Partial Concurrence and Partial Dissent by Chief Judge THOMAS;
Partial Concurrence and Partial Dissent by Judge CHRISTEN
OPINION
CLIFTON, Circuit Judge:Petitioner Eric Owen Mann lured Richard Alberts and Ramon Bazurto to his house in 1989 with a promise to sell them cocaine for about $20,000. Instead, he took the money and shot both men to death. Mann was eventually arrested and tried in Arizona state court in 1994. A jury found him guilty on two counts of first-degree murder, and the trial judge sentenced him to death, noting his long criminal history and his apparent lack of remorse for the killings. Following the affirmance of his convictions and sentence by the Arizona Supreme Court, Mann filed a petition for post-conviction relief in state court in 2000 asserting, among other claims, that he had received ineffective assistance of counsel at trial and sentencing. That petition was assigned to the same judge who had presided over Mann’s trial and imposed the capital sentence upon him a few years before. The judge denied in full Mann’s petition for post-conviction relief, and the Arizona Supreme Court thereafter denied Mann’s petition for review. Mann then filed a petition for habeas corpus under 28 U.S.C. § 2254 in federal district court. That petition was denied, and Mann appeals.
Of primary importance here is Mann’s argument that the state post-conviction court applied the wrong standard to his claim of ineffective assistance of counsel at sentencing, in violation of Supreme Court precedent in Strickland v. Washington, 466 U.S. 668, 104 S.Ct. 2052, 80 L.Ed.2d 674 (1984). Because of this error, Mann argues, our review of that claim should not be constrained by the Antiterrorism and *1147Effective Death Penalty Act of 1996 (AEDPA), which requires federal courts to defer to state court decisions on the merits of habeas corpus claims unless they are contrary to clearly established federal law. 28 U.S.C. § 2254(d). After careful review of the underlying state court decision, we conclude that its invocation of the Strickland prejudice standard might have been ambiguous but was not clearly incorrect. As a result, Mann’s petition presents us with the question: When, in the absence of clarity, can we conclude that a state court has applied the wrong standard to review a Strickland claim raised in a petition for post-conviction relief? The answer is that we must approach this question with the deference AEDPA requires and grant ha-beas relief only if no fairminded jurist could conclude that the adjudication was consistent with clearly established Supreme Court precedent. See Harrington v. Richter, 562 U.S. 86, 101, 131 S.Ct. 770, 178 L.Ed.2d 624 (2011).
Here, fairminded jurists could conclude that the state court’s review of Mann’s claim of ineffective assistance of counsel comported with Strickland. Applying AEDPA deference to Mann’s claims, we conclude that the state post-conviction court’s denial of post-conviction relief was not contrary to, or an unreasonable application of, federal law. We therefore affirm the district court and deny Mann’s petition for habeas relief.
I. Background
A. The murders
In the fall of 1989, a few weeks before Thanksgiving, Mann arranged to sell Richard Alberts a kilogram of cocaine in exchange for about $20,000. Mann never intended to provide Alberts with the drugs. Instead, he planned to hand Alberts a shoebox full of newspaper and take his money in exchange. He knew he would have to kill Alberts after this transaction took place.
On the evening of Thanksgiving Day, Alberts arrived at Mann’s house to complete the sale, accompanied by another man, Ramon Bazurto. Mann’s girlfriend, Karen Miller, was also at the house to provide Mann with backup. Miller testified that when Mann saw Alberts had brought along someone else, he became upset, but after a brief hesitation decided he would nevertheless go through with his plan to rip Alberts off. The three men and Miller went to Mann’s bedroom, where Alberts handed Mann the money and Mann gave Alberts the newspaper-filled shoebox. When Alberts lifted the top of the box, Mann shot him and then Bazurto.
Alberts died almost instantly from a shot through the heart. Bazurto took longer to die. He fell onto his back and lay on the ground, attempting to move his hand towards the pistol he was carrying in his waistband. According to Miller’s testimony, Mann stood over Bazurto’s prone body and described what was happening as Ba-zurto lost motor control and died. Miller testified that Bazurto continued to move for three to five minutes after being shot.
Mann then called a friend, Carlos Alejandro, to help him dispose of the two bodies. They dumped the bodies on the side of a gravel road. The next day, Mann and Miller cleaned the apartment and patched the walls where the bullets that had killed Alberts and Bazurto had entered. Mann gave Alberts’s car to a friend and disposed of his guns and the recovered bullets. When police later questioned Mann as to the whereabouts of Alberts and Bazurto, Mann responded that the two men had come to his house on the evening of the murders to conduct a drug deal, but had left after Mann and Alberts failed to agree on a price for the drugs.
*1148The ease remained unsolved for four years. Then, in late 1993, Miller ended her relationship with Mann because of increasing domestic violence. She told the police about the murders in January of 1994. Mann was arrested after the police corroborated Miller’s story with Alejandro and the man to whom Mann had sold Alberts’s car.
B. Trial
Mann’s jury trial took place over five days in the . Superior Court of Pima County, Arizona. Judge John F. Kelly presided. David Sherman, a criminal defense attorney in private practice, was appointed as Mann’s counsel. Sherman pursued the theory that Mann had acted in self-defense.
The day before trial commenced, Sherman told the judge that he would not be calling any witnesses “except for Mr. Mann.” Sherman successfully persuaded the trial court that, should Mann testify, his previous conviction for burglary would be excluded from evidence and any mention of his previous conviction for felony possession of a firearm would be limited to the fact that he had been convicted of a felony.
At trial, the state presented testimony from six witnesses, including Miller and Alejandro, both of whom had been granted immunity from prosecution for their roles in the crime. Miller and Alejandro testified that the murders had been premeditated. Mann ultimately did not testify, and Sherman did not call any witnesses of his own, instead attempting to discredit the state’s witnesses through cross-examination. The jury found Mann guilty of two counts of first-degree murder on November 1, 1994.
