Opinion by Judge THOMAS; Concurrence by Judge REINHARDT; Dissent by Judge RAWLINSON.
THOMAS, Circuit Judge.In this appeal we consider whether the district court erred in denying a writ of habeas corpus sought as to petitioner’s conviction and death sentence. We affirm the district court’s judgment as to the conviction. However, we conclude that the Supreme Court’s decision in Ring v. Arizona, 536 U.S. 584, 122 S.Ct. 2428, 153 L.Ed.2d 556 (2002), applies retroactively so as to require that the penalty of death in this case be vacated.
I
It is the raw material from which legal fiction is forged: A vicious murder, an anonymous psychic tip, a romantic encounter that jeopardized a plea agreement, an allegedly incompetent defense, and a death sentence imposed by a purportedly drug-addled judge. But, as Mark Twain observed, “truth is often stranger than fiction because fiction has to make sense.”
There is no doubt that Warren Summer-lin is an extremely troubled man. He has organic brain dysfunction, was described by a psychiatrist as “functionally retarded,” and has explosive personality disorder with impaired impulse control. His father was a convicted armed robber who was killed in a shootout. As a youth, his alcoholic mother beat him frequently and punished him by locking him in a room with ammonia fumes. At his mother’s behest, he received electroshock treatments to control his explosive temper. He dropped out of school in the seventh grade due to dyslexia and committed numerous petty juvenile offenses. In 1975, he was diagnosed as a paranoid schizophrenic and treated with the anti-psychotic medication Thorazine.
Before his murder conviction, his only dangerous adult felony was a conviction for aggravated assault. That conviction arose out of a road rage incident in which a car veered off the road, jumped the curb and struck Summerlin’s wife, who was hospitalized for her injuries. At the scene, Sum-merlin brandished a pocket knife at the errant driver, an act that occasioned the filing of the criminal assault charge. Sum-merlin was not convicted of this offense until after these capital proceedings had commenced. However, this conviction later served as one of two statutory aggravating factors that qualified Summerlin for the death penalty.
On the morning of April 29, 1981, Bren-na Bailey, a delinquent account investigator for Finance America, went to Summer-lin’s home to speak with Summerlin’s wife about an overdue account. When Bailey’s boyfriend, Marvin Rigsby, learned that she had not returned to work as scheduled, he obtained the addresses of the places she had planned to visit that day and began to retrace her travel. In the early evening, he spoke with Summerlin, who told Rigsby that Bailey had left that residence at 10:30 a.m. The woman residing at the next address Bailey was slated to visit told Rigsby that she had been home all day and had not received a visit from anyone. After making additional attempts to find Bailey, Rigsby reported Bailey’s disappearance to the police that evening.
Later that evening, the police received a tip from a female caller to an anonymous *1085crime hotline service who stated that she believed that the missing woman from “Pacific Finance Company” had been murdered by Summerlin, who had rolled up the victim’s body in a carpet. The caller later was identified as Summerlin’s mother-in-law who testified that the basis of her information was her daughter’s extra-sensory perception.
Early the next morning, a road paving crew outside a market approximately a mile from the Summerlin residence alerted the market’s manager to a smell emanating from the trunk of a parked car, later determined to have been owned by Bailey. The manager recognized the odor as that of a decaying body and telephoned the police. Upon arrival, the officers observed a pair of panties, pantyhose, and shoes on the floor-board of the back seat. They forced open the trunk and found Bailey’s body, wrapped in a bloody bedsheet. Her skull was crushed, and she was partially nude.
The police obtained a search warrant for the Summerlin residence and found numerous items of incriminating evidence. After the search warrant was read to Sum-merlin, he stated, “I didn’t kill nobody.” When the detective did not respond, Sum-merlin asked: “Is this in reference to the girl that was at my house?” In response to the officer’s inquiry as to which girl he was referring, Summerlin described Bailey. Summerlin’s wife identified the bloody bedspread found with the victim as belonging to the Summerlin household. Summerlin was then arrested for the murder of Brenna Bailey. At the police station, Summerlin asked to speak to his wife and then made several incriminating statements while police officers were within listening distance.
The state trial court appointed the public defender’s office to represent Summer-lin. The' first attorney assigned to the case moved for a mental competency examination pursuant to Ariz. R.Crim. P. 11. Thereafter, the assigned attorney left the public defender’s office, and an attorney whom we shall refer to as “Jane Roe” was designated to represent Summerlin.
In Juné 1981, Summerlin was examined by two court-appointed psychiatrists, Drs. Maier Tuehler and Ótto Bendheim. Each found him competent to stand trial and legally sane under the M’Naghten standard, which then governed the determination of competency under Arizona law. Although there was no evidence of mental disease or defect, Dr. Tuehler observed that dyslexia and illiteracy made Summer-lin “functionally mentally retarded.” He further found that Summerlin’s impulse control was extremely impaired due to an explosive-type personality disorder and that he had an anti-social personality. Upon reading the reports, Judge David G. Derickson found Summerlin competent to stand trial.
During this time period, Roe had conversations with Dr. Leonardo Garcia-Bunuel, a psychiatrist who treated Summerlin at the Maricopa County Jail, regarding a possible diagnosis of psychomotor epilepsy. Summerlin had described to Dr. Garcia Buñuel details of the murder, particularly his experiences of sensing an intense perfume odor, and this led Dr. Garcia Buñuel to suspect that Summerlin may have had a temporal lobe seizure at the time of the killing. Subsequently, in August 1981, Roe arranged for neurological testing by Dr. Mark Winegard. An electroencephalogram (EEG) showed some slowing in Sum-merlin’s posterior temporal area but was insufficient to support a diagnosis of epilepsy. CAT scans and a second EEG performed in October 1981 were normal. As a result, Dr. Garcia Buñuel withdrew his concerns.
Roe also secured a psychological evaluation of Summerlin from Dr. Donald Tatro *1086in November 1981. Although concluding there was no evidence to support an insanity defense, Dr. Tatro found indications of organic brain impairment, border-line personality disorder, and paranoid personality disorder. In Dr. Tatro’s opinion, Summer-lin “is deeply emotionally and mentally disturbed, unaware of the motives underlying much of his behavior, and unable, because of his problems, to exercise normal restraint and control, once his highly unstable and volatile emotions are aroused.”
In November 1981, Roe began plea negotiations with the prosecution and obtained an extremely favorable plea agreement, which Summerlin entered into on November 17, 1981. The prosecutor, whom we will call “John Doe,” had been willing to enter into the agreement because he did not believe that the offense satisfied the Arizona legal standard of “heinous, cruel and depraved.” At the time, Summerlin had not been convicted of the aggravated felony arising out of the road rage incident, so it did not qualify as an aggravating factor under Arizona’s capital sentencing statute. See Ariz.Rev.Stat. § 13 — 703(F)(2) (1981) (amended in 1993). Thus, Doe did not believe that Summerlin had committed a capital offense.
Under the proposed plea agreement, Summerlin was to enter an Alford plea, see North Carolina v. Alford, 400 U.S. 25, 91 S.Ct. 160, 27 L.Ed.2d 162 (1970), which enabled him, without admitting guilt, to plead guilty to second-degree murder and aggravated assault and to be sentenced accordingly. The agreement stipulated that Summerlin would be sentenced to twenty-one years in prison for the murder of Ms. Bailey, of which he would be required to serve fourteen. In exchange, Summerlin would plead guilty to aggravated assault for the road rage incident with a maximum sentence of fifteen years and would, admit that he violated his probation in another case charging burglary. The plea agreement provided that Summerlin’s sentences on the three charges would run concurrently. However, the agreement was subject to court approval. If the court rejected the stipulated sentence, Summer-lin could either (1) allow his plea to stand and be sentenced to a term of up to thirty-eight-and-one-half years, according to the court’s sole discretion, or (2) withdraw from his plea of guilty and have the matters proceed to trial and disposition.
On the day he entered the plea, Sum-merlin properly answered all the questions required to validate his Alford plea. He had second thoughts a few days later, however, and formally sent to the court a pro se motion to withdraw his plea and to fire his public defender. In a court appearance on December 15, 1981, scheduled to address his motion, Summerlin openly registered dissatisfaction with the plea, the stipulated sentence, and Roe’s handling of his case.
After hearing his complaints, Judge Derickson denied Summerlin’s motions, but took the occasion to inform Summerlin that it was his intention on the upcoming sentencing date not to accept the stipulated sentence. Therefore, Summerlin would have the option either to withdraw from the plea or to allow it to stand and be sentenced accordingly.
Realizing that her client’s intention to withdraw from the agreement might again make him eligible for the death penalty, Roe promptly attempted to have the case transferred to another judge who might look more favorably on the deal. On December 18, 1981, the presiding judge denied her motion to disqualify Judge Der-ickson on the ground of prejudice towards her client and allowed Judge Derickson to continue with the case.
That same evening, Roe attended a Christmas party. She and prosecutor Doe left the party together and had'what she later described as a “personal involvement *1087... of a romantic nature.” As a result of that, as she later testified, she felt she “could no longer ethically represent Mr. Summerlin.” Because of the circumstances, she believed “that it would be appropriate for another Public Defender to handle the case and take it to trial, since it looked like it might be a trial at that point, because Mr. Summerlin indicated he wanted a trial and Judge Derickson had indicated he was going to reject the plea.” She reported the situation to her supervisor, and it was determined that the entire office probably was compromised.
Notwithstanding her belief that she could not represent Summerlin due to a personal conflict of interest, Roe took no immediate steps to accomplish her withdrawal. Neither she nor her office informed either the court or their client of her conclusion that she could no longer be Summerlin’s attorney. Instead, she accompanied him to and represented him at the next hearing before Judge Derickson on December 22,1981.
At the hearing, Judge Derickson advised Summerlin of his decision not to be bound by the sentencing part of the plea agreement, and that if Summerlin allowed his plea to stand, he was facing up to thirty-eight-and-one-half years in prison for the three offenses. After some confusion during which Summerlin told the court twice that he did not understand the Judge’s explanation of the sentence he now might face, Roe privately conferred with her client. Their discussion ended with Roe’s statement to the court, “I believe he understands, your Honor.” Summerlin’s immediate response was, “No, I don’t understand,” to which Roe replied, “Then what is your question?” Summerlin then asked about the number of years he might face on the three charges. Judge Derickson explained again the sentence that Sum-merlin would face if he permitted his plea to stand. To this Summerlin said that he finally understood, adding, “Okay. I would like to withdraw from my plea agreement. Is that what you want me to say?” Judge Derickson appropriately told Summerlin that he did not “want” Summerlin to say that, he simply wanted to make sure that Summerlin understood what would happen if he permitted the plea to stand. This exchange prompted another confidential discussion between Summerlin and Roe, followed by Summerlin’s statement that, “It says, if this plea agreement should be changed in any way, I can withdraw.” “Yes, that’s the question he asked you,” Roe replied. Summerlin then withdrew from the agreement. The court immediately reinstated his pleas of not guilty to the two consolidated cases, vacated its findings in the pending probation violation matter, and ordered that the matters be sent to the presiding judge for trial setting. Summerlin’s courtroom decision to withdraw his plea made him eligible for a conviction of first-degree murder and a sentence of death.
At this point in the hearing, Summerlin moved once again for new counsel. Roe remained silent. The court denied his motion, stating that “the record may further reflect that you failed to establish any grounds upon which counsel should be changed.” Judge Derickson later submitted an affidavit indicating that had he then known of the conflict, he would have granted Summerlin’s request to change counsel and would have continued the proceeding rather than proceeding with the plea colloquy-
On December 28, 1981, six days after Summerlin withdrew his plea, Roe finally broached the problem with Doe. On behalf of her client, she wanted Doe to stay on the case because he favored disposing of it with a lesser plea. He could discern no reason to step down as the prosecutor. After this discussion, a hearing was arranged for December 28, 1981, at which *1088Roe planned to move to withdraw as counsel and to permit the rest of the Public Defender’s Office to withdraw also. By this time, the case had been assigned to another judge, Judge Riddel, for trial. On December 28, Judge Riddel’s calendar was being handled by Judge McCarthy. Roe did not inform Summerlin of her intent to seek withdrawal or of the conflict that had precipitated her decision. At the hearing, Judge McCarthy began by noting that it had been “brought to the attention of the Court that defendant Mr. Summerlin is dissatisfied with the legal representation he is presently receiving.” The judge asked Summerlin if that was correct, to which Summerlin responded, “Yes, sir.” The judge then noted that he had spoken with counsel in chambers “and apparently it is their feeling that it would be in the best interest of justice that new counsel be appointed.” Judge McCarthy then appointed George Klink, a private practitioner, as new counsel. Following the reassignment of counsel, Roe did not advise Summerlin of her conflict of interest because she saw “no reason to beat a dead horse.” Klink then assumed representation of Summerlin in both of the charges stemming from Bailey’s murder and in the separate charge of aggravated assault arising out of the road rage incident, which was unrelated to the murder.
Approximately six weeks later, the Arizona Attorney General’s Office assumed control of the prosecution due to the conflict of interest between Doe and Roe. The Attorney General made it plain that the case would not be settled by way of a lesser plea. Klink filed a motion disqualifying Judge Riddel, and Summerlin’s murder case was then assigned to Judge Philip Marquardt.
Klink had intended to disqualify Judge Riddel as to the separate aggravated assault charge filed on the basis of the road rage incident. However, he failed to take the appropriate measures to accomplish this aim. After discovering this, Klink moved for a continuance of the assault trial because he was unprepared. His motion was denied and the assault trial proceeded. Klink called only one witness, Summerlin’s wife. Summerlin was convicted of aggravated assault and this conviction served several months later as one of the two aggravating circumstances in the penalty phase of his murder trial.
Klink spoke with Roe about the murder case and the medical reports. However, he never spoke with any of the experts who had interviewed Summerlin. He attempted to get another psychiatric expert, but he failed.
Klink’s main defense theory for the murder trial was Summerlin’s putative lack of premeditation. Klink presented no evidence supporting that theory. He cross-examined several prosecution witnesses in an attempt to east reasonable doubt on the rape charge as a way of proving lack of premeditation. Because the prosecution offered no psychiatric evidence, Klink could not cross-examine anyone regarding Summerlin’s psychiatric problems and how they would affect his ability to premeditate the murder. Klink called only one witness, Roe, and he only asked her a few questions in order to impeach one of the coroner’s statements regarding the length of time seminal fluid remains in the body. The entire case lasted only four days, and the jury was out for a little over three hours. The jury found Summerlin guilty of both first-degree murder and sexual assault.
The judge set a sentencing hearing to hear testimony and argument on aggravating and mitigating circumstances approximately one month after the verdict. In that month, Klink never met with Sum-merlin. Klink knew that the prosecution planned to call Drs. Tuchler and Bendheim at sentencing, but he never contacted ei*1089ther of them. Klink knew that Summerlin had been convicted of only one dangerous felony — the aggravated assault that Klink had tried before another judge. He also knew of mitigating circumstances surrounding that assault, including that the victim was not physically harmed and that Summerlin’s reaction was in response to witnessing the woman striking his wife with her car. Klink nonetheless did not present this information to the judge. Klink prepared no evidence regarding Summerlin’s social history despite references in Dr. Tatro’s report that Summerlin possibly experienced severe physical and emotional abuse in his childhood.
The sentencing hearing commenced on the afternoon of July 8, 1982. It was an extremely short proceeding, extending only twenty-six transcript pages, more than half of which constituted colloquy between counsel and the court. The court first entertained argument on the defense motion for a new trial, which the judge indicated he would consider over the weekend. The State’s aggravation case consisted of only one exhibit, specifically certified copies of documents relating to the aggravated assault conviction. The State then asked the judge to consider the trial testimony and rested its case. The entirety of the State’s aggravation case was recorded in one page of transcript.
For the defense case in mitigation, Klink called Dr. Tatro to the stand. However, before the witness could be sworn in, Sum-merlin interrupted and — although the conversation is not in the trial transcript— apparently requested that his attorney not present Dr. Tatro. Klink requested a five-minute recess, at the conclusion of which he stated: iCWith the consent of the Defendant, the Defendant has no witnesses in mitigation at this time and ... we’ll rest.” The judge then reminded Klink and Sum-merlin that this was the time set aside for the aggravation and mitigation hearing and that he planned to proceed with sentencing the next Monday. The judge then said, “so you tell me that you have one witness that you may present on Monday?” Klink replied: “Well, I would not call any witnesses at all.” The judge then indicated that he would allow Summerlin to make any statement that he wished to make, either at the present hearing or on Monday. Subsequently in the hearing, Klink noted that he would rely on the written report of Dr. Tatro attached to the presentence report. The State proceeded by presenting two rebuttal psychiatric witnesses.
