GILMAN, J., delivered the opinion of the court, in which CLAY, J., joined. McKEAGUE, J. (pp. 484-93), delivered a separate dissenting opinion.
OPINION
RONALD LEE GILMAN, Circuit Judge.A Michigan state jury convicted Robert Ferensic in 1999 of armed robbery, home invasion, and possession of a firearm during the commission of a felony. The entirety of the evidence against Ferensic was based upon eyewitness identifications made by the victimized couple, Alexander and Angie Kostoff. Ferensic appealed, arguing among other things that (1) the trial court had violated his right to present a defense by preventing two of his witnesses — Dr. Harvey Shulman, an expert on eyewitness identification, and Danny St. John, who had observed the robbers prior to their entering the Kostoffs’ home — from testifying, and (2) his counsel had been constitutionally ineffective in failing to ensure that these two witnesses were allowed to testify. The Michigan Court of Appeals upheld Ferensic’s convictions, essentially reasoning that the nonappearance of both Dr. Shulman and St. John, whether attributable to the actions of the trial judge or of defense counsel, did not prejudice Feren-sic.
Ferensic subsequently petitioned the federal district court for a writ of habeas corpus, again raising the two grounds mentioned above. Having determined that the Michigan Court of Appeals’s ruling on each ground constituted an unreasonable application of clearly established federal law, the district court conditionally granted Ferensic’s petition. The Warden now appeals. For the reasons set forth below, we AFFIRM the judgment of the district court.
*471I. BACKGROUND
The Michigan Court of Appeals laid out the following relevant facts and procedural history of this case in People v. Ferensic, No. 221806, 2001 WL 865089, at *1-2 (Mich.Ct.App. July 31, 2001):
The victims, an elderly, married couple, identified defendant as one of two perpetrators who broke into their house and robbed them of money and other valuables at gunpoint. After describing defendant to the investigating officers, the victims met with a police sketch artist, who sketched composites of the perpetrators based on the victims’ descriptions. One of the officers in charge of the investigation recognized defendant, a former police officer, and defendant’s roommate from these sketches, and compiled a photographic array that included defendant. Only one of the victims was able to identify defendant from the photographic array. However, both victims identified defendant from a live lineup and at defendant’s preliminary examination. The victims’ identification of defendant, along with the officer’s testimony that he recognized the two individuals based on the sketch artist’s renderings, was the only evidence that defendant was involved in the offenses.
In denying the prosecution’s relevant motion in limine six months prior to defendant’s trial, the court ruled that the expert [Dr. Shulman] would be permitted to testify so long as defendant furnished a copy of the expert’s report to the prosecution two months before trial. In violation of that order, defendant mailed a copy of the expert’s report to the prosecution only eleven days before trial. Following the parties’ opening statements, in which each party emphasized that the identity of the perpetrators was the central issue in the case and in which defendant repeatedly told the jurors that defendant would present expert testimony on the inherent unreliability of eyewitness testimony, the prosecution successfully moved to exclude defendant’s expert witness from testifying. The court reasoned that the prosecution was unable to retain its own expert witness on identification without delaying the trial, and that although defendant mailed the report on the same day he received it, it had been his responsibility to chase up the expert to ensure compliance with the order.
(Footnote omitted.) Later in the trial, after the defense had rested, the court also denied Ferensic’s separate motion for a brief adjournment to allow time for St. John to arrive at the courthouse and take the stand.
Both Dr. Shulman and St. John testified during an evidentiary hearing held by the district court in connection with Ferensic’s habeas petition. (Ferensic’s trial counsel also was a witness at the hearing, but his testimony is not directly relevant to this appeal.) The court provided the following summary of their testimony:
Danny St. John testified at the eviden-tiary hearing that he did not know Petitioner and never saw him before. Similar testimony at trial might have caused the jurors to question the Kostoffs’ testimony even though Mr. St. John could describe only one of the two men he saw. In addition, his description of one of the robbers as having black curly ham was consistent with the original description given by Mr. and Mrs. Kostoff, which they denied at trial.
Dr. Shulman testified at the evidentiary hearing that several factors, such as divided attention, stress, passage of time, photo arrays, collaboration of witnesses, and social conformity, affect memory. He noted that crime produces stress, *472which makes high resolution (detailed) memory of faces difficult. He opined that the guns carried by the robbers in this case were the most distracting factor, because a person’s attention is directed to a weapon. “Sticky attention” to a weapon reduces the ability to recall details and leads to inaccurate identification, according to Dr. Shulman. In addition, gaps in memory are filled in by world knowledge, post-event information, inferences, and talking to other witnesses. Eyewitnesses find it difficult to disregard these influences on memory. Dr. Shulman also stated that there is no statistical relationship between a witness’s level of confidence in his or her identification and the accuracy of memory.
II. ANALYSIS
A. Standard of review
Under the Antiterrorism and Effective Death Penalty Act of 1996 (AEDPA), Pub. L. No. 104-132, 110 Stat. 1214, a federal court
may not grant a writ of habeas [corpus] to a petitioner in state custody with respect to any claim adjudicated on the merits in state court unless (1) the state court’s decision “was contrary to, or involved an unreasonable application of, clearly established Federal law, as determined by the Supreme Court,” ... or (2) the state court’s decision “was based on an unreasonable determination of the facts in light of the evidence presented in the State court proceedings.”
Taylor v. Withrow, 288 F.3d 846, 850 (6th Cir.2002) (quoting 28 U.S.C. § 2254(d)). This standard requires the federal courts to give considerable deference to state-court decisions. Herbert v. Billy, 160 F.3d 1131, 1135 (6th Cir.1998) (“[AEDPA] tells federal courts: Hands off, unless the judgment in place is based on an error grave enough to be called unreasonable.”) (quotation marks omitted).
