Smith v. Baldwin

466 F.3d 805 | 9th Cir. | 2006

Lead Opinion

Opinion by Judge Reinhardt; Dissent by Judge Bybee

REINHARDT, Circuit Judge.

I.

This case presents the question whether a state prisoner who contends that he is actually innocent, but whose principal witness is coerced by the state into not testifying on his behalf, may pursue his federal constitutional claims in federal court notwithstanding his failure to comply with all of the applicable procedural prerequisites. Roger Smith is currently serving a life sentence with a 30-year minimum term. The district court dismissed on procedural grounds his petition for a writ of habeas corpus without reaching the merits of his claims. It found that he had not exhausted those claims in state court and that, because state procedural rules barred him from doing so now, the claims were procedurally defaulted. Like the district court, we do not consider the merits of his case. All we decide is that, under an exception to the applicable procedural rules, Smith may pursue his federal constitutional claims in federal court. Both the facts and the law are complex, however, as they tend to be these days in almost all habeas corpus cases.

Smith argues on this appeal that his petition has not been procedurally defaulted and that, if it has, the procedural default should be excused on the basis of his claim of actual innocence. The exception on which he relies is known as the Schlup “actual innocence” exception, named after the case of Schlup v. Delo, 513 U.S. 298, 315, 115 S.Ct. 851, 130 L.Ed.2d 808 (1995). We hold that because prosecutorial misconduct in connection with his federal habeas proceedings seriously interfered with *807Smith’s ability to make the necessary showing under Schlup, and because the resultant harm cannot be effectively remedied by less intrusive means, the exculpatory testimony withheld from the court as a result of the state’s actions must be presumed to be true. We deem the exculpatory statements to be truthful, however, only for the purpose of determining whether Smith’s contentions are sufficient to excuse any procedural default that may have occurred; we do not consider here what remedy would be appropriate with respect to his subsequent efforts to establish any claim on the merits that might entitle him to relief, including any claim of actual innocence.

In short, we conclude only that, affording the disputed witness statements the benefit of the presumption of truthfulness, Smith satisfies the Schlup “actual innocence” standard for overcoming a procedural default of his claims insofar as they relate to his felony murder conviction, and nothing more. As we explain below, Schlup is satisfied here, and thus Smith is entitled to proceed with his constitutional claims, not simply because the evidence at this point would likely preclude any reasonable juror from determining that he was the actual killer, but because that evidence would more likely than not cause any such juror to conclude that he had established, by a preponderance of the evidence, an affirmative defense to the felony murder charge under Oregon law. Accordingly, we reverse the district court and remand so that Smith may be afforded a hearing on the merits of his constitutional claims.1

II.

On April 3, 1989, two young men, Smith and Jacob Edmonds, burglarized and robbed the home of Emmett and Elma Konzelman. A third man, Marlin Bouse, drove with them to the Konzelmans’ home, but decided not to participate further in the criminal activity after initially entering the garage. Although there was no indication that the two who entered the house itself had planned to injure anyone, one of them, when out of the presence of the other, attacked the Konzelmans in their bedroom and bludgeoned Mr. Konzelman to death. Mrs. Konzelman, who survived the assault, told police that only one of the burglars had attacked the couple and that no one else entered the room at that time or saw the killing. There is no dispute that either Smith or Edmonds alone committed the murder. The ultimate question is which one.

There is substantial evidence in the record to suggest that Edmonds murdered Mr. Konzelman and that he committed the killing outside of Smith’s presence and without any advance knowledge on Smith’s part that he would engage in any violent conduct or that he possessed a dangerous or deadly weapon. The blood on the getaway truck and Mrs. Konzelman’s recollection of the killer’s attire, for instance, point to Edmonds as the killer.2 However, as *808part of a deal with the prosecutor, Edmonds named Smith as the murderer in return for the dismissal of his murder charges. The conditions of the plea agreement required Edmonds to pass a polygraph examination showing that all of his allegations were truthful and to give a complete statement memorializing his accusation against Smith. In return, the state agreed to recommend concurrent sentences with a minimum of 43 months on the two robbery counts to which he pled.3

After learning that he would be prosecuted for capital murder and that Edmonds was going to testify against him, Smith, who had continuously asserted his innocence, pled no contest to robbery and felony murder. In return, he received a life sentence with a minimum term of 30 years. Although Smith pled no contest to the lesser charge of felony murder, the trial judge explained that he was enhancing Smith’s sentence because he believed that Smith was the one who had committed the killing, even though the evidence was far from clear on this point. After the plea, Smith filed a direct appeal, which the Oregon Court of Appeals and the Oregon Supreme Court denied without opinion.

Smith then filed a petition for post-conviction relief in state court, setting forth claims that (1) his trial counsel was constitutionally ineffective because his lawyer coerced his plea agreement without advising him of or pursuing an affirmative defense to felony murder; (2) his due process rights were denied when Edmonds’s plea bargain wrongfully implicated him; and (3) his due process rights were violated because the effect of his medication and his trial counsel’s coercion rendered his no contest plea unknowing and involuntary. The state post-conviction judge denied Smith’s request to appoint new counsel and ruled against him on the merits.

New counsel was appointed on appeal, but Smith did not appeal the denial of his post-conviction claims and argued only that replacement counsel should have been appointed at his first post-conviction hearing. The Oregon Court of Appeals affirmed without opinion and the Oregon Supreme Court denied review. As a result of this appeal, Smith never presented his substantive post-conviction claims to the state supreme court and they remain unexhausted.

In a notarized statement to the district attorney seven years after the incident and two years after the Oregon Supreme Court’s denial of review, Edmonds recanted his earlier testimony against Smith, implicitly acknowledging that he had murdered Mr. Konzelman. Edmonds explained that although he “stated in a sworn statement that Roger Smith ... had in fact bludgeoned Emmit Konzleman [sic] to death,” he wanted “for the record [to] retract that statement. Roger Smith did not kill Emmit Konzleman [sic].” He then confessed to committing perjury in his statement against Smith and explained that “the only way for me to set the record straight is to write this confession now.”

Based on Edmonds’s recantation, Smith again requested post-conviction relief in state court. Smith’s second petition asserted that Edmonds’s 1996 recantation was new evidence that exculpated him and that his due process rights were violated both when the differences between aggravated murder and felony murder were not *809explained to him at the time of his plea and when the court failed to provide him with an adequate explanation of the charges against him, given that he was on drugs and of below-average intelligence. The court, however, granted summary judgment for the state, concluding that “post-conviction is not the for[u]m to— won’t allow this relief here on newly discovered evidence.” Smith appealed but grew frustrated with his attorney and with the delay caused by numerous extensions of time. As a result, he filed a motion for voluntary dismissal, which the court granted. Smith next turned his attention to the federal courts: In 1997, Smith asserted four grounds for relief in a petition for federal habeas corpus: (1) conviction obtained by a guilty plea which was unlawfully induced and made neither voluntarily nor intelligently; (2) denial of effective assistance of counsel during the investigative and trial preparation stages; (3) denial of effective assistance of counsel during the plea negotiation/ entry stage; (4) denial of effective assistance of counsel during hearings before the trial court.

Amidst the federal habeas proceedings, Edmonds again recanted his statements against Smith in a sworn affidavit. This time, Edmonds’s declaration even more strongly inculpated himself and exculpated Smith as the actual killer: “I know that Mr. Smith did not bludgeon or otherwise strike Mr. Konzleman [sic] and I know that Mr. Smith never entered the Konzleman’s [sic] bedroom, where the killing occurred.” Edmonds also explained that “[a]s a part of my plea agreement in the underlying case, I was required to take a polygraph concerning the murder and I was told that I neither passed [n]or failed that test but that the results were inconclusive.” By these statements, Edmonds largely discredited his initial testimony, in which he had labeled Smith as the attacker.

On the basis of the second recantation by Edmonds, Smith amended his petition adding two additional claims: (5) that he is actually innocent of felony murder and his conviction violates his right to due process under the Fourteenth Amendment and his right to be free from cruel and unusual punishment under the Eighth Amendment; and (6) that his federal due process rights were violated under Brady v. Maryland, 373 U.S. 83, 83 S.Ct. 1194, 10 L.Ed.2d 215 (1963), when the prosecution made a deal with the actual murderer and suppressed evidence tending to exculpate Smith.

After Smith added his new claims, the state for the first time turned over a copy of Edmonds’s polygraph results from 1989, even though Smith’s trial counsel had specifically requested the document before Smith’s plea. The 1989 polygraph examination consisted of a few questions, which included whether Edmonds “st[ruek] either of the people in the house on South Shore Drive” and whether he was “lying about anyone else’s involvement in this event.” Edmonds answered “no” to both questions. The evaluator concluded that although “[i]t is the opinion of this examiner that ... Edmonds answered these questions in a truthful manner, ... the scoring on the Backster Zone Comparison Test Technique is inconclusive.” Taking such test results into account in considering Edmonds’s plea agreement, Smith alleges, and the state does not dispute, that Edmonds failed to satisfy the conditions of his plea agreement. Furthermore, the document was not produced for 14 years, even though it was timely requested and contained exculpatory information.

Smith asserts that the state dramatically accelerated its efforts to suppress exculpatory evidence during the pendency of the federal habeas proceedings. The state threatened Edmonds with the revocation *810of his guilty plea and the institution of capital murder charges if he insisted on testifying in accordance with his recantations. At the same time, it made Edmonds an offer of immunity with respect to perjury if he would testify on behalf of the state and formally withdraw his recent confessions. To be clear, the record shows that, as a part of its attempt to deter Edmonds from testifying that Smith was innocent, the state informed Edmonds and his newly appointed counsel that if he persisted in recanting his testimony against Smith, the state would seek the death penalty against him. In response to these threats, Edmonds sent a letter to the district judge informing him that he would invoke his Fifth Amendment privilege if called to testify on Smith’s behalf. In reply, Smith contended that the state was engaging in prosecutorial misconduct intended to distort the factfinding function of the court and that, if the court did not grant him the requested habeas corpus relief, it should order an evidentiary hearing at which the government should be compelled to provide Edmonds with immunity.

Almost two years later, the district court ruled that it was unable to reach the merits of Smith’s claims. It held that the claims were procedurally defaulted because of his failure to exhaust them in state court.4 The district court also found that Smith did not establish either sufficient actual innocence or cause and prejudice to excuse the defaults. It stated, however, that “[a] logical inference” from Edmonds’s refusal to testify in the manner desired by the state, even after the prosecution’s promise of immunity if he would do so, is that “if Edmonds did testify truthfully, he would say that Smith did not kill Emmitt [sic] Konzelman, and implicate *811himself as the killer.” Nevertheless, it rejected Smith’s suggestion that Edmonds be granted immunity under United States v. Westerdahl, 945 F.2d 1083 (9th Cir.1991), on the ground that it could not “say that the government’s immunity decision [wa]s so capricious as to constitute an intentional distortion of the factfinding process that warrants judicial intervention.” It then denied Smith’s request for an evidentiary hearing because it believed that a hearing would be futile and would serve no purpose given Edmonds’s refusal to testify — in other words, that in the absence of Edmonds’s exculpatory testimony (which the prosecution had advised Edmonds would cause it to seek his execution), Smith would be unable to overcome the procedural default. Smith now appeals the district court’s procedural default rulings. We have jurisdiction under 28 U.S.C. § 2253.

III.

The petition in this case was filed after April 24, 1996, and is therefore governed by the Antiterrorism and Effective Death Penalty Act (AEDPA), 28 U.S.C. § 2254. Woodford v. Garceau, 538 U.S. 202, 210, 123 S.Ct. 1398, 155 L.Ed.2d 363 (2003). An appellate court reviews de novo the district court’s decision to deny a habeas petition. Hunter v. Ayers, 336 F.3d 1007, 1011 (9th Cir.2003). This court also reviews de novo a district court’s decision to dismiss a habeas corpus petition for procedural default. Manning v. Foster, 224 F.3d 1129, 1132 (9th Cir.2000). The district court’s interpretation of AEDPA standards governing the grant or denial of an evidentiary hearing is reviewed de novo, Baja v. Ducharme, 187 F.3d 1075, 1077 (9th Cir.1999), and its ultimate denial of an evidentiary hearing, based on the AEDPA standards, is reviewed for abuse of discretion, Davis v. Woodford, 384 F.3d 628, 638 (9th Cir.2004) (citing Lawson v. Borg, 60 F.3d 608, 611 (9th Cir.1995)). An error of law is, however, an abuse of discretion. Koon v. United States, 518 U.S. 81, 100, 116 S.Ct. 2035, 135 L.Ed.2d 392 (1996) (citing Cooter & Gell v. Hartmarx Corp., 496 U.S. 384, 405, 110 S.Ct. 2447, 110 L.Ed.2d 359 (1990)). The district court’s findings of fact are reviewed for clear error. Bonin v. Calderon, 59 F.3d 815, 823 (9th Cir.1995).

IV.

Although the district court stated that it questioned the adequacy of Oregon’s post-conviction procedures if, as the state post-conviction court held, newly discovered evidence of actual innocence could not serve as a basis for post-conviction relief in Oregon, it declined to decide whether those procedures were in fact inadequate because it concluded that Smith could not, at least in the absence of Edmonds’s testimony, establish that he was actually innocent. We, like the district court, need not decide that question, although for an opposite reason. We conclude that even if there is a default, Smith has, given the procedural determinations we reach below, made the requisite showing of actual innocence necessary to have his claims heard on the merits.5

A state prisoner must normally exhaust his available state remedies before a federal court may consider his petition for a writ of habeas corpus. 28 U.S.C. § 2254(b); Rose v. Dandy, 455 U.S. 509, 515, 102 S.Ct. 1198, 71 L.Ed.2d 379 (1982). The Supreme Court has held that if a *812“petitioner [has] failed to exhaust state remedies and the court to which the petitioner would be required to present his claims in order to meet the exhaustion requirement would now find the claims procedurally barred,”6 his claims are procedurally defaulted for purposes of federal habeas review. Coleman v. Thompson, 501 U.S. 722, 735 n. 1, 111 S.Ct. 2546, 115 L.Ed.2d 640 (1991). A petitioner can overcome procedural default and obtain federal review of the merits of his claims either by demonstrating evidence of actual innocence sufficient to bring the defendant within the “narrow class of cases ... implicating a fundamental miscarriage of justice,” Schlup, 513 U.S. at 315, 115 S.Ct. 851 (quoting McCleskey v. Zant, 499 U.S. 467, 494, 111 S.Ct. 1454, 113 L.Ed.2d 517 (1991)) (internal quotation marks omitted), or by making an “adequate showing of cause and prejudice,” Strickler v. Greene, 527 U.S. 263, 282, 119 S.Ct. 1936, 144 L.Ed.2d 286 (1999); see also Murray v. Carrier, 477 U.S. 478, 485, 106 S.Ct. 2639, 91 L.Ed.2d 397 (1986). Because we conclude that, in light of our procedural determinations, Smith has made a showing of actual innocence sufficient to overcome any procedural default with respect to all of his claims regarding his felony murder conviction, we find it unnecessary to decide whether Smith could also make the alternative showing of cause and prejudice with respect to certain of them.

A.

Smith’s procedural default on his constitutional claims will be excused if he can “demonstrate that failure to consider the claims will result in a fundamental miscarriage of justice,” Coleman v. Thompson, 501 U.S. 722, 750, 111 S.Ct. 2546, 115 L.Ed.2d 640 (1991), which occurs when a “constitutional violation has probably resulted in the conviction of one who is actually innocent” of the offense that is the subject of the barred claim. Murray, 477 U.S. at 496, 106 S.Ct. 2639; see also Schlup, 513 U.S. at 324, 115 S.Ct. 851. The Supreme Court has described this showing of “actual innocence” as a “gateway” that allows the court to consider otherwise procedurally defaulted claims of constitutional error. See Schlup, 513 U.S. at 315-16, 115 S.Ct. 851 (1995). A Schlup “actual innocence” inquiry is subject to a less stringent standard than a substantive “actual innocence” claim. See Carriger v. Stewart, 132 F.3d 463, 476 (9th Cir.1997) (en banc) (suggesting that “a habeas petitioner asserting a freestanding innocence claim must go beyond demonstrating doubt about his guilt, and must affirmatively prove that he is probably innocent”). This is because passing through the Schlup gateway only permits a federal court to review the underlying constitutional claims; Schlup does not entitle a defendant to a declaration of actual innocence or to any relief outright. Under Schlup, “a petitioner must show that it is more likely than not that no reasonable juror would have found [him] guilty beyond a reasonable doubt.” Schlup, 513 U.S. at 327, 115 S.Ct. 851. In determining whether this showing has been met, the habeas court may consider “all the evidence,” including evidence excluded at trial, admitted illegally, or available only after the trial. Id. at 328, 115 S.Ct. 851. To pass through the procedural gateway, Smith does not need to “show that he is ‘actually innocent’ of the crime he was convicted of committing; instead, he must show that ‘a court cannot have confidence *813in the outcome of the trial.’ ” Majoy v. Roe, 296 F.3d 770, 776 (9th Cir.2002) (quoting Schlup, 513 U.S. at 316, 115 S.Ct. 851). Although here the conviction was upon a plea of guilty rather than a jury verdict, the inquiry is the same in either case: can we have confidence that the petitioner committed the offense of which he was convicted?

The Supreme Court recently reaffirmed the viability of the Schlup “actual innocence” gateway in House v. Bell, — U.S. -, 126 S.Ct. 2064, 165 L.Ed.2d 1 (2006). In House, the Court emphasized that “the Schlup standard does not require absolute certainty about the petitioner’s guilt or innocence,” but rather requires only that a petitioner demonstrate that it is “more likely than not any reasonable juror would have reasonable doubt” regarding his guilt. Id. at 2077. In doing so, the Court also rejected any suggestion that AEDPA “ha[d] replaced the Schlup standard with a stricter test,” id. at 2078, holding instead that AEDPA’s standard of review is inapplicable to “a first federal habeas petition seeking consideration of defaulted claims based on a showing of actual innocence,” id.

