concurring in part and dissenting in part:
Once more unto the breach. Time and again, we have been admonished for disregarding Congress’s clear instruction that federal judges in habeas proceedings must adopt a “highly deferential standard” under which “state-court decisions [are] given the benefit of the doubt.” Woodford v. Visciotti, 537 U.S. 19, 24, 123 S.Ct. 357, 154 L.Ed.2d 279 (2002) (per curiam) (internal quotation marks omitted). In clear violation of this principle, the majority today seizes upon imprecise language in a single sentence of a state court’s otherwise well-reasoned and comprehensive opinion, and uses it to sweep aside AEDPA’s restrictions on the scope of our review. The majority not only fails to faithfully apply Supreme Court precedent, it also creates a split with two other circuits.
If we are not summarily reversed, Mann’s death sentence will surely be reimposed by the state court. One way or the other, Mann will be executed, if he doesn’t die of old age first. But only after he— and the families of the two people he killed 25 years ago — endure what may be decades of further uncertainty. Where’s the justice in that? I respectfully dissent from Part III of the majority’s opinion.
The majority’s conclusion rests upon a single phrase in the state court’s opinion: its statement that the new evidence intro*1224duced at the post-conviction proceeding “would not have changed the sentence imposed.” The majority infers from this ambiguous phrase that the state court was holding Mann to a more-likely-than-not standard, rather than a “reasonable probability” standard. Maj. Op. 1215. Never-mind that “the difference between Strickland’s prejudice standard and a more-probable-than-not standard is slight and matters only in the rarest case,” Harrington v. Richter, 562 U.S. 86, 131 S.Ct. 770, 792, 178 L.Ed.2d 624 (2011) (internal quotation marks omitted), the majority latches onto the distinction as its reason for “removing] AEDPA as a bar to relief.” Maj. Op. 1215.
We may not get around AEDPA quite so easily. First, we must apply “a blanket presumption that state judges know and follow the law.” Lopez v. Schriro, 491 F.3d 1029, 1046 (9th Cir.2007). This is particularly the case where, as here, the state judge cites to a case that articulates word-for-word the correct legal standard. Second, we must construe any ambiguity in language in the state court’s favor. See Visciotti, 537 U.S. at 24, 123 S.Ct. 357; see also Holland v. Jackson, 542 U.S. 649, 655, 124 S.Ct. 2736, 159 L.Ed.2d 683 (2004) (per curiam). We cannot “presume from an ambiguous record that the state court applied an unconstitutional standard.” Poyson v. Ryan, 743 F.3d 1185, 1199 (9th Cir.2014). Third, we cannot “demand a formulary statement” by the state court but must instead assess the “fair import” of its opinion. Early v. Packer, 537 U.S. 3, 9, 123 S.Ct. 362, 154 L.Ed.2d 263 (2002) (per curiam). The majority ignores each of these guiding principles, treating AED-PA as a straightjacket to be escaped, rather than a policy judgment to be obeyed. As the Supreme Court has made clear beyond peradventure, we may disturb a state court’s judgment in the sovereign administration of its criminal justice system only under the most exceptional circumstances. See Visciotti, 537 U.S. at 24, 123 S.Ct. 357; see also Holland, 542 U.S. at 655, 124 S.Ct. 2736. Federal “habeas corpus is a guard against extreme malfunctions ... not a substitute for ordinary error correction through appeal.” Richter, 131 S.Ct. at 786 (internal quotation marks omitted).
Where was the “extreme malfunction” here? The state court laid out Strickland’s two-prong framework, cited to a case explicitly articulating Strickland’s “reasonable probability” language and carefully explained the reasons why it did not find the evidence of organic brain injury persuasive. The majority does not claim that the state court’s result was an “unreasonable application” of federal law. Nor could it: The sentencing result was plainly reasonable. Instead, the majority says that the state court’s omission of the words “reasonable probability” in describing the Strickland prejudice standard is “contrary to” law and forecloses AEDPA’s further application, notwithstanding the overall reasonableness of the state court’s result.
I have misgivings about whether, in light of the Supreme Court’s decision in Richter, we are still entitled to reverse a state court’s reasonable decision based on what we consider to be its incorrect reasoning. Richter held that when confronted with a state- court’s summary denial, we “must determine what arguments or theories ... could have supported the state court’s decision; and then [ ] ask whether it is possible fairminded jurists could disagree that those arguments or theories are inconsistent with the holding in a prior decision of [the Supreme] Court.” 131 S.Ct. at 786. After Richter, it seems clear that we should assess the reasonableness of a state court’s decision, not its reasoning. After all, a habeas petitioner is not entitled to *1225any reasoning at all, so reversing a state court’s reasonable decision on the grounds of incorrect reasoning risks treating defendants inconsistently: Those who are given incorrect reasoning get relief while those who aren’t given any reasoning do not. And it has the perverse effect of encouraging state courts to deny relief summarily, to insulate their orders from tinkering by the federal courts.