Judge Marquardt advised the parties that he would deliberate over the weekend and announce his decision on Monday. Unbeknownst to Summerlin, Judge Mar-quardt was a heavy user of marijuana at the time, a fact that the State conceded in the federal habeas proceedings before the district court in this case.1
*1090The amount of marijuana that Judge Marquardt may have used during the trial or deliberations is unknown because the district court did not allow discovery on this issue, although there is record support for Summerlin’s claim that Judge Mar-quardt was either having difficulty concentrating or experiencing short-term memory loss.2
In any event, Judge Marquardt adjourned the penalty phase proceedings on Friday, indicating that he would deliberate over the sentence during the weekend and would also consider the motion for a new trial. However, on Monday, he either forgot or elected not to rule on the motion for a new trial and immediately proceeded with sentencing. Judge Marquardt began the hearing simply by announcing the case and inquiring whether Summerlin had anything to say or legal cause to show why judgment and sentence should not be pronounced. Klink stated he knew of no legal cause. Summerlin stated that he had a motion to vacate the judgment for the judge to consider. The judge examined the motion, took a five-minute recess, then denied it. The judge then heard brief oral arguments from the State and from Klink. He neither asked Summerlin whether he had anything further to say nor advised him of his right to allocution with respect to the sentence.
Judge Marquardt then sentenced Sum-merlin to death after finding two aggravating circumstances and no sufficiently substantial mitigating circumstances. The judge based his decision as to aggravating circumstances on two statutory grounds: (1) that the defendant had a prior felony conviction involving the use or threatened use of violence on another person, Ariz. Rev.Stat. § 13-703(F)(2) (1981) (amended in 1993); and (2) that Summerlin committed the offense in an especially heinous, cruel, or depraved manner, id. § 13-703(F)(6). He found no mitigating circumstances.
The same day, Judge Marquardt also sentenced James Clifford Fisher to death. Fisher had been charged with murdering Marguerite Bailey — no relation to Brenna Bailey — -with a blunt instrument. See State v. Fisher, 141 Ariz. 227, 686 P.2d 760, 758 (1984). As in the case with Sum-merlin, Judge Marquardt found two aggra*1091vating circumstances (including that the victim had been killed in an especially heinous and depraved manner) and no mitigating circumstances sufficiently substantial to call for leniency. Id. at 775-76. Fisher eventually received post-conviction relief on the basis of an unethical plea agreement that Judge Marquardt expressly entered into as a party and subsequently allowed into evidence at trial. See State v. Fisher, 176 Ariz. 69, 859 P.2d 179, 184 (1993). Summerlin alleges that Judge Marquardt confused some of the facts between the cases during Summerlin’s sentencing healing.
The Supreme Court of Arizona reviewed and affirmed Summerlin’s convictions and his sentence. See State v. Summerlin, 138 Ariz. 426, 675 P.2d 686 (1983), recons. den. Jan. 17, 1984. After an initial petition for writ of habeas corpus in federal district court and four unsuccessful post-conviction attempts in state court to overturn his conviction, Summerlin filed a second amended petition for writ of habeas corpus in the federal district court in Arizona on November 22, 1995. See 28 U.S.C. § 2254 (1994). The federal district court denied Summerlin’s second amended petition for writ of habeas corpus on October 31, 1997. Pursuant to Fed.R.Civ.P. 59(e), Summerlin moved to vacate the judgment on November 28, 1997. The district court denied this motion on January 12, 1998. However, the district court issued a certificate of probable cause enabling Summerlin to appeal pursuant to Fed. R.App. P. 22(b)(1). This timely appeal followed.
A divided three-judge panel of this Court issued its opinion on October 12, 2001, affirming the district court in part and reversing in part. See Summerlin v. Stewart, 267 F.3d 926 (9th Cir.2001). The case was remanded for an evidentiary hearing as to whether Judge Marquardt was competent when he was deliberating on whether to impose the death penalty. Id. at 957.
In the meantime, the United States Supreme Court granted certiorari in State v. Ring, 200 Ariz. 267, 25 P.3d 1139 (2001), cert. granted, 534 U.S. 1103, 122 S.Ct. 865, 151 L.Ed.2d 738 (2002), which involved a potential reexamination of Arizona’s death penalty statute in light of the Sixth Amendment. Because this was an issue that had been raised by Summerlin in his state and federal court petitions, the panel withdrew its decision and deferred submission of the case pending the Supreme Court’s resolution of Ring. See Summerlin v. Stewart, 281 F.3d 836, 837 (9th Cir. 2002). Later that year, the Supreme Court issued its decision in Ring v. Arizona, holding that Arizona’s capital sentencing scheme was incompatible with the Sixth Amendment right to a trial by jury. 536 U.S. at 609, 122 S.Ct. 2428.
Following the Supreme Court’s Ring decision, Summerlin moved to stay the proceedings in this case. Summerlin desired the stay so that he could request that the Arizona Supreme Court recall the mandate in his direct appeal to consider Ring’s application to his case. Such a procedure is cognizable under Arizona state law. See Lindus v. N. Ins. Co. of N.Y., 103 Ariz. 160, 438 P.2d 311, 313 (1968) (describing doctrine); see also State v. Ariz. Dep’t of Corrs., 187 Ariz. 211, 928 P.2d 635, 636 (1996) (applying procedure to consider ret-roactivity of judicial ruling). The panel granted the stay request. Subsequently, the Arizona Supreme Court denied Sum-merlin’s motion to recall the mandate. This decision exhausted all of Summerlin’s potential state remedies. Cf. Woods v. Kemna, 13 F.3d 1244, 1245-46 (8th Cir.1994) (noting that habeas petitioners need not ordinarily resort to extraordinary state remedies, such as recall of the mandate, to satisfy federal exhaustion requirements, *1092but holding that exhaustion might be required when there was a real possibility of relief under extraordinary and unique circumstances).
After the Arizona Supreme Court denied Summerlin’s motion to recall the mandate, the panel requested a vote of our Court as to whether this case should be reheard en bane. Following an affirmative vote of a majority of the non-recused active members of the Court, see Summerlin v. Stewart, 310 F.3d 1221 (9th Cir.2002), the case was reheard en banc on December 10, 2002.
We have jurisdiction pursuant to 28 U.S.C. § 2253. The Warden does not contend that Summerlin failed to exhaust his state remedies or that any of his claims are procedurally defaulted. Because this appeal was filed after the effective date of the Antiterrorism and Effective Death Penalty Act of 1996, Pub.L. No. 104-132, 110 Stat. 1214 (“AEDPA”), the right to appeal in this case is governed by AEDPA rules. See Slack v. McDaniel, 529 U.S. 473, 482, 120 S.Ct. 1595, 146 L.Ed.2d 542 (2000). However, because the petition for habeas corpus was filed before AEDPA’s effective date, pre-AEDPA law governs the petition itself. See Lindh v. Murphy, 521 U.S. 320, 327, 117 S.Ct. 2059, 138 L.Ed.2d 481 (1997).
Summerlin argues on appeal that:
1. He did not receive effective assistance of counsel during the guilt phase of his trial in violation of his rights under the Sixth Amendment;
2. The Arizona death penalty statute, as applied to him, is unconstitutional in that it permits a judge rather than a jury to determine the elements necessary for a capital sentence;
3. He did not receive effective assistance of counsel during the sentencing phase of his capital trial in violation of his rights under the Sixth Amendment;
4. His court-appointed public defender had a conflict of interest that adversely affected her representation at a critical stage of the proceedings, in violation of his rights under the Sixth Amendment;
5. He was deprived of his right to due process of law because the trial judge was addicted to marijuana during his trial and deliberated over his sentence while under the influence of marijuana; and
6. Cumulative errors require reversal of his sentence and conviction.
Summerlin’s only claim specific to the conviction phase alone is his argument that he received ineffective assistance of counsel during his guilt-phase trial. With the exception of his cumulative error contention, the remainder of Summerlin’s claims relate to the imposition of the death sentence.3
II
Summerlin alleges that he was denied the effective assistance of counsel at the guilt phase of his murder trial in violation of the Sixth Amendment as interpreted in Strickland v. Washington, 466 U.S. 668, 104 S.Ct. 2052, 80 L.Ed.2d 674 (1984). We review the dismissal of a habeas petition de novo, including the mixed questions of law and fact raised by claims alleging ineffective assistance of counsel. See Williams v. Woodford, 306 F.3d 665, 684 (9th Cir.2002); Hendricks v. Calderon, 70 F.3d 1032, 1036 (9th Cir.1995).
*1093To prevail on this claim, Summerlin must demonstrate first that the performance of his counsel fell below an objective standard of reasonableness, and second 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. Failure to satisfy either prong of the Strickland test obviates the need to consider the other. See id. at 687, 104 S.Ct. 2052.
We begin with the “strong presumption that counsel’s conduct falls within the wide range of reasonable professional assistance,” and with the acknowledgment that “[j]udicial scrutiny of counsel’s performance must be highly deferential.” Id. at 689, 104 S.Ct. 2052. However, defense counsel has “a duty to make reasonable investigations or to make a reasonable decision that makes particular investigations unnecessary.” Id. at 691, 104 S.Ct. 2052. “[Strategic choices made after less than complete investigation are reasonable precisely to the extent that reasonable professional judgments support the limitations on investigation.” Id. at 690-91, 104 S.Ct. 2052.
Summerlin alleges that his trial lawyer failed to investigate and to present his “only viable defense,” namely, that Sum-merlin had an organic brain dysfunction and an “impaired ability to premeditate or to exercise self-control.” To analyze this issue properly, we must examine the mental defenses then available under Arizona law. At the time, Arizona had adopted the M’Naghten test “as the sole standard for criminal responsibility.” State v. Ramos, 133 Ariz. 4, 648 P.2d 119, 121 (1982) (internal quotation marks omitted). To sustain a defense of legal insanity:
An accused must have had at the time of the commission of the criminal act: (1) Such a defect of reason as not to know the nature and quality of the act, or (2) If he did know, that he did not know he was doing what was wrong.
State v. Christensen, 129 Ariz. 32, 628 P.2d 580, 583 (1981).
At the time Summerlin was charged, Arizona already had rejected the affirmative defense of diminished capacity. See State v. Mott, 187 Ariz. 536, 931 P.2d 1046, 1051 (1997) (“Because the legislature has not provided for a diminished capacity defense, we have since consistently refused to allow psychiatric testimony to negate specific intent.”). Thus, the situation that confronted Klink was unlike that in Pirtle v. Morgan, 313 F.3d 1160, 1169-73 (9th Cir.2002) (holding that counsel had been constitutionally ineffective for not asserting a diminished capacity defense then available under Washington law).
The Arizona Supreme Court also had held, as a matter of law, that criminal defendants could not present psychiatric testimony to negate the element of specific intent. State v. Laffoon, 125 Ariz. 484, 610 P.2d 1045, 1047 (1980). Arizona did allow the introduction of psychiatric evidence as to a defendant’s tendency to act on impulse as probative of an absence of premeditation. Christensen, 628 P.2d at 582-83. However, under Arizona law, the standard for establishing premeditation is not high. The prosecution only need show that the defendant “had time to reflect after forming the intent to kill.” Clabourne v.Lewis, 64 F.3d 1373, 1380 (9th Cir.1995) (citing State v. Neal, 143 Ariz. 93, 692 P.2d 272, 276 (1984)). “This length of time could have been as instantaneous as it takes to form successive thoughts in the mind, and premeditation may be proven by circumstantial evidence.” Neal, 692 P.2d at 276 (citing State v. Lacquey, 117 Ariz. 231, 571 P.2d 1027, 1030 (1977)).
Thus, Summerlin’s trial counsel faced formidable legal hurdles in presenting a psychiatric defense at the guilt phase. *1094Nonetheless, counsel conducted a substantial amount of investigation into a potential psychiatric defense. Summerlin’s first counsel moved for a mental examination under Ariz. R.Crim. P. 11 to determine whether Summerlin was competent to stand trial. Upon examination, Dr. Tu-chler concluded that Summerlin, although “functionally mentally retarded,” did not have a mental disease or defect. Before she withdrew from the case, Roe thoroughly investigated Dr. Garcia-Bunuel’s suspicion of psychomotor epilepsy. She obtained neurological testing and pursued this possible diagnosis with Dr. Bendheim, as revealed in the following letter the doctor sent to Judge Derickson in December 1981:
We again discussed the possibility of psychomotor epilepsy, especially in view of Dr. Garcia Bunuel’s findings that this man had very vivid olfactory (smell) hallucinations preceding outbursts. I went over this whole situation again and told Miss [Roe] that the neurologists have been unable to find psychomotor epilepsy, although there was some slowing of the wave patterns in the temporal lobes, where psychomotor epileptic attacks usually originate.
While a positive electroencephalogram, which was not obtained here, would make a positive diagnosis, an essentially negative EEG does not entirely rule out the possibility of epileptic-type seizures, and for this reason I see absolutely no harm and potentially quite a bit of benefit to place this defendant on anti-epileptic, anti-seizure type medication, even though the diagnosis has not been established.
During post-conviction hearings, Roe testified that she met with trial counsel Klink on two or three occasions and spent a number of hours discussing her investigative efforts and the viability of a possible insanity defense. She stated that she discussed this aspect of the case with Klink “at great length,” explaining to Klink the examinations and conclusions of all of the examining doctors. Klink testified that, after consulting with Roe, he made a tactical decision not to pursue an insanity defense due to the lack of evidence.
Klink did not follow up on Dr. Garcia Bunuel’s earlier suspicion of psychomotor epilepsy because the doctor had changed his opinion and was out of the country at the time of trial. Instead, Klink made a decision to defend his client by arguing that the facts and circumstances of the prosecution’s case did not support a verdict of first-degree murder. Summerlin himself desired this fact-based defense.
In assessing an attorney’s performance, a reviewing court must make every effort “to eliminate the distorting effects of hindsight, to reconstruct the circumstances of counsel’s challenged conduct, and to evaluate the conduct from counsel’s perspective at the time.” Strickland, 466 U.S. at 689, 104 S.Ct. 2062. As the Supreme Court recently reiterated, this evaluation must include “an objective review of [counsel’s ] performance, measured for ‘reasonableness under prevailing professional norms,’ which includes a context-dependent consideration of the challenged conduct.” Wiggins v. Smith, — U.S. -, 123 S.Ct. 2627, 2536, 156 L.Ed.2d 471 (2003) (quoting Strickland, 466 U.S. at 688, 104 S.Ct. 2052).
A review of the record indicates that Klink’s trial performance did not fall below the objective standard of reasonableness required under Strickland. In deciding whether to pursue evidence of Summerlin’s mental state, Klink was entitled to rely on the opinions of the mental health experts who already had examined Summerlin. See Hendricks, 70 F.3d at 1038. At the time, none of the doctors would opine that Summerlin was suffering from a mental disease or defect that would *1095provide a foundation for an insanity defense. None of the physicians, including Dr. Garcia Buñuel, was able to diagnose Summerlin as clearly suffering from psy-chomotor epilepsy. It thus was reasonable for Klink not to investigate this possibility further. Cf. Wiggins, — U.S. at - -, 123 S.Ct. at 2536-38 (upholding an ineffective assistance claim against counsel who curtailed investigation despite promising leads in preliminary discovery).
Likewise, given the doctors’ inability to make a diagnosis, Klink’s tactical decision to forgo presenting what little evidence he had of epilepsy was certainly within the “wide range of professionally competent assistance.” Strickland, 466 U.S. at 690, 104 S.Ct. 2052; see also Harris v. Vasquez, 949 F.2d 1497, 1525 (9th Cir.1990) (“It is also acceptable trial strategy to choose not to call psychiatrists to testify when they can be subjected to cross-examination based on equally persuasive psychiatric opinions that reach a different conclusion.”).
Psychiatric testimony would have been admissible concerning Summerlin’s impulsive personality to show absence of premeditation. See Vickers v. Ricketts, 798 F.2d 369, 372-73 (9th Cir.1986). However, under the circumstances of the case, Summerlin has not shown that he was prejudiced by the failure to introduce such testimony. The basis of the State’s premeditation theory was not that Summerlin had planned the crime; rather, it was that he formed the required premeditation during the commission of the crime. To prove its point, the State relied on the circumstances surrounding the crime, including the fact that sexual assault occurred prior to the murder and the fact that Summerlin had retrieved a blunt object after the assault to commit the murder.
