dissenting in part, concurring in part:
The principal issue in this appeal is whether the state court ruling on Rosario’s motion to vacate his conviction pursuant to New York Criminal Procedure Law § 440.10 was objectively unreasonable in holding that Rosario received effective assistance of counsel in accordance with the Sixth Amendment of the United States Constitution under Strickland v. Washington, 466 U.S. 668, 104 S.Ct. 2052, 80 L.Ed.2d 674 (1984). As I believe it was, I must respectfully dissent. Rosario raises two additional claims on appeal. Because I would conditionally grant Rosario’s petition on the basis of ineffective assistance of counsel, I believe it unnecessary to reach his claim under Batson v. Kentucky, 476 U.S. 79, 106 S.Ct. 1712, 90 L.Ed.2d 69 (1986). I concur only in the majority’s rejection of Rosario’s actual innocence claim.
This appeal presents an extraordinarily troubling set of circumstances. During the pendency of his prosecution, Rosario consistently maintained, both to the police and to his criminal defense attorneys, that he was in Florida on the day of the Bronx murder and on multiple occasions provided a list of up to thirteen alibi witnesses to corroborate this claim. Rosario’s defense attorneys nevertheless failed to investigate his alibi defense adequately and did not contact many of these potential witnesses. They offer no strategic reason for not doing so and, indeed, concede that such an investigation was essential to Rosario’s defense. Their explanation for this failure is that they mistakenly believed that the state trial court had denied Rosario’s application for fees to cover the investigatory expenses, when in fact the court had clearly granted the application. Such conduct *129plainly falls below acceptable professional standards, satisfying Strickland’s performance prong. Strickland, 466 U.S. at 687, 104 S.Ct. 2052.
As a result of this colossal failure, Rosario’s trial counsel presented a relatively weak alibi defense, consisting of only two alibi witnesses who were subject to impeachment as interested witnesses because they were close friends with Rosario. It is now clear that had Rosario’s defense attorneys followed through in investigating his alibi defense, they would have had the opportunity to call at least seven additional alibi witnesses at trial. These witnesses would have provided corroboration and supplied distinct facts relating to Rosario’s presence in Florida on and around the day of the murder, adding further context and credibility to his alibi defense; moreover, a number of these additional witnesses would not have been as vulnerable to impeachment as interested witnesses as were the two trial witnesses because they are not as close with Rosario. Moreover, the prejudice in this case is worsened because the only evidence of Rosario’s' guilt was the testimony of two stranger eyewitnesses. There is no question, in my opinion, that had the additional alibi witnesses who were presented in connection with Rosario’s post-conviction motion testified at trial, there is a reasonable probability that the jury’s verdict would have been different, satisfying the prejudice prong of the Strickland analysis. Id.
While the majority appears to agree with this much of the analysis, our opinions diverge where I further conclude that the state court’s holding to the contrary was not merely error, but an unreasonable application of Strickland. I come to this conclusion, as I must, because there exists too much alibi evidence that was not presented to the jury, and too little evidence of guilt, to now have any confidence in the jury’s verdict. In sum, I would conditionally grant the petition because it was objectively unreasonable both to sanction counsel’s failure to investigate Rosario’s alibi defense as reasonable and to find no reasonable probability that the verdict would have been different if the jury had heard the significant alibi evidence that Rosario’s defense attorneys neither uncovered nor presented.
I. Ineffective Assistance of Counsel
The majority does not dispute that Rosario received constitutionally ineffective assistance of counsel under Strickland, but views the state court’s decision to the contrary as within the bounds of permissible error. Engaging in the Strickland analysis is helpful to underscore why I must disagree with the majority’s conclusion that the state court did not unreasonably apply the precedent.
Under Strickland, to establish ineffective assistance of counsel, Rosario “must (1) demonstrate that his counsel’s performance fell below an objective standard of reasonableness in light of prevailing professional norms; and (2) affirmatively prove prejudice arising from counsel’s allegedly deficient representation.” Carrion v. Smith, 549 F.3d 583, 588 (2d Cir.2008) (internal quotation marks omitted). “To satisfy the first prong' — -the performance prong — the record must demonstrate that ‘counsel made errors so serious that counsel was not functioning as the “counsel” guaranteed the defendant by the Sixth Amendment.’ ” Wilson v. Mazzuca, 570 F.3d 490, 502 (2d Cir.2009) (quoting Strickland, 466 U.S. at 687, 104 S.Ct. 2052). “[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. 2052, and even “strategic choices made after less than complete investiga*130tion do not amount to ineffective assistance — -so long as the known facts made it reasonable to believe that further investigation was unnecessary,” Henry v. Poole, 409 F.3d 48, 63 (2d Cir.2005) (citing Strickland, 466 U.S. at 690-91, 104 S.Ct. 2052), cert. denied, 547 U.S. 1040, 126 S.Ct. 1622, 164 L.Ed.2d 334 (2006). By contrast, “omissions [that] cannot be explained convincingly as resulting from a sound trial strategy, but instead arose from oversight, carelessness, ineptitude, or laziness,” may fall below the constitutional minimum standard of effectiveness. Wilson, 570 F.3d at 502 (alteration in original) (quoting Eze v. Senkowski, 321 F.3d 110, 112 (2d Cir.2003)). To satisfy the second prong — the prejudice prong — a “defendant must show that there is a reasonable probability that, but for counsel’s unprofessional errors, the result of the proceeding would have been different. A reasonable probability is a probability sufficient to undermine confidence in the outcome.” Strickland, 466 U.S. at 694, 104 S.Ct. 2052.
