Michael Wayne Hall was convicted of capital murder in state court. He sought state and federal habeas relief, requesting in each forum a live evidentiary hearing on his claim of mental retardation. Each court denied his request. We granted a Certificate of Appealability.

I

Michael Wayne Hall was tried in Texas state court for the murder of Amy Robinson and convicted of capital murder by the jury. Although he was convicted prior to the Supreme Court’s decision in Atkins v. Virginia, which held that the execution of mentally retarded defendants is unconstitutional,1 there was evidence regarding his mental abilities presented at trial in mitigation. On direct appeal, the Texas Court *367of Criminal Appeals (the “CCA”) affirmed the conviction, rejecting his Penry claim and his claim that the Constitution barred the execution of mentally retarded persons. Hall filed a state petition for writ of habeas corpus and a petition for certiorari with the United States Supreme Court. In his state habeas petition, Hall asserted his claim of a constitutional bar to execution of the mentally retarded and requested a “full and fair hearing,” urging, “There has never been a fact finding rendered by either the trial court or a jury as to the issue of whether Applicant is, in fact, mentally retarded.” While his state habeas claim was pending, the Supreme Court decided Atkins. Hall requested a live hearing on the mental retardation issue, urging, “Because the issue of whether or not Applicant is ‘mentally retarded’ has never been fully and fairly litigated and resolved by a fact-finder, this Court cannot rely solely on ... [its] recollection of the testimony from the trial which was conducted over two (2) years ago. This Court should, at a bare minimum, conduct a live hearing on this matter.” Hall also objected to the state court’s “conducting a hearing on Applicant’s Atkins claim via affidavit rather than via live hearing.” Despite Atkins, the trial court conducted a “hearing” by affidavit without awaiting the disposition of Hall’s pending certiorari petition, and the CCA, relying on the paper results, denied Hall’s state habeas claim.

The Supreme Court granted Hall’s petition for certiorari from his direct appeal to the state court, vacating and remanding to the CCA to reconsider its initial affirmance of Hall’s conviction in light of Atkins. The CCA, relying on the state habeas and direct appeal records and a “re-review[ ][of] the evidence” from the records, held that Hall was not mentally retarded.2 Hall appealed again to the Supreme Court, which denied certiorari, and he then filed a federal habeas petition, again arguing, “Mr. Hall has a right under the Fifth, Sixth, Eighth, and Fourteenth Amendments to a full and fair hearing in a court of law on the issue of his mental retardation.” The federal district court relied on the state record to conclude that Hall was not retarded, and denied Hall’s Atkins claim. We granted a COA.

II

In applying the deferential standard under the Antiterrorism and Effective Death Penalty Act (“AEDPA”), we are mindful of the unique facts of this case, in that Atkins was decided after Hall’s conviction, and the state’s paper hearing on the Atkins mental retardation issue was completed before Texas had defined mental retardation under the Atkins standard. We review the federal district court’s refusal to grant an evidentiary hearing on the Atkins issue for an abuse of discretion.3

Section 2254(e)(2) of United States Code Title 28 does not constrain the district court’s discretion here because Hall diligently developed the factual basis of his claim in state court.4 In the state habeas *368proceedings, Hall consistently raised his claim that he was mentally retarded and that execution of a mentally retarded individual is unconstitutional.5 He also provided affidavits of experts, affidavits of former teachers and other individuals who were familiar with Hall’s capabilities,6 and evidence of mental limitations, such as medical records, grade reports and special education screening results from school, and the results of Hall’s examination for competency to stand trial. These were more than adequate to establish a factual basis for his mental retardation claim.

Once a district court determines that a petitioner may be entitled to an evidentiary hearing, this is not the end of the inquiry, since “[i]n cases where an applicant for federal habeas relief is not barred from obtaining an evidentiary hearing by 28 U.S.C. § 2254(e)(2), the decision to grant such a hearing rests in the discretion of the district court.”7 In determining whether to grant a hearing, under Rule 8(a) of the habeas Court Rules “ ‘the judge must review the answer [and] any transcripts and records of state-court proceedings ... to determine whether an eviden-tiary hearing is warranted.’ ”8 And the Supreme Court has held since AEDPA that the court must also “consider whether such a hearing could enable an applicant to prove the petition’s factual allegations, which, if true, would entitle the applicant to federal habeas relief. Because the deferential standards prescribed by § 2254 control whether to grant habeas relief, a federal court must take into account those standards in deciding whether an eviden-tiary hearing is appropriate.”9 And finally, a district court abuses its discretion in not holding an evidentiary hearing only if *369the state court failed to provide a full and fair hearing.10

The facts before us are a core manifestation of a case where the state failed to provide a full and fair hearing and where such a hearing would bring out facts which, if proven true, support habeas relief. Hall alleges that he is mentally retarded under Texas’ Atkins test for mental retardation announced in Ex Parte Brise-no: under Briseno a defendant must prove that he has “(1) significantly subaverage general intellectual functioning [‘defined as an IQ of about 70 or below’]; (2) accompanied by related limitations in adaptive functioning; (3) the onset of which occurs prior to the age of 18.”11 Because neither Atkins nor the Briseno test had been established at his original trial, and Briseno was not decided until approximately two years after the state court’s paper hearing on mental retardation,12 upon which the CCA and the district court relied, Hall never had the opportunity to present a full range of evidence on this technical issue.13

The federal district court discussed the Briseno factors but did not conduct a hearing, relying instead on the state court’s findings of mental retardation—findings that were made prior to the Briseno test. The district court explained, “In his petition, Hall thoroughly reviews the voluminous evidence as to his mental capacity. The state does the same in its response. There is no reason for the court to do so again here.”14 The court examined the paper evidence and held, “Having independently reviewed all of the evidence, the court concludes” that “the state court’s finding that Hall is not mentally retarded was not unreasonable.”15

The district court also addressed Hall’s claim for a jury determination of mental retardation or, in the alternative, a live hearing. The district court found that Hall “failed to raise this issue in the state proceedings” and that it was procedurally defaulted.16 This conclusion is incorrect; Hall consistently raised this issue in his state and federal habeas petitions. The district court went on to hold that “[e]ven if not procedurally defaulted, Hall’s claim that he was entitled to a jury determination on mental retardation is without substantive merit. Nowhere in Atkins does the Supreme Court hold that a jury determination on this issue is required.”17 This *370conclusion is correct, but it misses a central element of Hall’s claim — namely, that Hall requested a jury trial or at minimum a live hearing.18

In reviewing the paper evidence, the district court failed to account for several clearly erroneous findings of the CCA which, if addressed in an evidentiary hearing, might have highlighted the unreasonableness of the state court’s determination of the facts and entitled Hall to habeas relief. The CCA’s habeas determinations rested on the state court’s findings from the paper hearing, as the CCA fully adopted those findings.19 In other words, errors in the state court’s factual findings were not corrected when they reached the CCA. In assessing the affidavits of experts, which address Hall’s IQ, the state court misread an IQ score of 67 reported in Dr. Church’s sworn affidavit, replacing a Wechsler Adult Intelligence Scale, third edition, exam (“WAIS-III”) score of 67 with 72. The state’s expert, Dr. Price, also erroneously relied on an IQ score of 72 in making his assessment, stating, “[I]f an individual is being assessed for the presence or absence of mental retardation and receives and [sic] IQ score of 72, then his or her actual IQ is 95% likely to fall between 67 and 77-a range of scores that may indicate mild mental retardation or borderline intelligence.”20 Relying at least partially on this error, the state trial court concluded that Hall’s intelligence level was “either in the borderline range of intellectual functioning or in the upper end of mild mental retardation.”21 A hearing would clarify whether Dr. Price’s conclusions with respect to Hall’s IQ, which influenced the state court’s finding, were premised on factual error.

