Full opinion text
PER CURIAM: Richard A. Leavitt, a State of Idaho prisoner under sentence of death, brought a petition for habeas corpus in the district court. 28 U.S.C. § 2254. He filed a myriad of attacks on his conviction and sentence, ranging from alleged evidentiary errors through instructional errors and onto attacks on the Idaho death penalty scheme. He also asserted ineffective assistance of counsel. The district court granted habeas corpus relief on one claim: the assertion that a burden of proof instruction violated Leavitt’s due process rights. However, it denied relief as:to all of his other claims. The State of Idaho appeals the former, and Leavitt appeals the latter. We reverse as to the former, affirm as to all of the latter, with the exception of an ineffective assistance of counsel claim, and remand for further proceedings. BACKGROUND In the small town of Blackfoot, Idaho, on July 17, 1984, the victim ’ of this brutal crime, Danette Elg, was viciously attacked in her own bedroom by a knife-wielding assailant. The relentless and merciless assault took place on her waterbed and with such implacable force that the bed itself was punctured and torn, while the victim sustained numerous cuts and slashes as she fought for her life. She was also stabbed multiple times: One thrust caused the knife to enter her right lung, another the right side of her heart, still another her left lung, and others penetrated her stomach, her chest cavity, and her neck. One even went through her eye and into her brain. Another exceedingly peculiar and unique wound inflicted during this attack was a' cut made by the attacker through which he then removed her sexual organs. He did that in a manner that showed that he had some knowledge of female anatomy, for it was done in a manner that is difficult to accomplish. The evidence pointing to Leavitt was powerful, if circumstantial — he was not caught redhanded, nor did he confess. Unfortunately, the victim’s body was not found for several days which caused the destruction of some evidentiary markers, but gave rise to others. On the night of July 16, the victim had been severely frightened and shaken when a prowler tried to enter her home. She called the emergency 911 number and the police came, but they found nothing other than signs of attempted entry and a petrified young lady, who thought that Leavitt was the culprit. They then searched the area and the town but, alas, failed to find Leavitt. Strangely enough, during the period between the murder and the discovery of the body with Leavitt’s help, he became exceedingly “interested” in the victim’s whereabouts. He finally obtained permission to enter the house with the police and discovered the body. Another strange aspect of the case was that a person supposedly named Mike Jenkins also called the police a couple of times during that period and showed knowledge of details of the crime that only the killer himself would know. Mike Jenkins was not known in Blackfoot and was not heard of thereafter. Leavitt, however, is adept at disguising his voice on the telephone, and could even fool his own wife when he did so. What else? On the very night of the killing, Leavitt suffered a severe cut to his finger, for which he was treated in an emergency room. The killer was also wounded and left behind his blood — Type 0 — which was mixed with the blood of his hapless victim — Type A. Of all the possible suspects, the only likely source of the Type 0 blood was Leavitt himself. How could that damning connection be explained? Well, said Leavitt, he had somehow cut his hand on a fan at home — a story that was shown to be a lie. At trial he changed that to a story that he had really sustained the cut while preventing his wife from committing suicide. And the crime scene blood? Leavitt could not, at first, imagine how his blood could have been found there, but he had an epiphany by the time of trial. At trial, he managed to recall that a week before the killing he had a nosebleed in the victim’s bedroom. That, supposedly, resulted in his blood being mixed with hers when she was killed on her bed a week later. It also supposedly explained how his blood was elsewhere in her room — on the walls and at the window, and even on her underclothes — he wiped his nose on them — as well as on shorts that she had worn between the date of the “nosebleed” and the date of her death. Along the way, Leavitt also tried to send his wife a letter from jail in which he sought to have her memorize a story he had concocted, which would, not surprisingly, tend to exculpate him. Neither the jury, nor any court which has since reviewed the evidence in this case, has been impressed with Leavitt’s stories. The jury found him guilty, and an Idaho judge sentenced him to death. The Idaho Supreme Court affirmed and denied post-conviction relief. But the district court found error in the jury instructions and issued the writ of habeas corpus, and these appeals followed. STANDARD OF REVIEW The district court had jurisdiction pursuant to 28 U.S.C. § 2254. We have jurisdiction pursuant to 28 U.S.C. §§ 1291, 2253(c). We review the district court’s decision to grant or deny a petition for habeas corpus de novo. Martinez-Villareal v. Lewis, 80 F.3d 1301, 1305 (9th Cir.1996). “To the extent it is necessary to review findings of fact made in the district court, the clearly erroneous standard applies.” Silva v. Woodford, 279 F.3d 825, 835 (9th Cir.2002). As usual, clear error review is “significantly deferential,” and “we must accept the district court’s factual findings absent a ‘definite and firm conviction that a mistake has been committed.’ ” Id. (quoting United States v. Syrax, 235 F.3d 422, 427 (9th Cir.2000)). Further, “[although less deference to state court factual findings is required under the pre-AEDPA law which governs this case, such factual findings are nonetheless entitled to a presumption of correctness unless they are ‘not fairly supported by the record.’ ” Id. (quoting Bean v. Calderon, 163 F.3d 1073, 1087 & n. 3 (9th Cir.1998)). Finally, we “may affirm on any ground supported by the record even if it differs from the rationale of the district court.” Martinez Villareal, 80 F.3d at 1305. Because the First Amended Habeas Corpus petition was filed before April 24, 1996, the Antiterrorism and Effective Death Penalty Act does not apply to this case. See Reutter v. Crandel 109 F.3d 575, 577 (9th Cir.1997). DISCUSSION While we are faced with numerous issues, if the district court properly granted habeas corpus on the innocence instruction issue, the others fall by the wayside. Thus, we will first take up the state’s appeal of the decision of that issue. We will thereafter consider the others. I. THE INNOCENCE INSTR UCTION The state argues that the district court created a new rule of law in violation of Teague v. Lane, 489 U.S. 288, 109 S.Ct. 1060, 103 L.Ed.2d 334 (1989), when it concluded that instruction 12 unconstitutionally lowered the state’s burden of proof. Leavitt responds that the state waived reliance on Teague, miscalculated the date of finality, and overlooked the fact that the controlling opinion of the United States Supreme Court is In re Winship, 397 U.S. 358, 90 S.Ct. 1068, 25 L.Ed.2d 368 (1970)— decided almost fifteen years before this case arose. Thus, in his view, the principle that he advocates (and that the district court adopted) is. neither a new rule nor otherwise exceptionable because there is