Even before Richter, we were on the wrong end of a circuit split on this issue. See Clements v. Clarke, 592 F.3d 45, 55-56 (1st Cir.2010) (“It is the result to which we owe deference, not the opinion expounding it.”); Holder v. Palmer, 588 F.3d 328, 341 (6th Cir.2009) (“[Mistaken analysis of Strickland’s performance prong does not move the state court’s decision out from under AEDPA....”); Malinowski v. Smith, 509 F.3d 328, 339 (7th Cir.2007) (“[E]ven if the [state court] had applied the wrong standard, the proper standard results in the same conclusion_”). After Richter, I believe our precedent on this matter, as set forth in Frantz v. Hazey, 533 F.3d 724, 734 (9th Cir.2008) (en banc), is no longer good law.
Regardless, the state court’s reasoning here was not unambiguously contrary to federal law. The unadorned statement that the new evidence “would not have changed the sentence imposed” says nothing about the state court’s view as to the probability that the new evidence would change the sentence. For example, the state court could have meant that the new evidence would definitely not have changed the sentence or it could have meant that the evidence would likely not have changed the sentence. If the state court had the first modifier in mind, Mann was held to Strickland’s reasonable probability requirement; if the court had the second modifier in mind, perhaps Mann was not.
While both are consistent with what the state court said, the majority chooses to stitch “likely” into the opinion’s text. I don’t see why. • In ordinary language, we read the lack of a qualifier as a -sign of certainty, not doubt. That counsels in favor of reading “definitely,” rather than “likely,” into the state court’s statement.1 In any event, the opinion is at worst ambiguous. Under such circumstances, we may not simply rely on our best guess. Rather, we must presume that the state court knew the law it was applying, and that it read and understood the case it was citing. The majority does the opposite: It ignores the presumption that the state court knows the law, disregards the state court’s citation to correct authority, fails to consider the overall import of the opinion and infers error from ambiguity. In short, the majority plainly fails to deploy the “highly deferential standard for evaluating state-court rulings” that AEDPA requires. Renico v. Lett, 559 U.S. 766, 130 S.Ct. 1855, 1862, 176 L.Ed.2d 678 (2010) (internal quotation marks omitted).
*1226The majority’s approach conflicts directly with that of the other circuits that have considered this exact question. The Fifth and Seventh Circuits have both held that a state court’s omission of the modifier “reasonable probability” in explaining the Strickland prejudice prong does not constitute legal error when the state court has cited to a case articulating the correct Strickland standard and the overall import of the state court’s reasoning suggests that it believed the new evidence had little or no value. See Charles v. Stephens, 736 F.3d 380, 392-93 (5th Cir.2013) (per curiam); Sussman v. Jenkins, 636 F.3d 329, 359-60 (7th Cir.2011).2 The view of those circuits is reasonable and faithful to Supreme Court precedent; the majority's is not. Once again, we have placed ourselves on the wrong side of a circuit split regarding the scope of AEDPA’s application. We seldom come out the winner in such contests.
❖ *
After muffing the AEDPA inquiry, the majority doubles down by engaging in de novo review that misapplies both prongs of Strickland.
A. The majority’s assessment of Mann’s counsel is neither “highly deferential” nor shorn of “the distorting effects of hindsight.” Strickland v. Washington, 466 U.S. 668, 689, 104 S.Ct. 2052, 80 L.Ed.2d 674 (1984) The record makes clear that trial counsel conducted considerable investigation into Mann’s background in an effort to dig up mitigating sentencing evidence. In fact, the sentencing court commended counsel on doing a “very thorough job” in presenting his mitigation case. Counsel consulted with the Arizona Capital Representation Project. He suggested the trial court appoint a psychologist from the court clinic to evaluate Mann. He subpoenaed documentary evidence from government agencies with previous contact with Mann. He had Mann write an autobiography, and investigated details of that autobiography. He arranged for Mann’s daughter, mother, former employer and former co-worker to testify on Mann’s behalf during sentencing. These efforts were fruitful: The sentencing judge determined that Mann had demonstrated seven mitigating factors, all of which arose out of counsel’s investigation. Despite all this, the majority asserts that “there can be little doubt that counsel’s performance fell below an objective standard of reasonableness.” Maj. Op. at 1216.