The state trial court also made erroneous findings with respect to the credibility *371of at least one of Hall’s experts22 — -an expert whose testimony was central to Hall’s allegations of limited adaptive functioning.23 The state concluded that Dr. Church’s submission should “not be given weight” because Texas law does not permit individuals with Dr. Church’s credentials to make mental health determinations. Texas’ Persons with Mental Retardation Act only permits physicians or psychologists licensed in Texas or certified by the Texas Department of Mental Health and Mental Retardation to make such determinations.24 Yet this Act is inapplicable in the Atkins context, and the state court’s conclusion to the contrary was clearly erroneous. “The PMRA, by its own terms, is irrelevant to the application of Atkins. For Eighth Amendment purposes, it neither defines mental retardation nor — more relevantly — establishes who may diagnose mental retardation.”25

The issue of mental retardation, defined by Atkins only after Hall was tried and defined by Texas only after Hall’s paper “hearing” on mental retardation, is fact-intensive and rests on nuanced determinations under broadly stated concepts such as “limitations in adaptive functioning.” If Hall can prove the facts that he has consistently alleged on appeal, he will be entitled to habeas relief.26

Furthermore, the state court’s erroneous factfinding and its refusal to accept more than paper submissions despite the development of a new constitutional standard and a lack of guidance from the state on that standard, deprived Hall of a full and fair hearing at the state level. Although we have found paper hearings adequate where “the trial court and the state habeas court were one in the same,” as was the case here, there is a crucial distinction. Following trial, the state trial judge on habeas review faced a new constitutional rule categorically barring the execution of mentally retarded persons. Atkins was so new that Texas had not had time to establish a definition of mental retardation or the associated burdens of proof.

Nor were the district court’s findings of “facts” from the disputed assertions of affidavits below adequate. As we have discussed, some of these accepted “facts” were both critical and incorrect.27 These *372errors would have been drawn out in a hearing with an opportunity of counsel to examine the witnesses. Other of the district court’s conclusions rely heavily on the conflicting expert opinions of psychologists, asserted in affidavits unaired in court and shielded from cross examination. Given the material errors in credibility determinations and factfinding at the state level, we are persuaded that the determination of Hall’s claim, caught in the immediate uncertainty following Atkins, was so freighted with a risk of error in fact finding that the failure of the district court below to conduct a meaningful hearing was an abuse of discretion in these unusual and unique circumstances.

Ill

Accordingly, we VACATE the judgment of the federal district court and REMAND to that court for further proceedings including an evidentiary hearing.

concurring in part and dissenting in part:

The Supreme Court left implementation of Atkins to the states, a license that implicitly insists upon faithfulness to its core holding and the constitutional constraints of due process. I am persuaded that because the state denied Hall the right to confront and cross examine state witnesses in the adjudication of his claim of retardation, Hall’s death sentence was contrary to and involved an unreasonable application of the clearly-established federal law of due process as determined by the Supreme Court of the United States.

Michael Wayne Hall was convicted of capital murder by a Texas jury and sentenced to death in 2000. Much of his mitigation case rested on the claim that Hall suffered from a learning disability. The Texas Court of Criminal Appeals affirmed the conviction and sentence, rejecting Hall’s Penny argument that the jury could not give expression to his mitigation evidence and his claim that execution of mentally retarded persons is unconstitutional. Hall filed a petition for certiorari with the Supreme Court and sought collateral habeas review with the state court.1 There he continued his contention that it was unconstitutional to execute a mentally retarded individual — the issue presented in the Atkins case — which was then sub judi-ce in the Supreme Court. Before the state answered his state habeas petition, the Supreme Court held in Atkins that executing mentally retarded individuals violates the Eighth Amendment. Despite Atkins, the trial court on habeas refused over Hall’s objection to conduct an evidentiary hearing2 to determine whether Hall was mentally retarded, refusing Hall the opportunity to cross examine witnesses. Rather, it recommended that the CCA reject Hall’s state habeas claim, relying upon *373the record from his original trial and affidavits submitted by order of the trial court. No live testimony was taken, and Hall was not allowed to cross examine any of the affiants despite the introduction of many new affiants and claims that had not been tested at trial. In May of 2004, addressing the Supreme Court’s opinion vacating and remanding the judgment of the CCA,3 which had rejected the Atkins claim on Hall’s direct appeal, a divided CCA repaired to the same record and again concluded that Hall was not mentally retarded. Hall again petitioned the Supreme Court for certiorari, which the Supreme Court denied in June of 2005. Hall then filed his federal habeas claim, again requesting a hearing on the Atkins issue. Without holding a federal evidentiary hearing, the U.S. district court deferred to the state adjudication.

I

Michael Wayne Hall was 18 years old when he killed Amy Robinson, a mentally retarded co-worker at a grocery store.4 Hall had low intelligence and lived in a chaotic family environment. He had never been incarcerated, although he had participated in minor shoplifting incidents with other individuals. Hall often associated with people younger than his age but he met an older friend, Robert Neville, in 1997. He appreciated Neville’s friendship and attention and spent many hours with him, accepting his offers of gifts, rides, and places to stay, and imitating his style of dress. Neville persuaded Hall to quit his job so they could work together at a grocery store. They decided to kill someone. With purchased weapons, they killed Amy, an easy target. Police arrested them attempting to walk to Mexico several hundred miles away.

The trial commenced in February of 2000. The state presented, inter alia, a videotape that the media had taken of Hall, Hall’s written statement to the police, the arrest warrant, the weapons, photographs of the area where the victim was found, and testimony of the agents and detectives who questioned Hall and took his statement. There were vague references to Hall’s acuity throughout this portion of the trial, but they were not drawn out. At one point, an agent indicated that detectives had asked Hall to take a polygraph test, but he said that “he couldn’t take it, his mind was kind of messed up.” The attorney cross examining the agent asked if Hall had expanded on this statement, but the agent replied in the negative. The attorney also inquired about the competency of the writing — grammar and spelling, for example — in Hall’s statement, as well as Hall’s ability to communicate with the agent (whether there was anything “unusual about his understanding of events or how he told them” to the agent). The defense asked a detective involved in the case whether Hall’s family members had indicated that Hall had any mental disabilities and if this was consistent with the detective’s impressions of Hall. The detective replied briefly that, “[i]f I recall, [Hall’s stepbrother] said he was slow,” and that according to the detective, “he was not the most intelligent person I had ever seen, but he’s also not the dumbest person I had ever seen, either.” On the third day of trial, the jury found Hall guilty.

*374In the sentencing phase of the trial, the defense offered evidence of Hall’s learning disability. Hall’s mother and brother testified about mental problems and learning disabilities in the family, Hall’s struggles to perform typical childhood tasks, childhood injuries to his head, and later difficulties when he started work and attempted to socialize with colleagues.5 In addition to testimony from Hall’s family, Hall’s teachers and psychologists discussed his mental capacity at school.6 The school psychologist and experts discussed various results from Hall’s IQ tests, ranging from 67 to 71 for full-scale IQ tests7 but differed as to their significance, reliability, relevance, and margin of error.8 Several experts also testified generally as to Hall’s adaptive functioning and knowledge, and his academic abilities. Dr. Cunningham assessed Hall’s adaptive functioning,9 con-*375eluding that the tests indicated that Hall was learning disabled, with “significant academic deficiency.” Dr. Price assessed Hall’s adaptive knowledge (not adaptive functioning, which is the applicable issue under the Briseno10 test for Atkins mental retardation) using a Street Survival Skills Questionnaire,11 indicating that Hall was “slow on a lot” of the questions but that his intelligence was “either borderline, right at the level of mild mental retardation, or he’s mildly mentally retarded.” Dr. Price concluded, “it’s sort of a judgment call.”

Finally, the state presented lay witnesses who had some past contact with Hall. A fellow grocery store employee of Hall’s testified that Hall taught him how to bag groceries, said that he had worked with mentally challenged children, and indicated that he did not notice anything slow about Hall. A waitress who had once served food to Hall and Neville in a restaurant after the murder testified that Hall ordered food for himself and that she did not notice him picking up food with his hands, although she did not discuss whether or not he was able to use a knife.

In their closing arguments to the jury, the state and the defense focused exclusively on Hall’s mental capacity as it related to mitigation. The state argued that Hall was “at worst mildly retarded” and that this was not mitigating, since he was “smart enough” to choose a trusting victim. Hall’s attorney only discussed his IQ in the broad sense (discussing one teacher’s estimation that it was “somewhere in the 60s,” and how the experts could not agree on a “precise” IQ number) and emphasized Hall’s mental capacity in light of his inability to understand the wrongness or “grasp the horribleness” of his crime. The state, in its final rebuttal, argued: “does it really matter to us, other than for labeling purposes, whether he’s labeled as mild mentally retarded or low normal? Does that really tell us anything that we need to know or help us in determining is this a mitigating factor?”

