examination, Denise testified that she could not estimate how long she had seen this individual. She indicated that there were many men who were attacking her, and that she was concentrating on the faces of the men who tried to put their penises in her mouth.2 Prior to viewing photographs and a line-up, she did not give a description of Hampton or her other assailants to the police.3
Martha N. testified that the assault began as a group of men began to march toward the stage, making signals at another group that was standing in front of the stage. As she looked over at the aisle, she saw Knight gathering men around him as they proceeded down the aisle. When she, Hugo, and Denise attempted to move into the aisle in order to leave, she found herself surrounded by a group of up to thirty or thirty-five men. Knight swung a chair at her and she ducked to avoid the blow. She then felt others pulling her by the hair toward the stage. She was punched and kicked, and her jewelry was taken. A man that she identified as Hampton tore her pants and attempted (unsuccessfully) to put his hand into her vagina. When she tried to get away, she heard Knight say, “Get her. Get her.” R. 48-2 at 856. She managed to run to a security guard and asked for his help, but he did nothing. She then located a second guard who did help her. Martha testified that she subsequently picked Hampton out of line-up that took place on December 31. She acknowledged that she had only seen his face for four to five seconds on the night of the assault and that during that time she was kicking and flailing at the people who were attacking her. Prior to viewing the December 31st line-up, she did not describe Hampton or her other assailants to the police beyond saying that they were young black males. R. 48-3 at 949-50.
Detective Thomas Ptak later acknowledged on cross-examination that according to a written investigative report that he had helped to prepare, Martha N. had identified Ezra Garner, not Hampton, as the man who had tried to put his hand into her vagina. R. 48-3 at 984. Ptak testified that the report was erroneous, and that Martha had actually identified Hampton.
William Heinrichs, a field supervisor for the Cook County Sheriffs office, was moonlighting as a security guard at the concert. Just after midnight, a young Latina in a torn blouse and pants (Martha N.) approached him and told him that a fight was taking place and that her brother and his girlfriend were being attacked. He left her in thé custody of another guard, then rounded up ádditional other guards to assist him. As he and other guards approached the scene of the attack, he saw a large group of black men gathered in a circle. In the center of that circle, a man and a woman were on the ground being beaten. As Heinrichs pushed his way through to the center of the circle, he saw a young black man that he later identified as Hampton bending over the woman, thrusting his arm toward her vaginal area. One of the other guards yanked him off *225the woman, and the guards helped the two victims — both of them naked and bruised — to safety. Heinrichs testified that he grabbed the assailant he identified as Hampton by the shoulder, but was unable to keep him and the others from fleeing. On the witness stand, Heinrichs indicated that he saw Hampton for no more than three to four seconds during the assault; he also testified that he had searched Hampton earlier in the evening, when Hampton was admitted to the theater. Following the incident, Heinrichs made no effort to contact the police for more than a week; ultimately, someone from the police department contacted him on January 6. Later that day, Heinrichs identified Hampton and three other individuals as participants in the attacks from a photograph of a line-up. On the previous day, Heinrichs had seen a television news report about the attack. That report featured a picture of Hampton (who by then was in police custody), and in Hein-richs’ estimation, the photograph of Hampton had been displayed on the air for as long as one minute.
Ricky Knight called no witnesses in his defense. He offered only a single stipulation to the effect that a detective would testify that the police investigation revealed it was Ronald Mallory, and not Knight, who had put his penis in Denise’s mouth.
Ronald Mallory presented four witnesses besides himself. A woman who had grown up in the same neighborhood as he testified that to her knowledge he was not a gang member. A second witness, who knew Mallory from the projects and considered him a friend, testified that she had attended the concert, that she had seen the attacks, that she was standing right next to Mallory while the attacks took place, that Mallory had not participated in the attacks, and that he was not, to her knowledge, a gang member. Gregory Hubbard testified that he had known Mallory for five to six years from the neighborhood, that he too was present at the concert, that he saw Mallory during the attacks, that Mallory had not taken part in the attacks, and that Mallory was not a gang member. Gregory Mallory (“Gregory” or “Gregory Mallory”), Ronald’s brother, testified that he too had attended the concert, that he had seen the attacks, that he saw where his brother was during the attacks, and that Mallory was not one of the attackers. Gregory acknowledged that he had a previous conviction for an unspecified crime. Finally, Mallory himself testified, acknowledging that he was present at the concert but denying any involvement in the attacks. Mallory said that he approached the police on his own and gave them a statement after hearing that they were looking for him. He conceded, however, that he had lied to the authorities when he told them he had not seen anyone he knew (other than his brother) at the concert.
Two of Mallory’s witnesses had favorable things to say about Hampton. The government asked Hubbard on cross-examination whether he had seen anyone he recognized participating in the attacks. Hubbard said that he had not, and on further questioning, testified that he had not seen either Knight, Hampton, or Mallory take part in the attacks. R. 48-3 at 1145. Gregory Mallory, also on cross-examination, denied that either Mallory or Hampton was a member of the Disciples gang. R. 48-3 at 1107.
However, because each defendant’s case was heard by a separate jury, Hampton’s jury did not hear this testimony. When the State was putting on its own witnesses, all three juries were present in the courtroom at the same time. But as counsel for each defendant took turns cross-examining *226the State’s witnesses, and later as each defendant put on his own evidence, the juries were rotated in and out of the courtroom such that each jury only heard one defendant’s case. Hampton’s jury thus was not present in the courtroom when Mallory’s witnesses testified.
Hampton’s defense case was limited to one witness. Detective Craig Cegielski testified that Powell, to his knowledge, had never picked Hampton out of a line-up. R. 48-3 at 1087.
In his opening statement, Hampton’s attorney, Rodgon, made two promises: first, that Hampton would testify that he was present at the concert and had seen what happened but had not participated in the attack (R. 48-2 at 543), and second, that the evidence would show that Hampton was neither a member of, nor involved with, any gang (id. at 544). Neither promise was kept. Hampton did not testify, and his jury heard no evidence that he had lacked involvement with a gang.
The theme that Rodgon sounded in closing argument was one that he focused on exclusively throughout the trial — the weakness of the government’s case against Hampton. Rodgon challenged both Powell’s credibility and Heinrichs’: Rodgon noted that Powell claimed to have identified Hampton in a line-up but that the evidence revealed he had not done so; and Heinrichs, despite his position as a law enforcement officer, had not come forward as a witness until the police contacted him more than a week after the assaults, and he had identified Hampton only after he had seen a news report featuring a photograph of Hampton. Rodgon also questioned the ability of Martha N. and Denise M. to identify Hampton as one of their attackers, noting that they had only gotten brief glimpses of the assailant under stressful conditions. Rodgon explained neither Hampton’s failure to testify nor the lack of evidence that Hampton was uninvolved in a gang; he simply noted that Hampton was not obliged to put on a case.
During jury deliberations, Hampton’s jury sent four notes to the judge. One of these indicated that the jury had arrived at a verdict as to five of the nine charges against Hampton but was deadlocked as to the four other charges. R. 62 at 189. Ultimately, the jury acquitted Hampton of the attempted rape of Martha N. (R. 62 at 193) but convicted him of deviate sexual assault, the attempted rape of Denise, aggravated battery, and robbery (R. 62 at 194-201). Knight was convicted on all charges. Mallory was acquitted of certain charges and granted a mistrial as to the remaining charges, as to which his jury could not render a verdict. He was acquitted of the remaining charges at a second trial.
Judge Strayhorn ordered Hampton to serve a prison term of sixty years. R. 62 at 214; R. 48-6 at 58. The most serious of his offenses, deviate sexual assault, was normally punishable by a maximum of thirty years. However, that maximum was doubled to sixty years upon a finding that the assault was accompanied by exceptionally brutal or heinous behavior indicative of wanton cruelty. Judge Stray-horn determined that Hampton’s use of a foreign object to assault Denise M. met this condition. He remarked:
It is probably the most ag[g]rievously cruel, wanton, depraved, brutal, [heinous], animalistic activity that I have run into in some thirty five years in the practice of law as a prosecutor, as a defense counsel and as a Judge. I’ve never run into a situation involving the [heinous] type facts that evolved in this case and that were shown to exist in this case and Mr. Hampton was actively involved to take a foreign object and to attempt to force it into the vagina of a *227woman being held prostrate on the ground, people on each arm and on each leg. That’s animalistic. That’s depraved. That’s cruel. That’s doing great harm and I can’t allow it to pass and I will not allow it to pass unnoticed.
R. 48-8 at 37. He was not imposing the sentence for the purpose of rehabilitation, Judge Strayhorn explained. “It is purely and simply for punishment for ... the most cruel, wanton, brutal, [heinous,] depraved, animalistic act that I have ever seen committed on a human being.” R. 48-8 at 38. The judge ordered Hampton to serve concurrent, lesser terms on the other charges. R. 62 at 214; R. 48-6 at 58.
On direct review, the Illinois appellate court affirmed Hampton’s conviction and sentence. People v. Knight, 139 Ill.App.3d 188, 93 Ill.Dec. 521, 486 N.E.2d 1356 (1985). The Illinois Supreme Court, as well as the United States Supreme Court, both declined to hear the case. Knight v. Illinois, 480 U.S. 905, 107 S.Ct. 1346, 94 L.Ed.2d 518 (1987).
In 1990, Hampton filed a petition for postconviction relief in state court. Through counsel, he filed a supplemental petition after the Cook County public defender’s office was appointed to represent him. R. 48-5 at 108 et seq. In his supplemental petition, Hampton asserted that Rodgon had been ineffective for failing to investigate and interview exculpatory occurrence and character witnesses (R. 48-5 at 112-13, 117-19), for failing to have Hampton testify in his own defense, and for failing to present evidence that Hampton was not a gang member as Rodgon had promised in his opening statement (R. 48-5 at 125-27). (Hampton’s petition included other claims not relevant here.) Hampton submitted an affidavit in support of the petition in which he averred, inter alia, that he had given Rodgon the names, addresses, and telephone numbers of Gregory Mallory, Clinton Williams, and Ronnie Garner, and he had told Rodgon that they would confirm that he had not participated in the attacks. R. 48-5 at 168 ¶ 12. Williams and Garner signed affidavits indicating that Hampton was not a gang member, that they had attended the concert with him, and that Hampton was not involved in the attacks R. 48-5 at 180-SI 115 (Williams); id. at 182-83 ¶ 5 (Garner). They also stated that Rodgon had never contacted them. R. 48-5 at 180 ¶ 4 (Williams); id. at 182 ¶ 4 (Garner). Hampton’s affidavit also indicated that he would have given Rodgon the names of potential character witnesses if Rodgon had asked him (R. 48-5 at 169 ¶ 15); and several of his friends and family members submitted affidavits averring, among other things, that Hampton was not a gang member (R. 48-5 at 184 ¶ 5g; id. at 187 ¶ 6b; id. at 189 ¶ 5h, i; id. at 191 ¶ 5d; id. at 193 ¶ 5d). Hampton’s supplemental petition ultimately was assigned to Circuit Judge Colleen McSweeney Moore.
On the State’s motion, Judge Moore summarily dismissed the bulk of Hampton’s petition, including his ineffectiveness claim. She determined that an evidentiary hearing was warranted on Hampton’s claim that he had been denied the right to testify on his own behalf. R. 48-7 at C38.4 *228But she concluded that no such hearing was warranted as to his allegations of ineffectiveness. The judge thought the ineffectiveness claim meritless in part because Hampton could not show that the outcome of his trial might have been different had Rodgon taken the steps that Hampton alleged he had neglected to take. R. 48-7 at C34, C36. In the judge’s view, Hampton’s postconviction counsel was attempting to substitute his own strategic judgment, informed by hindsight, for that of Rodgon. Id. at C34-35. Judge Moore went on to “find ... as a matter of fact from the record ... that Mr. Ro[dg]on’s trial tactics, his strategy in the manner in which he represented Patrick Hampton’s interest was not ineffective but rather was highly competent.” Id. at C35. She noted that Rodgon had moved to quash Hampton’s arrest and had obtained an evidentiary hearing on that motion, he had moved to suppress the identification of Hampton and also obtained a hearing on that request, he had pursued additional motions regarding discovery and other issues in the case, and during the trial he had effectively cross-examined the State’s witnesses. Id. at C32-34. The judge rejected the notion that Mallory’s acquittal supplied reason to believe that Hampton too might have been acquitted had Rodgon looked for the types of witnesses that Mallory’s counsel had presented. “[T]he evidence with regard to Mr. Mallory was so much weaker than the evidence with regard to Patrick Hampton,” she reasoned. Id. at C35. In Hampton’s case, a security guard as well as the victims of the assault were able to identify Hampton; the case against Mallory, by contrast, rested solely on the testimony of a single victim (Denise M.). See id.
A divided Illinois Appellate Court affirmed the dismissal of Hampton’s ineffectiveness claims, concluding that Hampton’s allegations did not satisfy either of the two criteria set forth in Strickland v. Washington, 466 U.S. 668, 104 S.Ct. 2052, 80 L.Ed.2d 674 (1984)—objectively unreasonable performance by counsel and prejudice resulting therefrom. People v. Hampton, 296 Ill.App.3d 1065, 244 Ill.Dec. 876, 726 N.E.2d 1187 (1998) (unpublished) (hereinafter, “App.Ct.Order”). Hampton could not show prejudice, the court reasoned, given the “overwhelming” proof of his guilt — namely, the fact that two of the victims of the attack, along with the security guard, had identified him as one of the *229perpetrators. App.Ct. Order at 5. Nor could Hampton establish that Rodgon’s performance as his attorney was objectively unreasonable. The decision whether to present a witness, the court stated, is a matter of strategy and “cannot support a claim of ineffective assistance of counsel.” Id. at 6. The court noted that an attorney is only obliged to make a reasonable investigation or to make a reasonable decision that renders particular investigations unnecessary. Id. Here, Rodgon knew that Hampton had attended the concert along with several other defendants and that he had been positively identified as one of the perpetrators by two victims and the guard. Against that backdrop, the court believed, Rodgon reasonably could have concluded that pursuing witnesses who would simply have confirmed that Hampton was present at the scene of the crime and linked to the other perpetrators of the offense posed as much of a detriment as a benefit to Hampton’s defense:
The testimony of these other witnesses would have been redundant and only serve[d] to emphasize the fact that defendant went to the concert with the perpetrators, was present during the assault, and left with the perpetrators. It was not unreasonable for defense counsel to find this troubling. Indeed, it was defense counsel’s assessment that defendant faced a probability of “guilt by association.”