a reasonable likelihood that the jury understood the instructions as a whole to allow, conviction based on proof less demanding than proof beyond a reasonable doubt. If a state properly argues that the district court granted a habeas petition on the basis of a new rule of constitutional law that is Teague-barred, we must address the Teague issue first. Caspari v. Bohlen, 510 U.S. 383, 389, 114 S.Ct. 948, 127 L.Ed.2d 236 (1994); Horn v. Banks, 536 U.S. 266, 267, 122 S.Ct. 2147, 153 L.Ed.2d 301 (2002) (per curiam). Here, the state has not waived the issue and so we should conduct a Teague analysis. This means that we must determine when Leavitt’s conviction became final; survey the legal landscape at that time to see whether the rule he advocates was dictated or compelled by existing precedent; and if not, consider whether that relief falls within one of two exceptions to nonretroactivity on habeas review. See Lambrix v. Singletary, 520 U.S. 518, 527, 117 S.Ct. 1517, 137 L.Ed.2d 771 (1997) (summarizing steps). A Whether the rule is “new” depends upon whether it was dictated by controlling precedent at the time when Leavitt’s conviction became final in 1989, not when his sentence became final in 1992. This is because the guilt phase and sentencing phase were bifurcated, Gretzler v. Stewart, 112 F.3d 992, 1004 (9th Cir.1997), and 1989 is when the Idaho Supreme Court rendered its guilt-phase decision and the time for petitioning for certiorari had passed. United States v. Colvin, 204 F.3d 1221, 1224 (9th Cir.2000), is not to the contrary because it involved the finality of a judgment for purposes of the statute of limitations under 28 U.S.C. § 2255. By contrast, in the context of collateral review of a bifurcated decision, finality should be measured from the time when the decision under review — be it the conviction or the sentence — was actually made because the whole purpose of Teague is to “validate[ ] reasonable, good-faith interpretations of existing precedents made by state courts even though they are shown to be contrary to later decisions.” O’Dell v. Netherland, 521 U.S. 151, 156, 117 S.Ct. 1969, 138 L.Ed.2d 351 (1997) (internal quotation marks omitted); see also id. (“[W]e will not disturb a final state conviction or sentence unless it can be said that a state court, at the time the conviction or sentence became final, would have acted objectively unreasonably by not extending the relief later sought in federal court.”); Graham v. Collins, 506 U.S. 461, 468, 113 S.Ct. 892, 122 L.Ed.2d 260 (1993); Saffle v. Parks, 494 U.S. 484, 488, 110 S.Ct. 1257, 108 L.Ed.2d 415 (1990) (instructing that we survey the legal landscape as it then existed to “determine whether a state court considering [the defendant’s] claim at the time his conviction became final would have felt compelled by existing precedent to conclude that the rule [he] seeks was required by the Constitution”). B The rule that Leavitt advocates, and that the district court adopted, is that there was a reasonable likelihood that the jury understood the instructions as a whole to allow conviction based upon proof less demanding than proof beyond a reasonable doubt. En route, Leavitt argues (and the district court held) that instruction 12 is erroneous because it eroded the reasonable doubt standard by allowing the jury to convict if it believed that he was “in fact” guilty. The principle that the Due Process Clause requires proof of guilt beyond a reasonable doubt was established in Win-ship. And the principle that an instruction may not shift the burden of proof or lift it by a presumption as to an element of the crime charged was established in Sandstrom v. Montana, 442 U.S. 510, 99 S.Ct. 2450, 61 L.Ed.2d 39 (1979). However, the principle upon which Leavitt (and the district court) rely for habeas relief — that there was a reasonable likelihood that the jury interpreted the instructions as a whole to allow a finding of guilt based on a degree of proof below that required by the Due Process Clause — was established in Cage v. Louisiana, 498 U.S. 39, 41, 111 S.Ct. 328, 112 L.Ed.2d 339 (1990) (per curiam). Cage pretty clearly created a new rule. See, e.g., Tyler v. Cain, 533 U.S. 656, 665, 121 S.Ct. 2478, 150 L.Ed.2d 632 (2001) (noting the habeas petitioner’s concession to this effect); Victor v. Nebraska, 511 U.S. 1, 5, 114 S.Ct. 1239, 127 L.Ed.2d 583 (1994) (acknowledging that only in-Cape had the Court held that a definition of reasonable doubt violated the Due Process Clause); see also Tillman v. Cook, 215 F.3d 1116, 1122 (10th Cir.2000) (holding that a habeas petitioner who was convicted in the 1980s, and who argued on habeas that his jury instruction misdefined reasonable doubt, was necessarily seeking a “new rule” because Cage was decided after he was convicted); Gaines v. Kelly, 202 F.3d 598, 602-03 (2d Cir.2000) (observing that it was not until 1990 in Cage that the Supreme Court first held that a state trial court’s definition of reasonable doubt violated constitutional due process, and thus holding that Cage announced a new rule within the meaning of Teague). Therefore, Leavitt seeks the benefit (and the district court gave him the benefit) of a “new” rule, as Cage was issued after Leav-itt’s conviction became final. Existing precedent did not “dictate” or “compel” the conclusion that there was a reasonable likelihood that the jury interpreted the instructions to allow for conviction by proof less than proof beyond a reasonable doubt. Reasonable jurists in 1989 would have known that a flawed instruction must be viewed in the context of the instructions overall.. See Cupp v. Naughten, 414 U.S. 141, 147, 94 S.Ct. 396, 38 L.Ed.2d 368 (1973). Reasonable jurists assessing the overall effect of Leavitt’s jury charge would have realized that it contained numerous individual instructions that correctly defined reasonable doubt and stressed the importance of finding every element beyond a reasonable doubt based solely on the evidence presented at trial. And reasonable jurists would have been aware that the Supreme Court had never held that an instruction like instruction 12 was erroneous, let alone constitutionally so — either by itself, or in combination with other instructions on the burden of proof. A good argument can be made that we should start and stop with the law as determined by the Supreme Court. See Bell v. Hill, 190 F.3d 1089, 1093-97 (9th Cir.1999) (Rymer, J., dissenting). A Teag-ue analysis applies to Supreme Court decisions, see Lambrix, 520 U.S. at 538, 117 S.Ct. 1517, and the state courts of Idaho were (and are) not bound to follow Ninth Circuit law. However, we have held that “circuit court holdings suffice to create a ‘clearly established’ rule of law under Teague.” Belmontes v. Woodford, 350 F.3d 861, 884 (9th Cir.2003) (citing Bell). Regardless of whether this view is right or wrong, it is law by which we are bound. It is also true that while the Supreme Court typically refers to its own prior cases in conducting a Teague analysis, see, e.g., Caspari, 510 U.S. at 391, 114 S.Ct. 948; Graham, 506 U.S. at 466-77, 113 S.Ct. 892; O’Dell, 521 U.S. at 160-64, 117 S.Ct. 1969; Saffle, 494 U.S. at 489-94, 110 S.Ct. 1257, the Court does not ignore “the experience of the lower courts” as illuminating whether the rule contended for is a development in the law over which reasonable jurists could disagree. See, e.g., Lambrix, 520 U.S. at 538, 117 S.Ct. 1517; Caspari, 510 U.S. at 393-95, 114 S.Ct. 