Full opinion text
EBEL, Circuit Judge. Petitioner Marcus Cargle was convicted in Oklahoma of first degree murder and sentenced to death in connection with the shooting of Richard and Sharon Paisley during a drug transaction at their home. Present with petitioner during the fatal incident were Christopher Todd Jackson and Christopher Todd Williams. Apart from the critical question of who shot Sharon, most of the material facts surrounding the homicides were not in dispute at trial. Petitioner was nineteen years of age when the events took place. He and several acquaintances, including Jackson, also nineteen, and Williams, twenty-four, were socializing at his uncle’s house in the late afternoon. Earlier in the day, petitioner had purchased $100 worth of marijuana from the Paisleys. Dissatisfied with the marijuana, petitioner and his two friends left his uncle’s house to recover the money he had paid to the Paisleys. Their arrival at the Paisley home did not initially spark any confrontation; Sharon served them beer and popcorn while Richard left to get the cash from a neighbor. Richard returned and gave petitioner the $100, and petitioner and Jackson prepared to leave. Just then, however, Williams returned from the bathroom brandishing a firearm and, without warning or provocation, shot Richard in the chest. Petitioner and Jackson did not move. Williams shot Richard again, this time in the head, and Richard fell to the floor. According to Jackson, at this point petitioner exclaimed “damn,” ran over to where Sharon had crawled from a couch to the floor, and shot her in the head. His gun jammed and, while he fumbled with it, Williams shot Richard for a third and last time. Then, according to Jackson, petitioner shot Sharon once more. The three men tried to wipe their fingerprints from the scene, Williams took a television and VCR, and they left. While they were driving away, Williams threatened petitioner and Jackson that whoever said anything would have to die too. Petitioner was brought to trial first. Largely on the basis of Jackson’s immunized testimony, he was convicted of murdering Sharon and aiding and abetting in Richard’s murder. The death penalty was imposed on both counts. The Oklahoma Court of Criminal Appeals (OCCA) affirmed the convictions and sentences on direct appeal, see Gargle v. State, 909 P.2d 806 (Okla.Crim.App.1995), and subsequently denied post-conviction relief almost entirely on waiver grounds, see Gargle v. State, 947 P.2d 584 (Okla.Crim.App.1997). Petitioner timely filed a petition for a writ of habeas corpus under 28 U.S.C. § 2254. Of the many claims asserted in the petition, only the following are before us for consideration on appeal: (1) ineffective assistance of counsel in the guilt and penalty phases of trial; (2) prosecutorial misconduct in the guilt and penalty phases; (3) admission of improper victim impact testimony in the penalty phase; (4) insufficient evidence as to the murder of Richard Paisley; (5) improper excusal of veniremen because of reservations expressed about the death penalty; (6) facial and as-applied challenges to the aggravating circumstances invoked by the prosecution to support the death penalty; (7) failure to disclose exculpatory evidence under Brady v. Maryland, 373 U.S. 83, 83 S.Ct. 1194, 10 L.Ed.2d 215 (1963); and (8) cumulative error. The district court rejected petitioner’s challenges to the guilt phase of trial but concluded that he was entitled to sentencing relief on the bases of ineffective assistance of counsel, improper victim impact testimony, and the cumulative effect of these two errors during the penalty phase of trial. Accordingly, the district court granted petitioner’s request for relief from his death sentences, allowing the State 180 days to conduct a new sentencing hearing. Both petitioner and the State have appealed from the district court’s rulings and judgment. As explained below, we agree with the district court that petitioner’s death sentences cannot stand. However, we hold that constitutional error undermines petitioner’s convictions as well, and thus his present convictions cannot stand either. In Part I of this opinion, we address five procedural and threshold legal issues of general application. We must resolve these issues in order properly to analyze the merits of the parties’ arguments. First, we address the district court’s application of our decision in Walker v. Attorney General, 167 F.3d 1339 (10th Cir.1999). In Walker, we considered the effect of the 1995 amendments to Oklahoma statutes governing capital post-conviction proceedings and refused to apply those amendments retroactively for purposes of federal application of procedural bar. Here, we conclude that the district court read the Walker procedural-bar exception too broadly. With regard to many of petitioner’s claims, we conclude that the Walker exception is inapplicable and we must apply the traditional federal standards for determining whether a procedural default is excused. See infra Part I A. Second, we address the standard applied by the OCCA for assessing claims of ineffective assistance of appellate counsel. That standard is relevant because petitioner has argued that the omission of certain issues from his direct appeal resulted from ineffective assistance of his appellate counsel (which should excuse his procedural default) and because the OCCA rejected this argument on the merits. We conclude that the OCCA applied an incorrect standard and, as a result, we do not defer to its disposition. See infra Part I B. Third, we consider the effect of the OCCA’s review of some of petitioner’s constitutional claims under a state plain-error standard. We conclude that some of the state plain-error rulings constitute merits determinations under federal law while others reflect reliance on independent state grounds for disposition which warrant application of federal procedural-bar principles. We also clarify the circumstances in which we defer under AEDPA to state court determinations of federal issues on plain-error review. See infra Part I C. Fourth, we address important issues regarding petitioner’s claims of cumulative error. We conclude that prejudice may be cumulated among different kinds of constitutional error, such as ineffective assistance of counsel and prosecutorial misconduct. We further conclude that prejudice may be cumulated among such claims when those claims have been rejected individually for failure to satisfy a prejudice component incorporated in the substantive standard governing their constitutional assessment. Finally, we conclude that prejudice from guilt-phase error may be cumu-lated with prejudice from penalty-phase error. See infra Part I D and E. Fifth, we address whether it was proper under 28 U.S.G. § 2254(e)(2) for the district court to consider evidence that was not first presented to the state courts. As to testimonial evidence, we note that the State did not challenge the district court’s decision to hold an evidentiary hearing. As to new documentary evidence, we conclude that petitioner attempted to develop this material in the state court proceedings and, as a result, this evidence should not be barred on the ground that he failed to develop the factual basis of his claim in state court. See infra Part I F. With these threshold matters resolved, we turn to the merits in Part II. We conclude that petitioner received ineffective assistance of counsel at the guilt and penalty phases of trial, that the prosecution engaged in prejudicial misconduct at both phases of trial, and that the State relied on improper victim impact evidence to support the death penalty. We hold that petitioner is entitled to relief from his capital convictions based, individually, on ineffective assistance of counsel and, cumulatively, on the combined impact of this error and prosecutorial misconduct. We also hold that petitioner is entitled to relief from his death sentences based, individually, on ineffective assistance of counsel and, cumulatively, on the combined impact of all three errors cited immediately above. However, we reject in Part III petitioner’s claims that there was insufficient evidence to support his convictions and death sentences. Hence, we recognize that the State may retry and resentence petitioner within a reasonable time without constraint by double jeopardy concerns. These rulings render it unnecessary to address petitioner’s other claims. 