Id. 7. Thus, “[a]pplying a heavy measure of deference to defense counsel’s judgment,” the court concluded that Rodgon made a reasonable decision not to interview the witnesses that Hampton had named. Id. As for Rodgon’s failure to fulfill the promise that Hampton would testify in his own defense,5 the court noted that Rodgon, at the time he made this statement, believed that Hampton would take the stand. Id. at 8. Subsequently, however, he discussed with Hampton the problem of guilt by association, and he also became concerned that Hampton might not be able to withstand the rigors of cross-examination. Id. In the end, Rodgon had advised Hampton that the choice whether or not to testify was his to make, and Hampton had decided not to testify. Id. The court viewed this as a change in trial strategy that could not support a claim of ineffectiveness. Id.
Judge Sheila O’Brien dissented, reasoning that Hampton should have had the same opportunity that the State had to present occurrence witnesses:
This crime occurred in a large area, with hundreds of people present. Defendant could have been present in the area and not have participated in the events and these occurrence witnesses could have corroborative information. These allegations make a substantial showing of a violation of defendant’s constitutional rights.
Id. at 21. The Illinois Supreme Court subsequently denied Hampton’s petition for leave to appeal. People v. Hampton, No. 85803, 179 Ill.2d 599, 235 Ill.Dec. 570, 705 N.E.2d 443 (Ill. Oct.6, 1998) (unpublished).
Having exhausted his state court remedies, Hampton filed a pro se petition for a writ of habeas corpus in the district court. The court appointed counsel to represent Hampton, and his attorneys subsequently filed an amended petition which, in relevant part, re-asserted Hampton’s claim of ineffectiveness. R. 27. Over the State’s objection, the district court conducted an *230evidentiary hearing to explore Rodgon’s failure to investigate exculpatory witnesses.
Hampton testified at that hearing that during the nine-month period that he was incarcerated in advance of trial, Rodgon had met with him at the jail on only one occasion, for about thirty to forty-five minutes. R. 59-1 at 10, 13.6 During this meeting, which took place shortly after Hampton was arraigned in January 1982, Hampton told Rodgon that he had attended the concert with Ronnie Garner, Clinton Williams, and Gregory Mallory, and that all three could verify that he was not involved in the attacks and that he was not a gang member. Id. at 13-14. All three lived in the same building as Hampton, and they had grown up together. Id. at 14. Hampton gave Rodgon their contact information. Id. He also pointed out Williams and Gregory Mallory to Rodgon at subsequent court proceedings that they attended. Id. at 16. Hampton assumed that Rodgon would contact them and call them to testify at trial. Id. at 14, 16, 17. When Rodgon rested the defense case without calling these witnesses, Hampton asked him why he had not done so and Rodgon told him not to worry about the matter. Id. at 18.
Rodgon testified that he believed the State’s case against Hampton was weak: there was no physical evidence implicating Hampton in the attacks; it was purely an identification case. R. 59-1 at 89. Hampton had not given him any information about potential defense witnesses, and for that reason Rodgon had not interviewed any such individuals. Id. at 99-100. Rodgon knew that Hampton lived in the Robert Taylor Homes and that “numerous people from there” had attended the concert. Id. at 91. However, he did not know who these individuals were. Had Hampton given him the names of potential occurrence witnesses, Rodgon testified, he would have followed through and spoken to such witnesses. Id. at 116. Rodgon did not visit the Robert Taylor Homes in an effort to identify potential witnesses, nor did he have an investigator do so. Id. at 100-02. With two immaterial exceptions, Rodgon’s case file contained no notes reflecting interviews with potential witnesses, although he agreed that such notes would have been in the file. Id. at 95-99, 151-52.
In a 1987 letter to the Illinois Attorney Registration and Disciplinary Commission (“ARDC”) responding to a complaint Hampton had filed about his performance, Rodgon had indicated that he did speak with Hampton about possible witnesses.
Prior to trial I did talk to [Hampton] concerning witnesses that may or may not be called on his behalf. As I recall he told me at that time that he was with people who he had left the Amphitheatre with on a bus. Most of these people were later arrested and charged with the crime and plead[ed] guilty. I did not think it appropriate for me to put on witnesses who had plead[ed] guilty to the crime to say that they were with Mr. Hampton at the time but that he did not participate in the crime.
R. 58 Petitioner’s Ex. ID at 3. When asked about the letter at the evidentiary hearing, Rodgon still maintained that Hampton had not given him the names of potential occurrence witnesses. Had Hampton done so, Rodgon testified, he would have sent someone to speak with them. R. 59-1 at *231121. Rodgon indicated that he did not contact co-defendant Ronnie Garner because when defendants plead guilty (as he knew Garner had), they typically incriminate not only themselves but their co-defendants. Id. at 128-24. He had not spoken with any of the other defendants who pleaded guilty for the same reason; Rodgon did not think it good strategy to put people who had pleaded guilty on the witness stand. Id. at 125. He acknowledged, however (at first reluctantly), that one cannot assess the credibility of a prospective witness without first talking to him or her. Id. at 147.
Harold Winston, Hampton’s postconviction attorney, also testified at the hearing before Judge Kennelly. Winston recalled that after obtaining Rodgon’s trial file and reviewing the contents, he asked Rodgon about the lack of any notes from interviews with Hampton or other witnesses. Rodgon told Winston that he had given him all that he had. R. 59-1 at 32. Winston also indicated that in the course of preparing Hampton’s supplemental post-conviction petition, he had spoken with Gregory Hubbard. Although Hubbard had given him helpful information, he had ultimately refused to sign an affidavit in support of Hampton’s petition. Id. at 54.
W. Michael Fay had represented Ronald Mallory at the trial. He testified before Judge Kennelly that he went to the Robert Taylor Homes along with two assistant public defenders who were representing Ricky Knight to speak with other people who had attended the concert and had witnessed the attacks. R. 59-1 at 68-70. Fay did not recall, however, whether the occurrence witnesses he called to testify in Mallory’s defense were among the people he spoke to at the Robert Taylor Homes. Id. at 71. Fay was confident that, pursuant to Circuit Court rules, he would have given Rodgon a copy of Mallory’s witness list. Id. at 73. Gregory Hubbard was among the witnesses included on that list. Id. Fay also recalled that he spoke with Rodgon regularly about the case as they encountered one another in the hallways of the Criminal Courts building. Id. at 79.
Having heard both Hampton and Rod-gon testify, Judge Kennelly credited Hampton’s testimony in relevant part. He found that Hampton had given Rodgon the names of potential witnesses but that Rodgon had failed to follow up with these witnesses. 2001 WL 1518533, at *7-*8. Judge Kennelly thought that a decision by Rodgon not to pursue thése witnesses might have been strategically justified vis á vis Garner, who had pleaded guilty to charges arising out of the incident and who had not contested the government’s version of events, which implicated Hampton. Id. at *8, *16 n. 7. However, the failure to investigate would not have been justified as to Mallory and Williams. Id. at *8, *16-*17. Not having interviewed those two individuals, Rodgon had no strategic basis for dismissing them as prospective witnesses. Id. at *8.
Judge Kennelly further observed that Rodgon admitted knowing that a group of people who lived at the same housing project as Hampton had attended the concert, and that some of those individuals might have witnessed the attacks. Id. at *8. But Rodgon had made no effort to locate such individuals. Id. Nor was there any indication that Rodgon had spoken with the attorneys representing Hampton’s co-defendants about their witnesses. Id. At least one of those witnesses — Gregory Hubbard — could have exculpated Hampton. Id. Finally, Rodgon had never asked Hampton for the names of witnesses who could verify that he was not in a gang, nor had he made any other efforts to locate such witnesses. Id.
*232Ronnie Garner also testified at the evi-dentiary hearing before Judge Kennelly. At the time of the concert, Garner had known Hampton for ten years and was his friend. According to Garner, Hampton was not a gang member. Garner had attended the concert along with Ronnie Jackson, Sandelle Poole, Ezra Garner (Ronnie’s brother), and Hampton; Hampton also left the concert with them. During the concert, Hampton was seated near Ronnie Garner and was within his sight at all times. Garner testified that Hampton was not part of the crowd that attacked the three Latinos. Garner said that no one had ever asked him what he knew about Hampton’s involvement in the attacks, nor had anyone asked him to testify on Hampton’s behalf. He would have testified in Hampton’s defense had he been asked. Garner himself had been charged with participating in the attacks, however; and he ultimately had pleaded guilty (against his lawyer’s advice) and was sentenced to the six months he had already served in pretrial detention. On the witness stand before Judge Kennelly, Garner insisted that he had not, in fact, participated in the attacks, but had elected to plead guilty in order to avoid a potentially much longer term had he been tried and convicted. And although he had not contested the version of events that the prosecutor recited at the change of plea hearing — a version that had implicated Garner and Hampton both — neither had he endorsed it as accurate. R. 57-8 at 21-27; see R. 58, Petitioner’s Ex. 11 at 31-32 (State proffers factual basis for plea and Judge Strayhorn finds it sufficient to accept plea but Garner not asked to endorse it); compare id. at 7 (co-defendant Ford asked to stipulate to State’s proffer). Judge Kennelly found Garner’s testimony to be credible. 2001 WL 1518533, at *9.
Gregory Mallory, who had been a next-door neighbor and friend of Hampton’s for about nine years at the time of the concert, also testified. Gregory had attended the concert, was seated about five to ten feet from Hampton, and said that Hampton was not near the attacking crowd. As noted above, Gregory’s brother Ronald Mallory was tried along with Hampton, and Gregory had testified in Ronald’s (successful) defense. Although Gregory said that he would have testified on Hampton’s behalf, Hampton’s attorney had never contacted him. Gregory had begun using heroin in 1989, and was still addicted to the narcotic when the evidentiary hearing took place in 2001. He was convicted of felony theft in 1981, and for heroin possession in 1994. Although Gregory did not believe that his longtime abuse of heroin had affected his memory, he answered a number of the questions put to him with answers like “okay” and “I guess so,” and he was unable to recall certain details about the concert (e.g., precisely what row he sat in) and about the trial (including the fact that it involved three juries). But Gregory had not been a heroin user at the time of the concert or at the time of Hampton’s trial. And although Judge Kennelly characterized Mallory’s memory as imperfect, he found his testimony credible nonetheless.
Unsurprisingly, given the passage of time and the effects of heroin use, [Gregory Mallory’s] memory of the incident has faded a bit. But the issue is not whether he would make a good witness if Hampton’s trial were held today, but what effect his testimony might have had at the 1982 trial. The Court finds that Mallory would have made an effective defense witness at that trial.
2001 WL 1518533, at *9.
Clinton Williams died in July 1995, six years before the evidentiary hearing took place. He too had been- a friend of Hampton’s and had attended the concert. Williams was alive at the time that Hamp*233ton had petitioned for postconviction relief in state court and had signed an affidavit in support of that petition indicating that Hampton had not participated in the attacks. R. 48-5 at 180-81. Judge Kennelly discerned no reason to believe that Williams would not have been a credible witness on Hampton’s behalf. 2001 WL 1518533, at *9. The judge pointed out that there was no evidence linking Williams to the group that had perpetrated the attacks. Id.
Before reaching the merits of Hampton’s failure-to-investigate claim, the district judge determined that there were no procedural obstacles that precluded either the evidentiary hearing he had convened or consideration of the merits of-the claim. The court first concluded that Hampton had fairly presented the claim in state court and thus had exhausted his state court remedies with respect to that claim. See 28 U.S.C. § 2254(b)(1), (c); O’Sullivan v. Boerckel, 526 U.S. 838, 844-45, 119 S.Ct. 1728; 1732, 144 L.Ed.2d 1 (1999); Picard v. Connor, 404 U.S. 270, 275-76, 92 S.Ct. 509, 512, 30 L.Ed.2d 438 (1971). This was so, in the court’s view, notwithstanding Hampton’s failure to submit an affidavit from Gregory Mallory in support of his postcon-viction petition. 2001 WL 1518533, at *10. As the court observed, a claim advanced in a federal habeas petition may be said to have been fairly presented to the state courts so long as that claim is fundamentally the same claim that the petitioner asked the state courts to resolve. Id. at *10, citing Boyko v. Parke, 259 F.3d 781, 789 (7th Cir.2001). This was true of Hampton’s ineffectiveness claim, the judge reasoned. Although Hampton had not tendered an affidavit from Gregory Mallory to substantiate the ineffectiveness claim (as he had from Williams and Gamer), Hampton had, in his own affidavit, averred that Gregory’s name was among those he had given to his attorney as a witness, and Hampton had argued in his petition that Rodgon should have contacted and interviewed Gregory. Under these circumstances, Hampton’s federal claim was not materially different from the one Hampton had pursued in state court. 2001 WL 1518533, at *10.
Nor did Hampton’s failure to submit an affidavit from Gregory constitute a procedural default that barred federal consideration of the ineffectiveness claim. Id. at *11; see Harris v. Reed, 489 U.S. 255, 261-62, 109 S.Ct. 1038, 1042, 103 L.Ed.2d 308 (1989). Although the Illinois Appellate Court had noted the absence of an affidavit from Gregory in its discussion of the claim, it had nonetheless addressed the merits of the claim in its entirety and without relying on the procedural flaw of the missing affidavit. Because the state court chose to ignore this default, Judge Kennelly reasoned, it posed no obstacle to federal consideration of the ineffectiveness claim. 2001 WL 1518533, at *11.
Next, the judge found that a federal statutory restriction on evidentiary hearings in habeas proceedings did not foreclose the court from hearing testimony in support of Hampton’s ineffectiveness claim. Id. at *12-*14. Subject to narrow exceptions, 28 U.S.C. § 2254(e)(2), added by the Antiterrorism and Effective Death Penalty Act of 1996 (“AEDPA”), forbids a district court from holding an evidentiary hearing on a habeas claim if the petitioner failed to develop the factual basis of that claim in state court. As the district court acknowledged, we have held that this limitation applies not just to an evidentiary hearing, but to any means used in lieu of such a hearing to expand the record in order to introduce new factual information. 2001 WL 1518533, at *12, citing Boyko, 259 F.3d at 790. The State argued that section 2254(e)(2) constrained the court’s ability not only to convene an evidentiary *234hearing, but also its ability to consider the affidavit from Gregory Mallory that Hampton had submitted in support of his habeas petition, given that Hampton had not submitted an affidavit from Gregory in support of his postconviction petition in state court. But this provision forecloses an expansion of the record only if the habeas petitioner’s failure to develop the record appropriately in state court was due to the petitioner’s lack of diligence or some larger fault attributable to the petitioner or his counsel. See Williams v. Taylor, 529 U.S. 420, 432, 120 S.Ct. 1479, 1488, 146 L.Ed.2d 435 (2000); see also Matheney v. Anderson, 253 F.3d 1025, 1039 (7th Cir.2001) (quoting Burris v. Parke, 116 F.3d 256, 258-59 (7th Cir.1997), cert. denied, 522 U.S. 990, 118 S.Ct. 462, 139 L.Ed.2d 395 (1997)), cert. denied, 535 U.S. 1030, 122 S.Ct. 1635 (2002).