The State also relied on the uncontro-verted evidence as to how the murder was committed, specifically that Bailey had been hit repeatedly and forcefully on each side of her head. As the Supreme Court of Arizona later noted, Summerlin’s “excessive and purposeful actions demonstrate more than just a ‘reactionary’ homicide.” State v. Summerlin, 675 P.2d at 694. The State underscored this theory with presentation of graphic photographic evidence of the numerous wounds sustained by Bailey. The State’s witness testified that any one of the blows to the victim’s head was sufficient to kill the victim, yet numerous, deep lacerations were evident from the photographs.
Klink was not questioned during the post-conviction hearings about his choice not to present psychiatric evidence of impulsiveness, so we do not know whether this decision was strategic. However, after carefully reviewing the record, the district court concluded that there is no reasonable probability the jury would have acquitted Summerlin of first-degree murder had Klink introduced evidence of Summerlin’s impulsive personality. The district court therefore concluded that Summerlin could not establish Strickland prejudice as to this claim. See 466 U.S. at 691, 104 S.Ct. 2052 (“An error by counsel, even if professionally unreasonable, does not warrant setting aside the judgment of a criminal proceeding if the error had no effect on the judgment.”).
The district court’s assessment was correct. The psychiatric testimony on this point would have been limited to a general description of Summerlin’s behavioral tendencies. Given the State’s theory, this would have had only marginal probative value in determining whether Summerlin formed premeditation during the commission of the offense. The jury was instructed properly on the State’s premeditation theory, which was a correct statement of Arizona law. In this context, and considering the “totality of the evidence,” addi*1096tional psychiatric testimony would not have generated a “reasonable probability that at least one juror would have struck a different balance.” Wiggins, — U.S. at -, 123 S.Ct. at 2543. Thus, Summerlin has not established a “probability sufficient to undermine confidence in the outcome” of the guilt phase of his trial. Strickland, 466 U.S. at 694, 104 S.Ct. 2052.
Ill
The first penalty-phase question presented to us is whether the Arizona death penalty statute, as applied to Summerlin, is unconstitutional in that it permits a judge rather than a jury to determine the elements necessary for a death sentence. The Supreme Court recently has held that Arizona’s capital sentencing scheme was incompatible with the Sixth Amendment right to a trial by jury “to the extent that it allow[ed] a sentencing judge, sitting without a jury, to find an aggravating circumstance necessary for imposition of the death penalty.” Ring, 536 U.S. at 609, 122 S.Ct. 2428. The Supreme Court did not decide whether the holding in Ring applied to petitioners, such as Summerlin, who raised the constitutional challenge in collateral post-conviction proceedings rather than on direct appeal.
Because the Warden has argued that Teague v. Lane, 489 U.S. 288, 109 S.Ct. 1060, 103 L.Ed.2d 334 (1989), bars relief on this issue, we must decide whether Ring has retroactive application to cases on federal habeas review. Horn v. Banks, 536 U.S. 266, 272, 122 S.Ct. 2147, 153 L.Ed.2d 301 (2002) (holding that the court of appeals erred by not performing a Teague analysis when the issue was “properly raised by the state”) (citing Caspari v. Bohlen, 510 U.S. 383, 389, 114 S.Ct. 948, 127 L.Ed.2d 236 (1994) (“[I]f the State does argue that the defendant seeks the benefit of a new rule of constitutional law, the court must apply Teague before considering the merits of the claim”) (emphasis in original)).4
In short, now that the Supreme Court has decided that Timothy Ring’s capital murder conviction must be vacated because the judge was constitutionally disqualified from deciding whether Ring was eligible for the death penalty, the question *1097is whether others who received the same constitutionally infirm sentence, including those who previously raised the identical issue,5 are eligible for the same relief or whether they should remain subject to execution.
The question of whether a newly announced constitutional rule will apply retroactively on collateral review is a relatively recent inquiry in American jurisprudence. As Justice Holmes observed at the turn of the century, “[¿judicial decisions have had retrospective operation for near a thousand years.” Kuhn v. Fairmont Coal Co., 215 U.S. 349, 372, 30 S.Ct. 140, 54 L.Ed. 228 (1910) (Holmes, J., dissenting). At common law, the ret-roactivity question never arose because judges were believed to be discovering rules rather than declaring them. John C. Gray, The Nature and Sources of the Law 222 (1st ed.1909). Even now, a presumption exists that “a court is to apply the law in effect at the time it renders its decision, unless doing so would result in manifest injustice or there is statutory direction or legislative history to the contrary.” Bradley v. Sch. Bd. of Richmond, 416 U.S. 696, 711, 94 S.Ct. 2006, 40 L.Ed.2d 476 (1974). As the Supreme Court noted, “ ‘[Bjoth the common law and our own decisions’ have ‘recognized a general rule of retrospective effect for the constitutional decisions of this Court.’ ” Harper v. Va. Dep’t of Taxation, 509 U.S. 86, 94, 113 S.Ct. 2510, 125 L.Ed.2d 74 (1993) (quoting Robinson v. Neil, 409 U.S. 505, 507, 93 S.Ct. 876, 35 L.Ed.2d 29 (1973)).
Following the Civil War and enactment of the Fourteenth Amendment, Congress expanded the scope of habeas corpus review to cover challenges brought by those in state custody, see Act of 1867, ch. 28, § 1, 14 Stat. 385 (codified as amended at 28 U.S.C. § 2241(c)(3)), prompting the Supreme Court to determine the proper scope of federal habeas jurisdiction. By 1953, the Supreme Court confirmed the cognizability of all federal constitutional claims filed by state prisoners. Brown v. Allen, 344 U.S. 443, 73 S.Ct. 397, 97 L.Ed. 469 (1953). The expanding scope of federal review, coupled with a significant increase in the filing of federal habeas petitions by state prisoners, provided the Supreme Court with the opportunity to review for the first time a number of alleged constitutional deprivations. Walter V. Schaefer, Federalism and State Criminal Procedure, 70 Harv. L.Rev. 1, 21-22 (1956). Epochal constitutional criminal procedural protections were announced, and in their wake, a novel discussion arose as to whether a new constitutional rule of criminal procedure should be applied retroactively on direct or collateral review. This debate, a “product of the Court’s disquietude with the impacts of its fast-moving pace of constitutional innovation in the criminal field,” Mackey v. United States, 401 U.S. 667, 676, 91 S.Ct. 1160, 28 L.Ed.2d 404 (1971) (Harlan, J., concurring in part and dissenting in part), culminated in Linkletter v. Walker, 381 U.S. 618, 85 S.Ct. 1731, 14 L.Ed.2d 601 (1965).
In Linkletter, a defendant was convicted based on evidence that was obtained during a warrantless search. A year after the defendant had exhausted his state appeals, the Supreme Court decided Mapp v. Ohio, 367 U.S. 643, 81 S.Ct. 1684, 6 L.Ed.2d 1081 (1961). Subsequently, the defendant filed a habeas petition arguing that Mapp required reversal of his conviction. The Supreme Court held that even though “the *1098Constitution neither prohibits nor requires retrospective effect,” Linkletter, 381 U.S. at 629, 85 S.Ct. 1731, a constitutional rule of criminal procedure would not be retroactive unless, under a case-by-case analysis, three factors — the purpose of the new rule, reliance on prior doctrine, and the effect of retroactivity on the administration of justice — favor retroactive application of the rule. Id. at 636, 85 S.Ct. 1731. The Linkletter rule applied to convictions pending on direct review as well as to final convictions challenged collaterally by a federal habeas petition. Johnson v. New Jersey, 384 U.S. 719, 732, 86 S.Ct. 1772, 16 L.Ed.2d 882 (1966).
The tripartite Linkletter test proved difficult to apply. Justice Harlan observed that it had fostered the creation of “an extraordinary collection of rules to govern the application of that principle.” Desist v. United States, 394 U.S. 244, 256-57, 89 S.Ct. 1030, 22 L.Ed.2d 248 (1969) (Harlan, J., dissenting). He contended that the test produced inconsistent results, leading to different treatment for similarly-situated defendants. See id.
Justice Harlan remained critical of the Linkletter test throughout a series of subsequent cases. See, e.g., Mackey, 401 U.S. at 675, 91 S.Ct. 1160 (Harlan, J., concurring in part and dissenting in part); Coleman v. Alabama, 399 U.S. 1, 19, 90 S.Ct. 1999, 26 L.Ed.2d 387 (1970) (Harlan, J., concurring in part and dissenting in part); Von Cleef v. New Jersey, 395 U.S. 814, 817, 89 S.Ct. 2051, 23 L.Ed.2d 728 (1969) (Harlan, J., concurring in result); Jenkins v. Delaware, 395 U.S. 213, 222, 89 S.Ct. 1677, 23 L.Ed.2d 253 (1969) (Harlan, J., dissenting); Desist, 394 U.S. at 256, 89 S.Ct. 1030 (Harlan, J., dissenting). He argued instead that new constitutional rules ought to apply to all cases that were not final or that were pending on direct review. Mackey, 401 U.S. at 678-80, 91 S.Ct. 1160. He also contended that new procedural (as opposed to substantive) due process rules ought not to apply retroactively on habeas review unless the claim implicated procedures “implicit in the concept of ordered liberty” or addressed rules that “alter our understanding of the bedrock procedural elements that must be found to vitiate the fairness of a particular conviction.” Id. at 693-94, 91 S.Ct. 1160 (internal quotation marks and citations omitted).
The analytical framework propounded by Justice Harlan ultimately proved persuasive. In Griffith v. Kentucky, 479 U.S. 314, 107 S.Ct. 708, 93 L.Ed.2d 649 (1987), the Supreme Court adopted the first portion of Justice Harlan’s analysis, noting that “[i]n Justice Harlan’s view, and now in ours, failure to apply a newly declared constitutional rule to criminal cases pending on direct review violates basic norms of constitutional adjudication.” Id. at 322, 107 S.Ct. 708.
The Court therefore held that “a new rule for the conduct of criminal prosecutions is to be applied retroactively to all cases, state or federal, pending on direct review or not yet final.” Id. at 328, 107 S.Ct. 708.
Two years later, the Court clarified its retroactivity jurisprudence in the habeas context in Teague. Importing Justice Harlan’s analysis, Teague held that “[u]n-less they fall within an exception to the general rule, new constitutional rules of criminal procedure will not be applicable to those cases which have become final before the new rules are announced.” 489 U.S. at 310, 109 S.Ct. 1060.
Teague also adopted Justice Harlan’s two exceptions, providing that a new rule of criminal procedure would be retroactive if it “place[d] certain kinds of primary, private individual conduct beyond the criminal law-making authority to proscribe,” or if the rule “require[d] the observance of those procedures that ... are *1099implicit in the concept of ordered liberty.” Id. at 311, 109 S.Ct. 1060 (internal quotation marks omitted). The Supreme Court explained that the second exception had two components, formulated by combining aspects from Justice Harlan’s dissents in Desist and Mackey. The Court thus limited the exception for “watershed rules of criminal procedure” to those procedures that both “alter our understanding of the bedrock procedural elements that must be found to vitiate the fairness of a particular conviction,” id. (quoting Mackey, 401 U.S. at 693, 91 S.Ct. 1160) (internal quotation marks omitted; emphasis added in Teag-ue ), and “without which the likelihood of an accurate conviction is seriously diminished.” Id. at 313, 109 S.Ct. 1060.
Before applying these concepts to the instant case, it is important to set the appropriate analytic framework. The threshold question in a Teague analysis is whether the rule the petitioner seeks to apply is a substantive rule or a procedural rule, because “Teague by its terms only applies to procedural rules.” Bousley v. United States, 523 U.S. 614, 620, 118 S.Ct. 1604, 140 L.Ed.2d 828 (1998). If the rule is procedural, the court then conducts a three-step analysis to determine whether Teague bars its application. See O’Dell v. Netherlands 521 U.S. 151, 156-57, 117 S.Ct. 1969, 138 L.Ed.2d 351 (1997). First, the reviewing court “must ascertain the date on which the defendant’s conviction and sentence became final for Teague purposes.” Caspari, 510 U.S. at 390, 114 S.Ct. 948. Second, the court must survey “the legal landscape as it then existed,” Graham v. Collins, 506 U.S. 461, 468, 113 S.Ct. 892, 122 L.Ed.2d 260 (1993), to determine whether existing precedent compelled a finding that the rule at issue “was required by the Constitution.” Lambrix v. Singletary, 520 U.S. 518, 527, 117 S.Ct. 1517, 137 L.Ed.2d 771 (1997) (internal quotation marks and citations omitted). If existing precedent already required application of the rule, the Teague bar does not apply. If, by contrast, the procedure at issue is considered a new rule for Teague purposes, the court must proceed to the third step and determine whether either of the two announced exceptions applies. Teague, 489 U.S. at 307, 109 S.Ct. 1060 (plurality). The presumption against ret-roactivity is overcome only if the new rule prohibits “a certain category of punishment for a class of defendants because of their status or offense,” Penry v. Lynaugh, 492 U.S. 302, 330, 109 S.Ct. 2934, 106 L.Ed.2d 256 (1989), abrogated on other grounds by Atkins v. Virginia, 536 U.S. 304, 122 S.Ct. 2242, 153 L.Ed.2d 335 (2002), or presents a new “watershed rule of criminal procedure” that enhances accuracy and alters our understanding of bedrock procedural elements essential to the fairness of a particular conviction. Teague, 489 U.S. at 311, 109 S.Ct. 1060 (plurality; citations omitted).
IV
We first consider the threshold Teague question, namely whether Ring announced a substantive rule or a procedural rule. See Bousley, 523 U.S. at 620, 118 S.Ct. 1604. Unlike strictly procedural rules, “new rules of substantive criminal law are presumptively retroactive.” See, e.g., Santana-Madera v. United States, 260 F.3d 133, 138 (2d Cir.2001) (citing United States v. Mandanici, 205 F.3d 519, 525 (2d Cir.2000)), cert. denied, 534 U.S. 1083, 122 S.Ct. 817, 151 L.Ed.2d 701 (2002). Thus, the Teague retroactivity bar does not apply if the rule Ring announced is substantive, rather than procedural, in nature. Bousley, 523 U.S. at 620, 118 S.Ct. 1604.
As Erie doctrine demonstrates in the context of civil litigation, the distinction between “substantive” and “procedural” is not always easy to divine. Erie R.R. Co. v. Tompkins, 304 U.S. 64, 58 S.Ct. 817, 82 L.Ed. 1188 (1938); see also Hanna v. Plu-*1100mer, 380 U.S. 460, 471-74, 85 S.Ct. 1136, 14 L.Ed.2d 8 (1965); Guar. Trust Co. of N.Y. v. York, 326 U.S. 99, 109, 65 S.Ct. 1464, 89 L.Ed. 2079 (1945). The Supreme Court acknowledged this problem in a pre-Teague consideration of the difference between substantive and procedural criminal law, noting that “[w]e would not suggest that the distinction that we draw is an ironclad one that will invariably result in the easy classification of cases in one category or the other.” Robinson, 409 U.S. at 509, 93 S.Ct. 876.
However difficult it is to locate, though, “[t]his distinction between substance and procedure is an important one in the habeas context.” Bousley, 523 U.S. at 620, 118 S.Ct. 1604. In giving shape to this important distinction, the Supreme Court has understood decisions of “criminal procedure” to be those decisions that implicate how the criminal trial process functions. Under Teague, only those decisions of “procedure” that insert into the criminal trial process a mechanism “ ‘without [which] the likelihood of an accurate conviction is seriously diminished’” apply retroactively. Id. (quoting Teague, 489 U.S. at 313, 109 S.Ct. 1060).
Decisions of “substantive criminal law,” by contrast, are those that reach beyond issues of procedural function and address the meaning, scope, and application of substantive criminal statutes. Id. (noting that a Supreme Court holding is “substantive” for Teague purposes when it impacts the scope and application of a “substantive federal criminal statute”); see also Davis v. United States, 417 U.S. 333, 346, 94 S.Ct. 2298, 41 L.Ed.2d 109 (1974) (including within the definition of “substantive” those decisions that remove primary conduct from the purview of criminal punishment). Thus, for Teague purposes, a new rule is one of “procedure” if it impacts the operation of the criminal trial process, and a new rule is one of “substance” if it alters the scope or modifies the applicability of a substantive criminal statute. Bousley, 523 U.S. at 620, 118 S.Ct. 1604.
