OPINION
MICHAEL, Circuit Judge.Leon J. Winston was convicted of capital murder and sentenced to death by the Circuit Court for the City of Lynchburg, Virginia. After failing to obtain relief in state post-conviction proceedings, he petitioned the district court for a writ of habeas corpus. He now appeals the district court’s denial of the writ. Winston presents the following issues, as authorized by the certificate of appealability: (1) whether he received ineffective assistance of counsel at the guilt phase of his trial; (2) whether the state trial court unconstitutionally denied his request for a lesser included homicide instruction; (3) whether Atkins v. Virginia, 536 U.S. 304, 122 S.Ct. 2242, 153 L.Ed.2d 335 (2002), prohibits his execution because he is mentally retarded; (4) whether he received ineffective assistance when his counsel failed to argue that Atkins prohibits his execution; and (5) whether he otherwise received ineffective assistance at the sentencing phase of his trial.
We affirm the district court’s denial of Winston’s ineffective assistance claims for both the guilt and sentencing phases of his trial; we also affirm the denial of his jury instruction claim. On Winston’s Atkins and Ai&ms-related claims, however, we conclude that further proceedings in the district court are required. As the district court recognized, because Winston presented material evidence in his federal habeas hearing with respect to the Atkins-related claims that was not considered by the Supreme Court of Virginia, these claims present complex questions concerning the application of the Antiterrorism and Effective Death Penalty Act of 1996, Pub.L. No. 104-132, 110 Stat. 1214 (AED-PA). A district court may not consider additional evidence in a federal habeas proceeding unless the petitioner has satisfied AEDPA’s dual requirements of exhausting state remedies and developing the factual record in the state courts. While these requirements present a high bar for the petitioner, we conclude in this case that Winston cleared the bar. Winston’s counsel for both his state and federal habeas applications did not find records of a test score relevant to proving his retardation until two weeks before the evidentiary hearing held by the district court. We hold that it was error for the district court to refuse to consider this evidence because the score does not fundamentally alter Winston’s claims and because habeas counsel was diligent in searching for it. We therefore vacate the district court’s judgment insofar as it denies Winston’s Atkins and Ai&ms-related claims, and we remand for further proceedings on these claims.
I.
A.
On the morning of Friday, April 19, 2002, two men broke into Rhonda and Anthony Robinson’s house. After confronting the Robinsons near their daughters’ second-floor bedroom, one of the men followed Anthony Robinson downstairs. There, the man shot Anthony several times. He then returned upstairs and shot Rhonda in front of her daughters, *540Niesha and Tiesha Whitehead. After the shootings the men left, leaving Rhonda and Anthony to die from their wounds. Leon Winston was later arrested and tried for the murders.
At trial two witnesses gave essentially uncontradicted testimony that Winston was present at the Robinsons’ when the murders occurred. Michelle Lipford, who had been sexually intimate with Winston, testified that she drove him and a man named Kevin Brown to the Robinson house at about 5:00 a.m. on the morning of the murders. After waiting nearby in the ear for an indeterminate time, Lipford heard gun shots and then drove off. Tranika Turner, Winston’s girlfriend, testified that in response to a telephone request from Winston at 6:00 a.m. that same morning, she picked him up at a carwash near the Robinson house. Turner noted that when she picked him up, Winston was wearing a black hoodie with gray stripes.
To establish that Winston actually shot the Robinsons, the prosecution offered physical evidence, eyewitness testimony, and a confession. For physical evidence, the prosecution introduced a gun that forensic experts had identified as the murder weapon and on which Winston’s DNA — and only Winston’s DNA — was found. Robin Wilson testified that shortly after the murders Winston had given him the gun to keep.
For eyewitness testimony, the prosecution relied on nine-year-old Niesha Whitehead. While Niesha witnessed many of the events that transpired the morning of the murders, her testimony was somewhat inconsistent, and she could not directly identify Winston. She did, however, testify that her mother’s voice woke her on the morning of the murders and that upon waking, she saw two black-skinned men outside her second floor bedroom. She saw one of the men, whom she called “Mr. No Name,” follow Anthony Robinson downstairs. She then heard gunshots and saw Mr. No Name return to shoot her mother, Rhonda Robinson. On direct examination Niesha testified that Mr. No Name had a tattoo that looked like a “big dog,” J.A. 39, and wore all black. It is undisputed that Winston has a tattoo on his arm matching the one described by Niesha, but Niesha could not remember whether Mr. No Name’s clothing had any stripes on it. On cross examination she answered “Yes” twice when asked whether one man wore all black and one wore black with white stripes. She also answered “Yes” when asked whether the man in white stripes tried to stop Mr. No Name from shooting her mother. Finally, Niesha contradicted her testimony on direct by identifying the man with white stripes as the one with the tattoo.
For evidence of a confession, the prosecution called Nathan Rorls, a long-time friend of Winston’s, as a witness. Rorls testified that Winston called him in April of 2002 and told him that he had “slumped two people” in Lynchburg — meaning he had “murdered somebody, killed somebody.” Id. at 94. Rorls further testified that the day after he received the call, he saw Winston and Winston’s cousin Pego (Peyton Carter) at a friend’s house. There, Rorls said that Winston told him he had “killed two people and robbed them and stuff.” Id. at 98. Rorls recounted that Winston proceeded to describe the crime in detail, noting that he (Winston) and “his codefendant” had taken two thousand dollars and two ounces of cocaine from the Robinsons. Id. at 101.
Winston did not take the stand. His counsel thus focused on undermining the credibility of the prosecution’s witnesses and highlighting inconsistencies in the physical evidence. Pointing to Niesha Whitehead’s testimony concerning the shooter’s appearance, defense counsel con*541tended that another man, Tywan Turner, was the likely shooter, and that Winston had tried to prevent the shooting. Counsel also challenged Rorls’s testimony as fabricated and given to avoid the long prison term he was facing on drug charges. Finally, counsel challenged the testimony of Lipford and Wilson as unreliable and implausible.
The jury returned a verdict of guilty, convicting Winston of capital murder, robbery, and several lesser crimes. In the sentencing phase defense counsel presented testimony from Winston’s family describing his troubled childhood. Winston’s mother testified that he had never known his father, that she had been in and out of prison for much of his childhood, and that she regularly used PCP, marijuana, cocaine, and alcohol both before and after Winston was born. Winston’s grandmother testified that she had frequently taken a very young Winston with her when she committed various larcenies. Finally, defense counsel introduced three psychological evaluations of Winston along with some related records, all from Winston’s childhood.
The psychological evaluations, however, were not introduced to prove that Winston was mentally retarded under Virginia law. Rather, they were admitted as ordinary mitigating evidence, describing Winston’s sub-average intellectual functioning and troubled childhood. Defense counsel took this approach despite the bounty of evidence in those evaluations and related records relevant to a determination of retardation under the applicable Virginia statute. The evaluations were accompanied by I.Q. scores of 77, 73, and 76, respectively. In one of the evaluations the psychologist commented that Winston was functioning in the “[bjorderline range of abilities” and that his “ability to recall specific verbal facts which are typically acquired through education and experience is extremely deficient and falls within the Mentally Retarded range (1st percentile).” J.A. 2133. In another the psychologist found that Winston had “borderline intellect and severe verbal processing problems.” Id. The psychologist also noted that Winston’s immaturity and passiveness “place him at a risk to be easily manipulated by others. He is likely to always follow the easiest path, the strongest leader.” Id.
Also in defense counsel’s possession, but not submitted to the jury, were additional records from the Fairfax County Department of Student Services and Special Education (Fairfax County). These records showed that Winston was eligible for special education due to “mild retardation,” that he “demonstrate[d] a reduced rate of intellectual development and a level of academic achievement below that of age peers,” and that he “concurrently demonstrate^] deficits in adaptive behavior.” Id. at 2133-34. Unlike the psychological evaluations presented to the jury, these records did not include an I.Q. score. An affidavit from a Fairfax County special education professional — obtained long after Winston’s trial — indicated that while an additional psychological evaluation and I.Q. test would necessarily have accompanied the County’s conclusions about Winston, these additional items had likely been destroyed.
In determining not to pursue a sentencing strategy based on mental retardation, defense counsel apparently relied primarily, if not solely, on the advice of Dr. Evan Nelson, the clinical psychologist appointed to assist Winston during the sentencing phase. Dr. Nelson did an assessment of Winston for use at his sentencing. He reviewed the psychological evaluations and records forwarded to him by counsel and interviewed Winston and members of his family. Dr. Nelson did not interview Win*542ston’s teachers and social workers, and he cannot recall whether he saw the Fairfax County record changing Winston’s disability classification to mild retardation. Based on the materials before him at the time, Dr. Nelson told defense counsel that “he did not think [they] could prove [that Winston] was mentally retarded as defined by the Virginia statute.” J.A. 2135.