In the district court’s view, there was no evidence that Hampton had been anything but diligent in pursuing his ineffectiveness claim in state court. 2001 WL 1518533, at *12. Although the materials in support of Hampton’s post-conviction petition did not include an affidavit from Gregory, Hampton had identified Gregory in his own affidavit, averring that he had given Gregory’s name and contact information to Rodgon as an eyewitness who might testify on his behalf. The court also noted that Hampton had asked the lower state court to conduct an evidentiary hearing but that the State had opposed a hearing and the state court had refused to conduct one. Had a hearing been held, the court theorized, “it is overwhelmingly likely that [Gregory] Mallory would have been called to testify.” Id. at *12. Consequently, although the state court record lacked an affidavit from Gregory that outlined what his testimony on Hampton’s behalf might have been (had Rodgon pursued him as a witness), that omission could not be ascribed to a lack of diligence on Hampton’s part. “Having been rebuffed at the prosecution’s request in his attempt to make a complete record in state court, Hampton cannot be faulted for his failure to do so or accused of a lack of diligence.” Id.7
Having thus determined that the AED-PA posed no bar to taking additional evidence on Hampton’s claim, the court turned to pre-AEDPA standards. See Matheney, 253 F.3d at 1039. Under those rules, an evidentiary hearing on a habeas petitioner’s claim is required if the petitioner has alleged facts that would entitle him to relief and the state courts, for reasons beyond his control, did not consider his claim in a full and fair hearing. See Townsend v. Sain, 372 U.S. 293, 312-13, 83 S.Ct. 745, 757, 9 L.Ed.2d 770 (1963), overruled on other grounds by Keeney v. Tamayo-Reyes, 504 U.S. 1, 112 S.Ct. 1715, 118 L.Ed.2d 318 (1992); Wright v. Gram-*235ley, 125 F.3d 1038, 1044 (7th Cir.1997). A full and fair hearing is one that afforded the petitioner a complete opportunity to present the facts relevant to his constitutional claim. See Matheney, 253 F.3d at 1039; Spreitzer v. Peters, 114 F.3d 1435, 1456 & n. 9 (7th Cir.1997), disposition clarified, 127 F.3d 551 (7th Cir.1997), cert. denied, 522 U.S. 1120, 118 S.Ct. 1060, 140 L.Ed.2d 121 (1998). Judge Kennelly found that the Illinois post-conviction court, when it dismissed the' relevant portion of Hampton’s postconviction petition without a hearing, had deprived Hampton of this opportunity:
The state court did not permit Hampton to explore his trial counsel’s reasons — if he had any — for failing to interview and call the witnesses whose names Hampton had given him. Nor did it permit him to address the issue of prejudice, which would have required consideration of the effect of [those witnesses’] testimony.
2001 WL 1518533, at *13.8 The judge pointed out that in both Matheney and Bruce v. United States, 256 F.3d 592, 600 (7th Cir.2001), we had concluded that an evidentiary hearing was necessary in order to evaluate the ineffectiveness claims asserted in those cases. Judge Kennelly believed that this was true here as well. Testimony from Hampton’s counsel as well as Hampton’s proposed witnesses would enable the court to assess the adequacy of Rodgon’s representation of Hampton and to determine whether Hampton was prejudiced by Rodgon’s failure to interview (and summon to testify) Hampton’s witnesses. 2001 WL 1518533, at *13.
The Illinois Appellate Court had said that Rodgon’s failure to .follow up with Hampton’s occurrence witnesses was based on a strategic decision to avoid the specter of guilt by association that such witnesses might have raised or enhanced (App.Ct. Order at 7); but the district court found to the contrary. Although section 2254(e)(1) provides that state court findings of fact are owed a presumption of correctness, no such presumption was warranted here, in the district court’s view. 2001 WL 1518533, at *15. The state court’s finding was “entirely speculative,” with no basis in the record before the state courts. Id. The only evidence that Rodgon was concerned about guilt by association came from a limited hearing that the state trial court had conducted on Hampton’s separate claim regarding Rodgon’s failure to have Hampton himself testify. Id.; see n. 4, supra; R.48-7 at D32-33. Rodgon’s rationale for not calling Hampton to the witness stand could not be “transmogrified” into an explanation for his failure to conduct an adequate pretrial investigation, the district court reasoned. 2001 WL 1518533, at *15. Moreover, even if the appellate court’s characterization of Rodgon’s conduct as “strategic” were presumed correct, Hampton had succeeded in rebutting that presumption by clear and convincing evidence. Id.; see § 2254(e)(1). *236The court found that Rodgon had no reason not to follow up on the eyewitness information that Hampton had given him or not to make his own effort to identify other occurrence witnesses. Rodgon had conceded on the witness stand that an attorney cannot decide whether to call a witness who is not also a co-defendant without first interviewing that witness, barring some other external factor known to the attorney (such as a significant felony record or the immateriality of the witness’s testimony) that detracts from the value of that witness’s possible testimony. 2001 WL 1518533, at *15.
The district court also viewed the appellate court’s strategy determination as either contrary to or an unreasonable application of Stricklands observation that strategic choices based on something less than a complete investigation are reasonable to the extent that reasonable professional judgment would support limits on that investigation. Id. at *16; see Strickland v. Washington, supra, 466 U.S. at 690-91, 104 S.Ct. at 2066. Here, there was no evidence of any judgment at all on Rodgon’s part, and it was not the court’s duty to fill that void with its own judgment. 2001 WL 1518533, at *16.
Where the prosecution relies heavily on eyewitness testimony that might be rebutted by other eyewitnesses, the district court emphasized, the attorney’s duty to investigate such witnesses is critical. Id. In this case, Rodgon had cross-examined the State’s witnesses and exposed the weaknesses in their identifications of Hampton, but he made no effort to determine whether Hampton had more of a defense than that. Id. Cases including Washington v. Smith, 219 F.3d 620, 631-32 (7th Cir.2000); Williams v. Washington, 59 F.3d 673, 681 (7th Cir.1995); Harris v. Reed, 894 F.2d 871, 878-79 (7th Cir.1990); and Sullivan v. Fairman, 819 F.2d 1382, 1389 (7th Cir.1987), support the notion that the failure to investigate exculpatory witnesses can amount to ineffective assistance of counsel. 2001 WL 1518533, at *17-* 18. Rodgon’s ineffectiveness was not limited to his failure to interview the eyewitnesses that Hampton had identified for him, but included his failure to make an effort of his own to locate other eyewitnesses. Id. at *18. The court noted that if Rodgon simply had paid attention to the witness lists of Hampton’s co-defendants, he would have found at least one witness (Gregory Hubbard) who could have exonerated Hampton. Id.
The district court believed that Rodgon was also ineffective for failing to look for witnesses who could confirm that Hampton was not a gang member. Id. at *18-*19. The Illinois Appellate Court had written this omission off as a failure to call witnesses who might have testified that Hampton was a “decent person.” App.Ct. Order at 16. But Hampton’s supplemental postconviction petition, along with the affidavits he had submitted in support of that petition, made clear that the gist of such witnesses’ testimony was not simply that Hampton was of good character, but that he was not a gang member. 2001 WL 1518533, at *19. Gang affiliation was made relevant by the strong overtones of gang activity that Powell had observed in the group of men he saw marching toward the stage of the theater (the chanting of gang slogans and hand signals). The unfulfilled promise that Rodgon made in his opening statement — that the jury would hear evidence that Hampton was not involved in a gang — revealed that Rodgon was aware of the importance of this issue. Id. at *18. Proof that Hampton had no gang connections would have tended to undercut testimony that he was part of the group that attacked the three victims. Id. at *19. Yet Rodgon had never so much as asked Hampton for the names of individu*237als who could attest to his lack of gang involvement. Id.
The Illinois Appellate Court had also held that even if Rodgon had rendered ineffective assistance, Hampton could not demonstrate that he was prejudiced by his attorney’s performance. App.Ct. Order at 5. “In light of the overwhelming evidence of defendant’s guilt, ie., two of the victims and a security guard identified defendant, defendant cannot establish that but for his defense counsel’s performance the outcome of his trial would have been different.” Id. The district court concluded that this one-sentence disposition of the prejudice prong of the ineffectiveness inquiry amounted to an unreasonable application of Strickland.
In the district court’s view, the State’s case against Hampton was “far from unassailable.” 2001 WL 1518538, at *19. Heinrichs, the security guard, conceded that on the night of the attacks he had only seen the assailant he later identified as Hampton for three or four seconds, and he had not picked Hampton out of a lineup until after he had seen Hampton’s picture in a television news report. Id. Denise M. and Martha N. likewise had only gotten brief glimpses of Hampton and had been unable to describe him except as an African-American male for the police. Id. Hugo M. never identified Hampton notwithstanding the fact that he was at Denise M.’s side during the assault. Id. Each of the witnesses who identified Hampton had witnessed the assault “under extremely stressful .,. near-riot conditions.” Id. at *19.
The fact that the jury acquitted Hampton of the attempted rape of Martha N. demonstrated the likelihood that the testimony of exculpatory occurrence and other witnesses might have produced a different outcome on the other charges. Id. at *20. Martha N. had identified Hampton as one of her attackers, and had done so under the same circumstances as the other two witnesses who identified Hampton. But in Martha N.’s case, the record had supplied a basis to challenge her identification — the written report indicating that Martha N. had picked from the line-up someone other than Hampton. Notwithstanding testimony from a detective that the written report was erroneous, the jury’s decision to acquit Hampton on this charge suggested to the district court that the jury had doubts about the strength of Martha N.’s identification. Id. This in turn indicated to the court that if the jury had been given reason to doubt the identifications by Denise M. and the guard beyond arguments about the circumstances under which they had seen the assailant, it was reasonably likely that the jury’s decision on the other charges might have been different as well. Id.
Rodney Mallory’s acquittal on all charges supplied additional, “powerful evidence” that an adequate investigation into defense witnesses might well have produced a different result for Hampton. Id. at *20. Mallory’s attorney had put on the very type of eyewitness testimony that Rodgon had failed to pursue, including Rodney’s brother Gregory, whom Hampton had cited to Rodgon as an eyewitness. Id. Rodney Mallory was acquitted notwithstanding the testimony of Denise M. that he had attempted to place his penis in her mouth and the testimony of Keith Powell that Rodney was among the group of individuals who walked toward the stage of the Amphitheatre. Id. True, no security guard had identified Rodney. Id. Even so, the favorable outcome for Rodney Mallory persuaded the district court that it was reasonably probable that a defense founded in part upon exculpatory occurrence witnesses would have resulted in Hampton’s acquittal. Id.
*238The district court found further that Rodgon’s failure to investigate and present witnesses to establish Hampton’s lack of gang affiliation, although not dispositive by itself, was also “highly prejudicial” to Hampton. Id. Gang affiliation was an aspect of the prosecution’s case that tended to buttress the witnesses’ identification of Hampton. Id. Sowing doubt about the notion that Hampton would have participated in gang-related activity thus would have given the jury reason to question the identification of Hampton. Id.
Rodgon’s failure to keep the promise he made during opening that Hampton would testify in his own defense reinforced the district court’s conclusion that Rodgon’s representation of Hampton was ineffective. Id. at *21-*22. The Illinois Appellate Court had chalked up Hampton’s failure to testify to a change in trial strategy driven by Rodgon’s concern over the prospect of guilt by association. App.Ct. Order at 8. Had Hampton testified, he would have admitted that he was present at the scene of the crime and that he arrived and left with others implicated in the attacks. But if that prospect was what worried Rodgon, the district court noted, it was a possibility that was as apparent before trial as it was when the time came to decide whether Hampton would take the witness stand; there was no indication that anything relevant to the guilt-by-association concern had changed over the course of the trial. 2001 WL 1518533, at *21. Consequently, the district court believed, it was “foolhardy and objectively unreasonable for [Rodgon] to promise that Hampton would testify.” Id. The court acknowledged that an attorney’s failure to keep a promise made in an opening statement will rarely supply the basis for an ineffectiveness claim. Id. In this case, however, nothing occurred during the State’s case against Hampton that altered the pros and cons of Hampton taking the stand. Id. at *21-*22. True, had he testified, Hampton would have had to explain his presence at the scene of the crime, but, the court pointed out, Rodney Mallory had been able to do so successfully. Id. at *22. Under the circumstances, the court found it objectively unreasonable for Rodgon to make and then abandon the promise that Hampton would take the stand. Id. “The Appellate Court’s contrary conclusion, based on a finding that was without any support in the record, was an unreasonable application of Strickland.” Id. Rodgon’s failure to fulfill the promise, the district court went on, was also prejudicial to Hampton. Id. Hampton’s failure to take the stand as promised gave rise to a negative inference about what the content of his testimony might have been. Id. The court could not say that this alone likely affected the outcome of the trial. However, the decision not to keep the promise “significantly buttresse[d]” the court’s conclusion that Rodgon’s overall performance likely did influence the outcome. Id.
Having concluded that Hampton was deprived of the effective assistance of counsel, Judge Kennelly granted his petition for a writ of habeas corpus. He ordered Hampton released from prison unless, within thirty days, the State announced its intent to retry Hampton. Id. at *25. The State appealed, and subsequently it asked the district court to stay its judgment. Hampton in turn asked the court to order his release pending appeal. By this time, Hampton had been incarcerated for twenty years, but even with credit for good time, he could not expect to be discharged from prison prior to January 2012. After weighing the equities, Judge Kennelly stayed the judgment to the extent of relieving the State, pending appeal, of the obligation to announce its intent to retry Hampton within thirty days. Hampton v. Leibach, 2001 WL 1618737, at *3 (N.D.Ill. *239Dec.18, 2001). The judge also ordered Hampton released from prison on bond pending appeal. Id. at *2-*3. Hampton’s release was conditioned on his sister’s willingness to execute a quitclaim deed on her home as security. Id: at *3. He also required Hampton to “live with his sister in her home, restrict his travel to the State of Illinois, make reasonable efforts to seek employment, and avoid contact with the victims of the offenses with which he was charged in state court.” Id. Although this court subsequently stayed Hampton’s release pending appeal, United States Supreme Court Justice John Paul Stevens, acting as Circuit Justice, granted Hampton’s application to vacate that stay, and the full Court subsequently declined to overturn his order. On January 28, 2002, this court ordered Hampton released at such time as the district court’s conditions on his release were satisfied.