1. PROCEDURAL AND THRESHOLD LEGAL RULINGS These appeals raise numerous issues, but our overall approach is framed by several procedural and threshold legal rulings of general application. A. LIMITED SCOPE OF WALKER EXCEPTION TO PROCEDURAL BAR The district court reached the merits of many claims defaulted by petitioner in state court, based on an overly broad application of the exception to state procedural bar recognized in Walker v. Attorney General, 167 F.3d 1339. In Walker, we refused to bar a claim that was defaulted under Oklahoma’s newly amended (in 1995) post-conviction scheme, when the procedural default occurred before the amended statute was enacted. We noted that the 1995 amendments made it harder in the state courts to raise new post-conviction claims based on intervening changes in law, and we declined to give federal effect to a state default of a claim that would not have been defaulted before the amendments. We reasoned: “A defendant ... should not be deprived of a claim for failing to comply with a rule [i.e., the more stringent duty to anticipate changes in law and urge currently foreclosed issues or face later waiver consequences] that only comes into being after the time for compliance has passed.” Walker, 167 F.3d at 1345; see Mitchell, 262 F.3d at 1047. However, if claims omitted on direct state appeal would have been barred on state post-conviction anyway, even under Oklahoma’s pre-1995 law (for example, if they rested on authority established at the time of direct appeal), it would not make sense to isolate the new 1995 standard as the operative cause of the default. Thus, we have not applied Walker to excuse petitioner’s default in failing to raise claims in his direct appeal that were at the time of the default clearly established. Here, the Walker exception to procedural default is inapplicable to many of petitioner’s claims, including ineffective assistance of counsel, prosecutorial misconduct, and State suppression of evidence. We will consider whether those claims should be resolved on the merits notwithstanding state procedural default, under the traditional federal standards for evaluating when state procedural default should be excused. B. OCCA’s DEPARTURE FROM FEDERAL STANDARD FOR ASSESSING INEFFECTIVE ASSISTANCE OF APPELLATE COUNSEL With respect to several issues omitted from his direct appeal, petitioner argues ineffective assistance of appellate counsel to excuse his default. See generally Murray v. Carrier, 477 U.S. 478, 106 S.Ct. 2639, 91 L.Ed.2d 397 (1986). Because the OCCA denied petitioner’s claims of ineffective assistance of appellate counsel on the merits, ordinarily our review would be limited to determining whether the petitioner can show that the state court’s disposition is “contrary to, or involved an unreasonable application of, clearly established Federal law, as determined by the Supreme Court of the United States.” 28 U.S.C. § 2254(d)(1); see, e.g., Ellis v. Hargett, 302 F.3d 1182, 1187 (10th Cir.2002). However, this deferential standard of review does not apply if the state court employed the wrong legal standard in deciding the merits of the federal issue. See, e.g., Revilla v. Gibson, 283 F.3d 1203, 1220 n. 14 (10th Cir.), cert. denied, — U.S. -, 123 S.Ct. 541, 154 L.Ed.2d 430 (2002). The OCCA applied the wrong standard here. The proper standard for assessing a claim of ineffectiveness of appellate counsel is that set forth in Strickland v. Washington, 466 U.S. 668, 104 S.Ct. 2052, 80 L.Ed.2d 674 (1984). Smith v. Robbins, 528 U.S. 259, 285, 120 S.Ct. 746, 145 L.Ed.2d 756 (2000) (following Smith v. Murray, 477 U.S. 527, 535-36, 106 S.Ct. 2661, 91 L.Ed.2d 434 (1986)). Thus, the petitioner must show both (1) constitutionally deficient performance, by demonstrating that his appellate counsel’s conduct was objectively unreasonable, and (2) resulting prejudice, by demonstrating a reasonable probability that, but for counsel’s unprofessional error(s), the result of the proceeding — in this case the appeal — would have been different. Id. at 285, 120 S.Ct. 746 (applying Strickland). A claim of appellate ineffectiveness can be based on counsel’s failure to raise a particular issue on appeal, although it is difficult to show deficient performance under those circumstances because counsel “need not (and should not) raise every nonfrivolous claim, but rather may select from among them in order to maximize the likelihood of success on appeal.” Id. at 288, 120 S.Ct. 746 (following Jones v. Barnes, 463 U.S. 745, 103 S.Ct. 3308, 77 L.Ed.2d 987 (1983)). Thus, in analyzing an appellate ineffectiveness claim based upon the failure to raise an issue on appeal, “we look to the merits of the omitted issue,” Neill v. Gibson, 278 F.3d 1044, 1057 (10th Cir.2001) (quotation omitted), cert. denied, — U.S. -, 123 S.Ct. 145, 154 L.Ed.2d 54 (2002), generally in relation to the other arguments counsel did pursue. If the omitted issue is so plainly meritorious that it would have been unreasonable to winnow it out even from an otherwise strong appeal, its omission may directly establish deficient performance; if the omitted issue has merit but is not so compelling, the case for deficient performance is more complicated, requiring an assessment of the issue relative to the rest of the appeal, and deferential consideration must be given to any professional judgment involved in its omission; of course, if the issue is meritless, its omission will not constitute deficient performance. See, e.g., Smith, 528 U.S. at 288, 120 S.Ct. 746; Banks v. Reynolds, 54 F.3d 1508, 1515-16 (10th Cir.1995); Mayo v. Henderson, 13 F.3d 528, 533 (2d Cir.1994). ' The OCCA, however, utilized a different analysis by applying a three-step approach for capital cases that it had drawn from Okla. Stat. tit. 22, § 1089, in Walker v. State, 933 P.2d 327, 333-35 (Okla.Crim.App.1997). Under this analysis, (1) the threshold inquiry is whether appellate counsel actually committed the act which gave rise to the ineffective assistance allegation. If a petitioner establishes appellate counsel actually did the thing supporting the allegation of ineffectiveness, this Court then (2) determines whether the performance was deficient under the first of the two-pronged test in Strickland v. Washington, 466 U.S. 668, 687, 104 S.Ct. 2052, 2064, 80 L.Ed.2d 674, 693 (1984). If this burden is met, (3) this Court then considers the mishandled substantive claim, asking whether the deficient performance supports a conclusion either