Defense counsel were almost completely dependent on Dr. Nelson’s analysis. They did not read Winston’s psychological records in their entirety before forwarding them to Dr. Nelson, nor is there any evidence that they reviewed them after receiving Dr. Nelson’s opinion. The lawyer tasked with gathering mitigation evidence admitted that he had been unable to review all the records and had “relied on Dr. Nelson to thoroughly review the records and tell [him] what [he] needed to know.” J.A. at 2134. Neither of Winston’s lawyers read the records containing Fairfax County’s change of Winston’s disability classification to mild retardation.
After considering the evidence relevant to sentencing, the jury recommended a sentence of death for each of the murders, 73 years for the robbery and related crimes, and $400,000 in fines. The trial court sentenced Winston according to the jury’s recommendations, except for suspending the fines. The Supreme Court of Virginia affirmed Winston’s convictions and sentence, Winston v. Commonwealth, 268 Va. 564, 604 S.E.2d 21 (2004), and the United States Supreme Court denied certiorari, Winston v. Virginia, 546 U.S. 850, 126 S.Ct. 107, 163 L.Ed.2d 120 (2005).
B.
Winston later petitioned for habeas relief in the Supreme Court of Virginia. Winston v. Warden of the Sussex I State Prison, No. 52501, 2007 WL 678266 (Va. Mar. 7, 2007). In his petition Winston requested an evidentiary hearing and raised multiple claims centering on actual innocence, Atkins, and ineffective assistance of counsel. The Virginia Supreme Court denied all relief without allowing an evidentiary hearing. Id. at *18. It concluded that Winston’s actual innocence claim was not cognizable in habeas, that his Atkins claim had not been raised on direct appeal and therefore was procedurally defaulted, and that his ineffective assistance of counsel claims failed on the merits. Id. at *1, 3, 6-17. The Virginia Supreme Court did not then have before it Winston’s claim that the trial court erred in denying his requested jury instruction on lesser included homicide charges.
Following the denial of his state habeas petition, Winston filed the pending petition in federal district court. In its first opinion, dated May 30, 2008, the district court rejected on the merits Winston’s claims of actual innocence, ineffective assistance of counsel at the guilt phase, failure to instruct the jury on lesser offenses, and ineffective assistance of counsel in presenting mitigation evidence at the sentencing phase. However, with respect to Winston’s Atkins and Aifcms-related ineffective assistance of counsel claims, the court reserved decision and ordered an evidentiary hearing. The district court “d[id] not decide that Winston [wa]s entitled to an evidentiary hearing, but rather exercise[d] its discretion” to order one. J.A. 617.
At the hearing the district court heard from nine witnesses over two days and received documentary evidence not presented to the state courts. Of particular importance was the introduction of Winston’s missing 1997 psychological evaluation and I.Q. test score from Fairfax County. Just two weeks before the hearing, Winston’s habeas counsel found Marilynn Schneider Lageman, the psychologist who administered the test. Lageman testified that she had kept personal copies of Win*543ston’s records, and she produced documentation reflecting Winston’s I.Q. score of 66. Dr. Daniel Reschly, a mental retardation expert, testified for Winston, opining that “to a reasonable degree of psychological certainty,” Winston was mentally retarded under Virginia law before he reached the age of 18. J.A. 2141. After Reschly made what he concluded were proper adjustments for both the Standard Error of Measurement (SEM) and the Flynn effect, the low ranges for Winston’s other I.Q. scores were more than two standard deviations below the mean. Reschly further opined that Winston had conceptual, social, and practical adaptive deficits before age 18. In response to Dr. Reschly, the Commonwealth offered an expert, Dr. Leigh Hagan. Dr. Hagan opined that Winston was not mentally retarded before age 18, that there was no basis for adjusting Winston’s I.Q. scores due to SEM or the Flynn effect, and that limitations in Winston’s adaptive behavior did not significantly impair his functioning.
Winston’s trial counsel and mitigation expert also testified at the district court hearing. Counsel admitted that if they had seen the additional I.Q. score of 66, they would have argued that Winston was retarded and therefore ineligible for the death penalty under Atkins. Dr. Nelson testified that, in addition to not having access to the Fairfax County records and I.Q. score, counsel had not provided him with information from certain teachers, social workers, and family members that was relevant to Winston’s adaptive functioning. Nelson acknowledged that his prior opinion — that a mental retardation defense was not sustainable — might have been different had he possessed this information.
Following the hearing the district court issued a second opinion, dated March 6, 2009, denying Winston’s remaining claims. While the district court reaffirmed its decision to grant an evidentiary hearing, it refused to consider the additional records from Fairfax County, including the I.Q. score of 66, because Winston had failed to exhaust that aspect of his ineffectiveness claim in the state courts. The district court further noted that while Winston had diligently pursued his ineffectiveness claim in the state courts, he had failed to subpoena Lageman or the records of Winston’s fourth I.Q. test. After rejecting the additional records from Fairfax County, the district court concluded that nothing excused Winston’s procedural default of his Atkins claim and that his ineffective assistance claim failed on the merits. This appeal followed.
II.
We first consider Winston’s claims that his conviction should be vacated because he received ineffective assistance of counsel at the guilt phase of his trial or because the trial court erroneously refused to give a lesser-included homicide instruction. We affirm the district court’s decision to reject these claims.
A.
Winston claims that his Sixth and Fourteenth Amendment right to effective assistance of counsel was violated during the guilt phase of his trial. He argues that his trial counsel failed (1) “to prove, as they easily could have, that Rorls’ testimony was a lie” and (2) to pursue effectively their strategy of “flying] [Tywan] Turner to this crime any and every way [they] could.” Appellant’s Br. at 11, 13. We agree with the district court that the state court did not unreasonably apply federal law in denying this ineffective assistance claim.
When a state court has adjudicated a claim on the merits, a federal court may grant a writ of habeas corpus only if *544the state court decision “was contrary to, or involved an unreasonable application of, clearly established Federal law” or “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). We “review[ ] de novo the district court’s application of the standards of 28 U.S.C. § 2254(d) to the findings and conclusions of the [state] court.” McNeill v. Polk, 476 F.3d 206, 210 (4th Cir.2007). To prevail on an ineffective assistance claim, a petitioner must establish (1) that counsel’s performance was deficient and (2) that the deficiency prejudiced the defense. Strickland v. Washington, 466 U.S. 668, 687, 104 S.Ct. 2052, 80 L.Ed.2d 674 (1984). To establish deficient performance, a petitioner “must show that counsel’s representation fell below an objective standard of reasonableness.” Id. at 688, 104 S.Ct. 2052. Courts “indulge a strong presumption that counsel’s conduct falls within the wide range of reasonable professional assistance” to “avoid the distorting effects of hindsight.” Yarbrough v. Johnson, 520 F.3d 329, 337 (4th Cir.2008) (quoting Strickland 466 U.S. at 689, 104 S.Ct. 2052). To establish prejudice under Strickland, the petitioner “must show that there is a reasonable probability that, but for counsel’s unprofessional errors, the result of the proceeding would have been different.” Strickland, 466 U.S. at 694, 104 S.Ct. 2052.
Rorls, in a recent affidavit, said that his testimony at Winston’s trial was “made up” to get a “bigger sentence cut” in his own drug case. J.A. at 325. Winston argues that Rorls’s fabrication would have been apparent to the jury if his counsel had presented testimony from Winston’s cousins, Peyton Carter and Joe Lewis. Carter and Lewis attest in affidavits that they were present at the small gathering at which, according to Rorls’s trial testimony, Winston confessed his role in the murders to Rorls for the second time. Both Carter and Lewis swear that they were with or within earshot of Winston the entire time at the gathering, and they did not witness a confession or overhear any private conversation between Winston and Rorls in which Winston confessed. Counsel’s failure to interview Carter and Lewis and call them as witnesses, Winston argues, constituted deficient performance that prejudiced his defense.
