While serving a long sentence in state prison for a series of armed robberies, John Grant won a job as a kitchen worker. The job brought him under the supervision of Gay Carter, a civilian prison employee, but it wasn’t one that lasted very long. Mr. Grant was soon fired after he was caught fighting with another inmate — and Mr. Grant didn’t take getting fired very well. He came to bear a grudge against Ms. Carter, a woman he used to get along with and even considered a friend.
Seeing Ms. Carter one day during a morning breakfast service, he told her, “I’ll get you, bitch.” The next morning he followed up, “You’re mine.” Mr. Grant then proceeded to make good on his threats. After breakfast, he lingered in the dining hall with no obvious purpose, but not altogether out of place either because he used to work there. After about ten .or fifteen minutes, Ms. Carter passed near him and he grabbed her, put a hand over her mouth, and dragged her into a small closet. With a shank he had secreted into the dining hall, Mr. Grant stabbed Ms. Carter, sixteen times in all.
The State of Oklahoma charged Mr. Grant with first degree murder and sought the death penalty. At trial, the government had little trouble proving that it was Mr. Grant who stabbed Ms. Carter to death. In his defense, Mr. Grant testified that he had no recollection of killing or wanting to kill Ms. Carter. A defense expert also testified that Mr. Grant suffered from borderline personality disorder, though the expert added that Mr. Grant was of average intelligence and didn’t show any signs of an organic brain disorder. The expert also refused to offer any view on whether Mr. Grant did or didn’t understand the consequences of his acts at the time of the murder. In the end, the jury found Mr. Grant guilty as charged.
At the penalty phase, the government argued that Mr. Grant deserved the death penalty on the basis of three aggravating factors surrounding the murder: (1) he had been convicted previously of violent felony offenses, (2) he murdered Ms. Carter while serving a felony prison sentence, and (3) he posed a threat of future violent criminal acts. By this point in the proceedings, the first two factors weren’t in much dispute. For its case on the third, the government pointed to other prison fights Mr. Grant had been involved in, including a fight with a prison guard; pointed to the fact that Mr. Grant killed a civilian kitchen worker while in prison; and argued that the evidence suggested he *1011might well strike at prison workers or inmates again.
The defense responded that any threat Mr. Grant posed could be mitigated with adequate care. A psychiatrist explained that Mr. Grant had not received mental health counseling or anti-psychotic medications in prison, though he then refused to speculate whether and to what extent Mr. Grant would benefit from either. Mr. Grant also briefly recounted for the jury his troubled childhood.
In the end, the jury found in the government’s favor on all the aggravating factors, found no mitigating factors outweighing those aggravating factors, and voted to impose the death penalty. The Oklahoma Court of Criminal Appeals (OCCA) denied relief on appeal. See Grant v. State (Grant I), 58 P.3d 783 (Okla.Crim.App.2002); Grant v. State (Grant II), 95 P.3d 178 (Okla.Crim.App.2004). Neither did the OCCA find relief warranted in two separate post-conviction proceedings Mr. Grant attempted. See Grant v. State, No. PCD-2002-347, slip op. (Okla.Crim.App. Apr. 14, 2003); Grant v. State, No. PCD-2006-690, slip op. (Okla.Crim.App. Nov. 6, 2006).
Mr. Grant then filed a habeas petition in federal court but the district court denied relief, too. See Grant v. Workman (Grant II), No. 05-cv-0167-TCK-TLW, 2010 WL 5069853 (N.D.Okla. Dec. 2, 2010). The district court did, however, issue Mr. Grant a certificate of appealability that allowed him to bring his case to this court. Mr. Grant’s certificate allows us to review the district court’s decision on the five grounds we discuss below. Ultimately, we agree with all the courts that have come before us and hold none warrants relief.
I
We begin with the question whether the guilt phase jury instructions satisfy the demands of federal due process doctrine. In Beck v. Alabama, the Supreme Court held that the Due Process Clause of the Fourteenth Amendment sometimes requires a state charging a defendant with a capital offense to permit the jury to consider alternative, lesser included offenses that do not carry with them the prospect of a death sentence. 447 U.S. 625, 627, 100 S.Ct. 2382, 65 L.Ed.2d 392 (1980); see also Schad v. Arizona, 501 U.S. 624, 647, 111 S.Ct. 2491, 115 L.Ed.2d 555 (1991). In this case, Mr. Grant was charged with and convicted of first degree murder. He argues that the state court trying him violated Beck by failing to give the jury the option of finding him guilty instead of the lesser included — and noncapital — offenses of first degree manslaughter and second degree murder. But we soon encounter two difficulties with this submission.
The first is that Mr. Grant never asked for a lesser included jury instruction at trial. This is a problem because in Hooks v. Ward, 184 F.3d 1206 (10th Cir.1999), “we h[e]ld that a state prisoner seeking federal habeas relief may not prevail on a Beck claim as to a lesser included instruction that he or she failed to request at trial.” Id. at 1234. The requirement of a contemporary request isn’t one with roots, as one might imagine, in state procedural law. The Hooks rule is federal in nature, an explanation of what’s required as a matter of federal due process doctrine to invoke Beck. As Hooks explained, “a proper request for a lesser included instruction [is] an essential requirement under the federal rules,” and “[g]iven principles of comity, ... this rule applies with even greater force when [a federal court] sit[s] in review of a state conviction in a § 2254 action.” Id. at 1235, 1234 (internal quotation marks omitted). Simply put, this court won’t impose a requirement on sovereign states that we don’t impose on the *1012federal courts under our direct supervision. So a state generally won’t be said to offend a defendant’s due process right to particular jury instructions when it has no occasion to refuse a request for them. See id. at 1234 (“In such cases, ... it is the defendant him or herself that precludes the jury from considering a non-capital option.... ”).
Mr. Grant replies that .the portion of Hooks claiming to hold this much — section III.C of the opinion — doesn’t really contain a holding at all. He points out that two judges concurred separately, indicating they joined all but section III.C of the main opinion in Hooks. Id. at 1241 (Anderson, J., concurring). For its part, Oklahoma rejoins that the separate concurrence took issue with other aspects of section III.C,,not this' one — and that, properly viewed, the panel was unanimous on the need for a defendant to request a lesser included offense instruction to trigger Beck.
At the end of the day, who’s right about the optimal reading of Hooks doesn’t much matter. It doesn’t because since Hooks issued this court has expressly and repeatedly held that “a state prisoner seeking federal habeas relief may not prevail on a Beck claim as to a lesser included instruction that he or she failed to request at trial.” Thornburg v. Mullin, 422 F.3d 1113, 1126-27 (10th Cir.2005); see also Darks v. Mullin, 327 F.3d 1001, 1007 (10th Cir.2003) (the “[fjailure to request such an instruction precludes a petitioner seeking habeas relief from prevailing on a Beck claim”); Hogan v. Gibson, 197 F.3d 1297, 1303 n. 3 (10th Cir.1999) (same); Smith v. Gibson, 197 F.3d 454, 464 (10th Cir.1999) (same). The point is by now long settled in this circuit and by many more cases than just Hooks itself. At this late date we simply do not see how we might hold otherwise.