A. Performance Prong
Defense counsel has a “duty to make reasonable investigations or to make a reasonable decision that makes particular investigations unnecessary.” Wiggins v. Smith, 539 U.S. 510, 521, 123 S.Ct. 2527, 156 L.Ed.2d 471 (2003) (quoting Strickland, 466 U.S. at 690, 104 S.Ct. 2052). Rosario’s pre-trial and trial counsel did neither. From his arrest to the present, Rosario has consistently maintained that he was in Florida on the day of the murder. At every juncture of this case, he has disclosed the substance of his alibi defense and the names of the individuals who could corroborate it, including in his post-arrest statement on the day he voluntarily surrendered to the police and to both of his defense counsel thereafter. Nevertheless, his attorneys abdicated their duty to investigate a majority of these individuals because of their mistaken belief that the state trial court had denied the application for fees to cover the expenses of such an investigation. This clearly satisfies the deficient representation prong of Strickland.
To be more specific, the record is undisputed that Rosario’s first counsel, Joyce Hartsfield, retained investigator Jessie Franklin, and, after Franklin’s unsuccessful attempt to contact several potential alibi witnesses by telephone, concluded that an on-the-ground investigation in Florida was necessary. Accordingly, Hartsfield applied to the trial court for fees to cover the cost of sending Franklin to Florida. The court ultimately granted the application, but Hartsfield failed to disclose this fact to Franklin. Franklin assumed the court had denied the application because Hartsfield never informed her otherwise and never ordered her to conduct the investigation. Steven Kaiser, Rosario’s second counsel, similarly labored under the erroneous impression that the court had denied the application and neglected to make any further inquiry into the matter. Whatever their reasons for harboring this mistaken belief, an on-the-ground investigation in Florida was never conducted. The direct and proximate result of this mistake was that Rosario’s defense team never contacted most of Rosario’s alibi witnesses.
To be clear, neither Hartsfield nor Kaiser claim that the failure to conduct this investigation was strategic; they admit it was a mistake. Hartsfield testified that in this case it was “critical” for the investigator to be able to meet the witnesses “in person and have a face-to-face conversation,” and that had Hartsfield realized that the application for fees had been granted she would have asked Franklin to go to Florida. Hartsfield unequivocally confirmed that her failure to interview additional witnesses was not strategic. Kaiser *131likewise testified that he relied upon the erroneous belief that the fee motion had been denied in limiting his investigation of Rosario’s alibi to evidence that could be gathered from New York, and repeatedly testified to the effect that he would have “loved to” call additional alibi witnesses if only they had been available to him.
Under these circumstances, there is simply no question that this mistake on the part of Rosario’s defense attorneys — and their resulting failure to investigate Rosario’s alibi properly — was constitutionally deficient under the Sixth Amendment. See, e.g., Williams v. Taylor, 529 U.S. 362, 396, 120 S.Ct. 1495, 146 L.Ed.2d 389 (2000) (concluding that counsel’s failure to uncover and present voluminous mitigating evidence at sentencing could not be justified as a tactical decision to focus on the defendant’s voluntary confessions because counsel had not “fulfilled] their obligation to conduct a thorough investigation of the defendant’s background”); Pavel v. Hollins, 261 F.3d 210, 220 (2d Cir.2001) (noting that “an attorney’s failure to present available exculpatory evidence is ordinarily deficient, unless some cogent tactical or other consideration justified it”); Maddox v. Lord, 818 F.2d 1058, 1061-62 (2d Cir.1987) (concluding that counsel would be constitutionally deficient if he “was aware of — but failed” for non-strategie reasons “to interview — a potential witness ... who was prepared to testify ... that he had diagnosed [petitioner] as being extremely emotionally disturbed prior to, and during, the commission of the crime”); Garcia v. Portuondo, 459 F.Supp.2d 267, 287-88 (S.D.N.Y.2006) (“[T]here is no reasonable trial strategy that would have excluded at least conducting interviews of the alibi witnesses to determine whether they could provide exculpatory evidence.”).4
B. Prejudice Prong
I also conclude that Rosario has satisfied the prejudice prong of Strickland. Because of defense counsel’s failure to properly investigate Rosario’s alibi defense, the only two alibi witnesses presented at trial were John Torres and Jenine Seda, who both testified that Rosario stayed with them in Deltona, Florida from approximately the end of April or beginning of May until about June 20, 1996. Specifically, they testified that Rosario was in Florida on June 19, 1996, the day of the murder, and that they remembered this because it was the day before the birth of their son. John5 further explained that on June 19, his car broke down and he spent the day with Rosario looking for car parts before they returned *132to his apartment together. Seda also testified that Rosario was at her apartment to see the baby on June 21, 1996, when she returned from the hospital. Rosario took the stand on his own behalf and testified that he was in Florida on the day of the murder and was staying with John and Seda for most of June 1996.
The prosecution successfully discredited the alibi defense presented at trial by convincing the jury that John, Seda and Rosario were lying. The first words from the prosecution during its summation were: “You [the jury] have to determine which witnesses were credible, which witnesses were believable, which witnesses had an interest in the outcome of this case.” The prosecution went on to argue, Rosario’s “saying he was in Florida. Look at the testimony to determine, can you rely on it? Is it believable? Is it credible?” Discrediting Seda and John, the prosecution argued:
First the two witnesses we heard. Jenine Seda and John Torres, the defendant’s friends. I would suggest to you, ladies and gentlemen, that those witnesses are interested witnesses, interested because they have an interest in the outcome of the case. They don’t want to see their friend go to jail. They don’t want to see their friend in trouble.
With respect to Rosario’s testimony, the prosecution noted at the outset that “the Judge will instruct you he is an interested witness.” The prosecution also emphasized that Rosario lied about staying with a woman in Florida during March and April of 1996 named Shannon Beane, whom he claimed to have been with every day, when in fact he had been incarcerated between March 13 and April 12. The prosecution argued:
Ladies and gentlemen, he took the stand. He put his hand on the Bible. He swore to tell the truth, and he told you I was with Shannon [B]ean[e]. I was with her daily, every day, and we know, ladies and gentlemen, from Captain Bolton that’s not true.
Ladies and gentlemen, I would suggest to you he doesn’t want you to know the truth about June 19th because to know the truth is to know that he was on White Plains Road, to know that he was on Turnbull Avenue, and to know that he was pumping a bullet [into] the head of George Collazo, and ending his life.