As the district court observed, Winston’s claim that defense counsel should have called Carter and Lewis boils down to the argument that counsel “should have further emphasized that Rorls lied about Winston’s admissions in order to reduce Rorls’ own prison time.” J.A. 594. Defense counsel devoted considerable effort to attacking Rorls’s credibility. The very first questions to Rorls on cross-examination were successful in getting him to admit that he was facing a potential life sentence on cocaine conspiracy charges and that he had “dangle[d] out” information to the authorities about the Robinson murders in an effort to shorten his sentence. J.A. 110. Rorls, who had been released from custody, admitted to the jury that he had “bought [his] freedom” with his cooperation. J.A. 109-10. Defense counsel “clearly and repeatedly” used Rorls’s admissions to press to the jury the theme that Rorls was lying to reduce his prison time. J.A. 594.
It is easy for Winston to argue now that the testimony of Carter and Lewis should have been offered as part of the attack on Rorls’s credibility. But trial strategy decisions are not evaluated “through the distorting effects of hindsight.” Yarbrough, 520 F.3d at 337. Defense counsel’s strategy of attacking Rorls’s credibility in the manner chosen, which was undeniably focused and aggressive, “falls within the wide range of reasonable professional assistance.” Id.
*545Winston challenges another strategy decision of defense counsel in arguing that they did not sufficiently develop the theory that Tywan Turner committed the murders. Specifically, Winston argues that counsel should have called Patti Whitehead, the sister of one of the victims and the mother of Turner’s children, because she had made a statement to police that Turner was in Lynchburg at the time of the murders. However, one of Winston’s defense counsel explained by affidavit that while he and his co-counsel had subpoenaed Whitehead, they did not call her as a witness because they “were unable to talk to her beforehand” and “were unsure what she would say.” J.A. 1881. “On habeas review, a federal court generally credits plausible strategic judgments in the trial of a state case.” Vinson v. True, 436 F.3d 412, 419 (4th Cir.2006). Here, because defense counsel could not determine beforehand what Whitehead’s testimony would be, their decision not to call her was a plausible strategic judgment.
We conclude that defense counsel’s representation was not deficient at the guilt phase of Winston’s trial. We likewise conclude that the state court did not unreasonably apply Strickland in denying Winston’s ineffective assistance claim with respect to the guilt phase.
B.
Winston next argues that his due process rights were violated when the state trial judge refused to instruct the jury on lesser included homicide charges. Although Winston did raise this claim on direct appeal, he did not raise it with the Supreme Court of Virginia in his state habeas petition. The state did not argue in the district court, however, that the claim was unexhausted. Because we conclude that the claim should be denied on the merits, we need not reach exhaustion. See 28 U.S.C. § 2254(b)(2).
“In capital cases, due process requires the court to give an instruction on any lesser included offense when the evidence warrants such an instruction.” Bates v. Lee, 308 F.3d 411, 418 (4th Cir. 2002) (citing Beck v. Alabama, 447 U.S. 625, 637-38,100 S.Ct. 2382, 65 L.Ed.2d 392 (1980)). “Beck requires a lesser-included offense instruction when the evidence at trial ... casts ‘some doubt’ on a necessary element of the capital charge.” Hyatt v. Branker, 569 F.3d 162, 174 (4th Cir.2009). Because each state inevitably has its own rule for when its courts should instruct a jury on lesser included offenses, a federal court first determines whether the state rule offends the federal rule. See Hopper v. Evans, 456 U.S. 605, 611-12, 102 S.Ct. 2049, 72 L.Ed.2d 367 (1982); see also Beck, 447 U.S. at 628-29, 100 S.Ct. 2382. Here, there is no material distinction between the Virginia rule and the federal one. See Remington v. Commonwealth, 262 Va. 333, 551 S.E.2d 620, 632-33 (2001) (“[a] second degree murder instruction is only appropriate where it is supported by the evidence”); Briley v. Commonwealth, 221 Va. 532, 273 S.E.2d 48, 54 (1980) (noting that Beck left the Virginia rule unchanged). The Virginia rule therefore poses no constitutional problem on its face; indeed, its application is equivalent to application of the federal rule for the purposes of AED-PA. Accordingly, we consider only whether the state courts’ application of the Virginia rule was reasonable.
Winston argues that the evidence permitted the inference that he was not the triggerman and therefore supported an instruction on non-capital murder. Except in the case of murder for hire, Virginia law bars conviction of a capital crime when the accused was not the actual perpetrator. Frye v. Commonwealth, 231 Va. 370, 345 S.E.2d 267, 280 (1986). Given Tranika Turner’s unrebutted description of Win*546ston’s clothing that evening as being black with grey stripes, Winston contends that a reasonable jury could have concluded from Niesha Whitehead’s testimony that his accomplice fired the fatal shots. Winston points specifically to Niesha’s testimony that the shooter wore all black and the other man present wore black with white stripes. The jury could therefore have credited Niesha’s testimony and disqualified Winston as the triggerman.
Winston ignores the fact that the state court considered this slice of testimony in the context of a wealth of evidence indicating that he was the shooter. As the Virginia Supreme Court held on direct appeal, “[i]n light of the overwhelming evidence indicating that Winston was the trigger-man responsible for Rhonda’s death, we cannot say that a short passage excerpted from Niesha’s testimony was sufficient to merit jury instructions on first or second degree murder, or accessory after the fact.” Winston, 604 S.E.2d at 44. The evidence at trial pointing to Winston included: Rorls’s testimony that Winston confessed to being the shooter; testimony that the gun used in the killings was in Winston’s possession and had his DNA on it; Niesha’s testimony that a man with a tattoo matching Winston’s was the shooter; and Winston’s own admission that he was present when the shootings occurred. When the several pieces of evidence pointing to Winston’s guilt are measured against the small portion of Niesha’s testimony concerning the shooter’s clothing— which is itself inconsistent — we cannot conclude that the trial court unreasonably applied controlling law by failing to give the lesser-included offense instruction.
The dissent reaches the opposite conclusion by minimizing the significance of relevant evidence and failing to exercise sufficient deference under AEDPA. The dissent argues that Niesha Whitehead’s testimony was consistent with Winston not being the triggerman. However, there is little doubt that Niesha’s testimony was inconsistent, with even the dissent noting that Niesha identified both the man in black and the man with stripes as having the “big dog” tattoo. This inconsistency tends to undermine Niesha’s testimony as a whole rather than provide reliable evidence supporting the inference that Winston was not the triggerman. Similarly, Niesha’s initial failure to remember whether the shooter wore white stripes undermines the black and white distinction defense counsel tried to draw between the two perpetrators.
The dissent appears not to be troubled by the weaknesses in Niesha’s testimony because it concludes that “[ojther than the testimony of Rorls, whose credibility was questioned at trial, no evidence relied on by the majority can be characterized as ‘overwhelming’ evidence that Winston was the triggerman.” Infra at 563. While the dissent is quick to minimize Rorls’s testimony, it was properly admitted into evidence and clearly implicates Winston as the shooter. The fact that Rorls later recanted is irrelevant to our analysis under AEDPA. See Holland v. Jackson, 542 U.S. 649, 652, 124 S.Ct. 2736, 159 L.Ed.2d 683 (2004) (“[WJhether a state court’s decision was unreasonable must be assessed in light of the record the court had before it.”). Similarly, the presence of Winston’s DNA on the murder weapon, which was itself entrusted to another by Winston after the murders, permits an inference that Winston was the triggerman. In light of this additional evidence, it was not unreasonable for the Virginia Supreme Court to characterize as “overwhelming” the evidence that Winston was the triggerman.
We cannot see how the dissent can conclude otherwise under AEPDA’s deferential standard. The dissent states that “the state court made an unreasonable determi*547nation of the facts” when it found that there was overwhelming evidence that Winston was the triggerman. Infra at 563-64. The “finding” that the dissent points to is a legal determination that the evidence did not support giving a lesser included homicide instruction. Proper consideration of this determination is under § 2254(d)(1) rather than § 2254(d)(2). But even if the dissent has identified the correct standard, it has not demonstrated why the state court’s determination was not only incorrect but also unreasonable. The Virginia Supreme Court concluded that no rational juror could have discounted Rorls’s testimony, the evidence concerning the murder weapon, and Winston’s presence at the scene and still credited Niesha’s testimony that a man with white-striped clothing and a tattoo tried to stop the shooting. Assuming this conclusion was error because it crossed the line between making the legal determination of whether an inference is rational and weighing the evidence, it was not an unreasonable conclusion at variance with AEDPA.1
Finally, we disagree with Winston that we should grant him relief because the facts here closely resemble the facts of Beck. In Beck the evidence was sufficient to establish that the defendant was guilty of a serious, violent offense but left some doubt as to whether he had committed a capital offense. The jury, however, was instructed only on the capital charge. Thus, the jury potentially faced the choice of either convicting the defendant of capital murder because it believed that he was “guilty of some serious crime and should be punished” — -an impermissible reason— or acquitting because “whatever his crime, the defendant does not deserve death” — an equally impermissible reason. Beck, 447 U.S. at 642, 100 S.Ct. 2382. The Supreme Court held that an instruction on the lesser included offense was necessary in these circumstances. Winston’s jury, however, did not face the choice of death or acquittal because it was also instructed on robbery and related charges. More important, unlike the state court in Beck, the Supreme Court of Virginia concluded that the evidence against Winston on the capital charge was overwhelming. This reasonable conclusion did not leave room for the quantum of doubt required by Beck. Accordingly, Beck did not mandate a lesser-included offense instruction.