II.
A. Evidentiary Hearing
The State’s initial contentions on appeal concern the district court’s decision to conduct an evidentiary hearing on Hampton’s ineffectiveness claim. The State does not quarrel with the lower court’s determination that Hampton fairly presented his failure-to-investigate claim to the state courts. See § 2254(b)(1), (c); Boyko, 259 F.3d at 788-89. But for three reasons, the State contends that the decision to hold a hearing on that claim was, either in part or in whole, erroneous: (1) Hampton did not submit an affidavit from Gregory Mallory in support of his supplemental petition for postconviction relief, and to that extent failed to develop the factual basis for his claim in state court; (2) Hampton’s failure to tender an affidavit from Gregory in state court, which the Illinois Appellate Court noted, amounts to a procedural default that bars federal consideration of the ineffectiveness claim; and (3) there was, in any case, no need for an evidentiary hearing because there were no factual disputes that required a hearing to resolve.
1. Failure to Submit Affidavit from Gregory Mallory.
As the district court recognized, section 2254(e)(2)' bars a district court from conducting an evidentiary hearing or otherwise permitting an expansion of the evidentiary record in support of a habeas claim if the petitioner failed to develop the factual basis for his claim in state court. In the supplemental postconviction petition that Hampton filed in state court, Hampton identified the occurrence witnesses whose names he had allegedly given to Rodgon, he submitted his own affidavit averring that he supplied Rodgon with their names and contact information, and he submitted affidavits from two of those witnesses — Williams and Garner — which in turn confirmed that they would have given exculpatory testimony on Hampton’s behalf. But Hampton did not submit an affidavit from Gregory Mallory. As we have noted, Hampton’s postconviction counsel, Harold Winston, testified at the evidentiary hearing in the district court, and Winston explained that he had interviewed Gregory, that Gregory had provided helpful information, but that Gregory had been unwilling to sign an affidavit reflecting that information and that Winston lacked the authority to compel him to do so. R.59-1 at 54-55. The State contends that in failing to tender an affidavit from Gregory Mallory, Hampton failed to develop the basis for his ineffectiveness claim.
It is not entirely clear from the briefing whether the State believes that this omission wholly forecloses all expansion of the record in support of Hampton’s ineffectiveness claim, or only the submission and *240consideration of Gregory’s affidavit and testimony. We shall assume that the State means only to argue that Hampton is foreclosed from tendering Gregory’s affidavit and testimony, as it is otherwise clear that Hampton tendered affidavits from Williams and Garner in state court and thus developed his ineffectiveness claim to that extent. This is consistent with the State’s presentation at oral argument. There its counsel argued, in essence, that a failure-to-investigate claim akin to Hampton’s is both defined and limited by the particular witnesses that the petitioner charges his counsel with failing to contact. In other words, a failure to investigate claim supported by the affidavits of two exculpatory eyewitnesses (like Garner and Williams) is a claim based on those two witnesses alone; and if the petitioner at a later date presents affidavits from additional eyewitnesses that counsel failed to contact, he has inevitably transformed his ineffectiveness claim into a new claim with a distinct factual basis. Thus, in the State’s view, if the affidavits of those additional witnesses were not presented to the state courts, then section 2254(e)(2) precludes the petitioner from relying on those witnesses in federal court, because they represent a factual basis for the ineffectiveness claim that was never developed in state court.
Whether the State is correct to parse the failure-to-investigate claim and its factual basis in this way is a matter on which we reserve judgment. The essence of Hampton’s claim is that there were multiple eyewitnesses to the attacks who could have exonerated him and that his counsel failed to contact any of them. In support of that claim, he presented the affidavits of two of those witnesses (Garner and Williams) to the state postconviction court, in this way documenting the nature of the exculpatory testimony that was available had his counsel pursued this line of investigation. Mallory’s proffered testimony was not materially different from that of Garner or Williams and so did not alter the basic nature of Hampton’s ineffectiveness claim; his was just one more iteration of the exculpatory testimony that was available to the defense but not investigated by Rodgon. In that sense, the failure to tender an affidavit from Gregory to the state court arguably did not constitute a failure to develop the factual basis for the ineffectiveness claim for purposes of section 2254(e)(2) — that basis was fleshed out with the affidavits of Garner and Williams. We need not resolve the point, however. We may assume that the omission of Gregory’s affidavit from Hampton’s postconviction petition might preclude him from presenting Gregory’s affidavit and testimony in federal court if the omission was attributable to a lack of diligence on Hampton’s part.
As the district court noted, a petitioner’s failure to develop the factual basis of his habeas claim in state court will bar expansion of the record in federal court only if this failure was due to a lack of diligence or some greater fault attributable to the petitioner himself. Williams v. Taylor, supra, 529 U.S. at 432, 120 S.Ct. at 1488; Burris v. Parke, supra, 116 F.3d at 258-59; see also, e.g., Newell v. Hanks, 283 F.3d 827, 838 (7th Cir.2002). The relevant inquiry is thus not simply whether the petitioner theoretically could have discovered the evidence while he was still in the state forum, but whether he made appropriate efforts to locate and present that evidence to the state courts. See Williams, 529 U.S. at 435, 120 S.Ct. at 1490. Winston’s testimony in the district court reveals that he had spoken with Gregory, that he had obtained information from Gregory that would have supported Hampton’s claim, but that, in the end, Gregory refused to sign the affidavit. R. *24159-1 at 54. Hampton argues that his failure to submit an affidavit from Gregory was thus due to circumstances beyond his control (Gregory’s refusal to cooperate) and not to his own lack of diligence or some other fault that may deemed his responsibility. Notably, the State does not quarrel with the notion that a witness’s refusal to execute an affidavit may excuse the petitioner’s failure to develop the record as to that witness’s prospective testimony. See State Br. at 34-39; see also Newell, 283 F.3d at 838. It makes two other arguments instead. First, it suggests that this court erred in Burris when it held that the failure to develop the factual basis of a claim will préelude an evidentia-ry hearing under section 2254(e)(2) only if the failure is attributable to the petitioner. In the State’s view, the AEDPA bars expansion of the record at the federal level even when the petitioner and his counsel are not to blame for failing to develop the factual basis for the claim in state court. State Br. at 35. Second, the State argues that if Hampton attempted but was unable to procure an affidavit from Gregory, his attorney should have executed an affidavit to that effect and presented that affidavit to the state postconviction court. State Reply Br. at 2.
The State’s first argument is untenable given the Supreme Court’s intervening opinion in Williams. The Court in Williams did not construe section 2254(e)(2) to bar an evidentiary hearing whenever the underlying facts were not developed in state court. The Court instead held that evidentiary expansion is foreclosed only when the failure to develop the factual record sooner is due to “a lack of diligence, or some greater fault, attributable to the prisoner or the prisoner’s counsel.” 529 U.S. at 432, 120 S.Ct. at 1488. The rule that we announced in Burris is consistent with Williams; indeed, the Court cited Burris (among other authorities) with approval. Ibid.
The State’s second argument misapprehends the nature of the inquiry called for by section 2254(e)(2). The State faults Hampton and his counsel for not making a record in state court as to counsel’s inability to obtain an affidavit from Gregory. But when the petitioner has not fully developed the factual basis for his claim in state court, it is the federal court that must decide whether that omission forecloses expansion of the record pursuant to section 2254(e)(2). To the extent that the state court record sheds light on whether the petitioner could have developed the facts underlying his claim while he was still in the state forum, it is obviously relevant to this inquiry. Cf. Williams, 529 U.S. at 437-40, 120 S.Ct. at 1491-92 (references to psychiatric record in state court proceedings demonstrated that petitioner, if diligent, could have located report and raised its non-disclosure in state habeas proceeding). But when the reason is not self-evident from the state record, nothing in section 2254(e)(2) precludes the petitioner from supplying the explanation when he arrives in federal court. See id. at 440-44, 120 S.Ct. at 1492-94 (relying on federal court record to determine that petitioner’s failure to develop factual basis for two of his claims was not due to lack of diligence on petitioner’s part); Boyko, 259 F.3d at 791-92 (remanding for determination by district court as to petitioner’s diligence). Hampton’s postconviction counsel, in his testimony before the district court, detailed the unsuccessful effort he. had made to obtain an affidavit from Gregory. In deciding whether expansion of the record was appropriate, the district court was free to consider that and any other evidence that placed into context Hampton’s failure to completely develop the basis for his ineffectiveness claim in *242state court.9 As we have said, the State has not disputed the notion that a witness’s refusal to cooperate with the petitioner may suffice under section 2254(e)(2) to excuse the petitioner’s failure to tender that witness’s testimony in the state forum. In the absence of such an argument, we shall assume that Hampton did demonstrate diligence not withstanding his failure to submit an affidavit from Gregory Mallory in support of his postconviction petition.
2. Procedural Default
When it addressed Hampton’s ineffectiveness claim, the Illinois Appellate Court pointed out that Hampton had attached affidavits from both Williams and Garner to his petition for postconviction relief, but not one from Gregory. App.Ct. Op. at 6. The Illinois Post-Conviction Hearing Act requires a petitioner to attach to his petition affidavits (or other evidence) supporting the factual allegations of the petition. See Ill. Stat. Ann. Ch. 38 ¶ 122-2 (Smith-Hurd 1990), now codified at 725 Ill. Comp. Stat. 5/122-2. Thus, a claim that trial counsel failed to investigate and call to testify a particular witness should be supported by an affidavit from that witness. People v. Enis, 194 Ill.2d 361, 252 Ill.Dec. 427, 743 N.E.2d 1, 13 (2000), cert. denied, 533 U.S. 908, 121 S.Ct. 2254, 150 L.Ed.2d 241 (2001). Hampton’s failure to tender an affidavit from Gregory was in apparent violation of that requirement. The State contends that this procedural default, which the Appellate Court noted, constitutes an adequate and independent state procedural ground for the court’s decision that precludes the federal courts from reaching the merits of his constitutional claim insofar as it rests on Gregory’s testimony. See Stewart v. Smith, 536 U.S. 856, 860, 122 S.Ct. 2578, 2581, 153 L.Ed.2d 762 (2002); Coleman v. Thompson, 501 U.S. 722, 729, 111 S.Ct. 2546, 2553-54, 115 L.Ed.2d 640 (1991).
However, because the state court did not rely on Hampton’s purported default in disposing of his ineffectiveness claim, we may reach the merits of the claim. A petitioner’s procedural default will bar federal habeas review only if the state court actually relied on that default as an independent basis for its decision. Harris v. Reed, supra, 489 U.S. at 261-62, 109 S.Ct. at 1042. Here, the Illinois Appellate Court (the last state court to address the ineffectiveness claim) did no more than note Hampton’s failure to submit an affidavit from Gregory Mallory. App.Ct. Order at 6. The court gave no hint that the missing affidavit constrained or influenced its resolution of the claim in any way. The court *243proceeded to plenary consideration of the merits of the claim, referring collectively to the witnesses that Hampton claimed his attorney had failed to investigate as “these witnesses,” without drawing any distinction between Williams and Garner, whose affidavits had been attached to Hampton’s petition, and Gregory Mallory. App.Ct. Order at 7. Under these circumstances, we find no adequate and independent state procedural ground that forecloses or limits our consideration of Hampton’s ineffectiveness claim. See, e.g., Farmer v. Litscher, 303 F.3d 840, 846 (7th Cir.2002); Moore v. Bryant, 295 F.3d 771, 774 (7th Cir.2002).
The State makes a similar argument with respect to Gregory Hubbard. Hubbard’s name was not among those that Hampton had given to Rodgon. Rather, the district court cited Hubbard as an example of an exculpatory witness whom Rodgon would have discovered had he made a rudimentary effort of his own to identify such witnesses. 2001 WL 1518533, at *8. Recall that Hubbard had testified on Ronald Mallory’s behalf and, among other things, claimed that Hampton did not participate in the attacks. R. 48-3 at 1132-33, 1145. As was the case with Gregory Mallory, Hampton did not attach an affidavit from Hubbard to his postconviction petition, and the State thus contends that Hampton procedurally defaulted his ineffectiveness claim insofar as it relied on Hubbard as a prospective witness.10
Hubbard, however, occupies a unique place among the witnesses that Rodgon might have summoned on Hampton’s behalf, in that the trial record itself revealed the substance of Hubbard’s testimony. When he testified for Ronald Mallory, Hubbard had not only exculpated Mallory but also Hampton (Tr. 1132, 1145); it was simply the case that Hampton’s jury never heard this testimony. Given the existing record of Hubbard’s testimony, there was no real need for Hampton to submit an affidavit from Hubbard in order to substantiate the claim that he would have been a helpful witness. See People v. Johnson, 183 Ill.2d 176, 233 Ill.Dec. 288, 700 N.E.2d 996, 1003 (1998) (lack of affidavit or other evidence not fatal to postcon-viction claim if petitioner’s allegations are uncontradicted and clearly supported by record), cert. denied, 526 U.S. 1009, 119 S.Ct. 1150, 143 L.Ed.2d 217 (1999). As it was, Hampton not only named Hubbard in his supplemental postconviction petition, but cited the testimony he had given on behalf of Ronald Mallory. R. 48-5 at 119. That testimony was in the record, the entirety of which the postconviction judge said she had considered before dismissing his failure-to-investigate claim. R. 48-7 at C3, C31. Hampton did not, therefore, commit a procedural default in state court that precludes him from relying upon Hubbard’s testimony.11
*2443. Need for an Evidentiary Hearing
Finally the State appears to argue in passing that the district court erred in ordering an evidentiary hearing because “there were no factual disputes which warranted a hearing.” State Br. at 39. Because the State has not developed this argument, we need not consider it. See, e.g., Palmquist v. Selvik, 111 F.3d 1332, 1342 (7th Cir.1997) (“Even an issue expressly presented for resolution is waived if not developed.”). Instead, we believe it sufficient to make two brief points.
First, a habeas petitioner is entitled to an evidentiary hearing under Townsend v. Sain, supra, 372 U.S. at 312-13, 83 S.Ct. at 757, if he has alleged facts that would entitle him to relief and the state courts, for reasons not attributable to him, denied him a full and fair hearing to explore those facts. See Matheney, 253 F.3d at 1039; Porter v. Gramley, 112 F.3d 1308, 1317 (7th Cir.1997), cert. denied, 522 U.S. 1093, 118 S.Ct. 886, 139 L.Ed.2d 873 (1998).12 That is the case here. Hampton’s habeas petition, together with the affidavits before the district court at the time it decided to convene a hearing, made out a prima facie case of ineffectiveness. Hampton had preliminarily established that there were eyewitnesses to the attacks who could have exculpated him, that these witnesses had either been identified to Hampton’s trial counsel Rodgon by Hampton himself or were readily discoverable, but that Rodgon had not contacted these witnesses in order to determine whether they might have aided Hampton’s defense. The state postconviction court was presented with those same basic averments but summarily disposed of Hampton’s petition without an evidentiary hearing; and the Appellate Court affirmed that disposition.