The jury answered “no” to the question of whether there were mitigating circumstances to spare Hall’s life. On automatic appeal, the CCA affirmed the trial court’s conviction and sentence on January 16, 2002,12 rejecting Hall’s contention that it is unconstitutional to execute mentally retarded persons. On January 17, 2002, Hall filed a state application for a habeas writ, re-emphasizing his argument that “[t]o inflict a death sentence on a person suffering from mental retardation” violates Article 1, § 12 of the Texas Constitution and Eighth Amendment of the U.S. Constitution, an issue that was then before the Supreme Court in Atkins. He requested a full and fair hearing, a “live” hearing.13

On June 10, 2002, Hall petitioned the Supreme Court for a writ of certiorari on the mental retardation issue, seeking review of the CCA’s judgment on direct appeal affirming his conviction and sentence. On June 20, 2002, the U.S. Supreme Court in Atkins held that the “[Eighth Amend*376ment of the] Constitution ‘places a substantive restriction on the State’s power to take the life’ of a mentally retarded offender,”14 thus setting a constitutional bar against the execution of mentally retarded individuals.15

The state filed its reply to Hall’s state habeas application on July 16, 2002, acknowledging the Atkins decision but maintaining that the trial court’s procedure for determining mental retardation was adequate.16 The habeas trial judge, who had also presided over the original trial, refused to order a hearing or to allow cross examination of experts on the new Atkins issue, announcing that the issue would be submitted based on affidavits and the trial record. Hall again objected, urging that the court conduct a live hearing. The parties submitted affidavits from Dr. Price and Dr. Cunningham, as well as affidavits from many individuals who had not testified at trial, including psychologists Dr. Denkowski and Dr. Church, an advocate for mentally retarded death row inmates, prison guards, a mitigation specialist, investigators who assisted Hall with his ha-beas claim, an inmate who resided next to Hall on death row, more of Hall’s teachers, and Hall’s trial counsel.

Free of cross examination, experts presented new evidence of Hall’s IQ and further disputed the margin of error associated with the various IQ scores. These IQ scores included the school’s WISC-R result of 71 with a measurement error of 2.6, when Hall was 12/£ years old; Dr. Cunningham’s WAIS-III test, which resulted in Hall’s score of 67 at age 20; Dr. Church’s WAIS-III test at age 22, which again showed a score of 67 (wrongly reported by Dr. Denkowski as a 72); and TONI-2 and TONI-3 scores of 84 and 77 at ages 16 and 20, respectively (with associated information that the TONI “yields a 7-point higher score than the WISC-R Full Scale IQ” and does not measure general intellectual functioning). Dr. Price’s affidavit discussed the IQ findings from the original trial but presented new allegations regarding the proper standard for determining mental retardation under Atkins, and how his testing related to this standard. He criticized Dr. Denkowski for basing his affidavit on records rather than a personal examination of Hall but indicated that he and Dr. Denkowski “basically agree on the current working definition of mental retardation that is used by the psychological community (i.e., IQ below approximately 70 with measurement error considered, significant adaptive behavior deficits in 2 or 3 skill areas, and origination of these factors in the developmental period.)” Dr. Price also described the results of his adaptive knowledge testing of Hall. He alleged that “Mr. Hall’s knowledge of adaptive behavior ranged from the borderline level (16th percentile) in several areas (understanding and telling time, handling money, and using common measurements) to the average level (8th to 63rd percentile) in other areas” (“recognizing and understanding basic spatial/quantitative concepts, functional signs, tools, kitchen utensils, hygiene, safety skills, and the *377use of public services”). He did not indicate whether any of the tests demonstrated that Hall was below borderline. Dr. Price also alleged: “It is very important to note that Michael Hall was placed in special education due to having been classified as having a learning disability — not mental retardation .... [M]y review of this case does not clearly indicate that Michael Hall is mentally retarded.” Yet he concluded elsewhere that “[i]t is obvious that the issue of the diagnosis of mild mental retardation is controversial and determining whether or not someone receives that diagnosis can literally turn on a word or a number.”17

Dr. Denkowski’s affidavit explained that he did not interview Hall because a meeting in a high security prison would “not provide reliable information for gauging [Hall’s] functional status.” He contested Dr. Price’s IQ data, indicating that “Dr. Price ... misrepresented Mr. Hall’s SSSQ [adaptive behavior] data through tacit implication that they were contrasted with those of typical adults,” whereas in reality the test derives “adult norms” from “two secondary school prevocational programs in Indiana and Texas.” Dr. Denkowski concluded that “how Dr. Price presents SSSQ data is unacceptable for any diagnostic purpose, even outside the courtroom” and that Dr. Price’s other adaptive skills test, the K-FAR, showed that Hall’s math skills were “less competent than those of 99% of typical same-aged persons” and that his reading skills were “less proficient than 95% of people of his age.” Dr. Denkowski also examined relevant authorities that determine mental retardation under Atkins in Texas, concluding that the relevant numbers required for a finding of mental retardation of a defendant in Texas are a “WAIS-III full scale IQ” below 75,18 “significant adaptive behavior deficits must exist in three skills areas and the measurement error adjusted standard score for the overall instrument must be below 71,” and “[t]hese impairments must have originated prior to the 18-22 age range.” Further, he found that “[s]ince Mr. Hall’s IQ and adaptive behavior quality plainly fall within the area that the DSM-IV and Texas law consider to define mental retardation, it seems reasonable to express confidence in his diagnosis of mental retardation.” Hall’s adaptive behavior tests indicated that his behavior was “of the quality of higher-level mentally retarded adults.” Dr. Denkowski also emphasized that Texas’s definition of mental retardation recognized only a WAIS test for IQ determination, since only that test “yield[s] a ‘full scale intelligence quotient’ ” and that the TONI test (which had been discussed at trial) does not yield an actual IQ or assess general intellectual function, thus making it useless for diagnosing mental retardation in Texas. Most significantly, Denkow-ski’s affidavit, upon which the trial court relied in finding that Hall was not mentally retarded, indicated incorrectly that Dr. Church’s examination of Hall produced an IQ score of 72; the score was in fact a 67. Dr. Denkowski’s affidavit, in addition to commenting on other experts’ evaluations of Hall, discussed the weaknesses of the state’s lay witness affidavits as well as some of the state’s claims regarding mental retardation.

There were gaps and inconsistencies throughout the record; we mention only a small sample here. The state posited in its reply to Hall’s state habeas petition that Hall, while in prison, wrote a note that said, “You have to get me out of here because there’s no call button. The sink is *378stopped up. Also there are roaches and a small ass bed. My feet go all the way to the wall because I’m six feet four.” The document actually reads, as Hall’s attorneys point out in their briefing: “you half to get out of here because there is No call button, the senk is stopd up also rauch’s and small ass bed by feet go all the way to the wall because I am 6f4”.” This cleaning up of a statement lies buried in the paper records but would have been brought to light by any trial lawyer given the opportunity to cross examine the opposing party. Mr. Harris, one of Hall’s trial attorneys, indicated in an affidavit that Hall’s defense team had chosen not to reveal the victim’s IQ score, since she was clearly an innocent victim. However, the state at trial had focused strongly on how Amy was mentally impaired and was clearly more mentally impaired than Hall.19 In fact, as Mr. Harris’s affidavit reveals, her IQ of approximately 83 (the exact number remains unclear, as it was only put forth on paper) was meaningfully higher than Hall’s.

Hall’s affidavits from teachers alleged that he drooled in class, that he had “extreme difficulty speaking complete sentences,” and that he had trouble spelling his own name. An affidavit of a mitigation specialist appointed to Hall’s case indicated that Hall did not understand the appeal or how the attorney would help him with the appeal. He could only communicate basic information about himself, such as his name and his mother’s name, and that, during the interview, “Hall would agree with or go along with whatever he thought [the mitigation specialist] wanted him to say.” Finally, he indicated that Hall mispronounced words or used them out of context and appeared to use words that he did not understand “in an attempt to mask or hide his disabilities.”

The state filed “affidavits” of prison staff, all of which appeared to have been similar “fillable forms” employed by a state investigator. The affidavits alleged that Hall appeared “normal” to the guards and that he could follow instructions. To demonstrate experience with mental illness, one guard wrote that he “knew some kids in school with Down’s syndrome,” while another said that he had an uncle who was mentally retarded. Dr. Denkow-ski pointed out that the observations of teachers and prison guards, indicating that Hall was able to obey, follow rules, and function and cope, were painted by the state as indications of Hall’s mental capacity but in fact simply affirm that “Hall has long demonstrated the capacity to learn and abide by institutional rules and expectations,” as many mildly mentally retarded persons are capable of doing.

