dissenting.
I concur in all but Section IV(B) of the majority opinion, in which my esteemed colleagues hold that Hancock “has failed to prove that the OCCA’s decision was based on an unreasonable determination of the facts.” (Majority Op. 1016.) In so hold*1032ing, my colleagues fail to credit Hancock’s proof that the OCCA concluded that the trial court admitted the evidence under Okla. Stat. tit. 12 § 2609(B), when in fact the trial court expressly excluded the evidence under that section. My colleagues acknowledge that “[t]he OCCA did not tell us what it thought the district court had relied on to admit the evidence,” (Majority Op. 1016), but they mistakenly read ambiguity into this silence. They fail to recognize that the only logically consistent reading of the OCCA decision is that it is improperly predicated on a misreading of the trial court decision.
This leaves a situation in which Hancock will be executed because the majority finds the OCCA ruling to be so confusing and ambiguous that my colleagues are unable to discern what the OCCA’s factual determinations were. But judges of the OCCA are not pro se litigants whose “pleadings” must be construed liberally; they are experienced appellate jurists. • AEDPA deference is meant to prevent federal judges from reversing state-court merits decisions based on mere disagreement. This purpose is not implicated when the grounds for a state court decision are so inscrutable that they cannot be identified with some degree of certitude.1
On my review of the record, I conclude that the state trial court admitted evidence of Hancock’s 1982-manslaughter conviction and self-defense plea for impermissible reasons. On appeal, that evidentiary admission was nevertheless upheld by the OCCA based on its clear misapprehension of the trial court’s rationale. Yet, according to today’s majority, that misapprehension is unreviewable because the reasoning of the OCCA is exceptionally abstruse. This seems to me little less than compounding of error, and compels my dissent.
I
I agree with the majority that the OCCA made a decision on the merits when it reviewed the evidentiary ruling at issue for plain error. (Majority Op. 1011-12 (citing Thornburg v. Mullin, 422 F.3d 1113, 1124-25 (10th Cir.2005)).) I further agree that Hancock waived any argument that the evidentiary ruling constituted an unreasonable application of federal law. (See id. at 1016-17.) Accordingly, we may address the merits of his habeas claim de novo only if the OCCA’s decision was “based on an unreasonable determination of the facts in light of the evidence presented.” § 2254(d)(2). I agree that Hancock can meet his burden by proving that the OCCA plainly misapprehended or misstated the record. (Majority Op. 1011-12 (citing Byrd v. Workman, 645 F.3d 1159, 1171-72 (10th Cir.2011))); accord Brumfield v. Cain, — U.S. -, 135 S.Ct. 2269, 2276-77, 192 L.Ed.2d 356 (2015).
But I cannot agree that the decision of the OCCA is entitled to deference under § 2254(d)(2). The OCCA concluded that the trial court admitted Hancock’s 1982 manslaughter conviction and self-defense plea under the Oklahoma rule for impeachment by a prior felony conviction, Okla. Stat. tit. 12, § 2609(B). My colleagues and I agree that conclusion is incorrect. Although the basis of the trial court’s decision may have been unclear, what is clear is that the trial court ruled the evidence inadmissible under § 2609(B). In proceeding on the theory that the OCCA *1033might have affirmed the trial court on alternate grounds, the majority proceeds on a false premise. If the OCCA had in fact affirmed the trial court on alternate grounds, it would have said so, just as we do in our opinions. It did not do so. Q.E.D. — a false premise leads to an invalid conclusion, no matter the degree of rationalization.
A
Before trial, the state moved to admit evidence of Hancock’s 1982 manslaughter conviction and self-defense plea under § 2404(B). Under § 2404(B),
[ejvidence of other crimes, wrongs, or acts is not admissible to prove the character of a person in order to show action in conformity therewith. It may, however, be admissible for other purposes, such as proof of motive, opportunity, intent, preparation, plan, knowledge, identity, or absence of mistake or accident.