948; Stringer v. Black, 503 U.S. 222, 236-37, 112 S.Ct. 1130, 117 L.Edud 367 (1992); Butler v. McKellar, 494 U.S. 407, 415, 110 S.Ct. 1212, 108 L.Ed.2d 347 (1990). So circuit holdings need to be (or at minimum, can be) considered. We had held before Leavitt’s conviction was final that instructing a jury that the presumption of innocence is not intended to aid the guilty-in-fact was “prejudicial error.” Reynolds v. United States, 238 F.2d 460, 463 (9th Cir.1956). However, we did not hold that the instruction was constitutional error. As the Supreme Court has made clear, it is not enough that an instruction is “undesirable, erroneous, or even ‘universally condemned’ ” —it must have violated some constitutional right. Estelle, 502 U.S. at 72,112 S.Ct. 475 (quoting Donnelly v. DeChristoforo, 416 U.S. 637, 643, 94 S.Ct. 1868, 40 L.Ed.2d 431 (1974)). In addition, Reynolds was decided on direct appeal, which means that it did not determine how a reasonable jury likely interpreted the charge, and it considered the challenged instruction in isolation rather than in the context of the instructions as a whole. Thus, Reynolds says nothing about whether the prejudice from a similar instruction can be cured by other instructions. Reasonable jurists could have interpreted the law to allow this, as the Second Circuit did in United States v. Doyle, 130 F.3d 523, 539 (2d Cir.1997), and as we did in Shaw v. United States, 244 F.2d 930 (9th Cir.1957). In Shaw, we upheld a conviction despite the fact that the trial court had given substantially the same instruction as in Reynolds. We reaffirmed our previous disapproval of the instruction, but held that “we are not bound to reverse in every case where the instruction may have been given.” Id. Accordingly, it was evident in 1989 that we disapproved an individual instruction similar to instruction 12, but this does not dictate a conclusion that Leavitt’s jury likely interpreted their set of instructions to allow for conviction on proof less than beyond a reasonable doubt. Therefore, assuming that our circuit decisions are appropriately part of the mix that reasonable jurists would have considered in 1989, and that they may have “inform[ed], or even controlled] or governed], the analysis” that we would have applied to a case such as Leavitt’s, Saffle, 494 U.S. at 491, 110 S.Ct. 1257, reasonable jurists in 1989 would still not have felt compelled by Reynolds and Shaw to find that instruction 12 was constitutional error that automatically required reversal without regard to the entire package of instructions and the record as a whole. Other federal courts of appeals had considered similar instructions, but no consensus had emerged by 1989. Compare Gomila v. United States, 146 F.2d 372, 373 (5th Cir.1944) (“The statement that the presumption of innocence ‘was not intended, nor has it ever been intended, as extending an aid to one, who in fact is guilty ... ’ ” is not a correct statement of the law.), and United States v. Bridges, 499 F.2d 179, 186 (7th Cir.1974) (objectionable to state that “reasonable doubt” is not for the purpose of “permitting guilty men to escape”), with Moffitt v. United States, 154 F.2d 402, 404-05 (10th Cir.1946) (finding nothing in instruction that “the presumption of innocence was ‘not intended to shield those who are actually guilt/ ” from which it could be inferred that the presumption does not benefit the guilty defendant as well as the innocent), and United States v. Farina, 184 F.2d 18, 20-21 (2d Cir.1950) (instruction that the presumption of innocence “was not intended as a bulwark behind which the guilty might hide” would not “lead a jury to suppose the presumption could not be invoked until a defendant had dispelled proof of his guilt”). Therefore, assuming that this universe of case law counts for purposes of Teague, a 1989 Idaho state court would not have felt compelled to conclude that there was a reasonable likelihood that Leavitt’s jury understood the charge as a whole to allow a finding of guilt based on a degree of proof below that required by the Due Process Clause. State-court decisions (assuming they have some relevance as a reference point for determining what federal law was established at the time) were split as well. Compare Gilleylen v. State, 255 So.2d 661, 664 (Miss.1971) (reversible error to give instruction that the “presumption of innocence ... is not' intended to shield, from punishment anyone who is in fact guilty”), with State v. Farnsworth, 51 Idaho 768, 10 P.2d 295, 299 (1932) (upholding essentially the same instruction given in this case, and in State v. Gilbert, 8 Idaho 346, 69 P. 62, 64 (1902)). Since 1989 two state courts have rejected a constitutional challenge to a similar instruction, which indicates that the instruction was not regarded as unacceptable in those states as of 1989. See Sipress v. State, 562 N.E.2d 758, 762 (Ind.Ct.App.1990) (noting that an instruction that “the presumption of innocence and requiring the state to establish beyond .a reasonable doubt, every material fact ... is not intended to shield those who are actually guilty” was approved by the Indiana Supreme Court in Heald v. State, 492 N.E.2d 671 (Ind.1986)); State v. Schiappa, 248 Conn. 132, 728 A.2d 466, 486 (1999) (disapproving, but rejecting a constitutional challenge to, an instruction that “the principle requiring the state to establish guilt beyond a reasonable doubt is a ‘rule of law ... made to protect the innocent and not the guilty’ ” (omission in original)). In sum, jurists in 1989 considering Supreme Court, circuit, and state court precedent would not have felt compelled to hold that Leavitt’s jury convicted him on a diluted burden of proof solely because some courts had disapproved instruction 12. Even so, Leavitt contends that the other reasonable doubt instructions were themselves fraught with error, such that they could not undo the misleading impression left by instruction 12. In particular, he faults instructions 10, 11, 13, 36 and 39, which he claims (and the district concluded) were confusing, ambiguous, and possibly misleading to the jury. We disagree. The trial judge unequivocally told the jury at least nine times that the prosecution had the burden of proving the crime beyond a reasonable doubt. Set against these correct admonitions, existing precedent did not compel the conclusion argued for by Leavitt (and accepted by the district court) that, taken as a whole, the giving of instruction 12 created a reasonable likelihood that the jury interpreted the instructions as allowing Leavitt to be convicted on proof less than beyond a reasonable doubt. Whatever error there was in instruction 10 was immediately cured. To the extent that the judge’s statement at the beginning of this instruction that the jury “should” require proof beyond a reasonable doubt may have misstated the obligation, the judge immediately followed up by explaining that if “you entertain a reasonable doubt of the truth of any one of these material allegations, then it is your duty to give the Defendant the benefit of such doubt and acquit him,” and by summing up with the unequivocal statement: “There must be proof beyond a reasonable doubt.” Thus, even if a layperson would have understood “should” as precatory rather than mandatory, any such impression was promptly corrected. Cf. Boyde v. California, 494 U.S. 370, 381, 110 S.Ct. 1190, 108 L.Ed.2d 316 (1990) (cautioning against “technical hairsplitting” of jury