In this pre-AEDPA case, we are called upon to review counsel’s performance in a twenty-two year old capital murder trial. Finding that the record is clear that counsel requested critical jury instructions that were correct under federal law but clearly in error under Idaho law and that the error was not harmless, we reverse the judgment of the district court and grant the writ.

I. FACTS AND PROCEEDINGS

Robert and Cheryl Bravence were reported missing when they failed to arrive at a relative’s house following an extended camping trip in the summer of 1983. One week later, officials discovered their van abandoned at a Los Angeles bus terminal. In September, hunters found the couple’s remains in a remote Idaho campground, not far from their last known campsite. The skulls of both were badly damaged, indicating they had died of blunt trauma to the head.

Two brothers, Mark and Bryan Lank-ford, were arrested for the murders. Investigators found Mark’s car about a quarter of a mile from the victims’ bodies and his handwriting on credit slips for food and lodging purchased with the victims’ stolen credit cards. Investigators found both brothers’ fingerprints in the Bravences’ abandoned van, and they found Mr. Bra-vence’s knife — with Bryan’s initials newly carved into the scabbard — in Bryan’s possession.

The brothers were convicted, in separate trials, and sentenced to death on the same day. Bryan’s sentence was overturned on appeal based on the sentencing agreement he received in return for his testimony against Mark. Although there was strong circumstantial evidence connecting the Lankfords to the murders, there were no witnesses to the crime, and no murder weapon was discovered. Bryan’s testimony was critical to establishing the government’s theory that Mark actually killed the Bravences. At Mark’s trial, the state relied heavily on Bryan’s uncorroborated eyewitness testimony that Mark committed the murders, beating the Bravences to death with a small club or nightstick. When interviewed by police, Bryan claimed that both brothers went to the campsite. While Bryan engaged the campers in conversation,

At Mark’s trial, Bryan claimed Mark hit Robert Bravence “twice. It could have been once.” At his own trial, Bryan described one blow and added, “I think he struck the man another time.” He said the blows were aimed “both times in the back of the neck actually. Not in the head. Kind of across, you know, across the neck in the back.” He testified that Mark then struck Cheryl, also “across the back of the neck,” “one time as I can remember,” “only once.... [I]t could have been more. I don’t know.” Although Bryan admitted that his emotions might have interfered with his perception, in both his trial and in Mark’s trial he maintained that Mark hit the couple in the back of the neck, striking Robert twice and Cheryl once. Throughout the two trials, and in statements to investigators, Bryan described the murder weapon variously as possibly a “pipe,” “sort of a nightstick” made of wood, “a little limb deal,” and a “little club” only twelve inches long. Investigators found a nightstick in Mark’s abandoned car, but could not tie it to the murders and did not introduce it into evidence.

Mark did not testify at either trial, but he maintained (in statements to his attorney and to the police) that Bryan had confessed to committing the murders and enlisted his help to hide the bodies. Mark claimed that Bryan went to the couple’s camp while the brothers were separated and attacked the campers, crushing their skulls with a large rock. Mark said the brothers had traveled from Texas to Idaho because Mark was trying to get away from what he perceived to be his “materialistic” life in Texas, while Bryan wanted to escape an impending arrest for parole violations. They camped in Idaho at a remote site near Grangeville. After Bryan decided he wanted to return to Texas, Mark left Bryan to hitchhike his way into town. He gave Bryan fifty dollars, told him not to do anything “stupid,” and started walking back to the campsite. Before he reached the site some miles away, Bryan drove up in the stolen van. He initially refused to tell Mark how he stole the van, exclaiming “I got it. That’s what counts” and complaining “what are all these questions, I told you, don’t worry about it!” Finally after several questions, Bryan told Mark about his attack on the Bravences, claiming he hit Robert “in the head with my shotgun stock. I didn’t hit him too hard, just to knock him out. I pinched his leg and he didn’t move so I thought he was out of it.” When Cheryl appeared, he struck her, too “but not as hard. She was out like a light.” Mark insisted the two return to the Bravences’ camp, telling Bryan “[Ijt’s not too late to save your ass, maybe.” As the two neared the site, Bryan grew increasingly nervous. According to Mark, “[Bryan’s] driving was getting reckless and he was chain-smoking the entire time.” “The closer we got to the place, the more nervous [Bryan] became.” At the site, Mark immediately spotted the two figures, each lying with its head in a pool of blood. He checked for pulses and told Bryan “they’re both dead.” Bryan replied “Are you sure?” “They can’t be!” Mark then confronted Bryan: “shotgun butt? Tell me the truth.” “It was just like I said,” Bryan explained,