Two weeks later, the court granted Sherman’s request that Mann undergo psychological evaluation prior to sentencing. Sherman recommended that the court appoint Dr. Todd Flynn, a psychologist with the court psychological clinic. With the consent of the attorney for the state, the judge agreed to appoint Dr. Flynn to perform an evaluation.
Dr. Flynn diagnosed Mann with alcohol abuse or dependence, polysubstance abuse or dependence, and antisocial personality disorder. He also noted that Mann had scored highly for “acting out” on the Minnesota Multiphasic Personality Inventory-2 scale, and made the following observation:
As a group, people with acting out as a primary psychological defense tend to be emotionally shallow, deficient in their capacity for empathy, poor at frustration tolerance, unwilling or unable to delay gratification, and poor at anticipating the consequences of future actions. When combined with a history of victimization of others, this group becomes the worst of offenders because there is little sense of conscience or inhibition to stop the criminal lifestyle. Those who act out intensively and in a criminal manner may also qualify for the designation of Psychopath ....
The report concluded, “It is more probable than not that Mr. Mann fits this designation.”
The state’s presentence report concluded that three aggravating factors were present in the crime: it was committed for pecuniary gain, there were two victims, and Bazurto’s murder was committed “in an especially heinous, cruel or depraved manner.” Sherman argued for ten mitigating factors: Mann’s positive relationship with his daughters, his positive influence on his mother, his unstable and abusive family background, his poor educational experience, his history of substance abuse, his remorse, his cooperation with authorities, his non-violent history, his good behavior while incarcerated, and the disparity between his sentence and Miller and Alejandro’s sentences.
*1149At the sentencing hearing, which took place on January 31, 1995, four witnesses testified in support of Mann: a former boss, a former co-worker, his older daughter, and his mother. Mann also submitted a handwritten autobiography.
Judge Kelly found that the state had demonstrated all three of the aggravating circumstances they had identified: (1) the crime was committed for pecuniary gain, (2) there were two victims, and (3) Bazur-to’s murder was committed “in an especially heinous, cruel or depraved manner.”
As for mitigating factors, he found that Mann had a positive relationship with his daughters and his mother, an abusive and unstable family background, a poor educational experience, a history of substance abuse, good conduct while incarcerated, and recent stable employment. The judge declined to find that Mann had cooperated with the authorities, that he had a nonviolent history, or that a capital sentence would be disproportionate to sentences imposed on Miller and Alejandro.
The judge concluded that Mann, “by his actions for a period of four years ... showed no indication of any remorse, particularly his actions as introduced at trial after the murders.” The court noted that in his autobiography Mann described getting into a car accident in which two people died, but observed that “he indicates no remorse in that situation.” Ultimately, the judge concluded: “[T]he defendant is not capable of remorse. The psychological evaluation is that he has an anti-social personality disorder and that he is a psychopath. Basically, that he has no conscience. The Court believes that his expression[s] of remorse are not genuine.” The judge sentenced Mann to death on both counts of murder.1
Mann appealed to the Arizona Supreme Court, which affirmed. State v. Mann, 188 Ariz. 220, 934 P.2d 784 (1997) (en banc). The United States Supreme Court denied Mann’s petition for a writ of certiorari. Mann v. Arizona, 522 U.S. 895, 118 S.Ct. 238, 139 L.Ed.2d 169 (1997) (mem.).
C. Post-conviction relief proceedings
Mann filed for post-conviction relief under Rule 32 of the Arizona Code of Criminal Procedure. He alleged, among other claims, that he had been denied effective representation of counsel at trial and sentencing and that newly discovered evidence probably would have changed the verdict at the trial or sentencing. Among the arguments in his post-conviction brief, Mann claimed that Sherman “did little mitigation investigation or preparation for the mitigation hearing.” The brief described Mann’s 1985 automobile accident in great detail, including his remorse at the deaths of his two passengers, the injuries to his leg and head, and the changes in his personality and circumstances that resulted.
Mann also filed two additional expert reports: one from Dr. Richard Hinton, a clinical psychologist, and one from Dr. James Comer, a clinical neuropsychologist. Dr. Hinton’s report concluded:
Despite Mr. Mann’s extremely difficult childhood and his relatively long pedigree as a violator of the law, it does appear that his functioning changed dra*1150matically following the automobile accident. At that time he experienced a severe injury to his leg which caused him to be unable to work and which resulted in mounting debt. It also appears likely that he was tormented by a sense of guilt that he was somewhat responsible for the death of the two passengers in his car. From that time forward he began to behave much more aggressively and to use cocaine much more regularly.
Dr. Comer’s report concluded that “Mann appears to have experienced a T[ramuatic] B[rain] I[njury] leading to subtle though lasting cognitive deficits and other symptoms of post-concussional syndrome.”
The post-conviction proceedings were assigned to Judge Kelly, the same judge who had presided over Mann’s trial and sentencing. Judge Kelly held several evi-dentiary hearings, in which he heard testimony from Miller,2 Mann, Sherman, and Dr. Comer. The bulk of the post-conviction factfinding concerned the 1985 car crash. Miller testified that “[tjhings took a drastic turn” after the accident, that the doctors who examined Mann after the accident were “very concerned that he had had some head injuries,” and that he became “very depressed.” In particular, she said, the information that Mann’s 14-year-old passenger had been decapitated in the accident “really tore him up” and “many times he said he wished he had died in the accident too.” Mann testified that he had “suffered some type of concussion” as a result of the accident and that the accident “was a huge burden emotionally, physically, the way it affected friends, family.” Sherman testified that he had not learned of the severity of the accident until after Mann’s conviction and that he was “not focused on mitigation” during his representation.
Dr. Comer testified that some of Mann’s test results suggested “that he may have experienced a head injury, from which he recovered rather well in most respects, but which had left [him] with some residual, subtle cognitive deficits.” Dr. Comer noted that it was the “typical course of events” for the effects of a brain injury to be more severe closer in time to the accident, and that behavioral changes stemming from a brain injury generally “take the form of increased aggression, sexual indiscretions, poor ability to monitor one’s behavior, difficulty in appreciating the effect of one’s behavior on other individuals, increased irritability, agitation at times, [and increased] egocentrieity.” He also noted that the records that he reviewed “certainly indicated that there was a significant behavioral and emotional change, perhaps some change in personality as well” that occurred after the 1985 accident. He clarified that any one of the test results “could easily have been produced by other factors as well,” but that the significant majority of the test results “fell into that cluster of tests that are known to be sensitive to subtle, longlasting effects” from brain damage.