Id. The state argued that the evidence was admissible based on the “visible connection” between the 1982 manslaughter and the 2001 homicides, because “the defendant makes a claim of self-defense when he is caught committing a violent offense.” After hearing argument on the motion, the trial court ruled that the evidence was inadmissible, but that the issue could be revisited if Hancock testified.
The state then shifted tactics and moved to admit the evidence under § 2609(B). Section 2609, titled “Impeachment by Evidence of Conviction of Crime,” provides that “[ejvidence of a conviction ... is not admissible if a period of more than ten (10) years has elapsed since the date of ... the release of the witness from confinement imposed for that conviction ... unless the court determines ... that the probative value of the conviction ... substantially outweighs its prejudicial effect.” Id. At a subsequent hearing, the trial court deferred ruling on the motion until trial.
At trial, the issue resurfaced before Hancock testified. The trial court stated:
I advised you all of [sic] the past I didn’t believe it was appropriate to impeach him with, per se, the manslaughter conviction. However, the state has indicated that they were interested in impeaching — not impeaching, but questioning about the fact that he — that he claimed self-defense before when he killed somebody and then is claiming self-defense again, and therefore he’s well-versed in that — how that defense works, in essence, is my understanding of your wish to do that.
(Emphasis added.) This statement evidences confusion on the part of the trial court, which had earlier ruled that the 1982 manslaughter conviction was not admissible to show proof of knowledge under § 2404(B), but had deferred ruling on impeachment. If the purpose of admitting the 1982 manslaughter conviction was to prove Hancock’s knowledge, then it should have been admitted as part of the state’s case-in-chief, not after the state had rested and the defense was preparing to call Hancock as a witness.
Nevertheless, the prosecutor agreed that the 1982 manslaughter conviction was admissible only to show knowledge. But she said she “base[d]” her argument “on 2609.” Co-counsel affirmed that they were discussing “impeachment through evidence of a prior conviction,” and cited Dyke v. State, 716 P.2d 693, 699 (Okla.Crim.App.1986), a case involving § 2609. Perplexingly, the state’s theory of admissibility continued to change. In addition to the § 2404(B) and § 2609(B) theories, the state also argued that it would use the evidence to show “that [Hancock] offered this defense before, it’s worked for [him] *1034successfully in the past because the jury has found [him] guilty of manslaughter and gave [him] the minimum sentence allowed by law.” This theory fits neither § 2609(B) nor § 2404(B). Rather, this is a clear statement that the evidence was offered to show that Hancock acted in one way in the past, and thus had the propensity to act that way again in the future — a purpose plainly forbidden by § 2404(B).
After Hancock disputed the state’s theories, the trial court issued its ruling. It first explained that “I noticed that [Hancock] uses legal phraseology when he talks to the police in describing his actions, in describing he was in fear of great bodily injury. He, in essence, quotes the statute back.” This statement articulates why Hancock’s 1982 self-defense plea might be admissible under the knowledge exception to § 2404(B). The trial court then explicitly ruled that the evidence “doesn’t come in under the traditional impeachment by a prior conviction. It comes in under the fact that he’s claimed that defense before and it was successful for him to a great degree and that he is claiming it again.” This statement makes abundantly clear that the trial court did not admit the evidence under § 2609(B).2 The stated rationale seems to match most closely the prosecutor’s earlier argument for admissibility to show propensity, rather than knowledge. However, the trial court later instructed the jury that evidence of Hancock’s past offenses was admitted “solely on the issue of the defendant’s alleged ... knowledge.” During closing argument, the state repeatedly told the jury that the 1982 manslaughter conviction was admissible only to show Hancock’s knowledge of self-defense.
Given the trial court’s apparent confusion, it is unsurprising that neither party argued on appeal that the evidence was admitted under any one section of the Oklahoma Evidence Code. Hancock argued that “the evidence was admitted under what amounts to a hybrid theory of Sections 2404 and 2609.” Rather than defend the trial court’s ruling under either § 2404(B) or § 2609(B), the state acknowledged that “the trial court’s ruling on the admissibility of the defendant’s prior manslaughter conviction does not appear to have relied on the Evidence Code,” and that “the record does not reflect the specific law upon which the trial court rests her decision regarding this matter.” The state attempted to defend the trial court’s ruling as properly “relying upon common law principles and not a specific provision or combination of provisions from the Evidence Code.”