Ask yourself to what extent he would go to preventing you from knowing the truth. If he didn’t want you to know where he was in March and April of 1996, a time period which is insignificant since it has nothing to do with the commission of this crime, what would he do for the time period that really matters?
Ladies and gentlemen, use your common sense. Keep in mind that [Rosario] has an interest in this case.
The prosecution thus presented the jury with a choice: it could choose to believe two, disinterested eyewitnesses, or it could believe Rosario and his two good friends. It was a credibility battle. It is not shocking, therefore, that the prosecution secured a conviction. I conclude, however, that if the jury had been presented with the additional alibi evidence unearthed only by Rosario’s post-conviction team, there is a reasonable probability that the outcome at trial would have been different.
Rather than the slim alibi defense actually presented at trial, the jury would have been presented with a much stronger and more credible account of Rosario’s presence in Florida on the day of the murder and in the immediately surrounding period. Instead of disbelieving two alibi witnesses who were good friends with Rosario *133and Rosario himself, the jury would have had to discredit at least seven additional witnesses, who would have corroborated Rosario’s alibi, provided further context to his defense and testified to additional facts that had not been elicited at trial. Moreover, many of the additional witnesses are less interested in the outcome of the trial than were the trial witnesses and thus would have been less vulnerable to impeachment as interested witnesses.
The following alibi evidence was presented in connection with Rosario’s post-conviction motion. First, Chenoa Ruiz, neighbor of John and Seda and wife of John’s brother Robert Torres, testified that she saw Rosario about five times a week when he was living with John and Seda in June 1996, but that he moved out of their house and in with a friend named Ray who lived nearby when the baby was born. Chenoa testified that on the night of June 18, 1996 (the night prior to the murder), John, Seda, Robert and Rosario were at her and Robert’s apartment when Seda began to have contractions. Chenoa and another woman took Seda to the hospital without John, who chose instead to remain with Robert and Rosario. Seda was not kept in the hospital that night, but was told to return the next day for a scheduled appointment. The next morning (June 19, the day of the murder), Chenoa saw Rosario when she arrived at John and Seda’s apartment to take Seda to the doctor and saw him again when she and Seda returned home several hours later. Chenoa testified that she remembers this day in particular because she was annoyed that John was' “hanging out” with Rosario instead of tending to his pregnant girlfriend. Chenoa would have provided less interested testimony than John and Seda because she did not consider Rosario a friend.
Second, Fernando Torres, John’s father, testified that Rosario lived with John and Seda around the time that his grandson was born on June 20, 1996. Fernando testified that he was with Rosario in Florida on June 19, 1996, the day of the murder, because John’s car broke down and Fernando accompanied John and Rosario to purchase car parts. Fernando also saw Rosario in Florida the following morning: early on June 20, Fernando went to his son’s house and learned from Rosario that John and Seda were at the hospital. Finally, Rosario was again present at John and Seda’s apartment when Fernando met his grandson for the first time on June 21, the day Seda returned from the hospital. Fernando invited Rosario to church that day. In addition to providing additional facts not supplied by either John or Seda, a jury may have found Fernando more credible because he was not a friend of Rosario and thus undoubtedly a less interested witness. Additionally, Fernando is a generation older than Rosario, John and Seda, who were all in their twenties, which may have further bolstered his credibility over the trial witnesses. See, e.g., United States v. Liporace, 133 F.3d 541, 545 (7th Cir.1998) (approving instruction to jury that it may consider a witness’s age in assessing that witness’s credibility); cf. Washington v. Schriver, 255 F.3d 45, 59-60 & n.10 (2d Cir.2001) (implicitly approving same).6
*134A third witness — Michael Serrano, a corrections officer — testified that in June of 1996, he saw Rosario two or three times a week in the apartment complex where John, Seda, Robert and Chenoa lived, including in the days prior to the birth of John and Seda’s child. Though he did not know Rosario’s whereabouts on the day of the murder, Serrano testified that on the night that the baby was born (i.e., the day after the murder), Serrano and several other people, including Rosario, held an impromptu celebration in the parking lot of the apartment complex to congratulate John when he came home from the hospital.7 As with Ruiz, Serrano did not consider himself to be good friends with Rosario.
Fourth, Denise Hernandez, Rosario’s ex-girlfriend, testified that she saw Rosario in Florida around the time of the murder because they were dating throughout June 1996, and recalled in particular a big argument at some point in the middle to end of that month. Hernandez explained that one day, she and her friend were at her friend’s house getting ready to go out to a movie when Rosario took her car, without her permission, on a “joyride.” Hernandez was particularly upset because this incident occurred a few days before her sister’s birthday, which is on June 26, and her sister’s birthday present was in the car. As a result of this and other issues in their relationship, Hernandez broke up with Rosario at some point between her sister’s birthday and when Rosario returned to New York. It is true that Hernandez has maintained a close relationship with Rosario, even visiting him in prison on several occasions, and thus the prosecution presumably would have attacked her as an interested witness. Nevertheless, her testimony would have provided additional and distinct facts relating to Rosario’s whereabouts around the date of the murder and would have provided further context to his alibi defense.
Furthermore, a fifth witness, Hernandez’s friend Lyssette Rivera, testified that she was present when Rosario took Hernandez’s .car on the joyride and recalled the ensuing argument between Hernandez and Rosario and its proximity to Hernandez’s sister’s birthday (recalling that the argument occurred between five days and a week prior to the sister’s birthday). Thus, to the extent that Hernandez would have been subject to impeachment in light of her relationship with Rosario, defense counsel could have corroborated her testimony with that of Rivera, who — though she also had communicated with Rosario since his incarceration — did not have as close a relationship with him.
