Opinion by Judge Paez; Partial Concurrence and Partial Dissent by Judge Tallman.
OPINION
PAEZ, Circuit Judge.A California jury convicted Petitioner-Appellee William Charles Payton of the first degree murder and rape of Pamela Montgomery, and the attempted murder of Patricia Pensinger and her son, Blaine Pensinger. The jury imposed the death penalty. Payton appealed both the underlying conviction and the death sentence.
At the penalty phase of a trial in which a death sentence is at stake, a state may not preclude the jury from considering any mitigating circumstance “that the defendant proffers as a basis for a sentence less than death.” Eddings v. Oklahoma, 455 U.S. 104, 110, 102 S.Ct. 869, 71 L.Ed.2d 1 (1982) (internal quotations and citations omitted). The California death penalty statute channels the jury’s assessment of the appropriate penalty into an eleven-factor test that structures the jury’s weighing and balancing of the aggravating and mitigating circumstances. The first ten factors instruct the trier of fact to evaluate various circumstances specific to the crime and to account for the defendant’s age and prior convictions. The eleventh factor- — factor (k) — functions as a catch-all, enabling the jury to consider any other circumstance that the defendant presents in mitigation of a death sentence.
We are confronted here with the issue of whether, in Payton’s trial, the jury instructions regarding factor (k) impermissibly limited its constitutionally-mandated role as a vehicle for permitting the jury to consider all the mitigating evidence presented regarding whether Payton deserved a life term rather than a death sentence. In instructing the jury, the trial court employed the then-existing model jury instructions which incorporated the multi-factor test in the statute. 1 California Jury Instructions, Criminal (“CALJIC”) 8.84.1 (4th ed.1979). That instruction simply quotes factor (k), directing the jury to consider any circumstance “which extenuates the gravity of the crime even though it is not a legal excuse for the crime.” Id.; Cal.Penal Code § 190.3 (1978). The Supreme Court, reviewing the same jury instruction in Boyde v. California, 494 U.S. 370, 110 S.Ct. 1190, 108 L.Ed.2d 316 (1990), held that the text of factor (k), as clarified by the trial court, enabled the jury to consider pre-crime character and background evidence. The Court did not address the question of factor (k)’s application to post-crime evidence of rehabilitation, and did not have occasion to evaluate the effect on the jury of a prosecutor’s contention that such evidence could not be considered. Those questions squarely confront us here.
At the penalty phase of Payton’s trial, the only evidence offered in mitigation was Payton’s post-crime conversion to Christianity and his good works while in jail, which were offered under factor (k). The defense offered no other evidence then or *819at any other time during his trial. In closing argument, the prosecutor erroneously told the jury that factor (k) did not encompass the only evidence Payton offered to mitigate a sentence of death. Although defense counsel objected to the prosecutor’s argument, the trial court failed to cure the error.
On automatic appeal to the California Supreme Court, Payton argued, among other things, that he was deprived of a fundamentally fair trial because the trial court’s instructions and the prosecutor’s erroneous argument led the jurors to believe that they were not permitted to consider Payton’s mitigating evidence. The California Supreme Court affirmed the conviction and sentence. People v. Payton, 3 Cal.4th 1050,13 Cal.Rptr.2d 526, 839 P.2d 1035 (1992), cert. denied, 510 U.S. 1040, 114 S.Ct. 682, 126 L.Ed.2d 649 (1994). Subsequently, the California Supreme Court denied Payton’s petition for a writ of habeas corpus. Payton then filed a petition for habeas corpus relief in federal court under 28 U.S.C. § 2254 (1994). The district court concluded that the penalty phase of the trial was fundamentally unfair and granted a writ of habeas corpus requiring either a new penalty trial or the reduction of Paytbn’s sentence to a life term without parole. A divided three-judge panel of our court reversed the grant of the writ as to the penalty phase. Payton v. Woodford, 258 F.3d 905 (9th Cir.), reh’g en banc granted, 273 F.3d 1271 (2001).
We then agreed to rehear this case en banc. We affirm the district court’s judgment in full. We hold that it is reasonably likely that the text of factor (k) and the trial court’s failure to correct the prosecutor’s misstatements about the reach of factor (k) caused the jury to disregard relevant mitigating evidence, and that this error was not harmless.1
Background2
In 1980, while spending the night at Patricia Pensinger’s home, Payton raped Pamela Montgomery and stabbed her to death. Payton then entered the bedroom of Pensinger and her son Blaine, stabbed each of them repeatedly, and fled. Payton was charged with the first degree murder and rape of Montgomery, and the attempted murders of Pensinger and her son.
At the guilt phase of Payton’s jury trial, the prosecution presented testimony from the law enforcement officers who observed the crime scene; forensics experts who confirmed that saliva and semen samples taken from Montgomery’s body were consistent with Payton’s; Patricia and Blaine Pensinger who gave victims’ accounts of the attacks; Payton’s wife, who stated that soon after the attacks she saw blood on Payton’s clothes, face, hands and penis as well as fingernail scratches and digs on his legs and back; and a fellow inmate, Alejandro Garcia, who recounted that Payton admitted that he raped and stabbed Montgomery and stabbed the Pensingers be*820cause he “had this urge to kill.” The defense called no witnesses, and the jury-convicted on all counts.