Federal courts cannot grant habeas relief based on a state court’s erroneous application of state law. See Estelle v. McGuire, 502 U.S. 62, 67, 112 S.Ct. 475, 116 L.Ed.2d 385 (1991). We thus need not definitively ascertain which section of the Oklahoma Evidence Code, if any, the trial court relied upon to admit the evidence of Hancock’s 1982 manslaughter conviction and self-defense plea. Instead, we must determine if the OCCA made “an unreasonable determination of the facts in light of the evidence presented.” § 2254(d)(2). To do so in this instance, we only need to know what theory the trial court did not rely upon to admit the evidence. The trial court was clear and firm on this point: “it doesn’t come in under the traditional im*1035peachment by a prior conviction” rule. (Emphasis added.)
B
Despite the trial court’s explicit statement that the evidence was inadmissible under § 2609(B), the only reasonable interpretation of the state appellate court opinion is that the OCCA nevertheless based its opinion on the proposition that the trial court did admit the evidence under that section. Although the OCCA opinion is not easy to decipher, my colleagues err by concluding it can be plausibly interpreted in two ways.
In its March 9, 2007 opinion, the OCCA first explained that it proceeded on plain error review. Hancock v. State, 155 P.3d 796, 813-14 (Okla.Crim.App.2007).3 It noted that the Oklahoma Evidence Code authorizes the admission of prior felony convictions under § 2609. Hancock, 155 P.3d at 814. In acknowledging that the parties had characterized the basis for the trial court’s ruling in adversarial positions, the OCCA cast doubt on Hancock’s “hybrid” theory, and rejected the state’s common law theory, “because the Evidence Code itself sets the terms for admissibility of prior convictions in criminal cases.” Id. It also stated that the trial court had initially rejected the state’s motion to admit the evidence to show Hancock’s knowledge under § 2404(B). Hancock, 155 P.3d at 814. Although the OCCA did not explicitly say that the trial court admitted the 1982 manslaughter conviction and self-defense plea under § 2609(B), that conclusion is necessarily implied. The OCCA described § 2609 immediately before it rejected both Hancock’s “hybrid” theory and the state’s common law theory, and stated that the Evidence Code provided a basis for admissibility. It then acknowledged that the trial court had ruled the evidence inadmissible under § 2404(B). By process of ehmination, the only section of the Evidence Code that the OCCA could possibly have concluded was the basis for the trial court’s decision is the section that it actually cited at the beginning of its analysis— § 2609(B). The majority contends that “[tjhere are many other sources of authority that the district'court might have relied on.” (Majority Op. 1014.) Yet the OCCA did not specify any other sources of authority, or hint that the trial court relied upon them. These omissions stand in sharp contrast to the voluminous textual and contextual evidence suggesting that the OCCA determined that the trial court admitted the evidence under § 2609(B).