Mr. Grant replies by directing us to the OCCA’s decision in Shrum v. State, 991 P.2d 1032 (Okla.Crim.App.1999). As Mr. Grant sees it, Shrum requires Oklahoma state courts, as a matter of state law, to provide lesser included offense instructions in capital cases always and automatically. Because of this, he suggests, Shrum relieved him of any duty to request a lesser included offense instruction at trial.
This suggestion, however, fails on its own terms. One premise on which Mr. Grant’s argument depends raises some interesting questions. His argument surely rests on the (if entirely implicit and unexplored) premise that state law can relieve a party of its duty under federal law to invoke a federal right. Whether this premise is sound undoubtedly warrants investigation. But there’s no need to tangle with that project in this case. No need because another premise on which Mr. Grant’s argument rests has problems of its own. Shrum relieves defendants of their obligation to request lesser included offense instructions only as a matter and for purposes of state law. Oklahoma still requires defendants who’ wish to assert federal constitutional complaints about proposed jury instructions to raise them in a timely fashion with the trial court. See, e.g., Barnard v. State, 290 P.3d 759, 769 (Okla.Crim.App.2012); Warner v. State, 144 P.3d 838, 881 (Okla.Crim.App.2006); McGregor v. State, 885 P.2d 1366, 1384 (Okla.Crim.App.1994). This includes federal Beck claims in particular: Oklahoma expressly requires defendants who wish to raise Beck claims to make “specific objection[s] to the instructions administered or request alternative instructions” they believe will satisfy Beck. Douglas v. State, 951 P.2d 651, 672 (Okla.Crim.App.1997). So it’s plain enough that at least one essential premise on which Mr. Grant’s argument depends — that Oklahoma has sought *1013to relieve him of his burden to press a federal entitlement to a lesser included charge under Beck — is simply wrong.1
A second and entirely independent problem confronts Mr. Grant’s Beck argument. Even overlooking his failure to object before the state trial court, the argument fails on its merits. As Beck explained, federal due process doctrine does not require a lesser included offense instruction to be given unless “the evidence would ... support[ ] ... a verdict” on that lesser included offense. ,447 U.S. at 627, 100 S.Ct. 2882. In this case, the OCCA didn’t assess Beck directly but it did reject Mr. Grant’s claim that he was automatically entitled to lesser included offense instructions as a matter of state law under Shrum, and it did so on precisely these grounds, holding that no lesser included instructions were required under Shram because the evidence adduced at trial could not rationally support a verdict for either first degree manslaughter or second degree murder. Because Beck, like Shram, does not require lesser included offense instructions when the evidence is insufficient to support a verdict on them, the OCCA’s state law conclusion under Shram — if correct — would also spell the end to any effort to secure federal relief under Beck.
The OCCA’s conclusion turns out to be correct; In so holding, we are mindful of the limits of our authority as a federal court reviewing the work of a state court. When it comes to the question what is required as a matter-of law to . establish a manslaughter or second degree murder charge, we may not second guess Oklahoma authorities. Matters of state law are theirs, not ours, to answer. See 28 U.S.C. § 2254(a); Boyd v. Ward, 179 F.3d 904, 916 (10th Cir.1999). Further, in the course of rejecting Mr. Grant’s state law Shram argument, the OCCA made certain factual findings. In our own Beck analysis we are obliged to presume those findings are correct unless and until shown to be erroneous by clear and convincing evidence, as indeed a federal court sitting in review of a state court in collateral proceedings must always do. See 28 U.S.C. § 2254(e)(1); Hooks, 184 F.3d at 1231. At the same time, because-the OCCA didn’t purport to rule, as a legal matter, on a federal Beck claim but only on a different (if related) state Shram claim, and seeking to afford Mr. Grant the benefit of the most generous standard of review we can, we assess de novo the legal question under Beck, asking whether a jury could “rationally ... find [the defendant] guilty of the lesser offense and acquit him of the greater.” Hogan, 197 F.3d at 1307.2 ,
*1014With these points in mind we turn first to the question of manslaughter. As a matter of state law, the OCCA explained that a conviction for first degree manslaughter under Oklahoma state law “requires that a person act with a ‘heat of passion’ caused by ‘adequate provocation.’” Grant I, 58 P.3d at 795 (quoting Okla. Stat. Ann. tit. 21, § 711). The court then proceeded to explain that it could find in the record “[n]o evidence ... to support either of these elements” and for that reason held no manslaughter instruction was required. Id. Before us, Mr. Grant doesn’t point to any facts suggesting (let alone clearly compelling the inference) Ms. Carter did anything to him that was “calculated to provoke an emotional response” or that would “ordinarily cause serious violence” — actions that under Oklahoma law amount to “adequate provocation.” Washington v. State, 989 P.2d 960, 968 n. 4 (Okla.Crim.App.1999). Neither does Mr. Grant direct us to facts suggesting that he acted in the “heat of passion” — namely, that the “homicide occurred while the passion still existed and before a reasonable opportunity for the passion to cool.” Davis v. State, 268 P.3d 86, 111 (Okla.Crim.App.2011). In these circumstances, we must respect the OCCA’s factual finding about the state of the record and from that finding it’s plain as a matter of federal law that a first degree manslaughter charge was not required.
It is likewise evident that a second degree murder charge was not warranted by the evidence. Under Oklahoma law, a second degree murder conviction is permissible only when the defendant acts “without any premeditated design to effect the death of any particular individual.” Williams v. State, 22 P.3d 702, 712 (Okla.Crim.App.2001) (emphasis added). The OCCA found that the evidence in the record strongly suggested some degree of premeditation: Mr. Grant bore a prison-made shank, he threatened Ms. Carter twice in two days, he waited for Ms. Carter to come near the closet, and then he forced her into the closet, covered her mouth, and stabbed her repeatedly near her vital organs. See Grant I, 58 P.3d at 795. Neither, the OCCA found, was there evidence in the record that Mr. Grant “suffered mental infirmities that would have rendered him incapable of forming the specific intent necessary” for first degree murder. Id.
Mr. Grant challenges these findings. He says his testimony showed “he was in a dazed state of mind” at the time of the incident and couldn’t remember committing the murder. He reminds us that there was a video taken shortly after the murder that showed him highly agitated. And he points out that his expert believed he suffered from borderline personality disorder.