During the penalty phase, the prosecution presented as a witness a fellow inmate who testified to his jailhouse conversations with Payton in which Payton admitted that he had “severe problems with sex and women,” that he wanted to “stab them and rape them,” and that every “wom[a]n on the street he [saw] was a potential victim, regardless of age or looks.” Payton’s former girlfriend related that she had once awakened to find Payton holding a kitchen knife to her neck, and that he had stabbed her chest and arms. After she pushed him off, he stayed with her and held a towel around her bleeding arm until the police arrived.
The defense presented eight witnesses, including Payton’s pastor, a deputy sheriff, four inmates, his mother, and the director of a religious organization ministering to prisoners. Their testimony, taken as a whole, tended to show that Payton had been “born again,” made a sincere commitment to God, and was performing good works in jail.
Payton’s pastor testified that in his opinion, Payton’s conversion was credible and that he was “sincere in his statement and commitment to the Lord.” The director of a religious outreach organization ministering to prisoners testified to her numerous conversations with Payton about his spiritual commitment and its manifestation in the bible study groups he established with other inmates. She described his conversion of other inmates, his admission to a correspondence bible college, and his writings.
Four inmates testified that they believed that Payton’s religious conversion was sincere and that he had a calming influence on other inmates. One testified that Pay-ton’s intervention prevented him from committing suicide. A deputy sheriff assigned to Payton’s jail facility related that Payton led prayer meetings and had a positive influence on other inmates. Pay-ton’s mother described praying together with her son and discussing religion on a weekly basis. Asked if she had noticed a change in her son, she responded: “Oh, yes.... He’s totally immersed in the Lord.... He’s an instrument of the Lord as far as he’s concerned.”
Prior to closing arguments in the penalty phase, the judge held an in-chambers conference with the attorneys about the jury instructions. They discussed the application of the multi-factor CALJIC instruction that guides the jury in determining whether to impose a sentence of life imprisonment or death.3 Factor (k), the *821eleventh and final factor, directed that the jury consider “[a]ny other .circumstance which extenuates the gravity of the crime even though it is not a legal excuse for the crime.” CALJIC 8.84.1. Payton’s counsel sought an amendment to the instruction that expressly would have directed the jury to consider “evidence of the defendant’s character, background, history, mental condition and physical condition.”4 Although the trial judge agreed with the defense counsel’s interpretation of factor (k), he declined the request because he was reluctant to alter the instruction insofar as it reflected verbatim the text of California Penal Code § 190.3. He stated that he would allow counsel to argue the point. The judge also denied defense counsel’s separate proposal to amend the instruction to permit the jury to consider Payton’s “potential for rehabilitation.”
During closing argument, the prosecutor argued to the jury that factor (k) applied to “some factor at the time of the offense that somehow operates to reduce the gravity for what the defendant did” but that it did not “refer to anything after the fact or later.” He asserted that factor (k) did not encompass Payton’s conversion to Christianity and good conduct in jail because they occurred “well after the act of the crime,” and the factor “seems to refer to a fact in operation at the time of the offense.” At one point, the prosecutor said:
“What I am getting at, you have not heard during the past few days any legal evidence of mitigation. What you’ve heard is just some jailhouse evidence to win your sympathy, and that’s all. You have not heard any evidence of mitigation in this trial.
Concluding, the prosecutor told the jury that he did not “want to spend too much time on [Payton’s religious conversion] because I don’t think it’s really applicable and I don’t think it comes under any of the eleven factors.”
In response to the prosecutor’s factor (k) argument, the defense moved for a mistrial, objecting that the prosecutor’s argument was “completely contrary to what we all agreed in chambers on the record ‘k’ was designed to apply to.” The court responded that it was a “fair comment on either side” and “I think-you can argue it either way.” The court told the jury that “the comments by both the prosecution and the defense are not evidence. You’ve heard the evidence and, as I said, this is argument. And it’s to be placed in its proper perspective.”
Defense counsel’s closing argument acknowledged that factor (k) “may be awk*822wardly worded.” He argued that the factor was designed as a catch-all to include the kind of evidence in mitigation he had presented, and that, for Payton, it was the most critical of the factors.
After the closing arguments, the judge instructed the jury as noted above. Upon receiving instructions that it must reach a unanimous result, the jury retired to deliberate. The jury returned a verdict of death.
Discussion
We hold that the district court properly granted the writ of habeas corpus. As a preliminary matter, we confirm that this case is governed by the legal standards in effect prior to the effective date of the Anti-Terrorism and Effective Death Penalty Act of 1996, Pub.L. No. 104-132, 110 Stat. 1218 (April 24, 1996) (“AEDPA”). We then conclude that the relevant inquiry in this case is whether there was instructional error under Boyde, 494 U.S. at 380, 110 S.Ct. 1190. We hold that, under Boyde, there is a “reasonable likelihood” that the jury applied factor (k) in a way that prevented the consideration of constitutionally relevant evidence. Id. We further hold that this error was not harmless because of the likelihood that it precluded consideration of the only mitigating evidence that Payton presented at trial.