that the outcome of the trial would have been different but for the errors or that the defendant is factually innocent. Cargle, 947 P.2d at 587-88 (quotation omitted). These three steps each perform very distinct functions. Step one is merely a fact finding step. Step two is where the court analyzes whether there was ineffective assistance of appellate counsel sufficient to excuse what would otherwise be procedural default. If ineffective assistance of appellate counsel is found in step two, the OCCA then proceeds to step three, where the omitted claim of trial error -is fully reviewed on the merits. Walker, 933 P.2d at 333-34 (“Under [§ 1089], [petitioners] must prove deficient performance as a precondition to having their underlying claim reviewed.”). Thus, for purposes of deciding whether to give federal deference under § 2254(d) to the OCCA’s finding of no ineffective assistance of appellate.counsel, we focus on’step two of this Walker test. As the OCCA acknowledged in Walker, the analysis of ineffective assistance of counsel at step two “does not include the traditionally applied two-pronged Strickland analysis,” because it eliminates the prejudice prong. Walker, 933 P.2d at 333 n. 23; see Le v. State, 953 P.2d 52, 54 (Okla.Crim.App.1998) (upholding constitutionality of § 1089 despite acknowledged departure from Strickland analysis). The issue before us, then, is whether the OCCA’s evaluation of appellate counsel’s performance in step two of its Walker analysis comports with constitutional standards as determined by the United States Supreme Court. The OCCA has stated that because § 1089 eliminated the Strickland requirement of prejudice, it therefore need not consider the merits of the omitted claim in evaluating whether appellate counsel’s performance was deficient. In Walker, the OCCA explained: Before the enactment of the new statute, post-conviction petitioners would simply argue their barred claims and summarily conclude that their appellate counsel’s decision not to raise them or manner of raising them constituted ineffective assistance under the two-pronged Strickland test. In applying the prejudice prong of the Strickland test to these allegations, this Court was essentially forced to examine the merits of the allegedly mishandled but technically waived claim in order to determine whether it was so serious as to deprive the defendant of a fair trial and thus would have forced a reversal or sentence modification on direct appeal. By eliminating a capital post-conviction petitioner’s burden to prove prejudice for ineffective assistance of appellate counsel allegations, the pivotal and narrow threshold issue is now simply whether appellate counsel’s performance was deficient under prevailing professional norms. This issue may be fully analyzed without examining the merits of the technically waived, substantive claim which was allegedly mishandled. Walker, 933 P.2d at 334 (emphasis added). Walker’s step-two truncation of the Strickland test thus enables the OCCA to reject appellate ineffectiveness allegations without any assessment of the merits of underlying predicate claims, so that the OCCA has been able to declare that a “failure to raise even a meritorious claim does not, in itself, constitute deficient performance.” Slaughter v. State, 969 P.2d 990, 996 (Okla.Crim.App.1998) (following Mitchell v. State, 934 P.2d 346, 350 (Okla.Crim.App.1997), and Walker, 933 P.2d at 336-37). The OCCA’s post -Walker case law consistently reflects its understanding that something over and above the merits of the omitted issue must be shown before a petitioner can establish ineffective assistance of appellate counsel. The OCCA has on several occasions rejected claims that procedural default should be excused because of ineffective assistance of appellate counsel by observing that, “beyond arguing the substantive merits of the [omitted] claim, [the petitioner] fails to present any facts showing why this omission constituted deficient performance.” Johnson v. State, 952 P.2d 1003, 1006 (Okla.Crim.App.1998); Bryan v. State, 948 P.2d 1230, 1234 (Okla.Crim.App.1997); see also Douglas v. State, 953 P.2d 349, 353 (Okla.Crim.App.1998) (invoking same principle with slightly different formulation); McGregor v. State, 935 P.2d 332, 336 (Okla.Crim.App.1997) (same). Since omission of a clearly meritorious claim is not, by that fact alone, sufficient under Oklahoma law to establish ineffective assistance of appellate counsel, and since the merits of the omitted claim need not even be considered in evaluating whether appellate counsel’s performance was deficient, we look to see what Oklahoma does require in order to find ineffective assistance of appellate counsel. What the OCCA requires to establish ineffective assistance of appellate counsel under step two of the Walker test is suggested by Powell v. State, 935 P.2d 378 (Okla.Crim.App.1997), which held that an ineffectiveness claim could not succeed where the petitioner failed to provide “any evidence why appellate counsel failed to raise the substantive claims [now] argued,” id. at 384 (emphasis added). This indicates that petitioners must establish not only a meritorious omitted issue but also an improper motive or cause behind counsel’s omission of the issue. See, e.g., Walker, 933 P.2d at 335-36 (rejecting ineffectiveness claim for lack of evidence that counsel “purposefully omitted meritorious appeal issues because he wished to avoid angering [the trial] Judge”). This appears to involve the very inquiry that the Supreme Court specifically repudiated in Murray, 477 U.S. at 487, 106 S.Ct. 2639, which rejected a test for appellate ineffectiveness that would have required courts “to determine what prompted counsel’s failure to raise the claim in question.” It is clearly wrong, as a matter of federal law, to require as a necessary condition for relief under Strickland, something beyond the obvious merit of the omitted claim. The very focus of a Strickland inquiry regarding performance of appellate counsel is upon the merits of omitted issues, and no test that ignores the merits of the omitted claim in conducting its ineffective assistance of appellate counsel analysis comports with federal law. A sufficiently meritorious omitted claim certainly can, by itself (or in relation to other issues that counsel did pursue), establish constitutionally deficient performance by appellate counsel. Because the OCCA’s analysis of petitioner’s appellate ineffectiveness allegations deviated from the controlling federal standard, see Cargle, 947 P.2d at 588-89 (repeatedly invoking principle contrasted with Strickland here), it is not entitled to deference. Of course, in this and in every case raising an ineffective appellate counsel issue, whether the OCCA decision should be accorded AEDPA deference will depend upon a case-specific determination of whether the OCCA followed established Strickland standards, in-eluding the principle that ineffective appellate assistance can be established on the basis of the demonstrable merit of the issue omitted by counsel on the petitioner’s direct appeal. C. STATE PLAIN ERROR REVIEW, PROCEDURAL BAR AND § 2254(D) DEFERENCE The OCCA reviewed some of the issues in this case for plain error. This raises threshold questions about the effect of state plain-error review on procedural bar and § 2254(d) deference principles. As for procedural bar, the question is: does a state court’s plain-error review of an issue