After Bryan and his brother were convicted in 1984, Bryan recanted his testimony against Mark on multiple occasions. He also recanted his recantations. In a conversation with a newspaper reporter, in a written statement after his own sentencing, and again at a hearing on Mark’s second motion to reopen, Bryan admitted responsibility for the murders, claiming he killed the Bravences with a rock. Bryan contacted a local reporter in June of 1984, after the brothers had been convicted, but before either was sentenced. In October of 1984, at a hearing on Mark’s first motion for a new trial, the reporter testified that Bryan telephoned him to confess, saying he “[w]anted the truth to come out,” and admitting that he killed the Bravences by knocking Mr. Bravence unconscious with a stick and then beating them both with a rock. In this version of Bryan’s story, Mark was not on the scene. Bryan also testified at Mark’s hearing, admitting to the confession but claiming it was a lie that Mark directed him to make so “he would get out, therefore, they would have to let me out.”

Later, in 1985, Bryan wrote a letter to Mark’s counsel again accepting full responsibility. This prompted a subsequent hearing on Mark’s second motion for a new trial. This time Bryan testified that he entered the Bravences’ camp alone, carrying a 12-gauge shotgun, and ordered the couple to lie down. He crushed their skulls with “a rock, a large boulder” that was “about a foot in diameter [and] probably weighed about thirty-five or forty pounds or more.” He struck each camper three or four times, and then walked back to the road to enlist Mark’s help in hiding the bodies. Bryan described the Bra- vences’ campground and his attack on the campers in detail, and admitted that his earlier testimony against Mark was an attempt to “save [him]self.” The court denied Mark’s motions for a new trial. In 1998, while he was awaiting resentencing, Bryan again contacted Mark’s counsel, asking to speak with him. Bryan’s lawyer objected to the proposed interview, threatening to withdraw if Bryan insisted upon meeting with Mark’s attorney. As late as April 2003 in a letter to Mark’s appellate counsel, Bryan continued to assume full responsibility, confessing that he struck both the Bravences “several times” with “a head size rock.”

Mark pursued his state appeals, post-conviction relief, and state habeas relief, then (after several false starts) he filed a federal habeas petition. His ineffective assistance claims were initially dismissed because they had not been properly presented to the Idaho courts. However, the parties stipulated to an evidentiary hearing on Sixth Amendment claims after we ruled in Hoffman v. Arave, 236 F.3d 523 (9th Cir.2001), that Idaho procedural rules for state habeas claims “frustratefd] [prisoners’] exercise of a federal right” and were “inadequate to preclude federal [review].” Id. at 531 (internal quotation marks omitted).

Among the grounds raised in his state and federal habeas proceedings, Mark contended that ineffective assistance from his trial counsel, Gregory FitzMauriee, deprived him of his Sixth Amendment right to counsel. Mark argued that FitzMauriee failed to consult with an independent forensic pathologist, which would have shown the weaknesses in the state’s theory that Mark committed the murders; that FitzMauriee failed to impeach Bryan’s testimony concerning the murder weapon; and that FitzMauriee requested an erroneous jury instruction that reduced the state’s burden to provide corroborating evidence. The district court concluded that as to the forensic evidence, FitzMauriee made a “reasonable strategic decision” to present an alibi defense. The district court also concluded that FitzMauriee did impeach Bryan. Finally, with respect to the jury instruction, the district court found that FitzMaurice’s requested jury instruction was erroneous and that his performance in this regard was deficient. The court concluded, however, that Mark did not suffer prejudice as a result.

FitzMauriee requested and received the following instruction on accomplice testimony, cited in the record as instruction number 15:

(emphasis added).

FitzMauriee also submitted two additional instructions which contradicted instruction 15 by requiring that the jury find corroboration before accepting any accomplice testimony. Instruction 18 read:

Instruction 19 provided:

Mark claims that the instructions were not only inconsistent, but that instruction 15 is an incorrect statement of Idaho law. After three days of testimony, the district court held that the instruction was in error and found that FitzMaurice was ineffective in offering the erroneous instruction. The court also found FitzMaurice deficient in failing to research and present mitigating evidence at Mark’s sentencing. However, the court ultimately denied the petition, finding there was no prejudice.