Additional papers presented by affidavit included Hall’s medical records (showing various incidents such as a fall on the sidewalk resulting in an abrasion to his mouth and cuts that resulted from glass falling on Hall), grade reports and special education/disability documents from school, prison medical and other records, the results of a psychiatric examination to determine Hall’s competency to stand trial *379(indicating that Mr. Hall “presented his charge very well” but that “[h]e could perform only two of the four basic elementary mathematical functions”), and psychology texts, such as the American Association on Mental Retardation’s book entitled, Mental Retardation: Definition, Classification, and Systems of Supports.

The CCA relied entirely upon these affidavits and the original trial court record in reviewing Hall’s habeas claim of mental retardation,20 which was now controlled by the decision of the Supreme Court in Atkins. The CCA adopted the trial court’s paper findings in full — at least one of which was based on an erroneous conclusion of law21 — on February 26, 2003, and denied Hall’s habeas application.22 The CCA then rejected the mental retardation claim a second time on the Supreme Court remand of Hall’s direct appeal case23 but once again it simply adopted the trial court’s findings based on the paper record and relied solely on its habeas findings and portions of its original case affirming Hall’s conviction, as well as a “re-review[ ]” of the record evidence, to conclude that Hall was not mentally retarded.24 The court considered no new evidence in these trials and ordered no hearing on the Atkins issue. It justified its actions by arguing that the process was sufficient:

[T]he trial court and this Court did have the benefit of Atkins during the habeas proceedings. The parties had ample opportunity to present evidence at that time on the specific issue of mental retardation. And we can consider the ha-beas proceedings and evidence in the current posture of this appeal ... we are faced in a direct appeal with an issue that has already been presented to us on habeas corpus. Consequently, we address appellant’s mental retardation in light of both the direct appeal and the habeas records ... we believe that taking judicial notice of the habeas proceeding and its outcome satisfies the Supreme Court’s remand order in the present case.25

*380The dissenting justices disagreed with the “unique” precedent the court elected to follow to determine the Atkins claims on remand of the direct appeal, emphasizing that “[n]o trier of fact in this case has ever heard live testimony, subject to testing on cross examination, on the specific issue of whether appellant is mentally retarded.”26 The dissent concluded, “[W]e will never know [whether Hall is mentally retarded] unless we order [a] full hearing and have before us both the tested testimony of persons who are knowledgeable in the mental-health field and relevant lay testimony about his adaptive behavior.”27

In June of 2006, Hall filed a federal habeas petition after the Supreme Court had denied his petition for certiorari.28 He again specifically argued that the denial of a full evidentiary hearing denied him due process rights, stating, “Mr. Hall has a right under the Fifth, Sixth, Eighth, and Fourteenth Amendments to a full and fair hearing in a court of law on the issue of his mental retardation.” The federal district court, like the state trial court and court of appeals, refused to conduct an evidentiary hearing or to allow cross examination of the state experts on the Atkins claims, denying the claims, deferring to the state adjudication under AEDPA, and conducting its own examination of the paper record.29 We granted a certificate of appeala-bility on the Atkins claim.

II

Our decisions have accorded deference to state adjudications of claims for habeas relief from state criminal convictions under AEDPA even if made without a live hearing.30 Under AEDPA there is a presumption that the state court’s findings of fact are accurate “unless the petitioner can rebut the findings of fact through clear and convincing evidence.”31 Following this line of deference, in May32 and similar cases both before and after AEDPA, we have found that where the trial judge who presided over the initial case later considers the habeas claim, we presume that reliance upon the completed record is acceptable.33 This is sensible in many cases, at least those in which the trial judge experienced first-hand evidence directly relevant to the habeas issue. Due process requires a hearing, not two. Despite the deference that we give the adjudication of state courts under AEDPA, this court has highlighted — post-AEDPA—that “ ‘[t]he fundamental requirement of due process is the opportunity to be heard “at a meaningful time and in a meaningful manner”.’ ”34 This said, allowing the original trial judge to proceed without an evidentiary hearing *381or live cross examination finds trouble when a central constitutional principle changes between the initial trial court decision and that same court’s decision of a habeas claim — acutely so where the claim turns on sharply contested issues of fact and witness credibility.

AEDPA provides,

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.35

The Supreme Court has explained that AEDPA does not “require the federal courts, including this Court, to defer to state judges’ interpretations of federal law.”36 Rather, our duty is to make an “ ‘independent evaluation’ ” of “whether or not a rule is clearly established at the time a state court renders its final judgment of conviction.”37 A state court decision is contrary to clearly established federal law in two instances — first, where it is “diametrically different, opposite in character or nature, and mutually opposed to [the Supreme Court’s] clearly established precedent,”38 and second, where a “state court confronts a set of facts that are materially indistinguishable from a decision of [the Supreme Court] and nevertheless arrives at a result different from our precedent.”39 Under AEDPA, “state-court judgments must be upheld unless, after the closest examination of the state-court judgment, a federal court is firmly convinced that a federal constitutional right has been violated.”40

We must decide whether the state court’s decision to deny Hall the right of confrontation and cross examination in determining whether he was mentally retarded under Atkins was contrary to federal law. Looking to federal law of due process and the right of cross examination and confrontation as announced by the Supreme Court, I am persuaded that the relevant law is clear, and dictates the conclusion that the state court’s decisional process was contrary to this clearly-established law. The life or death of a defendant, determined without hearing cross examination to resolve disputed material facts, here violates the core principles of due process and Hall’s right of confrontation as announced by the Court.

It is helpful to lift up the questions that either are not presented or have otherwise been dispositively decided. I accept here the state’s refusal to grant a jury trial on the issue of mental retardation. Hall asked for a jury trial, but he also maintained that given the posture of the case, *382due process assured him more than a submission “on the papers.” I also give full force to the Supreme Court’s holding in Atkins and in Ford v. Wainwright leaving “to the State[s] the task of developing appropriate ways to enforce the constitutional restriction upon [their] execution of sentences,”41 while addressing the constitutional minimum required of the state proceeding. The relevant question to my eyes is the process required by federal law for the death eligibility determination made here, and whether the state court’s process directly conflicts with this law.42

The inquiry could begin and end with Ford, but it is helpful to ground this case in its place within the larger stream of due process precedent. Atkins holds that “the mentally retarded should be categorically excluded from execution”43 and that “death is not a suitable punishment for a mentally retarded criminal”;44 Ford establishes a “bar against executing a prisoner who has lost his sanity.”45 While the factors establishing insanity and mental retardation differ and, under each case, are established by the state,46 the result of the constitutional determinations required by Ford and Atkins is identical: both determinations, if established in the affirmative, render a defendant ineligible for a death sentence.

Ford was convicted of murder and sentenced to death in Florida47 Following trial and sentencing, Ford began to exhibit strange behavior and eventually became incomprehensible, speaking in a code known only to him. A psychiatrist diagnosed him with paranoid schizophrenia.48 His counsel requested that the state determine Ford’s competency. The Governor appointed a panel of three psychiatrists to determine Ford’s competency. Ford had one, thirty-minute meeting with the psychiatrists who collectively questioned him. Each then wrote a short report concluding that he had some form of psychosis but was competent to understand the death penalty and its consequences. Armed with these reports, the Governor decided that Ford was not insane and signed Ford’s death warrant.49 The state court denied Ford’s request for a hearing on his insanity. A district court, on habeas review, denied Ford’s petition for an evidentiary *383hearing on insanity.50

The Supreme Court found that since “the Eighth Amendment has been recognized to affect significantly both the procedural and the substantive aspects of the death penalty,” the adequacy of the procedures for determining Ford’s insanity rested on whether or not the Constitution barred the execution of an insane prisoner.51 In other words, a determination that the Constitution substantively bars the execution of certain types of people brings due process concerns to the forefront— with the substantive guarantee came assured procedures.52 Having determined that the Eighth Amendment was a constitutional bar to the execution of insane prisoners, the court turned to “whether the [court] was under an obligation to hold an evidentiary hearing on the question of Ford’s sanity.”53 The plurality held,

The adequacy of a state-court procedure ... is largely a function of the circumstances and the interests at stake. In capital proceedings generally, this Court has demanded that factfinding procedures aspire to a heightened standard of reliability. This especial concern is a natural consequence of the knowledge that execution is the most irremediable and unfathomable of penalties; that death is different.