otherwise waived, for lack of a trial objection constitute a merits decision under Harris v. Reed, 489 U.S. 255, 109 S.Ct. 1038, 103 L.Ed.2d 308 (1989), thus negating application of procedural bar, or does OCCA’s use of the heightened standard of plain error constitute the enforcement of a state waiver rule under Harris, thus necessitating application of procedural bar? Courts addressing this question have arrived at very different answers. As for § 2254(d), the question is: does a state court’s use of a plain-error standard affect the deference that the federal court owes to the state court’s determination? Case law explicitly addressing this question is scarce, though the Eighth Circuit appears to hold that state plain-error review is entitled to § 2254(d) deference. See James v. Bowersox, 187 F.3d 866, 869 (8th Cir.1999) (applying § 2254(d) after noting state appellate court had summarily rejected claim on plain-error review). In our view, the answer to both questions depends on the substance of the plain-error disposition. A state court may deny relief for a federal claim on plain-error review because it finds the claim lacks merit under federal law. In such a case, there is no independent state ground of decision and, thus, no basis for procedural bar. See Hux v. Murphy, 738 F.2d 737, 739 (10th Cir.1984) (following approach we adopt here, though without consideration of analytical complexities subsequently added by Harris and AEDPA). Consistent with that conclusion, the state court’s disposition would be entitled to § 2254(d) deference because it was a form of merits review. On the other hand, a state court could deny relief for what it recognizes or assumes to be federal error, because of the petitioner’s failure to satisfy some independent state law predicate. In such a case, that non-merits predicate would constitute an independent state ground for decision which would warrant application of procedural-bar principles on federal habeas. If the state procedural bar were then excused for some reason, the federal court would be left to resolve the substantive claim de novo, unconstrained by § 2254(d). See, e.g., McCracken v. Gibson, 268 F.3d 970, 977 (10th Cir.2001), cert. denied, — U.S. -, 123 S.Ct. 165, 154 L.Ed.2d 64 (2002); Hale v. Gibson, 227 F.3d 1298, 1328 (10th Cir.2000), cert. denied, 533 U.S. 957, 121 S.Ct. 2608, 150 L.Ed.2d 764 (2001). D. CUMULATIVE ERROR INVOLVING SEPARATE CATEGORIES OF ERROR As explained in detail below, petitioner has established many instances of constitutionally deficient performance by counsel, prosecutorial misconduct, and the admission of improper victim impact evidence at trial. In addition to claiming that each of these distinct categories of error gave rise to sufficient prejudice, in itself, to warrant habeas relief, petitioner argues that they provide an even more compelling basis for relief when their combined prejudicial effect is collectively assessed under cumulative-error principles. Petitioner exhausted this claim by asserting cumulative error on both direct appeal and post-conviction. The OCCA denied relief on appeal because the minor trial errors it recognized did not combine to create sufficient prejudice to warrant reversal, Cargle, 909 P.2d at 833-34, and it denied relief on post-conviction because the substantive claims raised by petitioner were rejected on procedural grounds, Cargle, 947 P.2d at 589. Neither of these decisions assessed the aggregate prejudice arising from the several constitutional errors we find here. Thus, our cumulative-error review is not restricted by § 2254(d)(l)’s limited focus on whether a state court decision was “contrary to, or involved an unreasonable application of, clearly established Federal law, as determined by the Supreme Court of the United States.” This limitation on federal habeas review applies only when there is an antecedent state court decision on the same matter. Lacking such a decision, we approach the question of cumulative error as we would have prior to AEDPA’s passage, applying the legal principles we conclude properly govern its disposition, whether they have been stated specifically in decisions of the Supreme Court or derived more generally from the federal case law. “A cumulative-error analysis aggregates all errors found to be harmless and analyzes whether their cumulative effect on the outcome of the trial is such that collectively they can no longer be determined to be harmless.” United States v. Toles, 297 F.3d 959, 972 (10th Cir.2002) (quotation omitted); see United States v. Rivera, 900 F.2d 1462, 1470 (10th Cir.1990) (en banc). Consistent with the unqualified reference to all errors, our cases reflect application of cumulative-error review to legally diverse claims such as those here. See, e.g., Toles, 297 F.3d at 972 (prosecuto-rial misconduct and improper limit on defense cross-examination); United States v. Wood, 207 F.3d 1222, 1237-38 (10th Cir.2000) (reversed conviction because of the cumulative effect of the improper denial of mid-trial acquittal on first and second degree murder and evidentiary error); Moore v. Reynolds, 153 F.3d 1086, 1113 (10th Cir.1998) (Brady error and prosecu-torial misconduct); United States v. Oberle, 136 F.3d 1414, 1423 (10th Cir.1998) (prosecutorial misconduct and evidentiary error). There is a further point to be made, arising from the fact that the particular types of error considered here are governed in the first instance by substantive standards which already incorporate an assessment of prejudice with respect to the trial process as a whole: Strickland errors require us to assess whether there is a reasonable probability that counsel’s deficient performance affected the trial outcome; Brady errors require us to look for the same reasonable probability that the trial outcome was affected in order to assess the necessary “materiality” of withheld evidence; and claims of prosecutorial misconduct and admission of improper victim impact evidence require a showing of fundamental unfairness in order to provide habeas relief, unless they involve the violation of specific constitutional rights, in which case the principles governing such rights control. These substantive prejudice components essentially duplicate the function of harmless-error review. Thus, such claims should be included in the cumulative-error calculus if they have been individually denied for insufficient prejudice. Indeed, to deny cumulative-error consideration of claims unless they have first satisfied their individual substantive standards for actionable prejudice “would render the cumulative error inquiry meaningless, since it [would] ... be predicated only upon individual error already requiring reversal.” Willingham v. Mullin, 296 F.3d 917, 935 (10th Cir.2002). E. CUMULATING GUILT-PHASE ERROR AND PENALTY-PHASE ERROR Petitioner did not partition his cumulative-error allegations into separate guilt and penalty-phase claims. Rather, he broadly asserted that “accumulation of all the trial errors entitled [him] to relief,” collectively claiming that these errors “re-sult[ed] in the first and second stage of [ ]his trial [being] unreliable.” Thus, his cumulative-error claim raises the possibility of guilt-phase error having a continuing, cognizable effect on the penalty phase. This commonsense notion that sentencing proceedings may be affected by errors in the preceding guilt phase is not novel. One especially clear affirmation of the principle, with