None of these facts, however, clearly and convincingly unseats the OCCA’s finding that all of the evidence surrounding the killing suggested a degree of premeditation. Neither do they undermine its finding that the record lacked evidence suggesting Mr. Grant suffered mental infirmities that rendered him incapable of forming an intent to kill, especially given that his own expert refused to say Mr. Grant couldn’t understand right from wrong. Indeed, were we to consider the factual questions ourselves, without the layer of deference due under § 2254(e), we would reach the same view about them as the OCCA did. We simply cannot say that, given the facts in this record, a rational jury could have found that Mr. *1015Grant acted “without any premeditated design” to kill Ms. Carter.3
II
Mr. Grant contends the state trial court violated his Sixth Amendment Confrontation Clause rights. During the guilt phase, the government called Dr. Frederick Smith, a psychologist. Dr. Smith testified that he examined Mr. Grant a few days after the murder and found no serious mental health problems at that time. During cross examination, the defense responded with an aggressive attempt to discredit Dr. Smith — and in doing so got Dr. Smith to admit that he spent at most twenty minutes with Mr. Grant, and that he did not know whether Mr. Grant was of sound mind at the moment he killed Ms. Carter. Defense counsel wanted to ask Dr. Smith, too, about an out-of-court report prepared by another psychologist, Dr. Elliot Mason, relating yet another out-of-court-statement by Mr. Grant a few days after the murder. Before counsel could get there, however, the trial court held that any questioning about Dr. Mason’s report was outside the scope of Dr. Smith’s direct examination. This, Mr. Grant says, wasn’t true and the trial court’s failure to permit the inquiry violated his Sixth Amendment right to confront his accusers.
The OCCA agreed with Mr. Grant that the trial court erred. But it proceeded to deny relief on the ground that the error was harmless beyond a reasonable doubt. “The failure to allow cross-examination on this single, self-serving statement made three days after Grant murdered the kitchen worker and contained in a secondhand report,” it held, “had no impact on the jury’s determination of guilt or the sentence in this case.” Grant I, 58 P.3d at 795.
As a federal court sitting in collateral review of a state court’s work, we may not reverse the OCCA’s determination that a constitutional error was harmless beyond a reasonable doubt unless we can be sure for ourselves that the error had a “ ‘substantial and injurious effect’ on the jury’s decision.” Banks v. Workman, 692 F.3d 1133, 1139 (10th Cir.2012) (citing Fry v. Pliler, 551 U.S. 112, 119-20, 127 S.Ct. 2321, 168 L.Ed.2d 16 (2007)). “This standard precludes reversal of a conviction on habeas unless we have a grave doubt about *1016the effect of the error on the verdict.” Id. (internal quotation marks omitted).
That much we cannot say. Dr. Mason’s report was hearsay. Mr. Grant’s putative statement contained within it (being offered by Mr. Grant himself) was hearsay within hearsay. As Mr. Grant’s present counsel conceded at oral argument before us, the defense could have lawfully used such hearsay materials only to impeach Dr. Smith, not as substantive evidence with respect to any aspect of the case. See Mackey v. State, 526 P.2d 1161, 1165 (Okla.Crim.App.1974) (“Evidence to prove a substantive fact cannot be introduced by a purely hearsay statement under the guise of impeaching a witness.”). Neither do we see how questioning Dr. Smith about the report could have had any material additive effect in his impeachment. From cross-examination the jury already knew that Dr. Smith performed only a brief examination of Mr. Grant and that his diagnosis was at best tentative. Anything Dr. Smith admitted about overlooking statements contained in Dr. Mason’s report would have been cumulative of points already well made on cross-examination. See Delaware v. Van Arsdall, 475 U.S. 673, 684, 106 S.Ct. 1431, 89 L.Ed.2d 674 (1986).4
Ill
During the penalty phase, the government introduced two “victim impact” statements. Ms. Carter’s daughter wrote that she had hoped to follow in her mother’s footsteps by pursuing a career in the corrections field, but that her mother’s death caused her to doubt herself. Ms. Carter’s brother wrote that his sister had been kind and loving; he added that he missed his sister and had trouble sleeping at night. Both statements concluded by asking the jury to impose the death penalty.
While victim impact statements may be admissible for other purposes, this court has understood the Supreme Court to have banned their use to the extent the statements expressly request a death sentence. See, e.g., Lockett v. Trammel, 711 F.3d 1218, 1236 (10th Cir.2013); see also Payne v. Tennessee, 501 U.S. 808, 830 n. 2, 111 S.Ct. 2597, 115 L.Ed.2d 720 (1991); Booth v. Maryland, 482 U.S. 496, 508, 107 S.Ct. 2529, 96 L.Ed.2d 440 (1987). There is no question, then, that the introduction of the victim impact statements in this case was plainly erroneous as a matter of federal law to the extent the statements included a plea for the death penalty.
At the same time, we cannot say the error in this case was sufficient to warrant reversal. To reverse we must be able to say (once again) that the error had a “substantial and injurious effect,” one leaving us with a “ ‘grave doubt’ about the effect of the error on the jury’s verdict.” Selsor v. Workman, 644 F.3d 984, 1027 (10th Cir.2011). Here, two things, taken together, persuade us that standard isn’t met.
First is the nature of the evidence at the penalty phase against Mr. Grant. The government’s presentation on two aggravating circumstances was all but indisputable and its evidence on the third and *1017perhaps most important aggravating circumstance — Mr. Grant’s potential for continued dangerousness even if incarcerated — -was potent. The government showed that Mr. Grant had a history of violent felonies well before the murder; it showed that he was fired from his position on the prison dining staff for fighting with another inmate; it showed that he had engaged in still other fights while in prison, including with a prison guard; and it showed, of course, that he killed a prison employee who was previously his friend. All of this suggested that Mr. Grant would continue to pose a danger to others, including civilian prison workers, even while he remained imprisoned. To be sure, Mr. Grant did respond with evidence of his amenability to treatment and evidence about his troubled childhood. But even viewed in its totality the case against him remained considerable. See Welch v. Workman, 639 F.3d 980, 1003 (10th Cir.2011) (finding victim impact statements harmless where, among other things, “the evidence supporting the three aggravating factors ... provided strong support for a death sentence”).
Second is the nature of the particular statements before us. They conclude with the fine, “I believe [John Grant] should be given [or should receive] the death penalty.” No other embellishment is made on the subject. This court has held far more extensive pleas to lack’the required “substantial and injurious” effect on a jury’s verdict when the evidence against the defendant at sentencing was strong. See, e.g., DeRosa v. Workman, 679 F.3d 1196, 1236 (10th Cir.2012) (ruling harmless a statement that “[o]ur family has suffered enough because of this man. My family pleads with you to give the death penalty.”); Welch, 639 F.3d at 1000 (ruling harmless a statement that “[w]e can now only put our faith first in God and then our courts, and you, the jury. And I would beg you, please, don’t let this happen to another family. And, again, I say I feel that he should be imposed the death penalty.”); Welch v. Sirmons, 451 F.3d 675, 701 (10th Cir.2006) (noting five separate requests for the death penalty). Given that this case contains similarly strong evidence against the defendant and yet a comparatively muted pair of pleas,. we are hard pressed to see how we could, faithful to our precedent, find the admission of the statements here reversible error.