Moreover, in its recitation of the events at trial, the OCCA explained that the state moved to admit the prior manslaughter conviction under § 2609(B) to *1036impeach Hancock if he testified. Hancock, 155 P.3d at 814. The OCCA stated that the trial court’s “subsequent instructions informed the jury Appellant’s prior convictions were offered as impeachment evidence, to show that the defendant’s testimony is not believable or truthful, and that [the] jury might consider it only to the extent that you determine it affects the believability of the defendant, if at all.” Id. (citing Oklahoma Uniform Jury Instructions — Criminal 9-23 (2d ed.)) (quotations omitted). In referencing the jury instruction, the OCCA did not distinguish between the 1982 manslaughter conviction and Hancock’s other convictions. The OCCA did not mention that the trial court also instructed the jury that evidence of Hancock’s prior offenses should be considered only to prove knowledge. Later, the OCCA reiterated that the conviction was relevant to Hancock’s credibility because he testified that he acted in self-defense. Id. at 815. Together, these statements further establish that the OCCA determined that the trial court admitted the evidence under § 2609(B). Finally, the OCCA held:
Plain errors are violations of legal rules clear from the appellate record that go to the foundation of the case or take from the defendant a right essential to his defense. Despite the creative exchange of theories for and against the District Court’s ruling, the State’s manifest purpose in offering evidence of the prior conviction and Appellant’s prior plea of self-defense was to attack the credibility of Appellant’s current plea of self-defense. The record as a whole shows the District Court weighed both the probative value of the evidence for this purpose and its prejudicial effect. Appellant’s prior conviction and his plea of self-defense were directly relevant to his credibility as a witness, as they tended to show that Appellant had minimized his responsibility when prosecuted in a previous. case and might do so again. We reject Appellant’s argument that by admitting the evidence under the particular facts and circumstances here the District Court committed plain error going to the foundation of the case or taking from Appellant a right essential to his defense.
Hancock, 155 P.3d at 815 (quotations and citations omitted).
That the OCCA was conducting plain error review is significant. It belies the majority’s suggestion that the OCCA conducted an independent analysis to conclude that the evidence was admissible under § 2609(B), and affirmed “because the result ... was correct even though the reasoning in the district court was wrong.” (Majority Op. 1012.) • Rather, the OCCA was evaluating what the trial court did. As the majority explains, the OCCA may perform quasi-de novo review through plain error analysis by analyzing whether an error is harmless. (See id. at 1014-15.) But the OCCA did not purport to do so. If the OCCA intended to conduct harmless error analysis, it would have said so.4 Only a strained and implausible reading of the OCCA opinion suggests that “the OCCA could have considered what the trial court did and still relied on the [harm*1037less error] prong of the plain-error standard.” (Id. at 1015.) The majority states that the OCCA might have determined that the trial court admitted the evidence under § 2403, and merely signaled approval of the trial court’s § 2403 balancing. (Id. at 1015.) If this was true, one would expect the OCCA to at least reference § 2403 in the relevant portion of the opinion. No such reference exists. Yet the OCCA was clearly aware of § 2403; it discussed that provision just a few paragraphs later while addressing a different evidentiary argument. Hancock, 155 P.3d at 818. Reliance on § 2403 is a post-hoc justification, but the language used by the OCCA unambiguously shows that the court was reviewing what the trial court did, not conducting an independent analysis.
Notably, the OCCA stated that “[t]he record as a whole shows the District Court weighed both the probative value of the evidence for this purpose and its prejudicial effect,” (emphasis added), referring to the state’s intended purpose of impeachment. Although the trial court did weigh the probative value and prejudicial effect of admitting Hancock’s 1982 manslaughter conviction and self-defense plea, it did so after definitively ruling the evidence inadmissible under § 2609(B). The OCCA thus made an unreasonable determination of the facts by claiming that the trial court weighed the probative value of Hancock’s 1982 manslaughter conviction “for this purpose” — the purpose of admitting the evidence under § 2609(B) — because the trial court did no such thing.5
Moreover, if the OCCA was conducting independent analysis to affirm the trial court because it reached the correct result for the wrong reason, as the majority suggests, then it would be necessary for the OCCA to address the prejudice stemming from the admission of the evidence.6 Section 2609(B) only allows admission of convictions that are more than ten years old “if the probative value of the conviction ... substantially outweighs its prejudicial effect.” (Emphasis added.) But the OCCA explained only why Hancock’s 1982 manslaughter conviction was probative. It was entirely silent as to prejudice. It is true that both reviewing the decision of the trial court and conducting independent analysis would require the OCCA to address prejudice. (See Majority Op. 1015-16 n.7.) But the omission of the weightier part of the § 2609(B) test is more difficult to explain in the context of independent review than in a deferential review of whether the trial court erred. Thus, the omission lends further support to the plain *1038meaning of the OCCA opinion: that the court was reviewing what the trial court did, not conducting independent analysis.