In Bousley, the Supreme Court applied this substantive procedural logic, rejecting the government’s Teague-based non-retro-activity argument because the case called for a construction of a federal statute. Teague, Chief Justice Rehnquist explained, “is inapplicable to the situation in which this Court decides the meaning of a criminal statute enacted by Congress.” Id. For the same reason, we recently determined that the rule announced in Richardson v. United States, 526 U.S. 813, 815, 119 S.Ct. 1707, 143 L.Ed.2d 985 (1999), requiring jury unanimity on individual violations alleged as part of a continuing criminal enterprise, is substantive, not procedural, under Teague. See United States v. Montalvo, 331 F.3d 1052 (9th Cir.2003). All of our sister circuits that have considered the question agree with this categorization.6
“[Significant” to both this court’s and our sister circuits’ understanding of Richardson’s rule as substantive is the fact that Richardson “was ‘deciding ] the meaning of a criminal statute.’ ” Montalvo, 331 F.3d at 1056 (citation omitted; alteration in original); see also Murr, 200 F.3d at 906. Explaining or redefining elements of an offense, we observed in Montalvo, constitutes a decision of substantive criminal law *1101for Teague purposes. 331 F.3d at 1055-56; see also United States v. Dashney, 52 F.3d 298, 299 (10th Cir.1995). Thus, because Richardson expressly “analyz[ed] what constitutes ‘elements’ as opposed to brute facts or ‘means’ ” of an offense, the rule Richardson announced was substantive for Teague purposes. Montalvo, 331 F.3d at 1056.7
In the habeas context in particular, as Chief Judge Becker has observed, there are those cases that do “not fall neatly under either the substantive or procedural doctrinal category.” United States v. Woods, 986 F.2d 669, 677 (3d Cir.1993). In such cases, “the best approach is to recognize that [the case at issue] is neither entirely substantive nor procedural.” Id. at 678. Ring is such a decision.
In one sense, Ring — like Apprendi — announced a procedural rule: Ring mandated that a jury, rather than a judge, must find aggravating circumstances in a capital case. Ring’s holding thus addressed, at least in part, the procedure by which any capital trial must be conducted. See Cannon v. Mullin, 297 F.3d 989, 994 (10th Cir.2002) (assessing the operation of Ring on Oklahoma law).8
In the context of substantive Arizona criminal law, however, Ring did more than answer a strictly procedural question. Thus, Ring is unlike Apprendi, in which the Supreme Court expressly declared that its decision had no impact on substantive criminal law, noting that “[t]he substantive basis for New Jersey’s enhancement is not at issue.” 530 U.S. at 475, 120 S.Ct. 2348. By important contrast, the substantive basis for Arizona’s capital sentencing scheme was precisely at issue in Ring.9 Ring rendered Arizona’s substan*1102tive capital murder statute unconstitutional. More than a procedural holding, Ring effected a redefinition of Arizona capital murder law, restoring, as a matter of substantive law, an earlier Arizona legal paradigm in which murder and capital murder are separate substantive offenses with different essential elements and different forms of potential punishment. That is, as applied to the particular Arizona murder statute at issue here, Ring’s holding was “substantive” for Teague purposes. See Bousley, 523 U.S. at 620, 118 S.Ct. 1604 (noting that a Supreme Court holding is “substantive” when it impacts the scope and application of a “substantive federal criminal statute”). A careful analysis of the structure and history of the relevant Arizona statutes, coupled with a close examination of the underlying rationale of Ring and the Supreme Court’s related jurisprudence, reveals that Ring is, as to Arizona, a “substantive” decision, even if its form is partially procedural.
In 1901, the Territory of Arizona enacted its first death penalty statute, leaving capital sentencing to the discretion of the jury except where the defendant entered a plea of guilty. See Ariz. Territorial Rev. Stat., tit. 8, § 174 (1901). In relevant part, the 1901 Arizona statute provided that
[e]very person guilty of murder in the first degree shall suffer death or imprisonment ... for life, at the discretion of the jury trying the same, or, upon the plea of guilty, the court shall determine the same.
Id.
In 1916, Arizona abolished the death penalty by state initiative, see Act of Dec. 8, 1916, 1917 Ariz. Session Laws, Initiative and Referendum Measures, at 4-5, but the 1901 death penalty statute was restored through similar political means in 1918. See Act of Dec. 5, 1918, 1919 Ariz. Sess. Laws, Initiative and Referendum Measures, at 18.
Following the 1918 initiative, Arizona’s death penalty scheme remained largely unchanged for more than 50 years. From 1919 until 1972, Arizona committed the decision as to whether to impose the penalty of death following a criminal trial to the complete discretion of the jury. See Hernandez v. State, 43 Ariz. 424, 32 P.2d 18, 20-21 (1934) (“It is clear from this that the question of punishment in first-degree murder cases is wholly within the jury’s discretion and that the court has no duty in connection therewith other than to advise it that it must determine which of the penalties-death or life imprisonment-shall be imposed upon the defendant if it finds him guilty of that offense.”).
In 1972, however, the Supreme Court held that death penalty statutes vesting complete discretion in the judge or in the jury, like Arizona’s, were unconstitutional. Furman v. Georgia, 408 U.S. 238, 92 S.Ct. 2726, 33 L.Ed.2d 346 (1972) (per curiam). The impact of Furman particularly is in-*1103struetive in this context. There was no doubt, importantly, that Furman had retroactive effect. See Moore v. Illinois, 408 U.S. 786, 800, 92 S.Ct. 2562, 38 L.Ed.2d 706 (1972) (applying Furman on habeas review); United States v. Johnson, 457 U.S. 587, 550, 102 S.Ct. 2579, 73 L.Ed.2d 202 (1982) (discussing same). Following the dictates of Furman and Stewart v. Massachusetts, 408 U.S. 845, 92 S.Ct. 2845, 33 L.Ed.2d 744 (1972), the Supreme Court vacated a number of Arizona death sentences, both on direct appeal and on collateral habeas review. See, e.g., Alford v. Eyman, 408 U.S. 939, 92 S.Ct. 2874, 33 L.Ed.2d 762 (1972) (habeas); Kruchten v. Eyman, 408 U.S. 934, 92 S.Ct. 2853, 33 L.Ed.2d 748 (1972) (habeas); Sims v. Eyman, 408 U.S. 934, 92 S.Ct. 2850, 33 L.Ed.2d 746 (1972) (habeas); Gause v. Arizona, 409 U.S. 815, 93 S.Ct. 192, 34 L.Ed.2d 71 (1972) (direct review). In light of this Supreme Court precedent, the Arizona Supreme Court declared the Arizona death penalty statute to be unconstitutional under the Eighth and Fourteenth Amendments. See State v. Endreson, 109 Ariz. 117, 506 P.2d 248, 254 (1973). The Arizona court acknowledged, after examining the structure of the relevant Arizona criminal statutes, that the Supreme Court, through Furman, had “abolished ‘capital offenses’ in Arizona” substantively. In re Tarr, 109 Ariz. 264, 508 P.2d 728, 729 (1973). In short, the effect of Furman in declaring Arizona’s capital murder statute unconstitutional was unquestionably substantive.
A year later, in 1973, Arizona enacted a new “capital offense” statute. This new statute established sentencing standards in capital cases and provided for sentencing by judge, rather than by jury. See Act of May 14, 1973, ch. 138, § 5, 1973 Ariz. Sess. Laws 966, 968-70. The 1973 statute identified six aggravating circumstances and four mitigating circumstances for sentencing courts to consider and required the court to impose a death sentence only if it found (1) one or more aggravating circumstances to exist and (2) no counter-vailing mitigating circumstances “sufficiently substantial to call for leniency.” Id. In State v. Richmond, 114 Ariz. 186, 560 P.2d 41 (1976), the Arizona Supreme Court upheld the constitutionality of the 1973 death penalty statute.
But the Supreme Court’s decisions in Lockett v. Ohio, 438 U.S. 586, 98 S.Ct. 2954, 57 L.Ed.2d 973 (1978), and Bell v. Ohio, 438 U.S. 637, 98 S.Ct. 2977, 57 L.Ed.2d 1010 (1978), which declared unconstitutional death penalty statutes that restricted the right of the defendant to show mitigating circumstances in capital cases raised anew questions of whether the 1973 Arizona statute could pass constitutional muster. To this end, in Bishop v. Arizona, 439 U.S. 810, 99 S.Ct. 69, 58 L.Ed.2d 103 (1978), the Supreme Court vacated and remanded an Arizona death sentence for reconsideration in light of Lockett. Following this remand, the Arizona Supreme Court held that Arizona’s 1973 death penalty statute was unconstitutional insofar as it precluded the defendant from proving non-statutory mitigating circumstances. State v. Watson, 120 Ariz. 441, 586 P.2d 1253, 1257 (1978).
Soon thereafter, in 1979, the Arizona legislature amended the State’s death penalty statute to conform to Lockett by defining as relevant mitigating circumstances “any factors offered by the state or the defendant which are relevant in determining whether to impose a sentence less than death.” These factors included, but were not limited to, the factors enumerated in the statute itself. Act of May 1, 1979, ch. 144, § 1, 1979 Ariz. Sess) Laws 449, at 450-51. The legislature added various aggravating and mitigating factors to the terms of the statute in 1977, 1978, 1984, and 1985, but, during this period, the *1104essential structure of Arizona’s death penalty statute remained the same. In 1983, the United States Supreme Court confirmed that, to be constitutional under the Eighth Amendment, a state’s capital sentencing scheme must “genuinely narrow the class of persons eligible for the death penalty and must reasonably justify the imposition of a more severe sentence on the defendant compared to others found guilty of murder.” Zant v. Stephens, 462 U.S. 862, 877, 103 S.Ct. 2733, 77 L.Ed.2d 235 (1983). Finding at least one aggravating factor “narrows the class of persons eligible for the death penalty.” Lowenfield v. Phelps, 484 U.S. 231, 244, 108 S.Ct. 546, 98 L.Ed.2d 568 (1988). Arizona’s substantive revisions to its capital murder statute were designed to pass these constitutional requirements. Thus, as a result of Furman and its progeny, the crime of capital murder in Arizona was substantively altered. As Justice Thomas has observed: “[I]n the area of capital punishment, unlike any other area, we have imposed special constraints on a legislature’s ability to determine what facts shall lead to what punishment-we have restricted the legislature’s ability to define crimes.” Apprendi, 530 U.S. at 522-23, 120 S.Ct. 2348 (Thomas, J., concurring).
In 1988, we considered a defendant’s Sixth Amendment challenge to the Arizona death penalty statute. See Adamson v. Ricketts, 865 F.2d 1011 (9th Cir.1988) (en banc). In Adamson, we noted that, “[u]n-der Arizona’s revised code, all murder is not capital murder,” id. at 1025, and that, under the code, Arizona had in effect defined capital murder to be a substantive offense separate from non-capital murder. Id. at 1026; see id. at 1025 (“[W]e recognize that the mere use of labels ... to compartmentalize the functions of judge and jury[ ] does not negate the very real possibility that what are called ‘sentencing’ decisions may in fact usurp jury factfind-ing responsibilities.”). We concluded in Adamson that, because the Arizona statute required the finding of aggravating factors before the death penalty could be imposed, the Arizona statute made these aggravating factors elements of the “distinctive offense of capital murder,” not mere sentencing factors relevant to increasing the punishment for a lesser offense. Id. at 1026-27 (emphasis in original). Accordingly, we held that “Arizona’s aggravating circumstances function as elements of the crime of capital murder requiring a jury’s determination.” Id. at 1027. Based on this understanding of Arizona law, we found Arizona’s identification and treatment of the “elements of the crime of capital murder as sentencing factors for determination by a judge,” rather than as offense elements to be determined by the jury, to be “impermissible]” and “in violation of the Sixth Amendment.” Id. at 1029.
In Walton, the Supreme Court abrogated our decision in Adamson. In pertinent part, Walton held that aggravating circumstances under Arizona law were only “sentencing considerations,” not “elements of the offense” of capital murder. Id. at 648, 110 S.Ct. 3047 (citing Poland v. Arizona, 476 U.S. 147, 156, 106 S.Ct. 1749, 90 L.Ed.2d 123 (1986)). But cf. id. at 710-14, 110 S.Ct. 3047 (Stevens, J., dissenting) (suggesting that, “under Arizona law,” the aggravating factors are “elements of a capital crime [ ] [that] must be determined by a jury”) (citations omitted). Thus, Walton refuted our decision in Adamson and concluded “that the Arizona capital sentencing scheme does not violate the Sixth Amendment.” Id. at 649.
Ring expressly overruled Walton in relevant part. 536 U.S. at 589, 122 S.Ct. 2428. In considering the same statutory scheme at issue in Walton and Adamson, Ring squarely rejected Walton’s interpre*1105tation of Arizona law, holding that “Arizona’s enumerated aggravating factors operate as ‘the functional equivalent of an element of a greater offense.’ ” Id. at 609, 122 S.Ct. 2428 (quoting Apprendi 530 U.S. at 494 n. 19, 120 S.Ct. 2348).
In so doing, Ring restored, as a matter of substantive law, the pre-Walton structure of capital murder law in Arizona; and, in so doing, Ring confirmed what we stated in Adamson: Under substantive Arizona law, there is a distinct offense of capital murder, and the aggravating circumstances that must be proven to a jury in order to impose a death sentence are elements of that distinct capital offense. 865 F.2d at 1025-28; see also Sattazahn v. Pennsylvania, 537 U.S. 101, 123 S.Ct. 732, 739, 154 L.Ed.2d 588 (2003) (“Put simply, if the existence of any fact (other than a prior conviction) increases the maximum punishment that may be imposed on a defendant, that fact — no matter how the State labels it — constitutes an element, and must be found by a jury beyond a reasonable doubt.”) (opinion of Scalia, J.). That is, when Ring displaced Walton, the effect was to declare Arizona's understanding and treatment of the separate crime of capital murder, as Arizona defined it, unconstitutional. And when Ring overruled Walton, 2’epositioning Arizona’s aggravating factors as elements of the separate offense of capital murder and reshaping the structure of Arizona murder law, it necessarily altered both the substance of the offense of capital murder in Arizona and the substance of Arizona murder law more generally. Cf. Jones v. United States, 526 U.S. 227, 229, 119 S.Ct. 1215, 143 L.Ed.2d 311 (1999) (holding that the federal carjacking statute established three separate offenses rather than a single crime with a choice of three maximum penalties). In response to Ring, the Arizona Supreme Court vacated all death sentences in cases pending on direct appeal, see State v. Smith, 203 Ariz. 75, 50 P.3d 825, 831 (2002), and the Arizona legislature once more changed the substantive law pertaining to capital punishment — this time providing for jury sentencing in capital cases. See Act of Apr. 27, 2001, ch. 260, § 1, 2001 Ariz. Sess. Laws 1334, 1334.
Ring’s understanding of capital murder as an offense both greater than and distinct from other murder crimes is neither unusual among the various States nor unrecognized by the Supreme Court. See, e.g., Atkins v. Virginia, 536 U.S. 304, 307 n. 1, 122 S.Ct. 2242, 153 L.Ed.2d 335 (2002) (noting that the two defendants were both “indicted for capital murder” but “[t]he prosecution ultimately permitted [one] to plead guilty to first-degree murder in exchange for [ ] testimony against” the other); Beck v. Alabama, 447 U.S. 625, 628, 100 S.Ct. 2382, 65 L.Ed.2d 392 (1980) (noting that Alabama law treats “[flelony murder [as][ ] a lesser included offense of the capital crime of robbery-intentional killing”).
In assessing the operation of Apprendi, in fact, Justice Scalia recently explained that “the underlying offense of ‘murder’ is a distinct, lesser included offense of ‘murder plus one or more aggravating circumstances.’ ” Sattazahn, 123 S.Ct. at 739. Noting that there was “no principled reason to distinguish ... what constitutes an offense for purposes of the Sixth Amendment’s jury-trial guarantee and what constitutes an ‘offence’ for purposes of the Fifth Amendment’s Double Jeopardy Clause,” Justice Scalia concluded “that ‘murder plus one or more aggravating circumstances’ is a separate offense from ‘murder’ simpliciter.” Id. at 739-40 (also citing Ring for the proposition that “ ‘first-degree murder’ ... is properly understood to be a lesser included offense of ‘first-degree murder plus aggravating circum-stanee(s)’ ”); see also Apprendi 530 U.S. at 501, 120 S.Ct. 2348 (Thomas, J., concur*1106ring) (“[I]f the legislature defines some core crime and then provides for increasing the punishment of that crime upon a finding of some aggravating fact[,] ... the core crime and the aggravating fact together constitute an aggravated crime.... The aggravating fact is an element of the aggravated crime.”).