The state charged Smith with felony murder because he committed a burglary and a robbery, and a killing occurred during the commission of those acts.7 Oregon allows a defendant to prove an affirmative defense to establish innocence of felony murder. OR. REV. S TAT. § 163.115(3) (2003). Smith satisfies Schlup’s actual innocence requirement if he shows that it is more likely than not that any reasonable juror would have had a reasonable doubt as to his guilt in light of the evidence in the district court record, including Edmonds’s affidavits, supporting his affirmative defense. See Jammillo v. Stewart, 340 F.3d 877, 883 (9th Cir.2003) (demonstrating use of affirmative defense for Schlup’s “actual innocence” analysis). At trial, Smith would have been required to prove his affirmative defense only by a preponderance of the evidence. See State v. Counts, 311 Or. 616, 622, 816 P.2d 1157 (1991). Therefore, to pass through the Schlup “gateway,” Smith must demonstrate only that it is more likely than not that no reasonable juror would have found that he had failed to establish the elements of the affirmative defense by a preponderance of the evidence.

In order to establish an affirmative defense to felony murder, Smith must show that he:

(a) Was not the only participant in the underlying crime;
(b) Did not commit the homicidal act or in any way solicit, request, command, importune, cause or aid in the commission thereof;
(c) Was not armed with a dangerous or deadly weapon;
(d) Had no reasonable ground to believe that any other participant was armed with a dangerous or deadly weapon; and
(e) Had no reasonable ground to believe that any other participant intended to engage in conduct likely to result in death.

§ 163.115(3). The state does not dispute that Smith has established the first element. The disagreement is over the last four. As we explain below, because the *814prosecutor’s misconduct deprived Smith of the benefit of Edmonds’s testimony, we presume Edmonds’s affidavits to be true for purposes of determining whether Smith is procedurally barred from proceeding in federal court. Affording that presumption to 17742 Edmonds’s recantations, we conclude that Smith has presented sufficient evidence to make it more likely than not that any reasonable juror would conclude that he has satisfied all five elements of the affirmative defense to felony murder by a preponderance of the evidence, and that he is “actually innocent” for purposes of Schlup. Thus, Smith may present his substantive habeas claims in the district court.

As to the second element, the evidence demonstrates that it is more likely than not that no reasonable juror would have found that Smith killed Konzelman. Edmonds’s recantations confessing that Smith did not kill Konzelman, which we presume to be true for purposes of this analysis, combined with the testimony of Mrs. Konzelman, which corroborates such recantations in more than one respect, the testimony of Bouse, and the physical evidence lead to the conclusion that Edmonds, and not Smith, was the killer.8 Although Edmonds has never explicitly confessed that he committed the murder, he has twice sworn that Smith did not kill Konzelman. In his latest affidavit, he said that Smith neither struck Konzelman nor “entered the Konzleman’s [sic] bedroom, where the killing occurred.”9 Given Mrs. Konzelman’s testimony that the attacker *815who struck her husband was the only attacker in the bedroom at the time — as the other burglar stood in the doorway briefly and then continued down the hall- — the only reasonable inference to be drawn, assuming Edmonds’s affidavit to be credible, is that Edmonds was the killer. Even the state acknowledged this inference when it stated that “Edmonds has implicitly admitted that he, rather than [Smith], was the killer, [even though] he has never directly admitted as much.”

Further, Edmonds’s initial account of the murder implicating Smith must be viewed with considerable skepticism, even aside from the presumption of truthfulness we afford his subsequent recantations. Edmonds failed to pass the polygraph examination required as part of his plea bargain,10 and the 17744 consideration he received in exchange for placing the blame on Smith was substantial. That consideration, the district judge observed, “arguably gave him incentive to lie.” Moreover, the mere fact of a subsequent confession in itself raises doubt. The state points to a few instances in which Edmonds’s initial story is corroborated. While it is true that Edmonds may have been able to describe how he avoided the newspaper delivery person while in the garage or how he and Smith went to Smith’s parents’ house shortly after the incident, these descriptions show only that he was present when the crime occurred and are irrelevant to the question whether he or Smith committed the murder in the Konzelmans’ bedroom.11

The physical evidence of blood near Edmonds’s area in and on the truck provides strong support for the conclusion that he was the killer. It is clear from the record that Edmonds was the person who actually drove the truck away from the crime scene. The state crime lab found blood only on the areas of the truck that Edmonds, as the driver, would have touched: “the left front part of the hood, the front part of the driver door, the back part of the driver’s door and the ignition area on the interior of the vehicle.” As it is apparent from the record that the scene was quite bloody, the magistrate judge speculated that the blood on the vehicle could be explained by the fact that Edmonds, as he told his jail cellmate, “may have helped wipe down the bloody crowbar.” Particularly in light of his recantations, such statements by Edmonds lack credibility;12 *816had he only helped to wipe down the tire iron, that would not explain why all of the blood ended up in his area of the truck.

Mrs. Konzehnan’s testimony regarding the attire of the man who attacked her husband also tends to support the conclusion that Edmonds was the killer. First, Mrs. Konzelman told police that the killer had a skull or stocking cap covering his head. On the night of the incident, Smith wore a hat with a brim, possibly the fedora-style hat that was found outside the Konzelmans’ house. Edmonds described Smith’s hat as “floppy” with “a brim all the way around it”; he called it an “old man hat,” an “old gangster hat,” or a “golf hat.” A fedora, or one of the hats Edmonds described, would have had some sort of brim that distinguished it from the hat Mrs. Konzelman described. Indeed, Mrs. Konzelman specifically noted that the attacker wore a different type of hat from the fedora found near the Konzelman home. Edmonds stated that he wore a baseball cap which, worn backwards, could easily be seen as resembling a skull or stocking cap.

Second, Mrs. Konzelman testified that the attacker was wearing a light-colored jacket, which, because the tags were showing, appeared to be turned inside out. Refuting the state’s suggestion that Smith wore a jacket with fleece lining during the crime, Marlin Bouse — the co-defendant who abandoned the burglary after he, Edmonds, and Smith initially entered the Konzelman’s garage — testified that Edmonds wore a jacket while Smith wore a sweatshirt, which Bouse believed to be red.13 Bouse further testified that Edmonds and Smith did not change clothing at any point during the night. Edmonds originally tried to evade responsibility for the jacket, but the police found his Levi’s jacket, which had a fleece or wool lining, in the possession of his girlfriend. Although Edmonds asserted, after earlier telling authorities otherwise, that he had loaned his Levi’s jacket to Smith, the magistrate judge who conducted the factfinding was highly skeptical of this claim, noting that Edmonds had confided in a friend that he had lent his jacket to Smith “in an effort to place [Smith] in that jacket.”

Third, Mrs. Konzelman testified that she did not notice gloves on the attacker. As the dissent notes, “it is not entirely clear who wore which gloves.” Dis. op. at 831. Edmonds testified that he wore white latex surgical gloves while Smith wore two black leather gloves; Smith agreed that Edmonds wore latex gloves but claimed that he, Smith, wore either one or two brown leather gloves. No matter whose account is more credible, though, Mrs. Konzelman would have been much more likely to fail to notice translucent latex gloves than either brown or black opaque leather gloves — particularly on a white attacker— which tends to support the conclusion that Edmonds was the killer.

Contrary to the state’s contentions and the magistrate judge’s findings, the remaining evidence does not support the claim that Smith was the killer. For example, simply because black jeans were recovered from Smith’s house and Mrs. Konzelman stated that the killer wore dark or black pants, it does not follow that Smith wore black jeans on the night of the burglary or that Smith was the killer.14 *817Indeed, there is nothing unusual about having a pair of dark jeans in one’s house. Moreover, contrary to the magistrate judge’s contention, the code names— “High” and “Low” — attributed to Smith and Edmonds do not demonstrate that Smith was the killer. Mrs. Konzelman told police that the voice of the person yelling “Low” into the phone in the other room was probably not the voice of the killer. It is true that at one point Edmonds claimed that his code name was “High”; therefore, the magistrate judge reasoned that it must have been Edmonds who called out “Low” to Smith. But Edmonds’s claim that he was “High” should be viewed with skepticism, given his later testimony at the sentencing hearing that he could not “recall what the code was” and the fact that he responded affirmatively to the question whether his code name was “High” and Smith’s “Low” only after being told about Mrs. Konzelman’s statement. Furthermore, while maintaining that the person who jerked out the telephone cord was the same person whom she had heard yelling into the phone, Mrs. Konzelman told detectives in a later interview that the person who jerked out the telephone cord was the same person whom she had seen hit her husband. Such statements would seem to contradict her earlier statements that the person yelling into the phone was not the attacker. Any evidence regarding the use of code names appears inconclusive with respect to the identity of her husband’s assailant.15

Finally, if Edmonds’s recantations are deemed credible, as we presume them to be, his initial testimony is necessarily false to the extent that it points to Smith and not himself as the actual killer, and lacks any force in countering the other available evidence.16 Assuming Edmonds’s recantations to be credible, the available evidence shows that it is more likely than not that no reasonable juror would have found that the second element of the affirmative defense — that Smith was not the actual killer' — -had not been established.

As to the third factor of the affirmative defense — whether Smith was armed with a dangerous or deadly weapon — there is no evidence to suggest that anyone other than the killer was so armed. Because the evidence discussed above demonstrates that it is more likely than not that every reasonable juror would find that Edmonds was the only person inside the Konzelmans’ bedroom17 and was the actual killer, *818it logically follows that it is more likely than not that Smith was not armed and that no reasonable juror would have found to the contrary.

The dissent argues that Smith cannot satisfy this element of the affirmative defense because he admitted to carrying a rope into the Konzelman residence. Dis. op. at 846-47, 851-52. This argument is concocted out of whole hemp by our creative dissenting colleague. The state wisely does not contend that the rope was a dangerous or deadly weapon; nor does Oregon law support such a contention. Under Oregon law, a “dangerous weapon” is defined as “any weapon, device, instrument, material or substance which under the circumstances in which it is used, attempted to be used or threatened to be used, is readily capable of causing death or serious physical injury” and a “deadly weapon” is defined as “any instrument, article or substance specifically designed for and presently capable of causing death or serious physical injury.” Or. Rev. Stat. §§ 161.015(1) & (2) (2003). There is no evidence suggesting that the rope in this case qualified under either of these provisions. Certainly it was not “specifically designed” for the purpose of causing death or serious physical injury. Nor does any evidence in the record suggest that the rope was “used” or “attempted to be used” during the commission of the offense-Mrs. Konzelman testified that she first saw the rope when she got out of bed after the attaek was complete, and after waiting a period to make sure that the intruders were gone from the house. Finally, there is no evidence suggesting that either intruder “threatened” to use the rope, let alone threatened to use it in a manner “readily capable of causing death or serious physical injury.” Indeed, Mrs. Konzelman unequivocally denied that any threats were made in the following exchange with the detectives:

Q: There was something mentioned about a rope.
A: There was a rope there on the floor, but I didn’t see them. They didn’t do anything with it.
Q: OK. Did he, did he threaten to tie you up or anything?
A: No.
Q: Did he say anything about he would, anything to you at all?
A: Not that I can tell you.
Q: OK. There was, I thought that maybe some statements were made earlier about saying something about rolling you over so he could tie you up.
A: No.

The record could not be any clearer on the question whether the intruders used, attempted to use, or threatened to use the rope — plainly, they did not.18

Under Oregon law, the question whether an object is a dangerous weapon turns entirely on the circumstances in which it is *819used, attempted to be used, or threatened to be used. No one doubts that, in certain circumstances, a rope can be used to asphyxiate, and that when it is, it qualifies as a dangerous weapon. This realization, however, is very different from concluding that the rope in this case met Oregon’s definition of a dangerous weapon. It most certainly did not. The cases the dissent cites show how far off-the-mark its conclusion really is. The dissent points to cases finding that a can opener, cowboy boots, or a concrete sidewalk can be a deadly weapon. Again, it all depends on the circumstance — if used to stomp someone to death, cowboy boots might meet the statutory definition; if simply worn while engaged in a fatal fist fight, the boots would not constitute a dangerous weapon. The issue of the rope begins and ends with the fact that Smith did not use, attempt to use, or threaten to use it in any way, let alone in a way that might cause death or serious injury.

The fourth element of the affirmative defense requires Smith to show that he had “no reasonable ground to believe that any other participant was armed with a dangerous or deadly weapon.” The district court acknowledged that, at the outset, Smith and Edmonds probably did not intend to kill the Konzelmans. Otherwise, it concluded, “there would be no need to disguise themselves [or] to speak in code names, unless they expected the Konzelmans to survive the intrusion and be capable of giving an account to the police.” Further, the haphazard and unplanned nature of the burglary — during which the burglars ran in and out of the house due to unexpected noises and left several items behind, including Mr. Konzelman’s money and wallet — suggests, contrary to the dissent’s assertions, see dis. op. at 845-46, that the burglars’ planning would not have included a debate as to whether they should arm themselves or any discussion as to who would carry the weapon, and that a juror would not be likely to credit the pair with such an organized or detailed level of planning.

It is not disputed that the crowbar came from somewhere in the Konzelmans’ garage. The state has never argued that either Smith or Edmonds brought it with him and Mrs. Konzelman testified that the crowbar used in the attack was one that she and her husband used frequently and that they kept in the garage. The district court opinion acknowledges that “[b]oth the garage and the interior of the house were dark when [Smith and Edmonds] entered, so Smith might not have noticed if Edmonds was carrying a dark-colored crowbar. ... Indeed, Edmonds claimed he was unaware of the crowbar until he first saw it in Smith’s hands in the bedroom.” The district court’s subsequent observation that “[t]he converse might also have been true, if Edmonds is reversing the roles,” is supported by Edmonds’s recantations, which identify him and not Smith as the actual attacker. Under those circumstances, it would have been Smith who was unaware that Edmonds was armed.19 Furthermore, as the district court noted, the two men were in different parts of the house for most of the time.20 *820The attacker apparently was not holding the crowbar in plain view when the second intruder passed by the room, as Mrs. Konzelman did not see anything in the attacker’s hands until he picked the crowbar up from the ground just prior to attacking her husband.21 Smith was not present when Edmonds struck Konzelman, as evidenced by Mrs. Konzelman’s statements about a lone attacker. According to Mrs. Konzelman, the second burglar “came from the den, up through the hall, and stopped at the door long enough for me to see his bandanna and headed out,”22 sometime before Edmonds picked up the crowbar and struck the victim. This version of events finds further support in Edmonds’s more recent affidavit asserting that Smith never entered the bedroom: “I know that Mr. Smith did not bludgeon or otherwise strike Mr. Konzleman [sic] and I know that Mr. Smith never entered the Konzleman’s [sic] bedroom where the killing occurred.” In sum, the evidence as to precisely what occurred in the garage and in the house is far from clear. As with the third element, however, having decided that it is more likely than not that Edmonds was the killer and thus the person who found, carried, and used the crowbar, and assuming the truthfulness of Edmonds’s recantations, we conclude, although this issue is close, that any reasonable juror would have determined that a preponderance of the evidence shows that Smith had no reasonable ground to believe that Edmonds had armed himself immediately prior to entering the house.

The last element of the affirmative defense requires Smith to show that it is more likely than not that he had no reasonable ground to believe that Edmonds intended to engage in conduct likely to result in death. As discussed supra, it is more likely than not that Smith was not aware that Edmonds had picked up a crowbar and was armed. The evidence supporting that conclusion would also support the conclusion that Smith had no reason to believe that Edmonds intended to do anything other than burglarize the house. Edmonds himself stated that prior to entering the house, there had never been any discussion between him and Smith about hitting anyone or using any kind of weapon on anyone. In a deposition, Smith stated with respect to the killing of Mr. Konzelman that “I didn’t know [Edmonds] was going to do it.”

The district court never found that Edmonds and Smith planned to engage in any violent conduct or even discussed doing so. Rather, it reasoned that because Edmonds and Smith were aware that the Konzelmans were home, they should each have been aware that the robbery might result in confrontation with the residents, and that they therefore each had reason to believe that they would engage in conduct *821likely to result in death. We disagree that a generic residential robbery is likely to result in death — especially when there are no plans to engage in violent conduct and, to Smith’s knowledge, no weapons on hand with which to do so.23

If the district court were correct in concluding that any residential burglary is likely to result in death simply because one can foresee a remote possibility of death occurring such as the possibility that a “resident might awaken startled, and suffer a fatal heart attack” or that a “[f]atal accident[ might] occur when police or ambulance crews are racing to the scene” — then the affirmative defense would be inapplicable in every case of felony murder. The district court’s reasoning would effectively foreclose the affirmative defense to anyone who committed, or even attempted to commit, any of the ten offenses that qualify for felony murder. See Or. Rev. Stat. § 163.115(l)(b) (2003). Additionally, the term “likely” in Oregon’s affirmative defense to felony murder implies a stronger possibility than a merely remote one.24 See, e.g., State v. Watkins, 146 Or.App. 338, 932 P.2d 107, 109 (1997) (in which a victim’s death was deemed “a significantly greater harm than is typical for the crime of robbery” and the defendant was acquitted of felony murder after raising the same affirmative defense); cf. State v. Dickerson, 112 Or.App. 51, 827 P.2d 1354, 1355 (1992) (in which the defendant was found to have reasonable grounds to believe another participant in the crime intended to engage in conduct likely to result in death based on conversations defendant had with other participants about “how to handle the situation” and the decision to kill the victim “to ensure her silence”). Indeed, if the commission or attempted commission of any of the listed felonies in section 163.115(l)(b) — which include arson, assault, and robbery — -automatically constituted conduct likely to result in death, one would assume that Oregon would not have made the affirmative defense available at all for such charges. Because it cannot be assumed that generic robbery is likely to result in death and the evidence in this case rebuts any claim that Smith had reasonable cause to believe that this particular robbery was likely to do so, Smith has more likely than not satisfied the fifth and final element of the affirmative defense.25

*822In reaching the opposite conclusion, that it is not “more likely than not” that a reasonable juror would find that all five elements of the affirmative defense had been met, the dissent makes the mistake of approaching the Schlup analysis exactly as we stated it should not be applied in Carriger: “We do not agree ... that our test under Schlup is to decide how a hypothetical jury would regard each bit of new evidence.” Carriger, 132 F.3d at 474 n. 4. Instead, we held, under Schlup, “[o]ur task is to determine whether confidence in the actual verdict is undermined.” Id. The dissent misconstrues the Schlup test, which ultimately requires only that Smith demonstrate that it is more likely than not that any reasonable juror would find that he had satisfied the elements of the affirmative defense to felony murder by a preponderance of the evidence — or, in other words, that there is a 51% or greater chance that any reasonable juror would conclude that the affirmative defense had been met in this case. Because an analysis of all of the evidence now before the court — and particularly Edmonds’s recantations — would more likely than not lead any reasonable juror to believe that Smith was not guilty of the crime of felony murder, we hold not that Edmonds rather than Smith is the actual killer, but that Smith has successfully passed through Schlup’s “actual innocence” gateway and is entitled to a hearing on the merits of his constitutional claims.