The first line of analysis under AEDPA focuses on the consistency of the state-court decision with existing federal law. A state-court decision is considered “contrary to ... clearly established Federal law” if it is “diametrically different, opposite in character or nature, or mutually opposed.” Williams v. Taylor, 529 U.S. 362, 405, 120 S.Ct. 1495, 146 L.Ed.2d 389 (2000) (emphasis and quotation marks omitted). Alternatively, to be found an “unreasonable application of ... clearly established Federal law,” the state-court decision must be “objectively unreasonable,” not simply erroneous or incorrect. Id. at 409-11, 120 S.Ct. 1495. If a state-court decision meets either of these two “preconditions” for habeas relief — thereby establishing a constitutional error — the reviewing federal court must still determine whether the error is harmless within the meaning of Brecht v. Abrahamson, 507 U.S. 619, 113 S.Ct. 1710, 123 L.Ed.2d 353 (1993). Fry v. Pliler, — U.S. -, 127 S.Ct. 2321, 2327-28, 168 L.Ed.2d 16 (2007) (noting that AEDPA “sets forth a precondition to the grant of habeas relief ..., not an entitlement to it,” and that “in § 2254 proceedings a [federal] court must assess the prejudicial impact of constitutional error in a state-court criminal trial under the ‘substantial and injurious effect’ standard set forth in Brecht ”). Brecht applies “whether or not the state appellate court recognized the error and reviewed it for harmlessness under the ‘harmless beyond a reasonable doubt’ standard set forth in Chapman [v. California], 386 U.S. 18[, 24, 87 S.Ct. 824, 17 L.Ed.2d 705 (1967) ].” Id. at 2328.
The second line of analysis under AED-PA examines the findings of fact made by the state courts. AEDPA requires federal courts to accord a high degree of defer*473ence to such factual determinations. “A federal court is to apply a presumption of correctness to state court findings of fact for habeas corpus purposes unless clear and convincing evidence is offered to rebut this presumption. The appeals court gives complete deference to the federal district court’s and state court’s findings of fact supported by the evidence.” McAdoo v. Elo, 365 F.3d 487, 493-94 (6th Cir.2004) (citations omitted).
B. The Michigan Court of Appeals’s decision
The Michigan Court of Appeals was the last state court to review Ferensic’s claims during postconviction proceedings. Because its review was on the merits, we must apply AEDPA deference to that decision. The Court of Appeals first held that the trial court had not abused its discretion in prohibiting Dr. Shulman from testifying as a witness for Ferensic. Ferensic, 2001 WL 865089, at *2. Specifically, the Court of Appeals noted that, although this prohibition was a “concededly severe sanction” for Ferensic’s failure to comply with the trial court’s production order, it “was nothing more than the court had warned of six months before trial.” Id. at *2. In addition, the Court of Appeals concluded that Ferensic was not prejudiced by the exclusion of Dr. Shulman’s testimony “because defense counsel was otherwise able to effectively challenge inconsistencies in the victims’ identification testimony.” Id.
The Court of Appeals also held that the trial court had not abused its discretion in denying Ferensic’s motion for a brief adjournment to allow St. John time to arrive at the courthouse and testify. Essentially, the Court of Appeals determined that Fer-ensic had not been prejudiced by St. John’s absence because “[t]he purported testimony was not especially strong, ... and inconsistency inherent in the victims’ identification was otherwise shown.” Id.
The Court of Appeals used similar reasoning — -that is, an absence of prejudice— to reject Ferensic’s ineffective-assistance-of-counsel claim. “Although counsel’s failure to secure the testimony of these two witnesses was objectively unreasonable, defendant has not shown that absent this deficient performance the outcome of his trial would have been different.” Id. This conclusion derived from the Court of Appeals’s determination that “[t]he victims’ identification of defendant was effectively attached [sic] through other means, and standing alone, the testimony of the officer who recognized defendant from the sketch artist’s rendering would arguably have been sufficient to convict.” Id.
C. The district court’s habeas review
The district court agreed with the Michigan Court of Appeals that “defense counsel was able to challenge inconsistencies in the victims’ identification testimony” even without the two excluded witnesses. But the district court concluded that these alternative challenges “were not an effective substitute” for what the two witnesses, especially Dr. Shulman, would have provided. The district court explained its reasoning as follows:
Identity was the primary issue. Dr. Shulman would have demonstrated how an eyewitness’ memory and recognition are not as reliable as one might initially believe. Without his testimony there was no evidence to support counsel’s argument [regarding the inherent unreliability of eyewitness identifications].
Mr. St. John also would have helped the defense, because his description of a suspect with curly black hair matched the description that the Kostoffs gave to the police. At trial, the Kostoffs denied telling the police that Petitioner had curly black hair. Therefore, Mr. St. John’s *474testimony might have led the jurors to believe that the Kostoffs tailored their trial testimony to fit Petitioner’s actual appearance. In addition, Mr. St. John testified at the evidentiary hearing that he had never seen Petitioner before the hearing.
Ferensic, in the district court’s opinion, was therefore “deprived of a substantial defense by the exclusion of Danny St. John and Dr. Harvey Shulman,” and the conflicting decision of the Michigan Court of Appeals was “contrary to, or an unreasonable application of, clearly established federal law as determined by the Supreme Court.”