III.
We next consider Winston’s claims that his sentence should be vacated because: (1) Atkins v. Virginia, 536 U.S. 304, 122 S.Ct. 2242, 153 L.Ed.2d 335 (2002), bars his execution; (2) he received ineffective assistance when his counsel failed to argue that Atkins barred his execution; and (3) he received ineffective assistance when his counsel failed to adequately present mitigating evidence during the sentencing phase. Our treatment of Winston’s Atkins-related ineffective assistance claim renders analysis of Winston’s Atkins claim unnecessary. Winston concedes that his Atkins claim is procedurally barred absent a valid excuse, and ineffective assistance is one such excuse. Because we vacate the district court’s decision and remand on Winston’s Atkins-related ineffective assistance of counsel claim, we need not decide on this appeal whether any other excuse applies.
*548A.
Winston claims that he received ineffective assistance when his counsel failed to mount a defense under Atkins v. Virginia, 536 U.S. 304, 122 S.Ct. 2242, 153 L.Ed.2d 335 (2002). In Atkins the Supreme Court held that the Eighth Amendment prohibits the execution of the mentally retarded, but it left to the states the task of enforcing the constitutional restriction. Id. at 306, 317, 122 S.Ct. 2242. The Virginia legislature enacted a statute defining mental retardation shortly after Atkins was decided. To prove mental retardation under the statute, the accused must establish a disability, “originating before the age of 18 years,” characterized by both
(i) significantly subaverage intellectual functioning as demonstrated by performance on a standardized measure of intellectual functioning administered in conformity with accepted professional practice, that is at least two standard deviations below the mean and (ii) significant limitations in adaptive behavior as expressed in conceptual, social and practical adaptive skills.
Va.Code. Ann. § 19.2-264.3:1.1(A). The Supreme Court of Virginia later held that “[w]hether [a defendant] is mentally retarded [is] a factual issue for the jury to determine.” Atkins v. Commonwealth, 272 Va. 144, 631 S.E.2d 93, 98 (2006).
The district court rejected Winston’s Atkins-related ineffective assistance claim, concluding that Strickland’s prejudice requirement had not been satisfied. Proving prejudice required Winston to demonstrate a reasonable probability that a fact-finder would have determined he was mentally retarded under Virginia law. In rejecting Winston’s claim in his state habeas proceeding, the Virginia Supreme Court focused on whether Winston had demonstrated “significantly subaverage intellectual functioning.” To this end, Winston had presented the state court with I.Q. scores of 77, 73, and 76 as well as evidence that Fairfax County had diagnosed him with mild retardation in 1997. The Virginia Supreme Court rejected the three I.Q. scores as evidence of retardation because it had previous held “that the maximum score for a classification of mental retardation [was] an I.Q. score of 70.” J.A. 305. The court rejected the Fairfax County diagnosis because it had been made for the purpose of establishing special education eligibility, and the definition of mental retardation used by the county did not require an I.Q. score below 70. The district court concluded that the Virginia Supreme Court’s determination was not objectively unreasonable.
In reaching its decision, however, the district court refused to consider new evidence that Fairfax County did indeed give Winston an I.Q. test and that the score was 66 — -clearly below 70. In Winston’s state habeas proceedings, the Virginia Supreme Court had denied, without explanation, his motion for discovery and an evidentiary hearing to further develop the factual basis of his claims. Consequently, it was not until two weeks before the evidentiary hearing scheduled by the district court that Winston, through the efforts of his counsel, was able to obtain the additional test score. The district court noted that the new evidence “rais[ed] interrelated questions of whether Winston failed to develop in state court proceedings the factual basis he now relies on to support his ineffective assistance claim and whether Winston has failed to exhaust that claim.” J.A. 2148. After the evidentiary hearing, however, the district court concluded that both questions were “perhaps best viewed analytically as exhaustion questions.” Id. It determined that the potentially dispositive new evidence should not be considered on the merits because it rendered Winston’s claim unexhausted.
*549The application of AEDPA is, of course, implicated in the question of whether the district court should have considered Winston’s new evidence. We agree with the district court’s original observation that the question raises both an exhaustion issue and a failure-to-develop issue. We disagree that the two issues should be collapsed into an exhaustion issue post hearing, and we therefore consider each in turn. See Boyko v. Parke, 259 F.3d 781, 788-92 (7th Cir.2001). Finally, we note that because mental retardation is at bottom a factual issue, review of Winston’s new evidence implicates § 2254(d)(2) and § 2254(e)(1), both provisions of AEDPA governing federal habeas review of state court factual findings. Because we conclude that the district court should have considered the evidence, we will discuss the appropriate application of these provisions on remand.
1.
“An application for a writ of habeas corpus on behalf of a person in custody pursuant to the judgment of a State court shall not be granted unless it appears that ... the applicant has exhausted the remedies available in the courts of the State.” 28 U.S.C. § 2254(b)(1). The exhaustion requirement applies as much to the development of facts material to a petitioner’s claims as it does to the legal principles underlying those claims. See Breard v. Pruett, 134 F.3d 615, 619 (4th Cir.1998). For a claim to be exhausted, “both the operative facts and the controlling legal principles must be presented to the state court.” Matthews v. Evatt, 105 F.3d 907, 911 (4th Cir.1997) (quotation marks and citations omitted). If the rule was otherwise, petitioner's could “expedite federal review by deliberately withholding essential facts from the state courts,” thus upsetting the careful balance struck by Congress between respecting the finality of state court judgments and enforcing federal rights. Vasquez v. Hillery, 474 U.S. 254, 260, 106 S.Ct. 617, 88 L.Ed.2d 598 (1986). The exhaustion requirement does not, however, prohibit a district court from considering evidence not presented to the state courts. “[Supplemental evidence” that does not “fundamentally alter the legal claim already considered by the state courts” can properly be considered by a district court. Id. The question of when new evidence “fundamentally alters” an otherwise exhausted claim “is necessarily case and fact specific.” Morris v. Dretke, 413 F.3d 484, 491 (5th Cir.2005).
In Vasquez v. Hillery the Supreme Court addressed the tension between the exhaustion requirement and consideration of new evidence. In that case the petitioner alleged that African-Americans had been systematically excluded from the grand jury that indicted him. In the federal habeas proceeding, the petitioner introduced two categories of evidence that had not been presented to the state courts: (1) affidavits stating that qualified African-Americans had been available but had not been called; and (2) statistical evidence— based on census data that had been available to the state courts — assessing the mathematical probability that chance could have accounted for the exclusion of African-Americans from grand jury panels. The Court held that the affidavits supported a proposition already well-substantiated by the record before the state courts and that the statistical analysis “added nothing to the case that th[e] Court had not considered intrinsic to the consideration of any grand jury discrimination claim.” 474 U.S. at 259, 106 S.Ct. 617. Without further explanation, the Court concluded that this evidence could be considered because it did not fundamentally alter the petitioner’s equal protection claim.
*550We read Vasquez to permit a district court to consider new evidence if it supports factual allegations for which there is already at least some support in the state record. The evidence in Vasquez was supplemental insofar as there was already evidence in the state record from which a fact-finder could reasonably determine the existence of facts essential to the petitioner’s claim. This reading is consistent with circuit precedent requiring petitioners to present the operative facts of their claims to the state courts. Matthews, 105 F.3d at 911. We note that the strength of the evidence in the state court record is irrelevant to our interpretation. Suppose, for example, that a petitioner on federal habeas introduces new evidence to establish the existence of fact X, a fact required to prove his claim. The claim will inevitably be stronger, regardless of the evidence the petitioner presented to the state courts. However, if the petitioner presented no evidence to the state courts to establish the existence of fact X, the claim will be fundamentally altered by the new evidence presented to the district court.