B.

To the extent that our analysis relies on the truthfulness of Edmonds’s recantations, there remains the question of his credibility. Under other circumstances, we might remand for an evidentiary hearing to allow the district court to assess the trustworthiness of Edmonds’s more recent statements. See Jaramillo, 340 F.3d at 883 (holding that new evidence presented by petitioner would, if credible, be sufficient to support a finding of actual innocence and remanding for an evidentiary hearing for the court to make the necessary credibility determinations). The circumstances here, however, are not conducive to such a remand. In this case, due to the prosecution’s threats to pursue capital charges against him, Edmonds has refused to testify, even after being promised immunity from perjury (if he testifies as the prosecution wishes). The facts regarding the prosecution’s conduct have all been established in the record26 and, as the *823district court recognized, a further hearing would serve no purpose.27 We must now consider, then, whether the prosecution engaged in misconduct that substantially interfered with Smith’s efforts to establish his Schlup claim and, if so, what the remedy should be.

1. Prosecutorial Misconduct

The prosecution effectively prevented Edmonds from testifying by threatening to institute capital murder charges against him — specifically, as the district court put it, to “seek the death penalty” — if he testified in a manner that was consistent with his two affidavits. The state also offered to immunize him from perjury if he would withdraw his recantations and testify that Smith was the killer. Due to the drastic nature and the unusual character of the prosecution’s threats and, specifically, due to the state’s declaration of its intentions to subject him to the death penalty, Edmonds understandably invoked his Fifth Amendment privilege and refused to testify about any and all facts relating to the murder of Konzelman, even — as the district court notes — after he received the state’s offer of immunity if he testified against Smith. The prejudicial effect of the state’s actions is starkly demonstrated by the district court’s finding that “[a] logical inference” from Edmonds’s refusal to testify in the manner the state desired, even after the prosecution’s promise of immunity if he would do so, is that “if Edmonds did testify truthfully, he would say that Smith did not kill Emmitt [sic] Konzelman, and implicate himself as the killer.”

Threatening a potential witness for the defense with execution constitutes prosecutorial misconduct far more coercive than that present in any reported case of which we are aware. The cases in which courts have considered the prosecution’s threats to charge witnesses with perjury or other criminal offenses, have all involved the possibility of far less serious punishment. See, e.g., United States v. Vavages, 151 F.3d 1185, 1188, 1192 (9th Cir.1998); United States v. Lord, 711 F.2d 887, 889, 892 (9th Cir.1983); see also United States v. Morrison, 535 F.2d 223, 225, 228-29 (3d Cir.1976). Here, the prosecution’s unprecedented threat to seek the death penalty against Edmonds if he testified that Smith was not the killer was unquestionably coercive and constituted substantial interference with Edmonds’s decision whether to testify. Cf. Earp v. Omoski, 431 F.3d 1158, 1170 (9th Cir.2005) (“It is well established that ‘substantial government interference with a defense’s witness’s free and unhampered choice to testify amounts to a violation of due process.’ ”) (citing Vavages, 151 F.3d at 1188).28

*824The dissent repeatedly insists that we should remand for an evidentiary hearing on the question whether the prosecutor’s misconduct was intentional. See, e.g., dis. op. at 840. There is no cause for such a hearing in this case. The “intent” to which the dissent refers is the intent to cause a witness not to testify in a particular manner or not to testify at all. See Lord, 711 F.2d at 891. The prosecution here intimidated Edmonds into refusing to testify by making a threat of the gravest nature, a threat that was clearly improper, especially given the prosecution’s prior determination in Smith’s case that an appropriate punishment for the actual killer was 30 years to life. In cases in which the harassment, intimidation, or coercion of a witness is evident, intent is not at issue; rather, the issue under our prior cases has been whether the interference was “substantial” and whether it affected the witness’s decision to testify.29 Cf. Williams v. Woodford, 384 F.3d 567, 601-02 (9th Cir. 2004) (“Undue prosecutorial interference in a defense witness’s decision to testify arises when the prosecution intimidates or harasses the witness to discourage the witness from testifying, for example, by threatening the witness with prosecution for perjury or other offenses.... The prosecution’s conduct must amount to a substantial interference with the defense witness’s free and unhampered determination to testify before the conduct violates the defendant’s right to due process.”); Vavages, 151 F.3d at 1189 (“A defendant’s constitutional rights are implicated only when the prosecutor or trial judge employs coercive or intimidating language or tactics that substantially interfere with a defense witnesses] decision whether to testify.”).30 *825In sum, our cases have ordered a remand only where necessary to make a showing of intent, see Lord, 711 F.2d at 890; given the prosecutorial conduct involved, no such showing is required here.

It is evident that the prosecution’s threat to seek the death penalty against Edmonds was intended to coerce Edmonds into changing his testimony or refusing to testify. “Where, under the totality of the circumstances, ‘the substance of what the prosecutor communicates to the witness is a threat over and above what the record indicates is necessary, and appropriate, the inference that the prosecutor sought to coerce a witness into silence is strong.” Vavages, 151 F.3d at 1190 (quoting United States v. Pierce, 62 F.3d 818, 832 (6th Cir.1995)) (internal quotation marks omitted).31 The prosecution warned Edmonds that if he testified in a particular manner — one that would be consistent with Smith’s defense — it would attempt to have him executed. In contrast, it informed him that if he withdrew his recantations and testified in accordance with its case against Smith, it would provide him with immunity from perjury. Interestingly, the state did not seek to prosecute Edmonds on the basis of his submission of the affidavits exculpating Smith and inculpating himself; it was only his testimony on behalf of Smith that, the state declared, would provoke it to seek to execute him. There can be no dispute that by constructing the options as it did, the prosecution clearly intended to compel Edmonds either to testify in the manner that would favor its case against Smith or not testify at all. Facing the possibility of a death sentence, Edmonds did not have a realistic option to testify in accordance with his previous recantations, even if — as the district court inferred — those recantations were likely truthful. Because it is clear that the prosecution’s threats intimidated Edmonds into *826invoking his Fifth Amendment privilege, even after he was offered immunity, we conclude that no further hearing regarding intent is required and that the prosecution’s interference constituted serious misconduct that denied Smith a fair hearing on his Schlup claim. We therefore proceed to a discussion of the appropriate remedy.

2. Remedy

In determining what the remedy should be, we observe that in all of the prior cases we have uncovered, the prosecutorial misconduct affected the outcome of a trial to determine the defendant’s guilt or innocence. The remedy we imposed in those cases was to reverse or vacate the defendant’s conviction. Here, by contrast, the underlying question is only whether the federal courts should be permitted to hear a petitioner’s constitutional claims on the merits. Because of that difference, we approach the remedy differently. We may look to our precedents regarding prosecutorial interference with an actual trial for guidance, but those cases do not control this appeal. The novel procedural posture of the case before us permits us to remedy the prosecutorial misconduct by far less drastic means.

As an initial matter, we cannot accept the dissent’s suggestion to remand to the district court for an evidentiary hearing on actual innocence under Schlup. Dis. op. at 840-43. Such a remedy would be both futile and ineffective, even were we to require the state to grant Edmonds “use immunity” for his testimony at that hearing.32 The district court found that holding a Schlup hearing would be a “pointless exercise” so long as Edmonds refused to testify. See supra note 27. Although use immunity sometimes serves to encourage defense witnesses to testify after they have been deterred from testifying previously, the present case is not the usual one. Death is indeed different, and threatening to seek a witness’s execution if he testifies in a manner that would exculpate the defendant is fundamentally different from threatening to charge a witness with perjury or with some other less serious non-capital offense.

In this case, use immunity would not effectively counter the threat of execution, as the state would be free to seek the death penalty even if barred from relying on Edmonds’s testimony.33 If Edmonds did decide to testify, there would be no way to ensure that the looming prosecutorial threat of execution would not significantly influence his testimony. Without assurances that he would not be facing capital charges as a result of his testimony — assurances that use immunity would not provide — Edmonds could not, in light *827of the prosecutor’s threats, be expected to risk the imposition of a death sentence by providing testimony exculpating Smith and inculpating himself, regardless of its truthfulness. Any testimony that contradicted his affidavits would be of doubtful reliability-

We are faced with three alternatives, then, for remedying prosecutorial misconduct in a case such as Smith’s, in which the misconduct occurs at the Schlup stage and the petitioner has presented affidavits from the witness regarding the facts to which he would have testified had the state not caused him to refuse to do so. First, we could issue the writ. Second, we could prohibit the state from asserting the procedural bar that blocks the petitioner from presenting the merits of his claims. Third, we could afford the affidavits a presumption of truthfulness and consider the Schlup inquiry on that basis. On balance, we conclude that the third alternative— presuming Edmonds’s recantations to be true for the limited purpose of determining whether Smith may pass through the Schlup gateway — is the least intrusive on the state’s interests and most consistent with the procedural posture of the case.

In this regard, we emphasize that we are not here considering how a state must conduct a trial in its courts or how the state courts must remedy instances of prosecutorial misconduct; instead, we are determining whether a federal proceeding has fairly afforded a petitioner his federal rights and what remedy will best allow a federal court to fulfill its proper role. We do so mindful that our remedy should be narrowly tailored and as respectful of the state’s interests as possible.

The first alternative — issuing the writ of habeas corpus — would have the effect of requiring the state to retry Smith. This is the usual remedy when the prosecutor’s misconduct interferes with the defendant’s ability to present his defense at trial. In most instances of prosecutorial misconduct causing defense witnesses to withhold testimony, this court has reversed or vacated the defendant’s conviction. See, e.g., Vavages, 151 F.3d at 1193; Young, 86 F.3d at 946; Westerdahl, 945 F.2d at 1088; Lord, 711 F.2d at 891-92. In the context of a hearing to determine whether a prisoner should be precluded from presenting his constitutional claims because of his failure to comply with procedural prerequisites, however, it is not necessary to reverse the conviction or issue the writ; less consequential and more appropriate remedies are available.

The second alternative — prohibiting the state from asserting the procedural bar— is far less intrusive on the state’s interests. It would simply allow the federal proceedings to continue. This remedy is better tailored than reversal to the procedural posture of the case. It is, nevertheless, overbroad. Such a remedy could turn prosecutorial misconduct into a windfall by granting relief to petitioners who have suffered no prejudice as a result of the prosecutor’s misconduct. The remedy would allow any petitioner who was unable to present a witness because of prosecutorial misconduct to overcome his procedural default, even if the witness’s statements, along with the rest of the evidence, would not be sufficient to meet Schlup’s strict standard.

We conclude that the third alternative— the more measured remedy of deeming Edmonds’s affidavits credible for purposes of the Schlup determination — is clearly preferable. Such a remedy provides relief only to a petitioner who has been deprived of the opportunity to present testimony that, if believed, would enable him to make the showing Schlup requires. Such a resolution emphatically does not provide the witness with immunity for a murder *828he may have committed. Although our remedy may be different than those we have applied in the past, it is because the procedural posture of this ease is different. Thus, the remedy is not “extraordinary,” dis. op. at 843. Indeed, given that neither we nor the district court has yet considered the merits of the alleged constitutional violations, it would be far more extraordinary were we to reverse Smith’s conviction and remand for retrial, see, e.g., Vavages, 151 F.3d at 1193, or reverse his conviction and order acquittal if the state does not request use immunity, see, e.g., Westerdahl, 945 F.2d at 1088.34 Here, we stop far short of that drastic relief, holding merely that Edmonds’s affidavits must be presumed to be true for purposes of the Schlup inquiry. Even considering the consequences of the remedy we apply— when Edmonds’s affidavits are presumed to be true, Smith is able to pass through the Schlup gateway and present the merits of his constitutional claims — our relief is far less drastic than that afforded the defendants in our earlier cases. In the end, Smith’s case will rise or fall on the validity of his constitutional challenges. Only if he succeeds on the merits will a writ issue and his case be returned to the state courts for a new trial.35

Because the prosecution’s threat of a death sentence was so coercive as to prevent Edmonds from testifying, and because the state’s conduct has, at the very least, severely tainted any testimony not exculpatory of Smith that Edmonds might offer at a Schlup hearing, Edmonds’s affidavits must be presumed to be credible, for purposes of resolving the question whether Smith’s procedural default should bar him from presenting his habeas claims on the merits. On the basis of that presumption, as well as the other available evidence, Smith has made the requisite showing of actual innocence necessary to proceed to a hearing on his constitutional claims.

V.

We hold that Smith’s constitutional claims are not proeedurally barred insofar as they relate to his felony murder conviction. We reverse and remand to the district court for a determination of those claims on the merits. Additionally, because it never decided whether there should be an evidentiary hearing to develop the factual basis of Smith’s constitutional claims,36 we leave it to the district court to decide in the first instance whether such a hearing should now be conducted.37

*829REVERSED and REMANDED for further proceedings consistent with this opinion.

. On the basis of the oral argument and briefs submitted to this court, we construe Smith's appeal as challenging the procedural default of his claims only with respect to his felony murder conviction and not with respect to his robbery conviction. Therefore, a Schlup showing of "actual innocence” as to the felony murder charge will suffice to overcome the challenged procedural default with respect to that conviction and will allow Smith to proceed on the merits of the claims relevant to that charge only. Smith offers no facts or arguments that would support a claim of actual innocence of the robbery offense or would otherwise excuse his procedural default as to that conviction. Thus, the robbery conviction stands.

. The evidence suggesting Edmonds was the person responsible for the killing will be explored in more detail in the "actual inno*808cence” section of the opinion. See infra Part IV.A.

. Although the record regarding Edmonds’s actual sentence is not clear, it appears from his subsequent criminal record that his actual sentence may have been substantially less than the purported recommendation.

. The district court held that because it found all of Smith's claims to be procedurally defaulted, it did not need to address any of them on the merits. In doing so, the district court failed to address the freestanding claim of innocence asserted in Smith’s amended federal habeas petition, even though it noted that innocence claims are not cognizable in Oregon courts. Cf. Herrera v. Collins, 506 U.S. 390, 393, 113 S.Ct. 853, 122 L.Ed.2d 203 (1993) (in which petitioner asserted a freestanding claim of actual innocence based on newly discovered evidence and contended that his execution would therefore violate the Fourteenth and Eighth Amendments). In his amended federal habeas petition, Smith, like Flerrera, argued that he is actually innocent of felony murder and that his conviction therefore violates the Fourteenth Amendment's guarantee of due process and the Eighth Amendment's prohibition against cruel and unusual punishment. Smith appears, however, to have intended to assert both a freestanding claim and the type of claim that serves to excuse procedural default (a Schlup claim). See id. at 404-05, 113 S.Ct. 853 (distinguishing between the "colorable showing of actual innocence” that is used to "seek excusal of a procedural error ... [of] an independent constitutional claim” and a "freestanding claim[] of actual innocence”).

We need not decide now whether, as the magistrate judge concluded in his Findings & Recommendations report, any freestanding (or "Herrera ”) claim is cognizable only in capital cases, see id. at 417, 113 S.Ct. 853 (assuming, for the sake of argument, that "in a capital case a truly persuasive demonstration of 'actual innocence’ made after trial would render the execution of a defendant unconstitutional, and warrant federal habeas relief if there were no state avenue open to process such a claim”); for the district court did not adopt that ruling, but instead held, erroneously, that Smith's freestanding claim of actual innocence was procedurally barred. A freestanding "actual innocence” claim, however, may not be procedurally defaulted. Rather, this type of claim constitutes an exception to the procedural default provisions. Moreover, in this case, there was, as noted earlier, no way for Smith to bring the claim in state court. The district court should have addressed the freestanding claim directly. Given the "extraordinarily high” standard established in Herrera, however, see id., Smith's prospects for prevailing on that claim would appear to be minuscule at best.

. For the same reason, we need not reach the question whether Smith was denied a fair opportunity to present his federal constitutional claims during the state post-conviction process.

. In Smith’s case, he is no longer able to obtain state review of his claims, given Oregon’s procedural bar regarding time limits for filing petitions for post-conviction relief. See Or. Rev. Stat. § 138.510 (2003).

. Bousley v. United States, 523 U.S. 614, 118 S.Ct. 1604, 140 L.Ed.2d 828 (1998), also requires Smith to prove his Schlup claim with regard to the aggravated murder charge foregone by the state as a result of the plea agreement. See id. at 624, 118 S.Ct. 1604. Here, a determination that we cannot have confidence in his felony murder conviction would necessarily imply the same finding of "actual innocence” with regard to the more serious crimes foregone by the state.

. The fact that, just three years after the date of his plea agreement, Edmonds committed a residential burglary, kidnapping, and rape, also tends to support the conclusion that Edmonds, not Smith, was the killer in the instant case. Of course, if Edmonds were on trial, the introduction of this "bad act" evidence to prove his guilt would be unfair to him and would almost certainly violate the Federal Rules of Evidence. This habeas proceeding, however, concerns Smith, not Edmonds, and ignoring this evidence would prejudice Smith in his effort to prove his innocence. On habeas review under AEDPA, we may consider such evidence, which was “unavailable at [Smith’s] trial,” without regard to "the rules of admissibility that would govern at a trial.” Schlup, 513 U.S. at 327-28, 115 S.Ct. 851 (expressly permitting consideration of evidence that was "excluded” or "alleged to have been illegally admitted”).

The dissent cites Old Chief v. United States, 519 U.S. 172, 180-82, 117 S.Ct. 644, 136 L.Ed.2d 574 (1997), to warn against the dangers of propensity evidence. Dis. op. at 838 n.17. Old Chief, however, describes "bad act” evidence as "relevant,” "logically ... persuasive,” and "probative.” 519 U.S. at 180-81, 117 S.Ct. 644 (quoting United States v. Moccia, 681 F.2d 61, 63 (1st Cir.1982) (Breyer, J.); Michelson v. United States, 335 U.S. 469, 475-76, 69 S.Ct. 213, 93 L.Ed. 168 (1948)). It then explains that this relevant, probative evidence is nonetheless excluded at trial when offered against a criminal defendant because it is highly prejudicial and its admission would create a risk that the defendant could be convicted on less than evidence beyond a reasonable doubt. Id. at 180-82, 117 S.Ct. 644. Because there is no such risk to any person in this case, consideration of Edmonds’s bad acts is not only permitted under Schlup, but also makes sense under the rationale of Old Chief.