Second, there were points that the district court could not reasonably resolve without conducting a hearing. For example, although the Illinois Appellate Court had ascribed Rodgon’s failure to investigate and present exculpatory witnesses to a legitimate concern that such witnesses might have presented a problem of guilt by association for Hampton (App.Ct. Order at 7), there in fact was no record as to Rodgon’s efforts and thinking as to such witnesses. The only available evidence as to Rodgon’s thinking came from the post-conviction hearing regarding Hampton’s own failure to testify. See supra at 227-228 & n. 4. Because the postconviction court had summarily dismissed Hampton’s ineffectiveness claim, the state court record was never developed as to what Rod-gon knew about exculpatory occurrence witnesses, what steps he may have taken to identify and speak with such witnesses, or his reasons (if any) for neither investigating nor presenting such witnesses. Indeed, as the evidentiary hearing in the district court revealed, there were disputes between Hampton and Rodgon as to whether or not Hampton had identified potential witnesses for Rodgon, disputes that could only be resolved after the court heard their testimony and assessed their credibility. See Bruce v. United States, supra, 256 F.3d at 598-99 (district court *245abused its discretion in refusing to conduct an evidentiary hearing on failure to investigate claim made under 28 U.S.C. § 2255, where the affidavits of trial counsel, defendant, and prospective alibi witnesses presented questions of fact as to whether counsel adequately assessed the potential testimony of these witnesses). Moreover, the district court could not assess what impact the exculpatory eyewitnesses likely would have had upon Hampton’s trial without hearing their testimony. Id. at 600. We observed in Matheney that “[a]n adequate record is imperative to properly evaluate ineffective assistance claims,” 253 F.3d at 1040 (citing United States v. Draves, 103 F.3d 1328, 1335 (7th Cir.), cert. denied, 521 U.S. 1127, 117 S.Ct. 2528, 138 L.Ed.2d 1028 (1997)), and in this case an evidentiary hearing was necessary in order to supply such a record. See also United States ex rel. Cosey v. Wolff, 682 F.2d 691, 693-94 (7th Cir.1982) (per curiam).
B. Merits
“The Antiterrorism and Effective Death Penalty Act of 1996 modified a federal habeas court’s role in reviewing state prisoner applications in order to prevent federal habeas ‘retrials’ and to ensure that state-court convictions are given effect to the extent possible under law.” Bell v. Cone, 535 U.S. 685, 693, 122 S.Ct. 1843, 1849, 152 L.Ed.2d 914 (2002), citing Williams v. Taylor, 529 U.S. 362, 403-04, 120 S.Ct. 1495, 1518, 146 L.Ed.2d 389 (2000). In order to obtain habeas relief under the AEDPA, a petitioner must establish that the proceedings in state court resulted in a decision (1) “that was contrary to, or involved an unreasonable application of, clearly established Federal law, as determined by the Supreme Court of the United States,” or (2) “that was based on an unreasonable determination of the facts in light of the evidence presented in the State court proceeding.” 28 U.S.C. § 2254(d). The relevant decision, for purposes of this assessment, is the decision of the last state court to rule on the merits of the petitioner’s claim—here, the order of the Illinois Appellate Court. E.g., Schultz v. Page, 313 F.3d 1010, 1015 (7th Cir.2002), cert. denied, — U.S. —, 123 S.Ct. 2220, 155 L.Ed.2d 1107 (2003).
Hampton appropriately does not contend that the Illinois Appellate Court’s decision is “contrary to” clearly established law; the court cited and applied the familiar rules governing claims of attorney ineffectiveness as set forth in the seminal Supreme Court precedent on that subject, Strickland v. Washington, supra, 466 U.S. 668, 104 S.Ct. 2052, 80 L.Ed.2d 674. See Woodford v. Visciotti, 537 U.S. 19, 123 S.Ct. 357, 359-60, 154 L.Ed.2d 279 (2002) (per curiam). Instead, Hampton maintains that the court unreasonably applied those standards. “A state-court decision that correctly identifies the governing legal rule but applies it unreasonably to the facts of a particular prisoner’s case certainly would qualify as a decision ‘involving] an unreasonable application of ... clearly established Federal law.’ ” Williams, 529 U.S. at 407-08, 120 S.Ct. at 1520. “Unreasonable” means something more than “mistaken,” however. Visciotti, 123 S.Ct. at 360, 361; Williams, 529 U.S. at 410, 120 S.Ct. at 1522; Hardaway v. Young, 302 F.3d 757, 762 (7th Cir.2002), cert. denied, — U.S. —, 123 S.Ct. 1802, 155 L.Ed.2d 668 (2003). A state court decision is unreasonable for purposes of section 2254(d)(1) if its application of Supreme Court precedent “l[ies] well outside the boundaries of permissible differences of opinion.” Id., citing Williams, 529 U.S. at 411, 120 S.Ct. at 1522; see also Visciotti, 123 S.Ct. at 361 (“[t]he federal habeas scheme ... authorizes federal-court intervention only when a state-court decision is objectively unrea*246sonable”). “Our task is to uphold those outcomes which comport with recognized conventions of legal reasoning and set aside those which do not.” Ward v. Sternes, 334 F.3d 696, 703 (7th Cir.2003).
In relevant part, Hampton’s habeas petition contends that he was denied his Sixth Amendment right to the effective assistance of trial counsel. In order to succeed on a claim of attorney ineffectiveness, a petitioner must demonstrate both that his counsel’s conduct fell below an objective standard of reasonableness and that his counsel’s sub-standard performance prejudiced him. Strickland, 466 U.S. at 687, 104 S.Ct. at 2064.
The first Strickland showing entails proof that the petitioner’s trial attorney “made errors so serious that counsel was not functioning as the ‘counsel’ guaranteed the defendant by the Sixth Amendment.” Ibid. In assessing the adequacy of an attorney’s performance, the court’s scrutiny of course must be “highly deferential,” id. at 689, 104 S.Ct. at 2065, allowing ample room for differences of professional opinion among attorneys as to how one might best represent the defendant, id. at 689-90, 104 S.Ct. at 2065-66.
A fair assessment of attorney performance requires that every effort be made to eliminate the distorting effects of hindsight, to reconstruct the circumstances of counsel’s challenged conduct, and to evaluate the conduct from counsel’s perspective at the time. Because of the difficulties inherent in making the evaluation, a court must indulge a strong presumption that counsel’s conduct falls within the wide range of reasonable professional assistance; that is, the defendant must overcome the presumption that, under the circumstances, the challenged action “might be considered sound trial strategy.” See Michel v. Louisiana, [350 U.S. 91, 101, 76 S.Ct. 158, 164 (1955) ].
Strickland, 466 U.S. at 689, 104 S.Ct. at 2065; see also Cone, 535 U.S. at 698, 702, 122 S.Ct. at 1852, 1854.
The second prong of the Strickland inquiry requires a demonstration of prejudice, that is, proof that “counsel’s errors were so serious as to deprive the defendant of a fair trial, a trial whose result is reliable.” 466 U.S. at 687, 104 S.Ct. at 2064. Prejudice is established when “there is a reasonable probability that, but for counsel’s unprofessional errors, the result of the proceeding would have been different.” Id. at 694, 104 S.Ct. at 2068. This standard does not require the petitioner to convince the court that his attorney’s ineffectiveness “more likely than not altered the outcome in the case.” Id. at 693, 104 S.Ct. at 2068; see also Visciotti, 123 S.Ct. at 359; Nix v. Whiteside, 475 U.S. 157, 175, 106 S.Ct. 988, 998, 89 L.Ed.2d 123 (1986); White v. Godinez, 301 F.3d 796, 804 (7th Cir.2002). Rather, a “reasonable probability” is one “sufficient to undermine confidence in the outcome.” Id. at 694, 104 S.Ct. at 2068. Even if the odds that the defendant would have been acquitted had he received effective representation appear to be less than fifty percent, prejudice has been established so long as the chances of acquittal are better than negligible. Miller v. Anderson, 255 F.3d 455, 459 (7th Cir.2001), judgment modified, 268 F.3d 485 (7th Cir.2001).
1. Failure to Investigate Exculpatory Eyewitnesses
Hampton’s ineffectiveness claim is based primarily upon what he contends was Rodgon’s failure to conduct a reasonable pretrial investigation by contacting the occurrence witnesses whose names had been give to him and by failing make any *247effort of his own to locate and contact other eyewitnesses.
The duty to investigate derives from counsel’s basic function, which is “ ‘to make the adversarial testing process work in the particular case.’ ” Kimmelman v. Morrison, 477 U.S. 365, 384, 106 S.Ct. 2574, 91 L.Ed.2d 305 (1986) (quoting Strickland, 466 U.S. at 690, 104 S.Ct. 2052, 80 L.Ed.2d 674). “Because that testing process generally will not function properly unless defense counsel has done some investigation into the prosecution’s case and into various defense strategies, [the Supreme Court has] noted that ‘counsel has a duty to make reasonable investigations or to make a reasonable decision that makes particular investigations unnecessary.”’ Id. (quoting Strickland, 466 U.S. at 691, 104 S.Ct. 2052, 80 L.Ed.2d 674).
Brown v. Sternes, 304 F.3d 677, 691 (7th Cir.2002). The Illinois Appellate Court characterized Rodgon’s decision not to present testimony from eyewitnesses who might have testified favorably for Hampton as a strategic decision that “cannot” support a claim of attorney ineffectiveness. App.Ct. Order at 6. The Supreme Court has, true enough, observed that “strategic choices made after thorough investigation of law and facts relevant to plausible options are virtually unchallengeable.” Strickland, 466 U.S. at 690, 104 S.Ct. at 2066 (emphasis supplied). The Court added, however, that “strategic choices made after less than complete investigation are reasonable precisely to the extent that reasonable professional judgments support the limitations on investigation.” Id, at 690-91, 104 S.Ct. at 2066 (emphasis supplied). We also pointed out in Crisp v. Duckworth that “[although there may be unusual cases when an attorney can make a rational decision that investigation is unnecessary, as a general rule an attorney must investigate a case in order to provide minimally competent representation.” 743 F.2d 580, 583 (7th Cir.1984), cert. denied, 469 U.S. 1226, 105 S.Ct. 1221, 84 L.Ed.2d 361 (1985). Thus, among the questions posed by Hampton’s claim are whether Rodgon’s omission to contact and present exculpatory eyewitness was indeed a strategic decision, whether that decision was based on a complete or incomplete investigation, and whether, if based on an incomplete investigation, “reasonable professional judgment” supported a limited investigation. See Montgomery v. Petersen, 846 F.2d 407, 413 (7th Cir.1988) (“counsel [has] a duty to contact a potential witness unless counsel ‘can make a rational decision that investigation is unnecessary’ ”) (quoting Crisp, 743 F.2d at 583).
The district court made several threshold factual determinations that have a substantial impact on our evaluation of the ineffectiveness claim insofar’ as the claim rests on Rodgon’s failure to investigate exculpatory eyewitnesses. The ■ court found first that Hampton had given Rodgon the names of three eyewitnesses (Garner, Gregory Mallory, and Williams), and in so finding rejected Rodgon’s testimony to the contrary. 2001 WL 1518533, at *8. It found further that Rodgon had not only failed to contact not only these witnesses, but had also failed to independently pursue other available eyewitnesses. Id. Third, except as to Garner, the court found that Rodgon had no tactical or strategic reason for not following up with such witnesses. Id. at *8, *16. Although the State does not expressly challenge these findings as clearly erroneous, see Foster v. Schomig, 223 F.3d 626, 634 n. 4 (7th Cir.2000), cert. denied, 532 U.S. 944, 121 S.Ct. 1407, 149 L.Ed.2d 349 (2001), it does nibble about their edges, taking issue with the way in which the court evaluated some of the evidence underlying its findings.
*248The State does not appear to quarrel with the district court’s finding that Hampton gave the names of Garner, Gregory, and Williams to Rodgon. In its recitation of the evidence presented to Judge Kennelly, the State points out that Rodgon denied having been given these names by Hampton, insisting that he would have investigated had he been given the names of such witnesses. State Br. at 25. Hampton, on the other hand, testified that he did give Rodgon this information. The choice of whom to believe fell to Judge Kennelly, who heard them both testify. See Foster, 223 F.3d at 634 n. 4; see also Sullivan v. Fairman, supra, 819 F.2d at 1392-93. Rodgon’s 1987 letter to the ARDC, which indicated that he and Hampton did discuss potential (unnamed) witnesses, figured into the judge’s credibility assessment. ' See 2001 WL 1518533, at *7. The State suggests that the district court gave the ARDC letter too much weight, noting that Rodgon’s mere reference to potential witnesses “does not mean that he failed to investigate any witnesses.” State Br. at 50. That is true enough. However, the principal inference that Judge Kennelly drew from the letter was that Hampton, contrary to Rodgon’s testimony, had identified potential defense witnesses for Rod-gon. 2001 WL 1518533, at *7. Considered along with Hampton’s testimony, that inference was not unreasonable. Judge Kennelly’s finding that Hampton did give the names of Garner, Williams, and Gregory Mallory to Rodgon is amply supported by the record and is not clearly erroneous.
The State appears to take some issue with the district court’s, next finding — that Rodgon made no attempt to contact the witnesses whose names Hampton had given him or to identify other potentially exculpatory eyewitnesses on his own. The record, we should point out, is devoid of any evidence that Rodgon actually did make such an attempt. Rodgon himself simply denied that Hampton had given him the names of any witnesses and said that he did not know who, other than Hampton’s co-defendants, might be favorable occurrence witnesses.13 The district court found it noteworthy that Rodgon’s trial file lacked any memoranda or notations indicating that Rodgon (or an investigator) had spoken with any witnesses. 2001 WL 1518538, at *7. The State suggests that there may be explanations other than a failure to interview witnesses for the lack of documentation in Rodgon’s file. State Br. at 49. Rodgon did testify that he had made notes that, by the time of the hearing before Judge Kennelly, were missing from the file. “I had notes. I know I had notes. I don’t know where they are because the file got split up, and it has been 20 years almost, 19 years.” R. 59-1 at 93. But Rodgon did not specify what type of notes these were, nor was he able to identify any occurrence witness that he did contact. In the absence of any evidence that Rodgon had spoken with such witnesses, Judge Kennelly permissibly concluded that Rodgon made no effort to locate and interview potentially exculpatory eyewitnesses. Again, his finding is amply supported by the record and is not clearly erroneous.14
Notably, the State mounts no direct challenge to the district court’s finding *249that Rodgon’s failure to pursue exculpatory eyewitnesses was not a strategic decision.15 The district court observed that although the Illinois Appellate Court had labeled it as such, this finding “ ‘rests on thin air’ ” and consequently was entitled to no deference. 2001 WL 1518538, at *15 (quoting Mendiola v. Schomig, 224 F.3d 589, 592 (7th Cir.2000), cert. denied, 533 U.S. 949, 121 S.Ct. 2591, 150 L.Ed.2d 750 (2001)). “Neither the state trial court nor the Appellate Court had any basis to determine why Rodgon had not interviewed Hampton’s witnesses (either the occurrence witnesses or the ‘character’ witnesses); the finding that it was a strategic move was entirely speculative.” Id. Our own review of the record confirms the district court’s assessment: there is no evidence to support the notion that Rod-gon made a strategic decision not to look for exculpatory eyewitnesses. As the district court put it, Rodgon’s omission to pursue such witnesses was one that occurred by default rather than design. Id. at *16.