II. STANDARD OF REVIEW

Lankford’s petition predates the Anti-terrorism and Effective Death Penalty Act (AEDPA) and is therefore not governed by it. We review the district court’s denial of the petition de novo. Turner v. Calderon, 281 F.3d 851, 864 (9th Cir.2002). Because ineffective assistance of counsel is a mixed question of law and fact, we review this claim de novo. See Silva v. Woodford, 279 F.3d 825, 835 (9th Cir.2002). We review factual determinations for clear error. Id.

III. ANALYSIS

Under Strickland v. Washington, 466 U.S. 668, 687 , 104 S.Ct. 2052 , 80 L.Ed.2d 674 (1984), Lankford may only prevail in his claim of ineffective assistance of counsel if he can show deficient representation “so serious that counsel was not functioning as the ‘counsel’ guaranteed the defendant by the Sixth Amendment.” He must also show “that counsel’s errors were so serious as to deprive the defendant of a fair trial, a trial whose result is reliable.” Id. These two elements of Strickland are generally referred to as the “performance” and “prejudice” prongs. We evaluate each in turn.

A

As we review FitzMaurice’s performance, we must refrain from second-guessing his strategies and acknowledge the “wide range of reasonable professional assistance.” Id. at 689 , 104 S.Ct. 2052 . Nevertheless, we must hold FitzMaurice to his “duty to bring to bear such skill and knowledge as will render the trial a reliable adversarial testing process.” Id. at 688 , 104 S.Ct. 2052 . Even considering “conduct from counsel’s perspective at the time,” in at least one respect, FitzMau- rice’s representation falls below the standard of “reasonable professional assistance.” Id. at 689 , 104 S.Ct. 2052 . We agree with the district court that there was no reasonable tactical advantage in requiring an erroneous jury instruction that would allow the jury to give greater weight to Bryan’s testimony. In this case, “[cjounsel’s errors with the jury instructions were not a strategic decision to fore-go one defense in favor of another. They were the result of a misunderstanding of the law.” United States v. Span, 75 F.3d 1383, 1390 (9th Cir.1996).

The error in the instructions requested by FitzMaurice is obvious. Instruction 15 left out a critical element of Idaho law controlling the use of accomplice testimony: the requirement of corroboration. Idaho law provides:

IDAHO CODE § 19-2117 (2005) (emphasis added). As the Idaho Supreme Court has explained, “ ‘[the] statute ... absolutely prohibits a conviction in a criminal case upon the uncorroborated testimony of an accomplice, even although[sic] the jury may believe such testimony to be entirely true, and that it establishes the defendant’s guilt beyond a reasonable doubt.’ ” State v. Emmons, 94 Idaho 605 , 495 P.2d 11, 15 (1972) (quoting State v. Carr, 28 Or. 389 , 42 P. 215, 216 (1895) (evaluating a statute that the Emmons court deemed “essentially the same as I.C. § 19-2117”)).

The Idaho statute reflects the state’s unwillingness to hand down convictions based solely on accomplice testimony. As the court in Emmons observed:

Id. (quoting Carr, 42 P. at 216 ); see id. (‘We concur with the reasoning of the Oregon Supreme Court....”). The Idaho Court of Appeals has explained that “[t]his statutory corroboration requirement is intended to protect against the danger that an accomplice may wholly fabricate testimony, incriminating an innocent defendant in order to win more favorable treatment for the accomplice.” Matthews v. State, 136 Idaho 46 , 28 P.3d 387, 390 (2001).

FitzMaurice explained that he performed his research at the library of the University of Idaho Law School, that there were no model instructions for Idaho at the time, and it seems that he took the instruction from a collection of federal in structions. In this, FitzMaurice was dutiful, conscientious — and quite in error. Federal law permits bare accomplice testimony, while Idaho expressly forbids it. Compare Emmons, 495 P.2d at 15 , and State v. Gillum, 39 Idaho 457 , 228 P. 334, 334 (1924), with Darden v. United States, 405 F.2d 1054, 1056 (9th Cir.1969) (noting that “a conviction in federal court may be based on the uncorroborated testimony of an accomplice”). FitzMaurice simply overlooked important differences between Idaho law and federal law. It was a young lawyer’s mistake, akin to failing to check the pocket part, but it was a mistake, plainly enough. As the district court noted, “[s]imply reviewing the three instructions would have revealed that two instructions ... conformed with Idaho’s law regarding accomplice testimony, while [the third] was substantially different, as well as not being in conformance with Idaho law.”