A Application of AEDPA
Because Payton filed his petition for the appointment of habeas counsel pri- or to the effective date of AEDPA, we review the district court’s order under pre-AEDPA standards. See Calderon v. United States Dist. Court (“Kelly”), 163 F.3d 530, 540 (9th Cir.1998) (en banc) (holding that a petition for appointment of habeas counsel, coupled with a motion for a stay of execution, fixes the date for determining whether AEDPA applies). We decline Respondent’s invitation to reconsider our decision in Kelly.
Applying pre-AEDPA standards, we presume that state court determinations of historical fact are correct. 28 U.S.C. § 2254(d) (1994). In contrast, the application of legal standards to historical facts does not warrant a presumption of correctness under § 2254(d) (1994). Thompson v. Borg, 74 F.3d 1571, 1573 (9th Cir.1996).
B. Instructional Envr
The central question in this case is whether the jury received a constitutionally adequate instruction guiding consideration of Payton’s mitigating evidence. The Constitution requires a capital jury to consider all relevant mitigating evidence. Boyde, 494 U.S. at 377-78, 110 S.Ct. 1190; Eddings, 455 U.S. at 113-14, 102 S.Ct. 869 (“Just as the State may not by statute preclude the sentencer from considering any mitigating factor, neither may the sen-tencer refuse to consider, as a matter of law, any relevant mitigating evidence.”). This broad permission includes authority to consider evidence of Payton’s good conduct after the crime. Skipper v. South Carolina, 476 U.S. 1, 106 S.Ct. 1669, 90 L.Ed.2d 1 (1986) (holding that post-crime good behavior must be considered as mitigating evidence). The trial court’s instructions to the jury must impart this constitutional directive.
Respondent urges us to apply the standard for prosecutorial misconduct rather than instructional error and consider whether the prosecutor’s argument “so infected the trial with unfairness as to make the resulting conviction a denial of due process.” Darden v. Wainwright, 477 U.S. 168, 181, 106 S.Ct. 2464, 91 L.Ed.2d 144 (1986). Applying this standard, the district court concluded that Payton’s trial *823had been unconstitutionally infected with unfairness.
We need not go down that road. At bottom, the constitutional violation here flows from the lack of guidance that the jury received regarding its duty to consider mitigating evidence. The prosecutor’s arguments cannot be isolated from the instruction itself or from the failure of the trial judge properly to instruct the jury or to correct the prosecutor’s error. Thus, the focus of our inquiry is whether, viewing the case as a whole, the court’s instructions properly guided the jury to consider Payton’s mitigating evidence.
Our approach here is consistent with Boyde. In Boyde, the Court first determined whether there was a reasonable likelihood that the jury applied the factor (k) instruction in a way that prevented consideration of the mitigating background and character evidence that Boyde presented. 494 U.S. at 381-84, 110 S.Ct. 1190. It then turned to Boyde’s claim that the prosecutor’s argument reinforced an impermissible interpretation of factor (k). Id. at 384-86, 110 S.Ct. 1190. Significantly, the Court did not discuss the prosecuto-rial misconduct standard. Instead, as we do here, the Court analyzed how the jury would have interpreted the instruction in light of the prosecutor’s argument. Id.
Under Boyde, we must reverse for instructional error if the challenged instruction is potentially ambiguous and there is a “reasonable likelihood” that the jury applied the challenged instruction in a way that prevents the consideration of constitutionally relevant evidence. Id. at 380, 110 S.Ct. 1190. We must also determine whether the error was harmless. Id. at 380, 110 S.Ct. 1190; Calderon v. Coleman, 525 U.S. 141, 147, 119 S.Ct. 500, 142 L.Ed.2d 521 (1998) (per curiam).
1. Ambiguity in unadorned factor (k)
The meaning of the factor (k) model instruction as it existed at the time of Payton’s trial was far from clear.5 That instruction directed the jury to consider “any other circumstance which extenuates the gravity of the crime even though it is not a legal excuse for the crime.” Cal.Penal Code § 190.3. We “approach jury instructions in the same way a jury would — ■ with a ‘commonsense understanding of the instructions in the light of all that has taken place at the trial.’ ” Penry v. Johnson, 532 U.S. 782, 800, 121 S.Ct. 1910, 150 L.Ed.2d 9 (2001) (quoting Boyde, 494 U.S. at 381, 110 S.Ct. 1190). Most naturally read, the phrase “extenuates the gravity of the crime” refers to evidence relating to or ameliorating the crime itself. On its face, factor (k) does not encompass the kind of post-crime evidence of good works, leadership and religious beliefs that Payton presented at the penalty phase of his trial.6 *824Certainly, the prosecutor’s interpretation of this factor as excluding post-crime evidence bolsters the conclusion that the jury instruction was ambiguous in its application to Payton’s mitigating circumstances.