The district court concluded that the addition of Winston’s 1997 I.Q. score of 66 fundamentally altered his ineffective assistance of counsel claim under Vasquez. It relied primarily on two rationales, the first centering on the potentially dispositive nature of the evidence:
Had the Supreme Court of Virginia confronted a full-scale I.Q. score of 66, it might have viewed the Strickland prejudice prong differently. Instead, Winston presented three scores over 70 and evidence that a likely non-retrievable fourth score might have been under 70. Therefore, there is a substantial difference between the operative facts that the Supreme Court of Virginia confronted and the operative facts here.
J.A. 2150. As an alternative rationale, the district court cited counsel’s failure to ask the Virginia Supreme Court for subpoena power. With that power, counsel could have subpoenaed Lageman and discovered the score during state habeas proceedings. Because counsel could have obtained and presented the score to the state court, the district court reasoned that they cannot now present it to the federal courts.
We conclude that the district court erred in its analysis. First, it is irrelevant to the “fundamentally alters” inquiry whether the Virginia Supreme Court would have decided the claim differently had it seen the new evidence presented by Winston. As noted above, whether new evidence fundamentally alters a claim does not depend on whether it makes the claim stronger. The key is whether Winston offered some evidence in state court to support the factual claim that he possesses significantly sub-average intellectual functioning as measured by a standardized test. Winston presented evidence to the Virginia Supreme Court that Fairfax County had diagnosed him as mentally retarded based on standardized testing, that one I.Q. score should be considered below 70 due to the Flynn effect, and that he possessed substantial deficits in adaptive behavior. While the Virginia Supreme Court was unpersuaded in the absence of an actual score below 70, the court’s determination was based on the weight of the evidence rather than the lack of any evidence to support Winston’s claim.
Wise v. Warden, Maryland Penitentiary, 839 F.2d 1030 (4th Cir.1988), a case cited by the district court, does not preclude our analysis. In Wise the petitioner claimed that the prosecution had violated Giglio v. United States, 405 U.S. 150, 92 S.Ct. 763, 31 L.Ed.2d 104 (1972), by withholding a cooperation agreement with a key witness. In the state courts the petitioner had offered little more than “bald *551assertions” and the affidavit of a fellow inmate attesting to the existence of the agreement. Wise, 839 F.2d at 1032, 1034. In the district court, however, the petitioner presented the agreement itself as well as statements by the inmate-witness to the police. Wise held that the claim had not been exhausted, noting that there was “a world of difference between conjecture and proof’ and that mere conjecture was all that supported the petitioner’s claim when it was presented to the state courts. Id. at 1034. While Wise did refer to the “significantly different and stronger evidentiary posture” of the petitioner’s claim in the district court, id. at 1033 (quoting Brown v. Estelle, 701 F.2d 494, 495 (5th Cir. 1983)), it used the words “significantly different and stronger” to distinguish a claim without evidentiary support from one with evidentiary support. Unlike the case before us today, no reasonable fact-finder in Wise could have found the facts necessary to support the petitioner’s claim from the evidence presented to the state courts.
The district court’s alternative rationale fares no better. Winston’s failure to request a subpoena from the Virginia Supreme Court is simply not relevant to the “fundamentally alters” inquiry. Indeed, considering this “failure” conflates the exhaustion inquiry with the standard for obtaining an evidentiary hearing under 28 U.S.C. § 2254(e)(2), which we discuss below. We note, however, that even if Winston’s efforts in the state proceedings were relevant, the district court would have been wrong to hold Winston responsible for failing to seek a subpoena. Death penalty habeas petitioners seeking discovery in Virginia are required by statute to request an evidentiary hearing from the Virginia Supreme Court. See Va.Code. § 8.01 — 654(c)(1); see also Hedrick v. Warden of the Sussex I State Prison, 264 Va. 486, 570 S.E.2d 840, 862 (2002). Only if that court grants the hearing request and refers the petitioner to a circuit court for discovery would a subpoena request be appropriate. Here, Winston did request an evidentiary hearing from the Virginia Supreme Court, and that hearing was denied. Requesting a subpoena directly would have been procedurally improper, and Winston cannot be faulted for failing to take that step.
2.
In district court the state did not argue that the introduction of the 1997 I.Q. score fundamentally altered Winston’s ineffective assistance claim. Instead, it argued that the court could not consider the score because Winston had failed to develop the factual basis of his claim in the state courts. At oral argument before us, the state continued to advance this failure-to-develop argument despite tacit acknowledgment that the district court had correctly applied the “fundamentally alters” test. We agree that the question of whether Winston failed to develop the factual basis of his claim in the state courts is independent of the exhaustion question. We conclude, however, that Winston did not fail to develop the factual record in state court.
The requirement that petitioners develop the factual basis of their claims in the state courts has its source in 28 U.S.C. § 2254(e)(2):
(2) If the applicant has failed to develop the factual basis of a claim in State court proceedings, the court shall not hold an evidentiary hearing on the claim unless the applicant shows that—
(A) the claim relies on—
(1) a new rule of constitutional law, made retroactive to cases on collateral review by the Supreme Court, that was previously unavailable; or
(ii) a factual predicate that could not have been previously discovered *552through the exercise of due diligence; and
(B) the facts underlying the claim would be sufficient to establish by clear and convincing evidence that but for constitutional error, no reasonable factfinder would have found the applicant guilty of the underlying offense.
The Supreme Court has held that the word “failed” in the opening line of this section connotes fault. Michael Williams v. Taylor, 529 U.S. 420, 431-32, 120 S.Ct. 1479, 146 L.Ed.2d 435 (2000). Specifically, it held that “a failure to develop the factual basis of a claim is not established unless there is a lack of diligence, or some greater fault, attributable to the prisoner or the prisoner’s counsel.” Id. at 432, 120 S.Ct. 1479. If a district court concludes that the petitioner was diligent in developing the factual basis of his claim, it may grant an evidentiary hearing as an exercise of discretion. Schriro v. Landrigan, 550 U.S. 465, 473, 127 S.Ct. 1933, 167 L.Ed.2d 836 (2007) (“Prior to [AEDPA], the decision to grant an evidentiary hearing was generally left to the sound discretion of district courts. That basic rale has not changed.”). This discretion should, however, be exercised with restraint. “[F]ederal evidentiary hearings ought to be the exception, not the rule.” Pike v. Ghiarino, 492 F.3d 61, 70 (1st Cir.2007). They are not “intended to provide a forum in which to retry state cases,” but rather their “prototypical purpose [is] to fill a gap in the record or to supplement the record on a specific point.” Id. As the Supreme Court has said: “Because the deferential standards prescribed by [AEDPA] control whether to grant habeas relief, a federal court must take into account those standards in deciding whether an evidentiary hearing is appropriate.” Schriro, 550 U.S. at 474, 127 S.Ct. 1933.
In its first opinion the district court concluded that it had discretion to grant Winston a hearing on his Atkins and Atkins-related ineffective assistance claims, and it exercised that discretion. After citing to the requirements of § 2254(e)(2), the district court held that: “(1) Winston diligently pursued his habeas corpus claims that Atkins bars his execution and that he received ineffective assistance of counsel through the failure to investigate this claim; and (2) it is not wholly implausible that he could establish his claims even in light of AEDPA’s deferential standards.” J.A. 617. In its post-hearing second opinion the district court declined to reconsider its finding of diligence. It is therefore somewhat puzzling that the state goes so far as to declare in its brief that “[t]he district court’s finding with respect to Winston’s failure to develop this claim is exceptionally clear: Winston did not exercise due diligence.” Appellee’s Br. at 37. The state is wrong.
Perhaps the district court confused the state by considering Winston’s failure to seek a subpoena in its analysis of the exhaustion question. Whether Winston could have or should have sought a subpoena is relevant to the issue of Winston’s diligence, not the issue of exhaustion. In any event, we do not read the district court’s discussion of the subpoena matter as a retreat from its otherwise clear finding of diligence. Moreover, as we have noted, whatever fault the district court assigned to Winston for failing to seek a subpoena was simply misplaced.
Finally, we cannot say that the district court abused its discretion in holding an evidentiary hearing. The district court specifically considered whether to hold a hearing in light of AEDPA’s deferential standards, and the court concluded that it was not implausible that Winston would succeed if he proved the facts alleged in his petition. Winston had never had a hearing on his Atkins and Atkins-related ineffective assistance claims be*553cause the Virginia Supreme Court denied him one, and the district court acted properly to provide him such an opportunity on a colorable claim. The district court narrowly tailored the hearing to supplement the record on a particular issue, and there was no attempt to retry the entire case against Winston.