FitzMaurice’s error is perhaps understandable, given his limited experience and resources, but it is constitutionally inexcusable. By inviting a jury instruction that misstated state law and made it easier for the jury to convict his client, counsel unwittingly undermined the very “adversarial testing process” he was supposed to protect. Strickland, 466 U.S. at 688 , 104 S.Ct. 2052 . We agree with the district court that in this regard his performance fell below the “range of reasonable professional assistance.” Id. at 689 , 104 S.Ct. 2052 .

B

Whether the erroneous instruction rendered the ultimate verdict in this case unreliable is a harder question. “It is not enough for the defendant to show that the errors had some conceivable effect on the outcome of the proceeding.” Id. at 693 , 104 S.Ct. 2052 . To reverse we must find a “reasonable probability that ... the result of the proceeding would have been different” had the erroneous instruction not been given. Id. at 694 , 104 S.Ct. 2052 ; see also Span, 75 F.3d at 1390 . And we must consider the instructions “as a whole” in evaluating the magnitude of the error. Middleton v. McNeil, 541 U.S. 433, 437 , 124 S.Ct. 1830 , 158 L.Ed.2d 701 (2004). “[A] single instruction to a jury may not be judged in artificial isolation, but must be viewed in the context of the overall charge.” Boyde v. California, 494 U.S. 370, 378 , 110 S.Ct. 1190 , 108 L.Ed.2d 316 (1990) (internal quotation marks omitted); see also Estelle v. McGuire, 502 U.S. 62, 72 , 112 S.Ct. 475 , 116 L.Ed.2d 385 (1991). Accordingly, Mark need not establish error for certain. “[A] defendant need not establish that the jury was more likely than not to have been impermissibly [influenced] by the instruction.” Boyde, 494 U.S. at 380 , 110 S.Ct. 1190 . “If the charge as a whole is ambiguous, the question is whether there is a reasonable likelihood that the jury has applied the challenged instruction in [an impermissible] way....” Middleton, 541 U.S. at 437 , 124 S.Ct. 1830 (internal quotation marks omitted) (emphasis added).

The government argues that any error was cured because there were additional, proper instructions (instructions 18 and 19) that arguably contradicted instruction 15. Instruction 18 stated that “a defendant cannot be found guilty based upon the testimony of an accomplice unless such testimony is corroborated by other evidence which tends to connect such defendant with the commission of the offense.” (emphasis added). Instruction 19 defined “corroboration” and instructed the jury to “assume the testimony of the accomplice has been removed from the case. You must then determine whether there is any remaining evidence which tends to connect the defendant with the commission of the offense.”

When instructions 15 and 18 are read together they can be reconciled, but not in a way that is helpful to the state. Instruction 18 plainly states that “a defendant cannot be found guilty based upon the testimony of an accomplice unless such testimony is corroborated by other evidence.” That, of course, contradicts instruction 15 which states that an accomplice’s testimony “may be received in evidence and considered by the jury, even though not corroborated by other evidence.” The two statements seem to be bridged by the last sentence of instruction 15: ‘You should never convict a defendant upon the unsupported testimony of an alleged accomplice, unless you believe that unsupported testimony beyond a reasonable doubt.” This line appears to create an exception to instruction 18’s warning not to credit uncorroborated testimony: The testimony can be credited if the jury believes it beyond a reasonable doubt. To the trained legal eye, the “beyond a reasonable doubt” adds nothing to instruction 15 because a criminal jury generally must find proof beyond a reasonable doubt to return a guilty verdict. See Cool v. United States, 409 U.S. 100, 104 , 93 S.Ct. 354 , 34 L.Ed.2d 335 (1972); In re Winship, 397 U.S. 358, 362 , 90 S.Ct. 1068 , 25 L.Ed.2d 368 (1970). Nevertheless, a conscientious jury might read instructions 15 and 18 together to state a general rule and its exception: A defendant cannot be found guilty based upon the testimony of an accomplice unless either (a) such testimony is corroborated by other evidence, or (b) the jury believes the unsupported testimony beyond a reasonable doubt. The instructions, thus reconciled, remain contrary to Idaho law. Accordingly, the jury could have convicted Lankford either because they found Bryan’s testimony to be adequately supported, or because they believed his uncorroborated testimony beyond a reasonable doubt. “[Wjhen a case is submitted to the jury on alternative theories the [impermissible legal error in] any of the theories requires that the conviction be set aside.” Leary v. United States, 395 U.S. 6, 31-32 , 89 S.Ct. 1532 , 23 L.Ed.2d 57 (1969); see also Griffin v. United States, 502 U.S. 46, 59 , 112 S.Ct. 466 , 116 L.Ed.2d 371 (1991) (“[T]he term ‘legal error’ means a mistake about the law, as opposed to a mistake concerning the weight or the factual import of the evidence.”).

