Ervine Davenport v. Duncan MacLaren ( 2020 )


Menu:
  •                          RECOMMENDED FOR FULL-TEXT PUBLICATION
    Pursuant to Sixth Circuit I.O.P. 32.1(b)
    File Name: 20a0197p.06
    UNITED STATES COURT OF APPEALS
    FOR THE SIXTH CIRCUIT
    ERVINE LEE DAVENPORT,                                 ┐
    Petitioner-Appellant,   │
    │
    >      No. 17-2267
    v.                                              │
    │
    │
    DUNCAN MACLAREN, Warden,                              │
    Respondent-Appellee.     │
    ┘
    Appeal from the United States District Court
    for the Western District of Michigan at Grand Rapids.
    No. 1:14-cv-01012—Ellen S. Carmody, Magistrate Judge.
    Argued: May 7, 2019
    Decided and Filed: June 30, 2020
    Before: COLE, Chief Judge; STRANCH and READLER, Circuit Judges.
    _________________
    COUNSEL
    ARGUED: Tasha J. Bahal, WILMER CUTLER PICKERING HALE AND DORR LLP,
    Boston, Massachusetts, for Appellant. Jared D. Schultz, OFFICE OF THE MICHIGAN
    ATTORNEY GENERAL, Lansing, Michigan, for Appellee. ON BRIEF: Tasha J. Bahal,
    Reuven Dashevsky, WILMER CUTLER PICKERING HALE AND DORR LLP, Boston,
    Massachusetts, for Appellant.   Aaron D. Lindstrom, OFFICE OF THE MICHIGAN
    ATTORNEY GENERAL, Lansing, Michigan, for Appellee.
    STRANCH, J., delivered the opinion of the court in which COLE, C.J., joined.
    READLER, J. (pp. 26–46), delivered a separate dissenting opinion.
    No. 17-2267                        Davenport v. MacLaren                                 Page 2
    _________________
    OPINION
    _________________
    JANE B. STRANCH, Circuit Judge. Ervine Lee Davenport was convicted of first-degree
    murder after a jury trial in Michigan state court. He challenges his conviction in a habeas corpus
    petition under 
    28 U.S.C. § 2254
     because he was visibly shackled at the waist, wrist, and ankles
    during trial. The State of Michigan admits Davenport’s shackling was unconstitutional but
    argues that the habeas petition should be denied because this error was harmless. The district
    court agreed it was harmless error and denied the petition. Because “shackling is ‘inherently
    prejudicial,’” Deck v. Missouri, 
    544 U.S. 622
    , 635 (2005) (quoting Holbrook v. Flynn, 
    475 U.S. 560
    , 568 (1986)), and the evidence of premeditation and deliberation necessary to a first-degree
    murder conviction was not overwhelming, the State has not met its burden to show the restraints
    did not have a “substantial and injurious effect or influence in determining the jury’s verdict.”
    Brecht v. Abrahamson, 
    507 U.S. 619
    , 637 (1993). We therefore REVERSE the judgment of the
    district court, GRANT Davenport a conditional writ of habeas corpus, and REMAND the case
    for further proceedings.
    I. BACKGROUND
    A. The Trial
    Davenport killed Annette White in the early hours of January 13, 2007. At his 2008 trial,
    he testified that he had been drinking beer and smoking crack cocaine with White and some
    friends when White was asked to leave her friends’ house due to her aggressive behavior.
    According to Davenport’s testimony, he tried to calm her down as he drove her home. While he
    drove, she was saying that she was hot and taking off her clothes. She demanded that he take her
    to a specific apartment building and tried to grab the steering wheel. Each time she tried to grab
    the wheel, Davenport pushed her back. She then started yelling and kicking and pulled out a
    boxcutter, which she swung at Davenport, cutting his arm. Davenport testified that he was afraid
    of the knife and trying to avoid oncoming traffic. As he continued to drive, Davenport pinned
    her against the side of the car with his fully extended hand pressed against her neck. Right as he
    No. 17-2267                         Davenport v. MacLaren                                   Page 3
    was about to let up, she scratched him on the face, and he “pinned her back up against the other
    side of the car.” At some point, he noticed that she was no longer struggling; initially, he
    thought that she had calmed down or passed out but then he realized she was not breathing. He
    panicked and left her partially clothed body in a field. He testified that he was not sure how long
    he held White back by the throat but that it “seemed like . . . everything happened fast.”
    Some of this testimony was corroborated by other evidence at trial. Medical evidence
    established that White had consumed a substantial amount of alcohol and crack cocaine shortly
    before her death. An independent witness also testified that he had consumed beer and crack
    cocaine with Davenport and White that night. This witness further testified that he asked White
    to leave at about 2:30 a.m. that morning because she was acting “agitated” and “getting crazy.”
    She was “ranting and raving,” though he would not describe her as “violent.” Other witnesses
    testified that White would get angry when she smoked crack cocaine, and that she was “a
    spitfire” who had a reputation for fighting.