Ring compelled Arizona to reorder its substantive murder law in order to recognize this two-offense structure. With regard to Arizona murder law, then, Ring did more than announce a procedural rule vis-a-vis whether a judge or a jury is to decide if elements of a particular offense have been proven satisfactorily. Ring reintroduced “capital murder” as a separate substantive offense under Arizona law, redefining, in the process, what the substantive elements of this “separate offense” of capital murder are. See Apprendi, 530 U.S. at 541, 120 S.Ct. 2348 (O’Connor, J„ dissenting) (observing that the Arizona first-degree murder statute “authorizes a maximum penalty of death only in a formal sense” and only to the extent it explicitly cross-references the separate Arizona statutory provision requiring the finding of an aggravating circumstance before imposition of the death penalty). In this sense, Ring had an inescapably substantive impact in Arizona for Teague purposes.10
To be sure, states must ensure that their capital sentencing schemes comply with the minimal procedural requirements set forth in Ring. Still, in the context of Arizona capital murder law, Ring’s rule is not limited to procedure. Ring did, as to Arizona, announce a substantive rule: It “decide[d] the meaning of a criminal statute,” see Bousley, 523 U.S. at 620, 118 S.Ct. 1604, and it did so in a manner that both redefined the separate substantive offense of “capital murder” in Arizona and reinserted the distinction between murder and capital murder into Arizona’s substantive criminal law structure. Under the Supreme Court’s articulation of “substantive” decisions in Bousley, then, Ring announced a “substantive” rule, Bousley, 523 U.S. at 620, 118 S.Ct. 1604, for it “altered the meaning of [Arizona’s] substantive criminal law.” Santana-Madera, 260 F.3d at 139; cf. Cannon, 297 F.3d at 994 (holding Ring’s rule to be procedural in a different capital murder context). When a decision affects the substantive elements of an offense, or how an offense is defined, it is necessarily a decision of substantive law. Dashney, 52 F.3d at 299. And because Ring is a “substantive” decision with regard to the meaning, structure, and ambit of the relevant provisions of Arizona’s criminal law, Teague does not bar retroactive application of Ring to cases decided under those Arizona provisions, regardless of whether those cases are considered on direct or collateral review.
The Arizona Supreme Court considered this question in State v. Towery, 204 Ariz. 386, 64 P.3d 828 (2003), and concluded that Ring was not a substantive decision. Id. at 833. More recently, the Arizona Supreme Court considered related issues in State v. Ring, 204 Ariz. 534, 65 P.3d 915 (2003) (“Ring II ”). In each case, the Arizona Supreme Court’s conclusion was founded on an interpretation of federal law, namely a construction of Teague and Allen v. Hardy, 478 U.S. 255, 106 S.Ct. 2878, 92 L.Ed.2d 199 (1986) (per curiam), in Towery and an analysis of the Ex Post Facto Clause in Ring II. Because the decisions in Towery and Ring II rest on federal law, and not state law, they do not bind us. Moore v. Sims, 442 U.S. 415, 428, 99 *1107S.Ct. 2871, 60 L.Ed.2d 994 (1979) (reiterating the familiar maxim that state courts possess final interpretive “authority” only regarding “laws of the state”) (citing R.R. Comm’n v. Pullman Co., 312 U.S. 496, 498, 61 S.Ct. 643, 85 L.Ed. 971 (1941), and Gilchrist v. Interborough Co., 279 U.S. 159, 49 S.Ct. 282, 73 L.Ed. 652 (1929)); Kern-Limerick, Inc. v. Scurlock, 347 U.S. 110, 121, 74 S.Ct. 403, 98 L.Ed. 546 (1954) (noting that a federal court “decide[s] for itself facts or constructions upon which federal constitutional issues rest”); Crew Levick Co. v. Pennsylvania, 245 U.S. 292, 38 S.Ct. 126, 62 L.Ed. 295 (1917) (noting that federal courts determine federal questions independently); see also Haynes v. Washington, 373 U.S. 503, 515-16, 83 S.Ct. 1336, 10 L.Ed.2d 513 (1963); Congoleum Corp. v. DLW Aktiengesellschaft, 729 F.2d 1240, 1242 (9th Cir.1984); Calkins v. Graham, 667 F.2d 1292, 1295 n. 1 (9th Cir.1982). This is particularly true in the Eighth Amendment context. Wainwright v. Goode, 464 U.S. 78, 84, 104 S.Ct. 378, 78 L.Ed.2d 187 (1983). Nonetheless, the two opinions are worthy of analysis.
In Towery, the Arizona Supreme Court correctly concluded that Ring’s rule is partially procedural under Teague. For the reasons previously stated, however, we respectfully disagree with Towery’s conclusion that Ring’s rule is entirely procedural. Ring’s invalidation of Arizona’s capital murder statute under the United States Constitution did more than alter “who decides.” Towery, 64 P.3d at 833. It restructured Arizona law and it redefined, as a substantive matter, how that law operates. It is, thus, incorrect to conclude that the repositioning of aggravating factors as elements of a separate offense did not constitute a “substantive” rule. Such a construction ignores that Ring’s restructuring of the elements of the separate offense of capital murder is, at the very least, a determination of the “meaning of a criminal statute,” which is precisely the kind of decision that Towery itself recognizes as “substantive.” Id. at 832. Of equal importance is the fact that Ring’s revival • of the pre-Walton two-offense structure of Arizona murder law does “address the criminal significance of certain facts,” another kind of decision that Tow-ery recognizes as “substantive.” Id.
The Arizona Supreme Court’s analogy to Apprendi in Towery is flawed as well. As noted above, in Apprendi, “[t]he substantive basis for New Jersey’s enhancement [was ] not at issue.” 530 U.S. at 475, 120 S.Ct. 2348. In Ring, conversely, the substantive basis of Arizona’s capital murder regime was at issue, so much so that Ring restored as a matter of substantive law the pre-Walton capital murder paradigm in Arizona. As we held in Adamson, this regime had defined capital murder as a substantive offense separate from non-capital murder. 865 F.2d at 1026. This distinction was required to satisfy the Eighth Amendment’s requirement that a capital sentencing scheme “genuinely narrow the class of persons eligible for the death penalty.” Zant, 462 U.S. at 877, 103 S.Ct. 2733. Ring further required the Arizona legislature to amend the Arizona murder statute to conform to the requirements of the United States Constitution.
The Arizona Supreme Court’s Ex Post Facto analysis in Ring II likewise does not alter our analysis. In Ring II, the Arizona Supreme Court concluded that applying Arizona’s new sentencing statutes to previously convicted defendants did not violate the federal or state prohibitions against Ex Post Facto application of laws. Ring II, 65 P.3d at 928. To reach this conclusion, Ring II relied on three decisions: the Supreme Court’s decisions in Dobbert v. Florida, 432 U.S. 282, 97 S.Ct. 2290, 53 L.Ed.2d 344 (1977), and Collins v. Youngblood, 497 U.S. 37, 110 S.Ct. 2715, 111 L.Ed.2d 30 (1990), and the Arizona Su*1108preme Court’s decision in State v. Correll, 148 Ariz. 468, 715 P.2d 721 (1986), rev’d in part on other grounds by Correll v. Stewart, 137 F.3d 1404 (9th Cir.1998). Ring II, 65 P.3d at 926-28. Like Dobbert, Ring II concluded, “the statutory change between the two sentencing methods was ‘clearly procedural’ ” in that this legislative enactment merely addressed the “who decides” question. Id. at 927 (quoting Dobbert, 432 U.S. at 293-94, 97 S.Ct. 2290).
In contrast to the post -Ring legislative changes at issue in Ring II, the Supreme Court’s Ring decision itself was not merely procedural. Ring declared a portion of Arizona’s prior law unconstitutional, demanded a redefinition of the meaning of that criminal statute, and prompted the legislative response at issue in Ring II, not by announcing a purely procedural rule, but by announcing, as a matter of substantive law, that Arizona’s understanding and treatment of the separate crime of capital murder was unconstitutional. This is exactly the kind of decision that is “substantive” under Bousley, 523 U.S. at 620, 118 S.Ct. 1604. Indeed, the very case law on which Ring II relies understands precisely this kind of rule to be of a “substantive nature.” See Collins, 497 U.S. at 51, 110 S.Ct. 2715 (noting that a rule is substantive in the Ex Post Facto context where it implicates “the definition of crimes, defenses, or punishments”); Correll, 715 P.2d at 735 (holding that changes relating to aggravating circumstances constituted substantive changes to the offense of capital murder).
Teague requires a different analytical lens from the one used by the Arizona Supreme Court in Ring II. We do not necessarily assess whether the action of the Arizona legislature, in response to Ring, effected a “substantive” change to Arizona law; rather, we examine whether the rule announced by the Supreme Court in Ring was a “substantive” one for Teag-ue purposes. Analyzed under Teague, the rule announced by the Supreme Court in Ring, with its restructuring of Arizona murder law and its redefinition of the separate crime of capital murder, is necessarily a “substantive” rule. See Bousley, 523 U.S. at 620, 118 S.Ct. 1604. Thus, Teague does not bar its application in this case.
V
In addition to Ring’s substantive effect on Arizona law, a full Teague analysis of the unique procedural aspects of Ring provides an independent basis upon which to apply Ring retroactively to cases on collateral review.
A
In undertaking a Teague procedural analysis, we first must ascertain the date that Summerlin’s conviction became final. Caspari, 510 U.S. at 390, 114 S.Ct. 948. Here, the Arizona Supreme Court denied rehearing of its opinion affirming his conviction and death sentence in 1984, see State v. Summerlin, 675 P.2d at 686, and Summerlin did not file a petition for a writ of certiorari with the Supreme Court. The relevant date thus is 1984. See Lambrix, 520 U.S. at 527, 117 S.Ct. 1517 (noting that the defendant’s conviction became final when his time for filing a petition for a writ of certiorari expired).
Next, we “survey the legal landscape” as it existed in 1984 to determine whether the result in Ring was dictated by then existing precedent. Graham, 506 U.S. at 468, 113 S.Ct. 892. Such examination is not limited to Supreme Court decisions. Williams v. Taylor, 529 U.S. 362, 381, 120 S.Ct. 1495, 146 L.Ed.2d 389 (2000). Sum-merlin’s conviction became final before the Supreme Court had decided Ring, which overturned Walton, which in turn abrogated our decision in Adamson.
*1109Summerlin contends that Ring does not announce a new rule because Adamson had found that Arizona’s sentencing scheme denied defendants the “right to a jury decision on the elements of the crime in violation of the Sixth and Fourteenth Amendments.” 865 F.2d at 1023. Regardless of the merits of such argument for convictions made final in 1988, the operative time period is 1984. At that time, Summerlin in fact raised this exact challenge to the Arizona Supreme Court, which soundly rejected it. State v. Summerlin, 675 P.2d at 695. The state supreme court reasoned that Proffitt v. Florida, 428 U.S. 242, 96 S.Ct. 2960, 49 L.Ed.2d 913 (1976), foreclosed such a challenge for two reasons. First, Proffitt found that jury sentencing in the capital context never has been held .to be “constitutionally required,” and second, the United States Supreme Court speculated that judicial sentencing “should lead, if anything, to even greater consistency” in capital punishment. State v. Summerlin, 675 P.2d at 695 (quoting Proffitt, 428 U.S. at 252, 96 S.Ct. 2960) (internal quotation marks omitted).
“[T]he Teague doctrine Validates reasonable, good faith interpretations of existing precedents made by state courts even though they are shown to be contrary to later decisions.’” O’Dell, 521 U.S. at 156, 117 S.Ct. 1969 (quoting Butler v. McKellar, 494 U.S. 407, 414, 110 S.Ct. 1212, 108 L.Ed.2d 347 (1990)). Further, “there can be no dispute that a decision announces a new rule if it expressly overrules a prior decision,” Graham, 506 U.S. at 467, 113 S.Ct. 892, which Ring indisputably did. Ring, 536 U.S. at 608-609, 122 S.Ct. 2428.
We therefore cannot say that a state court in 1984 “would have acted objectively unreasonably by not extending the relief later sought in federal court.” O’Dell, 521 U.S. at 156, 117 S.Ct. 1969. Summerlin’s argument fails because there is no doubt that Ring announced a new rule as that term is construed for Teague purposes. We must then proceed to the third stage of analysis, namely whether either of Teag-ue’s two exceptions apply.
B
The first Teague exception examines whether certain primary conduct has been decriminalized or whether certain classes of individuals are immunized from specified forms of punishment by the newly announced rule. Saffle v. Parks, 494 U.S. 484, 494, 110 S.Ct. 1257, 108 L.Ed.2d 415 (1990) (permitting retroactive application of a new rule “if the rule places a class of private conduct beyond the power of the State to proscribe or addresses a substantive categorical guarantee accorded by the Constitution”) (internal citations, quotation marks, and modifications omitted). Because Ring did not “decriminalize a class of conduct nor prohibit the imposition of capital punishment on a particular class of persons,” Graham, 506 U.S. at 477, 113 S.Ct. 892 (quoting Saffle, 494 U.S. at 495, 110 S.Ct. 1257), the first exception is inapplicable to the instant ruling.
C
To fall within the second Teague exception, a new rule must: (1) seriously enhance the accuracy of the proceeding and (2) alter our understanding of bedrock procedural elements essential to the fairness of the proceeding. Sawyer v. Smith, 497 U.S. 227, 242, 110 S.Ct. 2822, 111 L.Ed.2d 193 (1990).
1
In considering Ring’s effect on the accuracy of the proceeding, it is important to note that this is a capital case. “Where the State imposes the death penalty for a particular crime, ... the Eighth Amendment imposes special limitations on that *1110process.” Payne v. Tennessee, 501 U.S. 808, 824, 111 S.Ct. 2597, 115 L.Ed.2d 720 (1991). Under Teague, the focus on the “accuracy of the ... proceeding” generally is understood to mean “aceurac[y][in] determining] ... innocence or guilt.” Graham, 506 U.S. at 478, 118 S.Ct. 892 (internal quotation marks omitted). However, as the United States Supreme Court has stated, penalty-phase proceedings also “ ‘have the hallmarks of the trial on guilt or innocence.’ ” Sattazahn, 123 S.Ct. at 737 (quoting Bullington v. Missouri, 451 U.S. 430, 439, 101 S.Ct. 1852, 68 L.Ed.2d 270 (1981)). For this reason, a verdict rejecting the imposition of the death penalty prohibits a subsequent capital prosecution. See id.; Bullington, 451 U.S. at 439, 101 S.Ct. 1852. This is consistent with the substantive consideration of capital murder as a crime distinct from ordinary murder. Sattazahn, 123 S.Ct. at 739. Accordingly, in the capital context, Teague’s reference to “accuracy in the proceedings” contemplates the ultimate verdict in both the conviction and penalty phases.
The Supreme Court has long recognized that, in the capital context, “the Eighth Amendment requires a greater degree of accuracy and factfinding than would be true in a noncapital case.” Gilmore v. Taylor, 508 U.S. 333, 342, 113 S.Ct. 2112, 124 L.Ed.2d 306 (1993).11 Indeed, as Justice Kennedy has observed, “[a]ll of our Eighth Amendment jurisprudence concerning capital sentencing is directed toward the enhancement of reliability and accuracy in some sense.” Sawyer, 497 U.S. at 243, 110 S.Ct. 2822. Reformation of capital sentencing procedures has been presumed to meet the first requirement that the new rule substantially enhance the accuracy of the legal proceeding at issue. See id. (but emphasizing that the second “bedrock procedural element” requirement also must be met). Thus, on its face, the procedure at issue in Ring is sufficient to meet the first component of Teague’s second exception. Moreover, upon close examination, the actual impact of procedural change dictated by Ring provides further support for finding that the Ring rule enhances the accuracy of the determination of capital murder in Arizona.
The Supreme Court recently observed that, in light of the past thirty years of actual experience, “the superiority of judicial factfinding in capital cases is far from evident.” Ring, 536 U.S. at 607, 122 S.Ct. 2428. An examination of the procedure at issue makes apparent several reasons why fact-finding by a jury, rather than by a judge, is more likely to heighten the accuracy of capital sentencing proceedings in Arizona.