It is hard to imagine a case in which such an instructional error could have caused more damage. Erroneous instruction 15 went to the heart of the case. Although there is overwhelming evidence that one or both of the Lankford brothers killed the Bravences, only Bryan’s testimony singled out Mark as the killer. There were no witnesses to the murder, and no murder weapon was admitted into evidence. There was no forensic or circumstantial evidence suggesting that Mark, rather than Bryan, beat the victims to death. Bryan’s testimony was vital to the prosecution, and instruction 15 allowed the jury to convict Mark on Bryan’s word alone, in obvious .violation of Idaho law. Moreover, Bryan had every incentive to lie: If Mark did not kill the Bravences, then Bryan must have done the deed. Additionally, Bryan thought he had a deal to avoid the death penalty in exchange for his testimony against his brother. See Lankford v. Idaho, 500 U.S. 110, 120-121 , 111 S.Ct. 1723 , 114 L.Ed.2d 173 (1991); State v. Lankford, 127 Idaho 608 , 903 P.2d 1305, 1309 (1995). This is precisely the declared purpose for Idaho’s corroboration requirement: “to offset the danger that an accomplice may wholly fabricate testimony, inculpating an innocent person, in order to purchase immunity from prosecution, or lenient treatment, for his own complicity in the crime.” State v. Pierce, 107 Idaho 96 , 685 P.2d 837, 842 (Ct.App.1984) (citations omitted).

Considering the instruction “in the context of the instructions as a whole and the trial record” does not inspire confidence in the reliability of Mark’s verdict. Estelle, 502 U.S. at 72 , 112 S.Ct. 475 . The state nonetheless attempts to salvage Mark’s conviction by relying on Idaho law, which provides that “[a]n error in failing to give [a corroboration] instruction may be harmless if ample corroborative evidence was presented.” State v. Hill, 140 Idaho 625 , 97 P.3d 1014, 1019 (Ct.App.2004). The state contends that the prosecution presented corroboration adequate to meet the requirements of Idaho Code § 19-2117 .

The Idaho Supreme Court has explained in general terms what corroborating evidence is required by § 19-2117:

State v. Garcia, 102 Idaho 378 , 630 P.2d 665, 672 (1981) (quoting State v. Bassett, 86 Idaho 277 , 385 P.2d 246, 248 (1963)) (citations omitted). Idaho courts have also explained that:

Gillum, 228 P. at 336 . We note that most of the Idaho cases addressing the statute arose as claims disputing the sufficiency of the evidence, not in the context of an erroneous instruction, and in sufficiency cases “[the c]ourt will construe all of the evidence in favor of upholding the verdict.” Hill, 97 P.3d at 1018 . In such cases, the court must uphold the jurors’ verdict if there is any basis for a finding of corroboration because they are “the judges of the weight and credibility of the testimony under proper instructions.” Garcia, 630 P.2d at 672 (emphasis added) (quoting Bassett, 385 P.2d at 248 ); see also State v. Gonzales, 92 Idaho 152 , 438 P.2d 897, 901 (1968).