The year after the jury announced Pay-ton’s death sentence, the California Supreme Court recognized the potential for jury confusion inherent in the wording of factor (k). People v. Easley, 34 Cal.3d 858, 196 Cal.Rptr. 309, 671 P.2d 813, 825-26 & n. 10 (1983). The court acknowledged that there was some force to the argument that a jury might reasonably construe the text of the instruction “to permit consideration only of circumstances that relate to the ‘gravity of the crime ’ and not of circumstances that relate to the general character, family background or other aspects of the defendant.” Id. at 825-26.
The United States Supreme Court held that the factor (k) instruction was not ambiguous as applied to pre-crime background and character evidence as long as the trial court provided clarification of its meaning.7 Boyde, 494 U.S. at 381-82 n. 5, 110 S.Ct. 1190. The Court held that factor (k) passed constitutional muster because there was no “reasonable likelihood” that the jury was misled into believing it could not consider Boyde’s mitigating evidence. Id. at 381, 110 S.Ct. 1190.
Boyde did not address the question whether, on its face, the unadorned factor (k) instruction is unconstitutionally ambiguous as applied to post-crime evidence. The fact that all of Payton’s mitigating evidence was post-crime distinguishes this case from the pre-crime evidence at issue in Boyde which “more readily fits within factor (k).”8 Payton, 258 F.3d at 928 (Hawkins, J., dissenting). Significantly, Boyde distinguished the pre-crime evidence at issue there from evidence — such as Payton's — that “pertain[ed] to prison behavior after the crime for which he was sentenced to death.” Boyde at 382 n. 5, 110 S.Ct. 1190.
Unlike the pre-crime evidence in Boyde, post-crime mitigation evidence is simply not covered by any natural reading of the words of the unadorned factor (k) instruction. Mitigation evidence occurring after the crime cannot possibly “extenuate the gravity of the crime.” Because the unadorned factor (k) instruction does not encompass post-crime evidence, it violates Skipper’s requirement that the jury be permitted to consider post-crime good behavior as mitigating evidence in deciding whether to impose the death penalty. See 476 U.S. at 5, 106 S.Ct. 1669. Standing alone, the factor (k) instruction is unconstitutional as applied to post-crime evidence.
2. The conflicting legal arguments of counsel
The trial court’s failure to correct the prosecutor’s erroneous interpretation of that instruction, by compounding the potential for confusion inherent in the text of the factor (k) instruction, roots more deeply our conclusion that there was constitutional error. There is no dispute that the prosecutor impermissibly narrowed the *825scope of factor (k) when he argued to the jurors that the factor did not “refer to anything after” the crime “or later” and that they should not consider Payton’s evidence in mitigation. See Payton, 13 Cal.Rptr.2d 526, 839 P.2d at 1048 (“It is true that the prosecutor during closing argument suggested a narrow and incorrect interpretation of factor (k).”); see also Payton, 258 F.3d at 916 (“In this case, there is no question that the prosecutor misstated what factor (k) refers to.”).
The prosecutor’s statements further distinguish this case from Boyde. The prosecutor in Boyde “never suggested that the background and character evidence could not be considered.” 494 U.S. at 385, 110 S.Ct. 1190. In contrast, the prosecutor here told the jurors that the statutory list of factors precluded them from considering the only mitigating evidence Payton presented — evidence of a post-crime religious conversion and its positive effects on other inmates and the administration of the jail. When a natural reading of the unadorned factor (k) instruction already favored the prosecutor’s stance, defense counsel faced an imposing hurdle to convince the jury of the proper interpretation.
S. The absence of instruction from the trial court
We recognize that arguments of counsel generally carry less weight with a jury than instructions from the trial court. Boyde, 494 U.S. at 384, 110 S.Ct. 1190. The trial court, however, did nothing to level this uneven playing field. Over the objection of Payton’s counsel, the trial court decided to allow each attorney to argue his own legal interpretation to the jury, rather than instructing the jury as to which interpretation was correct. In contrast, the Supreme Court’s holding in Boyde that the jury understood the scope of factor (k) relied heavily on the trial court’s clarifying instruction allowing the jury to consider “any other circumstance that might excuse the crime,” which included the defendant’s background and character. Id. at 381-82 & n. 5, 110 S.Ct. 1190 (emphasis in original).
Here, the only “curative” instruction given was that the comments by the prosecutor and the defense counsel were not evidence. The ineffectiveness of the trial court’s instruction is clear from the prosecutor’s return, after the trial court’s admonition, to his argument to the jury that factor (k) did not encompass Payton’s mitigating evidence.
Nor did the trial court’s final instructions to the jury cure the error here. Before the jury retired to deliberate, as noted, the trial court instructed:
In determining the penalty to be imposed on the defendant, you shall consider all of the evidence which has been received during any part of the trial in this case, except as you may be hereafter instructed. You shall consider, take into account and be- guided by the following factors, if applicable ....