instructions). Instruction 11 is virtually identical to the reasonable doubt instruction upheld in Vic- tor v. Nebraska, 511 U.S. 1, 7, 114 S.Ct. 1239, 127 L.Ed.2d 583 (1994), that also defined reasonable doubt in terms of “moral evidence,” “moral certainty,” and “not a mere possible doubt.” Cf. Cage, 498 U.S. at 41, 111 S.Ct. 328 (holding unconstitutional a differently worded instruction that used the terms “grave uncertainty,” “actual substantial doubt,” and “moral certainty,” and that was not surrounded by other, correct instructions that gave content to those terms). Instruction 13 is and always has been a perfectly correct statement of the law; the prosecution need not prove every fact in the case beyond a reasonable doubt so long as it proves every element beyond a reasonable doubt. See, e.g., Harris v. United States, 536 U.S. 545, 549, 122 S.Ct. 2406, 153 L.Ed.2d 524 (2002) (explaining that “not all facts ... are elements,” and that such facts “are thus not subject to the Constitution’s indictment, jury, and proof requirements”). And instruction 36 would not have left jurors confused about their duty to acquit if they entertained a doubt that was reasonable rather than derived from “fanciful suppositions” or “remote conjectures as to possible ... facts different from those established by the evidence.” Cf. Victor, 511 U.S. at 5, 114 S.Ct. 1239 (“[T]he Constitution does not require that any particular form of words be used in advising the jury of the government’s burden of proof.”). But cf. State v. Holm, 93 Idaho 904, 478 P.2d 284, 287-88 (1970) (disapproving this instruction, but declining to hold it unconstitutional). Instniction 39 is more troublesome because it imposed the burden of proving an alibi on Leavitt, which is clearly wrong. See Thomas v. United States, 213 F.2d 30, 33-34 (9th Cir.1954) (citing cases from six circuits to this effect). However, it is not reasonably likely that this instruction, as part of the package of instructions, caused Leavitt’s jury to base his conviction on a degree of proof below that required by the Due Process Clause. Instruction 39 by its terms pertained only to the alibi defense; the jury was otherwise clearly instructed that the prosecution had the burden of proving that Leavitt committed murder beyond a reasonable doubt, and that “the law never imposes upon a Defendant in a criminal case the burden or duty of calling any witnesses or producing any evidence.” There is no question that Leavitt had to be at Elg’s house on the evening of July 17, 1984, to commit the crime. In other words, Leavitt’s presence there was more or less the flip side of the prosecution’s burden to show that it was he who did it. To this extent the instructions have some internal inconsistency and could allow a jury to misunderstand the prosecution’s burden. Cf. Stump v. Bennett, 398 F.2d 111, 121-22 (8th Cir.1968) (a flawed alibi instruction can undermine a jury’s understanding of reasonable doubt). But it is not reasonably likely that this jury did misunderstand the burden of proof, or that instruction 39 contributed to any confusion about the burden of proof required to convict, for two reasons. First, assuming there was any misunderstanding, it would extend only to Leavitt’s obligation to come forward with evidence to create a reasonable doubt; it would not extend to the government’s burden of persuasion." Instruction 39 did not impose any burden upon Leavitt himself to persuade the jury that he was not present beyond a reasonable doubt, or by a preponderance of the evidence. The instruction simply said (albeit infelieitously) what jurors would figure anyway, that the defendant — who would know where he was if he asserted an alibi — would be expected to produce some evidence that he was somewhere other than the scene of the crime. If believed, that evidence would create a reasonable doubt that Leavitt did it. Second, for all practical purposes, there was no alibi. Leavitt testified that he was at home watching television when Elg was murdered, yet neither his wife nor anyone else corroborated this story. He was severely impeached by having lied about'cutting his finger on a fan, and by a letter that he wrote in jail making up a time-line for his wife to memorize. Most damning of all, Leavitt’s blood was mixed with the victim’s blood, for which there is no rational explanation other than that he was there when she spilt it. Therefore, in this case, imposing on' Leavitt the burden of creating reasonable doubt of his guilt, by supporting an alibi, could have played no significant role in the jury’s understanding of the requirement that it find guilt beyond a reasonable doubt. Accordingly, reasonable jurists in 1989 would not have felt compelled to hold that, on account of instruction 12 in the context of the instructions as a whole, there was a reasonable likelihood that the jury interpreted the instructions to allow conviction by proof less than beyond a reasonable doubt. Because it would be a new rule to decide so now, we must determine whether it should nevertheless apply retroactively to Leavitt’s case. If not, there is no need to consider Leavitt’s remaining argument that structural error- — which a Cage error is — precludes harmless error analysis even on habeas review. c Under Teague, “a new rule can be retroactive to cases on collateral review if, and only if, it falls within one of two narrow exceptions to the general rule of non-retroactivity.” Tyler v. Cain, 533 U.S. 656, 665, 121 S.Ct. 2478, 150 L.Ed.2d 632 (2001). One is for determinations that private conduct is beyond the power of the state to proscribe; the other, which is the relevant exception here, is for “watershed rules of criminal procedure implicating the fundamental fairness and accuracy of the criminal proceeding.” Id. (internal quotation marks omitted). To fall, under the “watershed” exception, a new rule must meet two requirements: “[1] Infringement of the rule must seriously diminish the likelihood of obtaining an accurate conviction, and [2] the rule must alter our understanding of the bedrock procedural elements.” Id. (emphasis and internal quotation marks omitted). Whether the district court announced (or we would be announcing if we affirmed) a “watershed” rule by holding that the jury charge given at Leavitt’s trial was unconstitutional under Winship and Cage is an open question in the Ninth Circuit. In Ramirez v. Hatcher, 136 F.3d 1209 (9th Cir.1998), the habeas petitioner argued that the reasonable doubt instruction given at his trial was unconstitutional in light of Cage, but we bypassed the state’s Teague argument and upheld the challenged instruction on the merits. We also considered a similar question in Harmon v. Marshall, 69 F.3d 963, 967 (9th Cir.1995) (per curiam), where we held that a habeas petitioner-could benefit retroactively from an erroneous instruction that failed to define any elements of the charged crime. But Harmon does not shed light on the distinct question of whether Cage is a “watershed” rule. Although we have never decided whether to apply Cage retroactively on habeas review, six other circuits have and they all concluded that. Cage announced a watershed rule primarily on the footing that Sullivan held that a Cage error is structural error. See Tillman v. Cook, 215 F.3d 1116, 1121-22 (10th Cir.2000) (relying on Sullivan and holding “[l]ike the Second, Third, Fourth, Fifth, and Eleventh Circuits, ... that the remedy for an