. The dissent argues that Edmonds's affidavits are "internally inconsistent and plainly inconsistent with each other.” Dis. op. at 842. The language cited by the dissent is taken out of context. Edmonds stated in his first recantation — after stating explicitly that he had committed perjury, that Smith did not kill Konzelman, and that he was attempting to "set the record straight” — that "[t]he reason[he] lied under oath was because [his] attorney ... led [him] to believe that Roger Smith was about to testify and leave [him] to do a life sentence for a crime he did not commit.” Given the unequivocal nature of every other statement relating to guilt in the affidavit, this confusing statement could have resulted either from Edmonds’s characterization of his defense attorney's advice or the fact that he swore out his first affidavit without the benefit of an attorney's supervision to help him clarify his statements. In contrast *815to his first recantation, Edmonds’s second recantation does not contain any such confusing language and even includes a statement that is likely a more accurate characterization of the allegedly inconsistent statement in his first recantation: 'T was motivated to falsely implicate Mr. Smith because my trial attorney and others counseled that, by doing so, I could avoid prosecution for murder or aggravated murder."

. Even though polygraph evidence may not be admissible at trial under Oregon law, we may consider this evidence under Schlup. Schlup, 513 U.S. at 328, 115 S.Ct. 851.

. Further, the details about the actual murder that Edmonds was able to provide could best be explained by the fact that he was the attacker and therefore would have known the details of what occurred in the bedroom. A conclusion that the second, non-attacking burglar could provide such corroborated details of the murder would conflict with Mrs. Konzelman's statement that she could only catch a “brief glimpse” of the second burglar "as he went down the hall.” Thus, it is doubtful that Edmonds would have known the details of the attacker’s interaction with the Konzelmans in the bedroom unless he had been there for the duration, and not just passing by — in other words, unless he was the actual attacker.

.The dissent contends that Edmonds’s statements to his cellmate Samuel Seaman were consistent with the statements that Edmonds gave to police and therefore support the argument that Smith was the killer. Dis. op. at 833. It should be noted, however, that Sea*816man was also skeptical of Edmonds's story and told officers that "the way [Edmonds] talk[ed] about it [was as if] he [was] trying to cover up for himself.”

. Edmonds testified at Smith’s sentencing hearing that when they got to the Konzelmans’, it was he, and not Smith, who had been wearing a Levi's jacket with a white lining.

. Mrs. Konzelman also stated in her interview with police that she believed the pants were more like dress pants, not jeans.

. Like the dissent, we understand that Mrs. Konzelman was confused during parts of her testimony. Nevertheless, certain parts of her testimony are clear and unequivocally point to Edmonds, not Smith, as the attacker.

. The dissent criticizes us for crediting Edmonds’s statements only when it is convenient. See dis. op. at 841-42. To the extent that Edmonds has recanted his earlier testimony that Smith was the actual attacker, it is true that we have deemed the earlier testimony not credible and his later testimony truthful. Such a conclusion does not preclude us from evaluating other testimony given by Edmonds — such as what the two of them were wearing on the night of the murder, or where they were with respect to one another once inside the house — and the extent to which a jury would be likely to find such testimony credible, in light of the other available evidence. Further, because Edmonds’s various descriptions of the incident — as provided by his testimony and prior to his recantations— are not always consistent, any factfinder would be forced to choose the elements of his story that it found credible and those that it did not; the dissent’s suggestion that we must either afford full credibility to his testimony or none at all creates an unreasonable and unrealistic choice.

.Because a rope was found just to the "left of the bedroom door,” Smith likely dropped the rope when he paused briefly in the bedroom doorway and before he continued down the hall. Therefore, the fact that Smith was carrying the rope does not mean that he entered the Konzelmans’ bedroom and can be *818reconciled with both Mrs. Konzelman’s recollection that only the attacker was in the bedroom and Edmonds's second recantation, in which he stated that Smith never entered the bedroom.

. Still, the dissent cites State v. Cornell, 314 Or. 673, 842 P.2d 394 (1992) (in banc), as support for its conclusion that the rope Smith carried was a dangerous weapon. In Cornell, the Oregon Supreme Court upheld the felony murder conviction of a defendant who stuffed toilet paper into his victim's mouth and "hogtied” him around the neck with a cord, which asphyxiated him. Id. at 396. The opinion says not a word about whether a cord or rope is a dangerous weapon — it had no need to, because the use of a dangerous weapon is not an element of felony murder in Oregon, only an element of the affirmative defense. See Or. Rev. Stat. § 163.115. The defendant in Cornell, however, had not raised the affirmative defense, and thus the court had no need to consider the question in that case.

. Bouse testified that when the three first entered, Smith went into the house while Edmonds searched the garage, and Edmonds testified that during the time that he and Smith were both exploring the garage, they were searching in separate parts of it. Therefore it is likely that Edmonds found the crowbar without Smith’s knowledge.

. Like much of the testimony provided, Edmonds's testimony as to how and when the two men entered the house is unclear. The accounts the dissent refers to in which Edmonds suggested that he and Smith entered together are far from precise. See dis. op. at 848 n. 44. Such accounts are unclear as to whether the two entered at the same time, so *820that each could observe what the other may have been holding, and do not suggest that they kept track of one another’s whereabouts while in the house.

. According to Edmonds's initial testimony, he first saw the crowbar in Smith’s hands when both he and Smith entered the Konzelmans’ bedroom together. As noted supra, Edmonds's version of the story, insofar as it places Smith as the killer, is discredited by his subsequent recantations. Further, although it is possible that Edmonds may have reversed his and Smith’s roles in providing his original version of events, the fact that Mrs. Konzelman' — a neutral witness — "recalled the killer picking the crowbar off the floor before striking the blows” and that "he had not been carrying the crowbar around in his hand” suggests not only that Edmonds was the one handling the weapon, but also that Smith was never aware that Edmonds was armed, either before or after seeing Edmonds in the Konzelmans’ bedroom when he paused by the doorway.

. Edmonds told a Linn County detective that both he and Smith wore bandannas as masks.

.See Enmund v. Florida, 458 U.S. 782, 799-800, 102 S.Ct. 3368, 73 L.Ed.2d 1140 (1982) (noting that "only about one-half of one percent of robberies result[] in homicide” and that "killings only rarely occur during robberies”). Although Enmund's data are somewhat outdated, more recent data support our conclusion as well. Using the same data source that the Supreme Court used in Enmund (with updated data), and the same methodology, see Enmund, 458 U.S. at 800 n. 24, 102 S.Ct. 3368, data from 2005 show that less than one quarter of one percent (0.22%) of all robberies result in homicides. See U.S. Dep’t of Justice, Federal Bureau of Investigation, Crime in the United States 2005.

The dissent opines that these statistics are not "sufficiently refined,” and that the proper measure of risk is the probability that a burglary that “might turn confrontational” will result in death. Dis. op. at 851 n. 52. By definition, however, every robbery included in the FBI report — the set of crimes to which the 0.22% figure applies — involves the use or threat of force or violence or putting the victim in fear. See Crime in the United States 2005. Burglaries, which are by definition less confrontational, are not included in the Report’s statistics unless they turn into robberies, such as in the crime involved in this case.

. The district court conceded that "[t]he word likely does give some pause,” but failed to require a stronger nexus than a theoretical possibility.

. In concluding otherwise, the dissent again cites State v. Cornell, 314 Or. 673, 842 P.2d 394, in which the Oregon Supreme Court upheld the felony murder conviction of a defendant who used a cord to "hog-tie” and *822asphyxiate his victim. The dissent asserts that "a reasonable juror would very likely conclude that Smith brought the rope into the house with the contemplated purpose of using it to restrain the victims” and argues that Cornell supports the conclusion that "[tjying victims up is highly dangerous conduct that is likely to result in death.” Dis. op. at 851 (emphasis added). Therefore, the argument goes, Smith must have known that death was likely to result from his and Edmonds's robbery. The dissent's argument fails on two levels. First, even if a reasonable juror did conclude that Smith "contemplated” using the rope to restrain the Konzelmans, this is a far cry from concluding that it was likely that he would restrain them. Second, and more important, even if it was likely that Smith would restrain the Konzelmans, this hardly means that it was likely that one of them would die. Contrary to the dissent's assertion, tying the hands, feet, or even body of another person is not likely to result in death. The fact that the defendant in Cornell tied his victim around the neck after stuffing toilet paper into his mouth, and that this extraordinary action caused death, does not even come close to supporting the conclusion that Smith had reason to believe that death was likely to ensue if he used the rope to restrain the Konzelmans.

. The district court made the following findings: "The Linn County prosecutor's office ... warned Edmond’s [sic] counsel that the district attorney would seek the death penalty if Edmonds testified that he, and not Smith, actually killed Emmett Konzelman. On the *823other hand, if Edmonds withdrew that affidavit and testified that Smith was the killer, the State agreed to give Edmonds immunity from charges of (allegedly) filing a false affidavit. After conferring with his counsel, Edmonds refused to testify on the ground that the answers might incriminate him.”

. The district court found that to hold an evidentiary hearing “would be a pointless exercise so long as Edmonds refuses to testify.” Given that Edmonds's testimony would be relevant only to Smith's claims of innocence, it is most likely that the evidentiary hearing to which the district court referred was an evidentiary hearing regarding actual innocence under Schlup. A more limited hearing to determine only the credibility of Edmonds's exculpatory statements would be equally pointless. Edmonds would have all the same reasons not to testify regardless of the scope of the hearing, and without his testimony either future hearing would be of no avail.

. We reject the dissent's argument that the prosecution’s suggestion that the court provide Edmonds with counsel demonstrates that it was not seeking to silence him but to lay a foundation for prosecuting him. Dis. op. at 836-37. Certainly the prosecution was aware *824that appointed defense counsel would greatly aid its efforts to achieve its goal' — appointed counsel almost certainly would advise Edmonds that he would run a substantial risk of receiving a capital sentence if he testified, consistent with his affidavits, that Smith was not the killer. Counsel might even advise Edmonds that, as the dissent now argues, counsel’s appointment was a necessary preliminary step to Edmonds’s execution.

. In cases of threat or coercion, no separate and additional showing of intent has been required. The prosecution's affirmative acts — in this case, the prosecution's threat to seek the death penally against Edmonds — • inherently demonstrate the requisite intent. See, e.g., Earp, 431 F.3d at 1168-71 (remanding not to determine prosecutorial intent, but only to evaluate the credibility of affidavits alleging that the prosecutor had threatened a potential defense witness); Vavages, 151 F.3d at 1187, 1190-93 (reversing a conviction without an evidentiary hearing where the prosecutor threatened a potential defense witness with perjury charges and the withdrawal of a plea agreement in an unrelated prosecution); Morrison, 535 F.2d at 225-29 (reversing a conviction without an evidentiary hearing where the prosecutor made repeated threats and warnings to a potential defense witness, culminating in a "highly intimidating personal interview”). Cases in which the prosecution merely denies immunity to potential defense witnesses, by contrast, typically do require remand for evidentiary hearings to determine prosecutorial intent. See, e.g., United States v. Young, 86 F.3d 944, 946-49 (9th Cir.1996) (remanding for an evidentiary hearing on intent where the prosecution granted immunity or favorable plea agreements to prosecution witnesses but withheld immunity from a potential defense witness); United States v. Westerdahl, 945 F.2d 1083, 1087 (9th Cir.1991) (same); Virgin Islands v. Smith, 615 F.2d 964, 969 (3d Cir.1980) (remanding where the prosecution withheld immunity from an important potential defense witness). Cases in which the record of prosecutorial conduct is unclear may also require remand for clarification. See, e.g., Lord, 711 F.2d at 891-92 (remanding for "further clarification of the prosecutor's pre-trial comments” to a potential defense witness). There is no need for a remand here, however, as the prosecution indisputably took affirmative steps to cause Edmonds not to testify in the manner set forth in his affidavits.

. The dissent argues that we cannot look to United States v. Vavages because Smith cited instead our decision in United States v. Westerdahl, and the two cases represent doctrines *825that are entirely distinct. Hence, the dissent's argument goes, Smith has waived his "Vavages argument.” We disagree. Although the two cases admittedly put forth slightly different doctrinal tests, they address precisely the same doctrinal issue: prosecutorial misconduct that materially affects a defendant's opportunity to present his case. Furthermore, the cases descend from the exact same doctrinal ancestor: Webb v. Texas, 409 U.S. 95, 93 S.Ct. 351, 34 L.Ed.2d 330 (1972). Indeed, the dissent itself points out that Vavages stems directly from Webb, dis. op. at 836; Vavages also leans heavily on United States v. Morrison, a decision that found Webb " controlling],” 535 F.2d at 227. Westerdahl is simply a bit more removed, relying primarily on United States v. Lord, which in turn relied on Morrison (which found Webb controlling). Once it is clear, as here, that a defendant alleges prosecutorial misconduct that interferes with the presentation of his witnesses, we would be remiss to ignore certain prior decisions of this court (e.g., Vavages) while focusing exclusively on others (e.g., Westerdahl). Certainly, it would not be consistent with fundamental concepts of justice to deprive an individual of a hearing on his constitutional claims because he cited one rather than the other of two related cases addressing the same basic issue.

. In Vavages, a potential witness invoked her Fifth Amendment right against self-incrimination based on the prosecution’s threats to charge her with perjury and withdraw a plea agreement in an unrelated prosecution if she testified for the defense. 151 F.3d at 1187— 88. There, we held that "[i]t [did] not require much of an interpretive gloss on the prosecutor’s warning to conclude that unless [the defense witness] changed her testimony or refused to testify at all, she would be prosecuted for perjury and suffer any attendant consequences” and that "there [was] no question that the prosecutor’s warnings were a 'but for’ cause of [the witness’s] refusal to testify,” thus constituting substantial interference. Id. at 1190-91. The Vavages court held that the district court had "clearly erred in finding that Vavages[ ] had not been prejudiced by the prosecutor's substantial interference with [the witness’s] decision whether to testify and [therefore] reverse[d] Vavages’[s] conviction and remand[ed] for retrial.” Id. at 1193.

. Use immunity provides that, "while the government may prosecute the witness for an offense related to the subject matter of the witness's testimony, the testimony itself and any 'fruits' thereof may not be used against the witness in any criminal case except a prosecution for perjury arising out of the testimony.” Lord, 711 F.2d at 890.

. If Smith's conviction were reversed, the state could subpoena him and compel him to testify, truthfully, that Edmonds and not he was the killer. Edmonds certainly has reason to be concerned that Smith, once freed from prison, might not be willing to accept the consequences of refusing to testify. It is also possible that the state could rely on the declarations against interest contained in Edmonds's affidavits as well as any statements that a jailhouse informant might make. To be sure, it is not — as the dissent asserts- — -Edmonds’s "eligibility] for renewed charges” that has prevented him from testifying, dis. op. at 837, but rather the state's warning that it -would seek the death penalty if he testified. Use immunity would provide no guarantee that the state would not succeed in carrying out its threat.

. In Westerdahl, we reversed the defendant’s conviction and remanded to the district court. Because, unlike this case, it was unclear in Westerdahl whether the prosecutor had acted with the intent to distort the factfinding process, we instructed the district court to make that determination first. We then directed the district court to acquit Westerdahl if it found intent, unless the government requested use immunity for his witness. See Westerdahl, 945 F.2d at 1088.

. The dissent asserts that by adopting the third of the three remedial alternatives we consider, we have merely chosen a “bulldozer” instead of a “wrecking ball”; in other words, the dissent thinks that all of our considered alternatives are intrusive. Dis. op. at 841 n. 22. Surely, however, permitting a petitioner to present his constitutional claims in federal court bulldozes far less of the state’s efforts than reversing his conviction and forcing the state to try its case a second time. The Westerdahl remedy, which the dissent appears to prefer, entails the far more drastic latter consequences. See Westerdahl, 945 F.2d at 1088.

. The district court did not consider that question because it deemed Smith's claims proeedurally barred and denied him an evidentiary hearing on that basis.

. As we mentioned earlier, the district court must rule on the Herrera claim, although, as we also noted, that claim appears to have *829only a minimal chance of success. See supra note 4.






Dissenting Opinion

BYBEE, Circuit Judge,

dissenting.

Roger Smith and Jacob Edmonds needed money. Bad. For an Anthrax concert. So they burglarized the Konzelmans’ garage to obtain tools. They then put bandannas on their faces to disguise themselves and made up code names. Smith took a rope into the house. One of them took a three-foot-long crowbar as well and savagely beat the Konzelmans with it, killing Mr. Konzelman and severely injuring Mrs. Konzelman. When Edmonds entered into a plea agreement and agreed to testify against Smith, Smith pled no contest to felony murder. Edmonds now claims that Smith did not kill Mr. Konzelman, and the majority, willing to believe him, thinks that Smith is innocent. The problem is that Smith was convicted of felony murder. Even accepting every word of Edmonds’s latest story as true— and there is good reason not to — it is entirely irrelevant that “Edmonds, and not Smith, was the killer,” or that “Smith did not kill Konzelman.” Maj. Op. at 814. Nonetheless, the majority, reviewing the record afresh and construing all evidence in favor of Smith, concludes that Smith is actually innocent because the evidence “would more likely than not lead any reasonable juror to believe that Smith was not guilty of the crime of felony murder.” Id. at 822.

My view of the majority’s analysis of the evidence can perhaps best be described by paraphrasing author Mary McCarthy: I disagree with nearly every word the majority has written, including “and” and “the.” My profound disagreement is not limited to the facts, but runs throughout the majority opinion. I cannot agree that the prosecution engaged in misconduct when it informed Edmonds, correctly, that he was exposing himself to criminal liability by making statements that Smith did not kill Mr. Konzelman. Even accepting, arguendo, that this did constitute prosecutorial misconduct, I see no justification for the presumption of truthfulness that the majority affords Edmonds’s affidavits. The majority’s newfound presumption is contrary tó our prior cases, all of which require that we remand to the district court for further findings as to the prosecution’s motives. Yet, all of this is largely a sideshow because even indulging the majority’s presumption, the evidence in this case is still insufficient to qualify Smith for Schlup v. Delo’s actual innocence gateway. See 513 U.S. 298, 314-15, 115 S.Ct. 851, 130 L.Ed.2d 808 (1995). Indeed, in the thousands of habeas cases that have applied Schlup, I have been unable to locate a single case where a petitioner convicted of felony murder was able to establish actual innocence. Not one. And there is no basis for making this case the first. By all accounts, Smith was present, gloved, disguised, and armed with at least a rope when Mr. Konzelman was murdered in his home during the burglary. There isn’t a jury in the country that wouldn’t convict Smith today on a charge of felony murder. I therefore respectfully, but vigorously, dissent.