Sixth, Ricardo Ruiz, the brother of Chenoa, testified that he saw Rosario at John and Seda’s apartment during the month of June 1996 “[a]ll the time,” including before and after their baby was born. In particular, he testified that Rosario was in Florida “[a]t the time that [Seda] gave birth to [the baby].” He also testified that after Rosario moved out of John and Seda’s apartment, Rosario moved in with a friend named Ray, who lived across the street from John and Seda.
The seventh witness — Minerva Godoy, Rosario’s ex-fiancee — testified that Rosario left New York for Florida in May 1996, to relocate and find a job, and she did not see him again until the morning of July 1, 1996, when he claims to have returned to *135New York. Godoy testified that she was in regular contact with Rosario while he was in Florida, calling him at Fernando’s Florida telephone number and once wiring him money in Florida via Western Union. In particular, she testified that Rosario called her from Florida the day after Seda gave birth and told her that he was going to go see the baby.8
Because the prosecution’s case hinged so much on discrediting Rosario’s alibi defense, these additional -witnesses could have made all the difference in the world. Godoy could have provided the necessary context by testifying about Rosario’s departure from New York to Florida in May 1996, essentially serving as the first chapter of his alibi defense, and then about their meeting on July 1, 1996, providing the final chapter immediately prior to his surrender to the police. All of the other witnesses discussed above would have filled in the middle by testifying that they saw Rosario in their Florida community throughout June of 1996. They would have provided specific facts regarding where he lived and what he was doing at that time. Several witnesses could have corroborated each other’s testimony that Rosario was in Florida on the exact day of the murder and in the immediately surrounding days. Chenoa would have testified that she saw Rosario both the night prior to the murder, when she took Seda to the hospital, and twice throughout the day of the murder, both before and after Seda’s doctor’s appointment. John’s father Fernando would have placed Rosario in Florida on three consecutive days beginning with the day of the murder and would have corroborated John’s testimony that Rosario was with him looking for car parts on the nineteenth. From Chenoa and Fernando alone, the jury would have been provided additional concrete facts that Rosario was in Florida the night prior to, at various points the day of, and the morning following the murder — indisputably critical data points in establishing that Rosario was in Florida, and not over 1000 miles away in New York, when the victim was murdered.
Additionally, Serrano would have testified that he was with Rosario in the parking lot of John and Seda’s apartment complex on the night after the murder; Hernandez and Rivera would have provided consistent testimony about the fight between Hernandez and Rosario around the date of the murder; and Ricardo could have further corroborated Rosario’s general presence in Florida throughout June.
This additional evidence that the jury never heard would have provided the necessary context and corroboration for Rosario’s alibi defense. Moreover, as discussed, many of these witnesses were not vulnerable to impeachment as interested witnesses because they were not close friends with Rosario.9
I conclude that this evidence, taken together, clearly establishes a reasonable probability that the outcome of the trial would have been different had defense counsel investigated and presented this additional alibi evidence, satisfying Strickland’s prejudice prong. “Overall,” as Rosario argues, “if presented with the additional evidence at trial, a jury must disregard nine witnesses, as opposed to *136two, as mistaken or lying about seeing Rosario in Florida on and about June 19, 1996, before convicting him of the Bronx murder.” Brief for Rosario at 34. See Stewart v. Wolfenbarger, 468 F.3d 338, 359 (6th Cir.2006) (finding prejudice when defense counsel failed to call two additional alibi witnesses to corroborate the one alibi witness called at trial who was impeached because of his close association with the defendant); Washington v. Smith, 219 F.3d 620, 634 (7th Cir.2000) (finding prejudice when defense counsel failed to call three additional alibi witnesses to corroborate the one alibi witness at trial who had knowledge of the defendant’s whereabouts during the robbery, particularly when none of the additional witnesses, unlike the trial witness, had a criminal record); Montgomery v. Petersen, 846 F.2d 407, 415 (7th Cir.1988) (finding prejudice in failure to call additional, disinterested alibi witness, noting that “the jury might well have viewed the otherwise impeachable testimony of the twelve witnesses who were presented at the ... trial in a different light had the jury also heard the testimony of this disinterested witness”).
Further highlighting the prejudicial effect of defense counsel’s error in this case is the paucity of the prosecution’s case, which consisted of only two stranger eyewitnesses. We have consistently acknowledged that this sort of evidence is “proverbially untrustworthy.” Kampshoff v. Smith, 698 F.2d 581, 585 (2d Cir.1983); see also Gersten v. Senkowski, 426 F.3d 588, 613 (2d Cir.2005) (characterizing direct evidence consisting only of eyewitness testimony as “underwhelming”), cert. denied sub nom., Artus v. Gersten, 547 U.S. 1191, 126 S.Ct. 2882, 165 L.Ed.2d 894 (2006); Lyons v. Johnson, 99 F.3d 499, 504 (2d Cir.1996) (“[Tjhis court has noted on more than one occasion that eyewitness testimony is often highly inaccurate.”). Indeed, each year thousands of defendants in the United States are convicted for crimes that they did not commit, and many experts estimate that eyewitness error plays a role in half or more of all wrongful felony convictions. Richard A. Wise, Clifford S. Fishman & Martin A. Safer, How to Analyze the Accuracy of Eyewitness Testimony in a Criminal Case, 42 Conn. L.Rev. 435, 440 & n.12 (2009) (citing study showing that eyewitness error accounts for nearly sixty percent of all wrongful convictions).