(emphasis added). The trial court’s directive to “consider all of the evidence” failed to correct the prosecutor’s error. In the same breath, the trial court stated that the jury should consider all the evidence “except as you may be hereafter instructed” and then instructed them to be “guided by” the eleven-factor test. Thus, the trial court confined the jury’s consideration of the evidence to the multi-factor test that the prosecutor had just declared did not allow consideration of Payton’s extensive mitigating evidence. The judge then instructed the jury that it was to apply the factors only “if applicable.”
In effect, the court’s instruction delegated to the jury the legal question whether factor (k) allowed consideration of Payton’s mitigating evidence. Nothing prevented *826the jury from refusing to consider Payton’s mitigating evidence and thereby reaching an unconstitutional result. See Eddings, 455 U.S. at 114-15, 102 S.Ct. 869 (“The sentencer ... may determine the weight to be given relevant mitigating evidence. But [it] may not give it no weight by excluding such evidence from [its] consideration.”). When “jurors have been left the option of relying upon a legally inadequate theory, there is no reason to think that their own intelligence and expertise will save them from that error.” Griffin v. United States, 502 U.S. 46, 59, 112 S.Ct. 466, 116 L.Ed.2d 371 (1991). We cannot expect a jury to reach the constitutionally correct conclusion that the multi-factor instruction compelled consideration of Pay-ton’s mitigating evidence when the jury must overcome both the text of factor (k) and the facially reasonable argument of the prosecutor. These circumstances likely stripped Payton of his only defense to the imposition of the death penalty.
Thus, Payton has satisfied Boyde’s standard requiring that he establish that there was a reasonable likelihood that the jury applied the instruction in a way that prevented consideration of his mitigating evidence. Boyde does not require that Payton show that “the jury was more likely than not to have been impermissibly inhibited by the instruction.” Boyde, 494 U.S. at 380, 110 S.Ct. 1190. However, Payton’s death sentence would be constitutional “if there is only a possibility of such an inhibition.” Id. In determining whether more than a “possibility” of inhibition existed, we do not limit our inquiry to how “a single hypothetical ‘reasonable’ juror could or might have interpreted the instruction.” Id. The claimed error must amount to more than speculation about the jury’s understanding of the instruction. Id.
Payton’s claim is more than speculative. Compounding the nebulous terms of the unadorned factor (k) instruction were the prosecutor’s erroneous argument and the trial court’s silence as to the jury’s constitutional obligation to consider all of the mitigating evidence. In Easley, the California Supreme Court stated that trial courts should, in instructing jurors on factor (k), tell juries that they can consider any aspect of the defendant’s character or record. 196 Cal.Rptr. 309, 671 P.2d at 826 n. 10. Here, defense counsel asked for an instruction similarly clarifying the breadth of the scope of factor (k). Despite agreeing that it was a “catch-all provision,” the trial judge refused. Instead, the jury was given the unadorned factor (k) instruction without any explanation by the court as to what was appropriate to consider under factor (k). In sum, the jury received the multi-factor instruction, including factor (k), on the same plate with the contentions of the prosecutor and defense counsel as to its applicability, and without further guidance from the trial court.
Penry v. Johnson, 532 U.S. 782, 121 S.Ct. 1910, 150 L.Ed.2d 9 (2001), confirms our conclusion that there was a reasonable likelihood that the jury did not consider Payton’s mitigating evidence. There, the Supreme Court condemned a similar tripartite error consisting of a jury instruction that excluded consideration of Penry’s mitigating evidence, the prosecutor’s exhortation to the effect that the jury should follow that instruction, and the trial court’s failure to provide a “vehicle” for the jury to “express [ ] the view that Penry did not deserve to be sentenced to death based upon his mitigating evidence.” Id. at 804, 121 S.Ct. 1910. In emphasizing that the jury must be able to “consider and give effect to a defendant’s mitigating evidence in imposing sentence,” the Court stated:
[I]t is only when the jury is given a vehicle for expressing its reasoned mor*827al response to that evidence in rendering its sentencing decision that we can be sure that the jury has treated the defendant as a uniquely individual human being and has made a reliable determination that death is the appropriate sentence.
Id. at 797, 121 S.Ct. 1910 (internal quotations, brackets, italics, and citations omitted).
Penry reminds us that we presume that jurors follow their instructions.9 Id. at 799, 121 S.Ct. 1910. When the effect of a mitigation instruction, viewed in the full context of the trial, is to confuse or mislead the jury in its duty to consider all relevant mitigation evidence, there has been constitutional error. By labeling the prosecutor’s incorrect contentions mere “argument,” the trial court not only failed to correct a critical misstatement of law but also effectively instructed the jury to consider the prosecutor’s erroneous legal position. See Caldwell v. Mississippi 472 U.S. 320, 339, 105 S.Ct. 2633, 86 L.Ed.2d 231 (1985). This directive is sufficient to establish constitutional error.