The only sentence in the OCCA opinion that plausibly could be construed as independent analysis states that Hancock’s “prior conviction and his plea of self-defense were directly relevant to his credibility as a witness, as they tended to show that Appellant had minimized his responsibility when prosecuted in a previous case and might do so again.” Hancock, 155 P.3d at 815 (quotations omitted). This smacks of forbidden propensity evidence: because Hancock was prosecuted for homicide and claimed self-defense in 1982, he is more likely to kill and plead self-defense in the future. Yet the OCCA itself acknowledged that propensity evidence is inadmissible under § 2609. See Hancock, 155 P.3d at 815 (“Standing alone, the 1982 manslaughter conviction had no legal relevance to whether Appellant acted in self-defense in 2001.”).
Ultimately, I cannot accept the majority’s conclusion that the OCCA’s opinion may plausibly be read as “affirming because the result (admission of the manslaughter conviction) was correct even though the reasoning in the district court was wrong.” (Majority Op. 1012.) I readily acknowledge that “§ 2254(d)(2) requires that we accord the state ... court[s] substantial deference.” Brumfield, 135 S.Ct. at 2277. But this “ ‘deference does not imply abandonment or abdication of judicial review,’ and ‘does not by definition preclude relief.’ ” Id. (quoting Miller-El v. Cockrell, 537 U.S. 322, 340, 123 S.Ct. 1029, 154 L.Ed.2d 931 (2003)). The abstruse and conclusory reasoning employed by the OCCA allows readers to project a range of more or less plausible interpretations onto the court’s words. But if a state court can effectively suspend the Great Writ by writing opinions that are so confusing that they cannot be definitively read, the promise of due process of law would be hollow indeed. Rather than adopt a liberally construed (and, in my view, manifestly inaccurate) interpretation of what the OCCA may have meant, I propose instead we take the opinion at face value. Under that facial reading, the OCCA based its decision on an unreasonable determination of the facts in light of the evidence presented in the record. § 2254(d)(2).
II
Because the OCCA’s decision was based upon an unreasonable determination of the facts, I would review the merits of the issue de novo. Magnan v. Trammell, 719 F.3d 1159, 1175 (10th Cir.2013). On de novo review, I would conclude that the admission of Hancock’s 1982 manslaughter conviction and self-defense plea rendered his trial fundamentally unfair, thereby violating his Fourteenth Amendment due process rights. See Medina v. California, 505 U.S. 437, 443-44, 112 S.Ct. 2572, 120 L.Ed.2d 353 (1992). We do not apply or consider Oklahoma evidence rules in this analysis, but rather must determine whether admission of the evidence, “considered in light of the entire record ... resulted in a fundamentally unfair trial.” Knighton v. Mullin, 293 F.3d 1165, 1171 (10th Cir.2002). We “will not disturb a state court’s admission of evidence of prior crimes, wrongs, or acts unless the probative value of such evidence is so greatly outweighed by the prejudice flowing from its admission that the admission denies the defendant due process of law.” Duvall v. Reynolds, 139 F.3d 768, 787 (10th Cir.1998).
There is arguably some marginal probative value to Hancock’s prior self-defense plea insofar as it shows that he knew about the existence of the defense of self-defense *1039when he killed Jett and Lynch. Such knowledge could have influenced the weight the jury gave his testimony. Before consulting with an attorney, Hancock apparently told police that he killed Jett and Lynch to prevent great bodily injury, showing his familiarity with the legal definition of self-defense. As the trial court remarked, “I noticed that he uses legal phraseology when he talks to the police in describing his actions, in describing he was in fear of great bodily injury. He, in essence, quotes the statute back.”