Judge Kelly responded to Dr. Comer’s testimony by commenting that it was “not clear how such a brain injury could have a role” in Mann’s crimes, as they had been planned out in advance without “any indication ... that the murders were committed out of anger or irritability.” Dr. Comer replied that he was not stating a finding that Mann’s head injury caused him to commit the crimes but rather that “some of the behavior we see postconcussion could produce the disinhibitory aggression” that followed Mann’s accident.
*1151Judge Kelly denied Mann’s Rule 32 petition in full in an order that will be discussed in more detail below. Subsequently, the Arizona Supreme Court summarily denied Mann’s petition for review on all issues except for the jury determination of aggravating factors, which is not at issue in this appeal. Mann then filed a federal habeas petition in the District of Arizona. The district court denied the petition but issued a Certificate of Appealability for three claims: that the Arizona Supreme Court violated Mann’s Eighth Amendment rights in reviewing his death sentence, that he received ineffective assistance of counsel at trial, and that he received ineffective assistance of counsel at sentencing. Mann has limited his appeal to the two claims of ineffective assistance of counsel.
A three-judge panel of this court unanimously affirmed the denial of habeas relief for ineffective assistance of counsel at trial, but, over a dissent, reversed the denial of habeas relief for ineffective assistance of counsel at sentencing. Mann v. Ryan, 774 F.3d 1203 (9th Cir. 2014). A majority of the nonrecused active judges on our court subsequently voted to rehear the case en banc. Mann v. Ryan, 797 F.3d 654 (9th Cir. 2015) (mem.). We affirm the district court’s denial of habeas relief on the ineffective assistance of counsel claims at both trial and sentencing.
II. Legal Standards
A. Antiterrorism and Effective Death Penalty Act
Because Mann filed this petition for habeas corpus after April 24, 1996, AEDPA applies to his claims. See Lindh v. Murphy, 521 U.S. 320, 326, 117 S.Ct. 2059, 138 L.Ed.2d 481 (1997). Under that statute we may not grant an application for habe-as corpus for any claim adjudicated on the merits in state court unless the state court adjudication:
(1) resulted in a decision that was contrary to, or involved an unreasonable application of, clearly established Federal law, as determined by the Supreme Court of the United States; or
(2) resulted in a decision that was based on an unreasonable determination of the facts in light of the evidence presented in the State court proceeding.
28 U.S.C. § 2254(d). We review the last reasoned state court decision according to this deferential standard. Towery v. Ryan, 673 F.3d 933, 944 (9th Cir. 2012) (per curiam). The district court’s application of AEDPA to the last reasoned state court decision is a mixed question of law and fact which we review de novo. Lambert v. Blodgett, 393 F.3d 943, 965 (9th Cir. 2004).
An adjudication is “contrary to” clearly established Supreme Court precedent “if the state court arrives at a conclusion opposite to that reached by [the Supreme] Court on a question of law or if the state court decides a case differently than [the Supreme] Court has on a set of materially indistinguishable facts.” Williams v. Taylor, 529 U.S. 362, 412-13, 120 S.Ct. 1495, 146 L.Ed.2d 389 (2000). It is an “unreasonable application” of clearly established Supreme Court precedent “if the state court identifies the correct governing legal principle from [the Supreme] Court’s decisions but unreasonably applies that principle to the facts of the prisoner’s case.” Id. at 413, 120 S.Ct. 1495. “[A]n unreasonable application of federal law is different from an incorrect application of federal law.” Id. at 410, 120 S.Ct. 1495. The federal habeas court “may not issue the writ simply because that court concludes in its independent judgment that the relevant state-court decision applied clearly established federal law erroneously or incorrectly.” Id. at 411, 120 S.Ct. 1495.
A state court’s adjudication is unreasonable only if the federal habeas court *1152concludes that no fairminded jurist could conclude that the adjudication was consistent with established Supreme Court precedent. See Richter, 562 U.S. at 102, 131 S.Ct. 770. In that event, AEDPA does not require deference to the state court’s decision, and the petitioner’s claims are reviewed de novo. See Williams, 529 U.S. at 406, 120 S.Ct. 1495.
B. Ineffective assistance of counsel
Mann’s claims are based on the alleged violation of his Sixth and Fourteenth Amendment rights to effective assistance of counsel during both the trial and penalty phases of his state court proceedings. Ineffective assistance of counsel claims are evaluated according to the familiar standard set forth in Strickland. To receive relief under this standard, “[flirst, the defendant must show that counsel’s performance was deficient.” 466 U.S. at 687, 104 S.Ct. 2052. “Second, the defendant must show that the deficient performance prejudiced the defense.” Id. “Unless a defendant makes both showings, it cannot be said that the conviction or death sentence resulted from a breakdown in the adversary process that renders the result unreliable.” Id.
The proper standard for attorney performance under the first prong of the Strickland test is “that of reasonably effective assistance.” Id. “When a convicted defendant complains of the ineffectiveness of counsel’s assistance, the defendant must show that counsel’s representation fell below an objective standard of reasonableness.” Id. at 687-88, 104 S.Ct. 2052. “Judicial scrutiny of counsel’s performance must be highly deferential,” and “a court must indulge a strong presumption that counsel’s conduct falls within the wide range of reasonable professional assistance.” Id. at 689, 104 S.Ct. 2052.
The second prong of the Strickland test requires that the defendant show “that there is a reasonable probability that, but for counsel’s unprofessional errors, the result of the proceeding would have been different.” Id. at 694, 104 S.Ct. 2052. “A reasonable probability is a probability sufficient to undermine confidence in the outcome.” Id. “It is not enough for the defendant to show that the errors had some conceivable effect on the outcome of the proceeding.” Id. at 693, 104 S.Ct. 2052.