First, Arizona penalty-phase presentations to judges bear much greater resemblance to traditional non-capital sentencing hearings than to proceedings required to “ ‘have the hallmarks of the trial on guilt or innocence.’ ” Sattazahn, 123 S.Ct. at 737 (quoting Bullington, 451 U.S. at 439, 101 S.Ct. 1852). Penalty phases in jury trials are characterized by the orderly presentation of evidence and argument. In contrast, penalty-phase presentations to Arizona judges are capable of being extremely truncated affairs with heavy reliance on presentence reports and sentencing memo-randa, and with formal court proceedings frequently limited to a brief argument by counsel. Whether this has been the product of the participants treating penalty-phase trials as mere sentencing hearings, or whether this is the natural product of the shorthand communication typical of non-capital sentencing proceedings is unknown. However, the results are clear. A *1111quick survey of recent Ninth Circuit cases from Arizona illustrates the point. See, e.g., Beaty v. Stewart, 303 F.3d 975, 988 (9th Cir.2002) (no mitigating evidence presented), petition for cert. filed, No. 02-1611, 71 U.S.L.W. 3530 (Jan. 23, 2003); Lambright v. Stewart, 241 F.3d 1201, 1202-03 (9th Cir.2001) (no argument presented, and mitigating evidence consumed three transcript pages), cert. denied, 534 U.S. 1118, 122 S.Ct. 930, 151 L.Ed.2d 892 (2002); Smith v. Stewart, 189 F.3d 1004, 1010 (9th Cir.1999) (attorney asked court for advice on what legal and evidentiary considerations could be relevant in establishing mitigation; only brief argument on day of sentencing); Correll v. Stewart, 137 F.3d 1404, 1410 (9th Cir.1998) (no defense witnesses presented; only brief argument); Clabourne, 64 F.3d at 1384 (no witnesses presented; only one mitigating circumstance argued).
In addition, because penalty-phase presentations to judges tend to resemble non-capital sentencing proceedings, the sentencing judge receives an inordinate amount of inadmissible evidence, which he or she is expected to ignore. Indeed, the focus of penalty-phase proceedings before judges has been the presentence report prepared by the probation officer, rather than evidence formally presented and tested at trial.12
Although presentence reports are an extremely useful sentencing tool, by their nature the information they contain is “generally hearsay, even remote hearsay at the second and third remove.” United States v. Frushon, 10 F.3d 663, 666 (9th Cir.1993) (quoting United States v. Fine, 975 F.2d 596, 603 (9th Cir.1992)). As a result, presentence reports are generally inadmissible at trial to prove any of the hearsay reports they contain. See United States v. Matta-Ballesteros, 71 F.3d 754, 767 (9th Cir.1995), as amended by 98 F.3d 1100 (9th Cir.1996). Because they are not subject to evidentiary standards, presentence reports may also contain factual errors.13 In Arizona capital cases, presentence reports have also frequently contained inadmissible victim impact statements, including sentencing recommendations from the victim’s family.14
*1112In addition, capital sentencing judges in Arizona have often received letters directly from the victim’s family and friends expressing their opinions about sentencing, prompting the Arizona Supreme Court to explain that: “[w]e have no way of preventing members of the community from writing judges.” Mann, 934 P.2d at 792.15 The net result, prior to Ring, was a capital sentencing system that allowed a large amount of inadmissible evidence to be submitted to capital sentencing judges that could not be considered by a penalty-phase jury.
The penalty-phase presentation in the instant case was typical of pre-Ring Arizona capital sentencing cases and illustrates both problems. The actual penalty-phase proceeding was exceedingly truncated and bore more resemblance to traditional non-capital judge sentencing than a trial. Before hearing any presentation by the parties, Judge Marquardt received a presentence report prepared by a probation officer who did not testify during the penalty phase. It contained numerous sentencing recommendations from the victim’s family and friends, police officers, and others. Attached to the presentence report were a large number of letters from members of the community expressing their opinions, including a petition with over 500 signatories. The presentence report also contained the probation officer’s opinion as to the heinous nature of the crime and expressed her opinion as to the sentence that the judge should impose.
In contrast, the formal presentations by the parties were extremely abbreviated. The admissible evidence actually presented to the judge paled in comparison with the inadmissible material contained in the pre-sentence report. The State submitted a four-page sentencing memorandum urging imposition of the death penalty. Summer-lin’s attorney submitted nothing. Neither side made an opening statement. The State’s formal evidentiary presentation on aggravation consisted of less than one transcript page. Summerlin’s counsel declined to introduce testimony as to mitigation; rather, he asked the judge only to review the material contained in the pre-sentence report. Thus, the cumulative presentation of each side’s case in chief resulted in less than one page of trial transcript. The only live testimony was a brief presentation by the State to rebut medical statements contained in the pre-sentence report. When viewed by volume, well over ninety percent of the material received by the sentencing judge in this case could not have been presented to a capital sentencing jury. Such a proeeed-*1113ing is not one that bears “the hallmarks of a trial on guilt or innocence.”
The point of this discussion is not to examine whether trial errors occurred in any particular case, including this one, but to analyze whether requiring a jury to make the relevant findings would reduce the risk of an erroneous decision. A review of the cases demonstrates that judge capital sentencing proceedings have been contaminated by a large volume of inadmissible evidence and marked by truncated presentations by the parties. We have presumed that the sentencing judge could sort out the truly relevant, admissible evidence from this morass. The relevant question is not whether judges have been able to do so, but whether subjecting penalty-phase evidence to the crucible of a formal trial by jury would reduce the risk of error.
There is little doubt that it would. As Harry Kalven, Jr. and Hans Zeisel described it in their seminal study on the jury system:
In addition to his wide experience with the likelihood that the defendant before him is guilty, the judge is exposed to prejudicial information which the law, in its regard for the right of the defendant, aims to screen out of the evaluation of his guilt or innocence. The law’s ideal in this situation may be something of a libertarian luxury. Our only point is that the law cannot easily achieve it without a jury.
Harry Kalven, Jr. & Hans Zeisel, The American Jury 127 (Little, Brown 1966).
If there is any place in which adherence to evidentiary rules, constitutional restraints, and the defendant’s confrontation rights is paramount, it must be when the defendant is exposed to the penalty of death. Subjecting penalty-phase presentations to the rigors and restrictions of a jury trial necessarily will improve the quality of presentation and diminish the risk of an erroneous verdict.
A second primary accuracy-enhancing role of a jury in capital cases is to make the important moral decision's inherent in rendering a capital verdict. The Supreme Court “has emphasized that a sentence of death must reflect an ethical judgment about the ‘moral guilt’ of the defendant.” Schiro v. Indiana, 475 U.S. 1036, 1038, 106 S.Ct. 1247, 89 L.Ed.2d 355 (1986) (Marshall, J., dissenting from the denial of cer-tiorari) (citing Enmund v. Florida, 458 U.S. 782, 800-01, 102 S.Ct. 3368, 73 L.Ed.2d 1140 (1982)). One of the critical functions of a jury in a capital case is to “maintain a link between contemporary community values and the penal system.” Witherspoon v. Illinois, 391 U.S. 510, 520 n. 15, 88 S.Ct. 1770, 20 L.Ed.2d 776 (1968). Thus, “in a capital sentencing proceeding, the Government has ‘a strong interest in having the jury express the conscience of the community on the ultimate question of life or death.’ ” Jones v. United States, 527 U.S. 373, 382, 119 S.Ct. 2090, 144 L.Ed.2d 370 (1999) (quoting Lowenfield v. Phelps, 484 U.S. 231, 238, 108 S.Ct. 546, 98 L.Ed.2d 568 (1988)).
“ ‘[T]he men and women of the jury may be regarded as a microcosm of the community.’ ” Harris v. Alabama, 513 U.S. 504, 517, 115 S.Ct. 1031, 130 L.Ed.2d 1004 (1995) (Stevens, J., dissenting) (quoting Royal Commission on Capital Punishment 191,9-1953, Report 200 (1953)). There could “therefore be no more appropriate body to decide whether the fellow-citizen whom they have found guilty of murder should suffer the penalty of death prescribed by the law or should receive a lesser punishment.” Id. Thus, as Justice Breyer noted in his concurring opinion in Ring, entrusting a jury with the authority to impose a capital verdict is an important procedural safeguard, because the jury *1114members “are more attuned to the community’s moral sensibility,” “reflect more accurately the composition and experiences of the community as a whole,” and act to “express the conscience of the community on the ultimate question of life or death.” 536 U.S. at 615-16, 122 S.Ct. 2428 (Breyer, J., concurring in the judgment) (internal citations and quotation marks omitted).16
This principle is not only true as a general matter in capital murder verdicts, but it has specific application to determination of some of the aggravating factors contained in Arizona’s death penalty scheme. For example, one of the two aggravating circumstances found by Judge Marquardt in this case was that the murder was committed “in an especially heinous, cruel or depraved manner.” Ariz.Rev.Stat. § 13 703(F)(6). The Arizona Supreme Court has noted that these are “admittedly broad subjective terms.” State v. Vickers, 159 Ariz. 532, 768 P.2d 1177, 1188 n. 2 (1989). The assessment of whether a crime is “heinous” depends on the “mental state and attitude of the perpetrator as reflected in his words and actions.” State v. Gretzler, 135 Ariz. 42, 659 P.2d 1, 10 (1983) (citations omitted). As we noted in discussing this aggravating factor in Adamson: “These assessments directly measure a defendant’s’ moral guilt and overall culpability-traditionally the jury’s domain of decision.” 865 F.2d at 1027.
These assessments may be influenced by the possible acclimation of the judge to the capital sentencing process. Most jurors in capital cases will never sit on another case in which the death penalty is sought. Judges, by contrast, confront death penalty cases on a regular and sometimes routine basis in Arizona. For instance, Judge Marquardt, who sentenced Summerlin to death, imposed capital punishment on James Fisher in a separate case on the same day. A reasonable inference from the habituation brought about by imposing capital punishment under near rote conditions is that a judge may be less likely to reflect the current conscience of the community and more likely to consider imposing a death penalty as just another criminal sentence. Indeed, when questioned about another capital case in which his judgment was being assailed because he purportedly slept through portions of the short penalty-phase hearing, Judge Mar-quardt answered that he was unable to recall the case, but “said he had no doubt that the death penalty was warranted.” Adam Liptak, Judge’s Drug Use at Issue in 2 Death Sentences, N.Y. TIMES, May 16, 2002, at Al. “These guys have sentenced themselves,” he is reported to have said. Id.
Of course, Judge Marquardt’s conduct is not at all representative of the Arizona judiciary — a point that must be underscored. However, the extremity of his actions highlights the potential risk of accuracy loss when a capital decision is reposed in a single decision-maker who may be habituated to the process, or who may not treat capital sentencing in accordance with the heightened requirements that the Eighth Amendment imposes. Obviously, *1115in Summerlin’s case, the concern is not merely theoretical.
In addition, unlike judges, juries do not stand for election in Arizona and therefore are less apt to be influenced by external considerations when making their decisions. As Justice Stevens has commented, “given the political pressures they face, judges are far more likely than juries to impose the death penalty.” Harris, 513 U.S. at 521, 115 S.Ct. 1031 (Stevens, J., dissenting). This postulate has empirical support: Judges who face election are far more likely to impose the death penalty than either juries or appointed judges. See Stephen B. Bright & Patrick J. Keenan, Judges and the Politics of Death: Deciding Between the Bill of Rights and the Next Election in Capital Cases, 75 B.U. L.Rev. 759, 793-94 (1995) (discussing studies documenting the existence of a statistically-signiflcant correlation between an increased override of juries’ recommendations against the death penalty by state judges and occurrence of judicial elections in Alabama, Florida, and Indiana); Fred B. Burnside, Comment, Dying to Get Elected: A Challenge to the Jury Override, 1999 Wis. L.Rev. 1017, 1039-44 (same); see also Sam Kamin, Harmless Error and the Rights/Remedies Split, 88 Va. L.Rev. 1, 62 (2002) (citing empirical examination of death penalty decisions issued by the California Supreme Court between 1976 and 1996). It also has anecdotal support.17
These reasons underscore Justice Breyer’s observation in Ring that “the danger of unwarranted imposition of the penalty cannot be avoided unless ‘the decision to impose the death penalty is made by a jury rather than by a single governmental official.’ ” 536 U.S. at 618, 122 S.Ct. 2428 (Breyer, J., concurring in the judgment) (quoting Spaziano v. Florida, 468 U.S. 447, 469, 104 S.Ct. 3154, 82 L.Ed.2d 340 (1984) (Stevens, J., concurring in part and dissenting in part)).
For all of these reasons, exemplified by the facts of this case, there is little doubt that the rule announced in Ring will significantly improve the accuracy of capital trials in Arizona. This conclusion is not— and should not be considered as — a negative assessment of the many excellent state trial judges in Arizona, many of whom have been national leaders in improving the jury system. See, e.g., B. Michael Dann & George Logan III, Jury Reform: The Arizona Experience, 79 Judicature 280 (1996). However, the structure of Arizona capital sentencing allows extra-judicial factors to enter into the ultimate judgment such as the consideration of inadmissible evidence, political pressure, truncated evidentiary presentation, and prior experience with other capital defendants that would be absent from a jury’s consideration of penalty-phase evidence.
If the allegations concerning Judge Mar-quardt are true, Summerlin’s fate was determined by a drug-impaired judge, habituated to treating penalty-phase trials the *1116same as non-capital sentencing, who relied upon inadmissible evidence in making the factual findings that sentenced Summerlin to death. Although no system is perfect, relying on a jury to administer capital punishment unquestionably reduces the risk of error by reposing trust in twelve individuals who must agree as to the presence of aggravating factors beyond a reasonable doubt, whose continued job security is not threatened by their decision, and whose consideration is based solely on admissible evidence subject to the rigors of cross-examination.
Taking into account the heightened attention that the Eighth Amendment obligates us to afford capital cases, the inevitable conclusion must be that a requirement of capital findings made by a jury will improve the accuracy of Arizona capital murder trials.
2
The second requirement of the Teague exception provides that the newly announced rule must be a “watershed rule” that alters our understanding of bedrock procedural elements essential to the fairness of the proceeding. Sawyer, 497 U.S. at 242, 110 S.Ct. 2822. Although Eighth Amendment concerns are implicated in Ring, the bedrock procedural element at issue is the provision of the Sixth Amendment right to a jury trial.
Ring not only changed the substantive criminal law of Arizona, but it fundamentally altered the procedural structure of capital sentencing applicable to all states. Ring established the bedrock principle that, under the Sixth Amendment, a jury verdict is required on the finding of aggravated circumstances necessary to the imposition of the death penalty. 536 U.S. at 609, 122 S.Ct. 2428. Ring requires the vacation of a capital judgment based on judge-made findings.
A structural error is a “defect affecting the framework within which the trial proceeds, rather than simply an error in the trial process itself.” Arizona v. Fulminante, 499 U.S. 279, 310, 111 S.Ct. 1246, 113 L.Ed.2d 302 (1991). If structural error is present, “a criminal trial cannot reliably serve its function as a vehicle for determination of guilt or innocence, and no criminal punishment may be regarded as fundamentally fair.” Rose v. Clark, 478 U.S. 570, 577-78, 106 S.Ct. 3101, 92 L.Ed.2d 460 (1986) (citation omitted).
Depriving a capital defendant of his constitutional right to have a jury decide whether he is eligible for the death penalty is an error that necessarily affects the framework within which the trial proceeds. Indeed, the trial has proceeded under a completely incorrect, and constitutionally deficient, framework. In short, allowing a constitutionally-disqualified factfinder to decide the case is a structural error, and Ring error is not susceptible to harmless-error analysis.