    But medical evidence seemingly contradicted other aspects of Davenport’s testimony. A
    forensic pathologist, Dr. Brian Hunter, testified that although it would take 30 seconds to cut
    someone’s air off sufficiently to cause them to pass out, it would take at least four to five
    minutes to suffocate someone to death. Dr. Hunter also testified in rebuttal that the injuries to
    White were not consistent with Davenport’s testimony that he did not choke White and instead
    “his hand was flexed and that all he was doing was pushing her against the door.” Dr. Hunter
    explained that the injuries to either side of White’s neck, but not the middle, were “more
    consistent with choking than . . . broad pressure there.”
    The prosecution also presented testimony that Davenport had strangled another woman
    until she was unconscious less than a week before White’s death. Another witness testified that
    Davenport had told him a couple of times that “if things got out of hand,” he would choke
    people. Davenport told this same witness that White “kept coming back at him and it just got out
    of hand, and that’s when he offed her.”
    No. 17-2267                               Davenport v. MacLaren                                            Page 4
    In its closing, the defense argued that this was a case of self-defense. In contrast, after
    giving 17 reasons why Davenport’s “‘self-defense’ claim was bogus,”1 the prosecution claimed
    that “[t]he only real issue is whether it’s first-degree” or second-degree murder. The only
    support for premeditation and deliberation the prosecution gave in its closing statement was the
    length of time it would take to choke someone to death. The prosecutor claimed, “[c]learly he
    had the opportunity to hesitate, stop, think about what he was doing, and not kill her. I submit to
    you there’s more than enough evidence of premeditation and deliberation for first-degree
    murder, but at the very least obviously this is second-degree murder.” After deliberating for six
    hours over the course of two days, the jury found Davenport guilty of first-degree murder.
    During the trial, Davenport had one hand cuffed, as well as shackles around his waist and
    ankles. The trial judge allowed “his right hand to be uncuffed so he could write notes to his
    counsel.” The judge also noted that there was a privacy curtain around the defense table.
    Defense counsel referred to the “[c]ourt’s policy regarding the shackles,” but there was no on-
    the-record justification given for the shackling.
    B. The State Court Appeals and Evidentiary Hearing
    On direct appeal, Davenport raised several issues, including that “he was denied his due
    process rights when the trial court required him to wear shackles during the trial.” People v.
    Davenport, Docket No. 287767, 
    2010 WL 3062279
    , at *1 (Mich. Ct. App. Aug. 5, 2010). The
    Michigan Court of Appeals found that this issue was unpreserved. 
    Id.
     Reviewing it for plain
    error, the court found that “it was error for the trial court to order defendant to be restrained
    without making the requisite findings,” but that Davenport had “not shown that his restraints
    were visible to the jury” and thus had “not demonstrated prejudice.” 
    Id.
     at *1–2. The Michigan
    Supreme Court reversed, finding that this issue was preserved 2 and remanding the case to the
    trial court for an evidentiary hearing. People v. Davenport, 
    794 N.W.2d 616
     (Mich. 2011). The
    1These   reasons included, among others, that Davenport dumped White’s body in the woods instead of
    calling 911, initially lied to the police, weighed almost three times as much as the victim, and stole some of White’s
    property after killing her.
    2During  jury selection, defense counsel asked that Davenport’s handcuff be removed because, “given the
    circumstances, the testimony, the evidence I believe is going to be presented, I don’t want the jurors to be unduly
    influenced and fearful of Mr. Davenport.”
    No. 17-2267                             Davenport v. MacLaren                                        Page 5
    Michigan Supreme Court directed the trial court to determine whether “the jury saw the
    defendant’s shackles” and, if so, “whether the prosecution can demonstrate beyond a reasonable
    doubt that the shackling error did not contribute to the verdict against the defendant.” 
    Id.
     (citing
    Deck, 
    544 U.S. at 635
    ).
    The trial court subsequently held an evidentiary hearing where all 12 jurors testified.
    This evidentiary hearing was held on June 24 and July 29, 2011, approximately three years after
    the trial. Five jurors testified that they saw Davenport’s waist chain, handcuffs, or ankle shackles
    at some point during jury selection or the trial. Two other jurors testified that they recalled
    comments by other jurors about Davenport’s shackles. One juror could not remember whether
    she saw the shackles. The remaining four jurors testified that they did not notice or hear about
    Davenport’s restraints during the trial.
    Several jurors recalled at the evidentiary hearing that they had thought Davenport might
    be dangerous when they saw him in shackles. Another juror recalled that she was sitting closest
    to Davenport when he testified and a fellow juror had asked her if that made her nervous. She
    also recalled that there were more guards when Davenport testified because he was not in
    shackles. But the jurors who testified that they saw Davenport’s shackles also all said that they
    believed shackling was routine practice given that he was on trial for murder or because he was
    in pre-trial incarceration.3 Every juror asked also testified that Davenport’s shackling did not
    affect their deliberations.
    After this hearing, the trial court issued an opinion ruling that, although some of the
    jurors saw Davenport’s shackles, the prosecution had proved beyond a reasonable doubt that the
    shackling did not affect the jury’s verdict. The trial court focused on the jurors’ testimony that
    Davenport’s shackling was not discussed during deliberations and did not affect their verdict. It
    also relied on the jurors’ testimony that they viewed the shackling as a routine security
    procedure. The Michigan Court of Appeals affirmed, holding that “[t]he trial court did not err in
    finding that the prosecution proved beyond a reasonable doubt that the shackling error did not
    affect the verdict.” People v. Davenport, Docket No. 306868, 
    2012 WL 6217134
    , at *3 (Mich.