III. Ineffective Assistance of Counsel at Trial
Mann first argues that he received ineffective assistance of counsel during the trial phase of his proceedings. This argument is based on the alleged negligence of Mann’s counsel, Sherman, in (1) failing to recommend that Mann testify at trial, and (2) breaking an “implicit” promise to the jury that Mann would testify in his own defense. Because the Arizona Supreme Court summarily denied the petition for review of this claim, we look through that decision to the decision of the state post-conviction court. See Towery, 673 F.3d at 944.
A. Sherman’s recommendation that Mann not testify at trial
Mann and Sherman presented conflicting accounts of their relationship to the state post-conviction court. Sherman testified that Mann told him “early on in th[e] case” that the murders had been premeditated. Despite this knowledge, Sherman advised Mann that a self-defense theory “would be the best way to go.” Sherman said he gave Mann the option to testify at trial that he had acted in self-defense, but told Mann he “would have to withdraw as his attorney” if Mann did so because Sherman knew this testimony would be false. In a letter to Mann approximately one month before trial, Sherman laid out conflicting considerations for Mann to weigh. *1153On the one hand, Mann was the only witness who would be able to testify to his story, which was that Bazurto had reached for his gun and Mann reacted without thinking. On the other hand, Mann could be cross-examined about his previous criminal activity, as well as allegations of abuse from his former girlfriends. The letter concluded: “I’d hate to have you on death row in Florence having a big regret because you did not testify. When the stakes are this high, I have to leave the final decision up to you.”
Mann, in contrast, testified that he told Sherman from the beginning that he had killed Alberts and Bazurto in self-defense. Mann said that he and Sherman “briefly” discussed the possibility that Mann would testify at trial, but that Sherman told Mann that “it would be bad” for him to testify because it would open the door to questions about his prior conviction for felony possession of a weapon. Mann said that he took his attorney’s advice and did not testify.
The state post-conviction court concluded that Sherman was more credible than Mann and accepted Sherman’s version of the facts as the truth.
Our review of the state habeas court’s credibility determinations is highly deferential. Marshall v. Lonberger, 459 U.S. 422, 434, 103 S.Ct. 843, 74 L.Ed.2d 646 (1983) (“28 U.S.C. § 2254(d) gives federal habeas courts no license to redetermine credibility of witnesses whose demeanor has been observed by the state trial court....”).
Mann points to some possible inconsistencies in Sherman’s story: Sherman did not mention in his letter that he would withdraw from representation if Mann testified, and he continued to prepare for the possibility that Mann might testify up until the day before trial. Sherman had plausible explanations, however. Sherman said he never told Mann he couldn’t testify, but rather that “if he cho[s]e to do it against my recommendation, I would have to withdraw as his attorney and he could get another attorney to do that, but I wasn’t going to put him on the stand.” As for the letter, Sherman testified that he was “finessing the situation” and did not want to put the potential perjury in writing. Given these considerations, and Mann’s obvious motivation to lie, it is impossible to conclude that the state court’s findings as to Mann and Sherman’s relative credibility “lacked even ‘fair[] support’ in the record.” Id. at 432, 103 S.Ct. 843.
Once the state court had found that Sherman was a more credible witness than Mann, it was effectively bound by Supreme Court precedent to conclude that Sherman’s decision to discourage Mann from testifying did not constitute ineffective assistance of counsel. Counsel’s duty to advocate for his client is “[p]lainly ... limited to legitimate, lawful conduct compatible with the very nature of a trial as a search for truth.” Nix v. Whiteside, 475 U.S. 157, 166, 106 S.Ct. 988, 89 L.Ed.2d 123 (1986). “Although counsel must take all reasonable lawful means to attain the objectives of the client, counsel is precluded from taking steps or in any way assisting the client in presenting false evidence or otherwise violating the law.” Id. Sherman would have been suborning perjury if, knowing that the murders had been premeditated, he had allowed Mann to testify in support of the theory of self-defense. Under such circumstances, there can be “no failure to adhere to reasonable professional standards that would in any sense make out a deprivation of the Sixth Amendment right to counsel.” Id. at 171, 106 S.Ct. 988.
B. Testimony promised in opening statement
Mann’s claim that Sherman broke an implicit promise to the jury that Mann *1154would testify in his own defense is similarly unavailing. Had Sherman made such a promise, and then broken it, Mann’s claim might have merit. Several circuits have held that it may violate Strickland for counsel to promise evidence in an opening statement and then fail to present that evidence at trial. See, e.g., United States ex rel. Hampton v. Leibach, 347 F.3d 219, 258 (7th Cir. 2003) (finding that it was “objectively unreasonable” for counsel to promise the defendant’s testimony and then break that promise); Ouber v. Guarino, 293 F.3d 19, 27 (1st Cir. 2002) (describing counsel’s decision to discourage the defendant from taking the stand as “indefensible” when he had already represented to the jury that the defendant’s testimony would be the centerpiece of her defense); McAleese v. Mazurkiewicz, 1 F.3d 159, 166 (3d Cir. 1993) (noting generally that an attorney’s failure to produce evidence promised during an opening statement is “sufficient of itself to support a claim of ineffectiveness of counsel”).
But Sherman did not promise the jury that Mann would testify. His opening statement detailed the alternative version of evénts that formed the basis for his defense strategy: he told the jury that Mann had cocaine he intended to sell to Alberts, and that as Mann was handing over the drugs to Alberts, Bazurto “did something that made [Mann] think he was about to be the victim of a drug rip-off.” In reaction, Mann “fired at both people and killed them both because he felt like he was forced to do so.” Sherman then said, “Now, look at what the facts are. After that we’ll see [Mann] didn’t plan this.” He told the jury to anticipate the testimony of Miller and Alejandro, which he described as “the only evidence that the [state will] have of premeditation,” and argued that “[w]hen you compare that to the whole story, it doesn’t make any sense that if it was planned it would be done this way.”