Although the condemned prisoner does not enjoy the same presumptions accorded a defendant who has yet to be convicted or sentenced, he has not lost the protection of the Constitution altogether; if the Constitution renders the fact or timing of his execution contingent upon establishment of a further fact, then that fact must be determined with the high regard for truth that befits a decision affecting the life or death of a human being. Thus, the ascertainment of a prisoner’s sanity as a predicate to lawful execution calls for no less stringent standards than those demanded in any other aspect of a capital proceeding. Indeed, a particularly acute need for guarding against error inheres in a determination that in the present state of the mental sciences is at best a hazardous guess however conscientious.54

Justice Powell, concurring in Ford in the narrowest opinion,55 rejected the need for a full “sanity trial” but required procedural protections and a “fair hearing,” holding that

*384At least in the context of competency determinations prior to execution, the [due process] standard is no different from the protection afforded by procedural due process. It is clear that an insane defendant’s Eighth Amendment interest in forestalling his execution unless or until he recovers his sanity cannot be deprived without a “fair hearing.” Indeed, fundamental fairness is the hallmark of the procedural protections afforded by the Due Process Clause.56

I am persuaded that the state court acted contrary to this clearly-established Supreme Court precedent. Like Ford, Hall challenged the adequacy of the process afforded to determine the crucial fact underlying a substantive constitutional right, the right against cruel and unusual punishment. Just as the state and district courts denied Ford an evidentiary hearing when he requested one, both the state court and the district court denied Hall a meaningful hearing. Neither Ford nor Hall received a determination of insanity or mental retardation at trial — in Ford, this was denied because insanity is a pre-trial question, or a post-trial question for condemned prisoners; here, because mental retardation at the time of his trial was solely a mitigating factor and yet to be defined by the constitutional strictures of Atkins. Both Ford and Hall were denied the minimal due process afforded for the determination of a substantive constitutional right, process with a “high regard for truth that befits a decision affecting the life or death of a human being.”57 Instead, they were afforded only a minimal process void of any guarantees afforded by an adversarial hearing, such as confronting and cross examining witnesses to question their credibility. Ford lacked even the guarantees of the judicial system and received only a “cursory form of procedural review” in an executive forum. Hall had the benefit of a judicial system but received the same administrative-style review.

Cases defining due process, while not treating eligibility, bear on the clarity of Supreme Court precedent. Sentencing decisions on the elements of an offense and aggravating factors of course affect whether a death sentence is “appropriate,” but they do not reach the determinative question of whether the defendant is eligible for the death penalty. While sentencing decisions with respect to the underlying offense and accompanying factors involve a range of gradation and substantial discretion, the question of eligibility as determined by mental retardation or insanity is not a sentencing “factor.” As the Court determined in Atkins and Ford, the determination of eligibility is itself a substantive, constitutional guarantee governed by higher standards. The Ford Court alluded to this distinction between elements of death penalty offenses and sentencing factors on the one hand, and the determination of death eligibility based on a defendant’s mental abilities on the other, holding that heightened constitutional protections apply to the determination of insanity where the “ultimate decision will turn on the finding of a single fact, not on a range of equitable considerations.”58

In Emnund v. Florida, the Court held that under the Eighth Amendment, a defendant could not be sentenced to death for aiding and abetting a felony murder if there was no determination that the defen*385dant himself killed or attempted to kill the victim.59 The circuits differed over the constitutionally secured process for making this determination. The Fifth Circuit held that the guilt-innocence or sentencing phase of the trial must establish whether a defendant participated in murder or attempted murder, and the Eleventh Circuit held that a judge could make the finding.60 The Court in Cabana v. Bullock resolved the split, agreeing with the Eleventh Circuit and holding that “the Eighth Amendment is satisfied so long as the death penalty is not imposed upon a person ineligible under Enmund for such punishment .... The Eighth Amendment is not violated ... regardless of who makes the determination of the requisite culpability.”61 Cabana addressed only whether a jury trial was necessary for a finding of an element of a death-eligible offense and did not address the level of factfinding necessary for due process, holding,

[T]he question whether the defendant killed, attempted to kill, or intended to kill might in a given case turn on credibility determinations that could not be accurately made by an appellate court on the basis of a paper record .... We shall not now attempt to determine what factfinding procedures would be adequate in the particular case before us, for, as we shall see, the state courts have not yet purported to engage in the requisite factfinding, and we decline to decide the hypothetical question of the adequacy of that which has not yet occurred.62

The State argues that the Mississippi Supreme Court itself made a finding sufficient to satisfy Enmund in the course of its direct review of [defendant’s] conviction and sentence. We are very doubtful, however, that these assessments of the record were sufficient in themselves to constitute a finding that Bullock killed, attempted to kill, or intended to kill Dickson. The Mississippi Supreme Court obviously was not addressing the specific requirements set forth in Enmund, for that case had not yet been decided.63

Cabana’s core holding that an element of a death-eligible offense may be determined by a judge has since been eroded. Walton v. Arizona relied on Cabana and similar cases in holding that a jury need not determine the “aggravating circumstances ‘elements’ of ... [a death-eligible] offense.”64 The Court in Ring v. Arizona compared Walton’s holding to Apprendi v. New Jersey, which held that a sentencing judge’s determination that a crime had been committed because of racial animus— a determination that triggered the application of the Hate Crimes Statute — violated the requirement that “a jury determin[e] that [defendant] is guilty of every element of the crime with which he is charged, beyond a reasonable doubt.”65 Ring found the two cases irreconcilable and “overrule[d] Walton to the extent that it allows a sentencing judge, sitting without a jury, to find an aggravating circumstance neces*386sary for imposition of the death penalty.”66 The primary distinction between Cabana and Walton was that Cabana permitted a judge to determine a requisite element of a capital offense, whereas Walton found the same for the determination of an aggravating circumstance leading to a capital offense determination. Little remains of Cabana after Ring.

In Williams v. New York, the Court upheld judicial discretion to impose a death sentence where a jury found first-degree murder but recommended a life sentence.67 The Court denied the invitation to “draw a constitutional distinction as to the procedure for obtaining information where the death sentence is imposed.”68 Much has changed since 1949. The Court in Woodson v. North Carolina, striking down North Carolina’s mandatory death penalty statute, held,

the penalty of death is qualitatively different from a sentence of imprisonment, however long. Death, in its finality, differs more from life imprisonment than a 100-year prison term differs from one of only a year or two. Because of that qualitative difference, there is a corresponding difference in the need for reliability in the determination that death is the appropriate punishment in a specific case.69

The court also held,

[W]e believe that in capital cases the fundamental respect for humanity underlying the Eighth Amendment ... requires consideration of the character and record of the individual offender and the circumstances of the particular offense as a constitutionally indispensable part of the process of inflicting the penalty of death.70

Regardless of what remains of Cabana, Walton, and Williams after Ring’s holding, Apprendi, Woodson, Cabana, and Walton all addressed the question of whether a jury must determine a necessary element or aggravating circumstance of a capital offense, a question that I do not here ask or answer.

This brings me to other cases that do not speak to the due process afforded a determination of mental retardation or insanity but set heightened due process requirements for similar determinations. In Specht v. Patterson the Supreme Court held that a defendant who was convicted under the Sex Offenders Act but not sentenced under the Act could not be sentenced without a hearing or the “right of confrontation and so on.”71 That Act allowed a district court to order punishment for “one day to life,” including life imprisonment, if the defendant had been convicted of a sex offense, if the court received a written report arising from a full psychiatric examination of the patient, and if the court determined that the defendant “constitutes a threat of bodily harm to members of the public, or is an habitual offender and mentally ill.”72 The court that sentenced Specht did not afford him a hearing, and the Court held that this violated due process, stating,

*387[The Act] makes one conviction the basis for commencing another proceeding under another Act to determine whether a person constitutes a threat of bodily harm to the public, or is an habitual offender and mentally ill. That is a new finding of fact that was not an ingredient of the offense charged.73

Similarly, Atkins came after Hall’s trial, and demanded a new finding of fact. Of course, Almendarez-Torres v. United States74 and McMillan v. Pennsylvania75 have, since Specht, held that certain sentencing factors are not to be treated as elements of an offense and thus do not require a jury determination; and that “a sentencing factor that comes into play only after the defendant has been found guilty ... of [a] crime[ ] beyond a reasonable doubt”76 does not prevent a state court from “pursuing its chosen course in the area of defining crimes and prescribing penalties.”77 But those cases both recognized that the “defendant in Specht was ‘confronted with a radically different situation from the usual sentencing proceeding.’ ”78 In McMillan, for example, a judge could increase a defendant’s sentence if the judge found visible possession of a firearm. The Court held, “The finding of visible possession of a firearm of course ‘ups the ante’ for a defendant ... but it does so only in the way that we have previously mentioned, by raising the minimum sentence that may be imposed by the trial court.”79 Similarly, in Almendarez-Torres, where a penalty provision authorized an enhanced sentence for a recidivist, the sentence was “up to, but no more than, 20 years.”80 Neither of these cases rise to the level of life imprisonment, as in Specht, or to the potential for a life sentence. Nor do they undermine Specht’s holding, as they expressly distinguish their facts as “radically different” from Specht’s.