Regarding Ferensic’s related ineffective-assistance-of-counsel claim, the district court agreed with the Michigan Court of Appeals that Ferensic’s attorney had in fact been deficient in failing to secure both Dr. Shulman’s and St. John’s testimony. But unlike the Court of Appeals, the district court concluded that this deficient performance had prejudiced Ferensic. The district court emphasized that “no physical evidence linked Petitioner to the crimes” and that, instead, the entirety of the state’s case against him was based on multiple eyewitness identifications. Both Dr. Shulman’s and St. John’s testimony, moreover, “might have led the jurors to question the Kostoffs’ identification testimony.” This possibility was enough, in the district court’s opinion, “to undermine confidence in the outcome of [Ferensic’s] trial.”
The district court also pointed out that the Michigan Court of Appeals’s no-prejudice determination rested in part on a faulty legal premise. As noted above, the Court of Appeals concluded not only that the victims’ identification of Ferensic was effectively attacked through other means, but that, “standing alone, the testimony of the officer who recognized defendant from the sketch artist’s rendering would arguably have been sufficient to convict.” Fer-ensic, 2001 WL 865089, at *2. This focus on the sufficiency of the evidence against Ferensic, in the district court’s opinion, was proof that the Michigan Court of Appeals had employed the wrong standard in analyzing his claim because “[t]he prejudice inquiry ‘is not the same as the sufficiency of the evidence analysis.’ ” (Quoting Richey v. Mitchell, 395 F.3d 660, 687 (6th Cir.), overruled on other grounds by Bradshaw v. Richey, 546 U.S. 74, 126 S.Ct. 602, 163 L.Ed.2d 407 (2005).) The district court further cited Richey for the proposition that the prejudice inquiry under Strickland is not tantamount to
the analysis that a court might perform when deciding a motion for summary judgment. As the Fifth Circuit has explained, [the Court] need not find “that a reasonable jury could not have reached the same verdict if counsel had performed effectively.” Johnson v. Scott, 68 F.3d 106, 109 n. 4 (5th Cir.1995). [Petitioner] “need not show that he could not have been convicted. Instead, he need only undermine [the Court’s] confidence in the trial’s outcome.” Foster [v. Lockhart, 9 F.3d 722, 726 (8th Cir.1993) ].
Richey, 395 F.3d at 687 (emphasis in original).
Because the Michigan Court of Appeals not only reached what the district court believed was an incorrect result, but did so by way of an erroneous legal analysis, the district court concluded that the decision “amounted to an unreasonable application of Strickland.”
D. Our review
We will confine our review to Ferensic’s right-to-present-a-defense claim, which we believe to be the stronger of his two grounds for habeas relief. Ferensic seems to agree, styling his ineffective-assistance-*475of-counsel claim as an “alternative” argument. We also note that much of what we would have to say about this alternative claim is necessarily encompassed by our harmless-error analysis in Part II.D.2. below.
1. Right to present a defense
The right of an accused to present a defense in a criminal trial derives from the Compulsory Process Clause of the Sixth Amendment to the U.S. Constitution, and “stands on no lesser footing than the other Sixth Amendment rights that we have previously held applicable to the States.” Taylor v. Illinois, 484 U.S. 400, 409, 108 S.Ct. 646, 98 L.Ed.2d 798 (1988). In fact, “[flew rights are more fundamental than that of an accused to present witnesses in his own defense.” Id. at 408, 108 S.Ct. 646 (quoting Chambers v. Mississippi, 410 U.S. 284, 302, 93 S.Ct. 1038, 35 L.Ed.2d 297 (1973)). In Taylor, the Supreme Court explained the origins and nature of the right as follows:
The right to compel a witness’ presence in the courtroom could not protect the integrity of the adversary process if it did not embrace the right to have the witness’ testimony heard by the trier of fact. The right to offer testimony is thus grounded in the Sixth Amendment even though it is not expressly described in so many words: “The right to offer the testimony of witnesses, and to compel their attendance, if necessary, is in plain terms the right to present a defense, the right to present the defendant’s version of the facts as well as the prosecution’s to the jury so it may decide where the truth lies. Just as an accused has the right to confront the prosecution’s witnesses for the purpose of challenging their testimony, he has the right to present his own witnesses to establish a defense. This right is a fundamental element of due process of law.”
484 U.S. at 409, 108 S.Ct. 646 (quoting Washington v. Texas, 388 U.S. 14, 19, 87 S.Ct. 1920, 18 L.Ed.2d 1019 (1967)).
The right to present a defense, however, is not absolute. See Taylor, 484 U.S. at 409, 108 S.Ct. 646 (“The accused does not have an unfettered right to offer testimony that is incompetent, privileged, or otherwise inadmissible under standard rules of evidence.”); see also Michigan v. Lucas, 500 U.S. 145, 152, 111 S.Ct. 1743, 114 L.Ed.2d 205 (1991) (noting that the Taylor court “rejected the defendant’s argument that, under the Compulsory Process Clause of the Sixth Amendment, preclusion is never a permissible sanction for a discovery violation”) (emphasis and quotation marks omitted). As the Supreme Court reiterated most recently in United States v. Scheffer, 523 U.S. 303, 308, 118 S.Ct. 1261, 140 L.Ed.2d 413 (1998) (quoting Rock v. Arkansas, 483 U.S. 44, 56, 107 S.Ct. 2704, 97 L.Ed.2d 37 (1987)), the exclusion of evidence in a criminal trial “abridge[s] an accused’s right to present a defense” only where the exclusion is “ ‘arbitrary’ or ‘disproportionate to the purpose[ ] [it is] designed to serve.’ ” See also Scheffer, 523 U.S. at 330, 118 S.Ct. 1261 (Stevens, J., dissenting) (“As the Court notes today, restrictions on the ‘defendant’s right to present relevant evidence’ ... must comply with the admonition in Rock ....”) (citations omitted).