unconstitutional reasonable doubt instruction must be applied retroactively”); West v. Vaughn, 204 F.3d 53, 55, 61 (3d Cir.2000) (Sullivan left no doubt that Cage retroactively applies to habeas petitions), overruled by Tyler, 533 U.S. 656, 121 S.Ct. 2478, 150 L.Ed.2d 632; Gaines v. Kelly, 202 F.3d 598, 604-605 (2d Cir.2000) (relying on Sullivan and concluding that, “because any criminal conviction rendered pursuant to an unconstitutional definition of reasonable doubt is necessarily unfair, ... the rule advanced by Gaines falls within the second exception to the Teague doctrine”) (citation omitted); Humphrey v. Cain, 138 F.3d 552, 553 (5th Cir.1998) (en banc) (adopting the reasoning of Humphrey v. Cain, 120 F.3d 526, 529 (5th Cir.1997), that Sullivan “made it plain that Cage-Victor errors fit with the second Teague exception”); Adams v. Aiken, 41 F.3d 175, 178-79 (4th Cir.1994) {Adams TV) (in light of Sullivan, “the rule that a constitutionally deficient reasonable doubt instruction violates the . Due Process Clause satisfies Teague’s second exception”); Nutter v. White, 39 F.3d 1154, 1157 n. 5, 1158 (11th Cir.1994) (prior view of two circuits that Cage did not have retroactive effect on collateral review “does not survive Sullivan ”). See also Tyler, 533 U.S. at 671-72, 121 S.Ct. 2478 (Breyer, J., dissenting) (explaining why “[t]o reason as the Court reasoned in Sullivan is to hold (in Teague’s language) (1) that infringement of the Cage rule ‘seriously diminish[es] the likelihood of obtaining an accurate conviction,’ and (2) that Cage ‘alter[s] our understanding of the bedrock procedural elements’ that are essential to the fairness of a criminal trial”) (internal citations omitted) (alteration in original); Ramirez, 136 F.3d at 1216 (Reinhardt, J., dissenting) (concluding that “[t]he Supreme Court’s .unanimous decision in [Sullivan ] leaves no doubt that the Cage rule meets this standard”). Ordinarily; the unanimity of six circuits on a question of first impression for us would counsel against our reaching a contrary result. Cf. Zimmerman v. Oregon Dep’t. of Justice, 170 F.3d 1169, 1184 (9th Cir.1999) (“We realize that our decision creates an inter-circuit split of authority. [W]e are hesitant to create such a split, and we do so only after the most painstaking inquiry....”). However, there is a compelling reason to do so here. These decisions were all reached between 1994 and 2000 — after Sullivan had held that Cage error is structural, but before the Court indicated in Tyler that “a holding that a particular error is structural does not logically dictate the conclusion that the second Teague exception has been met.” Tyler, 533 U.S. at 666-67, 121 S.Ct. 2478. In light of Tyler, pre-Tyler circuit authority to the contrary is no longer persuasive. Considering the issue afresh, it is clear that the first “watershed” requirement is met because a defective reasonable doubt instruction affects the accuracy of the finding of guilt beyond a reasonable doubt. Indeed, it destroys it. Misdescribing the burden of proof “vitiates all the jury’s findings,” has “consequences that are necessarily unquantifiable and indeterminate,” and transforms appellate review into “pure speculation.” Sullivan, 508 U.S. at 281-82, 113 S.Ct. 2078. In short, when a Cage error is committed, “a criminal trial cannot reliably serve its function.” Id. at 281, 113 S.Ct. 2078. From this it necessarily follows that a defective reasonable doubt instruction seriously decreases the likelihood of obtaining an accurate conviction. Cf. Gaines, 202 F.3d at 604 (“[I]t logically follows [from Sullivan ] that the reasonable doubt rule flowing from Cage plays a pivotal role in ensuring the accuracy of the findings that underlie any criminal conviction rendered by a jury.”). However, we now know that it does not necessarily follow from Sullivan that the Cage rule alters our understanding of bedrock procedures. Tyler, 533 U.S. at 666-67 & n. 7, 121 S.Ct. 2478. The second Teague exception is reserved for only a “small core of rules,” Graham, 506 U.S. at 478, 113 S.Ct. 892, that can truly be categorized as “groundbreaking.” Caspari, 510 U.S. at 396, 114 S.Ct. 948. The “watershed” exception does not apply to every rule that promotes accuracy and guarantees due process. Cf. Tyler, 533 U.S. at 667 n. 7, 121 S.Ct. 2478 (“Nor can it be said that all new rules relating to due process (or even the ‘fundamental requirements of due process’ ...) alter such understanding.”). Indeed, in the thirteen years since Teague was decided, the Supreme Court has never found that any rule falls within the “watershed” exception despite at least eleven opportunities to do so. See United States v. Mandanici, 205 F.3d 519, 529 (2d Cir.2000) (“Beginning with the rule at issue in Teague, the Court has measured at least eleven new rules, or proposed new rules, of criminal procedure against the criteria for the second exception and, in every case, has refused to apply the rule at issue retroactively.”); see also United States v. Sanchez-Cervantes, 282 F.3d 664, 669 n. 23 (9th Cir.2002) (“[T]he Supreme Court has not found any rule to qualify under the second exception since Teague came out.”). But cf. Saffle, 494 U.S. at 495, 110 S.Ct. 1257 (indicating that the guarantee of counsel in criminal trials established by Gideon v. Wainwright, 372 U.S. 335, 83 S.Ct. 792, 9 L.Ed.2d 799 (1963), is an example of the type of rule that would fall within Teague’s second exception). And in 2001, in Tyler, the Court reiterated that “it is unlikely that any of these watershed rales ha[s] yet to emerge.” Tyler, 533 U.S. at 667 n. 7, 121 S.Ct. 2478 (citation and internal quotation marks omitted) (alteration in original). It follows that it is “unlikely” that Cage is a watershed rule. Accordingly, even if we were to hold that instruction 12 and the instructions as a whole were unconstitutional in light of Cage because they were ambiguous and misled thé jury as to the meaning of proof beyond a reasonable doubt, we would not be making a sweeping or groundbreaking pronouncement. We would simply be holding that Leavitt’s trial was infirm because the instructions were so ambiguous and confusing that it is reasonably likely the jury understood them overall to allow conviction on proof less than beyond a reasonable doubt, which would, of course, violate due process under Winship. Given that Cage error of the type claimed here would be a new rale not available at the time Leavitt’s conviction became final, but not a watershed rule of the sort from which he can retroactively benefit on federal habeas review, his claim is Teague-barred. Therefore, it is unnecessary to determine whether the jury instruction actually ran afoul of Cage. For these reasons, the district court could not grant the petition and,its order must be reversed. II. PRETRIAL PUBLICITY Leavitt complains that pretrial publicity in Blackfoot required a change of venue. The Idaho trial court and the Idaho Supreme Court held to the contrary. The latter reviewed the record and concluded: “[T]he pretrial publicity had little if any effect on the potential jurors, and [we] find no indication that potential jurors would prejudge the case.” Leavitt I, 116 Idaho at 288, 775 P.2d at 602. We, too, “must independently examine the news reports for volume, content, and timing to determine if they were prejudicial.” Jeffries v. Blodgett, 5 F.3d 1180, 1189 (9th Cir.1993). Still, the jury bias issue presents a factual question, and