In any case, an attorney’s decisions are not immune from examination simply because they are deemed tactical. Miller, 255 F.3d at 458; Crisp, 743 F.2d at 584. Strickland itself makes clear that “strategic choices made after less than complete investigation are reasonable precisely to the extent that reasonable professional judgments support the limitations on investigation.” 466 U.S. at 690-91, 104 S.Ct. at 2066 (emphasis ours). Rodgon’s decision not to call exculpatory eyewitnesses in Hampton’s defense — if it was a decision at all — was necessarily one made after an incomplete investigation, for Rod-gon never spoke with such witnesses to find out what they had to say. Only if it was objectively reasonable for Rodgon to self-limit his investigation in this way may his “decision” not to present exculpatory eyewitnesses itself be considered reasonable. Ibid.; see also Montgomery, 846 F.2d at 413; Crisp, 743 F.2d at 584. As we conclude below, given the circumstances confronting Rodgon, it was not reasonable for Rodgon to believe that it was unnecessary to identify and interview potentially exculpatory eyewitnesses to the events underlying the charges against Hampton.
Eyewitness testimony was the linchpin of the State’s case against Hampton. *250There was no physical evidence tying Hampton to the attack upon the three Latino cbncertgoers. The only evidence implicating Hampton in the attack — beyond Hampton’s conceded presence at the concert — came from the three prosecution witnesses (victims Denise M. and Martha N., and William Heinrichs, the security guard) who identified Hampton as one of the attackers, and (to a lesser extent) from Keith Powell, who identified Hampton as among the group of individuals marching toward the stage of the auditorium but had not seen Hampton attack anyone. None of the three individuals who identified Hampton as an assailant knew Hampton, and none of those three witnesses had more than a momentary glimpse of the assailant whom they identified as Hampton. They got that look under conditions that were anything but ideal: the attack was chaotic and perpetrated by a large group of people, and the house lights in the Amphitheatre had been dimmed for the concert. Moreover, by their own account, none of the witnesses had, prior to seeing a line-up, provided a physical description of the assailant they identified as Hampton to the authorities.
Under these circumstances, the first and most obvious line of attack on the prosecution’s case was to emphasize the vulnerabilities in the identification testimony; and this Rodgon did with vigor. Through his cross-examination of the State’s witnesses, Rodgon competently highlighted each of the circumstances that might have given the jury reason to doubt the ability of the State’s witnesses to identify Hampton beyond a reasonable doubt as one of the assailants.
But whatever the weaknesses in the State’s ease that Rodgon succeeded in exposing in this fashion, his defense of Hampton did not elicit any testimony that Hampton was not among the group of individuals who attacked the three victims. The omission of such exculpatory evidence cannot be discounted. None of the witnesses who identified Hampton as an assailant was proven to be wholly incredible, and none of the weaknesses and inconsistencies in their identifications was fatal to the State’s case. Each of them testified that it was Hampton she or he had seen, notwithstanding the difficult conditions under which they had seen him. Opposing testimony from other eyewitnesses to the attacks, positing that Hampton was not a participant, would have given the jury a qualitatively different and more powerful reason to believe that the State’s witnesses were mistaken in their identifications of Hampton. See Washington v. Smith, supra, 219 F.3d at 634 (additional alibi witnesses that attorney failed to contact “would have added a great deal of substance and credibility to [petitioner’s] alibi”); Wright v. Gramley, supra, 125 F.3d at 1042 (where state’s case against petitioner relied almost exclusively on testimony of two eyewitnesses who saw assailant only briefly, occurrence witnesses who gave physical description of assailant “radically different” from petitioner’s appearance “would have transformed a relatively weak defense into a far stronger one”); Crisp, 743 F.2d at 585 (“[h]aving independent witnesses corroborate a defendant’s story may be essential”); United States ex rel. Cosey v. Wolff, 727 F.2d 656, 658 n. 3 (7th Cir.1984) (witnesses whose names petitioner had given to his counsel “would not only have corroborated [petitioner’s] story and further impeached the victim’s version, but ... if the witnesses were believed, their testimony alone would have entirely exculpated [petitioner]”), overruled on other grounds, United States v. Payne, 741 F.2d 887, 891 n. 4 (7th Cir.1984) (per curiam). The Illinois Appellate Court’s assertion that such testimony would have been “redundant” is plainly *251wrong; testimony by one eyewitness to a crime that the perpetrator was not the person named by another eyewitness is the antithesis of redundancy. See Washington, 219 F.3d at 634; Montgomery, 846 F.2d at 415; Crisp, 743 F.2d at 585; Cosey, 727 F.2d at 658 n. 3. For that very reason, the Fifth Circuit has recognized that “[t]he failure to interview eyewitnesses to a crime may strongly support a claim of ineffective assistance of counsel.” Bryant v. Scott, 28 F.3d 1411, 1415 (5th Cir.1994) (citing Gray v. Lucas, 677 F.2d 1086, 1093 n. 5 (5th Cir.1982), cert. denied, 461 U.S. 910, 103 S.Ct. 1886, 76 L.Ed.2d 815 (1983)).
As the evidence presented to Judge Kennelly makes clear, there were such exculpatory eyewitnesses available to the defense. Hampton provided Rodgon with the names and means of contacting three of these witnesses — Garner, Gregory Mallory, and Williams. Each of those witnesses knew Hampton, had attended the concert, had witnessed the attacks, and was prepared to testify that Hampton had not participated in those attacks. Attorneys representing Hampton’s co-defendants had located other eyewitnesses to the attacks who resided at the public housing project where Hampton lived and thus might have known or at least recognized Hampton. As the district court pointed out, Rodgon would have been able to identify those witnesses simply by consulting with his colleagues and/or investigating the people identified on their witness lists. At least one of those individuals, Gregory Hubbard, would have testified that Hampton did not participate in the attacks.
Like the district court, we shall assume that Rodgon had a valid reason not to contact Garner, who (unlike the other prospective -witnesses) had been charged with, and had pleaded guilty to, participating in the attacks. Rodgon was aware of Garner’s guilty plea, and he was also aware that the proffer of evidence that the State recited each time one of Hampton’s co-defendants pleaded guilty expressly implicated Hampton in the attacks. Garner was not asked to affirm the truth of the government’s proffer (neither did he contest its accuracy). Nonetheless, Garner’s guilty plea in the face of a proffer that expressly implicated both Hampton as well as himself obviously diminished his utility as a defense witness to some degree. But see Bryant, 28 F.3d at 1419 (without interviewing a co-defendant who had pleaded guilty, petitioner’s trial attorney “was ill equipped to assess his credibility or persuasiveness as a witness, despite the objective factors tending to impugn [his] credibility”); cf. Cosey, 727 F.2d at 658 n. 3 (although three of potential witnesses had reason to be biased in petitioner’s favor, “that alone is insufficient cause to automatically reject them”).
But Rodgon had no reason not to contact and interview the other eyewitnesses Hampton has identified. Those witnesses were not implicated in the attacks, nor did they suffer from any other disability that necessarily would have impaired their credibility in the eyes of the jury. Although the State has methodically identified various inconsistencies and imperfections among the recollections of those witnesses, such flaws — hardly surprising after the passage of nearly twenty years — cannot excuse Rodgon’s failure to contact and interview them. Certainly they would bear on the weight that a jury-assessing Hampton’s guilt might have attached to their testimony, and to that extent they are relevant to the assessment of prejudice from the failure to contact them, an assessment we are about to make. But they are not qualitatively different in kind from the weaknesses in the identification testimony of the State’s own witnesses. A decision that it was unnec*252essary to look for and contact such eyewitnesses cannot be described as reasonable. See Washington, 219 F.3d at 632 (trial counsel’s failure, inter alia, to ascertain what exculpatory evidence defendant’s witnesses might have was a “flagrant example[ ] of ineffectiveness”); Lord v. Wood, 184 F.3d 1083, 1095 (9th Cir.1999) (counsel cannot assess credibility and demeanor of prospective witness without “looking him in the eye and hearing him tell his story”), cert. denied, 528 U.S. 1198, 120 S.Ct. 1262, 146 L.Ed.2d 118 (2000). Notably, Rodgon himself offered no strategic reason for the failure; he simply denied having any leads on such witnesses — an explanation the district court found incredible. 2001 WL 1518533, at *7-*8.
The Illinois Appellate Court also thought that testimony from exculpatory eyewitnesses would have posed a problem of “guilt by association,” and for that reason it was reasonable for Rodgon not to contact such witnesses. App.Ct. Order at 7. Testimony from other eyewitnesses, the court explained, would have “only serve[d] to emphasize the fact that defendant went to the concert with the perpetrators, was present during the assault, and left with the perpetrators.” Id. The State sounds the same theme in its briefs. “Petitioner’s potential witnesses all would place him at the scene of the crime,” the State emphasizes. Reply Br. at 14. “The testimony of Williams, Mallory and Garner would have reinforce[d] the fact that Petitioner was [at the concert] and would have taken away from Rodgon’s strategy to distance Petitioner from the taint of association and vigorously pursue a misidentification theory.” State Br. at 54.
The “guilt by association” concern is altogether a red herring, however, given the context in which the attacks occurred. There was never any dispute, and the jury well knew, that Hampton was present at the concert; so were thousands of other people. Attending a rhythm and blues concert is neither a crime nor suggestive of participation in a crime; and the fact that a horrible assault occurred at the concert does not cast a pall of suspicion upon everyone who attended. Furthermore, only one of the four favorable eyewitnesses that Hampton has identified— Garner — was implicated in the attacks, and we have assumed that Rodgon had a valid reason not to contact him. The other three — Williams, Gregory Mallory, and Hubbard — were simply bystanders to the assaults, and so would not have fostered the notion that Hampton was linked to the perpetrators by testifying on his behalf. Even to the extent that these witnesses would have testified that they, along with Hampton, had attended and left the concert in the company of other individuals who were perpetrators (which is not surprising given that they all lived in the same neighborhood), they were no different in that respect from Powell, the State’s own witness, who knew and rode home on the bus with the individuals he heard bragging about the attacks.
For these reasons, we sustain the district court’s determination that Rodgon’s failure to investigate exculpatory eyewitnesses was objectively unreasonable. Such witnesses were readily available to Rodgon, the testimony of those witnesses, if believed, would alone have given the jury a reason to acquit Hampton, and Rod-gon had no legitimate strategic reason not to pursue such witnesses.
We further agree with the district court that Hampton was prejudiced by his attorney’s ineffectiveness. In explaining why it was unreasonable for Rodgon not to look for exculpatory eyewitnesses, we have already highlighted several of the circumstances that establish prejudice, so we will *253briefly recite them again here with belaboring them. With no physical evidence against Hampton, the prosecution’s case against him depended entirely on witnesses who testified that they saw him participate in the attacks. Those witnesses saw the assailant they identified as Hampton only briefly and under difficult circumstances, rendering their identifications of Hampton vulnerable to challenge. And although Rodgon effectively identified these vulnerabilities, he presented no testimony to affirmatively counter the prosecution’s witnesses — i.e., from eyewitnesses who would have testified that Hampton did not participate in the attacks. See, e.g., Wright, 125 F.3d at 1042-43; Cosey, 727 F.2d at 658 n. 3.
Given the central role that eyewitness testimony played in this case, the vulnerabilities in the testimony of the State’s eyewitnesses, and the shortcomings in human perception that so frequently render eyewitness testimony less rehable than other types of evidence, see Wright, 125 F.3d at 1043 n. 4 (collecting cases), we are more than satisfied that the failure to investigate exculpatory eyewitnesses likely affected the outcome of Hampton’s trial. The eyewitnesses that Hampton has identified, and whose testimony the district court found credible, would have given the jury a powerful reason to doubt Hampton’s culpability-
Two separate acquittals lend support to this notion. First, the jury acquitted Hampton of the attempted rape of Martha N. Recall that according to a written report of a line-up that Martha N. had viewed, she had picked Ezra Garner rather than Hampton as her assailant. Detective Ptak had characterized the report as inaccurate in this respect. Nonetheless, the report stood as evidence that Martha N., at least initially, had identified someone other than Hampton as the person who attempted to rape her. The jury’s decision to acquit Hampton on that charge, but not others, suggests that the report gave it reason to doubt the reliability of Martha N.’s identification in a way that it did not doubt the other witnesses against Hampton.16 Second, Ronald Mallory ultimately was acquitted of all charges. Recall that Denise M. had identified Mallory as one of the men who attempted to place his penis in her mouth. In his defense, Mallory had testified that he did not participate in the attack, and he called three additional witnesses who said the same thing (a fourth witness confirmed that he was not a gang member). Mallory’s acquittal demonstrates the importance of exculpatory eyewitness testimony and suggests that Hampton’s jury might have been swayed by such testimony. We recognize that the case against Mallory was not as strong as the case against Hampton, in the sense that only one person— Denise M. — identified him as an assailant. But the State’s case against Hampton was not qualitatively different from the one against Mallory; and Denise M., who was also one of Hampton’s accusers, testified on cross-examination that it was the men who attempted to put their penises in her mouth (including Mallory) on whose faces she had concentrated during the attacks. What substantively made Mallory’s case unique, beyond his own (self-serving) testimony that he did not participate in the attacks, was the exculpatory eyewitness testimony that Mallory presented. That testimony was virtually identical to the eyewitness testimony on which Hampton *254bases his ineffectiveness claim, and Mallory’s jury evidently found that testimony sufficiently persuasive to doubt his accuser.17
The State latches onto a number of omissions, potential biases, and inconsistencies in and among the testimonies and affidavits of Garner, Gregory, and Williams, and contends that in view of these vulnerabilities, the district court was wrong to believe that their testimony likely would have had any real impact on the outcome of Hampton’s trial. It notes, for example, that Garner, Gregory, and Williams were all long-time friends of Hampton’s who, like Hampton, were present at the scene of the crime; and Garner, it reminds us, pleaded guilty to participating in the attacks. Cf. Montgomery, 846 F.2d at 414 (stressing importance of independent witnesses); Crisp, 743 F.2d at 585 (same). The State also points to various incongruities among the affidavits that these three witnesses submitted. Hampton alleged in his affidavit that he snuck into the concert with Gregory Mallory and then sat in the center of the theater with Gregory and Garner until they left the concert. However, Gregory’s affidavit did not mention that he and Hampton had snuck into the theater together, nor did it specify with whom Gregory sat during the concert. Garner’s affidavit omitted any mention of Gregory. And Williams’ affidavit averred that Hampton snuck into the theater with Garner, and like Garner’s affidavit, it omits mention of Gregory. These omissions and inconsistencies, in the State’s view, give reason to doubt the veracity and exculpatory value of the eyewitnesses Hampton has identified. Moreover, the State asserts that Gregory Mallory proved himself to be a poor witness: as we have previously noted, Gregory did not recall exactly where he sat during the concert, did not remember that Hampton was tried along with Gregory’s brother Ronald and that the case was tried before three juries, and could not remember certain representations he had made in his affidavit; he also answered a number of the questions posed to him with answers like “I don’t know” and “Okay.”