But the probative value of Hancock’s 1982 self-defense plea is extremely limited, because he could have learned about the legal definition of self-defense in many other ways. Self-defense is a concept so deeply rooted in our jurisprudence and social fabric that it has been a background assumption in not only the law, but also in popular culture, literature, and everyday conversations for centuries. See, e.g., William Shakespeare, Hamlet act 5, sc. 1 (“How can that be, unless she drowned herself in her own defense?”). Anybody who watches Law & Order or reads Agatha Christie knows that self-defense is a defense to a murder charge. One would be hard-pressed to find many people who do not know this basic legal fact. To elicit this information from Hancock, the state needed only to have asked him if he knew about self-defense. There was no need to establish the basis of his knowledge, let alone that he had pled self-defense in the past.
Although there may have been vanishingly small probative value to Hancock’s 1982 self-defense plea, I fail to see any probative value in Hancock’s 1982 manslaughter conviction. The state claims that our decision in United States v. Tan, 254 F.3d 1204 (10th Cir.2001), makes all prior convictions for similar crimes probative. But Tan does nothing of the sort. In Tan, we held that a defendant’s seven prior driving-while-intoxicated convictions were properly admitted to prove malice in a second-degree murder trial. Id. at 1210-11. We reasoned that the prior convictions tended to show that the defendant knew driving while intoxicated could harm other people, and that he therefore had the mens rea necessary to commit murder when he drove intoxicated. Id. Far from creating a blanket rule that prior convictions for similar crimes are always relevant, Tan held that a specific type of prior conviction could be used to prove mens rea for a different charge. Tan does not make Hancock’s 1982 manslaughter conviction relevant.
The state also claims support from Welch v. Sirmons, 451 F.3d 675 (10th Cir.2006). In that case, the trial court admitted evidence that the defendant had previously committed a rape and murder in order to establish a common scheme or plan. Id. at 683. The previous rape and murder were similar to the rape and murder for which the defendant was on trial. Id. The OCCA affirmed, concluding that the evidence was more probative than prejudicial. Id. at 687. We denied Welch habeas relief because it was highly relevant that both murders and rapes bore similarly distinctive characteristics. Id. at 688. The same is not true of Hancock’s prior conviction. There is little similarity between the homicides Hancock committed in 1982 and 2001. In 1982, Hancock brought a rifle to a park and used it to kill an unarmed man who had publicly humiliated him, after telling many people that he planned to kill the victim and asking around for a gun for several weeks. That is a manifestly different situation than the unarmed Hancock using Jett’s gun to kill Jett and Lynch, against whom he bore no previous grudge. As we explained in Welch, this is a situation where the “minimal relevancy” of the prior conviction *1040“suggests the possibility the evidence is being offered to show a defendant is acting in conformity with his true character.” Id. at 685.
Weighing against this miniscule probative value is the immense prejudice flowing from Hancock’s 1982 self-defense plea and manslaughter conviction. The entire case centered on the issue of whether the jury credited Hancock’s testimony that he acted in self-defense or the state’s theory that Jett withdrew, based on Shawn Tarp’s testimony.7 And the prosecution told the jury that they should not credit Hancock’s word on the issue because he had pled self-defense before. Jurors could easily have inferred that the 1982 jury did not credit his self-defense plea because he was convicted of manslaughter, further damaging his credibility. Or they could have inferred that Hancock was a bad person whose self-defense plea was not credible because of a question the state asked Hancock on cross-examination: “You were violently attacked? A man leaned into your window and you shot him five times, that’s a violent attack?” Improperly tipping the balance against the defendant on the core issue in a trial is the very definition of prejudice.
It is true that the jury was repeatedly admonished that it was only to consider Hancock’s self-defense plea as evidence of his knowledge. But “[n]o less an authority than the Supreme Court ... has said, ‘[t]he naive assumption that prejudicial effects can be overcome by instructions to the jury ... all practicing lawyers know to be unmitigated fiction.’ ” United States v. Medina-Copete, 757 F.3d 1092, 1108 n. 5 (10th Cir.2014) (quoting Burgett v. Texas, 389 U.S. 109, 115 n. 7, 88 S.Ct. 258, 19 L.Ed.2d 319 (1967)). The presumption that juries follow their instructions is rebuttable, and is “rooted less in the absolute certitude that the presumption is true than in the belief that it represents a reasonable practical accommodation.... ” Richardson v. Marsh, 481 U.S. 200, 211, 107 S.Ct. 1702, 95 L.Ed.2d 176 (1987).