the Idaho courts’ finding that the jury was not biased is entitled to the usual presumption of correctness. Id. Having reviewed the record, we agree that Leavitt was not deprived of his right to “a panel of impartial, indifferent jurors.” Irvin v. Dowd, 366 U.S. 717, 722, 81 S.Ct. 1639, 1642, 6 L.Ed.2d 751 (1961) (internal quotation marks omitted). First, there was not such a barrage of inflammatory publicity as would lead to a presumption of prejudice. See Ainsworth v. Calderon, 138 F.3d 787, 795, amended, 152 F.3d 1223 (9th Cir.1998); Gallego v. McDaniel, 124 F.3d 1065, 1070 (9th Cir.1997). Although Blackfoot is a small community and the crime was vile, that alone did not require a change of venue. See Fetterly v. Paskett, 163 F.3d 1144, 1146 (9th Cir.1998); see also United States v. Rewald, 889 F.2d 836, 864 (9th Cir.1989), amended, 902 F.2d 18 (9th Cir.1990); Seattle Times Co. v. United States Dist. Court, 845 F.2d 1513, 1517-18 (9th Cir.1988). The published articles were essentially factual in nature, and the testimony showed that the community was not inflamed against Leavitt. See Harris v. Pulley, 885 F.2d 1354, 1362-63 (9th Cir.1988). Second, no actual prejudice was shown here. On the contrary, each individual who was seated on Leavitt’s jury swore that he or she could impartially judge Leavitt’s guilt or innocence. Cf Blodgett, 5 F.3d at 1189. Only a handful of prospective jurors indicated an inability to sit because they had formed an opinion about Leavitt’s guilt, and a few others knew his family and, actually, felt favorably disposed towards him. That does not suggest that ’Leavitt’s constitutional rights were invaded, and certainly does not suggest that the Idaho courts’ assessment of the situation was unwarranted or incorrect. III. EVIDENTIARY ISSUES Leavitt complains about the admission of some items of evidence against him and about the failure of the prosecution to disclose or preserve other items. A. Evidence Related to Cooperation Leavitt grumbles about the fact that the prosecutor elicited evidence about and commented upon, Leavitt’s failure to cooperate with the investigation and to tell his ultimate story before he testified at trial. Most of his complaints arise out of his confusing true silence with lies, and cooperation with feints at cooperation coupled with suggestions to the jury that he had, in fact, cooperated. A defendant who has received Miranda warnings can, thereafter, remain silent without running the risk that the prosecutor will comment upon' that fact. See Doyle v. Ohio, 426 U.S. 610, 617-18, 96 S.Ct. 2240, 2244-45, 49 L.Ed.2d 91 (1976). That part of the canon stated, however, it must be added that talking is not silence. Thus, when a defendant chooses to speak, the prosecutor can, surely, explore that speech and its implications. See Anderson v. Charles, 447 U.S. 404, 408, 100 S.Ct. 2180, 2182, 65 L.Ed.2d 222 (1980) (per curiam). In that event, as the Supreme, Court bluntly put it, “the defendant has not remained silent at all.” Id. Thus, the prosecutor may point out inconsistencies. See United States v. Ochoa Sanchez, 676 F.2d 1283, 1286 (9th Cir.1982). The omission of critical details may also be explored. See id.; see also United States v. Makhlouta, 790 F.2d 1400, 1404 (9th Cir.1986). Similarly, when the defendant seeks to convey the impression that he cooperated with the police, the prosecutor can explore facts which suggest that the defendant did not do so. See McMillan v. Gomez, 19 F.3d 465, 469-70 (9th Cir.1994). In this instance, statements that Leavitt made to the police psychologist were admissible to demonstrate inconsistencies with the story he told at trial and to cast doubt upon his claim of cooperation with the police. Leavitt conceded at trial that his decision to speak with the psychologist was voluntary, and there is no evi-dentiary support of Leavitt’s claim that the police promised him use immunity in exchange for agreeing to talk with the psychologist. There was no error in admitting that evidence. The same may not be quite as true of his silence at the special inquiry, which is arguably a judicial, rather than a police, proceeding. The distinction between the two is enough to suggest that his silence at the special inquiry may not have been relevant to his claim of cooperation with the police, and that it was arguably improper to comment upon the exercise of his right to remain silent as to certain questions. However, in the context of all of the evidence in this case, including the myriad of other inconsistencies in his stories, any error was harmless as far as this habeas corpus proceeding is concerned. See Brecht v. Abrahamson, 507 U.S. 619, 638, 113 S.Ct. 1710, 1722, 123 L.Ed.2d 353 (1993). Of the same ilk are Leavitt’s ululations about the prosecutor’s comment on the fact that a wholly new blood story surfaced at trial. When speaking to the police, Leavitt indicated that he had no idea how his blood could have been found at the murder scene, but by trial he had developed a wholly inconsistent explanation of its presence — the nosebleed scenario. His jeremiad about the prosecutor’s exploration of his revenant memory avails him nothing. That surely was proper impeachment. See Anderson, 447 U.S. at 408-09, 100 S.Ct. at 2182; United States v. Harris, 726 F.2d 558, 559-60 (9th Cir.1984). It underscored his lies as well as his actual lack of cooperation. Then there was the cut on Leavitt’s finger where, again, his trial explanation differed radically from his pretrial explanation. Again, his hope that he could misdirect the police investigation and claim cooperation at the same time must die aborning. See id.; Phelps v. Duckworth, 772 F.2d 1410, 1412-13 (7th Cir.1985). In a slightly different vein, Leav-itt complains of the prosecutor’s showing that of all of the suspects, only Leavitt had refused to give a blood sample voluntarily. Leavitt asserts that it was improper to comment on the exercise of his right under the Fourth Amendment to the United States Constitution to refuse to consent to a search. We have indicated that, taken by themselves, comments on the exercise of one’s Fourth Amendment rights are improper. See United States v. Prescott, 581 F.2d 1343, 1350-52 (9th Cir.1978); United States v. Taxe, 540 F.2d 961, 969 (9th Cir.1976); Newhouse v. Misterly, 415 F.2d 514, 518 (9th Cir.1969). But, again, Leavitt’s argument is misdirected. Regardless of whether that Fourth Amendment rule should generally apply to habeas corpus cases, Leavitt’s particular objection ig answered by the much more banal and obvious rule that admission of the evidence was proper to attack his claim of cooperation. See United States v. McNatt, 931 F.2d 251, 257-58 (4th Cir.1991). Before there was ever any mention of the blood test, Leavitt had already launched himself on his theme of cooperation. The prosecutor was entitled to question that theme by showing that the leitmotiv was actually one of resistance. Again, we find no error. B. ■ Other Acts Leavitt next argues that the admission of other-act evidence denied him a fair trial. Under Idaho law the evidence was admissible. See Leavitt I, 116 Idaho at 290-91, 775 P.2d at 604-05. We cannot question that. See Estelle v. McGuire, 502 U.S. 62, 67-68, 112 S.Ct. 475, 479-80, 116 L.Ed.2d 385 (1991). However, we can ask whether its admission rendered the trial so fundamentally unfair as to deny Leavitt due process. Id. at 68, 112 S.Ct. at 480. In that regard, we have