    3Davenport wore an orange jail jumpsuit on the first day of trial, during jury selection. He wore dress
    clothes for the rest of the trial. He does not raise any issues relating to his attire on appeal.
    No. 17-2267                        Davenport v. MacLaren                                  Page 6
    Ct. App. Dec. 13, 2012). The Michigan Supreme Court denied leave to appeal. People v.
    Davenport, 
    832 N.W.2d 389
    , 390 (Mich. 2013). It stated that, although “the Court of Appeals
    erroneously failed to consider defendant’s claim in light of the United States Supreme Court
    decision in Holbrook v. Flynn . . . , the error was harmless under the facts of this case. Given the
    substantial evidence of guilt presented at trial, we cannot conclude that there was an
    unacceptable risk of impermissible factors coming into play.” 
    Id.
    C. The § 2254 Petition
    Davenport next filed a timely petition for a writ of habeas corpus under 
    28 U.S.C. § 2254
    ,
    raising a single issue—his shackling. After the State of Michigan responded to this § 2254
    petition, the magistrate judge issued a Report and Recommendation (R&R) that the petition be
    denied. The magistrate judge found that “it was completely reasonable for the jury to reject
    Petitioner’s claim of self-defense and to instead find that Petitioner committed first degree
    murder.” Thus, the magistrate judge concluded that the state courts’ determination “that the
    prosecution had demonstrated beyond a reasonable doubt that [Davenport’s shackling] did not
    contribute to the jury’s guilty verdict” was “neither contrary to, nor involve[d] an unreasonable
    application of, clearly established federal law.”      The district court overruled Davenport’s
    objections and adopted the R&R in its entirety, denying the petition and a certificate of
    appealability.
    Davenport, now proceeding pro se, applied to this court for a certificate of appealability.
    We granted him a certificate of appealability and appointed counsel. This appeal followed.
    II. ANALYSIS
    A. Standard of Review
    We “review the decision of a district court to grant or deny a writ of habeas corpus de
    novo” and “review factual findings by that court for clear error, except where the district court
    has made factual determinations based on its review of . . . court records; in such cases we
    review such findings de novo.” Ruimveld v. Birkett, 
    404 F.3d 1006
    , 1010 (6th Cir. 2005).
    No. 17-2267                              Davenport v. MacLaren                                          Page 7
    There is a dispute, however, about what standard applies on habeas review when
    determining whether Davenport’s unconstitutional shackling was harmless error. The State
    argues that our review must entail two separate determinations. First, we must find that the state
    court’s conclusion that the shackling was harmless beyond a reasonable doubt “was contrary to,
    or involved an unreasonable application of, clearly established Federal law, as determined by the
    Supreme Court.” 
    28 U.S.C. § 2254
    (d). And, second, we must find that the shackling had a
    “substantial and injurious effect or influence in determining the jury’s verdict.” The former is
    the familiar test required by the Antiterrorism and Effective Death Penalty Act of 1996
    (AEDPA). See, e.g., Mitchell v. Esparza, 
    540 U.S. 12
    , 18 (2003) (“[H]abeas relief is appropriate
    only if the [state court] applied harmless-error review in an ‘objectively unreasonable’ manner.”
    (quoting Lockyer v. Andrade, 
    538 U.S. 63
    , 75–77 (2003))). The latter is the prejudice standard
    that habeas petitioners complaining of trial error are required to meet. See Brecht, 
    507 U.S. at 623, 637
    . Davenport, on the other hand, argues that the only question before this court is
    whether the shackling “had substantial and injurious effect or influence in determining the jury’s
    verdict,” 
    id. at 637
    , because the Brecht standard “‘subsumes’” AEDPA’s unreasonableness
    inquiry. Davis v. Ayala, 
    135 S. Ct. 2187
    , 2198–99 (2015) (citing Fry v. Pliler, 
    551 U.S. 112
    ,
    119–20 (2007)).
    Binding precedent resolves the issue. “The answer in this Circuit is that Brecht is always
    the test, and there is no reason to ask both whether the state court ‘unreasonably’
    applied Chapman4 under the AEDPA and, further, whether the constitutional error had a
    ‘substantial and injurious’ effect on the jury’s verdict.” Ruelas v. Wolfenbarger, 
    580 F.3d 403
    ,
    412 (6th Cir. 2009). We adopted this conclusion from the Supreme Court’s statement that it
    “makes no sense to require formal application of both tests (AEDPA/Chapman and Brecht) when
    the latter obviously subsumes the former.” 
    Id.
     (quoting Fry, 
    551 U.S. at 120
    ); see Reiner v.
    Woods, 
    955 F.3d 549
    , 556 (6th Cir. 2020) (“The Supreme Court and this court have made clear
    that ‘Brecht is always the test’ for evaluating harmless error on collateral review, even where
    AEDPA applies.”).
    4Chapman v. California, 
    386 U.S. 18
     (1967) (holding that, on direct appeal, the prosecution has the burden
    of proving beyond a reasonable doubt that a federal constitutional error was harmless).