respect to ineffective-assistance allegations, was recently voiced by the Fifth Circuit in Moore v. Johnson, 194 F.3d 586 (5th Cir.1999): The [State’s] argument may be reduced to the premise that deficient performance occurring at the guilt phase of a capital trial may not be deemed to prejudice a capital defendant during the punishment phase of a capital trial. We reject this notion. When, as here, the same jury considered guilt and punishment, the question is whether the cumulative errors of counsel rendered the jury’s findings, either as to guilt or punishment, unreliable. Id. at 619 (granting habeas relief from death sentence because counsel’s deficient guilt-phase performance prejudiced outcome of penalty phase); see also Smith v. Wainwright, 741 F.2d 1248, 1255 (11th Cir.1984) (granting evidentiary hearing on trial ineffectiveness claim because counsel’s failure to impeach witness during guilt phase “may not only have affected the outcome of the guilt/innocence phase, it may have changed the outcome of the penalty trial”). This court has likewise assessed guilt-phase errors, including prosecutorial misconduct and evidentiary error, for prejudicial impact on subsequent capital sentencing proceedings. See Coleman v. Saffle, 869 F.2d 1377, 1394-96 (10th Cir.1989). However, given the procedural and temporal distance of guilt-phase error from the penalty proceeding, we have noted that the chance or degree of carry-over prejudice to the penalty phase may be attenuated. See id. at 1396. Accordingly, our consideration of petitioner’s claim of error at the penalty phase may be cumulated with guilt-phase error, so long as the prejudicial effect of the latter influenced the jury’s determination of sentence. F. NEW EVIDENCE, § 2254(E)(2), AND WILLIAMSIMILLER EXCEPTION The district court held an evidentiary hearing to develop the facts relating to petitioner’s allegations of penalty-phase ineffective assistance. Ordinarily, a petitioner must satisfy the conditions imposed by 28 U.S.C. § 2254(e)(2) before being granted a hearing to augment the evidentiary basis for a claim. However, the State does not challenge the district court’s decision to hold the hearing, and, though we would reach the same result in any event, the factual clarification gained has certainly aided our informed review. Under the circumstances, we will not second-guess the appropriateness of the hearing. Bryan v. Gibson, 276 F.3d 1163, 1172 n. 6 (10th Cir.2001) (following Romano v. Gibson, 239 F.3d 1156, 1174 n. 9 (10th Cir.), cert. denied, 534 U.S. 1045, 122 S.Ct. 624, 151 L.Ed.2d 545 (2002)). Petitioner also supported some of his allegations, particularly those relating to guilt-phase ineffective assistance, with documentary evidence that he had not submitted in state court. While § 2254(e)(2) refers only to evidentiary hearings, it governs as well “[w]hen expansion of the record is used to achieve the same end as an evidentiary hearing.” Boyko v. Parke, 259 F.3d 781, 790 (7th Cir.2001); see Dorsey v. Chapman, 262 F.3d 1181, 1190 (11th Cir.2001), cert. denied, 535 U.S. 1000, 122 S.Ct. 1567, 152 L.Ed.2d 489 (2002). In this regard, however, petitioner requested the opportunity to develop relevant evidence through discovery and an evidentiar ry hearing in state court, which the OCCA summarily refused after rejecting his ineffectiveness claims as.procedurally defaulted (on what we explain infra were inadequate state grounds). See Cargle, 947 P.2d at 590. Thus, .petitioner did not fail “to develop the factual basis” of his claim in state court, as .that threshold condition in § 2254(e)(2) has been construed by this circuit and the Supreme Court. See Williams, 529 U.S. at 435-37, 120 S.Ct. 1479 (explaining “fault component” in statutory condition); Miller v. Champion, 161 F.3d 1249, 1253 (10th Cir.1998) (presaging Williams). Therefore, petitioner properly supported his allegations with the additional evidence submitted to the district court. II. MERITS ANALYSIS A. OVERVIEW Petitioner challenges both his convictions and sentences. We conclude he is entitled to relief in both respects on the basis of ineffective assistance of trial counsel and, alternatively, on the basis of cumulative error combining ineffective assistance of counsel with instances of pros-ecutorial misconduct and the use of improper victim impact evidence. We begin with a brief overview of coun-' sel’s grossly deficient efforts and the unusual external pressures upon counsel, as well as the weaknesses in the State’s case which would have rendered it particularly vulnerable to true adversarial testing. We will then take up the parties’ specific legal contentions regarding the errors which lead us to grant petitioner habeas relief from both his convictions and sentences. Petitioner’s parents retained counsel in November 1993 when, unbeknownst to them, counsel was embroiled in bankruptcy proceedings and a professional grievance investigation against him. By the time of petitioner’s preliminary hearing in February 1994, counsel had become the target of a criminal tax probe as well. Within a year and a half, counsel had been convicted on a federal tax charge and had resigned from the bar. The strain of these overlapping pressures on counsel is evident. While his grievance case was set, continued, and reset throughout the spring of 1994, he never visited petitioner at the county jail. Petitioner’s parents came to see. counsel a couple of times, but these visits primarily concerned matters pertaining to counsel’s fee. Days after yet another setting of the grievance case was continued on May 27— and just four weeks before petitioner’s trial was scheduled to commence — counsel finally requested a transcript of petitioner’s preliminary hearing, went to see him at the jail for the first time, and moved for a continuance in petitioner’s case, conceding he could not be prepared to try the case by the June 20 trial date. When the motions for continuance and for the preliminary hearing transcript came on for hearing on June 2, however, counsel failed to appear and the motions consequently were denied. Two weeks later, counsel visited petitioner for the second and last time at the jail. Petitioner claims counsel’s desperate advice at this eleventh-hour meeting was for him to feign incompetency when he appeared for trial the next week and, indeed, counsel moved for a competency determination on the first day of trial. The transcript suggests a halfhearted and promptly-abandoned acquiescence in this scheme by petitioner, who was quickly determined competent before voir dire began. The last-ditch efforts to forestall trial were undoubtedly driven by counsel’s woeful state of preparation. The only potential witnesses he interviewed were petitioner and his parents, and the value of these typically crucial discussions with the accused and his family was greatly limited by time (counsel evidently spent less than an hour with his client before trial) and counsel’s failure to engage himself with the substance of the case. Petitioner and his parents said counsel never discussed trial strategy with them, instead giving them generic assurances that he had never lost a case. He made little or no effort to draw exculpatory or mitigation evidence from them, or to develop and follow leads to other sources of such information. Indeed, counsel did not even explain the two-stage process of capital prosecution, leaving them ignorant in particular about the penalty phase