According to the Warden, the arbitrary- or-disproportionate standard is so “general” that a federal habeas court would be hard-pressed to find that a state court unreasonably applied the standard within the meaning of AEDPA. The Supreme Court, however, has furnished ample, specific guidance as to what will and will not qualify as arbitrary or disproportionate. In applying the Rock standard or some earlier formulation thereof, for example, the Court has “found the exclusion of evi*476dence to be unconstitutionally arbitrary or disproportionate only where it has infringed upon a weighty interest of the accused.” Scheffer, 523 U.S. at 308, 118 S.Ct. 1261 (citing Rock, 483 U.S. at 58, 107 S.Ct. 2704, Chambers, 410 U.S. at 302, 93 S.Ct. 1038, and Washington, 388 U.S. at 22-23, 87 S.Ct. 1920).
In addition, the Supreme Court has given special consideration to the nature of the exclusion-triggering discovery violation at issue, noting that only egregious violations involving, for example, “willful misconduct” on the part of the defendant or his counsel will justify the exclusion of material evidence. See Lucas, 500 U.S. at 152, 111 S.Ct. 1743 (quotation marks omitted) (emphasizing the trial court’s conclusion in Taylor that the “discovery violation amounted to willful misconduct and was designed to obtain a tactical advantage,” and that the Supreme Court’s decision to uphold the exclusion of an undisclosed witness in that case was explicitly “[b]ased on these findings”). Alternative, less severe sanctions than exclusion will thus be “adequate and appropriate in most cases.” Id. (quoting Taylor, 484 U.S. at 413, 108 S.Ct. 646). Stated differently, the exclusion of a defendant’s evidence should be reserved for only those circumstances where “a less severe penalty ‘would perpetuate rather than limit the prejudice to the State and the harm to the adversary process.’ ” Lucas, 500 U.S. at 152, 111 S.Ct. 1743 (quoting Taylor, 484 U.S. at 413, 108 S.Ct. 646).
Regarding the trial court’s exclusion of Dr. Shulman’s testimony, the Michigan Court of Appeals recognized that “the exclusion of otherwise admissible evidence should be limited to the most egregious cases, in which other less severe remedies would fail to protect the parties’ competing interests.” Ferensic, 2001 WL 865089, at *1 (quotation marks omitted). But the Court of Appeals never applied this proportionality-based proposition to the ease before it. In fact, it disregarded any less severe remedies. The Court of Appeals reasoned that the “concededly severe sanction of exclusion of this expert testimony” was appropriate because it was harmless to Ferensic. Id. at *2.
If harmlessness vis-á-vis the outcome were the test, however, courts would never have to consider whether “less severe remedies would ... protect the parties’ competing interests.” The proportionality test of Rock instead measures the restrictions on a defendant’s right to present a defense against the “purposes they are designed to serve.” 483 U.S. at 56, 107 S.Ct. 2704. Nevertheless, the Michigan Court of Appeals applied a more typical outcome-based prejudice standard, noting that the exclusion “did not put the prosecution in a better position than it would have enjoyed had the order been complied with because defense counsel was otherwise able to effectively challenge inconsistencies in the victims’ identification testimony.” Ferensic, 2001 WL 865089, at *2.
Moreover, even if a traditional prejudice-based standard were the proper test for analyzing a right-to-present-a-defense claim, the Court of Appeals’s conclusion is still unreasonable. The exclusion of Dr. Shulman’s testimony unquestionably “put the prosecution in a better position than it would have enjoyed had the order been complied with,” i.e., had Dr. Shulman actually testified. Dr. Shulman’s testimony was categorically different from what Fer-ensic’s father, the lone defense witness, offered to the jury. Ferensic’s father testified primarily to vouch for the presence of a large, easily observable scar on his son’s neck, which neither of the eyewitnesses had noticed. This testimony therefore provided the jury with a specific reason to distrust the eyewitnesses’ identifications — in other words, an example of *477how those identifications were inconsistent and thus unreliable.
Dr. Shulman’s testimony, in contrast, would have informed the jury of why the eyewitnesses’ identifications were inherently unreliable. This would have been a scientific, professional perspective that no one else had offered to the jury. Compare Spisak v. Mitchell, 465 F.3d 684, 696, 699 (6th Cir.2006) (upholding under Rock the exclusion of multiple expert defense witnesses principally because “[tjheir testimony, however reliable their reputations, could not be taken to establish Defendant’s legal insanity under Ohio law,” and “in fact, ... would likely have severely undercut such a defense”).
To be sure, as the district court noted, Ferensic’s counsel argued to the jury on multiple occasions that eyewitness identifications were inherently unreliable. But the jury was explicitly instructed, as it always is, that arguments by counsel are not evidence. The district court was therefore correct in concluding that “[wjithout [Dr. Shulman’s] testimony there was no evidence to support counsel’s argument.”
As the Michigan Court of Appeals noted, the principal reasoning behind the trial court’s decision to exclude Dr. Shulman from testifying was that “the prosecution was unable to retain its own expert witness on identification without delaying the trial.” Forensic, 2001 WL 865089, at *2. We acknowledge that the proper functioning of the adversary system is indeed a legitimate, and nonarbitrary, consideration as a general matter. See Taylor, 484 U.S. at 414-15, 108 S.Ct. 646 (“The integrity of the adversary process ... must also weigh in the balance.”). And “prejudice to the prosecution” is integral to that concern in a criminal case. See Lucas, 500 U.S. at 152, 111 S.Ct. 1743.