C. Harmless error
Having concluded that an error of constitutional magnitude impacted the penalty phase of Payton’s trial, we turn to whether that error was nevertheless harmless. We hold that the error had a “substantial and injurious effect or influence” oh the jury’s verdict. Brecht v. Abrahamson, 507 U.S. 619, 637, 113 S.Ct. 1710, 123 L.Ed.2d 353 (1993); O’Neal v. McAninch, 513 U.S. 432, 436, 115 S.Ct. 992, 130 L.Ed.2d 947 (1995).
Our jurisprudence is divided as to whether the petitioner or the state, or neither, bears responsibility for demonstrating the significance of the error under the Brecht/O’Neal harmlessness standard. Compare Rodriguez v. Marshall, 125 F.3d 739, 744 (9th Cir.1997) (stating that petitioner bears the burden of showing harm); Franklin v. Henry, 122 F.3d 1270, 1273 (9th Cir.1997) (same); with Keating v. Hood, 191 F.3d 1053, 1062 (9th Cir.1999) (as amended) (noting that the state bears the burden of showing harmlessness); Fisher v. Roe, 263 F.3d 906, 917 (9th Cir.2001) (same); and with Gray v. Klauser, 282 F.3d 633, 651 (9th Cir.2002); Thompson, 74 F.3d at 1575 (rejecting burdens of proof in favor of an independent determination of whether a trial error had a substantial and injurious effect).10
It is clear from O’Neal that the petitioner does not bear the burden of showing harm. 513 U.S. at 437-45. Because the harmless error analysis is a purely legal question that lies outside the realm of fact-finding, we dispense with burdens of proof and presumptions. See O’Neal, 513 U.S. at 437, 115 S.Ct. 992 (explaining that the court must determine *828whether the error affected the judgment “without benefit of such aids as presumptions or allocated burdens of proof that expedite fact-finding at the trial)” (quoting R. Traynor, The Riddle of Harmless Error 26 (1970)). O’Neal directs us to ask a “conceptually clearer” question in reviewing the record in a habeas case: “Do I, the judge, think that the error substantially influenced the jury’s decision?” 513 U.S. at 436, 115 S.Ct. 992.
In the course of this inquiry, it is the State that bears the “risk of doubt.” Id. at 438, 115 S.Ct. 992. When issues arise during our analysis which create uncertainty, the petitioner is entitled to the benefit of the doubt. See id. at 436, 438-43, 115 S.Ct. 992. At the close of our inquiry, we step back to determine where we are on the spectrum of certainty about the harmlessness of the constitutional error. If we are convinced that “the error did not influence the jury, or had but very slight effect, the verdict and the judgment should stand.” Id. at 437, 115 S.Ct. 992 (quoting Kotteakos v. United States, 328 U.S. 750, 764-65, 66 S.Ct. 1239, 90 L.Ed. 1557 (1946)). If, on the other hand, we are not fairly assured that there was no effect on the verdict, we must reverse. Id.; Gray, 282 F.3d at 651. In the “narrow circumstance” in which we are in “grave doubt” as to the effect of the constitutional error, we must assume that there was such an effect, and grant the petition. O’Neal, 513 U.S. at 437, 115 S.Ct. 992; see also Thompson, 74 F.3d at 1575.
Thus, we look to the State to instill in us a “fair assurance” that there was no effect on the verdict. Gray, 282 F.3d at 651; United States v. Hitt, 981 F.2d 422, 425 (9th Cir.1992); see also O’Neal, 513 U.S. at 443, 115 S.Ct. 992 (“[T]he State normally bears responsibility for the error that infected the initial trial.”). Only if the State has persuaded us that there was no substantial and injurious effect on the verdict do we find the error harmless.
This framework is faithful to the balance the Supreme Court has struck between concerns of federal-state comity and finality in state criminal trials, and the irreversible harm caused by an execution resulting from an unconstitutional error. In weighing these concerns in a non-capital case, the Supreme Court has stated:
[T]he number of acquittals wrongly caused by grant of the writ and delayed retrial (the most serious harm affecting the State’s legitimate interests) will be small when compared with the number of persons whom this opposite rule (denying the writ) would wrongly imprison or execute. On balance, we must doubt that the law of habeas corpus would hold many people in prison “in violation of the Constitution,” for fear that otherwise a smaller number, not so held, may eventually go free.
O’Neal, 513 U.S. at 443, 115 S.Ct. 992. Placing the “risk of doubt” on the state is also consistent with the body of jurisprudence that has placed the burden of showing lack of prejudice on the party who would benefit from the constitutional error. Id. at 437-44, 115 S.Ct. 992; United States v. Olano, 507 U.S. 725, 741, 113 S.Ct. 1770, 123 L.Ed.2d 508 (1993) (stating that the government bears the “burden of showing the absence of prejudice”); Chapman v. California, 386 U.S. 18, 24, 87 S.Ct. 824, 17 L.Ed.2d 705 (1967) (noting that “the original common-law harmless-error rule put the burden on the beneficiary of the error ... to prove that there was no injury”). Kotteakos, which articulated the harmlessness standard that Brecht later adopted and that we now apply, “places the burden on prosecutors to explain why those errors were harmless.” O’Neal, 513 U.S. at 438-39, 115 S.Ct. 992 (quoting Brecht, 507 U.S. at 640, 113 S.Ct. 1710 *829(Stevens, J., concurring) (citing Kotteakos, 328 U.S. at 760, 66 S.Ct. 1239)).11
Considering the record before us, the State has not provided us with a “fair assurance” that the error did not prejudice the penalty phase of Payton’s trial. O’Neal, 513 U.S. at 437-38, 115 S.Ct. 992; Gray, 282 F.3d at 651. On one side of the balance sheet is Respondent’s evidence of aggravating circumstances. There is no question that this was a brutal crime. The prosecution introduced eyewitnesses to Payton’s actions, testimony as to his motives and character, and forensic and other evidence to demonstrate to the jury the devastating effects of the crime.