    No. 17-2267                        Davenport v. MacLaren                                 Page 8
    The dissent suggests that Ruelas and its progeny are called into question by the Supreme
    Court’s decision in Davis v. Ayala, 
    135 S. Ct. 2187
    . Ayala, however, noted that while a habeas
    petitioner’s obligation to meet the Brecht standard “does not mean . . . that a state court’s
    harmlessness determination has no significance under Brecht,” it further explained that though
    AEDPA remains a precondition to habeas relief, both the Brecht and AEDPA/Chapman tests
    need not be applied. 
    Id. at 2198
    ; see also Fry, 
    551 U.S. at
    119–20. The Supreme Court
    concluded that applying Brecht alone is appropriate because “the Brecht test subsumes the
    limitations imposed by AEDPA.” 
    Id.
     at 2199 (citing Fry, 
    551 U.S. at
    119–20). Ruelas relied on
    that precedential conclusion, explaining that though Brecht “handles the work” of both tests, a
    federal “habeas court remains free to, before turning to Brecht, inquire whether the state
    court’s Chapman analysis was reasonable. If it was reasonable, the case is over. But . . . a
    habeas court may [also] go straight to Brecht with full confidence that the AEDPA’s stringent
    standards will also be satisfied.” Ruelas, 
    580 F.3d at
    412–13.
    In fact, we have already concluded that Ruelas and Ayala are consistent: “Ruelas, which
    has not been affected by Ayala, . . . clearly announc[es] that in the Sixth Circuit on habeas
    review we always apply Brecht and need not also apply AEDPA/Chapman.”                   O’Neal v.
    Balcarcel, 
    933 F.3d 618
    , 625 (6th Cir. 2019). In Reiner, we explained that while “[t]he state
    argues that the Supreme Court’s subsequent decision in Davis v. Ayala changed this dynamic
    . . . [t]he problem for the state is that our precedent forecloses this approach.” 955 F.3d at 556–
    57; see also McCarley v. Kelly, 
    801 F.3d 652
    , 665 (6th Cir. 2015) (citing to both Ruelas and
    Ayala for the applicable standard); Sheard v. Klee, 692 F. App’x 780, 786 (6th Cir. 2017)
    (“Although Brecht is a pre-AEDPA case, the Supreme Court has subsequently held that
    the Brecht test ‘subsumes’ the AEDPA requirements such that a formal application of both tests
    is unnecessary.” (quoting Fry, 
    551 U.S. at 120
    )). We have again recently said, not that two tests
    must be performed, but that the proper test contains “a choice of prompts;” in which one
    “option—a shortcut of sorts—is to leapfrog AEDPA and jump directly to Brecht.” Hollman v.
    Sprader, 803 F. App’x 841, 843 (6th Cir. 2020) (citing Ayala, 
    135 S. Ct. at
    2198–99).
    The dissent supposes that our quotation from Reiner unfairly masks its holding. Dissent
    at 34. But Reiner expressly holds that the “Supreme Court and this court have made clear that
    No. 17-2267                               Davenport v. MacLaren                                            Page 9
    ‘Brecht is always the test.’” 955 F.3d at 556. Reiner is merely a recent installment in an
    ongoing chain of binding precedent that has sought to provide clarity to the relationship between
    AEDPA and Brecht—an issue that has preoccupied appellate courts and caused “consternation”
    among the lower courts since AEDPA was passed. Ruelas, 
    580 F.3d at
    412–13; see also 2 R.
    Hertz & J. Liebman, Federal Habeas Corpus Practice and Procedure § 31.1 (7th ed. 2019).
    There is no nefarious ellipsis plot to paint over the Reiner court’s recognition of the State’s
    argument. Simply put, where the Supreme Court has stepped in to provide clarity—as it did in
    Fry and again in Ayala—we are obliged to follow it. And here that clarifying law includes
    Ayala’s holding that where a habeas petitioner can succeed under the more demanding Brecht
    test, the state court’s “harmlessness determination itself is unreasonable,” which shows that both
    tests are satisfied. 
    135 S. Ct. at 2199
    .
    Faced with unambiguous precedent from both the Supreme Court and our circuit, the
    dissent next turns to challenges to timing, and word-splitting. First it contends that Brecht could
    not have functionally subsumed the AEDPA test because AEDPA was “still in the mind’s eye”
    when Brecht was decided. Dissent at 35. But, as noted above, the Supreme Court set out its
    holding that the Brecht test subsumed AEDPA after AEDPA was enacted. We are bound by its
    interpretation of AEDPA’s requirements. The dissent then challenges the meaning of the words
    used by the Supreme Court, contending that the Court held the Brecht test “subsumes” the
    AEDPA analysis but not that the Brecht test “consumes” the AEDPA analysis. Dissent at 35.
    This is word play. Black’s Law Dictionary defines “subsume” as a verb that means “[t]o judge
    as a particular instance governed by a general principle; to bring (a case) under a broad rule.”
    Subsume, Black’s Law Dictionary (11th ed. 2019). “Consume” is a verb for which Black’s
    offers five definitions, none of which apply to the question we face today.5 These definitions
    evidence the dissent’s underlying conceptual error. It uses the word “consumes” to assert that
    AEDPA’s requirements have been improperly eliminated. But the Supreme Court selects its
    5“(1)To destroy the substance of, esp. by fire; to use up or wear out gradually, as by burning or eating . (2) To expend wastefully; to waste; to squander . (3) To use up (time, resources, etc.), whether fruitfully or fruitlessly <45% of the paper we
    consume is recycled>. (4) To eat or drink; to devour . (5) To
    engage the attention or interest of fully; to obsess .”