But as the district court aptly remarked, in this specific case
the prosecutor never said that he intended to present a witness to rebut the defense expert witness’ testimony, and he did not claim to be prejudiced by the tardy disclosure of Dr. Shulman’s report. He must have been aware of Dr. Shulman’s identity, because defense counsel submitted a list of his witnesses about ten months before trial.
This latter fact distinguishes the present case from Taylor, where the Supreme Court upheld the exclusion of two defense witnesses whose very identity the defendant had failed to disclose in complete disregard of the prosecution’s pretrial discovery request. Taylor, 484 U.S. at 404, 108 S.Ct. 646 (emphasizing the trial judge’s concern about the possibility “that witnesses are being found that really weren’t there”).
The unlikelihood that the prosecution would have called its own witness is further underscored by Dr. Shulman’s testimony at the evidentiary hearing that, during the more than 20 times that he has testified as an expert on eyewitness identifications, not once has his testimony been “countered by an opposing expert.” In addition, the Warden’s attorney conceded at oral argument that nothing in the record indicates that the prosecution would have wanted to call its own “opposing expert.” Similarly, the attorney acknowledged that the only part of Dr. Shulman’s testimony that the prosecution might have wanted to oppose likely would have been handled through normal cross-examination.
The state trial court in Ferensic’s case, however, never considered these particular circumstances. In fact, the court’s own words reveal its indifference to whether the prosecution was actually prejudiced. When Ferensic’s counsel reminded the court that the prosecutor had not given any indication of wanting to rebut Dr. *478Shulman’s testimony with a separate witness, the trial judge said “I’m not worried about that” and “I don’t care what he says.” The trial judge, in short, would have ruled against Ferensic regardless of whether the prosecution had actually wanted to retain its own expert on identification — that is, even if the prosecution had explicitly disavowed such a desire.
Such disregard for the substantial rights of one party in the absence of any prejudice to the other raises an inference of arbitrariness. See Black’s Law Dictionary (8th ed. 2004) (defining an “arbitrary” judicial decision as one that is “founded on prejudice or preference rather than on reason or fact”). It is certainly proof of disproportionality, especially in light of the absence of harm to the prosecution, the lack of willfulness on the part of Ferensic’s counsel in violating the discovery order (he turned over Dr. Shulman’s report as soon as he received it 11 days before trial), and the lack of any delay caused by counsel’s misstep. Compare United States v. Nobles, 422 U.S. 225, 241, 95 S.Ct. 2160, 45 L.Ed.2d 141 (1975) (holding, in a pre-Rock case, that the trial court’s preclusion of a defense investigator’s testimony was “an entirely proper method of assuring compliance with its order” where defense counsel had affirmatively refused to submit the investigator’s potentially conflicting report to the prosecution). Here, a less severe sanction was appropriate or, at the very least, should have been considered by the trial judge.
In addition to these considerations, the very nonappearance of both Dr. Shulman and, later, St. John likely had an adverse effect on Ferensic. As the district court explained,
[t]he problem was exacerbated by the trial court’s refusal to instruct the jury that the two defense witnesses in question were not permitted to testify. When defense counsel failed to present either witness as promised in his opening statement, the jury might have concluded that the witnesses were unable or unwilling to testify as expected and that defense counsel could not live up to the claims that he made in his opening statement.
The trial judge thus effectively inflicted double punishment on Ferensic: Not only would he be unable to present two critical witnesses in his defense — by itself a “severe sanction,” Ferensic, 2001 WL 865089, at *2 — but the jury would likely hold that inability against him. These harms to Ferensic far outweighed the trial court’s ministerial, albeit legitimate, “need to manage its trial docket.”
Nor can there be any doubt that Feren-sic’s interest in having Dr. Shulman and St. John testify was “weighty.” See Scheffer, 523 U.S. at 308, 118 S.Ct. 1261 (quotation marks omitted). As discussed in Part II.D.2. below, eyewitness misidentification accounts for more false convictions in the United States than any other factor. Similarly, the current near-universal acceptance of the reliability of expert testimony regarding eyewitness identification, also discussed in Part II.D.2. below, distinguishes the exclusion of Dr. Shulman in the present case from the exclusion upheld as proportionate by the most recent Supreme Court case to have applied Rock. See id. at 309, 118 S.Ct. 1261 (upholding as constitutional a per se rule against the admission of polygraph evidence in court-martial proceedings, principally because “the scientific community remains extremely polarized about the reliability of polygraph techniques”).
The dissent responds by arguing that Supreme Court precedent “indicate[s] that a ‘weighty interest of the accused’ ... is limited to an interest in offering the accused’s own testimony, ... or an interest in using the only available means of ad*479dressing a relevant issue.” Dissenting Op. at 487-88 (citing Rock, 483 U.S. at 58 n. 15, 107 S.Ct. 2704, Chambers, 410 U.S. at 300-02, 93 S.Ct. 1038, and Washington, 388 U.S. at 22, 87 S.Ct. 1920). But this summary erroneously assumes that Rock, Chambers, and Washington collectively constitute an exhaustive list of potentially “weighty interests.” Although the dissent is correct that Scheffer specifically cites these three cases in support of the proposition that Rock provides relief only where an exclusion has infringed on a weighty interest of the accused, Scheffer in no way suggests that other cases involving different interests could not also meet the de facto weighty-interest standard.