I. FACTS

In the early morning hours of April 3, 1989, after consuming methamphetamine, Roger Smith, Jacob Edmonds, and Marlin “Hooter” Bouse1 decided to steal enough *830money or property to purchase tickets to the Headbanger’s Ball in Portland the following night. The trio drove through various neighborhoods in Edmonds’s truck, looking for a good target. They ultimately seized on the residence of Emmett and Elma Konzelman because the garage door was open.

The trio parked Edmonds’s truck a few houses up the road from the Konzelmans’ home, and all three entered the garage to look for valuables. While the men were in the garage, Smith leaned into the doorway connecting the garage to the house. The door then slammed shut, and, scared that they had awoken any inhabitants, the three took off running. At this point, Bouse became separated from Edmonds and Smith. Bouse took a beer, a hat, and a pair of gloves from the garage and hid in the bushes near Edmonds’s truck. He did not see Edmonds and Smith again until forty-five minutes later, when they returned to the truck together.

The exact events that transpired in those forty-five minutes are subject to conflicting accounts and inferences. What is clear is that Edmonds and Smith burglarized the Konzelman residence during this time, and that one of the two men took a large crowbar from the Konzelman residence and savagely attacked the elderly Konzelmans in their bed, killing eightyseven-year-old Emmett and severely injuring seventy-four-year-old Elma.

According to Elma Konzelman, she awoke to find two men standing in her small bedroom. One was positioned in front of her dresser, while the other stood in the doorway for a short time before walking down the hallway. The man by the dresser told her to roll over, and she woke Emmett, who started to climb out of bed. The man by the dresser told Emmett to “Lay down, old man,” and when Emmett did not respond, the burglar proceeded to beat him with the crowbar. He then beat Mrs. Konzelman in the head with the crowbar2 before resuming his bludgeoning of Mr. Konzelman.

Understandably, Elma Konzelman’s description of events is somewhat hazy. She could not recall at trial whether the lights were on or not, but stated that she could “see plainly.”3 However, because both men used bandannas to mask their faces and wore hats, she was not able to describe either man very well. She testified that she initially thought that the man by the dresser was Emmett, who also owned a similarly lightcolored jacket, but on further reflection thought it more likely that she saw the inner liner of a short jacket worn inside out, as she could see the tags. She described his pants as dark4 and his *831hat as a “skull cap” or a “stocking cap.” She said that he was white, not very tall, “not exactly skinny, but not heavy,” and that he did not have long hair. She described him as rocking back and forth on the balls of his feet, as if he couldn’t stand still.5 She did not remember seeing any gloves on the murderer’s hands. She could not describe the man in the doorway, who she said only stood there for a short time before leaving to resume ransacking her house. The second man also wore a bandanna over his face.

Mrs. Konzelman also remembered the attacker ripping out her phone lines after the attack to prevent her from calling the police, and that she heard a voice saying “Low” repeatedly.6 Mrs. Konzelman said that she thought that the voice saying “Low” was not the voice of the man who attacked her, but that it was hard for her to be sure because he had only said that one word over and over.7

Smith has never given a detailed account of that evening on the record.8 To quote the district court, however, he has “steadfastly denied hitting either victim with the crowbar, or even carrying the crowbar into the Konzelman home.” Smith wrote a letter to Elma Konzelman that was read during his sentencing. In this letter, he implicitly admitted his involvement in the burglary but maintained that he was not the killer, even though he could not remember certain parts of the evening. He also stated that Emmett Konzelman’s death was “a very violent and tremendous accident” that “never would have happened if it wasn’t for drugs.”

In contrast, Edmonds has given many accounts of that evening, which together encompass nearly every possible way that the burglary and murder might have occurred. Nonetheless, when considered in conjunction with the physical evidence recovered by the police and the testimony of the other witnesses, some things become clear. Both Smith and Edmonds were in the garage together before entering the house. There, they found two hats, and each put one on to disguise himself before entering the house. One burglar wore a baseball cap or a skull cap, while the other wore a floppier, fedora-type hat, with a brim running all the way around it. The men were also together when they both tied bandannas over their faces. Both burglars wore gloves, though it is not entirely clear who wore which gloves. Edmonds has stated that he wore a pair of white latex gloves, while Smith wore a pair of black leather gloves. However, evidence in the record also suggests that Smith wore a single brown work glove taken from the Konzelmans’ garage, and Smith asserts in his opening brief on ap*832peal that he wore one or two such gloves.9 In the garage, the pair found a crowbar and a rope. They brought both items into the house, with an eye toward possibly tying up the residents with the rope. According to affidavits from Smith’s two trial lawyers, Smith admitted to carrying the rope into the house. Police found the rope on the floor of the Konzelmans’ bedroom. The crowbar was found on the floor of the Konzelmans’ kitchen.

After leaving the Konzelman residence, Smith and Edmonds returned to the truck, where they rejoined Bouse. Edmonds drove the trio to Smith’s father’s house. Meanwhile, Elma Konzelman waited until she was sure that the burglars were gone before climbing out of her bed. When she realized that all of her phones had been disabled, she found her cane and carefully made her way next door to the neighbors’ house for help. The Konzelmans were rushed to the hospital. Mrs. Konzelman required surgery, including the implantation of a metal plate in her head, but she survived her injuries. Sadly, Mr. Konzelman was not as fortunate; he succumbed to his injuries sixteen hours later.

The trio was ultimately caught through a serendipitous convergence of good luck and sharp police work. While speaking with Edmonds about an unrelated offense, the police discovered that he had attended a particular Judas Priest concert in Seattle. A ticket stub from that same concert was found in the street about a block away from the Konzelmans’ on the morning after the burglary. After further investigation, the police confirmed that the ticket belonged to Edmonds. When confronted with this evidence, Edmonds admitted to his involvement with the burglary and identified Bouse and Smith as his companions that evening. Smith was charged with seven counts: aggravated murder, felony murder,10 burglary, two counts of robbery, and two counts of assault.

Other evidence that corroborated Edmonds’s version of events,11 Smith’s violent *833history, and Edmonds’s apparent credibility12 all pointed to Smith as the killer. While in custody, Edmonds also made statements to his cellmate Samuel Seaman consistent with the version of events Edmonds told the police. Ultimately, Edmonds was given a plea bargain contingent upon passing a polygraph examination. Although the results of the examination were inconclusive, the examiner’s opinion was that Edmonds was being truthful, and the state proceeded with the plea agreement. Meanwhile, Smith’s defense was under the impression that Edmonds had passed his polygraph examination. The defense requested a copy of Edmonds’s polygraph test results, but the prosecution, believing that the defense was not entitled to them because the prosecution could not introduce them as evidence, refused to produce them.

Faced with Edmonds’s testimony against him and the possibility of the death penalty if he proceeded to trial, Smith entered into a plea bargain. He pled no contest to felony murder and one count of robbery. In exchange, the state dropped the other five counts against him, including the capital crime of aggravated murder, as well as the charges pending against Smith’s girlfriend, Jeannie Simons. Smith received a life sentence, with a minimum term of 30 years. He appealed to the Oregon Court of Appeals and the Oregon Supreme Court without success.

Smith then sought post-conviction relief in state court, alleging violations of both his Sixth Amendment right to counsel and his due process rights. Smith’s petition was denied on its merits, as was his request for new counsel. On appeal, new counsel was appointed for Smith. However, on appeal, he argued only that his request for new counsel should have been granted at his first post-conviction hearing. He did not appeal the decision of the lower court on the merits. The Oregon Court of Appeals ruled against Smith and the Oregon Supreme Court denied review. Because Smith never presented his substantive post-conviction claims to the Oregon Supreme Court, he has not exhausted his state remedies.

Meanwhile, Edmonds was released from prison. He committed another crime and himself received a life sentence. Edmonds then sent the district attorney a short affidavit in which he recanted his testimony against Smith, declaring that Smith did not kill Mr. Konzelman and that Edmonds had knowingly committed perjury. Yet, at the same time, he maintained his innocence, stating that he had perjured himself out of fear that “Roger Smith was about to testify and leave me to do a life sentence for a crime I did not commit.”

Smith returned to state court to seek post-conviction relief based on Edmonds’s contradictory affidavit. The state court ruled against Smith, and he appealed. While this appeal was pending, he filed a motion with the state court requesting the voluntary dismissal of his appeal and the court granted his request.

Smith then sought federal post-conviction relief. During the course of these proceedings, Smith’s counsel obtained a second affidavit from Edmonds. This affidavit again stated that Smith did not kill Mr. Konzelman and that Edmonds had committed perjury. The affidavit also stated that Edmonds had not passed the polygraph test required by his plea bar*834gain. Smith amended his petition by adding two additional claims. The first claim contended that his conviction violated his constitutional rights because he was actually innocent of felony murder. The second claim, based on Brady v. Maryland, 373 U.S. 83, 83 S.Ct. 1194, 10 L.Ed.2d 215 (1963), derived from the prosecution’s failure to produce Edmonds’s polygraph test results when requested.

After the government saw Edmonds’s second affidavit, it arranged a meeting with Edmonds and his court-appointed counsel. There, the government informed Edmonds that if he persisted in testifying according to his affidavits, the state could have his plea agreement set aside. Without his plea agreement in place, Edmonds could be subject to capital murder charges for the killing of Emmett Konzelman. On the other hand, if Edmonds reaffirmed his prior testimony, which had identified Smith as the murder, the state would not pursue perjury charges against him based on the contents of his affidavits. After conferring with counsel, Edmonds decided to invoke his Fifth Amendment rights and refused to testify. In response, Smith argued that the prosecution had engaged in misconduct in order to prevent Edmonds from presenting exculpatory testimony. He sought a district court order compelling the prosecution to grant Edmonds use immunity under United States v. Westerdahl, 945 F.2d 1083 (9th Cir.1991).

The district court denied Smith’s petition on procedural grounds because he had failed to exhaust his state remedies, and his procedural default could not be excused because he had neither demonstrated actual innocence nor cause and prejudice. It rejected Smith’s claim of prosecutorial misconduct and denied his request for an evidentiary hearing. This appeal followed.

II. ANALYSIS

On post-conviction review under AED-PA, we “may not reach the merits of proeedurally defaulted claims.” Williams v. Stewart, 441 F.3d 1030, 1061 (9th Cir.2006); see also 28 U.S.C. § 2254(b) (2000). A claim is procedurally defaulted if “the petitioner failed to follow applicable state procedural rules in raising [it].” Sawyer v. Whitley, 505 U.S. 333, 338, 112 S.Ct. 2514, 120 L.Ed.2d 269 (1992). However, “[f]ederal courts retain the authority to issue the writ of habeas corpus in a further, narrow class of cases despite a petitioner’s ... procedural default.” McCleskey v. Zant, 499 U.S. 467, 494, 111 S.Ct. 1454, 113 L.Ed.2d 517 (1991). This narrow class of cases consists of those “extraordinary” cases where “a constitutional violation probably has caused the conviction of one innocent of the crime.” Id. In such a case, a “claim of innocencef ] ... is procedural, rather than substantive,” in that it allows a petitioner to overcome procedural hurdles so that a court will hear his actual substantive claim — that his constitutional rights were violated — on the merits. Schlup v. Delo, 513 U.S. 298, 314-15, 115 S.Ct. 851, 130 L.Ed.2d 808 (1995). Such a “claim of innocence is thus ‘not itself a constitutional claim, but instead a gateway through which a habeas petitioner must pass to have his otherwise barred constitutional claim considered on the merits.’ ” Id. at 315, 115 S.Ct. 851 (citing Herrera v. Collins, 506 U.S. 390, 404, 113 S.Ct. 853, 122 L.Ed.2d 203 (1993)).13

*835Smith raises various constitutional claims arising from the actions of prosecutors and his own counsel prior to his entry of his plea of no contest. However, these claims are not before us because Smith has procedurally defaulted on these claims by failing to exhaust his remedies before the Oregon courts. The only way we may consider his claims at this time is if he can prove his actual innocence. Smith must show that, considering all the evidence now available, it is more likely than not that no reasonable juror would convict him of felony murder, the crime to which he pled no contest. See House v. Bell, — U.S. -, 126 S.Ct. 2064, 165 L.Ed.2d 1 (2006).

The majority gets Smith over these formidable procedural hurdles in three steps. First, the majority finds that the prosecution engaged in misconduct by preventing Edmonds from testifying on Smith’s behalf at his post-conviction proceedings by telling Edmonds that if he recanted his prior testimony, he would be in breach of his plea bargain and he could therefore be charged anew with the murder. The majority then decides — and it is the first court to even consider such a remedy, let alone to adopt it — that the remedy for this misconduct is to presume the truthfulness of Edmond’s recantation for the purposes of Smith’s actual innocence inquiry. By thus having assumed that Edmonds likely wielded the crowbar, the majority finds that no reasonable jury would have convicted Smith and that therefore he is actually innocent of felony murder. I disagree with the majority at each step.

A. Prosecutorial Misconduct

In general, a criminal defendant is not entitled to compel the government to grant immunity to a witness. United States v. Shirley, 884 F.2d 1130, 1133 (9th Cir.1989). In Westerdahl, 945 F.2d 1083, we held that while the decision to grant or withhold immunity is generally the exclusive domain of prosecutors, there is a limited exception under which a court can compel a grant of use immunity. Westerdahl’s exception applies when “the fact-finding process is intentionally distorted by prosecutorial misconduct, and the defendant is thereby denied a fair trial.” 945 F.2d at 1086. A defendant can make a prima facie case of prosecutorial misconduct by demonstrating that “the government distorted the judicial factfinding process by denying immunity to [a] potential [defense] witness” whose testimony would have been relevant. Id. “If a defendant makes an unrebutted prima facie showing of prosecutorial misconduct that could have prevented a defense witness from giving relevant testimony, we will remand the case to the district court to determine at an evidentiary hearing whether the government intentionally distorted the fact-finding process.” United States v. Whitehead, 200 F.3d 634, 640 (9th Cir.2000) (emphasis added) (quotation omitted); see also United States v. Tam, 240 F.3d 797, 804 n. 4 (9th Cir.2001) (same); Westerdahl, 945 F.2d at 1086, 1088 (same); United States v. Lord, 711 F.2d 887, 891 (9th Cir.1983) (same). Other circuits take a similar approach. See, e.g., United States v. Tarricone, 21 F.3d 474, 476-77 (2d Cir.1993) (remanding a prosecutorial misconduct claim for an evidentiary hearing to address questions of the government’s intent and knowledge); Kirkpatrick v. Whitley, 992 F.2d 491, 497-98 (5th Cir.1993) (same); Virgin Islands v. Smith, 615 F.2d 964, 969 (3d Cir.1980) (same). If, on remand, the *836district court finds misconduct, then it can force the state to grant the defense witness use immunity.14

There is also a second line of cases which the majority opinion relies on. These cases stem from the Supreme Court’s decision in Webb v. Texas, 409 U.S. 95, 93 S.Ct. 351, 34 L.Ed.2d 330 (1972), where the trial judge “gratuitously singled out ... one witness for a lengthy admonition on the dangers of perjury,” thereby implying “that he expected [the witness] to lie.” Id. at 97, 93 S.Ct. 351. The trial judge then proceeded “to assure him that if he lied, he would be prosecuted and probably convicted for perjury, that the sentence for that conviction would be added on to his present sentence, and that the result would be to impair his chances for parole.” Id. The witness subsequently refused to testify; the Court stated that the judge “effectively drove that witness off the stand.” Id. at 98, 93 S.Ct. 351. The Court held that, in doing so, the trial judge had denied the defendant the right to present his own witnesses, thus violating his Due Process rights under the Fourteenth Amendment. Id.

The majority relies heavily on a number of cases following Webb, such as Earp v. Ornoski, 431 F.3d 1158 (9th Cir.2005), and, in particular, United States v. Vavages, 151 F.3d 1185 (9th Cir.1998).15 These cases hold that “substantial government interference with a defense witness’s free and unhampered choice to testify amounts to a violation of due process.” Vavages, 151 F.3d at 1188 (quotation omitted). If a defendant can establish both that the prosecution committed such a violation and that he suffered prejudice as a result, he is entitled to have his conviction vacated and to be retried. See, e.g., id. at 1188-93.

But, while both the Vavages and the Westerdahl lines of cases address prosecutorial misconduct, a petitioner who seeks relief under one line of cases presents a very different argument than a petitioner who seeks relief under the other. Vavages requires “substantial government interference with a defense witness’s free and unhampered choice to testify,” 151 F.3d at 1188, but Westerdahl requires something different; it requires a defendant to demonstrate that the “the fact-finding process [was] intentionally distorted by prosecutorial misconduct,” 945 F.2d at 1086. The biggest difference between these two standards is that Westerdahl has a more stringent prosecutorial intent requirement than Vavages does. This makes sense, as Westerdahl enables a defendant to compel the prosecution to grant immunity, a remedy that Vavages and its brethren do not provide. But here, Smith has opted not to seek relief under Vavages, and has only made an argument under Westerdahl. His opening and reply briefs do not cite to Vavages or to any other cases in its line. Nor does he argue that there was “substantial government interference with a defense witness’s free and unhampered choice to testify.” Instead, he narrowly focuses his attention on an argument under Westerdahl. The majority’s attempt to conjure a Vavages claim based on the argument that the cases “address[] the same basic issue,” Maj. Op. at 824-25 n. 30, is unconvincing. An argument that is not raised by a party in its opening brief is waived; accordingly, Smith has waived any Vavages argument he might make. See, *837e.g., Smith v. Marsh, 194 F.3d 1045, 1052 (9th Cir.1999).

I do not agree with the majority’s decision to raise Vavages sua sponte, nor do I agree that this case falls within Vavages’s purview. Under Vavages, we examine the totality of the circumstances to determine whether the prosecution “substantially interfere[d]” with a witness’s right to decide whether to testify. “Among the factors courts consider ... are the manner in which the prosecutor or judge raises the issue, the language of the warnings, and the prosecutor’s ... basis in the record for believing the witness might lie.” Vavages, 151 F.3d at 1190. I address each in turn.