Because we can find no legitimate justification for the district court’s exclusion of Winston’s 19971.Q. score, we conclude that remand is appropriate so that the court can reconsider Winston’s Atkins-related ineffective assistance claim with the score in evidence. The district court was wrong to exclude the score on the ground that it rendered Winston’s claim unexhausted, and the state is wrong that the score should have been excluded because the hearing in which the score was introduced was improperly granted. Winston is entitled to a decision on his claim based on all of the admissible evidence in the record. We remand for the district court to make this decision in the first instance.
3.
The issue remains as to what standard of review the district court should apply on remand. Winston insists that the district court should apply § 2254(e)(1) because that provision applies exclusively when an evidentiary hearing is appropriate in federal court. The state counters that the district court was right to apply § 2254(d)(2). While we reject Winston’s argument that § 2254(d)(2) will never apply once the district court has granted an evidentiary hearing, we conclude that it does not apply in this particular case. The district court should still apply § 2254(e)(1) to any factual findings made by the state court and make any additional findings necessary to resolve Winston’s claims. Once the facts are determined, the district court should apply Strickland ’s standards without any deference to the decision of the Supreme Court of Virginia. The Virginia court had its opportunity to consider a more complete record, but chose to deny Winston’s request for an evidentiary hearing. Accordingly, comity and finality do not require deference when material evidence later surfaces in a federal habeas hearing.
i.
Section 2254(d) provides AED-PA’s framework for reviewing habeas petitions:
(d) An application for a writ of habeas corpus on behalf of a person in custody pursuant to the judgment of a State court shall not be granted with respect to any claim that was adjudicated on the merits in State court proceedings unless the adjudication of the claim—
(1) resulted in a decision 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) resulted in a decision that was based on an unreasonable determination of the facts in light of the evidence presented in the State court proceeding.
Section 2254(d)(1) describes the standard of review to be applied to claims challenging how the state courts applied federal law, while § 2254(d)(2) describes the standard to be applied to claims challenging how the state courts determined the facts. Both provisions direct federal courts to assess the reasonableness of the state court determinations, and both assessments must be made in light of the evidence the state courts had before them. Holland, 542 U.S. at 652, 124 S.Ct. 2736. The only limitation on § 2254(d)’s application is that the claims submitted must have been “adjudicated on the merits” in state court. When a claim has not been adjudi*554cated on the merits by the state court, a federal court reviews the claim de novo. Cone v. Bell, — U.S.-,-, 129 S.Ct. 1769, 1784, 173 L.Ed.2d 701 (2009); Monroe v. Angelone, 323 F.3d 286, 297 (4th Cir.2003); Wilson v. Workman, 577 F.3d 1284, 1290 (10th Cir.2009); Pike, 492 F.3d at 67; Drake v. Portuondo, 321 F.3d 338, 345 (2d Cir.2003).
A state court’s judgment is unreasonable under § 2254(d)(1) “if the state court correctly identifies the governing legal principle from [Supreme Court] decisions but unreasonably applies it to the facts of the particular case.” McNeill v. Polk, 476 F.3d 206, 211 (4th Cir.2007) (quotation marks and citations omitted). The state court’s application, however, “must have been more than incorrect or erroneous.... [It] must have been ‘objectively unreasonable.’ ” Wiggins v. Smith, 539 U.S. 510, 520, 123 S.Ct. 2527, 156 L.Ed.2d 471 (2003) (quotation marks and citations omitted); accord Terry Williams v. Taylor, 529 U.S. 362, 410-12, 120 S.Ct. 1495, 146 L.Ed.2d 389 (2000). A similar analysis naturally applies to the analogous and adjacent language in § 2254(d)(2). For a state court’s factual determination to be unreasonable under § 2254(d)(2), it must be more than merely incorrect or erroneous. Schriro, 550 U.S. at 474, 127 S.Ct. 1933. It must be sufficiently against the weight of the evidence that it is objectively unreasonable.
An apparent tension between § 2254(d)(2) and § 2254(e)(1) arises because the latter section provides an alternate and seemingly inconsistent standard for review of state court factual determinations:
(e)(1) In a proceeding instituted by an application for a writ of habeas corpus by a person in custody pursuant to the judgment of a State court, a determination of a factual issue made by a State court shall be presumed to be correct. The applicant shall have the burden of rebutting the presumption of correctness by clear and convincing evidence.
This section does not concern itself with the reasonableness of factual determinations by the state courts but with the correctness or incorrectness of those determinations. It further assigns a burden of proof to the petitioner — clear and convincing evidence — for negating them. Finally, unlike § 2254(d), there are no limitations on the application of § 2254(e). The precise interplay between these provisions has split the courts of appeals. See Lambert v. Blackwell, 387 F.3d 210, 234 (3d Cir.2004) (“a comprehensive interpretation of AEDPA’s factual review scheme has yet to emerge from the federal courts”); Wood v. Allen, 542 F.3d 1281, 1303-05 & n. 23 (11th Cir.2008);2 Teti v. Bender, 507 F.3d 50, 57-58 (1st Cir.2007); Guidry v. Dretke, 397 F.3d 306, 324-27 (5th Cir.2005); Taylor v. Maddox, 366 F.3d 992, 999-1001 (9th Cir.2004).
The Supreme Court has held that § 2254(d)(2) and § 2254(e)(1) provide “independent requirements” for review by the federal courts. Miller-El v. Cockrell, 537 U.S. 322, 341, 123 S.Ct. 1029, 154 L.Ed.2d 931 (2003). The sections should not, for instance, be merged to “require *555[the] petitioner to prove that a decision is objectively unreasonable by clear and convincing evidence.” Id. Both provisions apply independently to all habeas petitions. “To secure habeas relief, petitioner must demonstrate that a state court’s [factual] finding ... was incorrect by clear and convincing evidence, 28 U.S.C. § 2254(e)(1), and that [it] was ‘objectively unreasonable’ in light of the record before the court.” Id. at 348, 123 S.Ct. 1029 (emphasis added); accord Miller-El v. Dretke, 545 U.S. 231, 266, 125 S.Ct. 2317, 162 L.Ed.2d 196 (2005). The Supreme Court has nevertheless left open the precise question of when and how § 2254(e)(l)’s presumption of correctness applies. See Rice v. Collins, 546 U.S. 333, 339, 126 S.Ct. 969, 163 L.Ed.2d 824 (2006).
Given these principles, we conclude that § 2254(d)(2) and § 2254(e)(1) will both ordinarily apply even after a district court has properly held an evidentiary hearing. First, we can discern nothing inconsistent with the concurrent application of § 2254(d)(2) and § 2254(e)(1) after a district court has held an evidentiary hearing. If a petitioner succeeds under § 2254(e)(1), he has merely proven that the state court finding was incorrect. To satisfy § 2254(d)(2), the petitioner must prove that the state court was not only incorrect, but also objectively unreasonable. Second, § 2254(e)(1) should apply because there is no limitation in the section’s text. While there might a situation when it would be improper to apply a presumption of correctness to state court factual findings, for example, when the state proceedings violated due process, this would be the exception rather than the rule. See Taylor, 366 F.3d at 1000; see also 1 Randy Hertz & James S. Liebman, Federal Habeas Corpus Practice and Procedure § 20.2c. Third, the only limitation on § 2254(d)(2) — that the claim have been adjudicated on the merits — will not automatically apply simply because the district court held an evidentiary hearing. If the record evidence after the hearing is substantially the same as the evidence presented to the state court, there is no reason why the district court cannot assess whether the state court made an “unreasonable determination of the facts in light of the evidence” presented to it.
When, however, the petitioner offers, for the first time in federal habeas proceedings, new, material evidence that the state court could have considered had it permitted further development of the facts, an assessment under § 2254(d) may be inappropriate. The requirement that § 2254(d) be applied only to claims “adjudicated on the merits” exists because comity, finality, and federalism counsel deference to the judgments of state courts when they are made on a complete record. Cf. Michael Williams, 529 U.S. at 436, 120 S.Ct. 1479. “In this respect, the ‘exhaustion’ and ‘adjudicated on the merits’ elements of federal habeas practice are mirror images.” Wilson, 577 F.3d at 1292. Exhaustion requires that the state courts have an opportunity to apply the law and consider all the evidence relevant to the petitioner’s claim. Baldwin v. Reese, 541 U.S. 27, 29, 124 S.Ct. 1347, 158 L.Ed.2d 64 (2004); Vasquez, 474 U.S. at 257, 106 S.Ct. 617. When a state court refuses to opine on the merits of a claim properly presented to it, exhaustion is satisfied. Cone, 129 S.Ct. at 1784; Pike, 492 F.3d at 67. Similarly, when a state court forecloses further development of the factual record, it passes up the opportunity that exhaustion ensures. Wilson, 577 F.3d at 1292-93; Drake, 321 F.3d at 345; see also Monroe, 323 F.3d at 297-98. If the record ultimately proves to be incomplete, deference to the state court’s judgment would be inappropriate because judgment on a materially incomplete record is not an adjudi*556cation on the merits for purposes of § 2254(d). Wilson, 577 F.3d at 1292; Drake, 321 F.3d at 345. New, material evidence, introduced for the first time during federal habeas proceedings, may therefore require a de novo review of petitioner’s claim.