and the nature of mitigation evidence. Counsel thus essentially foreclosed any helpful disclosures from those most likely to know, first-hand, the pertinent facts. And in the end, none of these often critical witnesses — people whom jurors understandably expect to hear from in the penalty phase in every death penalty case — took the stand. Counsel offered just two witnesses for the defense, one in the guilt phase and one in the penalty phase, and neither presentation bespeaks professional trial preparation. In the guilt phase, petitioner’s longtime friend, Steven Butler, testified that Chris Jackson told him Todd Williams shot both of the victims, contrary to Jackson’s trial testimony that petitioner shot Sharon Paisley. Significantly, it was Butler who contacted counsel just before trial to offer this testimony. This last-minute, fortuitous witness (who provided a mere seven pages of direct testimony), constituted the entire defense case in the guilt phase. The penalty-phase defense appears to have been, literally, an after-thought. Counsel, who had not prepared anyone to testify at this stage of trial, waived an opening statement. After the State completed its case in aggravation, counsel turned to the courtroom crowd and secured his only witness when the pastor at petitioner’s church offered to take the stand. The pastor’s brief, unprepared, personally remote, and fairly generic testimony was “more like [a] first-time interview[ ] conducted during discovery than the presentation of a defense at a capital murder trial.” Fisher v. Gibson, 282 F.3d 1283, 1294 (10th Cir.2002). Significantly, neither petitioner nor his parents testified. Indeed, from testimony at the habeas hearing found credible by the district judge, it appears counsel actively deterred petitioner’s parents from testifying, presumably to cover his blatant lack of preparation, when it finally became apparent that they might do so. Further, counsel arguably lied to the trial judge about petitioner’s alleged decision not to testify on his own behalf. Counsel told the judge, in chambers without petitioner present, that petitioner had, in front of his father and pastor, elected not to testify at the second stage. The habeas testimony of petitioner, his father, and the pastor all flatly contradicted the story counsel had told the trial judge, and the relevant portions of the state trial record contain nothing to suggest the incident counsel described had ever taken place. In all, counsel’s gross mishandling of the penalty-phase defense left his client’s fate to jurors who could only wonder why neither the man himself nor any member of Ms family would step up to explain, in personal human terms, why Ms life should be spared notwithstanding the reprehensible conduct of which he had been found guilty. Counsel’s failure to prepare any significant guilt or penalty-phase defense case, as dammng as that is, reflects just part of the meffectiveness claim here. Counsel also critically failed to investigate and effectively attack the State’s case, particularly with regard to the two absolutely critical, yet vulnerable, witnesses upon which the State’s case depended. The State had no physical evidence showrng that petitioner had shot either of the victims or that he had aided and abetted either murder. The State also had no eyewitnesses other than the three men involved in the crime, and one of them, Todd Williams, did not appear at petitioner’s trial. The other participant, Chris Jackson, testified he saw petitioner shoot Sharon Paisley. As a result, the State’s capital case against petitioner rested on Chris Jackson’s shoulders — a precarious position, m that Jackson himself had been charged in the murders until he deflected guilt onto petitioner (and, in the process, secured immunity from aiding and abetting charges). In essence, the case came down to a credibility battle between Jackson and petitioner, requiring the State to convince the jury beyond a reasonable doubt that Jackson was the one telling the truth. The other critical State witness was a third party, Luke Jones, -who was not present at the murders but who testified that petitioner had told him that he (petitioner) shot and killed Sharon. However, Jones’ testimony, like Jackson’s testimony, was highly impeachable by any competent lawyer. Nevertheless, petitioner’s counsel did virtually nothing to prepare an effective cross-examination to test their credibility. Significant grounds for such impeachment existed, both as to the content of their testimony and as to the consideration they received in exchange for their testimony. But, through counsel’s inaction, and partly also through the State’s misrepresentation, counsel simply did not have these means of impeachment at his disposal during trial. In sum, as we explain more fully below, counsel did not properly confer with and advise his client, did not interview and prepare even obvious defense witnesses, did not develop available exeulpatory/miti-gatmg evidence, did not investigate sources for impeachment of the crucial prosecution witnesses, disloyally impeached his own client, and failed to expose the State’s relatively weak case to meaningful adversarial testing. Petitioner was denied effective counsel on the capital charges agamst him. B. GUILT PHASE ERRORS 1. Ineffective Assistance of Counsel a) Standard of Review Some of petitioner’s allegations of trial ineffectiveness were raised for the first time in his federal petition, and we do not consider these in granting habeas relief. Most of the allegations we are concerned with were raised in petitioner’s application for post-conviction relief. The OCCA summarily rejected these claims of attorney ineffectiveness because they were not raised on direct appeal. However, these claims require consideration of matters outside the trial court record and we have repeatedly held under such circumstances that Oklahoma’s procedural bar rule is inadequate to bar federal review. See Revilla v. Gibson, 283 F.3d at 1220; Sallahdin v. Gibson, 275 F.3d 1211, 1234 (10th Cir.2002); Romano, 239 F.3d at 1179. Because the OCCA did not address the merits of these allegations, our review is not constrained by the deference principles in § 2254(d). See Revilla, 283 F.3d at 1220; Sallahdin, 275 F.3d at 1234-35. We also consider one ineffectiveness claim that was rejected on its merits in state court. The OCCA held counsel’s failure to obtain the preliminary hearing transcript “fell below acceptable levels of professionalism,” but denied relief because petitioner could not show that this error would likely have had an effect on the outcome of the proceeding. Gargle, 909 P.2d at 833. Because this prejudice determination was neither contrary to nor an unreasonable application of Strickland, we cannot grant relief on this error alone. However, our decision to grant relief on ineffective assistance grounds is a function of the prejudice flowing from all of counsel’s deficient performance — as Strickland directs it to be. See Strickland, 466 U.S. at 694-96, 104 S.Ct. 2052 (repeatedly stating prejudice inquiry in aggregate terms of reasonable probability counsel’s errors affected outcome of proceeding); see Fisher, 282 F.3d at 1307-11 (assessing prejudice from counsel’s “numerous shortcomings [and] omissions,” and holding “these errors” had a “devastating impact on the defense”); Turner v. Duncan, 158 F.3d 449, 457 (9th Cir.1998) (“it is appropriate to consider the cumulative