Other Supreme Court cases, like Specht, require an evidentiary hearing or similarly strict due process safeguards for findings far less weighty than mental retardation. Kennedy v. Mendoza-Martinez requires a full trial to “strip an American of his citizenship.”81 Mendoza-Martinez was convicted of draft evasion prior to the deportation proceedings that deprived him of citizenship.82 This did not change the Court’s due process conclusions. Mendoza-Martinez was “never tried for any crime the elements of which are identical with or totally inclusory of those” that led to his loss of citizenship,83 just as Hall never had the chance to address the question of mental retardation — the very question determining his life or death — in a live hearing. In Oyler v. Boles, the Court held that “[e]ven though an habitual criminal charge [under a recidivist sentencing statute] does not state a separate offense, the determination of whether one is an habitual criminal is ‘essentially independent’ of *388the determination of guilt on the underlying substantive offense.”84 Thus, “a defendant must receive reasonable notice and an opportunity to be heard relative to the recidivist charge.”85 As in Oyler, the determination of mental retardation in Hall is not a separate offense; it is a separate fact, and it determines whether or not Hall is a certain type of criminal — in Hall’s case, a death-eligible criminal.

When Atkins prohibited the execution of mentally retarded defendants, it did not add an element to a death-eligible offense or change a sentencing factor that could increase a defendant’s punishment from imprisonment to death. It established a core and freestanding constitutional principle, that determined whether a defendant would face life or death. A determination of mental retardation under Atkins implicates the due process principles established by Ford. “ ‘If there is one “fundamental requisite” of due process, it is that an individual is entitled to an “opportunity to be heard”.’ ”86 The question before that Court was whether a lower court — in that case the district court — “was under an obligation to hold an evidentiary hearing on the question”87 that determined Ford’s life or death, and it concluded that it was. I find no dearth of clarity in these basic principles of due process.

Ill

In applying the Supreme Court’s due process precedent in Ford and subsequent cases to the process here, it bears explaining that the CCA was not itself willing to decide the factual question of retardation on the trial record alone — that is, it conceded that it needed the written, and, I note, untested, statements “supplementing” the trial record.

Just as we must give deference to a lower court when it has appropriately addressed a habeas issue, the Court left to the states “the task of developing appropriate ways to enforce the constitutional restriction” in sentencing.88 Under Brise-no,89 decided after the CCA concluded that Hall was not retarded, Texas elected to follow the American Association on Mental Retardation definition, which requires: “(1) significantly subaverage general intellectual functioning [‘defined as an IQ of about 70 or below’90]; (2) accompanied by related limitations in adaptive functioning; (3) the onset of which occurs prior to the age of 18.”91 The process for determination of these three elements must of course meet constitutional standards. Evidence of mental retardation presented outside this framework as a mitigating factor of a capital sentence at trial, viewed retrospectively without a hearing, was contrary to the minimum due process required by federal law to determine Hall’s mental capacity under Atkins, as Texas has defined it.

It bears accenting that Hall’s claim was decided on the fly,92 and was decided be*389fore Briseno and subsequent cases that consistently allowed defendants to present live Atkins evidence at various stages of trial. The difference in process deployed in Hall as compared to Briseno and later Atkins cases says a great deal: Texas courts quickly found their footing post-Atkins, proceeding in a strikingly different way than they had with Hall’s claim of retardation. Briseno was also convicted before Atkins but, unlike Hall, received a five-day evidentiary hearing on his post-Atkins habeas claim of mental retardation.93 Similarly, in Morris, we remanded a defendant’s federal habeas claim to the district court for an evidentiary hearing to address his Atkins claim.94 In Briseno, when Texas set forth the Atkins test, the CCA’s language implied that an oral evi-dentiary hearing was necessary for a determination of mental retardation, finding that factors “which factfinders in the criminal trial context might ... focus upon in weighing evidence as indicative of mental retardation” included evidence such as whether the defendant responds “coherently, rationally, and on point to oral or written questions or do his responses wander from subject to subject” and whether “those who knew the person best during the developmental stage — his family, friends, teachers, employers, authorities— think he was mentally retarded at that time, and, if so, act in accordance with that determination.”95 Although these eviden-tiary standards could be ascertained to a limited degree on paper, the factfinder’s burden of making the final determination “based upon all of the evidence and determinations of credibility”96 cannot be met by a judge’s credibility assessment of conflicting affidavits. And in the Atkins case itself, the Virginia court on remand, considering “what type of proceeding is necessary and consistent with the Supreme Court’s opinion”97 and the fact that “the question of Atkins’ mental retardation has never been answered,”98 determined that the case should be sent to the county court “for a hearing on the sole issue of whether Atkins is mentally retarded”99 as defined by the Virginia Code. Further, the court explicitly found that in a trial where the jury was “required to consider evidence of mental retardation in mitigation of capital murder but ... not required to make a definitive determination whether [defendant] suffers from mental retardation,” the issue of mental retardation has not been fully addressed.100

Atkins both clarified the definition of mental retardation and moved it from a mitigating circumstance to a complete bar to execution after a hearing. And while that may not require another hearing where “[t]he question whether [defendant] is mentally retarded” was “highly contested” at trial,101 suggesting both that the defendant had ample opportunity for confrontation and cross examination, it is not this case. Here, mental retardation was solely a mitigating factor at trial, and the trial did not afford adequate process for *390reaching mental retardation, a technical, three-part test in Texas. In Louisiana, the courts have recognized their inability to rely upon mental retardation evidence presented for mitigation as evidence of Atkins mental retardation, stating: “Atkins changed what would be considered relevant. Prior to the trial, mental retardation was merely a factor in mitigation. Post Atkins, mental retardation is a complete prohibition against imposition of the death penalty .... ”102

The reality is that determining mental retardation in general and certainly under Atkins engages assessments of testifying witnesses as well as relevant records. As the CCA stated in Briseno when establishing the Atkins mental retardation standard for Texas:

[A]lthough experts may offer insightful opinions on the question of whether a particular person meets the psychological diagnostic criteria for mental retardation, the ultimate issue of whether this person is, in fact, mentally retarded for purposes of the Eighth Amendment ban on excessive punishment is one for the finder of fact, based upon all of the evidence and determinations of credibility.103

There is nothing new here. The Supreme Court has consistently held that evidentiary hearings are essential for determinations of credibility.104 The Advisory Committee to the Standing Committee on Federal Rules has agreed, stating: “When the issue is one of credibility, resolution on the basis of affidavits can rarely be conclusive, but that is not to say that they may not be helpful.”105 Indeed, affidavits can be helpful, but the credibility of the writer of an affidavit can be fairly tested only in a hearing.