At trial, Sherman successfully elicited testimony from Miller and Alejandro that was inconsistent with the prosecution’s theory that the murders were premeditated. For instance, he noted that Miller had testified to standing directly behind Bazur-to when he was killed, which she admitted was “awfully close to what the line of fire was going to be” for someone who allegedly knew Bazurto was about to be shot. Miller also testified that there had been no plan to dispose of the bodies. Alejandro testified that the first time he heard anything about Mann’s plan was when Mann showed up at his house the night of the crime asking for help. He described Mann as having been “[i]n desperate need,” and said that the first thing Mann said to him that night was, “I did it. I had no choice. I had to do it.” In other words, Sherman’s opening argument “merely summarized evidence that was later produced from which a jury could be left with a reasonable doubt” about whether the crimes were premeditated. McAleese, 1 F.3d at 167. This may not have been a successful trial strategy, but it was not improper.
IV. Ineffective Assistance of Counsel at Sentencing
Mann also argues that the state court erred in declining to grant habeas relief based on his claim that he had received ineffective assistance of counsel at the sentencing stage of his proceedings. His argument before the state post-conviction court focused on two alleged failings: (1) failure to present reasonably available mitigation evidence at sentencing, and (2) failure to seek adequate assistance in evaluating Mann’s mental health. The last reasoned state court decision on this issue came from the post-conviction court, so that is the decision we review. See Towery, 673 F.3d at 944.
*1155Under Strickland, Mann is eligible for post-conviction relief if he can show both (1) deficient performance, and (2) prejudice stemming from the deficient performance. 466 U.S. at 687, 104 S.Ct. 2052. When addressing Sherman’s alleged failure to present reasonably available mitigation evidence at sentencing, the post-conviction court did not consider whether Sherman’s performance was deficient. Rather, it elected to move directly to the second prong, as it was entitled to do, and found only that there had been no prejudice. Because the prejudice issue has been “adjudicated on the merits,” AEDPA requires us to allow the state court’s determination to stand unless it is contrary to or an unreasonable application of federal law or is based on an unreasonable determination of the facts. Richter, 562 U.S. at 98, 131 S.Ct. 770.
On appeal, Mann argues that the state post-conviction court’s analysis of prejudice was contrary to clearly established federal law. He also argues that, on the merits, Sherman’s performance caused Mann prejudice. We address these arguments in order because our evaluation of the first determines the standard of review to be applied to the second. First, we assess whether the state court’s decision was contrary to clearly established federal law as determined by the Supreme Court. Second, we assess whether Sherman’s performance caused Mann prejudice. If we conclude at the first step that the state court misapplied clearly established federal law, then our review at the second step is de novo. See Williams, 529 U.S. at 406, 120 S.Ct. 1495. If, however, the court’s interpretation of the law was permissible, then AEDPA governs our review at the second step as well, and we are required to defer to the state court’s conclusion that Mann suffered no prejudice from his counsel’s alleged failings unless that finding was unreasonable.
A. Applicability of AEDPA
Mann argues that we should review his Strickland claim de novo because the state post-conviction court made three different legal decisions that were contrary to or an unreasonable application of clearly established federal law. One, he argues that the state post-conviction court applied the wrong standard in assessing prejudice under Strickland. Two, he argues that the state court improperly excluded the mitigating evidence presented during post-conviction proceedings because Mann had not demonstrated a causal link between the evidence and the crime. Three, he argues that the court failed to re-weigh the evidence in aggravation against the totality of the mitigation evidence, including evidence introduced during the sentencing and during post-conviction proceedings. None of these three allegations gives us a basis to conclude that AEDPA does not apply to Mann’s claims.
i. Strickland’s prejudice standard
The common standard for a motion for a new trial based on newly discovered evidence requires a defendant to “show that counsel’s deficient conduct more likely than not altered the outcome in the case.” Strickland, 466 U.S. at 693-94, 104 S.Ct. 2052. The Supreme Court considered adopting this standard for the Strickland prejudice inquiry but determined that it was “not quite appropriate.” Id. at 694,104 S.Ct. 2052. Instead, the Court held that a defendant alleging ineffective assistance of counsel “must show that there is a reasonable probability that, but for counsel’s unprofessional errors, the result of the proceeding would have been different.” Id. This sets the bar a bit lower but not a lot lower. “[T]he difference between Strickland’s prejudice standard and a more-probable-than-not standard is slight and matters ‘only in the rarest case.’ ” Richter, *1156562 U.S. at 112, 131 S.Ct. 770 (quoting Strickland, 466 U.S. at 697, 104 S.Ct. 2052).
The claims presented by Mann in the state post-conviction proceedings at issue here included an argument that new mitigation evidence related to the 1985 accident merited a new sentencing. Those claims were identified as “Issues Three and Four.”3 Mann’s claim that Sherman’s failure to raise the evidence at the original sentencing constituted ineffective assistance of counsel was described as “Issue Seven.” Judge Kelly denied relief on both claims.
As to Issues Three and Four, he found: The record shows that Defendant provided this Court with evidence of the accident and its effects in the form of an autobiography he presented to show mitigation. Nonetheless, additional evidence, particularly that set forth in psychological reports, was submitted as part of this Petition. The Court finds, however, that nothing presented would have changed the verdict or the sentence imposed. The Court finds that Defendant has not proven the existence of a causal connection between the accident and its effects and the murders. The record shows, for example, that Defendant dealt drugs and used guns against others before and after the accident; that he did not misperceive a threat and overreact, but, rather, carefully planned to kill Alberts, then “made a choice, after a period of thought and said, ‘Well, I’ve got to do it,’ apparently meaning that to go through .with the plan he would also have to murder Bazurto.” Defendant committed these murders for pecuniary gain, not for reasons traceable to his 1985 accident. Accordingly, the claims are denied.
(emphasis added) (citation omitted).