IV
Mr; Grant charges his trial lawyer with doing a poor job of investigating and presenting evidence about his background at the penalty phase. To prevail on a Sixth Amendment claim of ineffective assistance of counsel under Strickland v. Washington, a defendant must show both that (1) counsel “committed serious errors in light of ‘prevailing professional norms’ such that his legal representation fell below an objective standard of reasonableness,” and (2) there is “a reasonable 'probability that, but for counsel’s unprofessional errors, the result of the proceeding would have been different.” Wackerly v. Workman, 580 F.3d 1171, 1176 (10th Cir.2009) (discussing Strickland, 466 U.S. at 688, 694, 104 S.Ct. 2052).
The OCCA held that Mr. Grant’s claim failed at both steps. On the question of deficient performance, the OCCA held that defense counsel’s alleged failure to investigate and present more evidence about Mr. Grant’s background was actually Mr. Grant’s fault, not his counsel’s, because Mr. Grant prohibited counsel from contacting his family members. On the question of prejudice, the OCCA held that none of the evidence Mr. Grant says his counsel should have uncovered was sufficiently *1018powerful that it would have made a difference.
The federal district court agreed in part and disagreed in part. It held that the OCCA’s deficient performance analysis was based on a clear error of fact. As the district court saw it, Mr. Grant never prevented his lawyer from contacting his family and the OCCA’s factual findings otherwise could not stand even under the lenient standard of review dictated by § 2254(e)(1). Correctly understood, the district court said, the facts suggested deficient performance. Even so, the district court agreed with the OCCA that the facts did not suggest prejudice because the evidence Mr. Grant’s lawyer could have found and presented about his client’s background was not reasonably likely to have made a difference to the outcome.
Before us, the State doesn’t dispute the district court’s assessment in any way but one. The district court assessed the question of prejudice de novo. In the State’s view, however, we must review the prejudice question through the deferential prism of § 2254(d). Meanwhile, Mr. Grant insists just the opposite. He suggests that the State’s failure to contest the district court’s finding of factual error on the deficient performance question requires us to review the prejudice question de novo.
The State is correct. This court has already held that when a state court’s deficient performance “analysis [is] ‘based on an unreasonable determination of the facts in light of the evidence presented in the State court proceeding,’ we [are] still ... bound to defer to [its] prejudice analysis under Strickland’s second prong.” Lott v. Trammell, 705 F.3d 1167, 1214 (10th Cir.2013) (quoting 28 U.S.C. § 2254(d)(2)). This much follows from the text of § 2254(d). That provision requires federal courts to defer to a state court’s decision on a defendant’s claim unless “the adjudication of the claim ... resulted in a decision that was based on an unreasonable determination of the facts.” 28 U.S.C. § 2254(d)(2). And even when a state court’s decision on Strickland’s first prong (deficient performance) is affected by an unreasonable determination of the facts, that does not necessarily mean its decision on Strickland’s second prong (prejudice) is similarly affected. This case illustrates why. By its very nature, any factual error the OCCA might have made in finding that Mr. Grant forbade counsel from contacting his family goes only to the question of deficient performance: whether counsel should or should not have done more to investigate Mr. Grant’s background. By the point the OCCA reached the second and separate Strickland question of prejudice, it had to and did assume that counsel was deficient in failing to contact family members. By that point, the OCCA had to and did ask only whether that presumed deficiency was prejudicial. Put differently, it is not possible to say the OCCA’s decision on the prejudice question was based on its putatively unreasonable factual determination that Mr. Grant forbade counsel from contacting his family, a factual determination that could have pertained only to the deficiency question.
We can thus say this about the standard of review binding us in this case. Under Strickland’s second element, the only element contested before us, Mr. Grant “must show that this deficient performance mattered — namely, that there is ‘a reasonable probability that, but for counsel’s unprofessional errors, the result of the proceeding would have been different.’ ” Wackerly, 580 F.3d at 1176 (quoting Strickland, 466 U.S. at 694, 104 S.Ct. 2052). In a system like Oklahoma’s, where only a unanimous jury may impose the death penalty, the question is whether it’s *1019“reasonably probabl[e] that at least one juror would have struck a different balance.” Wiggins v. Smith, 539 U.S. 510, 537, 123 S.Ct. 2527, 156 L.Ed.2d 471 (2003). A reasonable probability is one that is “substantial, not just conceivable.” Harrington v. Richter, — U.S.-, 131 S.Ct. 770, 792, 178 L.Ed.2d 624 (2011). But because we come at this question on habeas, we may approach it not directly and de novo but only indirectly through § 2254(d), asking whether the OCCA’s decision that the outcome would not have changed but for his counsel’s alleged deficiencies is itself unreasonable as a matter of fact or federal law. In asking that much narrower question we must, as well, limit our review of the record to the same record that was before the OCCA. See Cullen v. Pinholster, — U.S.-, 131 S.Ct. 1388, 1398, 179 L.Ed.2d 557 (2011); Black v. Workman, 682 F.3d 880, 895 (10th Cir.2012).5
In seeking to prove prejudice to the OCCA, Mr. Grant produced evidence from family members suggesting they had been prepared to testify during the penalty phase that Mr. Grant grew up poor, one of nine kids raised by a single mother who depended on social assistance. In his early years, Mr. Grant had been a quiet, sweet young boy with a fondness for dogs, football, and cars. But as Mr. Grant got older, things took a turn for the worse. He started hanging around with the wrong crowd and frequently found himself in trouble with the law, mostly for stealing (sometimes, his siblings insisted, for their benefit). He spent time incarcerated at juvenile facilities, an experience that intensified his relationships with friends who were up to no good. None of his other family members spent significant time in prison and most couldn’t comment on what Mr. Grant’s adult personality was like because they moved away to Oregon when Mr. Grant was a teenager. To be sure, Mr. Grant did send and receive the occasional letter — and. make, and receive an occasional phone call, too — but never very often in the twenty years Mr. Grant had spent in prison by the time of the murder. His siblings reported visiting only rarely, and less often the longer Mr. Grant was in prison. Only his mother and uncle together made an annual visit, and then only for about an hour each time.
The OCCA concluded that none of this would have made a difference to the outcome of Mr. Grant’s penalty phase proceedings. See Grant II, 95 P.3d at 181. And we cannot say — as we must to reverse — that the OCCA’s conclusion was contrary to, or involved an unreasonable application of federal law, or that it result*1020ed in a decision based on an unreasonable determination of the facts before the state courts.