Reciting the rote that oral hearings test credibility is an anemic and inadequate statement of their force. It is the hearing in open court which offers the opportunity to expose the very core of the evidence, its accuracy, and its weight. Our faith in live hearings is a judgment made at least two centuries ago, reaffirmed for so long as to become a central part of this country’s democratic tradition and of a piece with its sense of fairness, and its defining of the relationship of citizen and state. The examples are many. I pause only to remind of a few. The Supreme Court’s extensive examination of the role of magistrate judges in holding fact-finding hearings, which the trial judge then accepts on paper, demonstrates well the limits of the paper record. In holding that a magistrate under the Federal Magistrate Act cannot preside over jury voir dire, and reserving that function for a trial judge, the Supreme Court in Gomez spoke to the importance of the judicial function and the trial judge’s responsibility, which rises far above simple document review:

*391Like motions to suppress evidence, petitions for writs of habeas corpus, and other dispositive matters entailing evi-dentiary hearings, jury selection requires the adjudicator to observe witnesses, make credibility determinations, and weigh contradictory evidence. Clearly it is more difficult to review the correctness of a magistrate’s decisions on these matters than on pretrial matters, such as discovery motions, decided solely by reference to documents.106

As the Gomez Court further explained, the nature of voir dire requires the court to “scrutinize not only spoken words but also gestures and attitudes of all participants ... But only words can be preserved for review; no transcript can recapture the atmosphere of the voir dire, which may persist throughout the trial.”107 Peretz’s holding after Gomez — allowing a magistrate to preside over voir dire when the defendant waives his right to voir dire in the trial court — does not diminish Gomez’s findings. When the Court upheld this waiver for felony trials, it emphasized that “the duties that a magistrate may perform over the parties’ objections are generally subsidiary matters not comparable to supervision of jury selection. However, with the parties’ consent, a district judge may delegate to a magistrate supervision of entire civil and misdemeanor trials.”108 The Court in Raddatz, in examining the ability of the magistrate to conduct a hearing on suppression of evidence, discussed this country’s deep-rooted belief in the value of live trials:

More than 100 years ago, Lord Coleridge stated the view of the Privy Council that a retrial should not be conducted by reading the notes of the witnesses’ prior testimony: “The most careful note must often fail to convey the evidence fully in some of its most important elements .... It cannot give the look or manner of the witness: his hesitation, his doubts, his variations of language, his confidence or precipitancy, his calmness or consideration; ... the dead body of the evidence, without its spirit; which is supplied, when given openly and orally, by the ear and eye of those who receive it.”109

And as English scholars have aptly commented, “an artful or careless scribe may make a witness speak what he never meant, by dressing up the depositions in his own form at language; but the witness is here at liberty to correct and explain his meaning, if misunderstood, which he can never do after a written deposition is once taken.”110 There are markers of the need for live hearings along the way of this country’s entire history. In discussing the importance of jury voir dire and deference to a trial court’s findings on challenges to members of a venire during voir dire, the Supreme Court one hundred twenty-eight years ago highlighted the types of findings that cannot be simply drawn from written evidence:

The reading of the evidence leaves the impression that the juror had some hypothetical opinion about the case, but *392it falls far short of raising a manifest presumption of partiality. In considering such questions in a reviewing court, we ought not to be unmindful of the fact we have so often observed in our experience, that jurors not unfrequently seek to excuse themselves on the ground of having formed an opinion, when, on examination, it turns out that no real disqualification exists. In such cases the manner of the juror while testifying is oftentimes more indicative of the real character of his opinion than his words. That is seen below, but cannot always be spread upon the record.111

More recently, the Court has focused on the importance of deference to the findings of a trial judge, since the parties at trial are “forced to concentrate their energies and resources on persuading the trial judge that their account of the facts is the correct one,” and “only the trial judge can be aware of the variations in demeanor and tone of voice that bear so heavily on the listener’s understanding and belief in what is said.”112 Justice Powell in Ford discussed how the “competency determination depends substantially on expert analysis in a discipline fraught with subtleties and nuances.”113 And although Justice Powell’s opinion did not conclude that this determination required a full trial, just as I do not maintain that a formal trial would be necessary for an Atkins determination, he recognized the basic need for the parties to introduce arguments before an impartial decisionmaker.114

Courts’ emphasis, both historical and modern, on the need for a trial judge or an impartial decisionmaker to hear the most important issues in a case speaks directly to the necessity of a hearing to determine mental retardation under Atkins, a hearing that tolerates confrontation and cross examination. Without such a hearing, the judge lacks the most essential aspects of a mental retardation determination — observation of the witnesses, with the credibility of experts’ motivations and the confidence of the affiants exposed to cross examination in an open courtroom.

IV

Blind deference to the state’s decision in this case would deprive Hall of his most basic due process rights. The trial court’s examination of the mental retardation claim as one potential mitigating factor to a capital sentence, before Atkins was decided, tells us little about the evidence that is relevant to an Atkins claim. Nor do the criminal and district courts’ re-examinations of the mental retardation evidence presented for mitigation, as well as post-conviction affidavits which presented new and important evidence of mental retardation, serve as “hearings” on the Atkins claim to which we must apply a presumption of correctness and give deference. There is a backdrop to this deficient hearing that cannot be ignored and that is the state trial courts’ consistent and complete adoption of all of the State’s findings throughout this process:115 this creates a *393black hole of deference and assumption that is not defensible. Both courts attempted to weigh the mitigation evidence themselves, guessing as to the credibility, reliability, and accuracy of the purveyors of that evidence and the Atkins affidavits, and finally deciding who to believe by reading works not necessarily and likely not crafted by the “witness.”116 But in doing so, they demonstrated the very flaws that render its paper determination of mental retardation, made without a hearing, an inadequate purveyor of due process.

The only mental retardation evidence heard live in this case is the limited testimony and cross examination on mitigation from Hall’s trial, which occurred before the Supreme Court’s Atkins decision. Many of the affiants have only been “heard” on paper — including two new experts (Dr. Church and Dr. Denkowski) who have never been cross examined. And none of the witnesses, expert or lay, have been cross examined on the issue of mental retardation as defined by Atkins. The CCA relied wholly on the paper evidence from this trial court record and additional affidavits submitted by both parties in rejecting Hall’s Atkins claims. In the initial hearing, Hall and the state addressed mental retardation as a potential mitigating factor in Hall’s death sentence and not in light of the three specific Brise-no factors that define Atkins mental retardation in Texas. Much of the testimony focused on Hall’s family life, emotional abuse, and the question of Hall’s understanding of culpability and the “wrongness” of his acts, rather than on whether or not he was in fact mentally retarded according to Briseno’s “scientific” definition of that term. The question to which all of this evidence was directed did not even mention the word “mental retardation.” Rather, the question on mitigation (Special Issue Number 3) asked:

Whether, taking into consideration all of the evidence, including the circumstances of the offense, the Defendant’s character and background, and the personal moral culpability of the Defendant, there is a sufficient mitigating circumstance or circumstances to warrant that a sentence of life imprisonment rather than a death sentence be imposed.

The testimony and cross examination addressing Hall’s IQ showed that the issue was controversial and muddled,117 and the *394state’s only expert discussed Hall’s adaptive knowledge, not functioning. The closing arguments from the trial record most strongly illuminate the dearth of sufficient Atkins evidence in the trial court’s paper hearing, as shown by the state’s sweeping assertion that mental retardation was only for “labeling purposes” and had little relevance to the question of mitigation.

Ironically, the CCA, in finding that Hall was not mentally retarded for a second time and relying upon the paper trial record and habeas affidavits, stated: “At no point did appellant [during trial] request that the trial judge or the jury make a specific fact-finding as to whether appellant was in fact mentally retarded.”118 Had Hall and his attorneys possessed extrasensory abilities — and had they known that Hall’s initial trial would be their only opportunity to present live evidence of the very issue of Atkins mental retardation that would determine Hall’s fate — they would have of course seized the opportunity to present their proof of mental retardation in open court and not by mail. The state, in turn, would have offered rebuttal evidence and cross examined defendants’ witnesses and experts on their credibility as it specifically related to the three Brise-no factors of mental retardation in Texas. But neither side did so, because neither could have known that the Court would declare a new constitutional standard in Atkins. Further, neither side could have predicted that the state and district courts would have refused the parties an opportunity to confront and cross examine the expert witnesses who offered conflicting opinions. And the brief opportunity afforded to present Atkins evidence on paper failed to remedy the absence of genuine adversarial contest. Between Hall’s state application for habeas, which claimed that it was unconstitutional to execute a mentally retarded defendant in light of the pending Atkins case, and the state’s reply, the Supreme Court decided Atkins. Therefore, both the habeas application and the reply contained Atkins-type arguments, and both Hall and the state relied upon the Texas Legislature’s and Health Code’s definition of mental retardation, which is similar but not identical to the AAMR test for mental retardation under Atkins that the Briseno court later adopted. The TLHC definition of a mental retardation is “significantly subaverage general intellectual functioning that is concurrent with deficits in adaptive behavior Further, this condition must have originated during the developmental period.” Briseno’s AAMR test requires “limitations” in adaptive functioning that are “related” to the subaverage intellectual functioning, with onset before age 18. Although the trial court’s paper habeas hearing, which occurred just after Atkins was decided, ostensibly allowed the parties to “argue” the issue of mental retardation under a definition similar to the Briseno standards, the arguments put forth on paper left gaps in the focus demanded by Atkins. At best, the experts reiterated prior arguments over IQ and failed to resolve them, bickering over standard error and citing incorrect IQ results. Two new experts presented affidavits and neither of these experts were ever cross examined as to their competency. They asserted their own views on the proper means for testing IQ and adaptive functioning, as well as the proper measurement of margins of errors and application of test results to the Atkins factors, yet they never debated these views live. Significantly, Dr. Denkowski pointed out potentially major flaws in some of the analyses relied upon by Dr. Price and others (that the *395TONI test, for example, does not even technically measure IQ), yet most of these flaws have not been tested in a courtroom. And the lay affiants presented broad claims that would have likely been pierced, or substantially altered in strength, upon cross examination. The judges in each step of Hall’s case and collateral review decided they could sort through the complicated scientific evidence and conflicting lay opinions themselves, without the aid of adversarial truthseeking. This occurred despite the failure of the trial record or the supplemental affidavits to sufficiently address any of the Briseno factors. Regarding the question of onset before the age of 18, the state’s only argument directly addressing this issue covered less than one page of its reply to Hall’s habeas application and relied upon the controversial school IQ scores of 71 on the WISC-R and 84 on the TONI, the faults of which Dr. Cunningham had raised in trial and Dr. Denkowski further explored on paper.119 In alleging that Hall did not have significant limitations in adaptive functioning, the state relied wholly on descriptions of Hall’s participation in the murder (alleging that this demonstrated his ability to function competently), the testimony of his family from the trial testimony, lay affidavits from prison guards, and on Dr. Price’s description of the Street Survival Skills Questionnaire that he administered on Hall — a test for adaptive knowledge, not functioning. We expect a reasonable level of detail in determining whether a defendant meets the Briseno factors. As the Court in Atkins explained, “To the extent there is serious disagreement about the execution of mentally retarded offenders, it is in determining which offenders are in fact retarded.”120 This type of controversial determination requires a careful and reasoned investigation of relevant facts, and the general allegations of parties on paper fail to come close to the level of relevance required for an Atkins determination.