    Consume, Black’s Law Dictionary (11th ed. 2019).
    No. 17-2267                           Davenport v. MacLaren                                 Page 10
    words and it chose “subsumes”—a word routinely used in law to describe the interplay between
    legal standards—to show that the Brecht test satisfies AEDPA’s requirements, not eliminates
    them. We are bound to accept that choice and apply that word here. A habeas petitioner
    therefore “must satisfy Brecht, and if the state court adjudicated his claim on the merits,
    the Brecht test subsumes the limitations imposed by AEDPA.” Ayala, 
    135 S. Ct. at 2199
    . “[A]
    federal habeas court need not ‘formal[ly]’ apply both Brecht and “AEDPA/Chapman.” 
    Id. at 2198
     (quoting Fry, 
    551 U.S. at
    119–20) (second alteration in original).
    The dissent’s assertion that our Brecht-only framework jettisons the required AEDPA test
    is based on a misunderstanding—it conflates the habeas test that applies to an underlying
    constitutional error with the test that governs the harmlessness of that error. It relies chiefly on
    Yarborough v. Alvarado, 
    541 U.S. 652
     (2004) and Mendoza v. Berghuis, 
    544 F.3d 650
    , 655 (6th
    Cir. 2008) for its failure-to-satisfy AEDPA claim but those cases concerned whether there was
    an underlying constitutional error at the petitioners’ trials. That question, of course, requires
    asking whether the state court’s constitutional analysis “was contrary to, or involved an
    unreasonable application of, clearly established Federal law,” § 2254(d), before turning to
    Brecht.     Here, however, that question need not be asked because the State concedes that
    Davenport’s shackling was constitutional error.
    In fact, this concession neatly reveals the crux of the dissent’s confusion. It supposes that
    this opinion seeks to lay blame on the State for conceding “away AEDPA review of the
    Michigan courts’ harmless error determination.” Dissent at 36. Had the State made that issue-
    concluding concession, there would be no need to dispute the review standard. What the State
    conceded, unlike the cases on which the dissent relies, is the underlying constitutional error. It
    bears repeating—when there is a dispute over whether a constitutional error occurred, of course
    we would apply § 2254(d)’s “contrary to or . . . unreasonable application” of federal law test.
    But the State here concedes constitutional error. That is what makes this case unusual and is the
    reason that we can go straight to Brecht. The dissent’s confusion about what was conceded
    undergirds its misunderstanding of this majority opinion.
    Providing a list of cases, the dissent asserts that we stand alone in applying a Brecht-only
    test and that all other federal circuits addressing the issue “have granted AEDPA deference to a
    No. 17-2267                            Davenport v. MacLaren                              Page 11
    state court’s determination that a constitutional error was harmless,” thus applying both Brecht
    and AEDPA. Dissent at 36–37.        This statement illustrates the dissent’s confusion. Not a single
    case on the dissent’s list contains a habeas petitioner who prevailed under Brecht’s harmlessness
    inquiry and was then required to pass through the gauntlet of a second harmlessness test.
    Orlando v. Nassau Cty. Dist. Attorney’s Office, 
    915 F.3d 113
    , 127 (2d Cir. 2019) (no state court
    harmlessness determination; Brecht-only applied); Johnson v. Lamas, 
    850 F.3d 119
    , 137 (3d
    Cir. 2017) (petitioner failed to meet AEDPA test, therefore “necessarily cannot satisfy” Brecht
    (quoting Ayala, 
    135 S. Ct. 2199
    )); Long v. Pfister, 
    874 F.3d 544
    , 547 (7th Cir. 2017) (en banc)
    (faulting the original panel for failing to “apply the standard of Brecht” to its finding of
    constitutional error but reviewing only the constitutional holding below); Davis v. Grandienard,
    
    828 F.3d 658
    , 666 (8th Cir. 2016) (petitioner failed to meet AEDPA test, no Brecht analysis);
    Rademaker v. Paramo, 
    835 F.3d 1018
    , 1024 (9th Cir. 2016) (petitioner failed to meet AEDPA
    test, therefore “necessarily cannot satisfy” Brecht (quoting Ayala, 
    135 S. Ct. 2199
    )); Malone v.
    Carpenter, 
    911 F.3d 1022
    , 1030 (10th Cir. 2018) (petitioner failed to meet AEDPA test,
    alternatively failed to meet Brecht); Al-Amin v. Warden, 
    932 F.3d 1291
    , 1299 (11th Cir. 2019)
    (petitioner failed Brecht-only test).
    The dissent’s analytical error is perhaps best captured by its parallel citation to Sifuentes
    v. Brazelton, for the proposition that the Ninth Circuit rejects our court’s Brecht-only framework.
    
    825 F.3d 506
    , 534 (9th Cir. 2016). Sifuentes notes that AEDPA applies to the harmlessness
    assessment even after Ayala, and then explains how the Ninth Circuit’s harmless error analysis
    works—the same way as ours:
    In sum, a petitioner “necessarily cannot satisfy” the Brecht requirement of
    showing that he was “actually prejudiced” by the state court’s error . . . “if a
    fairminded jurist could agree with the [state appellate court] that this procedure
    met the Chapman standard of harmlessness.” [Ayala, 135 S. Ct.] at 2199. By the
    same token, if a petitioner does satisfy the Brecht requirement of showing that an
    error resulted in “actual prejudice,” then the petitioner necessarily must have
    shown that the state court’s determination that the error was harmless was
    objectively unreasonable.