First, the manner in which the prosecutor raised the issue contradicts the majority’s assertion that the prosecutor was trying to interfere in Edmonds’s decision to testify or not. It was the prosecution, not the defense, who asked the court to appoint counsel for Edmonds. And the prosecution did not attempt to discuss Edmonds’s affidavits with him until after the court had provided him with counsel. If the prosecutor was driven by a desire to scare Edmonds into silence, it would have been much easier to do so if he was not represented by counsel. The only reason that the prosecutor would have wanted Edmonds to have legal counsel is if he wanted to ensure that Edmonds was fully informed of the consequences of his actions so that he could be prosecuted if he was, in fact, the murderer.16 As in Vavages, “there is no question that the prosecutor was justified in ... cautioning [Edmonds’s counsel] against his client’s testifying falsely, and informing him of the possible consequences of [his] testimony.” Id. Thus, the manner in which the prosecutor raised the issue cuts against Smith. See also id. at 1191 (“The strongest factor in the government’s favor is that the prosecutor never directly admonished [the witness].”).

Second, the message itself was truthful and contained no more information than was necessary to communicate to Edmonds the implications of his change in testimony. Here, the message that the state would seek the death penalty against Edmonds if he testified that he, and not Smith, had killed Mr. Konzelman was, indisputably, a weighty statement, and if driven by an improper motive, would constitute prosecutorial misconduct of the highest order. See id. at 1190. But if Edmonds was lying, his testimony would be perjury that could potentially enable a convicted murderer to escape justice. Such a scenario would be sufficiently dire to merit such a stern warning.

Moreover, the prosecutor had very good reason to issue the warning to Edmonds. The state had previously entered into a plea agreement with Edmonds based on the premise that Smith, not Edmonds, had wielded the crowbar, and that Edmonds would so testify. If Edmonds recanted his testimony against Smith, he necessarily implicated himself and undermined the whole basis for his plea agreement. Under the circumstances — a vacillating, opportunistic witness — the state had every right, and perhaps even a duty, to advise Edmonds that his testimony imperiled his plea agreement and rendered him eligible for renewed charges. I suspect that we would not look favorably on a prosecutor who kept silent while a defendant violated his plea agreement and then declared him in breach and charged the defendant with the original offense.

*838Finally, the prosecutor had ample “basis in the record for believing [that Edmonds] might lie.” Id. “[A] direct conflict between the witnesses] proposed testimony and[his] own prior testimony” constitutes a “substantial basis in the record for believing the witness might lie” which could justify “unusually strong admonitions against perjury.” Id. Here, Edmonds was offering a version of events that completely contradicted testimony that he had given on multiple occasions, multiple statements he had given to the police, and statements that he had made to a cellmate.

But the prosecution also had another reason to think that Edmonds was lying: After Edmonds was released for his role in the Konzelman burglary and murder, he committed another crime and was sentenced to life in prison.17 Edmonds gave his affidavits while he was already in prison for this subsequent offense. It is quite plausible that Edmonds, believing that any testimony he gave would have no consequences for him, decided to lie in order to exculpate his old associate Smith.18 The prosecution had good reason to believe that this was indeed the case; the second *839affidavit was prepared and procured by Smith’s counsel in this proceeding.19 This fact, when combined with Edmonds’s testimonial reversal, reveals that the prosecution had very strong grounds for believing that Edmonds’s testimony would be a lie. On balance, then, even if we consider Smith’s claim under Viwages — and I do not think that we should do so — I do not think that Smith can meet its requirements.

The majority takes a different view, and, without even considering the fact that Smith has not raised a Vavages argument, finds that Smith has met its requirements. The majority’s efforts to supplant Smith’s actual claim is necessary to its result because Smith cannot succeed on the Westerdahl argument that he has made. It is not at all clear on this record that the prosecution distorted the factfinding process, intentionally or otherwise. Nor did the district court agree with the majority’s view that prosecutors “threatened” Edmonds. The district court found that the prosecutor “warned Edmond[s]’s counsel that the district attorney would seek the death penalty if Edmonds testified that he, and not Smith, actually killed Emmet[t] Konzelman.” While there is arguably a fine line between warning of future consequences and threatening retribution, the distinction is critical to a determination of prosecutorial intent.

The record in this case presents a number of legitimate explanations for the prosecution’s conduct. As I have discussed above, the prosecution had unusually strong reasons to believe that Edmonds, who thought that any statements he made would be without consequences, was lying. Under these circumstances, advising Edmonds that there could be real consequences to his actions was not an act taken to “intentionally distort[]” the factfinding process, but instead was merely designed to prevent Edmonds from distorting the process. More importantly, as I discussed above, the prosecutor was giving Edmonds fair warning that his changed testimony would breach his plea agreement, which entailed serious consequences. The majority’s approach seems to leave prosecutors no appropriate course of action other than rolling over and permitting defendants to violate their plea bargain agreements with impunity.

Nor do I believe that we should conclude that the prosecution’s actions were unwarranted because, as the majority claims, the prosecution had previously “determined] in Smith’s case that an appropriate punishment for the actual killer was 30 years to life.” Maj. Op. at 824. First, we should use the sentence the prosecution sought to impose on Smith, rather than the sentence he actually received after his plea bargain, as the basis of any comparison. Smith was indicted for aggravated murder, which is a capital crime in Oregon, see Or. Rev. Stat. § 163.105(l)(a) (2003), and both Smith’s own affidavit and those of his trial attorneys identify his eligibility for the death penalty as a primary reason for his plea bargain. For that reason, I see no significance in the majority’s comparison. Second, unlike the majority, I cannot say that a prosecutor’s decision to vigorously pursue charges against a person who claims to have committed murder must necessarily be motivated by an improper desire to prevent that person from testifying. I think that it is at least equally plausible that such decisions are motivated *840by an entirely proper impulse — a desire to bring that person to justice for his crimes.20

I am not suggesting that we should eviscerate Westerdahl by conclusively presuming that prosecutors always act with only the noblest of intentions when they make immunity decisions. Government malfeasance threatens the very legitimacy of our criminal justice system. Therefore, prosecutorial misconduct is an extremely serious charge that must be investigated carefully. But the question of the prosecution’s intent is a factual one, and must be resolved in each case by thoughtful and thorough consideration of the evidence. That task, we have sensibly held, should be undertaken by the district court. In Westerdahl cases, the defendant and the prosecution will often assert conflicting and contradictory versions of what happened and why. An appellate court, examining a cold and undeveloped record, is not the ideal institution for resolving such questions. Our precedent recognizes that an evidentiary hearing in the district court is a far superi- or mechanism. It is telling that in each of our Westerdahl cases, we have remanded to the district court for an evidentiary hearing. See, e.g., United States v. Young, 86 F.3d 944, 949 (9th Cir.1996) (remanding for an evidentiary hearing to determine whether the prosecutor withheld immunity from a defense witness in order to intentionally distort the factfinding process); Westerdahl, 945 F.2d at 1086 (same); Lord, 711 F.2d at 892-93 (same). The majority short-circuits this mechanism by avoiding the analysis required by a Westerdahl claim, and instead analyzing the case for a Vavages violation, which does not require the same showing of prosecutorial intent and the accompanying remand for an evidentiary hearing.

Thus, my problem with the majority’s conclusion goes far beyond the fact that it finds prosecutorial misconduct where I would find none. Far more important, the majority’s conclusion is premature. Instead of sifting through conflicting assertions with the aid of an undeveloped record, the majority should have followed our Westerdahl precedents and remanded this case to the district court for an evidentiary hearing on this question. The majority should not have deviated from our precedent by co-opting the Vavages line of cases and answering this question itself.

B. Remedy

Even if we could conclusively determine that there was prosecutorial misconduct here, I also take issue with the majority’s conclusion that the standard Westerdahl remedy for prosecutorial misconduct — ordering the prosecution to grant a defense witness use immunity — would be insufficient in this case. Maj. Op. at 825-27. The majority reasons that if Smith’s conviction is overturned as a result of Edmonds’s testimony, the prosecution could then charge Edmonds with murder and seek the death penalty. Id. Knowing this, the majority reasons, it is unreasonable to expect Edmonds to testify truthfully even if he is granted use immunity.21 Id. The *841majority declares that the issue of a remedy for Smith’s case presents a “novel procedural posture.” Maj. Op. at 826. However, this case is not novel. It presents a standard claim of prosecutorial misconduct under Westerdahl for which the remedy is clear, and clearly tied to the violation found. It is wholly inappropriate to treat this issue as open to fashioning new remedies when we are clearly bound by case law. However, this is what the majority does. Maj. Op. at 825-27. And after evaluating a series of proffered remedies,22 the majority concludes that the appropriate response is to presume the truth of Edmonds’s affidavits for purposes of the Schlup inquiry. Id. at 827-28.

If ever there were affidavits that were entitled to a presumption of truthfulness, these surely are not they.23 Edmonds has changed his testimony numerous times; at one point he even professed a version in which Smith’s girlfriend was their getaway driver.24 Cf. Schlup, 513 U.S. at 324, 115 S.Ct. 851 (“To be credible, [an actual innocence] claim ... [must be supported by] new reliable evidence — whether it be exculpatory scientific evidence, trustworthy eyewitness accounts, or critical physical evidence — that was not presented at trial.” (emphasis added)). Similarly, the reason that we typically afford credibility to confessions — i.e., that because such statements carry criminal consequences, one ordinarily has no incentive to falsely confess one’s guilt — does not apply here. See Fed. R. Evid. 804(b)(3). Edmonds, already serving a life prison sentence, and free from a charge of perjury for his prior testimony, see footnote 19, supra, might well have thought that any statement he might make carried no consequences for him. Indeed, once he found out that real consequences might attach to his testimony, he refused to testify.

The majority, undaunted by Edmonds’s fickle history and the risk-free circumstances under which the affidavits were made, sees fit to conclude that Edmonds has finally told the truth. The majority’s willingness to accept Edmonds’s latest version of events is remarkable. The ma*842jority’s treatment of Edmonds’s past statements is similarly troubling. When Edmonds’s testimony supports the majority’s views, it treats Edmonds as the embodiment of truth itself. Whenever his testimony does not mesh with the majority’s view of events, he is a pathological liar. For example, when Edmonds confessed his crime to a cellmate, “such statements by Edmonds lack credibility.”25 Maj. Op. at 845. Similarly, the majority finds that Edmonds’s previous claims that his code name was “Low,” “should be viewed with skepticism.” Id. at 817.26 On the other hand, Edmonds is to be fully credited when he “stated that prior to entering the house, there had never been any discussion between him and Smith about hitting anyone or using any kind of weapon on anyone,” id. at 820, never mind that “such statements by Edmonds [might] lack credibility,” id. at 815, or that Edmonds has given two other accounts of the evening where he and Smith discussed the possibility of using the crowbar against the occupants. The majority also credits Edmonds’s description of the hat Smith wore, Maj. Op. at 816, as well as his statements that he and Smith were in different parts of the dark house for much of the burglary, id. at 819-20 n. 20, even while concluding that “his initial testimony is necessarily false ... and lacks any force in countering the other available evidence,” id. at 817.27

But even if one were to ignore the affidavits’ troubling back story, they are both internally inconsistent and plainly inconsistent with each other. His first affidavit, from February 1996, states both that “Roger Smith did not kill Emmit Konzleman [sic]” and that Edmonds testified falsely out of his fear that “Roger Smith was about to testify and leave [Edmonds] to do a life sentence for a crime[Edmonds] did not commit.”28 Yet both Edmonds and Smith agree that they and the Konzelmans were the only people to enter the Konzelman house that night, and it is clear that the Konzelmans did not attack themselves. There is no way to reconcile this affidavit — short as it is29 — with reality.30

*843Smith’s 2001 affidavit is a slight improvement over its predecessor, in that it is slightly longer and better organized and that it does not blatantly contradict itself. It explains his perjury as simply motivated by a desire to avoid prosecution for murder. While this differs markedly from what he said in his first affidavit,31 it is certainly a more plausible reason. Nonetheless, this affidavit remains entirely undeserving of the weight the majority accords it. Entirely apart from the serious question of Edmonds’s honesty, there are also real issues as to his basis for making some of the statements in the affidavits, which consist entirely of bare, unsupported assertions.

Edmonds’s unexplained statement in his recent affidavits “that Mr. Smith never entered the Konzleman’sfsic] bedroom,” which the majority credits, see Maj. Op. at 820, provides an excellent example of both problems. Mrs. Konzelman’s testimony clearly refutes the affidavit in this respect: She testified that both burglars entered the bedroom for a short period of time.32 At the same time, it is somehow undisputed that the two burglars were in the Konzelman residence for approximately forty minutes, and the majority states both that Smith and Edmonds did not “ke[ep] track of one another’s whereabouts while in the house,” id. at 819-20 n. 20, and that “the two men were in different parts of the house for most of the time,” id. at 819. Thus, it seems that not only did Edmonds lie in his affidavits, but that he lacked a basis for even making many of the statements contained therein.

Yet all of these issues- — -the internal inconsistencies, the unsupported assertions, and the contradictions with other facts in the record — could be fully addressed if Edmonds testified at an evidentiary hearing. Indeed, if Edmonds were given use immunity — like that which we have granted every other witness under the Westerdahl exception — there is good reason to believe that this is precisely what would happen. Further, Smith has not even asked for the extraordinary remedy that the majority confers.33 I cannot agree with the majority’s decision to jettison an entire line of precedent by creating a new *844remedy that is entirely alien to our case law and that does not suit the facts of this case.

C. Actual Innocence and the Schlup Gateway

Even indulging the majority’s presumption of truthfulness, Smith is not actually innocent. Not even close. As a legal matter, Edmonds’s affidavits, even if believed, completely fail to exculpate Smith of felony murder. In order to pass through Schlup’s “actual innocence” gateway, the “petitioner [must] show that ‘a constitutional violation has probably resulted in the conviction of one who is actually innocent.’ ” Schlup, 513 U.S. at 327, 115 S.Ct. 851 (quoting Murray v. Carrier, 477 U.S. 478, 496, 106 S.Ct. 2639, 91 L.Ed.2d 397 (1986)). “To establish the requisite probability, the petitioner must show that it is more likely than not that no reasonable juror would have convicted him in the light of the new evidence.” Id. In determining whether the petitioner has carried his burden, courts consider all of the evidence, not just that which was presented at trial. Id. at 327-28. In doing so, however, courts must also consider the reliability of that evidence and augment or discount its probative value as may be appropriate. Id.34

In this case, the inquiry is more complicated than it would be ordinarily. Smith pled no contest to felony murder, because Emmett Konzelman was murdered in the course of Smith and Edmonds’s burglarizing the Konzelmans. Smith does not deny his involvement with the burglary. Instead, he rests his claim to innocence on Oregon’s affirmative defense to felony murder, which has five separate criteria that a defendant must satisfy. See Or. Rev. Stat. § 163.115(3) (2003).In Oregon, defendants bear the burden of proving affirmative defenses by a preponderance of the evidence. Id. § 161.055(2). Smith must therefore prove, by a preponderance of the evidence, that no reasonable juror would have found that he failed to meet each of these five criterion by a preponderance of the evidence. Or, put another way, Smith must show that it is more likely than not that every reasonable juror would find that Smith established every element of his affirmative defense. Or, rephrased a third time, Smith must show by a preponderance of the evidence that every juror who would find that Smith did not establish his affirmative defense was unreasonable. This is an extremely high burden to overcome: If a single reasonable juror could find that the evidence was in equipoise on any individual element of Smith’s affirmative defense, we must find for the state.

The elements of Oregon’s affirmative defense are not easy for a defendant to establish. A defendant must demonstrate that he:

(a) Was not the only participant in the underlying crime;
(b) Did not commit the homicidal act or in any way solicit, request, command, importune, cause or aid in the commission thereof;
(c) Was not armed with a dangerous or deadly weapon;
(d) Had no reasonable ground to believe that any other participant was armed with a dangerous or deadly weapon; and
(e) Had no reasonable ground to believe that any other participant intended to engage in conduct likely to result in death.

*845Id. § 163.115(3). The government concedes that Smith has satisfied element (a); however, each of the other four elements is contested.

Although I disagree with the majority on nearly every question presented in this case, I think there is at least one thing that is beyond dispute — it simply is not clear exactly what happened during the forty minutes that Smith and Edmonds were inside the Konzelman residence. Smith has never given a full account of what happened in this period, and Edmonds has given numerous contradicting accounts, each of which is marred by various biases. Elma Konzelman, the sole surviving, unbiased witness, was only conscious for a relatively small portion of the burglary, and her recollections are understandably somewhat hazy.

When one combines the vague and incomplete state of the evidence with the burden of proof imposed by Oregon’s affirmative defense to felony murder, which requires that Smith bear the burden of proof on multiple elements, and Schlup, which requires that a petitioner demonstrate that every juror who would find against him was unreasonable, it becomes clear that Smith cannot prevail. Thus, even if the majority was correct that the prosecution engaged in misconduct, Smith did not suffer any prejudice because, even if Edmonds were to testify consistently with his affidavits,35 there is no way that Smith can satisfy the daunting burden created by Schlup and Oregon’s expansive felony murder statute. Even if one accepts the majority’s dubious remedy and accords Edmonds’s affidavits the presumption of truth, Smith has not carried his burden of proof on elements (c), (d), and (e).36 I address each of these elements in turn.

1. Whether Smith “Was Not Armed with a Dangerous or Deadly Weapon”

The majority relies heavily on Edmonds’s recantations to conclude that Edmonds was the killer, not Smith. Based on this conclusion, the majority reasons that there is no evidence in the record to suggest that the other burglar — i.e., Smith — was armed at any point. Maj. Op. at 817-19. As I discuss below, this is simply not true. Yet even if it were true, Smith bears the burden of actually proving that he was not armed at any point; a lack of evidence is insufficient. If it was more likely than not that any reasonable juror could find that there was at least a fifty percent chance that Smith was armed at any point during the burglary, then the Schlup gateway is closed to Smith.