The error here is a more fundamental one. The jury was instructed to overlook the question of corroboration, and we must ask whether, given proper instructions, there is a “reasonable probability” that the jury would have found Bryan’s testimony unsupported. “A reasonable probability is a probability sufficient to undermine confidence in the outcome.” Strickland, 466 U.S. at 694 , 104 S.Ct. 2052 . The Idaho cases suggest that, where the trial court has refused a corroboration instruction under § 19-2117, the Idaho Supreme Court has demanded substantial evidence of corroboration before it will find the error harmless. In State v. Scroggins, 110 Idaho 380 , 716 P.2d 1152 (1985), the Idaho Supreme Court found that the trial court committed harmless error when it omitted the corroboration instruction in a felony murder conviction where Scroggins “admitted that he assisted [the accomplice] in taking the victim to the creek, that he, Scroggins, had handcuffed the victim, that [the accomplice] had used Scroggins’ knife, that he, Scroggins, had attempted to rape the victim and that he had been in the vicinity when [the accomplice] had, in fact, raped and murdered the victim.” Id. at 1158. The Court concluded that “although Scroggins’ testimony did not corroborate [the accomplice’s] version of the facts, it was sufficient to permit a finding that Scroggins was connected with the commission of the offense.” Id. In Matthews v. State, 136 Idaho 46 , 28 P.3d 387 (Ct.App.2001), Matthews sought post-conviction relief from his conviction for first degree murder. Id. at 389-90. Matthews argued that his counsel was ineffective because he failed to request a jury instruction on corroboration of accomplice testimony. The court concluded that the jury should have been instructed, but it found the error was harmless. Matthews had admitted that he joined four other juveniles in beating the victim; testimony from other witnesses also corroborated the testimony from Matthews’s accomplices. In this case, “the outcome of the trial would [not] have been different had Matthews’s counsel requested a jury instruction regarding the corroboration of accomplice testimony.” Id. at 391-92.

These eases, in which there was substantial corroborating evidence, contrast with the Idaho Supreme Court’s decision in State v. Gonzales, 92 Idaho 152 , 438 P.2d 897 (1968). In that case, the Idaho Supreme Court held that failure to give the corroborating evidence instruction was not harmless even though the evidence included witnesses who saw the defendant fight with the victim before the murder and who recognized the defendant’s car at the subsequent drive-by shooting. Id. at 899, 901. Investigators found a spent cartridge at the defendant’s house and a rifle at his mother’s apartment. Id. at 900. The defendant allegedly asked police at his booking, “How is the fellow I shot, is he ded [sic] or what?” Id. at 899. Even with all of this evidence, the Court reversed the conviction because the testimony of an accomplice (a passenger in the car) was not properly limited with a corroboration instruction. “[T]he [trial] court had a duty to instruct the jury on all matters of law necessary for their information,” the Court explained, “and the defendant was entitled to have his theory of the case submitted to the jury upon proper instructions.” Id. at 901.

The nature of the evidence sufficient to corroborate is further explained in State v. Campbell, 114 Idaho 367 , 757 P.2d 230 (Ct.App.1988). In that case, Campbell was accused of kidnaping and murder. Id. at 231. He argued that there was insufficient evidence to corroborate the testimony of his two accomplices, who testified that Campbell had repeatedly kicked the victim. The state introduced a pathologist who testified that the victims injuries were consistent with the testimony of the accomplices. The Idaho Court of Appeals held that this was not sufficient:

Id. at 232-33. Although the pathologist’s report was not sufficient corroboration, the court found sufficient corroboration in other evidence, including statements that Campbell made to others. Id. at 233-34.

The corroborative evidence the state points to in this case — including, principally, the testimony of the state’s pathologist, Dr. Robert Cihak — implicates the Lank-fords generally, not Mark specifically. There is no evidence that uniquely points to Mark as the perpetrator. All of the state’s evidence equally implicates Bryan. The only physical evidence presented at trial was circumstantial, and all of that evidence also linked Bryan to the murders. The only evidence tending to show that Mark was the killer was his brother’s word. We conclude there is a “reasonable probability” that, with proper instructions on the question of corroboration, the jury would not have convicted Mark of first degree murder. Strickland, 466 U.S. at 694 , 104 S.Ct. 2052 .