    
    Id. at 535
    .
    No. 17-2267                                Davenport v. MacLaren                                           Page 12
    There is no dispute that both Brecht and AEDPA must be satisfied for a habeas petitioner
    to show that a constitutional error was not harmless. The Supreme Court’s statement in Fry that
    the Brecht test subsumes the limitations imposed by AEDPA and our acknowledgment in Ruelas
    that the Brecht test handles the work of both tests show that our test accomplishes that principle.
    The critical point missed in the dissent’s analysis is that in the harmless error context, it is
    significantly harder for a habeas petitioner to meet Brecht’s actual prejudice standard than
    Chapman’s defendant-friendly standard or, in other words, easier for the State to prevail under
    Brecht than under AEDPA/Chapman. So much so that where a state court finds an error
    harmless under Chapman and the defendant is later able to surmount the imposing Brecht hurdle,
    the state court’s Chapman analysis (even though insulated by AEDPA deference) is necessarily
    objectively unreasonable under Harrington v. Richter, 
    562 U.S. 86
    , 103 (2011). See Ayala,
    
    135 S. Ct. at
    2198–99. The tests of Brecht and AEDPA/Chapman then both seek traces of the
    same poison but Brecht’s test covers both because it requires the petitioner to show enough
    poison to be fatal under either test.6 This is why the Court in Ayala held that “[i]n sum, a
    prisoner who seeks federal habeas corpus relief must satisfy Brecht, and if the state court
    adjudicated his claim on the merits, the Brecht test subsumes the limitations imposed by
    AEDPA.” Ayala, 
    135 S. Ct. at 2199
    . This standard of review, moreover, was a point of
    agreement among the Justices in Ayala.7 As a result, even though Brecht predated AEDPA, a
    6This   makes sense because Brecht intended to place a higher burden on a habeas petitioner at collateral
    review than on direct appeal. 
    507 U.S. at 637
    . As the dissent correctly notes, the habeas petitioner in Fry argued for
    the AEDPA/Chapman test (instead of Brecht’s) because it would have lowered his burden. Dissent at 31–32. The
    dissent is correct that AEDPA did not make relief more available to petitioners, but it overlooks the Fry Court’s
    rationale that Brecht is the test because AEDPA did not intend to lower the burden on petitioners and Brecht is
    harder on petitioners than AEDPA/Chapman. Fry, 
    551 U.S. at
    119–20. The point stands that the AEDPA/Chapman
    test is easier. “[I]t is implausible that, without saying so, AEDPA replaced the Brecht standard . . . with the more
    liberal AEDPA/Chapman standard.” 
    Id.
    7In   her dissent in Ayala, Justice Sotomayor explained:
    My disagreement with the Court does not stem from its discussion of the applicable standard of
    review, which simply restates the holding of Fry. Fry rejected the argument that [AEDPA]
    compels federal courts to apply any standard other than that set forth in Brecht, when assessing the
    harmlessness of a constitutional error on habeas review. . . . In addition to confirming
    the Brecht standard’s continued vitality, Fry established its exclusivity. Fry expressly held that
    federal habeas courts need not first assess whether a state court unreasonably applied Chapman
    before deciding whether that error was prejudicial under Brecht. Such a requirement would “make
    no sense when the latter standard obviously subsumes the former.” Nothing in the Court’s opinion
    today calls into question this aspect of Fry’s holding. If a trial error is prejudicial under Brecht’s
    No. 17-2267                               Davenport v. MacLaren                                          Page 13
    habeas court may choose to start with Brecht because AEDPA deference may be exacted through
    Brecht’s demanding standard. Id. at 2199. Brecht, moreover, not only contains AEDPA’s
    stringent commands of deference to state court merit determinations but also its spirit of
    federalism, comity, and finality.8 Fry, 
    551 U.S. at 116
    . We therefore turn to the question of
    whether Davenport’s shackling had a “substantial and injurious effect or influence” on the jury’s
    verdict. See Brecht, 
    507 U.S. at 637
    .
    B. The Law of Shackling
    “The law has long forbidden routine use of visible shackles during the guilt phase” of a
    criminal trial. Deck, 
    544 U.S. at 626
    . After discussing the roots of this precept in the common
    law, Deck summarized the near-universal consensus of lower courts and commentators that “a
    criminal defendant has a right to remain free of physical restraints that are visible to the jury; that
    the right has a constitutional dimension; but that the right may be overcome in a particular
    instance by essential state interests such as physical security, escape prevention, or courtroom
    decorum.” 
    Id. at 628
    . The Supreme Court concluded, “the Fifth and Fourteenth Amendments
    prohibit the use of physical restraints visible to the jury absent a trial court determination, in the
    exercise of its discretion, that they are justified by a state interest specific to a particular trial.”
    
    Id. at 629
    .
    standard, a state court’s determination that the error was harmless beyond a reasonable doubt is
    necessarily unreasonable.