Our conclusion is not new to this circuit. In Monroe v. Angelone, 323 F.3d 286 (2003), we held that “AEDPA’s deference requirement does not apply when a claim made on federal habeas review is premised on Brady material that has surfaced for the first time during federal habeas proceedings.” Id. at 297. In Monroe the petitioner had sought habeas relief from the Supreme Court of Virginia for violations of Brady v. Maryland, 373 U.S. 83, 83 S.Ct. 1194, 10 L.Ed.2d 215 (1963), based on nine categories of exculpatory evidence allegedly withheld by the prosecution. The petitioner had also sought discovery to seek disclosure of additional evidence that might have been withheld. The state court denied his claims and “refused, without explanation, to authorize ... discovery.” Monroe, 323 F.3d at 294. The federal district court, however, granted discovery and an evidentiary hearing, and the petitioner uncovered several additional categories of withheld exculpatory evidence. On appeal we held that de novo review was appropriate for those portions of petitioner’s Brady claim that were based on the new evidence. Id. at 297. In addition, we held that it was necessary to consider all the evidence in the record — not just the new evidence — because the materiality element of a Brady claim requires a collective assessment of whether introduction of the exculpatory evidence might have affected the outcome of the trial. As a result, we could not “accord AEDPA deference on an item-by-item basis to the ... items of exculpatory material considered in state court.” Id. at 298.
Our conclusion in this case is also consistent with holdings reached in similar cases in the Second and Tenth Circuits. In Drake v. Portuondo, 321 F.3d 338 (2d Cir. 2003), the Second Circuit concluded that de novo review of the petitioner’s due process claim was appropriate when “the state courts did not permit the development of the factual record” and therefore relied on an incomplete record in denying the claim. Id. at 345. Indeed, the Second Circuit actually remanded the case for discovery and an evidentiary hearing. In Wilson v. Workman, 577 F.3d 1284 (10th Cir.2009), the Tenth Circuit, sitting en banc, concluded that de novo review of the petitioner’s ineffective assistance claims was appropriate because the Oklahoma state courts denied the claims based “on a factual record that was, solely as a result of [a] state procedural rule, incomplete.” Id. at 1290-91. Like the Virginia Supreme Court in this case, the Oklahoma Court of Criminal Appeals denied the petitioner an evidentiary hearing and therefore passed up any opportunity to consider evidence material to the petitioner’s ineffective assistance claim.
The court in Wilson also addressed what might be an additional concern with our conclusion today: that habeas petitioners seeking de novo review of their claims need only produce some small sliver of new evidence in federal court. Like the Wilson court, we conclude that this concern is “greatly exaggerated.” Id. at 1294. As we have emphasized above, the requirements that petitioners exhaust their state remedies and diligently develop the record in state court are exacting burdens. Even if the state courts deny a petitioner’s request for an evidentiary hearing, new evidence surfacing in federal court that fundamentally alters a claim will render the claim unexhausted. Similarly, while requesting an evidentiary hearing from the state courts may be necessary to satisfy *557§ 2254(e)(2)’s diligence requirement, it may not always be sufficient. See Michael Williams, 529 U.S. at 437, 120 S.Ct. 1479 (“Diligence will require in the usual case that the prisoner, at a minimum, seek an evidentiary hearing in state court in the manner prescribed by state law.”) (emphasis added). “In light of these safeguards, we see little risk that habeas petitioners will misapply [our] holding to obtain unwarranted de novo review in the federal courts.” Wilson, 577 F.3d at 1295.
ii.
Turning more specifically to this case, we hold that § 2254(d) does not apply to Winston’s Ai/cms-related ineffective assistance claim and that the district court should not afford deference to the Supreme Court of Virginia’s application of Strickland. Winston’s additional I.Q. score is material to whether he is retarded under Virginia law and therefore material to whether he was prejudiced under Strickland by his counsel’s conduct. Moreover, like the petitioner in our Monroe opinion, Winston requested discovery and an evidentiary hearing from the state court to further develop the factual record. Like the state court in Monroe, the Virginia Supreme Court denied Winston’s request and passed on the opportunity to adjudicate Winston’s claim on a complete record. Finally, as in Monroe, the state court adjudicated a claim that was materially incomplete. The deficient performance and prejudice inquiries under Strickland present mixed questions of law and fact, Smith v. Moore, 137 F.3d 808, 817 (4th Cir.1998), that require a collective evaluation of the evidence rather than an analysis confined to a subset of the facts. See Kyles v. Whitley, 514 U.S. 419, 436-37, 115 S.Ct. 1555, 131 L.Ed.2d 490 (1995); see also Cargle v. Mullin, 317 F.3d 1196, 1206 (10th Cir.2003). Consequently, the Virginia Supreme Court’s legal conclusions are not neatly separable into those based on a complete record and those based on an incomplete record. Accordingly, on remand the district court should not afford any deference under AEDPA to the Virginia Supreme Court’s application of the Strickland standards.
AEDPA deference under § 2254(e)(1), however, will be appropriate to any relevant factual findings made by the Virginia Supreme Court. While that court does not appear to have found that Winston was not mentally retarded' — -a factual matter — we note that it did make several factual findings concerning the tests Winston took and the Fairfax County assessment procedures. On remand these findings are entitled to a presumption of correctness, and should Winston wish to overcome this presumption, he will have to do so by clear and convincing evidence. Where the Virginia Supreme Court did not make a factual finding, the district court must make its own without regard to what the state court might have done.
Finally, in making its factual determinations, the district court should not only consider Winston’s 1997 I.Q. score, but also decide whether to consider Winston’s proffered evidence concerning the Flynn effect and SEM. The only time the Virginia Supreme Court has even mentioned the Flynn effect in a published opinion was in this case, and it has never considered the effect of SEM. The Virginia court did not discuss the Flynn effect beyond stating that Winston had offered evidence concerning it. The court concluded that:
This Court has previously held that the maximum score for a classification of mental retardation is an I.Q. score of 70. Petitioner provides no documentation that he was diagnosed as being mentally retarded before the age of 18 in accordance with the legal definition of mental retardation established by the legisla*558ture. Thus, petitioner has failed [to satisfy the requirements of Strickland ].
J.A. 305-06 (citations omitted). The most that can be reasonably inferred from this holding is that the Virginia Supreme Court was unconvinced by Winston’s evidence concerning the Flynn effect and SEM. While a federal court sitting in habeas could not conclude that such a holding violated § 2254(d), see Green v. Johnson, 515 F.3d 290, 300 n. 2 (4th Cir.2008), the district court will not be applying Section 2254(d) on remand. Instead, it will be weighing Winston’s expert evidence, documentary evidence, and proffered testimony — taking § 2254(e)(1) into account when necessary — to determine the facts. Once determined, the district court will consider de novo whether there is a reasonable probability that, but for his counsel’s deficient performance, Winston would have succeeded on his Atkins claim.
B.
Lastly, we affirm the district court’s denial of Winston’s claim that he received ineffective assistance when his counsel failed to present an adequate case in mitigation. The Supreme Court of Virginia held that Winston had failed to demonstrate deficient performance and prejudice, and the district court held that this was not an unreasonable application of Strickland. We affirm because it was not unreasonable to conclude that there was no prejudice.
Because a verdict recommending death in Virginia must be unanimous, Va.Code Ann. § 19.2-264.4(E), the prejudice inquiry requires Winston to demonstrate that there was a reasonable probability that, but for his counsel’s failures, a single juror would have voted against death. Each juror has discretion to vote for death provided that: (1) the prosecution proves one of Virginia’s statutory aggravating factors beyond a reasonable doubt; and (2) the jury considers all the evidence in mitigation. See Watkins v. Virginia, 229 Va. 469, 331 S.E.2d 422, 437-38 (1985). Winston does not argue that either of these requirements was not satisfied. Accordingly, we consider whether the Supreme Court of Virginia unreasonably applied federal law when it concluded that no juror would have exercised his or her discretion to vote against death.