In this case, there are reasons to be concerned with the two eyewitnesses’ accounts: the porter, Robert Davis, saw the shooter at a distance of more than two car lengths for only a few seconds, and although Michael Sanchez testified that he got a good look at the shooter, it was only for a short moment under very stressful conditions.10 This is of course not to say there was insufficient evidence to convict Rosario. But Strickland makes clear that “a verdict or conclusion only weakly supported by the record is more likely to have been affected by errors than one with overwhelming record support.” 466 U.S. at 696, 104 S.Ct. 2052. Such is the case here. See Lindstadt v. Keane, 239 F.3d 191, 204-05 (2d Cir.2001) (finding prejudice and reversing denial of writ of habeas corpus where trial counsel failed to investigate evidence that could have corroborated the petitioner’s alibi claims, and where the prosecution’s case rested on only two eyewitnesses and limited corroborating evidence); see also Espinal v. Bennett, 588 F.Supp.2d 388, 402, 407-08 (E.D.N.Y.2008) *137(granting habeas relief when defense counsel failed to investigate a statement provided by a potential alibi witness who might have corroborated the petitioner’s own testimony regarding his whereabouts on the day of the murder in a prosecution consisting primarily of two eyewitnesses, one of whose credibility was impeached), aff'd, 342 Fed.Appx. 711 (2d Cir. Aug.18, 2009) (unpublished disposition).
II. Habeas Corpus Standards
The majority essentially concedes a Strickland violation and that Rosario would be entitled to relief if this case arose on direct review but denies the writ out of deference to the state court. Pursuant to 28 U.S.C. § 2254, a federal court may not grant a writ of habeas corpus to a state prisoner “with respect to any claim that was adjudicated on the merits” by the state court unless the state court’s decision “was contrary to, or involved an unreasonable application of, clearly established Federal law, as determined by the Supreme Court of the United States.” 28 U.S.C. § 2254(d)(1). Under this principle of deference, habeas relief may not be granted merely upon a “conclusion that counsel’s performance was constitutionally inadequate.” Carrion v. Smith, 549 F.3d 583, 591 n.4 (2d Cir.2008). Rather, “petitioner must identify some increment of incorrectness beyond error in order to obtain habeas relief.” Jones v. West, 555 F.3d 90, 96 (2d Cir.2009) (quoting Sorto v. Herbert, 497 F.3d 163, 169 (2d Cir.2007)). Moreover, as the majority notes, “because the Strickland standard is a general standard, a state court has even more latitude to reasonably determine that a defendant has not satisfied that standard.” Knowles v. Mirzayance, 556 U.S. -, 129 S.Ct. 1411, 1420, 173 L.Ed.2d 251 (2009). Nevertheless, “the increment of incorrectness beyond error need not be great; otherwise, habeas relief would be limited to state court decisions so far off the mark as to suggest judicial incompetence.” Georgison v. Donelli, 588 F.3d 145, 154 (2d Cir.2009) (internal brackets omitted) (quoting Hoi Man Yung v. Walker, 468 F.3d 169, 176 (2d Cir.2006)).
A close review of the state court’s decision makes it entirely clear, however, that — even affording the state court its due deference — its decision rejecting Rosario’s claim was an unreasonable application of Strickland and should not stand.
At the outset, I note that the state court’s use of the “meaningful representation” standard led it to focus on certain factors that have little bearing on a proper Strickland analysis. And it appears to have done so at the expense of determining whether the undisputed mistakes made by Rosario’s defense counsel fell below objectively reasonable standards and, moreover, whether they caused him prejudice, as required under Strickland. Indeed, the state court relied heavily upon its finding that Rosario’s pre-trial and trial attorneys “represented [him] in a thoroughly professional, competent, and dedicated fashion.” It emphasized that “[b]oth attorneys filed all appropriate motions; within the scope of the information that was then available to them, an investigation was conducted; witnesses were examined and cross-examined adeptly, professionally and with clarity; Mr. Kasier’s opening and closing statements were concise and to the point; and, most importantly, a credible alibi defense was presented to the jury.” The state court went on to emphasize that counsel’s mistake as to the denial of the application for investigative fees “was not deliberate” and “does not alter the fact that both attorneys represented defendant skillfully, and with integrity and in accordance with the standards of ‘meaningful representation’ defined by [the New York state] appellate courts.” It wrote:
*138Defendant has tried to second-guess his trial counsel at almost every level of their representation. He has questioned the depth of their investigation, the scope and focus of cross-examination and argued that his alibi defense could have been better if they had only followed through on [the state trial court’s fee] order. His criticisms ignore the fact that Ms. Hartsfíeld and Mr. Kaiser ably, and professionally represented him at every stage of the case with integrity and in ways that were consistent with the standards of ‘meaningful representation’ described above.
... And Mr. Kaiser at trial was prepared, skillful, purposeful, thoughtful and creative.
This type of analysis is entirely at odds with Strickland and is not dispositive of whether Rosario’s defense counsel were ineffective under the Sixth Amendment. It is axiomatic that, even if defense counsel had performed superbly throughout the bulk of the proceedings, they would still be found ineffective under the Sixth Amendment if deficient in a material way, albeit only for a moment and not deliberately, and that deficiency prejudiced the defendant. See, e.g., Henry v. Poole, 409 F.3d 48, 72 (2d Cir.2005) (“[R]eliance on counsel’s competency in all other respeets[] ... fail[s] to apply the Strickland standard at all.” (internal citation and quotation marks omitted)), cert. denied, 547 U.S. 1040, 126 S.Ct. 1622, 164 L.Ed.2d 334 (2006); cf. Kimmelman v. Morrison, 477 U.S. 365, 386, 106 S.Ct. 2574, 91 L.Ed.2d 305 (1986) (noting that while “[i]t will generally be appropriate ... to assess counsel’s overall performance throughout the case in order to determine whether the identified acts or omissions overcome the presumption that a counsel rendered reasonable professional assistance,” a “failure to make reasonable investigations or to make a reasonable decision that makes particular investigations unnecessary,” may be constitutionally deficient irrespective of trial performance (internal quotation marks omitted)).