A reasonable juror might well have made such a finding here. A reasonable juror could have concluded that the two burglars conferred in the garage for some time on their method, and emerged from this conference with a clear consensus. *846Both wore hats and similarly knotted bandannas to disguise their physical appearance. Both wore gloves to avoid leaving fingerprints inside the house. They procured these items together, with each taking items from both Edmonds’s truck and the Konzelmans’ garage. They established paired code names, “High” and “Low,” to further shield their identities. It hardly strains credulity to conclude that both decided to carry a weapon in the event that the residents should awaken while they were in the house.37 Even if only one weapon was available (which was unlikely to be the case in a garage full of tools), the two might have debated who would carry the crowbar. If Smith picked it up first, or held it for a minute before giving it back to Edmonds, that brief possession would suffice.38

Then there is the rope. Smith’s trial attorneys both swore out affidavits stating that Smith admitted to carrying the rope into the Konzelman residence and Edmonds stated that they took it in case they should need to restrain their victims. Nothing in the record contradicts these assertions. Our legal reporters are replete with cases where criminals have used ropes as weapons. See, e.g., Schad v. Arizona, 501 U.S. 624, 627, 111 S.Ct. 2491, 115 L.Ed.2d 555 (1991); Sims v. Brown, 425 F.3d 560, 564 (9th Cir.2005); State v. Langley, 314 Or. 511, 840 P.2d 691, 695 (1992) (en banc).39 Oregon’s statutory affirmative defense to felony murder requires a defendant to establish that he was not armed with any dangerous or deadly weapon; it is insufficient to establish that he was not carrying the murder weapon. Thus, even if a reasonable juror concluded that Smith never handled the crowbar, that juror may still conclude that Smith was armed with a dangerous weapon because he was carrying the rope.

The majority argues that it does not matter that Smith carried the rope because, the majority concludes, it is not a “dangerous weapon” under Oregon law.40 Maj. Op. at 818. Oregon law defines a “dangerous weapon” as “any weapon, device, instrument, material or substance which under the circumstances in which it is used, attempted to be used or threatened to be used, is readily capable of causing death or serious physical injury.” Or. Rev. Stat. § 161.015(1) (2003); Maj. Op. at 818. According to the majority, “[tjhere is no evidence suggesting that the rope in *847this case qualified [as a dangerous or deadly weapon].” Id. at 818. This contention is absurd. Oregon courts have interpreted a “dangerous weapon” quite broadly. See, e.g., State v. Hill, 298 Or. 270, 692 P.2d 100 (1984) (en banc) (automobile); State v. Reed, 101 Or.App. 277, 790 P.2d 551 (1990) (concrete sidewalk); State v. Bell, 96 Or. App. 74, 771 P.2d 305 (1989) (cowboy boots); State v. Gale, 36 Or.App. 275, 583 P.2d 1169 (1978) (can opener); State v. Jacobs, 34 Or.App. 755, 579 P.2d 881 (1978) (hot water). When burglars attempt to tie up awakened victims in the middle of the night, the rope they use is certainly “readily capable of causing death or serious physical injury.” Cf. State v. Cornell, 314 Or. 673, 842 P.2d 394, 396, 402 (1992) (en banc) (upholding a felony murder conviction where victim died as a result of being bound by burglar). At one point, Edmonds testified that they took the rope so they could bind the occupants if they awoke. The majority ignores this evidence. But given that the victim in Cornell died when a rope was used as a restraint, a juror hearing testimony of this plan could reasonably conclude that the rope was a dangerous weapon, likely to cause death or serious physical injury if used in this way.

These are only a few of the logical paths that reasonable jurors might have taken. In this case, where so many details are unclear, it is impossible for Smith to meet the evidentiary burden that Schlup imposes. Yet the majority gives short shrift to the legal standards at issue, blithely brushing aside any concerns by relying on its questionable assumption that Edmonds was likely the killer.

2. Whether Smith “Had No Reasonable Ground to Believe That [Edmonds] Was Armed with a Dangerous or Deadly Weapon”

The majority’s analysis of whether Smith knew Edmonds was armed is no more satisfying than its determination that Smith was not armed. Based on the fact that the house was dark,41 and the fact that Edmonds’s testimony as to when he and Smith entered the house was unclear, the majority concludes that every reasonable juror would likely conclude that Smith failed to notice that his companion was toting a three-foot-long crowbar. Maj. Op. at 819-20.

I cannot agree. It would have been difficult for Edmonds to conceal a three-foot-long crowbar, assuming he had wished to, and there is nothing in the record to suggest that Edmonds even attempted such a concealment.42 As discussed previously, Smith and Edmonds were in the garage together for some time disguising themselves, assigning code names, and planning for contingencies before they entered the house. A reasonable juror could easily conclude that Smith saw the weapon then.43

*848There is also plenty of evidence in the record to support a finding that Edmonds and Smith entered the house at or around the same time. The majority gives short shrift to one of Edmonds’s accounts of how they entered the house together and ignores others completely.44 Even without this evidence, a jury could easily rely on Edmonds’s descriptions of how the two entered together after they disguised themselves together — strongly corroborated by the undisputed fact that Smith and Edmonds adopted similar disguises — and conclude that Edmonds and Smith entered together, which would have given Smith an excellent opportunity to notice the three-foot-long crowbar that the majority posits Edmonds was carrying. That conclusion, however, would be inconvenient to the majority’s case. Moreover, according to Smith’s attorneys, even though they told him what he would have to prove in order for his affirmative defense to succeed, Smith never told them that he did not know Edmonds was carrying the crowbar. Indeed, it was concern over precisely these issues that persuaded Smith’s two trial attorneys to forgo this affirmative defense and seek a plea bargain.

The majority’s inferences drawn from other “facts” fare no better. From Mrs. Konzelman’s statement that she was attacked by one person and Edmonds’s affidavit’s bald assertion that “Smith never entered the Konzleman’s[sic] bedroom,” the majority concludes that “Smith was not present when Edmonds struck Konzelman.” Maj. Op. at 820. This reasoning has several problems. First, Mrs. Konzelman’s testimony refutes Edmonds’s affidavit on this point. She testified that both burglars entered the bedroom for a short period of time. Second, nothing in Mrs. Konzelman’s statement is inconsistent with Smith being present when she was beaten.45 Even ignoring the question of who *849was actually doing the striking, the fact that there was a lone attacker does not prove that the second burglar was not present during the beatings. Although Mrs. Konzelman saw the second burglar walk away from the doorway, she was hit in the head with a crowbar directly afterward, at which point she, understandably, seems to have lost visual track of her surroundings as the first burglar continued beating Mr. Konzelman. It is entirely possible that the second burglar returned to the doorway while the beating was going on, or that he moved to a vantage point in the hallway from which he would not be visible to Mrs. Konzelman but could still see the attack.46 Moreover, the second burglar need not have been present for the attack. All that is required is that he had reasonable grounds to believe that the first burglar was armed. Yet the majority concludes otherwise, adopting its own version of the facts as incontrovertible and ignoring all contrary evidence.

I have only scratched the surface of the broad realm of possible conclusions that a reasonable juror might have drawn. A reasonable juror easily could have credited Mrs. Konzelman’s testimony and concluded that Smith saw the crowbar while he was standing in the doorway of the Konzelmans’ bedroom and Edmonds was standing by the dresser, before the beating began.

Once again, the rope provides another sticking point for Smith’s case. Smith admitted to his attorneys that he took the rope into the Konzelmans’ home. The crowbar was found in the Konzelmans’ kitchen, and the rope was found on the floor of the bedroom. It is unlikely that the burglar carrying it would have abandoned the rope in the Konzelmans’ bedroom while they were asleep and untouched. Similarly, since the crowbar was found in the kitchen, the murderer had to continue carrying it after the beating. Thus, even assuming that Edmonds carried the crowbar and Smith the rope, the ultimate placement of both objects implies that Smith and Edmonds crossed paths while Edmonds had the crowbar and Smith the rope, or that Edmonds took the rope from Smith before the beating. In either event, it is strong evidence that, at some point during the course of the burglary,47 Smith knew that Edmonds was carrying the crowbar.48

*850Even a more modest hypothetical juror would recognize the implausibilities in the majority’s argument. Smith did not need to know that Edmonds was armed. It would suffice if Smith had reasonable grounds to believe that Edmonds was armed. This was almost certainly the case here. Recall again that Edmonds and Smith were in the garage together planning for some time before they entered the house. In the course of their discussion of disguises and code names and while Edmonds was rifling through the tools,49 Edmonds may well have indicated his intent to carry a weapon inside with him, just in case. This knowledge alone would be enough to ruin Smith’s affirmative defense. Yet all of this is lost on the majority, whose opinion takes the most narrow conceivable view of what a juror might have found, completely flouting the Supreme Court’s holding in Schlup.

3. Whether Smith “Had No Reasonable Ground to Believe That [Edmonds] Intended to Engage in Conduct Likely to Result in Death”

The majority also makes short shrift of what reasonable jurors might infer about Smith’s knowledge the evening of the burglary.50 While the fact that the burglars used disguises and code names could suggest that they did not intend to kill the Konzelmans, Maj. Op. at 819, the question is not whether the burglars originally intended to kill the Konzelmans. The question is whether Smith had reasonable grounds to believe that Edmonds intended to engage in potentially deadly conduct, and there is ample evidence to support such a finding. Indeed, a reasonable juror might well have found that Smith was put on notice as he stood in the bedroom doorway and saw that a visibly drug-addled Edmonds had turned on the lights and was waking up the Konzelmans. Even if one were to accept the majority’s highly contrived reasoning and conclude that Smith could not see any weapons as he looked into the room, the fact that Edmonds was intentionally arousing the house’s occupants was surely enough to put Smith on notice that extremely dangerous conduct — either outright violence from resisting victims, or the imposition of some form of dangerous restraint of the victims to keep them from seeking help — was likely to be forthcoming.51

*851And yet again, our reasonable juror need not even go this far. As discussed earlier, a reasonable juror would very likely conclude that Smith brought the rope into the house with the contemplated purpose of using it to restrain the victims. Burglaries often involve potentially perilous circumstances.52 Tying victims up is highly dangerous conduct that is likely to result in death — particularly in cases such as this one, where the victims are very elderly and are awakened unexpectedly by burglars in the middle of the night.

Indeed, there is Oregon case law directly on point. In State v. Cornell, 314 Or. 673, 842 P.2d 394, 396, 402 (1992) (en banc), a burglar tied up his victim and left; as a result, the victim died of asphyxiation. The defendant was convicted of felony murder, and the Oregon Supreme Court upheld the conviction.53 Many other jurisdictions have upheld convictions under similar fact patterns. See also, e.g., Hill v. Norris, 96 F.3d 1085 (8th Cir.1996) (same); Cowan v. Artuz, 96 F.Supp.2d 298 (S.D.N.Y.2000) (upholding felony murder conviction after medical examiner testified that cause of death was “heart attack possibly caused by stress from being struck ... and bound”); United States ex rel. Brodie v. Hilton, 496 F.Supp. 619 (D.N.J. 1980) (upholding the felony murder conviction of a burglar whose victim was asphyxiated as a result of being bound); Cannistraci v. Smith, 470 F.Supp. 586 (S.D.N.Y.1979) (upholding a reckless murder conviction where victim was asphyxiated after being bound); State v. Molitor, 729 S.W.2d 551 (Mo.Ct.App.1987) (upholding felony murder conviction after victim was asphyxiated as a result of being bound). The majority suggests that a rope may not qualify as a deadly or dangerous weapon. While that may be true generally, the rope in this case does qualify. Edmonds and Smith were prepared to use the rope as a restraint. Given that the victim in Cornell died when so restrained, the inherent danger in such use of a rope is clear.

Summarizing all of the factual and legal conclusions that the majority would require every reasonable juror to draw demonstrates just how unreasonable their *852construction of the evidence is. Every reasonable juror would have to conclude that the night’s events proceeded as follows: Smith and Edmonds, after deciding to burglarize the house to look for Mr. Konzelman’s wallet, returned to the garage together. They then disguised themselves and adopted code names. However, contrary to Edmonds’s prior testimony, they never discussed carrying a weapon in case the residents — whom they knew to be home — awakened. The two then entered the house, with Edmonds carrying a three-foot-long crowbar that Smith never touched. In fact, Smith not only didn’t touch it, he didn’t even have any reason to suspect that Edmonds was carrying a three-foot-long crowbar that Edmonds made no effort to conceal. Smith himself carried a rope he planned to use to tie up the victims if they awoke — an idea that, contrary to Edmonds’s testimony, he must have gotten on his own, since he and Edmonds never even contemplated what they would do if their victims woke up^ — but tying up resisting octogenarian victims who had been rudely awakened in the middle of the night and then leaving them there was not conduct likely to result in death. Smith and Edmonds then spent forty minutes burglarizing the Konzelmans’ residence, but even though Edmonds wasn’t keeping track of where Smith was or what he was doing during this entire time, Smith never had reason to believe that Edmonds was carrying a three-foot-long crowbar, or that he was contemplating any conduct that was likely to result in death. Smith then walked by the Konzelmans’ bedroom, where he saw that Edmonds, who had turned the bedroom light on, was standing at the foot of the Konzelmans’ bed and was waking them up. But this did not put Smith on notice that anything was amiss because Edmonds must have placed the crowbar on the ground at this point, so Smith didn’t see it, and he could not reasonably have foreseen any potential problems from the visibly drug-addled Edmonds waking the visibly geriatric Konzelmans in the course of the burglary. In fact, Smith felt so at ease that he decided to abandon the rope he was carrying right then and there, dropping it into the room without ever noticing the crowbar that was only a few feet away from it. Smith, confident that Edmonds and the Konzelmans would have a peaceful and harmonious discourse, then proceeded down the hall to resume burglarizing. He remained completely and blissfully unaware of what transpired in the bedroom until Edmonds had him pull the kitchen phone out of the wall (Edmonds himself was, at the same time, pulling the bedroom phone out of the wall). Smith then left the house without ever seeing the crowbar, which Edmonds must have tossed in the kitchen as he ran out of the house. Edmonds ran faster than Smith, though, since both Edmonds and Smith were able to make it back to Edmonds’s truck at the same time. Smith was as surprised as everyone else to learn that Mr. Konzelman had been murdered and Mrs. Konzelman had been assaulted.

Unfortunately for the people of Oregon, and fortunately for Smith, the majority does not even consider the inferences that a reasonable juror would likely draw. Instead, the majority engages in one-sided advocacy, cabining its analysis to narrowly limit the universe of findings that a reasonable juror might make. In doing so, the majority infers elaborate conclusions from the tiniest scraps of evidence, building narrow platforms that it leaps between in a complex game of judicial hopscotch. It is difficult enough to trace their path; I cannot join them in it.

*853III. CONCLUSION

I vigorously disagree with the majority. I think that the majority finds a prima facie case of prosecutorial misconduct where there was none; I think that the majority should have addressed the prima facie case it concocted by remanding to the district court for an evidentiary hearing; and I think that the majority should not have deviated from our Westerdahl precedent by presuming Edmonds’s internally and mutually conflicting affidavits to be true instead of remanding for an evidentiary hearing with a grant of immunity. But I need not be right on every single one of these points to render my position correct and the majority’s erroneous. If I am correct on any single one of these issues, the majority has erred in its disposition of this case.

But most egregiously, even if we indulge in all of the majority’s legal innovations, the majority cannot escape the facts of this case. Smith is guilty of felony murder. He has no affirmative defense. The Schlup gateway is closed to him, and no amount of prying and tugging — not even with a crowbar and a rope — can get it open.

I respectfully dissent.

. There is some ambiguity in the record as to whether Mr. Bouse's first name is “Arlen” or "Marlin.” The district and magistrate judges *830used the former, and the majority and the state court sentencing transcript use the latter. On this question, at least, I join the majority.

.Mrs. Konzelman believes that she was attacked because she protested when her husband was beaten. However, she stated that she based this opinion on speculation and that she didn't actually remember why the burglar decided to attack her. It is therefore entirely possible that she was attacked for a different reason, or — considering the methamphetamine induced state that both burglars were in at the time — that she was attacked for no reason at all.

. Contrary to the majority’s discussion of her observations, Maj. Op. at 816, Elma Konzelman would not have “fail[ed] to notice” what she could “see plainly.' ”

. The majority argues that “simply because black jeans were recovered from Smith's house and Mrs. Konzelman stated that the killer wore dark or black pants, it does not follow that Smith wore black jeans on the night of the burglary.” Maj. Op. at 816. The fact that officers found pants matching Mrs. Konzelmans description of what the killer was wearing in Smith's possession is undeniably inculpatory, particularly combined with the other evidence in tire record that Smith *831wore jeans that night. Obviously, the fact that a defendant owns black pants is of limited probative value in and of itself, but just as a brick is not a wall, the majority’s assertion that "it does not follow that Smith wore black jeans on the night of the burglary” wholly misses the point.

. The tendency to fidget and the inability to sit still is a common symptom among methamphetamine users and addicts.

. Her early statements to police suggested that this might have been a truncated version of the word “hello,” but the record suggests that the two burglars referred to each other by the code names “High” and "Low” during the robbery.

. Not to mention the fact that she was seventy-four, she had just been awakened unexpectedly at four in the morning to watch her husband get beaten in the head with a crowbar, and she herself had just been hit in the head with that same crowbar.

. Nor, under the protections of the Fifth Amendment, is he in any way obligated to do so.

. I find that this assertion by the defense actually inculpates Smith. Mrs. Konzelman testified that she did not see gloves on the killer’s hands. Her testimony suggests that Smith wore only one brown glove, leaving Edmonds as the wearer of the latex gloves. This evidence is made more incriminating by the fact that the work glove had a small blood stain on its palm. Even though the stain was too small to permit blood typing, the presence of blood suggests that it may have been worn by the killer. The matching glove was found in the Konzelmans’ garage. In light of the attention that the majority gives to the rest of Smith and Edmonds’s attire, Maj. Op. at 816-17, I find it strange that they ignore this piece of evidence.

. In Oregon, felony murder and murder are codified under the same statute, Or. Rev. Stat. § 163.115 (2003), but felony murder was clearly the theory under which Smith was charged and prosecuted in this case.

. The majority acknowledges that there is evidence corroborating Edmonds's initial accounts of the burglaiy, but finds that the corroborated elements "are irrelevant to the question [of] whether he or Smith committed the murder.” Maj. Op. at 815. This is not true. Edmonds stated that the Konzelmans woke up because Smith turned on the bedroom lights, which conforms with Mrs. Konzelman’s version of events. Edmonds testified that he was standing in the bedroom doorway behind Smith when Smith stepped into the room and turned on the lights. This positioning matches Mrs. Konzelman's testimony that before the beating she saw one burglar in her room by the dresser and another in the doorway. Edmonds testified that Mr. Konzelman started to rouse, and Smith told him to "Lay back down, old man,” before making a threatening gesture with the crowbar, at which point Edmonds left. This sequence of events synchronizes with Mrs. Konzelman’s description almost exactly, including that the second burglar was only in the doorway for a short period of time before walking away. In Mrs. Konzelman's testimony, the only other evidence that has been presented *833about the murder, there is actually significant corroboration for Edmonds's testimony.