In mental retardation cases, the CCA has specifically found that when a trial court determines an Atkins claim, “the trial court must not rely so extensively upon ... expert testimony as to commit the ultimate decision of mental retardation to the experts.”121 The state’s post-Atkins argument that Hall is not retarded rests in part on the affidavits of prison guards. These witnesses, given the nature of their job and its accompanying dangers, may not be inclined to volunteer evidence of mental retardation to state prosecutors. Additionally, as the dissent to the opinion of the CCA observes, the guards demonstrated only vague and largely irrelevant understandings of mental retardation122 while simultaneously asserting that Hall appeared normal. The Supreme Court has found — in civil cases — that an individual’s knowledge of the nature of the Government’s argument, protected by “the requirements of confrontation and cross-examination,” is “important in the case of documentary evidence,” but “even more important where *396the evidence consists of the testimony of individuals whose memory might be faulty or who, in fact, might be ... persons motivated by malice, vindictiveness, intolerance, prejudice, or jealousy.”123 While the right to cross examination rests largely on questions of “credibility and veracity,” “[a]n individual has those rights because facts are in issue.”124 And we know little about how the other lay witnesses actually “understood” mental retardation, or whether they observed Hall for a sufficiently long period of time to accurately observe his mental capacity. Many statements in affidavits from both the state and the defense indicate precisely how these issues begged for oral hearing. The affidavit of a waitress who served him once, for example, would not likely carry much weight in an Atkins hearing. Prison guards repeatedly made broad, sweeping allegations, such as “Michael Hall is just a normal inmate to me,” “I have never heard Michael Hall referred to as ‘Half Deck’ by anyone,” and “I had the usual contacts and conversations with Hall and observed nothing unusual,” while teachers claimed that Michael “demonstrated behaviors similar to those of a child with a diagnosis of Mental Retardation” and his trial counsel indicated that “Michael tried to mask his retardation ... he bitterly did not want people to think him ‘dumb’ .... ”

As Hall argued in his habeas application: For any fact-finder to make a determination as to whether Applicant is mentally retarded, it will be necessary to hear the testimony [of] people such as teachers, counselors, and mental health providers who had contact with Applicant before and after his incarceration concerning Applicant’s adaptive functioning. Additionally ... the fact-finder will be required to weigh the testimony of witnesses from both sides — psychologists who had interviewed Applicant ... The fact-finder will also have to hear from experts in the field of mental retardation so that the fact-finder [will] be able to gain an understanding of exactly what mental retardation is, and the variety of ways that it is manifested in those afflicted with it. It will be impossible for any fact-finder to make an informed and intelligent decision about which experts are most qualified and most credible in making their respective conclusion without hearing from the experts in court and subjecting them to the rigors of a Daubert/Rule 702 hearing and cross-examination.

A judge presiding in court at an oral hearing has “the superior opportunity of an observer of witnesses to comprehend their testimony,”125 an advantage in “essaying the truth of a matter when the facts are caught up with a witness’s manner of expression.”126 Our principles of deference to trial courts are clear, and we follow these principles because of the trust that we place on the trial process to do justice to the evidence placed before it.

Returning to basics, an oral hearing with cross examination of experts allows the trier to evaluate not only the raw contents of that evidence but the way in which the evidence plays out as presented live: the inconsistencies that arise, the phrases that went unnoticed yet carry great weight, and the responses to cross examination that may bring out weakness*397es in an assertion that appears bullet proof on paper. It is sometimes the silences in the courtroom that confess a realization of truth but are not preserved for reading by judges at another place and another time. Hearings in open court, or “little” trials, are not perfect but they are the best that we do in a system that remains as fair as any in the world. And they must not be jettisoned in the elusive and illusive pursuit of “efficiency.” Of course efficiency is an important component of a fair system, at least as much as the system can stand. The reality is that far less time would have been taken by a hearing than the time consumed in the struggle over its absence. Submissions on paper are the staple of the administrative state. I do not gainsay the role of our administrative state in saying that whether a man lives or dies at the hand of his government for his conduct is not to be decided by administrative agencies. Nor, I say, with all respect, by courts who unwittingly ape them by accepting the adequacy of administrative manners not prescribed for courts of law but as an alternative to them.

Here, judges distant from any live testimony or cross examination attempted to decipher the evidence for themselves, despite differing accounts of the accuracy of IQ tests and standard error, despite conflicting lay opinions on Hall’s mental abilities, and, most importantly, despite the state expert’s admission that the determination of mental retardation is a “judgment call” in this case and can “literally turn on a word or a number.” When a constitutional issue turns on a word or number — particularly here, where the first prong of Briseno’s mental retardation test requires an IQ of 70 or below for a finding of mental retardation — a judge’s analysis of paper arguments over the accuracy of IQ tests and complicated standard of error ranges cannot do justice to a defendant’s right to a determination of mental retardation. I have little doubt that the state trial judge would have never relied upon the erroneous misstatement of Hall’s IQ made in an affidavit of one of Hall’s expert witnesses had there been a hearing in more than name. It is the risk of just such errors that underpins the assurances of the procedural protections of due process. That a state is free to allocate the adjudicatory responsibility to trial or appellate courts does not reduce the constitutionally secured minimum for deciding a substantial claim of retardation under Atkins, bristling with sharply contested facts.

I have not invoked the familiar and reminded of primer rules and common understandings that came with the founding and have clung to our national psyche, powerfully informing our present elaboration because they have not been learned, but because in the daily mix of affairs they are occasionally forgotten.

This panel has unanimously concluded that the state did not afford Hall’s claim of retardation a full and fair hearing. It follows that we vacate the decision of the federal district court and remand for an evidentiary hearing. Here I would chart a different path. We should also make clear that because the finding that the state denied Hall a full and fair hearing on a claim of retardation is constitutionally footed, the state’s adjudication is constitutionally flawed and has resulted in a decision that was contrary to and an unreasonable application of federal law as determined by the Supreme Court. Refusing to allow confrontation and cross examination was an error of law, one that violated the federal law as established by the Supreme Court. We should vacate the decision of the district court with instructions to that court to enter an order that unless the state shall provide a constitutionally adequate evidentiary hearing on Hall’s claim within 120 days of the issuing of the man*398date, Hall shall not be eligible for death.127 To do otherwise does not give Hall his due^ — a hearing that not only affords the opportunity to confront and cross examine the state witnesses but also a hearing free of the deference the federal district court must give to the state adjudication of retardation — and passes over the reality that the state adjudication here was constitutionally flawed.