In seeking to persuade us otherwise, Mr. Grant relies primarily on the Supreme Court’s decision in Wiggins and similar cases. In Wiggins, the Supreme Court faced a Strickland claim predicated on a lawyer’s failure to conduct a reasonable investigation into the background of his capital defendant client. See 539 U.S. at 533-34, 123 S.Ct. 2527. In conducting its prejudice analysis, the Court highlighted the “sordid details of [the defendant’s] life history” that an investigation would have uncovered. Id. at 536, 123 S.Ct. 2527. Mr. Wiggins, it turned out, “experienced severe privation and abuse in the first six years of his life while in the custody of his alcoholic, absentee mother.” Id. at 535, 123 S.Ct. 2527. Once he secured a place in' foster care, things fared no better: “He suffered physical torment, sexual molestation, and repeated rape.” Id. As the Court saw it, “had the jury been able to place petitioner’s excruciating life history on the mitigating side of the scale, there is a reasonable probability that at least one juror would have struck a different balance.” Id. at 537, 123 S.Ct. 2527.
The Supreme Court and this one have found prejudice arising from an attorney’s failure to investigate his client’s background circumstances in other materially similar circumstances. For example, in Williams v. Taylor, 529 U.S. 362, 120 S.Ct. 1495, 146 L.Ed.2d 389 (2000), the Supreme Court relied on many of the same “nightmarish” circumstances that would later move it to find prejudice in Wiggins: privation, neglect, and abuse, all to an astounding degree and all in addition to borderline mental retardation. See id. at 395 & n. 19, 120 S.Ct. 1495. In Anderson v. Sirmons, 476 F.3d 1131 (10th Cir.2007), this court emphasized that counsel failed to realize that the defendant suffered from brain damage, was “borderline mentally defective,” and “grew up in poverty, the twelfth child of a physically and emotionally abusive mother” who neglected her children. Id. at 1147. In Smith v. Mullin, 379 F.3d 919 (10th Cir.2004), we noted that the defendant grew up in an “unstable” home and suffered abuse at the hands of his caregiver, two facts the jury never learned in the bare case for mitigation it did hear. Id. at 942. Neither did it learn that the defendant suffered from a brain injury and had the intellect and emotional maturity of a twelve-year old. Id. at 941. And in Hooks v. Workman, 689 F.3d 1148 (10th Cir.2012), we emphasized that the defendant suffered from organic brain damage, had a “traumatic and chaotic” childhood, ahd had been abused by his alcoholic father and neglected by his mother. Id. at 1203 n. 28; see also Rompilla v. Beard, 545 U.S. 374, 390-92, 125 S.Ct. 2456, 162 L.Ed.2d 360 (2005) (finding prejudice in light of the evidence of abuse, poverty, alcoholism, brain damage, and mental impairments counsel unreasonably failed to uncover).
Of course we can only fault the OCCA for failing to abide Supreme Court precedent, not our own. See 28 U.S.C. § 2254(d)(1). AEDPA permits reversal only if a state court’s decision is contrary to a Supreme Court decision, and “circuit precedent may [not] be used to refine or sharpen a general principle of Supreme Court jurisprudence into a specific legal rule.” Marshall v. Rodgers, — U.S. -, 133 S.Ct. 1446, 1450, 185 L.Ed.2d 540 (2013) (per curiam). But while Anderson, Smith, and Hooks are not controlling for our present purposes, they do suggest successful claims of prejudice involving personal history in capital cases all tend to follow the same general path Wiggins laid down. Meanwhile, other cases from our *1021circuit reveal the many different ways in which unsuccessful claims of prejudice involving personal history tend to fail and can find themselves — reasonably—distinguished from Wiggins and its kin.
For one, while juries may be moved by “[diagnoses of specific mental illnesses ..., which are associated with abnormalities of the brain and can be treated with appropriate medication,” this court has held that evidence suggesting a defendant suffers from “generalized personality orders” may be less powerful. Wilson v. Sirmons, 536 F.3d 1064, 1094 (10th Cir.2008), aff'd on reh’g sub nom. Wilson v. Workman, 577 F.3d 1284 (10th Cir.2009) (en banc). So, for example, in McCracken v. Gibson, 268 F.3d 970 (10th Cir.2001), we thought it “conceivable” but not reasonably likely that evidence suggesting the defendant suffered from borderline personality disorder and bipolar disorder could have helped him avoid the death penalty. See id. at 979-80. Likewise, in Young v. Sirmons, 551 F.3d 942, (10th Cir.2008), we held that a defendant could not establish prejudice with evidence that he suffered from “Compulsive Personality Disorder” and committed the murders in question while “under severe emotional distress.” Id. at 968-69. We didn’t think this evidence “provide[d] a compelling or sympathetic explanation for [the defendant’s] violent behavior.” Id.
For another, evidence of- childhood abuse or neglect isn’t always severe enough to earn a jury’s sympathies. In Pinholster, the Supreme Court didn’t think it was “significant” that the defendant there “always got the worst of it” when he and his siblings were disciplined and that he “didn’t get much love” from his “mother and stepfather.” 131 S.Ct. at 1410. And our own precedent suggests that a court can sometimes reasonably find mitigating evidence of childhood abuse outweighed by countervailing “evidence supporting -the conviction and evidence supporting -multiple aggravating circumstances,” such as a record of violent felonies. Foster v. Ward, 182 F.3d 1177, 1189 (10th Cir.1999); see also Lott, 705 F.3d at 1214; McCracken, 268 F.3d at 980; Walker v. Gibson, 228 F.3d 1217, 1234 (10th Cir.2000), abrogated on other grounds by Neill v. Gibson, 278 F.3d 1044 (10th Cir.2001); Smith, 197 F.3d at 463. It suggests, too, that such evidence is generally less persuasive when a defendant commits the capital offense later in life. See Boyd, 179 F.3d at 918.
Even the presence of evidence about mental health problems and abusive family environments doesn’t necessarily run only one way: we have said this kind of evidence often possesses a “double-edged nature.” Wackerly, 580 F.3d at 1178. In Pinholster, for example, the Supreme Court thought evidence suggesting -that the defendant’s family was afflicted -with “serious substance abuse, mental illness, and criminal problems” could backfire before a jury, leading it to conclude that the defendant “was sijnply beyond rehabilitation.” 131 S.Ct. at 1410. Our cases have likewise recognized the potential double-edged effect some mental health evidence can have. See, e.g., Gilson v. Sirmons, 520 F.3d 1196, 1250 (10th Cir.2008) (finding no prejudice where expert report would have indicated defendant could not easily conform to social norms because of “impulsivity, poor judgment, and the failure to see or understand the consequences of his actions”); McCracken, 268 F.3d at 980 (finding no prejudice where evidence would have established that defendant could be “unpredictable, moody and impulsive”); Cannon v. Gibson, 259 F.3d 1253, 1277-78 (10th Cir.2001) (finding that defendant’s evidence could show him to be an “unstable individual with very little impulse control”). And we have recognized that fami*1022ly background can sometimes cut both ways, as well. We explained in Young, for example, that the defendant “had a family, including parents, brothers, and sons, that loved and care for him,” and that this in some ways made him more culpable. 551 F.3d at 968. With a loving and supportive family behind him, it’s reasonable to think that a defendant who has committed a heinous crime may be seen by jurors to bear fewer excuses for his actions. See id.