It is far from clear whether the state court realized this basic principle. In fact, the state court noted in a footnote that New York case law, in particular People v. Benevento, 91 N.Y.2d 708, 674 N.Y.S.2d 629, 697 N.E.2d 584 (1998), “expressly rejected” Strickland’s requirement “that, but for the attorneys[’] errors, the result of the proceeding would have been different.” This footnote, viewed in context with the entirety of the court’s decision, begs the question whether the state court understood that New York state’s “ineffective assistance cases have departed from the second (‘but for’) prong of Strickland,” only to “adopt[] a rule somewhat more favorable to defendants.” People v. Turner, 5 N.Y.3d 476, 480, 806 N.Y.S.2d 154, 840 N.E.2d 123 (2005) (emphasis added) (citing People v. Caban, 5 N.Y.3d 143, 155-56, 800 N.Y.S.2d 70, 833 N.E.2d 213 (2005); People v. Stultz, 2 N.Y.3d 277, 284, 778 N.Y.S.2d 431, 810 N.E.2d 883 (2004); Benevento, 91 N.Y.2d at 713-14, 674 N.Y.S.2d 629, 697 N.E.2d 584). That is, it is unclear whether the state court appreciated that even if prejudice in the Strickland sense is not shown, a defense attorney can be found ineffective under the New York State Constitution if his performance was so below par that he did not provide “meaningful representation” to his client. See Caban, 5 N.Y.3d at 156, 800 N.Y.S.2d 70, 833 N.E.2d 213 (“[U]nder our State Constitution, even in the absence of a reasonable probability of a different outcome, inadequacy of counsel will still warrant reversal whenever a defendant is deprived of a fair trial---- [0]ur state standard thus offers greater protection than the federal test .... ”).
*139On a different note, at one point in the decision the state court sharply detoured into an analysis regarding newly discovered evidence. It wrote:
In order to prevail on a motion for a new trial based on a claim of newly discovered evidence, a defendant must establish by a preponderance of the evidence that evidence has been discovered since the trial which could not, with due diligence, have been produced at trial, and which is of such a character that, had it been presented at trial, there is a probability that the verdict would have been more favorable for him....
... [T]he existence of these witnesses was not new evidence discovered since the trial. They were known to defendant, who immediately gave their names to the police after his arrest, to his attorneys at their first and subsequent meetings, and to Jesse Franklin. Efforts were made to speak [to], and interview them and the substance of their testimony was known to the parties before the trial began.
It is unclear when, if ever, the court returned to the ineffective assistance of counsel analysis, and, more importantly, to what extent this detour infected that analysis. If this newly discovered evidence analysis did in fact bleed over to the ineffective assistance of counsel analysis, the harmful effect is patent, considering the obvious tension between a newly discovered evidence claim and an ineffectiveness claim based on an attorney’s failure to investigate an alibi that was disclosed to him by his client prior to trial.
It is true that a New York state court’s application of the meaningful representation standard does not necessarily result in error affording a petitioner habeas relief because the standard, properly construed, is more favorable to defendants. See Henry v. Poole, 409 F.3d at 68-71. It is also true that we do not grant habeas relief when a state court is merely inartful or unclear in its reasoning. But, in this case, it is entirely unclear to what extent the state court abandoned the Strickland analysis for a rule less favorable to defendants. Such an error would clearly be “contrary to” Strickland. 28 U.S.C. § 2254(d)(1).
The majority aptly pinpoints the “danger” of New York’s “meaningful representation” standard: though generally more protective of defendants’ rights than Strickland, it risks leading a court that “misunderstand[s] the New York standard” to “look past a prejudicial error as long as counsel conducted himself in a way that bespoke of general competency throughout the trial.” Ante at 126. The state court’s opinion provides strong indications that this is precisely what happened here. Yet the majority fails to address the very real likelihood that the state court fell victim to the danger it identified, merely concluding that, in general, when properly applied, the New York standard is not contrary to Strickland. Id. at 126.
Nevertheless, I “need not make a determination under the ‘contrary to’ clause, for [I] conclude that the ... Court’s rejection of [Rosario’s] ineffective-assistance-of-counsel claim was at least an objectively unreasonable application of Strickland.” Henry, 409 F.3d at 71. It is clear from the record that the state court not only unreasonably focused on counsel’s overall performance and minimized their mistakes, but also unreasonably discounted the alibi evidence adduced at the post-conviction hearing and thus undervalued its prejudicial effect.
In terms of Strickland’s performance prong, the state court recognized that counsel’s failure to complete their investigation was neither strategic nor the result *140of any sound trial strategy, but rather a “mistake.” The state court — as well as the majority — appears to excuse this mistake because it was “not deliberate,” counsel’s performance was otherwise “skillful[ ],” and counsel’ conducted some investigation leading to the presentation of a putatively “credible” alibi defense. But none of this excuses the fact that counsel essentially turned a blind eye to the existence of substantial potentially exculpatory evidence of which it was aware and, moreover, did so not on the basis of any “reasonable professional judgment,” Strickland, 466 U.S. at 690, 104 S.Ct. 2052, but rather as a result of pure inadvertence. Such conduct clearly falls below the threshold of minimal competence and, to the extent the state court found otherwise, I conclude that was an unreasonable application of Strickland.
With respect to prejudice, in relevant part, the state court reasoned:
[A]n alibi defense was presented through the two witnesses who had the best reason for remembering why defendant was present in Florida on June 19[,] 1996 — the birth of their son — an event that was more relevant for them than the events relied upon by the other witnesses .... Moreover, the alibi evidence offered by defendant at the hearing was in some cases questionable and in others raised issues which could have created questions for a deliberating jury. For example, two of the witnesses— Lisette Rivero [sic], and Denise Hernandez — could not say where defendant was on June 19 and 20. And Fernando Torres, when questioned about the purchase of auto parts years later, changed the date to three or four days before his grandson was born....
... It may not be cumulative to evidence presented at the trial — which largely was the case herein — and it must not be merely impeaching evidence....
For instance, Chenoa Ruiz recalled defendant’s presence in the Torres’ apartment on June 18 and 19, the two days prior to the birth of their child. And, Fernando Torres testified that he was with defendant and his son the day before his daughter-in-law gave birth. That testimony was cumulative to his son John’s trial testimony.