The focal point of the majority’s Strickland analysis is its assertion that counsel should have obtained Mann’s medical records, id. at 1217, and pursued further evidence of Mann’s alleged brain injury.3 As *1227the majority notes, counsel could have called for more evaluations and for expensive neurological testing (though, of course, there’s no indication the trial court would have paid for such expensive procedures). But why? The neutral, court-appointed psychologist didn’t suggest any reason to suspect brain injury. Counsel is entitled to rely on such an expert’s view in deciding whether or not to pursue a defense theory related to mental illness. See Kendricks v. Calderon, 70 F.3d 1032, 1038-39 (9th Cir.1995); see also Cullen v. Pinholster, — U.S.-, 131 S.Ct. 1388, 1405, 179 L.Ed.2d 557 (2011) (it was “a reasonable penalty-phase strategy to focus on evoking sympathy for Pinholster’s mother,” rather than Pinholster’s alleged brain injury, in part because the single psychiatric expert counsel retained “told counsel that Pinholster did not appear to suffer from brain damage”). From counsel’s perspective at the time of trial, “there were any number of hypothetical experts ... whose insight might possibly have been useful,” Richter, 131 S.Ct. at 789, and there was no indication that a brain trauma specialist was the magic bullet the majority now construes it to be.
Counsel was already armed with substantial evidence of Mann’s dysfunctional childhood and long-term substance abuse. Further evidence of diminished capacity could have proven counter-productive. Such evidence can be a “two-edged sword that [a sentencing judge] might find to show future dangerousness” or use to conclude that a defendant is “simply beyond rehabilitation.” Pinholster, 131 S.Ct. at 1410 (internal quotation marks omitted).
Moreover, Arizona courts in 1995 frequently gave short shrift to mitigation evidence not causally related to the offense. See Schad v. Ryan, 671 F.3d 708, 723 (9th Cir.2011) (per curiam). An experienced and competent Arizonian counsel could reasonably have concluded that, even if he found compelling evidence of brain injury that affected Mann’s temperament, there would be no way of establishing a causal nexus between such an injury and a premeditated crime. Therefore, in “balanc[ing] limited resources,” Richter, 131 S.Ct. at 789, defense counsel was entitled to “make a reasonable determination that [this] particular [course of] investigation [was] unnecessary.” Leavitt v. Arave, 646 F.3d 605, 609 (9th Cir.2011) (internal quotation marks omitted).
This is so even if, as the majority contends, pursuing medical records at the mitigation stage was common practice among defense attorneys at the time. The Supreme Court has squarely “rejected the notion that the same investigation will be required in every case.” Pinholster, 131 S.Ct. at 1406-07. Here, counsel had no indication from the court-appointed psychologist that there was anything worth pursuing; and, even as a best case scenario, more evidence of Mann’s inability to control himself would have limited value in mitigation and perhaps even risked weighing against Mann. We shouldn’t require a defense counsel to spend time and money conforming to common practice if there’s no indication that doing so will actually benefit a particular client.
The crux of the majority’s argument is simply that counsel, in hindsight, could have done more. But that’s true in almost any case. “[Hjindsight is a weak standard for fixing constitutional minima.” Hendricks, 70 F.3d at 1038. Counsel’s failure to accumulate further psychological evidence hardly constitutes an “error [] so serious that counsel was not functioning as the ‘counsel’ guaranteed the defendant by *1228the Sixth Amendment.” Richter, 131 S.Ct. at 787 (internal quotation marks omitted).
B. The majority also misapplies Strickland’s prejudice prong. Instead of “reweigh[ing] the evidence in aggravation against the totality of available mitigating evidence,” Wiggins v. Smith, 539 U.S. 510, 534, 123 S.Ct. 2527, 156 L.Ed.2d 471 (2003), the majority examines the new evidence in isolation and fails to meaningfully assess its added worth in the context of the other evidence presented.
Absent from the majority’s analysis of Strickland’s prejudice prong is any discussion of the aggravating factors. Mann committed a heinous crime. He deliberately lured two men into his home with the express intention of murdering them and stealing their money. After shooting one of his victims, Mann stood over him narrating his slow, agonizing death. These aggravating factors were significant to the sentencing court, understandably so. There is no indication whatsoever that it agreed with the majority’s facile assertion that “the aggravating and mitigating factors were already closely balanced.” Maj. Op. 1222.
The majority claims that the car accident evidence could “have significantly altered the sentencing profile presented to the sentencing judge,” -id. at 1219, because Mann’s reaction to the accident would demonstrate that he was “capable of profound remorse,” and neuropsychological testing would show that he “suffered a serious brain injury that altered his personality.” Id. at 1222. But, evidence that Mann was “capable of remorse” would have little value because the sentencing judge explicitly found that Mann failed to show any remorse for the murders about which he was convicted. None of the additional evidence presented to the state post-conviction court would disturb that determination. That Mann was abstractly “capable” of remorse he did not show or feel for the victims he killed for lucre may actually have cut against him at sentencing.