opined that other-act evidence is irrelevant and may violate the due process- clause if it goes only to character and “there are -no permissible inferences the jury may draw” from it. McKinney v. Rees, 993 F.2d 1378, 1380, 1384 (9th Cir.1993) (quoting Jammal v. Van de Kamp, 926 F.2d 918, 920 (9th Cir.1991)); see also Windham, 163 F.3d at 1103; Jammal v. Van de Kamp, 926 F.2d 918, 919-20 & n. 2 (9th Cir.1991). Leavitt demurs to the fact that his ex-wife testified that once, while- hunting, she came upon him as he carefully and rather surreptitiously was cutting at the female sexual organs of a deer. He then removed those organs, examined them, and played with them because, he said, he wanted to see how they worked. It will be recalled that the victim in this case (or her body if she was then deceased) was subjected to a highly unusual removal of her female organs. Other evidence showed that it would be difficult to accomplish that in the way it was done and that it would help to have knowledge of anatomy when doing it. We agree with the Idaho Supreme Court that the evidence in question was relevant to identifying the killer. We cannot find that the Idaho Supreme Court committed constitutional error when it opined that: In the instant case the corpse of the victim had been brutalized by the removal of her sexual organs by a person who clearly had certain anatomical knowledge. That evidence tended to indicate that the defendant had a morbid and sadistic interest in sexual organs, had a knowledge of anatomy, a possible motive for the crime, and a modus oper-aridi which tended to identify the defendant as the killer. Leavitt I, 116 Idaho at 290, 775 P.2d at 604. We need not pass on whether an evidence scholar would entirely agree; we only decide that the United States Constitution was not violated. The same can be said about the episode testified to by Leavitt’s girlfriend to the effect that he showed her a knife, which was never produced, at a rather peculiar point during a sexual encounter with her. Leavitt’s failure to produce that particular knife for the police officers went to the question of his alleged cooperation with them. Because that, of all knives, was missing, some inference was also possible that it was the murder weapon itself or the knife that was used to cut Elg’s screen. See McKinney, 993 F.2d at 1383-84. Still and all, the connection was pretty thin. Thinner still is the relevance of other knives, which were admitted into evidence. True, they could have been weapons used by the unknown intruder or the murderer, but nothing tended to show that they were; the missing knife was probably a better candidate for that. Nevertheless, considering the other evidence in this case, we are unable to say that, error or not, the knife evidence by itself or in tandem with other errors “ ‘had substantial and injurious effect or influence in determining the jury’s verdict.’ ” Brecht, 507 U.S. at 637, 113 S.Ct. at 1722 (quoting Kotteakos v. United States, 328 U.S. 750, 776, 66 S.Ct. 1239, 90 L.Ed. 1557 (1946)). C. Victim’s Statements The doomed victim of this crime had, as we have already noted, been severely frightened on the night before her death by a prowler, who tried to break into her home. In a great state of agitation, she called the police and spoke to dispatchers and to police officers. Among other things, she said that she thought the prowler was Leavitt, because he had tried to talk himself into her home earlier that day, but she had refused him entry. Leavitt claims that the admission of the hearsay testimony violated his rights under the Confrontation Clause of the Sixth Amendment to the United States Constitution. Of course, one cannot confront a hearsay declarant, but not all uses of hearsay violate the Confrontation Clause. See Idaho v. Wright, 497 U.S. 805, 813-14, 110 S.Ct. 3139, 3145-46, 111 L.Ed.2d 638 (1990). Hearsay can be admitted if it is sufficiently reliable. Reliability is shown when the hearsay “falls within a firmly rooted hearsay exception or is supported by particularized guarantees of trustworthiness.” Guam v. Ignacio, 10 F.3d 608, 612 (9th Cir.1993) (citation and internal quotation marks omitted). The Idaho courts relied upon the state’s residual exception, which is not firmly rooted, but the evidence could properly have come in under the excited utterance exception, which is. See Wright, 497 U.S. at 817, 820, 110 S.Ct. at 3147, 3149. We have considered the circumstances and have no doubt that the victim was speaking while under the baleful influence of an exceedingly stressful event — the attempt by an intruder to break into her home. Nor do we doubt that she lacked the time or the incentive to reflect upon and confabulate a story. Thus, the evidence properly came in as an excited utterance. There was no violation of Leavitt’s constitutional rights. D. Undisclosed and Lost Evidence That mysterious phantom, Mike Jenkins, looms large in an objection by Leavitt regarding an alleged Brady violation by the prosecutor. Jenkins, who evidenced detailed knowledge of the murder, called and spoke to two different police dispatchers — Lisa Pugmire and Theta Duchscher. He first spoke with Pugmire, who never gave out her first name, and later spoke with Duchscher to whom he mentioned Pugmire’s first name. Leavitt and Pug-mire had a friendly relationship and had often spoken to each other before. Both dispatchers testified. Pugmire testified that she could not say that Jenkins had Leavitt’s voice. Duchscher was not asked, and did not say, whether she recognized his voice. At a later time, however, it was revealed that a police investigation had asked Duchscher whether she could make a connection between the voices, and she had said no. Later on, she had said that the voice could have been Leavitt’s. Those facts were not disclosed to Leavitt and that, he says, constitutes the violation. Even if there were some error, it was entirely harmless; it simply is not reasonably probable that the result of the proceeding would have been any different if Leavitt had obtained the information in question. The Leavitb-Jenkins connection was not based on voice recognition, but rather on Mike Jenkins’s use of Pugmire’s first name and on Mike Jenkins’s response, when queried, that his address was near Leavitt’s address. See Williams v. Woodford, 306 F.3d 665, 697 (9th Cir.2002); Hayes v. Woodford, 301 F.3d 1054, 1075 (9th Cir.2002). The other evidence of which Leavitt deems himself improperly deprived was blood samples from the murder scene, which he could have subjected to further testing. But no more useable samples existed after the state serologists had performed their tests. Because it is undisputed that no bad faith was involved in the destruction of the possibly helpful blood samples, Leavitt simply cannot prevail on this claim. See Arizona v. Youngblood, 488 U.S. 51, 56, 109 S.Ct. 333, 336, 102 L.Ed.2d 281 (1988); California v. Trombetta, 467 U.S. 479, 488, 104 S.Ct. 2528, 2534, 81 L.Ed.2d 413 (1984); Cooper v. Calderon, 255 F.3d 1104, 1113-14 (9th Cir.2001); United States v. Hernandez, 109 F.3d 1450, 1455 (9th Cir.1997); Mitchell v. Goldsmith, 878 F.2d 319, 321-22 (9th Cir.1989). That is true, even though the results of the state’s serologists’ tests on the collected blood samples were used in the prosecution’s case in chief. That simply cannot change our conclusion. See Trombetta, 467 U.S. at 491, 104 S.Ct. at 2535; Mitchell, 878 F.2d at 322. IV. ALIBI