We find FitzMaurice’s error in drafting jury instructions an adequate grounds for reversal. Although we do not reach the other grounds on which Mark contends that FitzMaurice rendered ineffective assistance of counsel, the evidence Mark has presented in support of those claims reinforces our judgment that Mark was prejudiced in this case. At the federal eviden-tiary hearing, Mark presented experts who questioned the reasonableness of FitzMau-rice’s failure to hire an independent forensics expert to investigate Mark’s claim that Bryan beat the Bravences with a large river rock. The testimony of Dr. Todd Grey, forensic pathologist and chief medical examiner for the state of Utah, strongly suggests that an independent forensic investigation would have revealed serious flaws in Bryan’s story that Mark beat the Bravences with a stick. He pointed out that the “extensive fragmentation” of the victim’s skulls suggests the couple were bludgeoned with a larger object, such as the river rock Mark described. He described Cheryl Bravence’s skull as “a pile of bones” before it was reconstructed by forensics experts, and explained that her skull showed multiple linear and branching fractures. In at least one area, parts of the bone were missing where bone had been broken off and driven downwards into her brain. Robert Bravence’s skull also required reconstruction, and it had to be reinforced with clay where sections of bone were missing. Because the Bravences’ skulls showed “extensive firacturing,” “depression of bone,” and “fragmentation” on virtually all sides, Dr. Grey concluded that their injuries were “much more consistent with the scenario in which these individuals were struck with the large rock.” To cause this level of destruction with a twelve-inch stick, the expert testified, “would take a tremendous amount of force,” “an horrendous amount of force.” “I am not saying it is impossible,” Dr. Grey explained, “but it would certainly take a lot of effort.” Although Bryan claimed Mark dealt Robert a “hard blow” that necessarily dislodged portions of his skull, Bryan testified at his own trial that there were no sounds “such as bones breaking” and that there was no blood. The single skull-crushing strike that killed Cheryl was apparently also silent and bloodless. Bryan certainly did not describe a clubbing that used “an horrendous amount of force.”

Dr. Grey’s testimony not only cast doubt on the identity of the murder weapon and the amount of force employed, it clearly showed that the victims must have endured multiple blows. Bryan’s description of a quick, three-strike attack did not fit the physical evidence. It was more likely that the victims were struck several times than that they were each hit once or twice (as Bryan maintained), because skulls were not only severely fractured and fragmented, but sustained damage on nearly every side, suggesting blows from various angles. Cheryl Bravence’s skull showed fractures on the top and back of the skull, breaks near the right ear, and fractures in the left temple — hardly the work of a single silent strike to the neck. Her husband’s skull sustained fractures at the back and top of the head, was missing parts of bone from the left side, and suffered breaks in the left eye socket, right brow, and left jaw. A fracture also separated his alveolar plate — a portion of bone that supports the upper teeth. “To inflict the extent of damage that I see in this skull with three blows is pretty far beyond the pale of what I would expect to be able to see,” Dr. Grey explained. “I can’t see how that pattern would be produced by a stick with that scenario [of three blows] that you described, even with as much force as you could possibly muster, I don’t see how you could get that pattern.” On cross examination, Dr. Grey admitted that the injuries he observed could have been caused by as few as three blows to Cheryl and four to Robert — making seven strikes altogether — but more blows would likely be needed. In addition, he explained that Bryan’s description of the location of the blows did not match the physical evidence. The extensive injuries on both sides of the skulls could not have been produced during the attack Bryan described, where an assailant struck the victims only “in the back of the neck.... Not in the head.”

Had Dr. Grey been asked to serve as an advisor in a trial that presented this sort of forensic evidence, he would have advised the attorney cross examining Bryan to “get [a] detailed ... description of how the blows were inflicted, the number of blows, what kind of a swing was used, what the implement, the stick, was like, [and whether] it appear[ed] to be weighted or unweighted” so that the testimony could be tested against the pattern of injury. Dr. Grey affirmed that his type of questioning, together with detailed questioning of an expert pathologist, was “the sort of thing you customarily do when you are retained to work a case.” Dr. Grey did not disagree with Dr. Cihak’s testimony at trial that the injury was the result of blunt trauma that could have been caused by a pipe, nightstick, rock, or a gun. The problem was that Dr. Cihak was never “really pinned down on what [wa]s more likely given the pattern of -injury.” “He never was asked to say among those choices what, seeing this pattern of injury, what seems more likely....” By testifying to injuries caused by blunt trauma, Dr. Cihak merely affirmed “[a] myriad of possibilities” as to the murder weapon.