Even if a family member’s proffered testimony doesn’t stand to harm a defendant, it might still invite an opening for harmful questioning from the prosecution. See, e.g., Pinholster, 131 S.Ct. at 1410 (noting that “mitigating evidence” may sometimes “expose[ ]” a defendant “to further aggravating evidence”); Wilson v. Trammell, 706 F.3d 1286, 1306 (10th Cir.2013) (“[W]e must consider not just the mitigation evidence that Defendant claims was wrongfully omitted, but also what the prosecution’s response to that evidence would have been.”). We saw the potential for that effect in Davis v. Executive Director of the Department of Corrections, 100 F.3d 750 (10th Cir.1996). Although the petitioner in that case could have benefitted from his family’s testimony about his “troubled childhood,” we thought it just as likely that their testimony would “open the door to damaging evidence” about past criminal wrongdoing. Id. at 761. Likewise, in Lott, we didn’t think the OCCA was wrong to account for the risk that an expert witness would have to admit on cross-examination some of the less sympathetic facts about the defendant’s life. See 705 F.3d at 1214. Neither did we think the OCCA unreasonably determined that risk outweighed whatever mitigating effect the expert’s discussion of the defendant’s background would have. Id. And in Gardner v. Galetka, 568 F.3d 862 (10th Cir.2009), we thought it significant that, had the evidence the defendant urged us to consider actually been presented to the jury, the prosecution would have likely told the jury by way of rebuttal about a number of violent acts the defendant had previously committed. See id. at 881.
Many of our cases have also refused to find prejudice when the evidence the defendant says counsel should have presented would have been cumulative of the evidence the jury actually heard. See, e.g., DeRosa, 679 F.3d at 1221; Welch, 639 F.3d at 1013; Gardner, 568 F.3d at 880-81. In DeRosa, for example, we determined there was no prejudice in part because the mitigating witnesses had actually brought up at trial many of the points trial counsel failed to uncover with his own investigation. See 679 F.3d at 1221. And as our discussion in DeRosa made clear, the cumulative nature of mitigation evidence is especially problematic when the government’s case — either on guilt or for the death penalty — is strong. See id.; see also Walker, 228 F.3d at 1234; Clayton v. Gibson, 199 F.3d 1162, 1176 (10th Cir.1999); Castro v. Ward, 138 F.3d 810, 832 (10th Cir.1998).
In light of these precedents, we cannot agree with Mr. Grant that the OCCA was unreasonable in any respect when it concluded that he was unable to show prejudice. Mr. Grant’s proffered evidence in this case isn’t of the kind or quality of that in Wiggins and related cases — but seems a good deal more like what’s found in those cases where the Supreme Court and this court have refused to find prejudice. True, Mr. Grant’s family could have testified that Mr. Grant grew up poor and disadvantaged. But there is no evidence before us suggesting organic brain injuries or mental retardation, no evidence of physical abuse before he started down the path of illegal activity, and no evidence his other family members felt the need due to the difficulties of their family life to travel *1023down a similar path. Noteworthy, too, is that Mr. Grant murdered Ms. Carter when he was 36, well into adulthood.
Mr. Grant’s ease also tends to run into pitfalls this court has already recognized and we can hardly fault the OCCA’s decision as unreasonable for failing to find prejudice in such circumstances. For one, Mr. Grant’s proffered evidence has a clear double edge to it. While the jury would have learned Mr. Grant grew up poor and suffered abuse after he committed crimes and found himself in juvenile detention, the prosecution could have used that fact to emphasize Mr. Grant was already a criminal by the time he was a teenager and that none of his many years behind bars had dissuaded him from acts of violence against other prisoners, prison guards, or even prison kitchen staff. See Grant II, 95 P.3d at 183 (Lumpkin, J., specially concurring) (“Twenty years of structured incarceration has not been sufficient to ameliorate the defendant’s violent tendencies and that is what impacts the average juror as they view the savagery of this attack on an unarmed female food service worker.”). While the jury would have learned that Mr. Grant had been a sensitive child and this may have earned the jury’s sympathies, it’s just as likely it would have made him seem more culpable, proof that Mr. Grant wasn’t the type of murderer who lacked the ability to empathize with his victims.
Another pitfall would have been the damaging information the prosecution could have elicited from Mr. Grant’s family members on cross-examination. The prosecution could have asked about the fact that Mr. Grant appeared not to maintain any meaningful relationship with his family. It could have asked why, of all Mr. Grant’s many siblings, only he wound up in serious trouble with the law. And it could have pressed questions that revealed Mr. Grant’s family didn’t know much of anything about what kind of man twenty years in prison had turned Mr. Grant into, or bother to stay in close enough touch to find out.
Finally, the family’s evidence would have been at least partially cumulative of the evidence counsel did present at trial. Mr. Grant and his guilt stage expert both talked about his difficult childhood. That expert, for instance, revealed to the jury that Mr. Grant “functionally never knew his father at all,” that Mr. Grant “had problems in school” but was “never seen by any kind of school psychologist or clinician,” and spent time in a juvenile facility “which has since been closed down for abuse.”
Thus far, we have addressed Mr. Grant’s challenge to the legal reasoning the OCCA employed in analyzing Strickland’s prejudice prong — that is, his attempt at proving the OCCA decision rested on an unreasonable application of federal law. See 28 U.S.C. § 2254(d)(1). But Mr. Grant also challenges two-factual determinations the OCCA made in connection with its Strickland prejudice analysis. See . id. § 2254(d)(2). Of course, as we’ve seen, Mr. Grant challenges still other factual findings bearing only on the question of deficient performance. See supra pp. 1017-19. But when it comes to prejudice, he says the OCCA erred factually in two ways — when it said his turn to crime as an adolescent was a “matter of choice” and when it said his family “would have repeated [his] account of his childhood” if they had testified. In addition to these two factual errors asserted by Mr. Grant, the dissent claims to have spotted a third: a statement by the OCCA that Mr. Grant had no meaningful contact with his family while in prison.