INSTRUCTION Leavitt maintains that the alibi instruction, instruction 39, impermissibly shifted the burden of proof on his alibi, his only defense. We understand his argument to be that the issue was not procedurally defaulted based on the doctrine of invited error, as the district court held, because — as the court also held — the Idaho Supreme Court said that it had reviewed all trial issues on the merits, including instruction 39, and thus so would the habeas court. We agree with Leavitt that the issue is not procedurally defaulted, which means that the merits are before us. On the merits, we have already held that the instruction is erroneous, Thomas v. United States, 213 F.2d 30, 33 (9th Cir.1954), but the error was harmless in Leavitt’s case. Leavitt unquestionably invited the alibi instruction by proposing it to the trial court. The Idaho Supreme Court could have refused to consider his challenge to the alibi instruction on this basis, State v. Carlson, 134 Idaho 389, 3 P.3d 67, 80 (Ct.App.2000), but did not. Had the Idaho Supreme Court rested its decision on the invited error doctrine, then it might have been an independent state ground that would bar consideration of the issue on habeas review in federal court. Cf. Bennett v. Mueller, 322 F.3d 573, 580 (9th Cir.2003). Instead, the district court found that the Idaho Supreme Court had considered and rejected every possible allegation of trial error on the merits and failed to invoke an independent and adequate state procedural ground. The state acquiesced in this finding for purposes of this appeal. Consequently, the invited error doctrine cannot be invoked as an independent procedural reason for defeating Leavitt’s challenge to the alibi instruction. See Panther v. Hames, 991 F.2d 576, 580 (9th Cir.1993). We .are not persuaded by Idaho’s contention that federal habeas courts may independently rely on the invited error rule regardless of whether it was actually applied in state court. The state points to a number of cases in which a habeas claim was rejected on invited error grounds without the decision being expressly conditioned on prior invocation of the rule by the highest state court to consider the claim, but in these cases the state court had clearly and expressly invoked the invited error doctrine. There is no reason that we should treat the invited error rule differently from other state procedural bars. Turning to the merits, Thomas held that there is no burden of proof on the accused regarding an alibi. But habeas relief is not available for every flawed instruction; the question is “ ‘whether the ailing instruction by itself so infected the entire trial that the resulting conviction violates due process.’ ” Estelle, 502 U.S. at 72, 112 S.Ct. 475 (quoting Cupp, 414 U.S. at 147, 94 S.Ct. 396); cf. Johnson v. Muncy, 830 F.2d 508 (4th Cir.1987). (flawed alibi instructions reviewed for harmless error). Leavitt argues (and the district court agreed) that instruction 39 misallocated the burden of proof, but what it really did was incorrectly identify an alibi as an affirmative defense and then impose on Leavitt the burden of producing enough evidence to create a reasonable doubt as to his guilt, ie., as to his being at Elg’s house instead of somewhere else. If he had not done at least that much, he would have had no alibi defense to begin with. That’s exactly what happened in this case. Leavitt testified that he was home watching TV, but his testimony was uncorroborated. He had to be treated for a bad cut on his finger on the night of Elg’s murder. He was severely impeached on both accounts. And his blood was mixed with Elg’s in her room. It is inconceivable that any reasonable juror would have bought his alibi in these circumstances. Besides, the evidence overall was overwhelming. Therefore, the error in instruction 39 was harmless. Brecht v. Abrahamson, 507 U.S. 619, 637, 113 S.Ct. 1710, 123 L.Ed.2d 353 (1993); see also Neder v. United States, 527 U.S. 1, 119 S.Ct. 1827, 144 L.Ed.2d 35 (1999). V. PROSECUTOR’S ARGUMENT Leavitt asserts that the prosecutor committed misconduct in arguing the case to the jury. Leavitt first complains about arguments that would be erroneous only if we had decided the evidentiary issues discussed in Part III in Leavitt’s favor. Because we did not, we find no impropriety in those respects. Nor was it improper for the prosecutor to argue that Leavitt — by legal strategy or otherwise — had delayed revealing or suggesting the “true new” story until after he had heard all of the evidence against him at trial. To any cognoscente of the trial process, one of the transaction costs of our system is that the order of proof can allow a defendant to tailor his testimony, but “[t]he adversary system surely envisions— indeed, it requires — that the prosecutor be allowed to bring [the tailoring danger] to the jury’s attention.” Portuondo v. Agard, 529 U.S. 61, 69-70, 120 S.Ct. 1119, 1125-26, 146 L.Ed.2d 47 (2000). That, essentially, is what occurred in this case; it was not misconduct. Much more problematic is the prosecutor’s link-in-the-chain-of-law-enforcement argument. Here it is in all of its glory: In closing let me just say that you are part of a very important chain called the chain of law enforcement. And law enforcement and justice don’t work in our country unless you do your part. The police officers can be as well trained as you want them and the forensic sciences can be as well trained as you want in the sciences. And they can go out an[d] investigate crimes as competently and professionally as this group has done. And I think that Officer Robinson and those associated with him have done an excellent job. You can have the best prosecutors around. And I want to tell you that I believe Mr. Moss is one of the best prosecutors in the State. And they work together like this because they are part of the chain of law enforcement that keeps our community safe. But the third link in that chain is a jury, which when they’re given the proper evidence and they are given the proof beyond a reasonable doubt, they have the fortitude to be able to act upon that and to preserve that chain unbroken. And the fourth link in the chain, of course, is the judge who has the courage and also the wisdom to impose the appropriate sentence. Now, none of this works unless you do your job. This suggestion that the jury is simply a link in a chain of law enforcement which includes the police, the prosecutor, and the judge is just plain wrong. It minimizes the important role of the jury and tends to align neutrals — judge and jury — with a party to the case — the state itself. The Tenth Circuit put the matter very well when it was presented with a link-in-the-chain argument. It said: By suggesting that the jury is only the last link in a long decision, the statement tends to trivialize the jury’s importance. This argument also misstates the role of the jury, placing it in an adversarial position with the respect to the defendant. To the extent that the prosecutor’s argument portrayed the jury as part of a team opposing the defendant, it was improper. Coleman v. Brown, 802 F.2d 1227, 1238 (10th Cir.1986) (internal citation omitted); see also Tucker v. Kemp, 802 F.2d 1293, 1296 (11th Cir.1986). Although the argument is wholly undesirable, we cannot say that it alone— and it does essentially stand alone — is enough to result in a determination that the trial was so infected with unfairness as to be a denia