The forensic evidence, considered in light of Idaho’s corroborative requirement, underscores the inconsistencies in Bryan’s story. Bryan wavered in his physical description of the weapon and its origin. He described the murder weapon in ambiguous terms as a “sort of nightstick” made of wood, “a little limb deal” and a “little club” twelve inches long. In an interview with the sheriff, he had trouble remembering specifics about the stick. It “could have been” a piece of pipe, but Bryan did not know “for sure.” When asked whether the weapon was a foot long, he said “umm, probably about, yeah, probably about a foot and a half, maybe.” His descriptions changed over the course of the investigation. In another of his initial statements, given to police, he claimed that the weapon was “a stick like a night stick that a policeman wears but only about a foot long.” He told an FBI investigator that it was “a ‘night stick’ like a policeman uses. [Mark] had the ‘night stick’ for a long time and carried it in his car.” On direct examination at his own trial, Bryan said the weapon was “a thing about a foot long, which is a little club that he has had for a long time.” He claimed he “d[id]n’t know where [Mark] got it from.... Probably bought it somewhere.” Later, at Mark’s trial, Bryan remembered the weapon’s appearance and history clearly. It was “a sort of a night stick, in a way.... It’s about a foot long, reddish brown wood. I think it, I believe it came from me. I believe my wife — my ex-wife gave it to me, and I gave it to Mark a long time ago.” At his trial, Bryan claimed that he did not know Mark was carrying the club when the two entered the Bravences’ camp. “I never saw the club,” he claimed, explaining that it was so small it could “[v]ery easily” have been concealed in a pocket. He also said he had not seen the little club since the killings. It could have been at the Bravences’ camp site, he explained, or it could be at the spot where the bodies were unloaded or the place they were ultimately hidden. “I’m sure it would be in one of the three places,” he told the court. Mark did in fact carry a nightstick, and it was recovered. Police described it as a wooden nightstick, dark brown or black in color, about eighteen inches long. The stick was not found at the murder site, on the road, or at the Bravences’ grave site. It was found in Mark’s abandoned car. The nightstick the police found in Mark’s car would, of course, have been powerful corroboration of Bryan’s testimony, but the state — to its credit — declined to introduce it into evidence because it could not be linked to the killings.

In sum, Bryan’s testimony about the murder weapon, the number of blows, and the force of those blows was vague, contradictory, and — when considered in light of the Bravences’ fragmented skulls — not persuasive. It is this precise scenario — an accomplice who must implicate the defendant in order to hide his own culpability — • that Idaho’s corroboration requirement addresses. We agree with the district court that FitzMaurice’s requested jury instructions were simply wrong under Idaho law. The corrupted instruction was not corrected by other instructions. Although the question is close, we disagree with the district court as to the prejudicial effect of the error. There was ample evidence that either one or both of the Lankfords killed the Bravences, but there was no evidence that Mark attacked and killed the Bra-vences other than Bryan’s testimony, and there was strong evidence raised at the habeas proceedings suggesting that Bryan’s testimony is inconsistent with the forensic evidence. Counsel’s request that the jury be instructed that it could convict on the basis of Bryan’s testimony alone was plainly prejudicial.

IV. CONCLUSION

The judgment of the district court is reversed and the case is remanded with instructions to grant the petition for a writ of habeas corpus, with appropriate instructions to retry Mark Lankford within a reasonable time or release him.

REVERSED and REMANDED.

. Under cross examination, Bryan admitted at trial that he expected to receive an indeter-mínate life sentence in return for his testimony- . In the evidentiary hearing in district court, the state attempted to characterize FitzMau-rice’s selection of the instruction as a tactical move, arguing that there was sufficient corroboration and that the jury would not be able to find Bryan credible beyond a reasonable doubt without corroborating evidence. FitzMaurice could not confirm the theory, explaining "I can’t tell you at this point in time ... my motivations of asking [for] that jury instruction. That appears to be a plausible reason, but I can't tell you honestly that was my reason at the time.” Even if Fitz-Maurice did introduce the instruction in order to draw attention to Bryan’s credibility, a general instruction on reasonable doubt could have served the same purpose without eliminating the corroboration requirement. . FitzMaurice was a part-time public defender who had, it seems, only tried one previous major felony: a case of cattle-rustling. He had never defended a murder case, and had no training in capital defense. He delivered no opening statement to the jury, and called no witnesses. He did not secure a second attorney, an investigator (except for a fingerprint analyst), an independent pathologist, or an independent psychologist to aid in preparing a defense. . There is another possibility, also not helpful to the state. The jury could have read instruction 15 to mean that they could convict Mark on Bryan's unsupported testimony if they believed Bryan beyond a reasonable doubt. Or, they could convict Mark even if they did not believe Bryan beyond a reasonable doubt so long as there was some evidence to corroborate his testimony. This latter possibility would, of course, flatly violate In re Winship. . The stick did not show any blood or tissue residue. Although the two brothers have together given several versions of the murder story, never did either claim to have returned to Mark’s brush-covered car after the killings.