It is the other side of the balance sheet that undermines any assurance that the jury’s verdict was not affected. As required by California Penal Code § 190.3, the trial court further instructed the jury that “If you conclude that the aggravating circumstances outweigh the mitigating circumstances, you shall impose a sentence of death.” We have determined that there is a reasonable likelihood that the jury-accepted the prosecutor’s statement of the law rather than the defense counsel’s and that it therefore failed to consider the only evidence offered in mitigation of the death penalty. That left the jury bereft of any countervailing evidence to weigh against the prosecution’s evidence of aggravating circumstances.
We cannot know whether the jury would have returned a verdict of life or of death had it been properly instructed. Payton’s extensive evidence of his conversion to Christianity, positive influence on other inmates, and other good works in jail were offered to evoke to the jury his potential for rehabilitation.12 If the jury had been inclined to weigh favorably evidence of redeeming features of his character or his conduct while in custody pending trial, it would have felt constrained by law from considering that evidence. Without Pay-ton’s mitigating evidence, the jury was bound by California Penal Code § 190.3 to impose a death sentence. See Easley, 196 Cal.Rptr. 309, 671 P.2d at 827.
Having pondered “all that happened without stripping the erroneous action from the whole,” we do not arrive at a fair assurance that the error was harmless. Gray, 282 F.3d at 651 (quoting O’Neal, 513 U.S. at 437, 115 S.Ct. 992). As we have previously stated, “[bjecause a death sentence is qualitatively different from other forms of punishment, there is a greater need for reliability in determining whether it is appropriate in a particular case.” *830Coleman v. Calderon, 210 F.3d 1047, 1050 (9th Cir.2000); see also Mills v. Maryland, 486 U.S. 367, 376, 108 S.Ct. 1860, 100 L.Ed.2d 384 (1988) (“In reviewing death sentences, the Court has demanded even greater certainty that the jury’s conclusions rested on proper grounds.”). Far from a fair assurance that the error was harmless, the “possible jury confusion” arising from the trial court instruction leaves us in “grave doubt about the likely effect of [the] error on the jury’s verdict.” O’Neal, 513 U.S. at 435, 115 S.Ct. 992; see also Fisher, 263 F.3d at 917-18. We conclude, therefore, that the instructional error had a “substantial and injurious effect or influence on the jury’s verdict” that necessitates a new penalty phase trial. See Coleman, 525 U.S. at 147, 119 S.Ct. 500. Payton is entitled to a penalty trial before a jury that is properly instructed that it must take his post-crime evidence into account in determining whether to impose a sentence of life or death.
Conclusion
Accordingly, we AFFIRM the judgment of the district court granting Respondent’s motion for summary judgment as to all claims except Claim IVB, item 3 of the Petition for Habeas Corpus, and granting the writ of habeas corpus as to the penalty phase of the trial.
AFFIRMED.
. Payton also contested the underlying conviction, raising several challenges to the guilt phase of his trial. The district court found no constitutional error in his conviction. In appeal No. 00-99003, Payton challenges the district court's rulings rejecting his claims of ineffective assistance of counsel, prosecutorial misconduct during the guilt phase of the trial, and the cumulative effects of the alleged constitutional errors. The panel affirmed the district court's rulings on these issues, as do we. We adopt the panel’s reasoning on the guilt phase issues as our own. See Payton, 258 F.3d at 919-25.
. We summarize the pertinent facts only briefly. The facts surrounding Payton's conviction are set forth in detail in the opinions of the panel and the California Supreme Court. Payton, 258 F.3d at 910-14; Payton, 13 Cal.Rptr.2d 526, 839 P.2d at 1039-40.
. The instruction provided in full:
In determining which penalty is to be imposed on[each] defendant, you shall consider all of the evidence which has been received during any part of the trial of this case, [except as you may be hereafter instructed], You shall consider, take into account and be guided by the following factors, if applicable:
(a) The circumstances of the crime of which tire defendant was convicted in the present proceeding and the existence of any special circumstance[s] found to be true.
(b) The presence or absence of criminal activity by the defendant which involved the use or attempted use of force or violence or the expressed or implied threat to use force or violence.
(c) The presence or absence of any prior felony conviction.
(d) Whether or not the offense was committed while the defendant was under the influence of extreme mental or emotional disturbance.
(e) Whether or not the victim was a participant in the defendant's homicidal conduct or consented to the homicidal act.