A number of these asserted weaknesses are unremarkable. It is by no means surprising that the witnesses who would have exonerated Hampton were his friends— they were the people he was sitting with at the concert, and as such they were in a superior position to know whether or not he had joined the group of men who attacked the three Latinos. Their friendship with Hampton certainly is a circumstance that a factfinder would consider in weighing their credibility, but it is not so impeaching that one can wholly discount the import of their testimony and the effect that it might have had on Hampton’s jury. As for the asserted inconsistencies between the affidavits of Hampton and the other eyewitnesses, these are by and large inconsistencies of omission rather than outright conflicts. Because the affidavits were prepared many years after the relevant events took place, gaps and inconsistencies in the memory of the affiants are to be expected. See, e.g., Brice v. Nkaru, 220 F.3d 233, 240 (4th Cir.2000); Greben*255ick v. Chater, 121 F.3d 1193, 1200 (8th Cir.1997). The district judge, who in contrast to the state courts and to this court heard all but Williams testify, found them to be credible witnesses.
Two of Hampton’s witnesses — Garner and Gregory Mallory — did suffer from more serious credibility problems, but the district court by no means overlooked these flaws as the State suggests. The court agreed with the State that Rodgon could not be faulted for failing to contact Garner, who had pleaded guilty and in doing so had not contested the evidentiary proffer that inculpated Hampton; Garner’s testimony therefore played no role in the court’s decision to grant relief to Hampton. Gregory Mallory’s testimony did factor into the court’s assessment, but the court was candid about his shortcomings as a witness. Years of drug abuse may well have taken their toll on Gregory’s faculties, although Mallory denied that his memory was impaired. But the district court had to look backward and consider what effect Gregory’s testimony might have had in 1982, when Hampton and his co-defendants were tried. Gregory was not yet a heroin user then; moreover, he had not only testified in his brother Ronald’s successful defense, but he also had testified at that time that Hampton had not participated in the attacks. Under those circumstances, Judge Kennelly had reason to believe that Gregory’s current testimony was not the product of delusion or fabrication and that Gregory would have made a credible witness for Hampton in 1982.
For all of these reasons, Hampton was prejudiced by his attorney’s failure to look for and interview exculpatory occurrence witnesses. A diligent attorney would have made an effort to locate such witnesses and, upon learning what they had to say, surely would have put them on the witness stand to testify on Hampton’s behalf. Although we cannot say with confidence that it is more likely than not that Hampton would have been acquitted had such witnesses been presented, his chances of acquittal with the support of those witnesses certainly would have been better than negligible. The State’s case, although more than sufficient to convict Hampton, was not so overwhelming that the outcome of the trial was a foregone conclusion.
Although we have concluded that Hampton’s claim of ineffective assistance of counsel is meritorious, our work does not end there. Hampton is entitled to relief in habeas corpus only if the Illinois Appellate Court’s rejection of his ineffectiveness claim amounts to an objectively unreasonable application of Strickland. See Visciotti, 123 S.Ct. at 361; Williams, 529 U.S. at 411, 120 S.Ct. at 1522; ante at 245-246. For the reasons that follow—reasons that we have already touched upon—we believe that the Illinois Appellate court unreasonably applied both prongs of the Strickland test for ineffectiveness.
The Illinois Appellate Court recognized that Hampton’s trial counsel was required either to undertake a reasonable investigation or to make a reasonable decision that an investigation was unnecessary. App.Ct. Order at 6. As found by Judge Kennelly, Rodgon made no effort to locate and speak with potentially favorable occurrence witnesses. 2001 WL 1518533, at *8. Under Strickland, that failure to investigate can be excused only if the decision not to look for such witnesses was itself reasonable. 466 U.S. at 690-91, 104 S.Ct. at 2066. Yet, neither of the two explanations that the state court posited for Rodgon’s failure to pursue exculpatory eyewitnesses — (1) those witnesses would place Hampton at the scene of the crime and link him to the perpetrators of the assaults, and (2) their testimony would be redundant (App.Ct. *256Order at 7) — is plausible, as we have pointed out. It was never in dispute that Hampton attended the concert, and testimony that he did so would not have incriminated him any more than it would have incriminated the thousands of other people who attended but did not participate in the attacks. Nor would such testimony have linked Hampton to the other perpetrators. With the exception of Garner, whom we have assumed Rodgon was not obligated to contact in view of his guilty plea, none of the exculpatory witnesses Hampton has identified was implicated in the offenses. At the same time, eyewitness testimony to the effect that Hampton did not participate in the assaults logically cannot be written off as redundant. Whatever ammunition Rod-gon had amassed against the State’s witnesses, he had no witness pointing affirmatively to Hampton’s innocence. In short, the record supplies no sound reason for Rodgon to have concluded that it was unnecessary to investigate eyewitnesses who would have exculpated Hampton. The Illinois Appellate Court’s contrary assessment was therefore unreasonable. It is wholly inconsistent with the facts of the case. See, e.g., Rice v. McCann, 339 F.3d 546, 549 (7th Cir.2003) (state court’s application of federal law is reasonable where it is “ ‘at least minimally consistent with the facts and circumstances of the case’ ”) (quoting Sanchez v. Gilmore, 189 F.3d 619, 623 (7th Cir.1999), cert. denied, 529 U.S. 1089, 120 S.Ct. 1724, 146 L.Ed.2d 645 (2000)). It also stands in such tension with the duty to investigate acknowledged in Strickland that it was not only erroneous, but unreasonable. See Ward v. Sternes, supra, 334 F.3d at 705.
As for prejudice, the Illinois Appellate Court characterized the proof ■ against Hampton as overwhelming, so that even if Rodgon had looked for, contacted, and presented exculpatory eyewitnesses on Hampton’s behalf, there is no reasonable probability that the outcome of the trial would have been different. Again, we do not doubt that the trial record as it stands supports Hampton’s conviction. But as we have emphasized, the case against Hampton rested entirely on the testimony of eyewitnesses who said that he participated in the assaults, and the identification testimony of these witnesses was by no means invulnerable. Against that backdrop, one cannot summarily discount the possibility that the outcome of the trial might have been different had the jury heard credible testimony from other eyewitnesses that Hampton was not one of the assailants. Recall that under Strickland, Hampton need not convince the court that such testimony more likely than not would have resulted in his acquittal; he need only establish that this is a reasonable probability, a better than negligible likelihood. 466 U.S. at 693-94, 104 S.Ct. at 2068; Miller v. Anderson, supra, 255 F.3d at 459. The exculpatory eyewitnesses whom Rodgon never pursued have now been located, examined under oath, and found credible by the district judge. Obviously, we cannot predict how the jury would have resolved the credibility contest between the State’s witnesses and Hampton’s, but nothing in the record permits one to say that the jury inevitably would have discredited the exculpatory eyewitnesses. The appellate court’s one-sentence holding that Hampton could not establish prejudice (App.Ct. Order at 5) was not a reasonable application of Strickland.
The Illinois Appellate Court’s decision turned a blind eye both to the nature of the State’s case and to the importance of the eyewitnesses that Rodgon failed to locate and interview. Those witnesses, if believed, would alone have given the jury a reason to acquit Hampton, see Cosey v. Wolff, supra, 727 F.2d at 658 n. 3, and *257given the vulnerabilities in the testimony of the State’s witnesses, would have increased the odds of an acquittal. See Washington v. Smith, supra, 219 F.3d at 634; Wright v. Gramley, supra, 125 F.3d at 1042-43. Rodgon’s failure to seek out exculpatory eyewitnesses who were readily available to him cannot fairly be described as a reasonable decision. Nor can the possibility of a different outcome had Rodgon pursued this line of defense reasonably be quantified as no better than negligible. See Miller, 255 F.3d at 459. The state court’s application of Strickland was therefore unreasonable, and Hampton is entitled to relief.
2. Failure to Fulfill Promises Made in Opening Statement
In his opening statement, Rodgon made two representations that form the basis for Hampton’s second claim of ineffectiveness. He promised the jury first that “Mr. Hampton will testify and tell you that he was at the concert. Mr. Hampton will tell you that he saw what happened but was not involved with it.” R. 48-2 at 543. He went on to promise the jury, that “[t]he evidence will show that my client is not a member of any gang nor a part of - a gang .... ” Id. at 544. Rodgon did not deliver on either promise — Hampton did not testify, and the jury never heard any evidence establishing that Hampton lacked a gang affiliation.
It bears noting that the foundation for this claim is the broken promise as opposed to the decision not to pursue a particular line of testimony. We may assume, without deciding, that it was reasonable for Rodgon to advise Hampton not to testify and not to present testimony from other witnesses about his lack of gang ties; such decisions are often motivated by strategic considerations that command deference from the judiciary. E.g., Taylor v. United States, 287 F.3d 658, 662 (7th Cir.2002); Foster v. Schomig, supra, 223 F.3d at 631. But Rodgon promised the jury that it would hear from Hampton and that it would also hear evidence that he had no gang involvement, and he reneged on his promises without explaining to the jury why he did so. Turnabouts of this sort may be justified when “unexpected developments ... warrant ... changes in previously announced trial strategies.” Ouber v. Guarino, 293 F.3d 19, 29 (1st Cir.2002) (citing Dutton v. Brown, 812 F.2d 593, 598 (10th Cir.), cert. denied, 484 U.S. 836, 108 S.Ct. 116, 98 L.Ed.2d 74 (1987)); see also, e.g., Drake v. Clark, 14 F.3d 351, 356 (7th Cir.1994). However, when the failure to present the promised testimony cannot be chalked up to unforeseeable events, the attorney’s broken promise may be unreasonable, for “little is more damaging than to fail to produce important evidence that had been promised in an opening.” Anderson v. Butler, 858 F.2d 16, 17 (1st Cir.1988); see also Washington v. Smith, supra, 219 F.3d at 634 (failure to produce witness identified in notice of alibi and mentioned during voir dire gave rise to “negative inference” against the defendant). The damage can be particularly acute when it is the defendant himself whose testimony fails to materialize:
When a jury is promised that it will hear the defendant’s story from the defendant’s own lips, and the defendant then reneges, common sense suggests that the course of trial may be profoundly altered. A broken promise of this magnitude taints both the lawyer who vouchsafed it and the client on whose behalf it was made.
Ouber, 293 F.3d at 28.
In this case, the Illinois Appellate Court ascribed the unfulfilled promise that Hampton would testify to a change in trial strategy. App.Ct. Order at 8. In the post-*258conviction proceeding, Rodgon had testified that at the time of opening statements, he believed that Hampton would take the stand in his own defense. Subsequently, however, he raised with Hampton the possibility that his testimony might aggravate the possibility of the jury thinking him guilty.by association. He therefore communicated this concern to Hampton,-and Hampton made the decision not to testify. The postconviction judge found Rodgon credible on this point. R. 48-7 at D58-59. In the Appellate Court’s view, this change of strategy “cannot support an ineffective assistance of counsel claim.” App.Ct. Order at 8.
The potential disadvantages of Hampton’s testimony were ones that would have been obvious from the óutset of the case, however, and thus do not justify Rodgon’s decision to promise the jury that Hampton would testify and then renege on that promise. We háve already voiced our skepticism of the validity of the “guilt by association” theory; and the record does not shed any light on the reasons why Rodgon thought that Hampton otherwise might not hold up on cross-examination. But we may put these points aside and assume that Rodgon legitimately concluded that Hampton’s testimony posed a substantial risk to the defense for the reasons he cited. Nonetheless, the circumstances that gave Rodgon pause were entirely foreseeable at the time he made his opening statement. Hampton himself was obviously available to Rodgon before the trial started, such that Rodgon could have assessed his strengths and weaknesses as a prospective defense witness. If Hampton was a weak witness, Rodgon should have ascertained that before he announced that Hampton would take the stand; nothing in the record suggests that anything material to Hampton’s worth as a witness remained a secret until after Rodgon had made that promise. See Ouber, 293 F.3d at 29. Nor did the guilt by association issue emerge unexpectedly later in the trial. It was never a secret that Hampton was present at the concert, that he attended the concert with friends, or that he knew other individuals (for example, Garner) who were implicated in the attacks.
Thus, to the extent that Rodgon had legitimate reasons to conclude that Hampton should not testify, it was unreasonable for him to tell the jury that Hampton would take the stand. Nothing was to be gained from making that promise, only to renege upon it later without explanation. The jury was lead to believe that Hampton had a story to tell that was diametrically opposed to that of his accusers; it was told, in essence, that there were two versions of what occurred and that it would have the opportunity to evaluate Hampton’s own credibility in choosing between those versions. In the end, however, the jury never heard a second version of what occurred — from Hampton or any other eyewitness; it heard only the State’s account of events. And in that context, Hampton’s unexplained failure to take the witness stand may well have conveyed to the jury the impression that in fact there was no alternate version of the events that took place, and that the inculpatory testimony of the prosecution’s witnesses was essentially correct. See Harris v. Reed, supra, 894 F.2d at 879 (failure to present exculpatory testimony as promised “left the jury free to believe [the prosecution’s witness’s] account of the incident as the only account”); see also Ouber, 293 F.3d at 34. Rodgon’s decision to promise Hampton’s testimony and then to break with that promise was objectively unreasonable.