The Warden’s related argument that unreasonably applying the Rock standard for AEDPA purposes is virtually impossible because “the considerations to be applied [in analyzing a right-to-present-a-defense claim] present the most general of legal rules” is similarly misplaced. We fail to understand, and the Warden fails to explain, how the admittedly fact-intensive disproportionality inquiry mandated by Rock requires state appellate courts to do anything more than what other “clearly established” Supreme Court precedents, such as the evidence-weighing prejudice inquiry mandated by Strickland, already require. See Wiggins v. Smith, 539 U.S. 510, 534, 123 S.Ct. 2527, 156 L.Ed.2d 471 (2003) (“In assessing prejudice [under Strickland ], we reweigh the evidence in aggravation against the totality of available mitigating evidence.”).
Regarding the exclusion of St. John’s testimony, Ferensic faces a somewhat tougher task because that particular exclusion was simply the byproduct of the trial court’s denial of his motion for an adjournment. That determination, the validity of which is measured by a totality-of-the-circumstances test, is entitled to considerable deference. The Supreme Court explained the rationale behind this open-ended standard in Ungar v. Sarafite, 376 U.S. 575, 589, 84 S.Ct. 841, 11 L.Ed.2d 921 (1964), as follows:
The matter of continuance is traditionally within the discretion of the trial judge, and it is not every denial of request for more time that violates due process even if the party fails to offer evidence or is compelled to defend without counsel.... Contrariwise, a myopic insistence upon expeditiousness in the face of a justifiable request for delay can render the right to defend with counsel an empty formality.... There are no mechanical tests for deciding when a denial of a continuance is so arbitrary as to violate due process. The answer must be found in the circumstances present in every case, particularly in the reasons presented to the trial judge at the time the request is denied.
(Citations omitted.)
Although the governing federal law relating to continuances is a broad, totality-of-the-circumstanees standard, as opposed to a hard-and-fast bright-line rule, this again does not mean that its application cannot be unreasonable within the meaning of AEDPA. The state courts must at the very least consider the relevant circumstances. But neither the trial court nor the Michigan Court of Appeals even attempted to do so in the present case.
Per the instructions of Ferensic’s counsel, St. John was due to arrive at the courthouse at 11:00 a.m. on the morning in question. Ferensic’s father, the lone defense witness, concluded his testimony at 10:25 a.m., a mere 35 minutes earlier. The court then temporarily excused the jury. Although the court allowed Ferensic to make an offer of proof at that time regarding St. John’s anticipated testimony, it denied Ferensic’s motion for a brief adjournment to allow St. John to actually testify *480before the jury. The trial judge appeared to base her decision largely on the rationale that she had another trial scheduled to begin at 11:00 a.m. that same morning. Nevertheless, the court then formally recessed for a break at 10:30 a.m. and scheduled closing arguments to begin at 10:57 a.m.
Such a “myopic insistence upon expeditiousness in the face of a justifiable request for delay” is precisely what the Supreme Court forbids. See Ungar, 376 U.S. at 589, 84 S.Ct. 841. Even the Warden does not argue that Ferensic’s request for less than a 30-minute adjournment was unreasonable. Although St. John’s tardiness was entirely Ferensic’s fault, Feren-sic was asking for the briefest of delays. St. John was scheduled to arrive a mere three minutes after the trial court had scheduled closing arguments to begin. Given the offer of proof that the court itself requested, moreover, it was fully aware that St. John’s testimony, even including cross-examination, would have been brief.
Granting Ferensic’s motion would likely not have delayed the proceedings by more than half an hour. Even a longer delay would have been justifiable in light of the fact that, as Ferensic notes in his brief, “the trial judge had scheduled three days for trial, and the case was proceeding to closing argument in less than a day and a half.” The Michigan Court of Appeals never mentioned this fact, much less any of the other circumstances discussed above, in its opinion.
Instead, the Court of Appeals rested its affirmance of the trial court’s judgment solely on a no-prejudice determination; namely, that St. John’s “testimony was not especially strong, ... and inconsistency in the victims’ identification was otherwise shown.” Ferensic, 2001 WL 865089, at *2. Harmless error, however, is not the test mandated by the arbitrary-or-disproportionate inquiry of Rock, as noted above. And even if prejudice were material to the analysis, Ferensic was prejudiced by St. John’s failure to testify for the reasons already mentioned, as well as for those discussed below in the context of our independent harmless-error review. Indeed, the exclusion of St. John could in a certain sense be considered more prejudicial than the exclusion of Dr. Shulman because St. John, unlike Dr. Shulman, would have introduced defense-favorable “factual evidence” to the jury. See Scheffer, 523 U.S. at 316-17, 118 S.Ct. 1261 (concluding that a per se rule excluding polygraph evidence did “not implicate any significant interest of the accused” because “the Rule did not preclude [the defendant] from introducing any factual evidence,” but “merely from introducing expert opinion testimony to bolster his own credibility”).
In sum, Ferensic was denied his Sixth Amendment right to present a defense by the trial court’s exclusion of both Dr. Shul-man and St. John as defense witnesses. The contrary determination of the Michigan Court of Appeals was unreasonable for the reasons stated above. But before we can affirm the judgment of the district court conditionally granting Ferensic’s petition for a writ of habeas corpus on this ground, we must independently determine whether the state trial court’s constitutional error was harmless within the meaning of Brecht v. Abrahamson, 507 U.S. 619, 113 S.Ct. 1710, 123 L.Ed.2d 353 (1993). See Mitzel v. Tate, 267 F.3d 524, 534 (6th Cir.2001) (“In applying the harmless error analysis on habeas review for cases governed by AEDPA, we apply the harmless error standard set out in Brecht.”).