    Ayala, 
    135 S. Ct. at 2211
     (Sotomayor, J., dissenting) (quoting Fry, 
    551 U.S. at 120
    ) (brackets and citations omitted).
    See also Jimerson v. Payne, 
    957 F.3d 916
    , 929–30 (8th Cir. 2020) (adopting Justice Sotomayor’s rule statement for
    the harmlessness inquiry). This shows that the presence of the Eighth Circuit on the dissent’s list of circuits that
    require application of two tests is another mistaken entry.
    8In fact, counter to the dissent’s assertion that Brecht only does the work of AEDPA/Chapman where there
    was no Chapman analysis by the state court, dissent at 34–35, Fry held that precisely because Brecht performs the
    work of “finality, comity, and federalism. . . . Brecht’s applicability does not turn on whether the state appellate
    court recognized the constitutional error and reached the Chapman question.” Fry, 
    551 U.S. at
    117–18 (“[These]
    weighty reasons given in Brecht for applying a less onerous standard [on the State] on collateral review . . . hav[e]
    nothing to do with whether the state court actually applied Chapman.”). Therefore, “[i]n this Circuit, Brecht is the
    standard for reviewing all (non-structural) errors on collateral review; it applies whether or not the state appellate
    courts recognized the error.” Ruelas, 
    580 F.3d at 411
    . Ayala, moreover, makes clear that Fry’s holding that Brecht
    subsumes AEDPA deference, applies even where there is a harmless error determination by the state court. 
    135 S. Ct. at 2199
    .
    No. 17-2267                          Davenport v. MacLaren                                   Page 14
    This right is rooted in “three fundamental legal principles.” Id. at 630. First is the
    presumption of innocence, which “lies at the foundation of the administration of our criminal
    law.” Id. (quoting Coffin v. United States, 
    156 U.S. 432
    , 453 (1895)). “Visible shackling
    undermines the presumption of innocence and the related fairness of the factfinding process. It
    suggests to the jury that the justice system itself sees a ‘need to separate a defendant from the
    community at large.’” 
    Id.
     (quoting Holbrook, 
    475 U.S. at 569
    ) (citation omitted). Second is the
    right to counsel: “Shackles can interfere with the accused’s ‘ability to communicate with his
    lawyer’” and his “ability to participate in his own defense.” 
    Id. at 631
     (quoting Illinois v. Allen,
    
    397 U.S. 337
    , 344 (1970) (citations omitted). Finally, there is the need to preserve the integrity
    of the judicial process. 
    Id.
     “[T]he use of shackles at trial ‘affront[s]’ the ‘dignity and decorum
    of judicial proceedings that the judge is seeking to uphold.’” 
    Id.
     (second alteration in original)
    (quoting Allen, 
    397 U.S. at 344
    ).
    For these reasons, “shackling is ‘inherently prejudicial.’” Id. at 635 (quoting Holbrook,
    
    475 U.S. at 568
    ). “Thus, where a court, without adequate justification, orders the defendant to
    wear shackles that will be seen by the jury,” the defendant’s due process rights are violated. 
    Id.
    Here, the State of Michigan does not dispute that there was no on-the-record justification for the
    shackling and therefore it was unconstitutional. The State argues instead that this error was
    harmless under the standards applicable on habeas review.
    Certainly, Davenport is not entitled to habeas relief simply because he was
    unconstitutionally shackled. As noted above, we must also find that the shackling error “had
    substantial and injurious effect or influence in determining the jury’s verdict.” Brecht, 
    507 U.S. at 637
    . Habeas petitioners “are not entitled to habeas relief based on trial error unless . . . it
    resulted in ‘actual prejudice.’” 
    Id.
          But “[t]he inquiry cannot be merely whether there was
    enough to support the result, apart from the phase affected by the error. It is rather, even so,
    whether the error itself had substantial influence [on the jury’s decision]. If so, or if one is left in
    grave doubt, the conviction cannot stand.” McCarley, 801 F.3d at 665 (quoting O’Neal,
    513 U.S. at 438) (emphasis in original). The State “has the burden of proof in this analysis.”
    Stumpf v. Robinson, 
    722 F.3d 739
    , 752 (6th Cir. 2013) (en banc); see also Rosencrantz v. Lafler,
    
    568 F.3d 577
    , 590 (6th Cir. 2009) (Under Brecht, “[t]he state bears responsibility for showing
    No. 17-2267                         Davenport v. MacLaren                                Page 15
    that the error had no effect on the verdict.”). “Where things are ‘evenly balanced,’ O’Neal
    instructs that the state bears the ‘risk of equipoise.’” Reiner, 955 F.3d at 556 (first quoting
    O’Neal, 513 U.S. at 435) (citation omitted).
    The high bar of actual prejudice highlights the error in the dissent’s claim that our
    Circuit’s Brecht-only framework “pays no respect” to our sister state courts and “sets us apart
    from every other circuit court to have addressed” harmless error issues in habeas. Dissent at 39.