Meanwhile, as to Issue Seven, Judge Kelly found:
To succeed on this claim and on the others that follow below, Defendant must show that: (1) counsel’s performance was deficient, as defined by prevailing professional norms; and (2) the deficient performance resulted in prejudice to the defense. State v. Nash, 143 Ariz. 392, 694 P.2d 222 (1985). Here, Defendant has failed to show prejudice. Some of the evidence he accuses Sherman of not presenting, e.g., testimony from family members,-was in fact presented to this Court at sentencing. Additional evidence that pertains to the 1985 accident and its effects is discussed under “Issues Three and Four” above, where this Court found that it would not have changed the sentence imposed. For that reason, Defendant was not prejudiced by counsel’s performance and the claim is denied.
Mann argues that the state post-conviction court’s conclusion that the new evidence “would not have changed the sentence imposed,” combined with the court’s ' reference to the analysis for “Issues Three and Four,” suggests that it applied the wrong standard in analyzing Mann’s ineffective assistance of counsel claim. Rather than finding that there was no reasonable probability that the new evidence would have changed the sentence imposed, Mann argues, the post-conviction court imported the standard from its analysis of the motion for a new trial into its analysis of the ineffective assistance of counsel claim and concluded only that the new evidence more *1157likely than not would not have changed the sentence imposed.
Had the state post-conviction court applied the more-likely-than-not standard to analyze Mann’s ineffective assistance of counsel claims, its opinion would have been contrary to clearly established federal law under AEDPA. The Supreme Court considered and explicitly rejected the more-likely-than-not standard in Strickland. 466 U.S. at 693-94, 104 S.Ct. 2052. But the standard the court applied is not clear on the surface of its decision. Judge Kelly simply said that the additional information about Mann’s accident “would not have changed the sentence imposed.” This leaves us to draw inferences as to the precise standard the court used in reaching the decision to deny each of Mann’s claims.
Reading the opinion as a whole, the more logical inference here is that the standard used by the state post-conviction court was the correct reasonable probability standard. While the court did not clearly state the standard it applied, either for the newly discovered evidence claim or for the ineffective assistance of counsel claim, it indicated that the ineffective assistance of counsel claim was controlled by State v. Nash, 143 Ariz. 392, 694 P.2d 222 (1985), in which the Arizona Supreme Court adopted the reasonable probability standard.
Judge Kelly also made various factual findings that suggested he was deeply skeptical that evidence about the accident had any significant impact on Mann’s mitigation profile. Most importantly, he noted that Mann dealt drugs and used guns prior to the accident, premeditated both murders, and committed them for pecuniary gain. Taken in context, the judge’s discussion of Issues Three, Four, and Seven suggests that he knew the proper standard for both the new evidence claim and the ineffective assistance of counsel claim, but concluded that the new evidence made at most a minimal difference, such that the standard applied would have no effect on the outcome of either claim. This inference is particularly strong because Judge Kelly was the same judge who had handed down Mann’s initial sentence, giving him particular insight into how the additional evidence would, or would not, have changed Mann’s mitigation profile. See Schriro v. Landrigan, 550 U.S. 465, 476, 127 S.Ct. 1933, 167 L.Ed.2d 836 (2007) (noting that the judge presiding over the post-conviction review was “ideally situated” to interpret an interaction that had taken place in the trial court because she had also presided over the trial). Because of that fact, it is not surprising that Judge Kelly would have spoken in direct terms — “nothing presented would have changed the verdict or the sentence imposed” — rather than in words designed to express degrees of probability.
At worst, the state court opinion is ambiguous. But, as we have noted in the past, “[u]nder AEDPA, we must do more than find the statement ambiguous.” Lopez v. Schriro, 491 F.3d 1029, 1037-38 (9th Cir. 2007). Our analysis starts with a “presumption that state courts know and follow the law.” Woodford v. Visciotti, 537 U.S. 19, 24, 123 S.Ct. 357, 154 L.Ed.2d 279 (2002). Next, we “determine what arguments or theories supported or ... could have supported[] the state court’s decision.” Richter, 562 U.S. at 102, 131 S.Ct. 770. Finally, we ask “whether it is possible fairminded jurists could disagree that those arguments or theories are inconsistent with the holding in a prior decision of th[e Supreme] Court.” Id. Here, it is possible to read the state court’s decision in a way that comports with clearly established federal law. The judge, who had imposed the sentence originally and who was already familiar with Mann’s mitigation profile, concluded that the additional information provided at the post-*1158conviction hearing would have made little difference. Accordingly, he denied Mann’s petition for post-conviction relief. Under AEDPA, because we can read the decision to comport with clearly established federal law, we must do so. We cannot conclude • that the state post-conviction court’s application of Strickland misunderstood or misapplied clearly established federal law.
ii. Causal nexus
Mann also argues that the state post-conviction court violated clearly established federal law when it improperly refused to consider evidence about Mann’s car accident because he had “not proven the existence of a causal connection between the [1985 car] accident and its effects and the murders.” This argument relies on Eddings v. Oklahoma, in which the Supreme Court held that the “Eighth and Fourteenth Amendments require that the sentencer ... not be precluded from considering, as a mitigating factor, any aspect of a defendant’s character or record and any of the circumstances of the offense that the defendant proffers as a basis for a sentence less than death.” 455 U.S. 104, 110, 102 S.Ct. 869, 71 L.Ed.2d 1 (1982) (quoting Lockett v. Ohio, 438 U.S. 586, 604, 98 S.Ct. 2954, 57 L.Ed.2d 973 (1978)). “The sentencer, and the Court of Criminal Appeals on review, may determine the weight to be given relevant mitigating evidence. But they may not give it no weight by excluding such evidence from their consideration.” Id. at 114-15, 102 S.Ct. 869.