(f) Whether or not the offense was committed under circumstances which the defen*821dant reasonably believed to be a moral justification or extenuation for his conduct.
(g) Whether or not the defendant acted under extreme duress or under the substantial domination of another person.
(h) Whether or not at the time of the offense the capacity of the defendant to appreciate the criminality of his conduct or to conform his conduct to the requirements of law was impaired as a result of mental disease or defect or the affects [sic] of intoxication.
(i) The age of the defendant at the time of the crime.
(j) Whether or not the defendant was an accomplice to the offense and his participation in the commission of the offense was relatively minor.
(k)Any other circumstance which extenuates the gravity of the crime even though it is not a legal excuse for the crime.
CALJIC 8.84,1. In his instructions to the jury, the trial judge omitted the bracketed word "each” and retained the bracketed phrase "except as you may be hereafter instructed.”
. The proposed amendment read: "Any other circumstance which extenuates the gravity of the crime even though it is not a legal excuse for the crime, including evidence of the defendant's character, background, history, mental condition and physical condition.”
. In line with the suggestion of the California Supreme Court in People v. Easley, 34 Cal.3d 858, 196 Cal.Rptr. 309, 671 P.2d 813 (1983), the factor (k) instruction has since been amended to ensure that the jury may consider "any sympathetic or other aspect of the defendant's character or record [that the defendant offers] as a basis for a sentence less than death, whether or not related to the offense for which he is on trial.” See CALJIC 8.85(k) (6th ed.1996); see Easley, 196 Cal.Rptr. 309, 671 P.2d at 826 n. 10. We refer to the factor (k) instruction as it existed prior to this amendment as “unadorned.”
. The dissent casts Payton's religious beliefs as an overnight occurrence manufactured for the occasion, stating that the jury heard evidence of Payton's religious conversion "after Payton was apprehended for raping and murdering one individual and attempting to murder two others.” Infra, at 10782. In fact, a year and nine months spanned the date of the crime and the date of Payton’s death sentence, during which Payton's conversion and religious works took place.
. The trial court in Boyde defined the term “extenuate” to mean “to lessen the seriousness of the crime as by giving an excuse." Id. at 381, 110 S.Ct. 1190.
. In Babbitt v. Calderon, 151 F.3d 1170, 1178-79 (9th Cir.1998), reviewing the application of the factor (k) instruction to evidence of Babbitt's background, we noted the Supreme Court's holding in Boyde that the instruction did not mislead the jury to exclude consideration of Boyde's background and character evidence. Babbitt did not address the issue, squarely presented here, of the applicability of the factor (k) instruction to post-crime evidence.
. The dissent's reliance on Weeks v. Angelone, 528 U.S. 225, 120 S.Ct. 727, 145 L.Ed.2d 727 (2000) is misplaced. In Weeks, the Supreme Court considered an instruction that it had previously determined was unambiguous standing alone. Id. at 231, 120 S.Ct. 727 (citing Buchanan v. Angelone, 522 U.S. 269, 118 S.Ct. 757, 139 L.Ed.2d 702 (1998)). In contrast, the Supreme Court in Boyde determined that the factor (k) instruction was constitutional by relying heavily on the trial judge's clarifying instruction to the jury about its meaning. Moreover, in Weeks, the Supreme Court emphasized that the trial judge had separately instructed the jury to consider all mitigating circumstances. This is almost exactly the instruction that Payton's defense counsel requested and that the trial judge rejected. Id. at 231-32, 120 S.Ct. 727.
. This inconsistency was previously noted in Mancuso v. Olivarez, 282 F.3d 728, 737 n. 4 (9th Cir.), as amended 292 F.3d 939 (2002), and the court there attempted to clarify the issue. Our analysis here is not inconsistent with Mancuso.
. To the extent that they are inconsistent with this opinion, we overrule the statements in Rodriguez, 125 F.3d at 744; Franklin, 122 F.3d at 1273; and Thomas v. Hubbard, 273 F.3d 1164, 1170 (9th Cir.2002) (as amended) that appear to place the burden on the petitioner to establish that there was harm under Brecht.
. The dissent questions the sincerity of Pay-ton’s religious beliefs, calling his conversion a "miracle on the cellblock” and a "fortuitous epiphany.” Infra, at 10787, 10793. The testimony in mitigation permits a different inference. Payton's pastor testified that as a high school student Payton involved himself with a church group for several years. Reinitiating contact with the church after his arrest is consistent with his actions as a high school youth. Ultimately, resolving the question of the depth of Payton’s beliefs demands the kind of sifting and weighing of the evidence that is the jury's exclusive realm. Skipper, 476 U.S. at 9, 106 S.Ct. 1669 (remanding for new penalty phase trial when exclusion of post-crime mitigating evidence "impeded the sentencing jury’s ability to carry out its task of considering all relevant facets of the character and, record of the individual offender”). We steer clear of determining the value of the evidence in favor of ensuring that the jury had the opportunity to decide for itself whether Payton's religious beliefs were merely "fortuitous.”