In this case, any presumption that the limiting instructions cured the defect is rebutted by the sheer frequency with which the state discussed the instruction. During closing arguments, the state discussed Hancock’s 1982 self-defense plea, manslaughter conviction, and knowledge of self-defense no fewer than six times. The state urged the jury to “[rjecall back, 1982, he’s convicted of manslaughter, and he proudly told you he served two years of that four-year sentence ... [but] don’t consider that manslaughter. We don’t want to you [sic] consider that manslaughter and say, you know what, he did it before, he’s done it again.” When reminding the jury to follow its instructions, the state again stated that “evidence has been introduced of the former conviction of the defendant, that 1982 manslaughter.” . The state discredited Hancock’s statement to the police by reminding the jury to “keep in mind, when he’s talking to the police, he knows what he has to set up for self-defense, because he’s done it before, in 1982.” During its rebuttal argument, the state reminded the jury that “it is not our intent for you to decide that the defendant committed Murder in the First Degree by *1041malice aforethought because he was convicted previous [sic] in 1982 for manslaughter.” The prosecution waved Hancock's 1982 manslaughter conviction and self-defense plea in the jury’s face so many times that, even with the caveat that it was only to be considered as evidence of knowledge, the prejudicial impact is clear. There is a limit to the frequency with which a jury can be told “don’t think of an elephant” before we can but assume that the jury was being asked, with a wink and a nod, to think of an elephant.
Moreover, not all of the state’s references to the challenged evidence were limited to the theory that it proved his knowledge of self-defense. The state spun its knowledge theory into a propensity argument when it told the jury: “In 1982, when he’s convicted of that manslaughter, he puts up the same defense. He knows what to do this time around.” It told the jury that Hancock was less credible than Tarp because of the manslaughter: “[R]emember, in order to not believe Smokey, you have to believe the defendant____Who is the eight-time — seven-time convicted felon? Who knows what a defense of self-defense entails?” The prosecutor specifically highlighted Hancock’s manslaughter conviction to show that he was a convicted felon, and therefore not credible. The state also used the conviction to prove an element of the felon-in-possession of a firearm charge, even though Hancock had committed other felonies that were admissible to prove the same element.
Perhaps seeking to prevent the jury from accepting the state’s invitation to convict Hancock based on the events of 1982, Hancock’s counsel opened his closing argument by telling the jury, “what really I’m afraid of the most in this case ... [i]s Phillip Hancock’s lifestyle, his prior convictions.” It was as obvious to defense counsel at the time of Hancock’s trial as it is today that there was a significant risk the jury would convict Hancock and discredit his self-defense plea because of what happened in 1982.
The details of the 1982 incident were the first thing the state presented to the jury during the sentencing phase of Hancock’s trial. The state called multiple witnesses, who testified at length about the 1982 incident, and showed the jury pictures of the victim covered in blood. Although such evidence was permissible during the sentencing phase to prove the continuing threat aggravator, it crystallizes just how prejudicial evidence about the 1982 manslaughter conviction was when introduced and discussed extensively during the guilt phase of Hancock’s trial.
I would therefore conclude that this is the rare case in which the probative value of the evidence is “greatly outweighed” by its prejudice, such that.Hancock’s right to due process of law has been violated. Duvall, 139 F.3d at 787. Despite the prosecution’s repeated admonitions — and in part because they were so frequent — there can be no doubt but that prejudice occurred and that the jury convicted Hancock on a theory grounded on his past conviction and his asserted propensity to kill and then claim self-defense. The entire trial came down to whether the jury believed Hancock or Tarp about whether Jett retreated. And the prosecution used both Hancock’s prior self-defense plea and his 1982 manslaughter conviction to tell the jury that Hancock was not to be believed. Thus, the prejudice stemming from the admission of Hancock’s prior crime and plea went to the heart of the trial.