Our court has concluded that the law in Arizona at the time of Mann’s post-conviction proceedings, in spite of this clear command, required that defendants demonstrate “actual causation” linking their alleged mitigating circumstances with the crime for which they were being sentenced. McKinney v. Ryan, 813 F.3d 798, 815 (9th Cir. 2015) (en banc). The causal nexus test provided that, “[i]f the defendant fails to prove causation, the circumstance will not be considered mitigating. However, if the defendant proves the causal link, the court then will determine what, if any, weight to accord the circumstance in mitigation.” State v. Hoskins, 199 Ariz. 127, 14 P.3d 997, 1022 (2000) (en banc). In Tennard v. Dretke, the Supreme Court held that a comparable causal nexus test violated defendants’ Eighth Amendment right requiring the finder of fact to “ ‘be able to consider and give effect to’ a capital defendant’s mitigating evidence.” 542 U.S. 274, 285, 124 S.Ct. 2562, 159 L.Ed.2d 384 (2004) (quoting Boyde v. California, 494 U.S. 370, 377-78, 110 S.Ct. 1190, 108 L.Ed.2d 316 (1990)).
Mann does not and cannot argue here that the state post-conviction court violated clearly established federal law by applying Arizona’s unconstitutional causal nexus test. Mann did not present a post-conviction Eddings claim before the Arizona Supreme Court, so we have no jurisdiction to grant relief on that issue. See Rose v. Lundy, 455 U.S. 509, 515, 102 S.Ct. 1198, 71 L.Ed.2d 379 (1982). Mann contends that what effectively amounts to an Eddings claim nevertheless comes in on the back of his Strickland claim, since Judge Kelly applied the causal nexus test to bar evidence about the effects of the accident and therefore did not properly consider whether Sherman’s failure to present that evidence at sentencing led to 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.
This argument fails to acknowledge that Judge Kelly’s comment about the lack of a causal connection between the accident and the murders was directly responsive to the evidence Mann presented before the state post-conviction court. Dr. Hinton’s *1159neuropsychological report concluded that “it does appear that [Mann’s] functioning changed dramatically following the automobile accident,” noting that “[f]rom that time forward he began to behave much more aggressively and to use cocaine much more regularly.” Dr. Comer’s evaluation concluded that “Mann’s functioning had changed dramatically after the [accident], as exemplified by increased aggression.” Meanwhile, Miller testified that, after the accident, Mann “began to intimidate people,” “liked to use power to scare people,” and “was ‘pointing guns to people’s heads and telling them ... don’t fuck with me [or] I’ll blow your head off.” Miller stated that she believed that “the change in [Mann] was a result of the accident, the trauma, and the cocaine use.” In his petition for post-conviction relief, Mann argued:
There is no question that this evidence, if presented to the Court in mitigation, could have resulted in a sentence other than death. As described by Karen Miller, the killing of Bazurto and Alberts was of a piece with the change that occurred in Petitioner as a result of the accident. It was a crime of power and money, and the clear result of the fundamental personality and character changes that occurred in Eric Mann as a result of the accident.
The dissent notes, at 44, that Mann argued at the post-conviction hearing that the new evidence as to the effects of the accident would have suggested a mitigating reason for some of his behavior and, “would have provided some freestanding mitigation of itself.” As we have noted in the past, however, courts generally find explanatory mitigation evidence more convincing than humanizing mitigation evidence. See Allen v. Woodford, 395 F.3d 979, 1006 (9th Cir. 2005). Perhaps recognizing this, Mann focused his attention on arguing that the circumstances and the neuropsychological effects of the accident explained his decision to commit murder. Thus, while the dissent may be correct that the post-conviction court did not discuss how Mann’s brain damage could have reduced his moral culpability or altered his sentencing profile apart from having contributed to the crime, that was not the core of the argument to which the post-conviction court responded. Mann’s substantive arguments at the post-conviction hearing were focused on the link between the accident and the murders. He argued that after the accident, “the nature of his problems and degree of his problems and degree of his behavior changed radically,” that Dr. Comer “concluded either because of the accident or the aftermath of the accident or his heavy drug use, it’s not unreasonable to conclude that Mr. Mann might have perceived a threat and that he overreacted,” and that Dr. Comer had “ma[d]e a connection” between the murders and “what people with the kind of injury ... Mann was determined to have ... ordinarily do.” The court could not have properly addressed these arguments without considering the causal relationship between the new evidence and the crimes.
In addition, it is clear elsewhere in the record that Judge Kelly did not exclude evidence from his mitigation assessment based solely on the lack of a causal nexus with the crime when that evidence was not presented as causal. For instance, he found during sentencing that Mann had demonstrated several mitigating factors unrelated to the murders, including that Mann had an “unstable and abusive family background,” a poor educational background, and a history of substance abuse.
As Eddings itself makes clear, it is well within the sentencer’s power to “determine the weight to be given relevant mitigating evidence.” 455 U.S. at 114-15, 102 S.Ct. 869. Here, Judge Kelly considered Mann’s argument that the accident led to a fundamental change in his personality that set *1160him on a path towards murder and found it unconvincing. This conclusion was entirely within the court’s power. As such, there is no indication that Judge Kelly excluded evidence that would have led to a reasonable probability of a different sentence in violation of clearly established federal law, and Eddings cannot provide a back door to de novo review.
iii. Re-weighing old and new evidence
Finally, Mann argues that the state court erred in failing to reweigh the old and new mitigation evidence against the existing aggravation evidence during post-conviction proceedings. Courts considering additional evidence in post-conviction proceedings must “evaluate the totality of the available mitigation evidence — both that adduced at trial, and the evidence adduced in the habeas proceeding in reweighing it against the evidence in aggravation.” Williams, 529 U.S. at 397-98, 120 S.Ct. 1495.
There is no indication that the state court did not fulfill this duty here. The court did not explicitly state that it had reweighed the evidence, but that is not the standard. Instead, “[i]t is sufficient that a sentencing court state that it found no mitigating circumstances that outweigh the aggravating circumstances.” Ortiz v. Stewart, 149 F.3d 923, 943 (9th Cir. 1998) (quoting Poland v. Stewart, 117 F.3d 1094, 1101 (9th Cir. 1997)). In this case, it was enough that the court found that the additional mitigation evidence Mann provided at the post-conviction proceedings was cumulative, not credible, or would otherwise have had no effect on the sentencing decision.