An investigator was not sent to Florida to interview witnesses. Nevertheless, the fact remains that the People’s case was strong, which was acknowledged by the Appellate Division when it affirmed the conviction herein. The prospective witnesses now before the court, studied closely, were, for the most part, questionable and certainly not as persuasive as the two witnesses who did testify, and were rejected by the jury.
First, the state court’s finding that “a credible alibi defense was presented to the jury” is hardly relevant to whether there is a reasonable probability of a different result had defense counsel presented a substantially more credible alibi defense. Second, the state court’s recognition that “an alibi defense was presented through the two witnesses who had the best reason for remembering why defendant was present in Florida on June 19, 1996 — the birth of their son — an event that was more relevant for them than the events relied upon by the other witnesses” also misses the point. It overlooks the fact that John and Seda were subject to impeachment as interested witnesses, and at least seven additional witnesses were available, a number of whom were less interested in the outcome of the trial, to corroborate their testimony, as well as add additional facts.
Third, although the court did find that “the alibi evidence offered by defendant at *141the hearing was in some cases questionable and in others raised issues which could have created questions for a deliberating jury,” it provided just three examples from a voluminous record in support of this finding, none of which bear scrutiny. It noted that “two of the witnesses — -Lisette Rivero [sic], and Denise Hernandez — could not say where defendant was on June 19 and 20.” But, as discussed, these witnesses testified to additional, non-cumulative facts that placed Rosario in Florida around the day of the murder. See ante at 135-36. The relevancy of this evidence is indisputable. The court also noted that “Fernando Torres, when questioned about the purchase of auto parts years later, changed the date to three or four days before his grandson was born.” This is simply not supported by the record. In fact, when asked whether he told Rosario’s post-conviction counsel that he went looking for car parts with his son and Rosario three or four days before his grandson was born, Fernando responded, “No, I don’t recall that at all,” and maintained that the excursion occurred on June 19.
Fourth, the state court found that the additional alibi witnesses were “largely ... cumulative.” To the extent that the additional alibi evidence corroborated John’s and Seda’s testimony, it is only reasonable to conclude that this militates in favor of a showing of prejudice. Again, John’s and Seda’s credibility was attacked by the prosecution. Corroboration was thus desperately needed. See, e.g., Washington v. Smith, 219 F.3d 620, 634 (7th Cir.2000) (“Evidence is cumulative when it ‘supports a fact established by existing evidence,’ Black’s Law Dictionary 577 (7th ed.1999), but Washington’s whereabouts on the day of the robbery was far from established— it was the issue in the case. The fact that Pickens had already testified to facts consistent with Washington’s alibi did not render additional testimony cumulative.”).
Finally, the state court characterized the People’s case as “strong.” But, the fact remains that it was based solely on the eyewitness accounts of two strangers — the type of evidence that this Court has repeatedly characterized as weak.
At bottom, the problem with the state court’s decision is its application of the reasonable probability standard. Contrary to the state court’s apparent belief, this standard does not require that the reviewing court be convinced of Rosario’s alibi defense. “[T]he reasonable-probability standard is not the same as, and should not be confused with, a requirement that a defendant prove by a preponderance of the evidence that but for error things would have been different.” Wilson v. Mazzuca, 570 F.3d 490, 507 (2d Cir.2009) (quoting United States v. Dominguez Benitez, 542 U.S. 74, 83 n.9, 124 S.Ct. 2333, 159 L.Ed.2d 157 (2004) (citing Kyles v. Whitley, 514 U.S. 419, 434, 115 S.Ct. 1555, 131 L.Ed.2d 490 (1995))). “A reviewing court looks instead to whether the probability of a different result is sufficient to undermine confidence in the outcome of the proceeding.” Id. (internal quotation marks omitted) (quoting Dominguez Benitez, 542 U.S. at 83, 124 S.Ct. 2333 (quoting Strickland, 466 U.S. at 694, 104 S.Ct. 2052)); see also Porter v. McCollum, 558 U.S. -, 130 S.Ct. 447, 455-56, — L.Ed.2d -(2009) (per curiam) (“We do not require a defendant to show ‘that counsel’s deficient conduct more likely than not altered the outcome’ of his penalty proceeding, but rather that he establish ‘a probability sufficient to undermine confidence in [that] outcome.’ ” (alteration in original) (quoting Strickland, 466 U.S. at 693-94, 104 S.Ct. 2052)).
Under the present circumstances, it is unreasonable to conclude that the probability of a different result is not sufficiently *142likely so as to undermine the confidence in the verdict. Defense counsel failed to investigate Rosario’s alibi defense and, as a result, did not call at least seven additional alibi witnesses. Instead, they proceeded with only two witnesses, both of whom were impeached as interested. In a credibility battle, such as this case, there is, to some extent, power in numbers — that is, if presented with the additional evidence at trial, the jury would have had to disregard a total of at least nine defense witnesses claiming to have seen Rosario in Florida on and around the day of the murder, as opposed to just two interested witnesses. As discussed, the additional alibi witnesses would have provided further context to and corroboration of Rosario’s alibi defense, would have testified to non-cumulative facts, and a number of them would have been less subject to impeachment than John and Seda.
The prosecution’s principal argument is that the additional alibi witnesses are not as reliable or credible as John and Seda. It emphasizes that Fernando, Chenoa, Rivera and Godoy provided less detailed accounts of their recollection during interviews pri- or to the 440.10 hearing than they did on the stand during the actual hearing. We have noted, however, that such “silence is so ambiguous that it is of little probative force.” Victory v. Bombard, 570 F.2d 66, 70 (2d Cir.1978) (quoting United States v. Hale, 422 U.S. 171, 176, 95 S.Ct. 2133, 45 L.Ed.2d 99 (1975)). The prosecution also emphasizes that Chenoa did not recollect certain facts, such as when Rosario traveled back and forth between Florida and New York during his previous trips and the precise date he left Florida at the end of June 1996. The fact that witnesses do not remember all relevant details is hardly surprising and certainly not dispositive as to whether they are reliable witnesses to the ultimate fact at issue, such as Rosario’s whereabouts on or about June 19, 1996— particularly where, as here, there is a significant independent event to anchor memories surrounding the relevant date. The prosecution also argues that any harm created by defense counsel’s failure to call additional alibi witnesses is overwhelmed by the harm that Rosario caused himself by what it characterizes as lying on the stand when he did not disclose that he was incarcerated for part of March and April of 1996. This argument seems to cut the other way, however. That is, to the extent that the jury believed that Rosario was being deliberately deceptive, additional alibi witnesses were all the more necessary.