2. Harmless-error review
Brecht clarified that the harmless-error standard set forth in Kotteakos v. United States, 328 U.S. 750, 66 S.Ct. 1239, 90 *481L.Ed. 1557 (1946), governs federal courts’ review of habeas corpus petitions based on constitutional “trial error.” 507 U.S. at 687-38, 113 S.Ct. 1710. Pursuant to Brecht, a constitutional error is not harmless if it “had substantial and injurious effect or influence in determining the jury’s verdict.” 507 U.S. at 631, 113 S.Ct. 1710 (quoting Kotteakos, 328 U.S. at 776, 66 S.Ct. 1239). Although Brecht further states that habeas petitioners “are not entitled to habeas relief based on trial error unless they can establish that it resulted in ‘actual prejudice,’ ” 507 U.S. at 637, 113 S.Ct. 1710 (citation omitted), the Supreme Court in O’Neal v. McAninch, 513 U.S. 432, 438, 115 S.Ct. 992, 130 L.Ed.2d 947 (1995), clarified that this language “is not determinative,” and that petitioners do not bear an affirmative burden of proof as the language suggests. Instead, “it [is] conceptually clearer for the judge to ask directly, ‘Do I, the judge, think that the error substantially influenced the jury’s decision?’ than for the judge to try to the put the same question in terms of proof burdens (e.g., ‘Do I believe the party has borne its burden of showing ... ?’).” O’Neal, 513 U.S. at 436-37, 115 S.Ct. 992.
Uncertainty in answering this question, moreover, militates in favor of the habeas petitioner. “When a federal judge in a habeas proceeding is in grave doubt about whether a trial error of federal law had ‘substantial and injurious effect or influence in determining the jury’s verdict,’ that error is not harmless. And, the petitioner must win.” Id. at 436, 115 S.Ct. 992. The O’Neal Court explained that “[b]y ‘grave doubt’ we mean that, in the judge’s mind, the matter is so evenly balanced that he feels himself in virtual equipoise as to the harmlessness of the error.” Id. at 435, 115 S.Ct. 992. An error of constitutional magnitude will therefore satisfy Brecht if we determine that it “had a substantial and injurious effect or influence in determining the jury’s verdict,” or if we find ourselves in “grave doubt” about that determination. See Kotteakos, 328 U.S. at 765, 66 S.Ct. 1239 (“If [the error itself had substantial influence], or if one is left in grave doubt, the conviction cannot stand.”).
In the present case, we are in “grave doubt” as to whether the exclusion of St. John and especially Dr. Shulman had “a substantial and injurious effect or influence” on the outcome of Ferensic’s trial. Regarding St. John, we agree with the Michigan Court of Appeals that his testimony was not “especially strong.” Even though his description of one of the two men that he observed outside of the Kos-toffs’ home as having curly black hair was consistent with the Kostoffs’ initial description, which they later denied at trial, St. John effectively conceded at the evi-dentiary hearing that his testimony was not alone sufficient to establish Ferensic’s innocence. He admitted that his observation of the men was actually “[j]ust a glance” that he caught while driving past the Kostoffs’ home and, accordingly, that he was “not able to provide a full description of either one of these individuals.” On its own, therefore, the trial court’s exclusion of St. John is not sufficient to satisfy the “substantial and injurious” standard of Brecht, even though Ferensic was prejudiced to at least some degree by St. John’s failure to testify. Our decision to affirm the grant of Ferensic’s habeas petition instead rests on the combined effect of the trial court’s exclusion of both St. John and Dr. Shulman, with particular emphasis on Dr. Shulman.
Regarding Dr. Shulman, we add the following analysis to our foregoing discussion of his value to Ferensic. We agree with the district court that “other means” of attacking eyewitness identifications do not effectively substitute for expert testimony on their inherent unreliability. This *482court’s decision in United States v. Smithers, 212 F.3d 306 (6th Cir.2000), provides direct support for the district court’s conclusion that the typical methods of “challenging] inconsistencies” in the eyewitnesses’ testimony “were not an effective substitute” for what Dr. Shulman would have offered. In Smithers, the majority thoroughly rejected the dissent’s contrary argument as follows:
The Dissent counters by arguing that eyewitness identification experts are not necessary because cross-examination and jury instructions should be tools used in a trial to discredit and flush-out eyewitness testimony. Unfortunately, the Dissent’s homage to trial procedures does not extend to expert witness testimony. The same argument can be made for the admission of expert testimony: cross-examination and jury instructions can be used to question the qualifications of the proffered expert, undermine the basis of the expert’s theories, explain the limits of social science’s validation studies and pick apart research methods. The only reason given by the Dissent for why cross-examination and jury instructions can serve these goals for eyewitness testimony, but not for expert testimony, is that the jury may take the expert’s testimony as “scientifically irrefutable truth.” The Dissent’s selective faith in the collective intelligence, common sense and decision-making ability of the jury is disheartening, and is also inconsistent with the Dissent’s deference to the jury on other matters, including judging the credibility of eyewitness identifications.
212 F.3d at 316.
The Smithers court also recognized that expert testimony on eyewitness identifications, once thought to be unreliable and overly prejudicial to the prosecution, is now universally recognized as scientifically valid and of “aid [to] the trier of fact” for admissibility purposes. Id. at 315 (discussing expert testimony on eyewitness identifications in the context of the now-prevailing Daubert test for the admissibility of expert testimony in the federal courts). As the court explained,
jurors tend to be unduly receptive to, rather than skeptical of, eyewitness testimony. Further, accepting the district court’s analysis that all jurors are aware of their obligation to be skeptical would lead to absurd results: expert testimony on eyewitness identification would never be admissible. As demonstrated by abundant case law, this is not the conclusion that has been reached by courts addressing this issue. Today, there is no question that many aspects of perception and memory are not within the common experience of most jurors, and in fact, many facts that affect memory are counter-intuitive.