Winston argues that the Virginia Supreme Court unreasonably applied federal law because at least one juror would have exercised his or her discretion to vote against death if: (1) his trial counsel had elicited facts concerning his sub-average intellectual functioning through live witnesses rather than through the written psychological reports that were submitted to the jury; (2) the jury had seen Winston’s 1997 psychological records and I.Q. score of 66; and (3) his trial counsel had presented a more detailed narrative of his traumatic childhood. In rejecting these arguments, the Virginia Supreme Court emphasized that Winston “does not identify the substance of any additional evidence he contends counsel should have presented and does not explain how such evidence would not have been cumulative.” J.A. 306; accord id. at 308.
In rejecting Winston’s first argument about the need for live witnesses on the issue of low functioning intelligence, the Virginia Supreme Court rested its determination primarily on the rich content of the written psychological reports. The reports contained many assessments easily understood by the lay juror, including findings that Winston was of “mentally deficient to average intelligence” with “extreme problems maintaining attention and effort” and “many emotional concerns resulting from his abandonment and rejection from various family members.” Id. at 306. Hence, the core of Winston’s mitigation case was before the jury. Also, Win*559ston had not shown that any particular testimony would have materially added to the reports.
Winston responds by arguing that the manner rather than the substance of counsel’s mitigation case is the critical factor. He argues that the district court “treats manner of presentation as a complete irrelevancy.” Appellant’s Br. at 77. Winston does not, however, cite any case law in this circuit to support his argument. Moreover, what case law Winston does cite does not stand for the proposition that ineffective assistance follows from a failure to present evidence through live witnesses. See, e.g., Gray v. Branker, 529 F.3d 220, 234-38 (4th Cir.2008) (prejudice found when counsel offered no medical explanation for defendant’s behavior); Williams v. Allen, 542 F.3d 1326, 1342 (11th Cir.2008) (prejudice found when counsel failed to discover evidence that “paint[ed] a vastly different picture” than that presented to the jury). While we acknowledge that there might be a case when the manner in which counsel presents evidence is so abysmal that it supports a finding of prejudice, the line would have to be drawn carefully so as not to conflate deficient performance with prejudice. We simply cannot conclude in this case that the Virginia Supreme Court unreasonably applied federal law by discounting the manner in which counsel presented mitigation evidence.
Winston’s second argument — based on counsel’s failure to introduce his 1997 psychological records and I.Q. score — does not survive a careful comparison of the psychological records the jury did see and the records it did not. Winston’s 1997 psychological evaluation reads very much like his previous evaluations. Both his 1995 and 1997 evaluations agree that Winston was “cooperative” but very “immature.” J.A. 1917, 1979. Where Winston largely disclaims responsibility for his own bad acts in the 1995 evaluation, he “reports that he does not feel guilty after doing something wrong” in the 1997 evaluation. J.A. 1919, 1981. With regard to I.Q. scores, his verbal I.Q. for both the 1995 and 1997 evaluations was consistent at 60, which both evaluations describe as being within the range of “mild mental retardation”. J.A. 1918, 1980. His performance I.Q. scores on his 1987 and 1990 tests were actually lower than his score on the 1997 test, and his 1990 and 1997 evaluations both describe his nonverbal skills as “borderline.” J.A. 1913, 1980. Finally, each of the evaluations describes his intellectual functioning at various points as “limited”, “borderline”, or “mentally deficient.” J.A. 1910, 1915, 1921,1982.
Despite these similarities between what was introduced and what was not, Winston argues that “the information possessed by the jury was tremendously incomplete.” Appellant’s Br. at 77. The question, of course, is whether the information possessed by the jury was materially incomplete. Although the jury did lack certain details, Winston cannot suggest prospects for the material omission category except for the single one-page document in which Fairfax County changed his primary disability classification to “mild retardation”, J.A. 1876, and the 1997 I.Q. score of 66. While Winston’s classification and 1997 I.Q. score are material to his Atkins claim, there is no reason to believe that these items would have been persuasive to the jury as ordinary mitigating evidence. With all of the similarities between the submitted and unsubmitted evaluations described above, the question is whether the classification and the additional low score would have made a difference. Because the reports before the jury had already classified Winston’s verbal abilities as mildly retarded, we cannot conclude that a more general classification of mild retardation would have made a material differ*560ence. Similarly, we cannot conclude that the difference between a score of 73 and 66, while perhaps critical to a finding of mental retardation under Atkins, would have made a material difference to a lay juror considering a broader range of mitigation evidence about intellectual functioning.
The dissent contends that the jury did not possess at least ten items of material information. In most of the instances cited, however, the jury either had the information or had information that was essentially indistinguishable. Items 3 (deficient verbal skills) and 5 (twice failing kindergarten) cited in the dissent were noted in Winston’s 1995 and 1987 evaluations. J.A. 1909, 1918. Items 4, and 6-10 all comment on various specific deficiencies, some of which were discussed in the other evaluations, J.A. 1920-21 (items 7 and 9 — failure to understand social expectations or properly handle money), while others such as item 4 (evenly delayed functioning in all areas) are merely syntheses of previous observations. This again leaves only Fair-fax County’s retardation classification and the 1997 I.Q. score, and, as explained above, these omissions were not material.
Winston’s third and final argument— that his trial counsel did not adequately present the jury with evidence of his traumatic childhood — also fails. Defense counsel called four witnesses to illustrate his traumatic childhood: a social worker who testified as to his itinerant experience in foster care; and his mother, grandmother, and great-grandmother, who testified as to his poor upbringing and his exposure to drugs and crime at a very young age. While we have no doubt that Winston’s counsel could have painted a more detailed picture of his childhood, we cannot conclude that the picture they did paint was somehow materially incomplete. Winston’s argument appears to be that further embellishments, greater detail, and more forceful advocacy would have changed at least one juror’s mind. He does not argue, however, that the mitigation case or traumatic childhood his counsel did present to the jury was misleading or that it left out critical facts. Cf. Williams, 542 F.3d at 1342-43 (finding prejudice when counsel put on testimony describing defendant’s childhood beatings as “ordinary parental discipline” that occurred occasionally, rather than testimony describing the beatings ' as “serious assaults” involving “deadly weapons” that occurred “on a near constant basis”). The jury heard testimony concerning Winston’s abandonment by a drug-abusing mother and his exposure to criminal life by his grandmother. It also saw documentary evidence describing his many difficulties coping with his sub-average intellectual functioning. Because the jury still recommended death in light of this evidence, we cannot conclude that the Supreme Court of Virginia unreasonably applied Strickland when it held that a more detailed mitigation case would not have changed even a single juror’s mind.
In reaching a contrary conclusion, the dissent gives inadequate consideration to AEDPA’s standard of review. The dissent does argue that it “can find no reasonable theory under which the evidence disregarded could be considered cumulative,” infra at 570, because Winston’s counsel left out “even the most basic details of Winston’s childhood,” infra at 570. As we explained above, however, we do think that such a reasonable theory existed and that counsel did present the basic details of Winston’s admittedly troubled childhood to the jury. Again, the Virginia Supreme Court may have been wrong to adopt such a theory, but we cannot conclude that it acted unreasonably.
In summary, even if we assume deficient performance by Winston’s counsel, we conclude that the Supreme Court of Virginia *561did not unreasonably apply federal law when it held that counsel’s performance did not prejudice Winston’s case in mitigation.
IV.
For the foregoing reasons we affirm the district court’s denial of Winston’s claims except with respect to his Atkins and Atkins-related ineffective assistance claims. With respect to the Atkins and Atkins-related claims, we vacate the district court’s decision and remand for further proceedings consistent with this opinion.
AFFIRMED IN PART, VACATED IN PART, AND REMANDED
. The dissent stresses that Beck does not present a sufficiency of the evidence issue. We agree. The fact that the evidence supported the jury's verdict is not dispositive. However, we disagree with the dissent to the extent it assumes that evidence besides Niesha’s testimony is irrelevant to answering the Beck inquiry.
. The Supreme Court granted certiorari in Wood v. Allen and issued its decision on January 20, 2010, without addressing the question we consider above. The Court acknowledged that "[ajlthough we granted certiorari to resolve the question of how §§ 2254(d)(2) and (e)(1) fit together, we find once more that we need not reach this question, because our view of the reasonableness of the state court's factual determination in this case does not turn on any interpretive difference regarding the relationship between these provisions.” Wood v. Allen, 558 U.S. -, 130 S.Ct. 841, -, —• L.Ed.2d-, at 9 (2010). Nothing in our opinion today is inconsistent with the Court's decision.