The dissent’s attempt to fold Mann’s claim of Eddings error into his claim that the state post-conviction court failed to properly re-weigh the evidence does not change this conclusion. As discussed above, the state court reasonably concluded that the additional evidence about Mann’s 1985 ear accident did not cause a fundamental change in his personality or set him on the path that led to Alberts’ and Bazurto’s murders, and therefore gave the new evidence little to no weight in mitigation. Therefore, it was not contrary to, and did not involve an unreasonable application of, federal law for the court to conclude that Mann had failed to show that the mitigating circumstances he had presented before the sentencing and post-conviction courts outweighed the aggravating circumstances of his crimes.
B. Reasonableness of state court’s prej-udicefinding
Even though Mann has not demonstrated that the state post-conviction court’s opinion should be subject to de novo review, he could still secure habeas relief under AEDPA if he could demonstrate that it was “unreasonable” for the state court to conclude that Mann was not prejudiced by his counsel’s alleged failings. See Richter, 562 U.S. at 101, 131 S.Ct. 770. A court ruling is unreasonable if it is “so lacking in justification that there was an error well understood and comprehended in existing law beyond any possibility for fairminded disagreement.” Id. at 103, 131 S.Ct. 770.
Under Strickland’s prejudice prong, the petitioner seeking habeas relief must show that there is a “reasonable probability that, but for counsel’s unprofessional errors, the result of the proceeding would have been different.” 466 U.S. at 694, 104 S.Ct. 2052. “A reasonable probability is a probability sufficient to undermine confidence in the outcome.” Id. It is conceivable that the evidence that emerged in the state post-conviction proceedings might have changed Mann’s profile somewhat had Mann’s attorney presented it at sentencing. For example, Mann notes that while the state sentencing court concluded that he was “incapable of remorse,” evidence *1161was presented to the state post-conviction court that supported the contention that he had in fact been deeply affected by the 1985 car accident in which two people died. Mann also notes that Dr. Flynn’s damning suggestion that he might be a psychopath failed to account for the possibility of a traumatic brain injury, and that further testing would have revealed that the injury significantly changed his personality, effectively turning a significant aggravating factor into a significant mitigating factor.
But, as noted above, even if the accident had an effect on Mann’s personality, it hardly changed an altar boy into a callous criminal. The presentence report submitted by the probation department attached a long criminal history record that included 14 offenses predating Mann’s 1985 car accident. While Mann’s earliest crimes were minor, they escalated over time. In 1980, Mann was sentenced to probation for third-degree burglary. In 1982, he was arrested after pointing a gun at the victim of an attempted car theft, at which point his probation for the burglary offense was revoked and he was sentenced to two years in prison. The presentence report also noted that Mann stole to support his family starting around age 15. Mann himself acknowledges that his pre-accident record was “troubling” and that the sentencing court had already been presented with “credible and potent mitigation” that it found outweighed by the heinous nature of Mann’s crime.
Given this history, it was not unreasonable for the state post-conviction court to conclude that the accident had no significant effect on Mann’s behavior. He had set on a path of violent crime long before the accident occurred. Nor was it unreasonable for the court to conclude that additional evidence of Mann’s good character was not enough to outweigh the various aggravating factors that the sentencing court had found: (1) the crime was committed for pecuniary gain, (2) there were two victims, and (3) Bazurto’s murder was committed “in an especially heinous, cruel or depraved manner.” Indeed, the aggravating factors remained unaffected by the new mitigating evidence.4 And, while the new evidence did suggest that Mann had been remorseful about the car accident, it did not undermine the sentencing court’s conclusion that Mann had shown no remorse for the cold-blooded murders of which he had been convicted. We cannot conclude that it was unreasonable for the state post-conviction court to find that the new evidence as to Mann’s capacity for remorse. and the 1985 car accident did not “undermine confidence in the outcome” of the sentencing. See Strickland, 466 U.S. at 694, 104 S.Ct. 2052.
Y. Conclusion
The judgment of the district court denying habeas relief as to both Mann’s convictions and his sentence is affirmed.
AFFIRMED.
. Until 2002, Arizona’s sentencing scheme in capital prosecutions allowed the trial judge to determine "the presence or absence of the aggravating factors required by Arizona law for imposition of the death penalty.” Ring v. Arizona, 536 U.S. 584, 588, 122 S.Ct. 2428, 153 L.Ed.2d 556 (2002). In that year, the Supreme Court held that this system violated the Sixth Amendment right to a jury trial. Id. at 609, 122 S.Ct. 2428. Two years later, in Schriro v. Summerlin, 542 U.S. 348, 358, 124 S.Ct. 2519, 159 L.Ed.2d 442 (2004), the Court clarified that the new rule set forth in Ring did not apply retroactively to cases already final on direct review.
. By the time of the post-conviction proceedings, Karen Miller had married and changed her name. For purposes of clarity and consistency, this opinion will continue to refer to her by her name at the time of Mann's trial.
. Mann did not receive a Certificate of Ap-pealability as to these claims and does not raise them on appeal. We discuss them here only because they are relevant to determining the standard that the state post-conviction relief court applied to the ineffective assistance'of counsel claim.
. Mann argues that Miller “admitted during the PCR proceedings!]] that neither victim suffered and that they died or became unconscious almost simultaneous with the shootings.” Had she said this, it could have undermined the sentencing court's conclusion that Bazurto's murder was especially cruel, heinous, or depraved. However, Miller actually said at the post-conviction hearing that she didn't know whether Bazurto’s movements after he was shot were voluntary or involuntary, that it was "possible” that Mann had not accurately described the last moments of Ba-zurto's life to her, and that it was "possible” that. he had been unconscious even though she believed him to have been awake and struggling. To the extent that her testimony before -the post-conviction court was different from her testimony at trial, the post-conviction court " discounted] it and continuefd] to believe the evidence she provided at trial.” As noted above, we defer to the reasonable credibility determinations made by the state post-conviction court. Lonberger, 459 U.S. at 434, 103 S.Ct. 843.