. When officers informed Edmonds that his ticket stub had been found at the scene of the homicide, he put his head down on the interrogation table and started to cry.

. The majority states that such a claim would not be subject to procedural default. See Maj. Op. at 810 n. 4. But it is not at all clear from Herrera that this is the case. The Herrera Court only assumed that establishing substantive actual innocence in a capital case "would ... warrant federal habeas relief if there were no state avenue open to process such a claim." Heirera, 506 U.S. at 417, 113 S.Ct. 853 (emphasis added). Thus, so long as a *835habeas petitioner has a state forum in which he could have his actual innocence claim heard, but eschews it in favor of federal review, we would still be barred from adjudicating it on the merits because of the petitioner’s procedural default.

. Technically, in our Westerdahl cases, the state is not compelled to offer use immunity. However, if the state declines to do so, the court enters a judgment of acquittal for the defendant.

. As these cases make clear, Webb's rule has been extended to apply to prosecutors as well as judges.

. The majority notes that appointed counsel was likely to advise Edmonds of the risks of testifying. While the majority's cynical reading that appointed counsel was "a necessary preliminary step to Edmonds's execution,” Maj. Op. at 823-24 n. 28, one cannot seriously question the benefits of acting with counseled advice.

. Although I rely on this evidence to evaluate the lessened deterrent force of potential punishment, I do not use the conviction, as the majority does, as propensity evidence to suggest that it indicates Edmonds was more likely to have been the Konzelmans' killer. See Maj. Op. at 814 n.8. The majority misreads Schlup as definitively allowing this inadmissible evidence.

It is a well-established proposition that Fed. R. Evid 404(b) strictly forbids propensity evidence as improper because its prejudicial effect out-weighs its probative value. See Old Chief v. United States, 519 U.S. 172, 180-82, 117 S.Ct. 644, 136 L.Ed.2d 574 (1997). This evidence and the majority's unqualified inference is no more acceptable under the Schlup inquiry than it would be in court. The majority interprets Schlup as providing that "[ujnder AEDPA, we may consider such evidence on habeas review without regard to the rules of admissibility that would govern at a trial.” Maj. Op. at 814 n. 8 (referring to Schlup, 513 U.S. at 327-28, 115 S.Ct. 851). It is true that Schlup notes that a reviewing court "is not bound by the rules of admissibility that would govern at trial.” 513 U.S. at 327, 115 S.Ct. 851. That is, a court may consider "relevant evidence that was either excluded or unavailable at trial.” 513 U.S. at 327-28, 115 S.Ct. 851. Such evidence, the Court explains, includes "that alleged to have been illegally admitted (but with due regard to any unreliability of it) and evidence tenably claimed to have been wrongly excluded or to have become available only after the trial.” Id. (citation omitted); see also House v. Bell,-U.S. -,-, 126 S.Ct. 2064, 2077, 165 L.Ed.2d 1 (2006) (characterizing Schlup as allowing for review of "old and new, incriminating and exculpatory” evidence).

But it is not clear that Schlup allows for consideration of all inadmissible evidence. Schlup and its progeny emphasize that a reviewing court need not make reliability or credibility determinations in assessing whether a petitioner passes through the Schlup actual innocence gateway. See Schlup, 513 U.S. at 331, 115 S.Ct. 851 (noting that the new evidence was potentially unreliable). Instead, reviewing courts may assume the truthfulness of new evidence in assessing whether no juror would find a petitioner guilty beyond a reasonable doubt. See id. (assuming statements to be true and then evaluating whether a juror hearing such evidence would vote to convict).

At no point has the Supreme Court provided for evaluation of flatly and notoriously inadmissible evidence like the propensity evidence discussed by the majority. While Schlup frees a reviewing court from admissibility considerations such as assessing reliability and weighing credibility, Schlup does not permit us to consider evidence that is inadmissible independent of such considerations.

. By the same token, Edmonds may have reasoned that, since he would be spending the rest of his life behind bars, he ought to acknowledge his true role in the murder of Emmett Konzelman and thereby exonerate the wrongfully convicted Smith. Indeed, this is Smith’s theory. But the alternate posited motivation is at least as plausible, and the prosecutor cannot be faulted for refusing to accept Edmonds’s questionable new testimony at face value.

. A perjury charge for Edmonds's testimony at Smith’s sentencing was barred by the statute of limitations. The record indicates that the prosecution was concerned that Smith's defense attorney might have told Edmonds this, but neglected to tell him that his plea agreement could be set aside and that he could be prosecuted for capital murder.

. I note that, under Oregon law, there is no statute of limitations for either murder or manslaughter. Or. Rev. Stat. § 131.125(1) (2003).

. While the death penalty is of course the most severe punishment a defendant can receive, nothing in the majority’s logic is limited to cases involving capital punishment. The spectre of a life sentence is still a highly coercive punishment that could have nearly as strong an effect on a defendant's behavior, and the same logic would apply to any long prison sentence. Indeed, the defense witness who was granted immunity in 'Westerdahl faced a conviction for robbery which carried a potential sentence of at least twenty years. See Westerdahl, 945 F.2d at 1085; United *841States v. Westerdahl, 727 F.Supp. 1364, 1365 (D.Or.1989).

. Although the majority settles on what it terms "the least intrusive” of the three remedies it invents, Maj. Op. at 827, one cannot ignore the utter obtrusiveness of all of the alternatives. Indeed, the majority’s analysis is like comparing a wrecking ball to a bulldozer and settling on the latter to avoid some of the dust.

. The majority does limit its opinion by only adopting the presumption of truthfulness for the purposes of the Schlup actual innocence gateway. See, e.g., Maj. Op. at 807 ("We deem the exculpatory statements to be truthful ... only for the purpose of determining whether Smith’s contentions are sufficient to excuse any procedural default that may have occurred.”); id. at 814 ("[W]e presume Edmonds’s affidavits to be true for purposes of determining whether Smith is procedurally barred from prosecuting in federal court.”); id. at 827 (adopting the remedy that "presum[es] Edmonds's recantations to be true for the limited purpose of determining whether Smith may pass through the Schlup gateway”); id. at 828 (”[W]e ... hold[] merely that Edmonds's affidavits must be presumed to be true for purposes of the Schlup inquiry.”). Yet the fact that the majority’s rule is limited to the Schlup inquiry is not particularly reassuring. Prisoners in our circuit file numerous claims under Schlup each year. Many prisoners will likely try to contort the facts of their cases so as to fit the majority's narrow, but wholly unwarranted, expansion of our case law.

.She was originally indicted as a result. Ultimately, she was not prosecuted after it was discovered that she had not been in the truck and was not involved in the Konzelman burglary. The government did pursue charges against her for hindering prosecution, but these were dropped as part of Smith's plea agreement.

. The majority's full statement begins “Particularly in light of his recantations, such statements by Edmonds lack credibility.” Id. at 815. Of course, this begs the question of which of his statements, if any, is true. A jury might just as easily conclude that "Particularly in light of such statements by Edmonds, his recantations lack credibility.”

. The majority also concludes that, because of Edmonds's subsequent inability to remember the specifics of the code, the "evidence regarding the use of code names appears inconclusive.” Id. at 817. But elsewhere, the majority concludes that the burglars did use code names and uses that fact to infer that they did not intend to kill the Konzelmans at the outset. Id. at 819.

. Contrary to the majority's claims, I do not argue that "we must either afford full credibility to [Edmonds's] testimony or none at all,” and I do not dispute that "any factfinder would ... choose the elements of his story that it found credible and those that it did not.” Maj. Op. at 817 n. 16. But the majority resolves every ambiguity in Edmonds’s testimony by concluding that a jury would choose the reading that most favors Smith. This is, at best, a highly dubious practice.

. The majority suggests that “this confusing statement could have resulted either from Edmonds's characterization of his defense attorney's advice or the fact that he swore out his first affidavit without the benefit of an attorney’s supervision to help him clarify his statements.” Maj. Op. at 814-15 n. 9. It strains credulity to suggest that Edmonds thought that he was innocent of murdering the Konzelmans because of "his defense attorney’s advice.” The suggestion that Edmonds thought that he was innocent because he wrote the affidavit "without the benefit of an attorney's supervision” is equally absurd.

. Edmonds's 1996 affidavit is only a few lines long, and essentially says only three things. It contains the two, conflicting statements previously discussed, and a third state*843ment that further impeaches Edmonds's credibility. This statement is the assertion that Edmonds committed perjury "with full knowledge of the consequences.” Thus, when the majority refers to the "unequivocal nature of every other statement relating to guilt in the affidavit,” Maj. Op. at 814-15 n. 9, it is relying on a remarkably small amount of text.

. As a matter of formal logic, I must admit to being somewhat baffled by how the majority manages to accord a presumption of truth to two conflicting statements.

. In the 1996 affidavit, he states that his perjury was motivated by a fear that he would be unjustly convicted of a crime that he did not commit, while the 2001 affidavit identifies a more mercenary desire to limit the term of his incarceration.

. It is possible to reconcile these two statements if Mrs. Konzelman considered the burglar standing in her bedroom doorway to be standing in her bedroom and Edmonds considered this to be standing outside the bedroom. This is a highly charitable reading of Edmonds's testimony, however. Moreover, as discussed later in my dissent, Smith has admitted to carrying a rope into the house from the Konzelmans' garage. This rope was found by police on the floor of the bedroom. This provides a strong independent basis, apart from Mrs. Konzelman's testimony, from which to conclude that Smith entered the Konzelman bedroom.

. The majority argues that its remedy is not extraordinary, Maj. Op. at 828, but I would say that according a presumption which we have never before accorded — in any context-— to an area where our cases have clearly established that a particular remedy should be applied is, by definition, extraordinary.

. As the majority notes, see Maj. Op. at 813, the Supreme Court's recent decision in House v. Bell, — U.S. -, 126 S.Ct. 2064, 165 L.Ed.2d 1 (2006), establishes that the Schlup actual innocence gateway survives AEDPA.

. Referring to the second affidavit, the majority states that “Edmonds's declaration even more strongly inculpated himself and exculpated Smith as the actual killer.” Maj. Op. at 809. But because Smith was convicted of felony murder it makes no difference whether or not he actually killed Mrs. Konzelman. The affidavit says nothing about the three contested elements of Oregon’s felony murder affirmative defense, and if he cannot establish any one of those elements, Smith’s claim of actual innocence fails.

. If we do not presume Edmonds’s affidavits to be true, it is even clearer that these elements have not been met. For example, a juror might find that Edmonds’s story has changed so drastically in each retelling that nothing he said was credible. That juror could rely on Elma Konzelman’s testimony to infer that the two burglars were near each other in the lighted bedroom and that, even if Edmonds had committed the murder, Smith could not establish that he had not seen Edmonds carrying the crowbar.

. The police found knives next to the single brown bloodstained work glove outside the Konzelmans' residence. At least one of those knives was established to have come from the Konzelmans’ garage. As Smith argues on appeal that he wore a single brown work glove, a reasonable juror could easily find that Smith carried a knife during the burglary.

. It is also worth noting that, even after their evidentiary value has been unduly magnified by the presumption of truth that the majority grants them, Edmonds's affidavits provide no direct insight as to whether Smith was armed at any point.

. I note also that this perception is hardly limited to the legal world. For example, the popular board game Clue features a rope as one of the possible murder weapons. The others are the revolver, the lead pipe, the wrench, the candlestick, and the knife. See also ROPE (Alfred Hitchcock 1948).

.The majority argues that I have "creativefiy]” “concocted [the rope argument] out of whole hemp,” and that the state does not raise this argument. Maj. Op. at 818. Since Smith bears the burden of proving that he was not armed, it seems entirely appropriate to rely on the undisputed fact that he carried the rope in assessing whether he has met that burden, regardless of whether the state has raised this issue. As for creativity, my own imagination pales in comparison to the majority's — its opinion draws sweeping, astounding conclusions from the tiniest scraps of evidence and applies an unrequested, never-before-seen remedy.

. The majority neglects to mention that the burglars had no visibility problems while they were picking through the Konzelmans' garage. The record suggests that this was because they each carried and used a light source: Edmonds and Bouse each had a lighter and Smith had a flashlight.

. Edmonds's affidavits do not say anything about Smith's knowledge of the presence of the crowbar.

.The majority attempts to downplay this possibility, emphasizing that the burglary was "haphazard and unplanned” because "the burglars ran in and out of the house due to unexpected noises and left several items behind, including Mr. Konzelman's money and wallet.” Maj. Op. at 819. It is undisputed that a loud noise initially drove the burglars out of the garage. Indeed, it was after they fled that they decided to return and burglarize the interior of the house instead of merely the garage. At that point, they returned to the *848garage, disguised themselves, established code names and obtained a weapon. Thus, the fact that they initially fled only serves to undermine the majority's position instead of supporting it. As for Mr. Konzelman's money and wallet, the record does not suggest that the burglars abandoned it in their haste to flee, but that they never found it. In any event, their treatment of the wallet sheds little light either on the discussion the two burglars had in the Konzelmans' garage or on the question of how likely Smith would have been to see a crowbar in Edmonds’s hand.

. There is plenty of evidence from which a jury could conclude that the burglars entered the house together. Consider, for example, this excerpt from a statement that Edmonds gave to police:

[A]t that point we go in there and uh, we’re standing by the garage and I figure, before we get [the wallet from inside the house], just in case they wake up, and we have to run, that we mask ourselves, uh, put hats on and I noticed that there was two hats hanging on hooks in the garage so, we grabbed those and we put them on and we masked ourselves with bandanas and uh, we went in the house....

Or consider this transcript of police questioning Edmonds:

Police: Okay, alright, and then you entered the house?
Edmonds: Yes.
P: You’re in the living room?
E: Yes.
P: Is Roger [Smith] with you then?
E: Uh, yeah.

Or consider Edmonds's testimony at Smith's sentencing proceedings:

Prosecutor: Who went into the house first?
Edmonds: I did.
P: Did you see Mr. Smith come into the house?
E: Yeah.
P: How long had you been in the house when Mr.
Smith came inside the house?
E: Ah, I don't know. He came in probably right behind me.

. Yet, when it fulfills other needs, the majority finds that "the state could subpoena [Smith] and compel him to testify, truthfully, that Edmonds ... was the killer.” Id. at 826 n. 33. The majority does not explain how *849Smith could not have been present, but could "testify!] truthfully” that Edmonds killed Mr. Konzelman. If the majority merely means that Smith can testify that he did not kill the Konzelmans, it is not at all clear why the jury would credit his view over contrary testimony from Edmonds. Of course, if Smith had seen Edmonds kill Konzelman, he would have to have seen the crowbar, which would completely foreclose his affirmative defense to felony murder.

. This also casts doubt on the majority's suggestion that "the details about the actual murder that Edmonds was able to provide could best be explained by the fact that he was the attacker.” Maj. Op. at 815 n. 11.

. The Oregon statute does not specify when during the course of the crime a defendant must be aware that his accomplices are armed. There is good reason to conclude that a defendant who persists in the felony after discovering that his co-felon is armed is precluded from succeeding under Oregon's affirmative defense, even if the murder has already been committed at that point. The same underlying inferences regarding a defendant’s mental culpability that animate this element of the affirmative defense — that the defendant persists in committing a felony despite the fact that the presence of an armed criminal greatly increases the risks that victims face — applies equally well before and after the beating in this case, since a second victim (Mrs. Konzelman) was still alive and vulnerable. Mrs. Konzelman also testified that the burglars continued to ransack her home after the attack.

. The majority's conclusion that "Smith likely dropped the rope when he paused briefly in the bedroom doorway and before he continued on down the hall” — a conclusion based solely on the fact that “the rope was found just to the 'left of the bedroom door' ” —is speculation of the highest order. Maj. Op. at 817-18 n. 17. There is absolutely nothing in the record that would suggest that Smith dropped the rope when the majority posits that he did. The majority cannot even provide a reason why Smith would have dropped the rope at that point. A reasonable juror could easily draw a multitude of very different conclusions from this evidence.

. A chainsaw from the Konzelmans' garage was among the items that Edmonds stole and brought back to his truck.

.The majority mischaracterizes the district court’s consideration of this issue. The majority states that "The district court never found that Edmonds and Smith planned to engage in any violent conduct or even discussed doing so.” Maj. Op. at 820. In fact, the district court did not reach the question of whether Smith and Edmonds had explicitly planned to engage in violent conduct or discussed that possibility because it concluded that the nighttime burglary of a home that they knew was occupied was sufficiently dangerous conduct to foreclose the affirmative defense. To imply that the district court considered whether Edmonds and Smith "planned to engage in any violent conduct,” or would have needed to, is highly misleading.

. The majority states that "Edmonds himself stated that prior to entering the house, there had never been any discussion between him and Smith about hitting anyone or using any kind of weapon on anyone.” Maj. Op. at 820. The record demonstrates that Edmonds has stated the exact opposite multiple times, a fact which the majority ignores. Edmonds also testified at Smith’s sentencing hearing that they took the rope so they could bind the occupants if they awoke.

. The majority relies on Enmund v. Florida, 458 U.S. 782, 799-800, 102 S.Ct. 3368, 73 L.Ed.2d 1140 (1982) for statistics to dispute that a "generic residential robbery is likely to result in death.” Maj. Op. at 821 & n. 23. But while Supreme Court opinions are binding precedent for their legal holdings, they are not so for their statistical references.

Even if, as the majority argues, the Enmund statistics are still valid today, the occurrence of homicide during a robbery is not a sufficiently refined measure of whether any given activity is likely to result in death. Smith and Edmonds discussed tying up any occupants of the house. They anticipated that their planned burglary might turn confrontational. Certainly the likelihood of serious injury or death occurring under these circumstances may be greater than average. See http:// www.fbi.gov/ucr/ 05cius/index.html (providing definitions and reporting statistics).

.The majority's statement that "there [we]re no plans to engage in violent conduct and, to Smith's knowledge, no weapons on hand with which to do so” is patently false. Maj. Op. at 821. The record is clear that the burglars were contemplating tying up the house's residents. This would clearly be violent conduct. As I have discussed earlier, the rope itself was a weapon.

midpage