This conclusion is compelled by the analysis in Sullivan v. Louisiana, 508 U.S. 275, 113 S.Ct. 2078, 124 L.Ed.2d 182 (1993). In that case, the Supreme Court considered whether a deficient reasonable-doubt instruction18 was subject to harmless-error analysis. Id. at 277, 113 S.Ct. 2078. In resolving the question, Justice Scalia first noted that the Sixth Amendment right to a jury trial “includes, of course, as its most important element, the right to have the jury, rather than the judge, reach the requisite finding.” Id. at 277, 113 S.Ct. 2078 (citing Sparf v. United *1117States, 156 U.S. 51, 105-06, 15 S.Ct. 273, 39 L.Ed. 343 (1895)). To determine whether harmless error exists, he noted, the Supreme Court must examine “the basis on which ‘the jury actually rested its verdict.’” Id. at 279 (quoting Yates v. Evatt, 500 U.S. 391, 404, 111 S.Ct. 1884, 114 L.Ed.2d 432 (1991) (emphasis added in Sullivan)). Given that, Justice Scalia observed the “illogic of harmless-error review” under those circumstances because a review could occur only if a court “hypothesize[d] a guilty verdict that was never in fact rendered.” Id. at 279-80, 113 S.Ct. 2078. The Court thus concluded that the Cage error was structural because any constitutionally defective reasonable doubt jury instruction “vitiates all the jury’s findings” and “[a] reviewing court can only engage in pure speculation — its view of what a reasonable jury would have done.” Id. at 281, 111 S.Ct. 328 (emphasis in original). “And when it does that, ‘the wrong entity judge[s] the defendant guilty.’ ” Id. (quoting Rose, 478 U.S. at 578, 106 S.Ct. 3101) (alteration in original).
In our case, the wrong entity found Summerlin to be guilty of a capital crime. Here, as in Sullivan, there was no jury verdict within the meaning of the Sixth Amendment and no constitutionally cognizable finding to review. A complete deprivation of the right to a jury is an error that does not arise during a presentation to a jury. Rather, such an error indisputably affects “the framework within which the trial proceeds.” Rose, 478 U.S. at 570, 106 S.Ct. 3101. This type of error cannot be cured, or determined harmless, by examining other mitigating circumstances that may have been presented at trial because, as Justice Scalia observed:
The Sixth Amendment requires more than appellate speculation about a hypothetical jury’s action, or else directed verdicts for the State would be sustainable on appeal; it requires an actual jury finding of guilty.
Sullivan, 508 U.S. at 280, 113 S.Ct. 2078 (citing Bollenbach v. United States, 326 U.S. 607, 614, 66 S.Ct. 402, 90 L.Ed. 350 (1946)).19
Sullivan’s, logic thus applies with even greater force where, as here, there was no jury finding at all. As Justice Scalia observed in Sullivan, under the Sixth Amendment, a judge “may not direct a verdict for the State, no matter how overwhelming the evidence.” Id. at 277, 113 S.Ct. 2078. If a judge is constitutionally precluded from directing a verdict for the State, then, perforce, a judge sitting without a jury cannot constitutionally enter a judgment of conviction for capital murder. In this sense, Ring error indisputably affects the framework of the trial and must therefore constitute structural error.
The Supreme Court’s recent opinion in Nguyen v. United States, — U.S. -, 123 S.Ct. 2130, 156 L.Ed.2d 64 (2003), reaffirms that any decision of an improperly constituted judicial body must be vacated. In Nguyen, the Supreme Court assessed a series of judgments rendered by a federal appellate court panel on which a “non Article III judge” served. — U.S. at-, 123 S.Ct. at 2133-34. Vacating this group of judgments, the Supreme Court reasoned that an appellate *1118panel that included a non-Article III judge proved an “impermissible” and “unauthorized” decisional body, one that necessarily conflicted with a “strong policy concerning the proper administration of judicial business.” Id. at 2135-36 (citation and internal quotation marks omitted). Because the “validity” of the relevant judicial body was fundamentally flawed, and because this “plain defect” was incurable, Nguyen explained, the decisions reached by that body must be vacated. Id. at 2137-38; see also N. Pipeline Constr. Co. v. Marathon Pipe Line Co., 458 U.S. 50, 83-85, 102 S.Ct. 2858, 73 L.Ed.2d 598 (1982) (holding that delegation to adjunct bankruptcy judges of powers beyond those conferred to non-Article III judges rendered an entire administrative scheme unconstitutional).
The principle animating Sullivan, Nguyen, and Northern Pipeline provides that where an improperly constituted or situated tribunal reaches a decision, that decision is infected with a “plain defect” and must be vacated. Nguyen, — U.S. at ----, 123 S.Ct. at 2137-38. Whether before an improperly constituted federal appellate panel, a flawed jury panel, a biased judge, or a judge without fact-finding authority in particular contexts, even an otherwise error-free trial is subject to reversal because the error affects the very framework within which the trial proceeds. Id. Such structural error indisputably arises here.20
Application of the heightened scrutiny commanded by the Eighth Amendment in capital cases underscores the structural nature of this Sixth Amendment constitutional infirmity. The Sixth Circuit recently considered whether harmless-error analysis could apply in a capital case under similar circumstances in Esparza v. Mitchell, 310 F.3d 414 (6th Cir.2002), petition for cert. filed, Nos. 02-1369, 02-8849, 71 U.S.L.W. 3613 (Mar. 4, 2003). In that case, the Sixth Circuit held that a death sentence could not be imposed when the state had failed to charge the aggravating circumstance and the jury had not found the aggravating circumstance beyond a *1119reasonable doubt. Id. at 420. In reaching its decision, the Sixth Circuit specifically rejected the theory that a harmless-error analysis under a Neder theory could apply, noting that “[tjhere is no suggestion in the Chief Justice’s opinion in Neder that harmless error would protect a directed verdict for the State on a crucial finding under the Eighth Amendment in a capital case.” Id. at 421. The court further noted:
Harmless-error review in [capital] cases should apply only when the jury has actually performed its function under the Eighth Amendment. The jury in this case never made a judgment at all on the only possible aggravating circumstance — a constitutionally indispensable requirement without which the death penalty cannot be imposed. The State’s argument that the error here can be excused as harmless would lead to the conclusion that any, or all, elements required by a state’s capital sentencing system may be supplied by judges rather than the jury. Neither the Eighth Amendment nor Ohio’s own statutes adopted in order to comply with it permit such a gross deviation from the principle of jury sentencing according to expressly stated, clear statutory standards.
Id. at 422.
Given Ring’s declaration that a defendant is entitled under the Sixth Amendment to a jury verdict in the penalty phase of a capital case, the substitution of a non-jury verdict cannot be subject to harmless-error analysis. Ring error is one “ ‘affecting the framework within which the trial proceeds, rather than simply an error in the trial process itself.’ ” Neder, 527 U.S. at 8, 119 S.Ct. 1827 (quoting Fulminante, 499 U.S. at 810, 111 S.Ct. 1246).
3
That Ring error is structural is a critical consideration in determining whether the second Teague exception has been satisfied. It does not, however, end our inquiry. See Tyler v. Cain, 533 U.S. 656, 666-67, 121 S.Ct. 2478, 150 L.Ed.2d 632 (2001). The Supreme Court has explained on numerous occasions that a “truly watershed case” is one of a “small core of rules” that is “groundbreaking.” See, e.g., O’Dell, 521 U.S. at 167, 117 S.Ct. 1969; Caspari, 510 U.S. at 396, 114 S.Ct. 948; Graham, 506 U.S. at 478, 113 S.Ct. 892. The newly announced rule must enhance accuracy, improve fairness, and dictate “observance of those procedures that ... are implicit in the concept of ordered liberty.” Teague, 489 U.S. at 311, 109 S.Ct. 1060 (plurality) (internal quotation marks omitted).
As previously noted, the procedure at issue in Ring had been reviewed by the Supreme Court on prior occasions and had been found constitutionally sound. See Apprendi 530 U.S. at 496-97, 120 S.Ct. 2348; Walton, 497 U.S. at 649, 110 S.Ct. 3047. In a sharp reversal of course, however, the Supreme Court in Ring determined that when the Arizona court adhered to prior Supreme Court law, it violated the defendant’s fundamental right to have his guilt determined by a jury trial.
The Supreme Court’s decision in Ring affects the structure of every capital trial and has rendered unconstitutional every substantive statute in conflict with its dictates. It involves the structure of penalty-phase trials which, unlike non-capital sentencing proceedings, are subject to the constraints of the Double Jeopardy Clause. Bullington, 451 U.S. at 439, 101 S.Ct. 1852. Thus, Ring’s effect on capital murder cases is akin to that of Furman, which declared that death penalty statutes vesting complete discretion in the judge or in the jury, like Arizona’s, were unconstitutional. Furman, as already noted, was given full retroactive effect for the pur*1120poses of federal habeas review. See Moore, 408 U.S. at 800, 92 S.Ct. 2562.
Ring’s impact thus is far greater than the impact of the Mills/McKoy rule, which some of our sister circuits have determined to be a “watershed rule” under Teague. See Gall v. Parker, 231 F.3d 265, 323-24 (6th Cir.2000), cert. denied, 533 U.S. 941, 121 S.Ct. 2577, 150 L.Ed.2d 739 (2001); Williams v. Dixon, 961 F.2d 448, 455-57 (4th Cir.1992).21
In Williams, Chief Judge Ervin analyzed the Mills/McKoy rule in detail and concluded that: “Given the history of the Eighth Amendment jurisprudence and the constitutional requirement of individualized sentences, we believe that a rule striking down an arbitrary unanimity requirement has the same ‘primacy and centrality’ of Gideon [v. Wainwright, 372 U.S. 335, 83 S.Ct. 792, 9 L.Ed.2d 799 (1963)].” 961 F.2d at 456. The Fourth Circuit underscored the language in Mills that “it would certainly be the height of arbitrariness to allow or require the imposition of the death penalty under the circumstances [where one holdout juror prevents the consideration of mitigating evidence].” Id. (quoting Mills, 486 U.S. at 374, 108 S.Ct. 1860) (emphasis added and alterations inserted in Williams ).22
Ring does not merely announce a supplemental procedural safeguard. Rather, it establishes bedrock procedural requirements that affect the structure of every penalty-phase hearing in a capital case. Application of the rule in Ring, like the Mills/McKoy rule and other watershed rules, will increase the accuracy of capital murder trials significantly. The absence of the rule’s protection necessarily constitutes structural error and deprives a defendant of a fundamental right.
Moreover, Ring implicates the foundation of the capital murder trial itself by declaring that judges are constitutionally unqualified to decide whether a defendant is eligible for the death penalty. Thus, the rule in Ring is central to the conduct of every capital murder trial. Ring “effectively deelare[d] five States’ capital sentencing schemes unconstitutional,” Ring, 536 U.S. at 621, 122 S.Ct. 2428 (O’Connor, J., dissenting), and cast doubt on the viability of at least four other state capital murder statutes. Id.; see also Brief for Amici Curiae Alabama, Colorado, Delaware, Florida, Idaho, Indiana, Mississippi Montana, Nebraska, Nevada, New York District Attorney’s Ass’n, Pennsylvania, South Carolina, Utah, and Virginia to the Supreme Court in Ring at *4 & n. 2, Ring (No. 01^188), available at 2002 WL 481140. In short, Ring directly impacted the substance of approximately one-fourth of the 38 state capital murder statutes and established irreducible minimum structural requirements for all. It fundamentally altered our view of how the Sixth Amendment right to a jury trial affected the Eighth Amendment’s requirement that state statutes narrow the class of individuals eligible for the penalty of death. By deciding that judges are not *1121constitutionally permitted to decide whether defendants are eligible for the death penalty, the Supreme Court altered the fundamental bedrock principles applicable to capital murder trials. When viewed in both theoretical and practical terms, Ring redefined the structural safeguards implicit in our concept of ordered liberty.
The Teague doctrine was based on the notion that one of the “principal functions of habeas corpusfis] to assure that no man has been incarcerated under a procedure which creates an impermissibly large risk that the innocent will be convicted.” Bousley, 523 U.S. at 620, 118 S.Ct. 1604 (alteration in original; internal quotation marks omitted). The rule announced in Ring, like other watershed rules, is designed to reduce significantly the risk of an erroneous capital verdict. Thus, the rule announced in Ring defines structural safeguards implicit in our concept of ordered liberty that are necessary to protect the fundamental fairness of capital murder trials. Ring satisfies the criteria of Teag-ue and must be given retroactive effect on habeas review.
D
The Warden’s primary argument against applying Ring retroactively relies on United States v. Sanchez-Cervantes, in which we held that Apprendi may not be applied retroactively on habeas review. United States v. Sanchez-Cervantes, 282 F.3d 664, 670 (9th Cir.2002). However, our analysis in Sanchez-Cervantes does not conflict with our conclusion that Ring must be applied retroactively. First, as we have noted, the decision in Apprendi clearly was not one of substantive criminal law. Unlike the result in Ring, Apprendi did not cause the relevant statute to be declared unconstitutional. Second, Apprendi errors are not structural and therefore are subject to harmless-error analysis. Id. at 669-70 (citing Buckland, 277 F.3d at 1184) (internal quotation marks omitted). Third, as we reasoned in Sanchez-Cervantes, Apprendi was neither a rule that greatly enhanced the accuracy of sentencing proceedings nor a “sweeping rule” in light of the finding that it “would apply only in a limited number of cases.” Id. at 669. Accordingly, it could not qualify as a “watershed” rule. Fourth, capital cases are structurally much different from non-capital criminal trials. As already noted, non-capital sentencing has historically been within the province of the judge. In contrast, the penalty-phase proceeding of a capital murder case “resembles a trial,” Walton, 497 U.S. at 704, 110 S.Ct. 3047, and, unlike the non-capital sentencing proceedings, is subject to the constraints of the Double Jeopardy Clause. See Bullington, 451 U.S. at 439, 101 S.Ct. 1852. Finally, the Eighth Amendment constraints applicable to capital trials demand a heightened analysis inapplicable to the usual Apprendi situation. Thus, neither Sanchez-Cervantes nor the retroactive application of Apprendi governs our analysis.
VI
In summary, we affirm the district court’s judgment denying Summerlin’s petition for habeas corpus for relief from his conviction for first-degree murder. We hold, both on substantive and procedural grounds, that the Supreme Court’s decision in Ring has retroactive application to cases on federal habeas review. Thus, we reverse the judgment of the district court insofar as it relates to the imposition of the penalty of death.
Given our resolution of the penalty-phase issues based on the retroactive application of Ring, we need not, and do not, reach the merits of any of the other penalty-phase errors raised on appeal. We also need not reach the issue of whether cumulative errors require reversal.
*1122AFFIRMED IN PART; REVERSED IN PART; REMANDED.
. During his later disbarment proceedings, Judge Marquardt admitted that he was addicted to the drug, although he did not specify how long he had been addicted. In support of his allegations against Judge Marquardt, Summerlin submitted to the district court an official report from the Phoenix Police Department dated June 3, 1991. The report details a purchase of marijuana by Judge Marquardt from Barbara Moffett in May of 1991, which was intercepted from the United States mail by the police. When the delivery went awry, the report states that Judge Mar-quardt called Barbara Moffett to see if she had spoken to the authorities about the purchase, and when she told him she had not. Judge Marquardt told her that everything would "work out okay” because his daughter Tiffany's boyfriend Butch "was going to take the rap for the marijuana.” The official police report also states that Barbara Moffett told Phoenix police in 1991 on the basis of first-hand knowledge that Judge Marquardt "was a frequent user of marijuana, had been when she met him [sixteen years earlier], and has continued to be so since.” The envelope *1090in which Judge Marquardt sent a cashier’s check to Ms. Moffett for the marijuana carried the printed official heading, "Philip Mar-quardt, Superior Court Judge, Phoenix, Arizona.”
In a separate episode, Judge Marquardt was convicted in 1988 in Texas of misdemeanor possession of marijuana which was found on his person at the port of entry in Houston. In re Marquardt, 161 Ariz. 206, 778 P.2d 241, 242-43 (1989). His apparently false explanation documented in that case was that a stranger gave him the marijuana wrapped in a small piece of plastic that he stuck in his watch pocket. Id. at 242, 778 P.2d 241. For that offense, the Supreme Court of Arizona suspended him from his judicial position without pay for one year from September 2, 1988, through September 2, 1989, a sanction considered by that court to be more severe than a mere censure or reprimand. Id. at 250, 778 P.2d 241. Despite this serious and career-threatening incident, Judge Marquardt continued to use marijuana. Eventually, Judge Marquardt stepped down from the bench and was ordered disbarred in Arizona and by the United States Supreme Court after the 1991 incident came to light. See In re Disbarment of Marquardt, 503 U.S. 902, 112 S.Ct. 1256, 117 L.Ed.2d 486 (1992); In re Marquardt, 169 Ariz. 500, 821 P.2d 161 (1991).
. There are instances during pre-trial hearings and at trial when Judge Marquardt exhibited confusion over facts that had just been presented to him. He also made some quite perplexing, if not unintelligible, statements at various times during the trial. Obviously, because there was no discovery or evidentiary hearing permitted, the question of whether these episodes were related to impairment was, and is, unresolved.