Id. at 315-16 (emphasis in original). Nor was there any dispute regarding the admissibility of Dr. Shulman’s expert testimony in the present case. Rather, the trial court excluded his testimony simply because Ferensic had failed to comply with a pretrial discovery order.
The significance of Dr. Shulman’s testimony cannot be overstated. Without it, the jury had no basis beyond defense counsel’s word to suspect the inherent unreliability of the Kostoffs’ identifications. The Supreme Court has long recognized this obliviousness as inimical to our system of criminal justice. See, e.g., Watkins v. Sowders, 449 U.S. 341, 352, 101 S.Ct. 654, 66 L.Ed.2d 549 (1981) (“[D]espite its inherent unreliability, much eyewitness identification evidence has a powerful impact on juries. Juries seem most receptive to, and not inclined to discredit, testimony of a witness who states that he saw the defendant commit the crime.”). Indeed, eyewitness misidentification is “the single most important factor leading to wrongful con*483victions in the United States.” United States v. Brownlee, 454 F.3d 131, 141 (3d Cir.2006) (quoting C. Ronald Huff et al., Guilty Until Proven Innocent: Wrongful Conviction and Public Policy, 32 Crime & Delinq. 518, 524 (1986)); see also Chris Conway, The DNA 200, N.Y. Times, May 20, 2007, § 4, at 14 (reporting on the first 200 U.S. inmates formally cleared on the strength of DNA evidence, and noting that “three-quarters [of the initial convictions] were marked by inaccurate eyewitness identification,” a greater percentage than that attributable to any other factor). This statistic alone strongly supports the conclusion that excluding Dr. Shulman’s testimony “had a substantial and injurious effect” on Ferensic. Brecht, 507 U.S. at 631, 113 S.Ct. 1710. The identity of the perpetrator, after all, was the only issue at trial.
It was, moreover, a very close issue. We acknowledge that, as the Michigan Court of Appeals concluded, “standing alone, the testimony of the officer who recognized defendant from the sketch artist’s rendering would arguably have been sufficient to convict.” Ferensic, 2001 WL 865089, at *2. But Kotteakos prohibits us from simply focusing on the sufficiency of the evidence, especially where it entails “stripping the erroneous action from the whole” and determining the sufficiency of what is left “standing alone.” 328 U.S. at 765, 66 S.Ct. 1239. We must instead “ponder[ ] all that happened.” Id. This wider lens is what leaves us with “grave doubt.” See id.
In “pondering all that happened” in the present case, we are particularly persuaded by the district court’s finding that “the record indicates that the jurors were unable to agree on a verdict at one point during their deliberations.” More specifically, the jury sent a note to the trial judge stating that “[w]e would like to see the police report,” and asked “[w]hat are our options if we don’t totally agree on a verdict?” The “police report,” of course, contained the police sketch as part of Feren-sic’s larger file. Thus the jury’s own words imply not only that it had doubts about the strength of the case against Ferensic, but also that those doubts related at least in part to the contents of his police file.
Although the jury did not explicitly question the “sufficiency” of the sketch-based identification vis-á-vis Ferensic’s guilt, its note, especially when considered in the context of the erroneously excluded testimony from Dr. Shulman and St. John, precludes us from saying “with fair assurance ... that the judgment was not substantially swayed by the error.” See Kotteakos, 328 U.S. at 765, 66 S.Ct. 1239; see also Fry v. Pliler, — U.S. -, 127 S.Ct. 2321, 2330, 168 L.Ed.2d 16 (2007) (Stevens, J., dissenting) (citing six federal appellate-court decisions, including Powell v. Collins, 332 F.3d 376, 401 (6th Cir.2003), for the proposition that “jurors’ evident uncertainty” as to a defendant’s guilt makes a finding of prejudicial error in a right-to-present-a-defense case almost impossible to effectively dispute).
We wish to emphasize just how significant the jury’s note is to our analysis, because it distinguishes the present case from many others in which the erroneous exclusion of an expert witness on eyewitness identification might well be harmless. For example, we recognize that this court has held on at least two occasions that the failure of counsel to hire an expert in eyewitness identification did not prejudice the defendant in a criminal trial. See Dorch v. Smith, 105 Fed.Appx. 650, 653 (6th Cir.2004) (upholding as reasonable the Michigan Court of Appeals’s conclusion that defense counsel’s failure to call an expert witness on eyewitness identification did not satisfy Strickland v. Washington, *484466 U.S. 668, 104 S.Ct. 2052, 80 L.Ed.2d 674 (1984), because counsel “presented several witnesses who testified as to [the habeas petitioner’s] whereabouts on the weekend of the incident” and cross-examined the eyewitness regarding inconsistencies in his identification of the petitioner); Tipton v. United States, No. 96-5026, 1996 WL 549802, at *1-2 (6th Cir. Sept.26, 1996) (holding that “any allegedly ineffective assistance” caused by counsel’s failure to “hir[e] an expert in eyewitness identification” did not prejudice the petitioner within the meaning of Strickland).
Although the failure to retain an expert as an initial matter presents a somewhat different problem than the exclusion of an already retained expert, we recognize the tension that our holding today might appear to generate with Dorch, Tipton, and similar cases. We therefore limit our holding to the situation here where the record reflects the doubts of the jury itself as to the identification of the perpetrator. Our analysis, in other words, should not be read to imply that the exclusion of an eyewitness-identification expert in a state-court criminal trial — even if “arbitrary” or “disproportionate” within the meaning of Rock — will always warrant relief on federal habeas corpus review.
III. CONCLUSION
For all of the reasons set forth above, we AFFIRM the judgment of the district court.