Ill
■I would REVERSE.
. I do not contest that the text of AEDPA unambiguously puts the burden on Hancock to prove that the OCCA made a mistake of fact. (See Majority Op. 1013 n.4. (discussing 28 U.S.C. § 2254(e)(1)).) But unlike the majority, I conclude that Hancock has met his burden.
. The trial court also "recognize[d]" that the evidence was prejudicial, but concluded that it was "more probative.” As the majority notes, such weighing sheds little light on the basis for the trial court’s ruling, because it is required for evidence admitted under § 2404(B), § 2609(B), and the § 2402 residual exception for other relevant evidence. (See Majority Op. 1013.)
. Hancock's trial counsel were unaware that they waived the issue of the admissibility of Hancock’s 1982 manslaughter conviction and self-defense plea. They filed written objections to the state’s motions to admit the evidence, argued against admissibility in several hearings, objected to the prosecutor's reference to facts underlying the 1982 conviction, and moved for a mistrial based on the admission of the evidence. The trial court repeatedly assured defense counsel that if they asked Hancock about the 1982 conviction on direct examination, they would not waive their right to appeal the decision to admit the evidence. But both Hancock's trial counsel and the trial court were incorrect. As the OCCA explained, under Oklahoma law, defense counsel waived the error by broaching the issue on direct examination. See Hancock, 155 P.3d at 813-14 (quoting Dodd v. State, 100 P.3d 1017, 1039 (Okla.Crim.App. 004)).
In both his direct appeal to the OCCA and in his habeas petition to the federal district court, Hancock alleged that his trial counsel were ineffective for broaching the issue on direct examination. Both courts denied his claim. See Hancock, 155 P.3d at 821. Hancock has not pursued a certificate of appealability on this issue.
. The majority concludes that the OCCA’s failure to specify which prong of the plain error test it relied upon indicates only that “the OCCA wrote the opinion differently than we might have.” (Majority Op. 1015 n.5.) I respectfully disagree. Because the OCCA was not explicit about the factual basis of its opinion, I consider the context of the OCCA opinion for indicia of that factual basis. That context unambiguously indicates that the OCCA was analyzing what the trial court did, rather than conducting harmless error analysis.
. The majority distinguishes between impeaching the credibility of Hancock and impeaching the credibility of his self-defense plea. (Majority Op. 1016.) This is a distinction without a difference. Perhaps there is a situation in which the credibility of a plea could be attacked without also attacking the credibility of the person making the plea. But in this context, it was manifestly Hancock's credibility that was at issue. The trial court admitted the evidence only because Hancock himself testified. Discussion of the evidence before the trial court centered on Hancock’s credibility. Nothing in the OCCA opinion indicates that Hancock’s credibility is ' in any way distinct from the credibility of his self-defense plea.
. If the OCCA was conducting independent review to determine that the evidence was admissible under § 2609(B), I would also expect it to at least reference its own test for conducting that analysis. See Croney v. State, 748 P.2d 34, 37-38 (Okla.Crim.App.1987) (instructing courts to consider "the nature of the conviction itself," whether it has "some bearing on the accused's propensity to tell the truth,” and “the similarity between the previous conviction and the present charge”). Yet the OCCA neither considered the applicable factors nor cited Croney or its progeny.
. As such, I reject the state’s argument that any error in admitting evidence of the 1982 self-defense plea and manslaughter conviction was harmless. See Webber v. Scott, 390 F.3d 1169, 1177 (10th Cir.2004) (concluding that due process violations reviewed de novo in a habeas case can be considered harmless error). All parties recognized from the outset that, because the case turned on whether the jury credited Hancock’s or Tarp’s testimony, there was a very real possibility that a jury might convict Hancock based on his past crimes, rather than what happened in 2001.