At the outset we note that it’s not at all clear that the OCCA’s decision on *1024the question of prejudice was actually “based on” any of these findings, as § 2254(d)(2) requires. At least some of these findings arguably concerned only subsidiary issues that the OCCA mentioned in passing. But even assuming the OCCA’s prejudice decision was “based on” all of these findings, another problem quickly arises: none is “unreasonable” within the meaning of § 2254(d)(2). That standard is a “restrictive” one, Williams, 133 S.Ct. at 1092, requiring a showing more powerful than that “the federal habeas court would have reached a different conclusion in the first instance,” Wood v. Allen, 558 U.S. 290, 301, 130 S.Ct. 841, 175 L.Ed.2d 738 (2010). We do not believe that standard met in this case.6
Take the OCCA’s finding that Mr. Grant “had no meaningful contact with the family members who would have testified.” Grant I, 58 P.3d at 800. As it happens, that finding was entirely consistent with the record. Only Mr. Grant’s mother and uncle visited Mr. Grant in the years leading up to the murder—and only once a year for an hour each time at that. Indeed, Mr. Grant himself freely concedes to this court that “it is true ... that [his] family had no meaningful contact with him.” Aplt.’s Opening Br. 49. ■
Consider next the OCCA’s determination that Mr. “Grant’s childhood ... was a matter of choice.” Grant II, 95 P.3d at 180. This may represent an unforgiving view of the facts, one we would not choose to make ourselves, but we can’t say it’s lacking evidentiary support. Mr. Grant’s mother pinned her son’s turn to crime at an early age on “skipping school,” which she explained “led to getting into trouble.” One of Mr. Grant’s brothers recounted that Mr. Grant started committing crime not because of the family’s circumstances but because he “just basically h[ung] around the wrong crowd and then, you know, [it] kind of snowballed.” A sister explained that while Mr. Grant stole in order to provide for his younger siblings, this was only true “sometimes.” And out of everyone in Mr. Grant’s household, it was only Mr. Grant who was sent to prison or even juvenile detention.
Finally, we aren’t persuaded that the OCCA made an unreasonable determination of the facts when it said that Mr. Grant’s family would have “repeated” what Mr. Grant • had said during the penalty phase of his trial. See Grant I, 58 P.3d at 800. To be sure, the family would have added nuance and detail to Mr. Grant’s brief recitation of his personal history. But an imperfect or even an incorrect determination of the facts isn’t enough for purposes of § 2254(d)(2). See Schriro v. Landrigan, 550 U.S. 465, 473, 127 S.Ct. 1933, 167 L.Ed.2d 836 (2007). A determination must be' unreasonable. And that much we cannot say in this case. Mr. Grant provided a sketch of his background and his family’s testimony would have done no more than fill in some gaps—and, as we have seen, to the extent they would have added anything they would have done so in ways often more harmful than helpful to Mr. Grant.
In the end, we do not question that the evidence Mr. Grant says his lawyer should have presented bears some mitigating effect. But in light of the Supreme Court’s *1025holdings and our own, we can’t say it was unreasonable for the OCCA to hold that Mr. Grant’s case falls on the side of the line where the potential mitigating impact of the unproduced evidence might have been “conceivable” but not “substantial” enough to think it would have altered the outcome. Harrington, 131 S.Ct. at 792. Neither can we say that any of the factual determinations underlying the OCCA’s conclusion on Strickland’s prejudice prong were more than just “debatable” but in fact altogether “unreasonable.” Wood, 558 U.S. at 303, 130 S.Ct. 841. For that matter, we do not believe de novo review would yield any different result. As we’ve seen, this case bears remarkably few similarities to cases in which we found prejudice while in several ways it is indistinguishable from those in which we haven’t.
V
That leaves us with Mr. Grant’s complaint that, even if the errors he has’ identified do not warrant reversal individually, they do when considered cumulatively. Mr. Grant’s only argument on this score, however, is that “this Court should consider the synergistic effect of all the errors and grant Mr. Grant relief.” Aplt.’s Opening Br. 96. Such a perfunctory assertion falls well short of what’s needed to overturn a judgment, let alone one as long-settled and repeatedly reviewed as this one. Even a capital defendant can waive an argument by inadequately briefing an issue, see Romano v. Gibson, 278 F.3d 1145, 1155 (10th Cir.2002), and we break no new ground by holding the same here.
Even if we were to overlook the deficiency of the argument made to us and try to develop a cumulative error theory for Mr. Grant, we would still fail to find enough here to reverse. At first glance, the cumulative error doctrine looks simple enough. This court has said that the task “merely” consists of “aggregating] all the errors that have been found to be harmless” and “analyzing] whether their cumulative effect on the outcome of the trial is such that collectively they can no longer be determined to be harmless.” United States v. Rivera, 900 F.2d 1462, 1470 (10th Cir.1990) (en banc). Only if the errors “so fatally infected the trial that they violated the trial’s fundamental fairness” is reversal appropriate. Matthews v. Workman, 577 F.3d 1175, 1195 n. 10 (10th Cir.2009). '
But as easy as the standard may be to state in principle, it admits of few easy answers in application. After all, “it is the rare trial that will be an ideal specimen in all respects, given that even the most well-intentioned trial participants may commit the occasional error.” United States v. Runyon, 707 F.3d 475, 520 (4th Cir.2013). Where and how, then, should a court draw the line between what’s ordinary (and ordinarily harmless) and what’s rare (and fundamentally unfair)? Especially when the errors we are called on to accumulate may be very different in kind (incommensurate) and involve separate aspects of the case (guilt versus penalty)? Our precedent doesn’t say except to suggest that wherever the cumulative error line may fall, it is not crossed often. Indeed, our search turns up only two published cases in the last many years in which this circuit has found cumulative error. See Cargle v. Mullin, 317 F.3d 1196, 1225 (10th Cir.2003); United States v. Wood, 207 F.3d 1222, 1237 (10th Cir.2000); see also Derden v. McNeel, 978 F.2d 1453, 1456 (5th Cir.1992) (en banc) (“[T]he possibility of cumulative error is often acknowledged but practically never found persuasive..... ”).
Confusing, too, is what it means to accumulate error. The task is undoubtedly more subtle than simply counting up the number of errors discovered. In one case, for example, we thought that the accumu*1026lation of as many as six errors, insufficient to reverse. See Darks, 327 F.3d at 1019. We explained that four guilt-stage errors had only “minor significance,” neither “significantly strengthen[ing] the State’s case [n]or diminish[ing]” the petitioner’s ease. Id. We explained, as well, that none of the errors “undermine[d]” the government’s evidence of the sole aggravating circumstance at sentencing. Id. The strength of the aggravator meant there was “no reasonable probability ... that the jury would have imposed a sentence less than death.” Id. By contrast, in one of the few cases in which we found cumulative error to warrant reversal, we paid special mind to the fact that the errors had “an inherent synergistic effect.” Cargle, 317 F.3d at 1221. At the guilt phase of the trial, the errors all went to “two absolutely critical witnesses” for the government. Id. As it concerned sentencing, we noted that the errors “greatly inflated” the government’s case for the petitioner’s guilt, and counsel’s ineffectiveness meant there was a “conspicuous absence of counterbalancing mitigation evidence.” Id. at 1224-25.
To be sure, we don’t understand the difference between Darks on the one hand and Cargle on the other to imply a need for some “synergistic effect” to prevail on a claim of cumulative error. ' The reason why becomes clear if we understand prejudice in terms of probabilities. One might “accumulate” probabilities by adding them together, taking into account the disjunctive probabilities of each error. One might also “accumulate” probabilities by multiplying them and finding reversible error only in the space where all errors are conjunctively appearing all at once. If the cumulative error doctrine means anything, it must be that prejudice can be accumulated disjunctively — that all a defendant needs to show is a strong likelihood that the several errors in his case, when considered additively, prejudiced him. If it were otherwise, the cumulative error doctrine would be a nullity. A finding that one error wasn’t prejudicial would necessarily preclude a finding that all of the errors were prejudicial. See generally Amos Tversky & Daniel Kahneman, Extensional Versus Intuitive Reasoning: The Conjunction Fallacy in Probability Judgment, 90 Psychol. Rev. 293 (1983) (describing the intuition that a conjunction is more probable than its constituents as the “conjunction fallacy”). So while one error may make another error in the same direction more egregious, a defendant can still show cumulative error by accumulating unrelated errors if their probabilistic sum sufficiently undermines confidence in the outcome of the trial.