impact of [counsel’s] errors in assessing prejudice”). Given the OCCA’s procedural rejection of nearly all of petitioner’s allegations of ineffectiveness, an adequate assessment of prejudice arising from the ineffectiveness of petitioner’s counsel has never been made in the state courts, so we have no state decision to defer to under § 2254(d) on this issue. b) Failure to Interview and Call Witnesses Counsel failed to speak with many obvious potential witnesses. Even a rudimentary investigation would have involved contacting, for example, those people who were with petitioner and Jackson at petitioner’s uncle’s house when they left to go to the Paisleys’ home and when they returned. In this group were petitioner’s aunt, Dewonna Cargle, and Jackson’s girlfriend, Angel Harris. Both of these women would have contradicted various aspects of Jackson’s “innocent” account of his conduct that cast blame instead on petitioner. Indeed, these two women would have testified that immediately upon his return from the Paisleys’, Jackson bragged about having killed one or both of the victims himself. Jackson evidently told yet another story about the murders to Mark Gaddis, another possible witness that counsel never pursued, even though the application for petitioner’s arrest warrant referred to an informant (Mark Gaddis), who later testified at Todd Williams’ trial. The prosecution did not call Gaddis as a witness, and defense counsel flatly ignored the matter. Gaddis was contacted in connection with these habeas proceedings, however, and gave a much cloudier picture of what Jackson had said about the murders. Gaddis revealed that Jackson made the much more equivocal statement that Williams had made either him (Jackson) or petitioner shoot Sharon. Further, after it became known he was to be a prosecution witness, Jackson assured Dewonna Cargle he would straighten it out and not let petitioner go down for something he did not do. Even more telling was Jackson’s alleged pretrial offer to petitioner’s parents, which they recounted in affidavits and at the habeas hearing, that he would not testify against petitioner if they paid him money. Thus, at least five potential witnesses stated that, before trial, Jackson had indicated his testimony would be materially different than it turned out to be, had suggested that his testimony remained open to influence and alteration, and had made statements flatly contradicting his testimony. All of this could have been discovered if counsel had properly interviewed these witnesses before trial, but he did not do so. A similar story of dereliction surrounds the State’s other key witness, Luke Jones. Luke Jones testified that petitioner had confessed to him that he had killed Sharon Paisley, while the two of them were staying with petitioner’s mother shortly after the murders. However, Jones’ credibility was open to impeachment through an obvious source — his wife, Tisha Harris, who was the victim of repeated protection-order violations for which Jones was confined when he offered his information against petitioner. Jones told Tisha Harris he was not worried about the protection-order violations because he had information to offer the authorities about a murder that petitioner had witnessed. Then, in direct contradiction of his later trial testimony, Jones expressly exonerated petitioner, telling Tisha Harris that petitioner said he had not killed anyone. Counsel, however, never interviewed Harris and, thus, was unaware of this impeachment evidence, which certainly would have amplified suspicions roused by what the jury knew of the consideration given by the State in exchange for Jones’ appearance at trial. Over and above the incremental benefit each of these six witnesses would have added to the defense in impeaching the government’s two central witnesses (Jackson and Jones), there is the larger point that they could have, collectively, provided an effective overall defense strategy (particularly in a case resting almost entirely on the credibility of these two inherently vulnerable prosecution witnesses) that counsel utterly failed to see, much less effectively employ: showing that the case involved such a tangle of inter- and intra-witness inconsistency that the jury could not be confident enough in any person’s word to justify holding petitioner responsible for first degree murder beyond a reasonable doubt. With the kind of trial preparation expected of any competent counsel in a death case, counsel could have left the jury faced with: (1) Angel Harris, Dewonna Cargle, Gaddis, and Butler, all contradicting Jackson’s trial testimony with his own prior inconsistent statements on several key points, including attribution of guilt for the murders to himself and/or Williams; (2) petitioner’s parents and Dewonna Cargle indicating that Jackson had held out his trial testimony as a matter subject to influence and change; (3) Gaddis and Butler relating petitioner’s accounts of his own innocence; and (4) Tisha Harris’ impeachment of Jones with his prior statement to her expressing petitioner’s innocence. Instead of exposing this Gordian knot of inconsistencies and lies, counsel effectively acquiesced in the case being tried as a neat three-act play: Jackson and Jones establishing petitioner’s guilt; Butler, alone, trying to save his friend by insisting Jackson had told him petitioner was innocent; and Jackson’s parents brought out at the end to confirm that Jackson had been telling a consistent story ever since the murders occurred. There is no plausible reason other than counsel’s self-inflicted ignorance for allowing the prosecution to control the presentation in this way and, in our view, the difference in these scenarios reflects a persuasive case for Strickland prejudice. There are, moreover, further omissions by counsel to consider. c) Failure to Impeach Jackson Based upon Undisclosed Bias and State Inducement for his Testimony Counsel failed to inquire into a deferred sentence pending (and not accelerated) against Jackson at the time of trial and counsel failed to discover and present to the jury the true extent of the immunity afforded Jackson when he agreed to testify against petitioner. Jackson’s written agreement with the district attorney was admitted at trial. It granted Jackson “immunity from prosecution for the crimes of Murder in the First Degree involving the deaths of Richard and Sharon Paisley and Accessory to Murder involving the deaths of Richard and Sharon Paisley.” The document affirms that it reflects “the entire agreement” between Jackson and the district attorney. However, there evidently was an additional, and quite significant, quid pro quo for Jackson’s cooperation, which was not recited in the agreement and never disclosed to the jury. In 1992, Jackson had been charged for assault with a deadly weapon, carrying a maximum twenty-year sentence. In exchange for his guilty plea to that offense, sentencing was deferred for five years. Thus, if Jackson had any other state law violation during this five-year period, he could be sentenced immediately on the assault offense. Jackson’s admitted conduct in this case (assisting in wiping off fingerprints at the scene and agreeing with the others not to tell the authorities anything) made him a potential accessory under Okla. Stat. tit. 21 § 173, thereby exposing him to an immediate sentence of up to twenty years’ imprisonment on the pre-existing assault conviction in addition