. Summerlin's claim that his counsel had a conflict of interest does not implicate the guilt phase because his argument is that he was deprived of a favorable sentence. Summer-lin’s counsel clarified at oral argument that the claim concerning the trial judge's impairment was limited to the judge’s penalty phase deliberations.
. Some of our sister circuits have addressed the question of whether Ring is retroactive as to cases governed by the AEDPA retroactivity rule, 28 U.S.C. § 2244(b)(2)(A). See Cannon v. Mullin, 297 F.3d 989, 994 (10th Cir.2002) (holding that Ring was not retroactive under AEDPA, but not reaching the question of ret-roactivity under Teague); see also Whitfield v. Bowersox, 324 F.3d 1009, 1012 n. 1 (8th Cir. 2003) (declining to address whether death sentence contravened Ring because Supreme Court had not expressly made Ring retroactive under AEDPA); Moore v. Kinney, 320 F.3d 767, 771 n. 3 (8th Cir.2003) (en banc) (same), petition for cert. filed, No. 02-10093 (Apr. 14, 2003). The question of whether a rule has retroactive application under AEDPA is a different inquiry from the question of whether Teague precludes retroactive application of a rule. Most importantly, AEDPA precludes retroactive application of a new rule of constitutional law unless "made retroactive to cases on collateral review by the Supreme Court.” 28 U.S.C. § 2244(b)(2)(A). Because the Supreme- Court has not addressed whether Ring should be applied retroactively, the analysis of the retroactively of Ring under AEPDA and Teague is necessarily distinct. As the Eighth Circuit noted in Moore, in analyzing whether Ring should be applied retroactively in a case governed by AEDPA, "[a]bsent an express pronouncement on retroactivity from the Supreme Court, the rule from Ring is not retroactive.” 320 F.3d at 771 n. 3. To date, only the Eleventh Circuit has addressed the retroactivity of Ring under a Teague analysis. See Turner v. Crosby, 339 F.3d 1247, 2003 WL 21739734 (11th Cir. 2003). In Turner, the Eleventh Circuit held that the petitioner’s claim was procedurally barred, but held in the alternative that Ring was a procedural rule that should not be applied retroactively. Id. at *30-*37, 339 F.3d at 1279-86.
. Most notably, Jeffery Walton, who raised the identical issue a decade ago, see Walton v. Arizona, 497 U.S. 639, 110 S.Ct. 3047, 111 L.Ed.2d 511 (1990), but whose claim was rejected by the Supreme Court.
. See, e.g., United States v. Barajas-Diaz, 313 F.3d 1242, 1245 (10th Cir.2002); Ross v. United States, 289 F.3d 677, 681 (11th Cir. 2002), cert. denied, 537 U.S. 1113, 123 S.Ct. 944, 154 L.Ed.2d 787 (2003); Santana-Madera, 260 F.3d at 138-39; United States v. Lopez, 248 F.3d 427, 431-32 (5th Cir.), cert. denied, 534 U.S. 898, 122 S.Ct. 222, 151 L.Ed.2d 158 (2001); Lanier v. United States, 220 F.3d 833, 838 (7th Cir.2000); Murr v. United States, 200 F.3d 895, 905-06 (6th Cir. 2000).
.In contrast, our opinion in United States v. Buckland, 289 F.3d 558 (9th Cir.) (en banc), cert. denied, 535 U.S. 1105, 122 S.Ct. 2314, 152 L.Ed.2d 1067 (2002), illustrates how a decision with some substantive implications may be considered a procedural rule for Teag-ue purposes. In Buckland, we assessed the impact of Apprendi v. New Jersey, 530 U.S. 466, 120 S.Ct. 2348, 147 L.Ed.2d 435 (2000), on "type and quantity” findings under 21 U.S.C. § 841(b). 289 F.3d at 562-63. Ap-prendi held that "[ojther than the fact of a prior conviction, any fact that increases the penalty for a crime beyond the prescribed statutory maximum must be submitted to a jury, and proved beyond a reasonable doubt.” 530 U.S. at 490, 120 S.Ct. 2348. Rejecting the contention that Apprendi rendered § 841 unconstitutional, id. at 564, 120 S.Ct. 2348, we concluded that Apprendi did not alter, restructure, or redefine as a matter of New Jersey law the substantive elements of the underlying offense there at issue, nor did it create or resurrect a separate substantive offense. Further, we noted, Apprendi did not demand that we do so when assessing "type and quantity” evidence under § 841. Id. at 565, 568, 120 S.Ct. 2348. Rather, Apprendi imposed a particular procedure through which the existing " ‘elements’ in § 841 [ ]” must be established, viz., through submission to the jury and proof beyond a reasonable doubt. Id. Accordingly, we determined in Buckland that the impact of Apprendi on "type and quantity" under § 841 was one of a procedural — not substantive — ilk. Id. at 563, 120 S.Ct. 2348 (noting that Apprendi was retroactive nevertheless because the case arose through direct review) (citing Griffith, 479 U.S. at 328, 107 S.Ct. 708).
. The Eleventh Circuit did not address the question of whether Ring had substantive impact on Florida law in its consideration of whether Teague barred the retroactive application of Ring. See Turner, 339 F.3d at 1282-86, *33-*37. Thus, our consideration in this respect is different from the issue addressed by the Eleventh Circuit. To the extent that the Eleventh Circuit relied on a pure analogy to Apprendi in its Ring analysis, we respectfully disagree with its conclusions.
. In its assessment of the "linkage” between Ring and Apprendi, the dissent contends that we hitch our distinction of Apprendi solely to the putative inapplicability of harmless-error analysis in the Ring context. However, the crux of the analysis is different. What "distance[s]” Ring from Apprendi is not simply a *1102harmless error consequence. Rather, the very focus of the Supreme Court’s analysis in the two cases proves Ring and Apprendi distinct: Apprendi expressly refused to reach "[t]he substantive basis” of law at issue in that case, see 530 U.S. at 475, 120 S.Ct. 2348; Ring, conversely, did reach the relevant substantive basis. See 536 U.S. at 589-90, 122 S.Ct. 2428. In eliding this aspect of Ring's analysis, the dissent both overstates Ring's affinity to Apprendi and mischaracterizes the first step of our "syllogism.” See Dissent at 12789. The “substantive” aspect of Ring rests on more than the creation of a separate substantive offense. It rests, in addition, on the Supreme Court's wholesale invalidation of Arizona’s capital sentencing scheme. See Bousley, 523 U.S. at 620, 118 S.Ct. 1604. This kind of consideration of the "substantive basis” of the law was wholly absent from the Supreme Court’s analysis and decision in Ap-prendi.
. Justice O'Connor recognized this in her dissent in Ring, noting that ''[t]he Court effectively declares five States' capital sentencing schemes unconstitutional.” Ring, 536 U.S. at 620, 122 S.Ct. 2428 (O’Connor, J., dissenting) (referencing Arizona, Idaho, Montana, Colorado, and Nebraska).
. This concern is not merely theoretical. See James S. Liebman, et al., Broken System: Error Rates in Capital Cases, 1973-1995 at 5 (2000) available at <http://justice.poli-cy.net/jpreport/>.
. The capital murder statute effective at the time contemplated the consideration of pre-sentence reports by capital sentencing judges. Ariz.Rev.Stat. § 13-703(C) (1998) (repealed Laws 1999, Ch. 104, § 1). Consideration of presentence reports was routine. See, e.g., State v. Bocharski, 200 Ariz. 50, 22 P.3d 43, 56 (2001); State v. Mann, 188 Ariz. 220, 934 P.2d 784, 792 (1997); State v. Kemp, 185 Ariz. 52, 912 P.2d 1281, 1295 (1996); State v. Gulbrandson, 184 Ariz. 46, 906 P.2d 579, 599 (1995); State v. Stokley, 182 Ariz. 505, 898 P.2d 454, 468 (1995); State v. Brewer, 170 Ariz. 486, 826 P.2d 783, 801 (1992); see also infra note 12.
. During the period relevant to this case, a study commissioned by the Federal Judicial Center noted that "[t]he principal problem inherent in the use of the presentence report is its potential for introducing inaccurate or misleading information into the sentencing decision.” Stephen A. Fennell & William N. Hall, Due Process at Sentencing: An Empirical and Legal Analysis of the Disclosure of Presen-tence Reports in Federal Courts, 93 Harv. L.Rev. 1615/ 1628 (1980). The study cited an Arizona state case as a primary example in which an inaccurate presentence report caused "defendants [to be] incarcerated for a significantly longer period than they should have been because of untrue statements in the presentence report.” Id. at 1628-29, 826 P.2d 783 (citing State v. Killian, 91 Ariz. 140, 370 P.2d 287, 290 (1962)).
.See, e.g., State v. Sansing, 200 Ariz. 347, 26 P.3d 1118, 1129 (2001), cert. granted and judgment vacated by Sansing v. Arizona, 536 U.S. 954, 122 S.Ct. 2654, 153 L.Ed.2d 830 (2002); Bocharski, 22 P.3d at 56; State v. Soto Fong, 187 Ariz. 186, 928 P.2d 610, 633 (1996); State v. Spears, 184 Ariz. 277, 908 P.2d 1062, 1077 (1996); Gulbrandson, 906 P.2d at 599; State v. Williams, 183 Ariz. 368, 904 P.2d 437, 454 (1995); State v. Bolton, 182 Ariz. 290, 896 *1112P.2d 830 (1995); State v. Apelt, 176 Ariz. 349, 861 P.2d 634, 644 (1993); Brewer, 826 P.2d at 800; State v. Greenway, 170 Ariz. 155, 823 P.2d 22, 29 (1992); State v. Amaya Ruiz, 166 Ariz. 152, 800 P.2d 1260, 1287 (1990).
. Of course, the proper admission of victim impact evidence by itself does not necessarily violate the Eighth Amendment. Payne, 501 U.S. at 827, 111 S.Ct. 2597. However, the Supreme Court has held that "the admission of a victim's family members' characterizations and opinions about the crime, the defendant, and the appropriate sentence violates the Eighth Amendment.” Id. at 830 n. 2, 111 S.Ct. 2597 (citing Booth v. Maryland, 482 U.S. 496, 107 S.Ct. 2529, 96 L.Ed.2d 440 (1987)). In addition, if the material "is so unduly prejudicial that it renders the trial fundamentally unfair, the Due Process Clause of the Fourteenth Amendment provides a mechanism for relief.” Id. at 825, 111 S.Ct. 2597 (citing Darden v. Wainwright, 477 U.S. 168, 179-183, 106 S.Ct. 2464, 91 L.Ed.2d 144 (1986)). The Arizona Supreme Court has acknowledged that a defendant’s constitutional rights would be violated if a capital sentencing jury received sentencing recommendations from the victim's family and friends, but has allowed capital sentencing judges to receive the information on the basis that judges will disregard "the irrelevant, inflammatory, and emotional factors.” Bolton, 896 P.2d at 856.
. The dissent assails the use of juries in capital sentencing. This criticism misses the central issue. The presence of some imperfections in jury sentencing does not affect the conclusion that juries are ultimately more accurate than judges. While individual jurors may hold emotional or legally inaccurate views, the requirement of unanimity across twelve individuals significantly reduces the chance that these views will hold sway. Moreover, the fact that some jurors feel sympathy or pity does not imply that the verdict is ultimately governed by these emotions. See, e.g., Stephen P. Garvey, The Emotional Economy of Capital Sentencing, 75 N.Y.U. L.Rev. 26, 65 (2000) (noting the lack of correlation between feelings of sympathy or pity and a juror’s final vote).
. See John Blume & Theodore Eisenberg, Judicial Politics, Death Penalty Appeals, and Case Selection: An Empirical Study, 72 S. Cal. L.Rev. 465, 470-75 (1999) (describing a variety of campaigns to unseat state judges based on their alleged failure to impose or affirm death sentences); Stephen B. Bright et ah, Breaking the Most Vulnerable Branch: Do Rising Threats to Judicial Independence Preclude Due Process in Capital Cases, 31 Colum. Hum. Rts. L.Rev. 123 passim (1999) (providing additional examples of judges under attack due to the outcomes of capital cases over which they presided); Symposium, Politics and the Death Penalty: Can Rational Discourse and Due Process Survive the Perceived Political Pressure?, 21 Fordham Urb. L.J. 239, 270-73 (1994) (presenting statements by judges participating in symposium in which they described criticism they faced during elections based on their decisions in capital cases).
. The instruction at issue in Sullivan was similar to the one held unconstitutional in Cage v. Louisiana, 498 U.S. 39, 111 S.Ct. 328, 112 L.Ed.2d 339 (1990) (per curiam). Thus, the question in Sullivan was whether Cage error was subject to harmless-error analysis under Chapman v. California, 386 U.S. 18, 87 S.Ct. 824, 17 L.Ed.2d 705 (1967).
. Likewise, we have held that a constitutionally deficient indictment is a structural defect requiring reversal because the indictment “failfed] to ensure that [the defendant] was prosecuted only 'on the basis of the facts presented to the grand jury.’ ” United States v. Du Bo, 186 F.3d 1177, 1179 (9th Cir.1999) (quoting United States v. Rosi, 27 F.3d 409, 414 (9th Cir.1994)). We noted that "[flailing to enforce this requirement would allow a court to ‘guess as to what was in the minds of the grand jury at the time they returned the indictment.’ ” Id. (quoting United States v. Keith, 605 F.2d 462, 464 (9th Cir.1979)).
. The Arizona Supreme Court did not have the benefit of Nguyen when it decided Towery and Ring II. Thus, the Court assumed that the question of "who[ ] decides” the existence of aggravating circumstances was susceptible to harmless-error review. Towery, 64 P.3d at 834-35. For the reasons already discussed, however, harmless error cannot be assessed in a void, and, without a jury finding, there is nothing for the appellate court to review. The Arizona Supreme Court primarily relied on Neder v. United States, 527 U.S. 1, 119 S.Ct. 1827, 144 L.Ed.2d 35 (1999), but Neder in fact buttresses, rather than controverts, our conclusion that Ring error is structural. In Neder, the Supreme Court held that the failure to submit a materiality instruction was subject to harmless-error analysis. 527 U.S. at 8-15, 119 S.Ct. 1827. This, the Arizona Supreme Court reasoned, was akin to depriving a defendant a jury trial in a capital case penalty phase. Towery, 64 P.3d at 834-35. But Neder itself provides the crucial analytic distinction by identifying structural errors not susceptible to harmless-error analysis, namely errors that "infect the entire trial process.” 527 U.S. at 8, 119 S.Ct. 1827 (citing Brecht v. Abrahamson, 507 U.S. at 630, 113 S.Ct. 1710). These are errors that "deprive defendants of'basic protections’ without which 'a criminal trial cannot reliably serve its function as a vehicle for determination of guilt or innocence ... and no criminal punishment may be regarded as fundamentally fair.' ” Id. (quoting Rose, 478 U.S. at 577-78, 106 S.Ct. 3101). Examples of structural error provided by Neder are racial discrimination in jury selection, see Vasquez v. Hillery, 474 U.S. 254, 106 S.Ct. 617, 88 L.Ed.2d 598 (1986); biased judges, see Tumey v. Ohio, 273 U.S. 510, 47 S.Ct. 437, 71 L.Ed. 749 (1927); and denials of public trials, see Waller v. Georgia, 467 U.S. 39, 104 S.Ct. 2210, 81 L.Ed.2d 31 (1984). Neder, 527 U.S. at 8, 119 S.Ct. 1827. As Neder makes clear, for structural error, it does matter "who decides” the case and in what context. There is a vast difference between not submitting the element of materiality to the jury for decision and having no jury decision at all.
. The Mills/McKoy rule struck down state procedures that limited any given juror's consideration of mitigating circumstances in capital sentencing to such evidence that the entire jury had found relevant. McKoy v. North Carolina, 494 U.S. 433, 110 S.Ct. 1227, 108 L.Ed.2d 369 (1990); Mills v. Maryland, 486 U.S. 367, 108 S.Ct. 1860, 100 L.Ed.2d 384 (1988). Both the Fourth and the Sixth Circuits have held that the "Mills/McKoy " rule is a "watershed rule” that warrants retroactive application.
. But see Cordova v. Collins, 953 F.2d 167, 173 (5th Cir.1992) (holding without analysis that Teague bars retroactive application); Miller v. Lockhart, 65 F.3d 676, 686 & n. 6 (8th Cir.1995) (holding that Teague bars retroactive application of Mills and observing that the defendant never raised argument that Mills falls into a Teague exception).