Even bearing all this in mind, and even approaching the question de novo, we could not say the accumulation of the three errors we’ve identified or assumed (concerning the cross-examination of Dr. Smith, the victim impact statements, and the background mitigation evidence) warrants reversal. As we have seen, none of the three errors was anything more than modest on its own terms. Adding them together undoubtedly leads to a somewhat less modest sum. But even still they do not collectively call into question the compelling case the government put on and they did not rob Mr. Grant of the ability to present anything more than a modest case for mitigation. Neither was there a synergistic effect that somehow, say, undermined a particular key question in the government’s case. For example, even if Mr. Grant’s family had testified at the penalty phase and even if they didn’t have to compete with victim impact statements for the jury’s sympathies, taken synergistically the jury still would have been left with an .at best mixed impression of Mr. Grant’s family support structure and social connections. The further (cumulative) im*1027peachment of Dr. Smith would not have touched on those questions at all, being unusable for any substantive purpose. In these circumstances, and given the guidance we have from our precedent, we simply do not see any basis on which we might reverse. See Bland v. Sirmons, 459 F.3d 999, 1029 (10th Cir.2006) (no cumulative error where “evidence supporting ... aggravating factors was overwhelming and the mitigating evidence weak”); Willingham v. Mullin, 296 F.3d 917, 935 (10th Cir.2002) (“[T]he strength of the State’s case ... effectively undercut[ ] [petitioner’s] assertion of actionable prejudice....”).
Affirmed.
. Of course, what due process doesn't compel the Sixth Amendment still might. A defendant’s failure to seek a lesser included offense instruction may not trigger Beck, but if the failure was the product of defense counsel's deficient performance and if that failure turns out to be prejudicial to the defendant, that could form the basis for a Sixth Amendment ineffective assistance of counsel claim under Strickland v. Washington, 466 U.S. 668, 688, 694, 104 S.Ct. 2052, 80 L.Ed.2d 674 (1984). We recognized this possibility in Hooks, 184 F.3d at 1235 n. 29. But here, as there, we are confronted with no ineffective assistance claim on this question.
. Since this case was argued the -Supreme Court has instructed us to apply a rebuttable presumption that a "federal claim was adjudicated on the merits” even when "a state court rejects a federal claim without expressly addressing that claim.” Johnson v. Williams, - U.S. -, 133 S.Ct. 1088, 1096, 185 L.Ed.2d 105 (2013). Presuming -this much would preclude the application of de novo review of the legal question under Beck and trigger instead the application of § 2254(d)'s additional layer of deference to the OCCA's decision, making Mr. Grant's task in this case all the more difficult. Because Mr. Grant hasn’t had a chance to address the potential application of Johnson and because he loses *1014whether or not § 2254(d) applies, our analysis here assumes without deciding that Johnson 's "presumption of merits adjudication” has been rebutted. See id. at 1096-97.
. Mr. Grant points out that the trial court granted his request for an insanity instruction. But that legal instruction doesn't answer the Beck question whether the facts in the record supported anything other than a first degree murder conviction. Indeed, if the trial court’s decision to provide that legal instruction suggests anything, it is that the trial court was generous to Mr. Grant given that even his own expert declined to testify that he was legally insane (unable to tell right from wrong) at the time of the killing.
Mr. Grant also argues that the OCCA violated Beck by focusing on the wrong question, asking whether the government’s evidence was sufficient to support a conviction for first degree murder rather than whether it permitted a conviction for second degree murder. But here again Mr. Grant simply misreads the OCCA’s opinion. As we’ve seen, the OCCA wasn’t purporting to apply Beck at all but state law (Shrum). Besides, while the OCCA did comment on the sufficiency of the government's case to sustain a first degree conviction, see Grant, 58 P.3d at 795 (“In this case, the evidence clearly establishes a premeditated design to kill.”), it then proceeded to hold as well that the evidence "simply d[id] not support a finding that [Mr. Grant] acted without a premeditated design to effect death.” Id.; see also id. ("[W]e do not ask a jury to consider a lesser offense if no jury could rationally find both that the lesser offense was committed and the greater offense was not.” (emphasis added)). And this determination— one we entirely agree with for reasons we’ve already explained — properly forecloses Mr. Grant's Beck claim.
. We are confident the error could have had little impact on the outcome for still other reasons. As we have already seen, the State’s evidence of Mr. Grant's premeditation at the guilt stage was strong, quite independent of anything Dr. Smith had to say: the evidence showed advance planning and threats, and even Mr. Grant’s own expert at the guilt phase couldn’t say he failed to understand the consequences of his actions. By the time of the penalty phase, the defense abandoned altogether any contention that the murder was the result of psychotic delusions and focused instead on Mr. Grant's personal history and amenability to treatment, making the error all the more immaterial to the proceedings.
. In connection with his Strickland claim, Mr. Grant asks us to consider new evidence he has offered for the first time only in these federal court proceedings. The evidence consists of affidavits from a family member, a friend, a psychiatrist, and a self-described "mitigation specialist” who researched Mr. Grant’s childhood. But our review under § 2254(d) is limited to the same “record that was before the state court.” Black, 682 F.3d at 895. In other words, a federal court cannot hold a state court decision to be legally or factually unreasonable in light of evidence the state court never saw. Id. Admittedly, in other circumstances "state prisoners may sometimes submit new evidence in federal court,” Pinholster, 131 S.Ct. at 1401 — though our case law has yet to clarify precisely when. Yet even if we were to view this case de novo on the amended record Mr. Grant now presents, our conclusion would not be any different. The new evidence is for the most part cumulative of the state court record. In places, as well, it would seem to do Mr. Grant as much if not more harm than good. There are indications, for example, that Ms. Carter feared Mr. Grant even before the murder, that Mr. Grant started using "PCP or harder drugs” as an adolescent, and that Mr. Grant to this day "suffers from ... outbursts of anger.”
. Of course, stale court factual findings are entitled to deference not just under § 2254(d)(2) but also under § 2254(e)(1). Just how these two provisions interact when it comes to the review of a state court's factual findings we need not decide. Because we conclude that the OCCA made no unreasonable determination of fact under (d)(2) and because (e)(1) is "arguably more deferential” than (d)(2), Wood, 558 U.S. at 301, 130 S.Ct. 841, nothing turns on the difference for purposes of our analysis.