At bottom, the prosecution’s brief takes each witness’s testimony in isolation, picks it apart, and makes an assessment as to whether there is a reasonable probability that the inclusion of that particular witness’s testimony would have affected the outcome of the trial. We cannot engage in such a piecemeal analysis. Rather, we must analyze the cumulative effect of counsel’s failure to call any of the additional alibi witnesses. See Lindstadt v. Keane, 239 F.3d 191, 199 (2d Cir.2001) (“Strickland directs us to look at the ‘totality of the evidence before the judge or jury.... We therefore consider these errors in the aggregate.” (quoting Strickland, 466 U.S. at 695-96, 104 S.Ct. 2052)). This principle, which the majority’s analysis seems to overlook, is essential to the proper application of Strickland, as we were yet again reminded by the Supreme Court in Porter v. McCollum, 558 U.S. -, 130 S.Ct. 447, 453-54, — L.Ed.2d -(2009) (per curiam).
I find defense counsel’s performance and the resulting prejudice in this case very troubling. “[Tjhere is nothing as dangerous as a poorly investigated alibi. An attorney who is not thoroughly prepared does a disservice to his client and runs the risk of having his client convicted *143even where the prosecution’s case is weak. A poorly prepared alibi is worse than no alibi at all.” 2 G. Schultz, Proving Criminal Defenses ¶ 6.08 (1991), quoted in Henry v. Poole, 409 F.3d 48, 65 (2d Cir.2005), cert. denied, 547 U.S. 1040, 126 S.Ct. 1622, 164 L.Ed.2d 334 (2006); cf. United States v. Parness, 503 F.2d 430, 438 (2d Cir.1974) (“It is axiomatic that exculpatory statements, when shown to be false, are circumstantial evidence of guilty consciousness and have independent probative force.”), cert. denied, 419 U.S. 1105, 95 S.Ct. 775, 42 L.Ed.2d 801 (1975). Defense counsel put forth a half-baked alibi defense, leaving substantial additional alibi evidence unexplored, and Rosario is paying the price. For all the foregoing reasons, I would grant the writ of habeas corpus on a conditional basis, providing the State with sufficient opportunity to commence a new prosecution against Rosario prior to his ordered release. Accordingly, I respectfully dissent.
I note that I agree with the majority’s implied denial of habeas relief on the basis of Rosario’s actual innocence claim. While I conclude it is unreasonable to hold that defense counsel performed adequately and that there is no reasonable probability that the verdict would have been different had the additional alibi witnesses testified at trial, I do not think that Rosario has surmounted the extraordinary hurdle required to succeed on an actual innocence claim, assuming such a claim exists under federal law. Finally, I would not so quickly dismiss Rosario’s claim of racial discrimination in the prosecutor’s use of peremptory challenges; however, I need not reach the merits of this claim, because I would grant a conditional writ of habeas corpus based upon Rosario’s receipt of ineffective assistance of counsel, which would warrant a new trial or his release from custody— the same or greater relief that would be provided by a successful Batson challenge.
. Kaiser's deficiency extended beyond his failure to investigate. In his limited investigation, Kaiser was able to contact a few individuals in Florida beyond the two witnesses he actually presented at trial. Specifically, he spoke with Fernando Torres — who, it will be seen, could have been an important witness— perhaps Fernando’s wife Margarita, and others whose names Kaiser could not recall. Kasier would have liked to call some of these witnesses; the limited recollection of the conversations he had, however, was that those he spoke with could not afford to come to New York and may have been reluctant to testify at least in part for financial reasons. Kaiser was unaware of a New York state statute providing reimbursement of certain expenses of out-of-state witnesses and, in any event, operating under the belief that the state court had denied the motion for fees to send Franklin to Florida, assumed that the court would have likewise declined to reimburse the witnesses any expenses. Thus, Kaiser’s decision not to pursue additional witnesses was also based on an erroneous belief rather than on any "plausible strategic calculus or an adequate pretrial investigation” of the facts and law. Pavel, 261 F.3d at 222.
. Because a number of relevant witnesses share the surnames "Torres” and "Ruiz,” I shall refer to those individuals by their first names.
. Margarita Torres, Fernando’s wife and John’s mother, filed an affidavit in connection with the post-conviction hearing stating that she saw Rosario in Florida on June 19, 1996, the day of the murder, and again when Seda came home from the hospital with her grandson. Along with Fernando, Margarita invited Rosario to church that day. Though she did not testify at the post-conviction hearing, she indicated that she would be "more than willing” to testify on Rosario's behalf. As with Fernando, had Margarita testified, a jury may have viewed her testimony as more credible than either John's or Seda’s because she was *134not a friend of Rosario and is a generation older than they.
. According to Serrano, John alone came home briefly to get a change of clothes before returning to the hospital that night. Accordingly, his testimony does not contradict Fernando’s account that Seda and the baby remained in the hospital until the following day.
. Another potential alibi witness — Jeremy David Guzman — filed a written statement in connection with the post-conviction hearing stating that he had spent "hours” with Rosario in Florida on June 19, 1996.
. Nor would have impeachment for criminal history been an issue. Notably, Ricardo was the only witness at Rosario's post-conviction hearing with any criminal record, consisting solely of misdemeanor convictions.
. A third eyewitness, Jose Diaz, believed that he might be able to identify the shooter, but failed to identify Rosario in court.