to the uncertain charges that could have been brought against him in this case. When asked by an investigator, in connection with this habeas proceeding, whether non-acceleration of the deferred sentence had been a (tacit) part of his agreement to testify in petitioner’s case, Jackson admitted that he had received an assurance from the district attorney that nothing would come up in court about the deferred sentence. Counsel knew about Jackson’s deferred sentence. Any competent attorney would have discerned the legal connection between Jackson’s conduct in this case and the conditions sufficient to trigger acceleration of the deferred sentence. Quite apart from the tacit quid pro quo assurances, Jackson’s mere exposure to this punitive threat was pertinent to the jury’s assessment of his motivation for testifying. Like a pending criminal charge or possible probation violation, this threat was “relevant to show pro-government bias on the part of the testifying witness, on the theory that the witness might tailor [his] testimony to please the prosecutor.” Stephens v. Hall, 294 F.3d 210, 224 (1st Cir.2002); see Davis v. Alaska, 415 U.S. 308, 311, 94 S.Ct. 1105, 39 L.Ed.2d 347 (1974). Yet counsel erroneously confessed a pretrial motion in limine preventing the defense “from mentioning, referring to, inferring or in any way informing the jury” about Jackson’s deferred sentence. Thus, counsel abandoned an important avenue of impeachment as to the State’s star witness. This involves a dimension of prejudice distinct from the prior-inconsistent-statement evidence noted earlier in Part II B 1(b): “A colorable showing of bias can be important because, unlike evidence of prior inconsistent statements— which might indicate that the witness is lying — evidence of bias suggests why the witness might by lying.” Hall, 294 F.3d at 224. And such impeachment “ ‘increases in sensitivity in direct proportion to [the] witness’s importance’ to [the] state’s case.” Jones v. Gibson, 206 F.3d 946, 955-57 (10th Cir.2000) (discussing counsel’s inability to question witness regarding pending charges against her). The State may permissibly offer certain forms of advantageous treatment, such as forbearance on potential charges or favorable recommendations as to sentencing, to secure the cooperation of a witness. See generally United States v. Singleton, 165 F.3d 1297 (10th Cir.1999) (en banc). Critical to the sanction of this practice, howev- er, are certain procedural safeguards, “prohibiting the government’s deliberate use of perjured testimony, requiring the government to timely disclose the [terms of witness agreements], and providing [the defense] an adequate opportunity to cross-examine the witnesses about those agreements.” United States v. Fria Vazquez Del Mercado, 223 F.3d 1213, 1215 (10th Cir.), cert. denied, 531 U.S. 1027, 121 S.Ct. 600, 148 L.Ed.2d 513 (2000); see also Romano, 239 F.3d at 1174 (“The government must disclose any understanding or agreement it has with its witnesses.”). A conviction based on testimony implicating concealed incentives to an important witness is potentially tainted. See Giglio v. United States, 405 U.S. 150, 154-55, 92 S.Ct. 763, 31 L.Ed.2d 104 (1972); Carriger v. Stewart, 132 F.3d 463, 479-82 (9th Cir.1997) (reversing conviction for nondisclosure of evidence relating to witness agreement and explaining that “criminals who are rewarded by the government [by being granted immunity] for their testimony are inherently untrustworthy”). While challenges in this regard are typically advanced in connection with allegations of misconduct by the prosecution, they may also be advanced, as here, in connection with a claim that petitioner’s “trial attorney was ineffective for failing to investigate adequately whether [a witness] had a deal with prosecutors.” Romano, 239 F.3d at 1173 n. 7; see also Hall, 294 F.3d at 214. The prejudice flowing from such counsel error is magnified here by the crucial nature of Jackson’s testimony to the State’s case. d) Counsel’s Failure to Challenge the State’s Bolstering of Jones’ Testimony through Mitchell Jones was the State’s other critical witness, because he testified petitioner had confessed to him about killing Sharon Paisley. The State apparently tried to bolster the credibility of Jones’ testimony by emphasizing its originality (and, hence, freedom from influence). Detective Ron Mitchell told the jury that before Jones came forward with his account of the murders, Mitchell did not know how the murders had occurred and had never even heard of petitioner, Williams, or Jackson. The district court found Mitchell’s representations of such prior ignorance incredible. We agree. Weeks before Jones offered any information, Gaddis had already spoken with the authorities, relating what petitioner had told him about the murders and identifying all three participants. As Mitchell was the detective in charge of the investigation, to suggest that Mitchell did not know those details when Jones first talked to authorities would have been ludicrous. Indeed, on the same day Jones talked to the State, Mitchell prepared an arrest warrant based on information from Jones and Gaddis. Even minimal effort on counsel’s part, such as reading the arrest warrant for his client, would have revealed enough of these facts to permit pointed impeachment of Mitchell’s professed ignorance of those facts before learning of them from Jones. Nevertheless, the district court discounted the matter, concluding there was no reasonable probability that impeaching Mitchell would have changed the trial outcome. Admittedly, Mitchell himself did not play an important role in the State’s case. Yet this impeachment would have shown the jury that even the police testimony in this case may not be believed, making the jury’s task of discerning the truth of petitioner’s guilt beyond a reasonable doubt even less attainable. Further, for the jury to see the police as lying on this point would have destroyed the State’s effort to bolster Jones — and left the jury understandably suspicious of this mercenary criminal informant whose dubious credibility had prompted such an improper vouching effort. e)Counsel’s Disparagement of his own Client Petitioner’s counsel failed to pursue a defense based on a severe injury to petitioner’s right hand that occurred less than two months before the shooting and instead suggested to the jury that petitioner (his own client) had lied about the significance of the injury during his videotaped police interview. While we agree with the district court that counsel’s decision not to press the issue of petitioner’s hand injury as an affirmative defense could be justified as a matter of defense strategy, there is disloyalty and impeachment by counsel of his own client implicated here which is not obviated by that strategy. When counsel mocked petitioner’s complaints of a medically established injury to his hand as an “old phony” ruse, counsel demonstrated a “fundamental violation of his duty of loyalty by exhibiting actual doubt and hostility toward his client’s case.” Fisher, 282 F.3d at 1300. In essence, counsel gratuitously called his client a bar, in a case that was all about credibility (though petitioner did not take the stand at trial, the jury was shown the police interview in which he denied committing the murder). “We can neither discern nor conceive of a trial strategy that would justify this treatment of one’s own client, which so clearly ca