Although less important, we also cannot ignore the failure to present testimony that Hampton was not involved with a gang, an omission that the state appellate court noted (App.Ct. Order at 8) but did *259not explicitly address (see id. at 7-9). Powell’s testimony that the group of men who perpetrated the attacks had approached the stage of the theater chanting “Third World Disciple Nation” and making apparent gang signals with their hands suggested that the attacks were gang-related. Against that backdrop, evidence that Hampton was neither a gang member nor involved with a gang arguably was relevant to an assessment of Hampton’s guilt or innocence in the sense that it tended to make his participation in the attacks less likely. Just as a defendant’s affiliation with a gang may be probative of his relationship with other participants in a crime and his connection to the charged offense, see United States v. Thomas, 86 F.3d 647, 652 (7th Cir.), cert. denied, 519 U.S. 967, 117 S.Ct. 392, 136 L.Ed.2d 307 (1996), proof that the defendant lacks any gang affiliation may bear on the likelihood that he participated in criminal activity with unmistakable gang overtones. To be sure, evidence of a defendant’s gang affiliation or lack thereof is by no means disposi-tive of his guilt or innocence. But to the extent it sheds light upon the defendant’s relationship with other perpetrators of the crime, and perhaps on his motive to commit the crime, it is relevant nonetheless. Rodgon himself plainly recognized as much when he informed the jury that Hampton was not a gang member and, more to the point, told jurors that they would hear evidence to that effect. As the record reveals and as Judge Kennelly found, testimony along these lines was readily available to Rodgon; he simply failed to pursue it. We may assume that it would have been entirely reasonable for Rodgon not to look for and introduce such evidence in the absence of a promise that such evidence would be presented. But having created an expectation that the jury would hear evidence tending to disassociate Hampton from the group of men who perpetrated the attack, it was unreasonable for Rodgon not to follow through by eliciting testimony on this point. His failure to do so could only have undercut the credibility of the defense with the jury.
The Illinois Appellate Court’s determination that it was reasonable for Hampton Rodgon to make and then break these promises as a matter of evolving trial strategy was unreasonable. Making such promises and then abandoning them for reasons that were apparent at the time the promises were made18 cannot be described as legitimate trial strategy. Promising a particular type of testimony creates an expectation in the minds of jurors, and when defense counsel without explanation fails to keep that promise, the jury may well infer that the testimony would have been adverse to his client and may also question the attorney’s credibility. In no sense does it serve the defendant’s interests. Abandoning the promise may be necessary when things do not pan out as the attorney reasonably expected or the detriments of the promised evidence become clear only later. But for the reasons we have discussed, that was not the case here. Rodgon’s failure to call Hampton as a witness or to establish his lack of gang ties cannot be chalked up to a change in trial strategy. The Illinois Appellate Court’s effort to do so is inconsistent with the facts of the case, and unreasonable in that regard. See Rice v. McCann, supra, 339 F.3d at 549.
*260Although we agree with the district court that Rodgon’s breach of the promises he made in the opening statement was not so prejudicial that it would support relief in and of itself, the breach serves to underscore the more important failure to investigate exculpatory occurrence witnesses. Although Rodgon exploited what weaknesses there were in the State’s case, he elicited no affirmative evidence that Hampton did not participate in the attacks. The jury was given reason to question how good of a look the State’s witnesses had gotten of the assailant they identified as Hampton, but was presented with no evidence that the assailant was not Hampton. What is more, Rodgon promised the jury that it would hear Hampton testify to his innocence, and then failed to deliver on that promise. He promised the jury that it would hear evidence that Hampton was not involved in any gang, and then failed to deliver on that promise as well. Those broken promises themselves supplied the jury with reason to believe that there was no evidence contradicting the State’s case, and thus to doubt the validity of Hampton’s defense. See Harris, 894 F.2d at 879.
III.
For the reasons we have identified, we conclude Hampton was deprived of the effective assistance of trial counsel, in violation of the Sixth and Fourteenth Amendments to the Constitution. The Illinois Appellate Court’s contrary conclusion was the result of an unreasonable application of the principles identified in Strickland v. Washington. After careful review of the record, we are left with the definite and firm conviction that the result of Hampton’s trial is not reliable. We therefore agree with the district judge, who handled this case with commendable thoroughness, that Hampton is entitled to a writ of habe-as corpus.
AFFIRMED
. We shall refer to the these three individuals by their first names and last initials to respect their privacy.
. Other testimony indicated that on January 3, 1982, Denise had identified Ronald Mallory, not Knight, as the person who put his penis into her mouth; subsequently, on January 8, she identified Knight as the one who did this. R. 48-3 at 952-53, 957. At trial, Denise testified that both men had attempted to do this, but that neither one had succeeded. R. 48-3 at 1018, 1019.
. Chicago police Detective Thomas Ptak testified that Denise M. did supply the police with a description of her assailants. However, no such description was included in the police report that he prepared, R. 48-3 at 960.
. At the subsequent hearing on that issue, Rodgon explained that he had advised Hampton not to take the witness stand because Hampton would have testified that he knew everyone involved in the case, including the other defendants, that he had gone to the concert with some of the other defendants, that he had left with them, and that he had taken the bus home with them. R. 48-7 at D32. In Rodgon’s view, the specter of guilt by association was a "terrible problem.” Id. at D33. Rodgon testified that he conveyed this concern to Hampton, but left it to his client to decide whether or not to testify. Id. at D31. *228Hampton, by contrast, testified that Rodgon had never informed him that he had a right to testify and did not explain that the decision whether or not to testify was his rather than Rodgon's to make. Id. at D7-8. Hampton said that he had told Rodgon during the State’s case that he wanted to testify. Id. at D9-11. But according to Hampton, Rodgon did not discuss the matter with him again until after Rodgon announced to the court that the defense was resting its case. Id. at D11. At that point, Hampton asked Rodgon why he could not take the witness stand, and Rodgon, according to Hampton, responded simply, “Don’t worry, everything will be all right.” Id.
After hearing both Rodgon and Hampton testify, Judge Moore credited Rodgon’s version of events over Hampton’s. See id. at D58-59. She found that Rodgon and Hampton had discussed the possibility of him testifying and what Hampton would say, that Rodgon had informed Hampton of his right to testify and that the decision was his as opposed to Rodgon's, and that Hampton had elected not to take the witness stand. Id. She also found, in view of the "overwhelming” testimony from the "numerous eyewitnesses” in the State’s case who identified Hampton as one of the perpetrators, that Rodgon had understandably changed his mind about Hampton testifying in his own defense. Id. at D57. "It is ... not unreasonable during the course of a trial or as a matter of concurring trial strategy between a defendant and his attorney to agree not to take the stand with the type of testimony that they had discussed Mr. Hampton testifying to, in light of the overwhelming evidence in the case.” Id.
. The court noted (App.Ct. Order at 8) but did not address Rodgon’s prediction that the evidence would show Hampton's lack of involvement with any gang. Hampton had raised this promise in his supplemental post-conviction petition (R. 48-5 at 125-27) and in his appeal (R. 17 Ex. B at 17-18).
. Hampton did have other two- or three-minute meetings with Rodgon during this period- — in court or in the holding cell adjacent to the courtroom — but, according to Hampton, on those occasions Rodgon simply explained to him what had occurred in court that day or what he expected to happen at the next court date. R. 59-1 at 11.
. The court reached a different conclusion with respect to the affidavit of Farod Poole, which Hampton had submitted in support of his supplemental habeas petition. According to that affidavit, Poole was with Hampton at the concert, he had witnessed the attacks, and he saw that Hampton was not a participant. R. 27 Ex. E ¶ 5. Poole was not mentioned in Hampton’s supplemental postconviction petition, however, and no affidavit from him was tendered to the state postconviction court. Although the district court remained convinced that Hampton had fairly presented his ineffectiveness claim to the state courts (because Poole’s affidavit did not alter the nature of Hampton’s claim), see 2001 WL 1518533, at *11, it found that consideration of Poole's affidavit was barred under section 2254(e)(2), id. at *14. Hampton had not established in state court what Poole’s testimony would have been, nor had he offered any explanation for the omission. Id. "Under the circumstances, the Court has no alternative but to find that Hampton lacked due diligence in state court in this regard, thus barring this Court’s consideration of Poole’s affidavit.” Id.
. Although the court’s analysis on this point was framed in terms of Gregory alone, the focus properly is on the need to hear him and the other witnesses Hampton has identified testify. Section 2254(e)(2) and procedural default rules deal with whether the. underlying factual basis for the claim was adequately developed in state court and whether state procedural rules were followed. The Townsend analysis, by contrast, addresses the need for the federal habeas court to conduct an evidentiary hearing in order to fairly resolve the petitioner’s claim. Here, for example, that analysis considers whether it is necessary to hear Hampton’s witnesses testify or whether his failure-to-investigate claim can be resolved solely on the affidavits, as the state courts did. This analysis therefore must take into consideration not only Gregory’s proffered testimony, but the entire body of evidence relevant to Hampton’s claim.
. Although section 2254(e)(2) does not require the petitioner to have explained his failure to develop the factual basis for his claim while still in the state forum, we acknowledge that the Illinois Post-Conviction Hearing Act did require Hampton either to attach to his postconviction petition "affidavits, records, or other evidence supporting its allegations” or to explain "why the same are not attached.” Ill. Stat. Ann. Ch. 38 ¶ 122-2 (Smith-Hurd 1990). Thus, Hampton and his counsel should have explained to the state postconviction court that their failure to submit an affidavit from Gregory Mallory was due to Gregory’s refusal to sign such an affidavit, just as the State suggests. Had the Illinois courts relied on the lack of such an explanatory affidavit as a reason to disregard Gregory’s potential testimony on Hampton’s behalf, Hampton’s failure to explain to the state courts why he had not submitted an affidavit from Gregory might have constituted a procedural default barring the federal courts from considering Gregory's testimony. As we note below, however, although the Illinois Appellate Court noted that Hampton had not submitted an affidavit from Gregory, it did not rely on the missing affidavit in disposing of Hampton’s ineffectiveness claim; neither did it note or rely on the lack of an explanation for the missing affidavit.
. It is not clear whether .the State also means to invoke section 2254(e)(2) in pressing this argument. However, for the same reasons we conclude below that the failure to submit an affidavit from Hubbard in state ■ court did not constitute a procedural default barring Hampton from relying on Hubbard's testimony, we would also conclude that Hampton did not fail to develop the factual basis for his failure-to-investigate claim insofar as it rests on Hubbard’s testimony.
. The State makes the same procedural default argument with respect to Farod Poole, an additional witness that Hampton did not name to Rodgon but, Hampton alleges, was readily discoverable had Rodgon simply made an effort to locate additional eyewitnesses. An affidavit from Poole was submitted in support of Hampton's supplemental habeas petition. R.27 Ex. E. The district court ultimately did not consider Poole’s affidavit, however, finding that Hampton’s failure to identify him as a potential witness and to tender an affidavit from him in support of his petition for *244postconviction relief in state court was due to a lack of diligence on Hampton's part. 2001 WL 1518533, at *14.
. The State suggests that section 2254(e)(2) displaces the pre-AEDPA Townsend analysis. However, our opinion in Matheney indicates that if section 2254(e)(2) by its terms does not apply (i.e., if the failure to develop the factual basis for the habeas claim in state court cannot be attributed to something that the petitioner did or failed to do), then the federal habeas court should consult pre-AEDPA standards to determine whether an evidentiary hearing on the petitioner's claim is warranted. 253 F.3d at 1039.
. Garner, Gregory, and Williams each averred that Rodgon had never contacted him.
. The State suggests that "Rodgon would [not] have been able to file numerous pretrial motions requiring extensive research, prepare discovery requests, conduct pretrial hearings and conduct lengthy and vigorous cross-examinations of all the State witnesses” without having conducted a more extensive pretrial investigation than reflected in the documentation remaining in the case file. State Br. at 49. This contention misses the point, however. The district court never suggested, let alone held, that Rodgon’s representation of *249Hampton was deficient in all respects. Consistent with the ineffectiveness claim that Hampton asserted, the court focused on Rod-gon's failure to look for eyewitnesses who might have exonerated Hampton. Rodgon’s representation of Hampton might well have been conscientious and effective in other respects, and we are obliged to consider Rodgon's performance as a whole. See Miller v. Anderson, supra, 255 F.3d at 458-59. But the possibility that Rodgon was competent in other aspects of his lawyering by no means rules out the possibility that his failure to investigate potential exculpatory defense witnesses amounted to constitutional ineffectiveness. Cf. Bryant v. Scott, 28 F.3d 1411, 1418-19 (5th Cir.1994).
. The State does suggest that in making this determination, the district court "closed its eyes” to the ramifications of one document found in Rodgon’s trial folder — -a copy of Ronald Mallory’s pretrial statement to prosecutors. State Br. at 49. Among other things, that statement asserts that Ronald Mallory did not see anyone except his brother Gregory at the concert, that he did not see Garner there at all, and that he never left the balcony of the Amphitheatre. R.58 Petitioner’s Ex. 4 at 4-5. In these particulars, Ronald Mallory's statement is inconsistent with the accounts given by Williams, Garner, Gregory Mallory, and Hampton. (Mallory admitted at trial that he had lied in his pretrial statement. R. 48-3 at 1158, 1175.) But those inconsistencies in and of themselves hardly would have justified a decision not to speak with the witnesses whose names Hampton gave to Rodgon or to look for additional eyewitnesses. See, e.g., Bryant v. Scott, supra n. 14, 28 F.3d at 1418.
. Recall that at one point during deliberations, the jury sent a note to Judge Strayhorn indicating that it was deadlocked on four of the charges against Hampton. R. 62 at 189. Although the deadlock obviously did not persist, it undercuts the notion that the case against Hampton was overwhelming in all respects.
. The postconviction court suggested (R. 48-7 at C35), as has the State (State Br. at 53, Reply Br. at 13) that the case against Hampton was significantly stronger given that the witnesses against him included William Heinrichs, the security guard who testified that he saw an assailant he identified as Hampton assaulting Denise M. But Heinrichs' credibility was far from unquestioned. Recall that Heinrichs, notwithstanding his employment with the Cook County Sheriff, did not come forward as a witness for more than a week after the attacks (after the police contacted him), and that he identified Hampton the day after he had seen a television news report featuring a photograph of Hampton.
. The record is silent as to Rodgon's reasons, if any, for abandoning the promise to present evidence that Hampton was not affiliated with a gang. In the absence of any indication that some subsequent development rendered Rodgon unable to present such evidence, we shall assume that Rodgon decided not to present such testimony for reasons akin to the concerns that led him not to call Hampton himself to the stand.