The worth of the brain injury evidence is even more dubious. Unlike Pinholster, who actually demonstrated brain injury, see Pinholster, 131 S.Ct. at 1397, Mann provided only speculation that he suffered brain .trauma. Dr. Comer testified that Mann’s, neurological scans suggested he might have suffered from past brain injury, but admitted that substance abuse could also explain his findings. Furthermore, Comer relied entirely on Miller’s testimony to conclude that Mann might have suffered from personality changes after the 1985 accident, yet Miller’s testimony points to Mann’s substance abuse as the cause for a shift in his personality. Adding the mere possibility of brain injury to the mitigation side of the ledger falls far short of creating a reasonable likelihood that the sentence would have been different.
More fundamentally, the majority’s underlying belief seems to be that all that is required to meet Strickland’s second prong is to “significantly alter[ ] the sentencing profile” of a defendant. Maj. Op. 1219. That’s a dangerous misreading of the Supreme Court’s ineffective assistance precedent. By definition, any new, noncumulative evidence makes' a defendant’s sentencing profile different. The majority appears to assume, without explaining, that these differences will typically create a reasonable likelihood of a different sentencing result. But, here, the sentencing judge would still have been presented with the profile of a violent, cold-blooded killer who revelled in his victims’ suffering; the fact that the killer had shown remorse to others — but not his victims — and had suffered emotional difficulties after a car accident, doesn’t add up to a reasonable likeli*1229hood of a more lenient sentence. As the Supreme Court has made clear, to prevail on Strickland’s prejudice prong, “[t]he likelihood of a different result must be substantial, not just conceivable.” Richter, 131 S.Ct. at 792. What the majority has shown here is just barely conceivable.
This is not an unusual case. The state court, though it may have used some loose language, did nothing unreasonable. Defense counsel, though he could have pursued more evidence, was entirely competent, even expert. Yet the majority holds that the state of Arizona is unable to carry out the punishment it lawfully imposed. It was almost twenty years ago that Eric Mann was sentenced to death for his self-serving and sadistic crimes. That’s how long it takes to navigate the tortuous path from initial sentencing to federal habeas review. Our justice system cannot function effectively if we are compelled to restart the arduous post-conviction process in even a typical case such as this. No judge or lawyer is perfect. Holding them to unreasonable standards means that capital defendants — and the families of their victims — live out their whole lives in an interminable cycle of litigation.
There’s no virtue to endless delay; we disserve all concerned when we paralyze the judicial process. I would respect the state of Arizona’s sovereign judgment and allow Eric Mann to suffer the punishment justly and lawfully imposed on him.
. The majority also finds it significant that the state court used the phrase "would not have changed the sentence imposed” in denying Mann's Arizona Rule of Criminal Procedure 32.1(e) claim, a claim for which the burden of proof was more-likely-than-not. In the majority's view, because the state court denied Mann’s Strickland claim “for the same reason” as it denied the Rule 32.1(e) claim, it must have held Mann to the same standard of proof for both claims. But that’s a non sequi-tur. Just because a court uses the same reason to reject two claims doesn’t mean it believes those claims have identical burdens of proof. Nor does the state court's cross-reference to a previous use of an ambiguous phrase diminish that phrase’s ambiguity. • If the phrase "would not have changed the sentence imposed” means "would definitely not have changed the sentence imposed,” then it makes perfect sense that the state court would reject both Strickland and Rule 32.1(e) claims “for that reason.”
. The majority's view is also in deep tension with at least two other circuits. See Bledsoe v. Bruce, 569 F.3d 1223, 1232 (10th Cir.2009) (stating that "despite its 'may is not good enough' language, the Kansas Supreme Court applied the correct [Strickland ] standard”); Parker v. Sec’y for Dep’t of Corr., 331 F.3d 764, 786 (11th Cir.2003) ("Despite the imprecise language used by the Florida Supreme Court, we conclude the court understood and applied the correct prejudice standard from Stiickland.”).
. The majority also faults counsel for not calling Mann's ex-girlfriend Karen Miller as a witness during sentencing. But it was Miller who turned Mann in to the police, and then offered negative testimony about him during trial. Counsel had no way of knowing what Miller would say if called — her testimony could easily have worsened things for Mann. Given the "strong presumption that counsel’s conduct f[ell] within the wide range of reasonable professional assistance,” Strickland, 466 U.S. at 689, 104 S.Ct. 2052, it certainly wasn’t a blunder not to call a volatile and heretofore adverse witness during sentencing. Indeed, had counsel called Miller, the majority would likely have pointed to that decision as proof that counsel was incompetent: *1227"What a moron! He even put the snitch on the stand.”