    The dissent’s inclusion of the Ninth Circuit on its list again illustrates its misunderstanding. The
    Ninth Circuit recently held that “[t]he Brecht standard is so stringent that it ‘subsumes’ the
    AEDPA/Chapman standard for review of a state court determination of the harmlessness of a
    constitutional    violation.    We    need     not   apply   both   a   Brecht   review    and   an
    AEDPA/Chapman review because ‘a determination that the error resulted in ‘actual prejudice’
    under Brecht necessarily means that the state court’s harmless error determination was not
    merely incorrect, but objectively unreasonable.’” Hall v. Haws, 
    861 F.3d 977
    , 992 (9th Cir.
    2017) (quoting Deck v. Jenkins, 
    814 F.3d 954
    , 985 (9th Cir. 2016) (brackets and citations
    omitted)).
    Hall shows again that the dissent’s list of circuit cases applying AEDPA/Chapman
    merely supports the undisputed conclusion that courts may choose to perform harmless error
    analysis under AEDPA/Chapman first, before turning to Brecht’s more onerous inquiry, if
    necessary. The list does not support the requirement of “two separate conclusions” that “the
    courts of appeals have universally accepted,” dissent at 39, nor does it stand for the separate
    proposition that AEDPA requires courts to apply a second analytical framework after applying
    Brecht.      See Jenkins, 814 F.3d at 985 (discussing Ayala and concluding “[a] separate
    AEDPA/Chapman determination is not required”).           The dissent, for another instance, also
    includes the Third Circuit on its list—and yet that court too applies a Brecht-only framework.
    See, e.g., Wharton v. Vaughn, 722 F. App’x 268, 277 (3d Cir. 2018), cert. denied, 
    139 S. Ct. 594
    (2018) (“[A]lthough the District Court reviewed [the state court’s harmlessness assessment]
    under § 2254(d), our review here will focus on whether he has met the Brecht standard.”). In
    sum, this opinion does not flout AEDPA or misapprehend Supreme Court precedent, dissent at
    No. 17-2267                         Davenport v. MacLaren                                 Page 16
    39, and it does respect the judgment of our sister circuits, which routinely apply a test
    comparable to our own.
    Applying Brecht, we turn to cases that analyze harmlessness by assessing the weight of a
    shackling error in light of the evidence presented. For example, in Robinson v. Gundy, we held
    that the unconstitutional shackling was harmless error because “[t]he evidence against Robinson
    was truly overwhelming.” 174 F. App’x 886, 893 (6th Cir. 2006). Similarly, in Lakin v. Stine,
    our circuit determined that the shackling error was harmless because “the evidence of guilt is
    overwhelming.” 
    431 F.3d 959
    , 966 (6th Cir. 2005). These cases were preceded by Ruimveld v.
    Birkett, which held that the petitioner proved his entitlement to habeas relief by “showing the
    harm to the presumption of innocence that the Supreme Court has found to be inherent in indicia
    of guilt such as shackles, by showing that there was no good reason for the shackling, by
    showing that his was a close case based on purely circumstantial evidence,” and by noting that
    the State had failed to show any factors (other than his imprisonment) that would make the
    shackles less prejudicial. 
    404 F.3d at
    1017–18.
    Taken together, these cases stand for the proposition that the shackling of a defendant
    without justification is highly prejudicial if viewed by the jury because it vitiates the presumption
    of innocence and undermines the fairness of the factfinding process. Indeed, the Supreme Court
    in Deck instructed that “shackling is ‘inherently prejudicial.’” Deck, 
    544 U.S. at 635
     (quoting
    Holbrook, 
    475 U.S. at 568
    ). The dissent mistakenly views the Supreme Court’s conclusion as
    limited to direct review. Dissent at 43. Of course, in Davenport’s habeas case, unlike Deck’s
    direct appeal, actual prejudice is a prerequisite to relief. Deck, 
    544 U.S. at 635
    . That is precisely
    our position in the disagreement with the dissent over the standard of review.              But the
    “inherently prejudicial” nature of visible shackling, as determined by the Supreme Court, does
    not rise and fall with the standard of review. Instead, as the Supreme Court itself explained,
    “[t]hat statement is rooted in our belief that the practice will often have negative effects, but—
    like ‘the consequences of compelling a defendant to wear prison clothing’ or of forcing him to
    stand trial while medicated—those effects ‘cannot be shown from a trial transcript.’”             
    Id.
    (quoting Riggins v. Nevada, 
    504 U.S. 127
    , 137 (1992)).
    No. 17-2267                                Davenport v. MacLaren                                           Page 17
    The governing caselaw reveals Brecht’s “actual prejudice” standard in action.
    AEDPA/Chapman and Brecht provide two mechanisms for assessing harmless error. Brecht’s is
    “broader.” Ruelas, 
    580 F.3d at 412
    . While error found under Brecht necessarily means error
    under AEDPA/Chapman, the opposite cannot be said. As the Eleventh Circuit (another court
    included by the dissent on its list) correctly explained, “if a petitioner satisfies
    the Brecht standard, he necessarily also satisfies the AEDPA standard, though the reverse is not
    true.” Hammonds v. Comm’r, Alabama Dep’t of Corr., 712 F. App’x 841, 850 (11th Cir.
    2017), cert. denied sub nom. Hammonds v. Dunn, 
    139 S. Ct. 106
     (2018) (citation omitted).9 By
    this same token, if a petitioner cannot prevail under AEDPA/Chapman, he cannot prevail under
    Brecht. 
    Id.
     This is why courts, including this one—and many on the dissent’s list—often test
    the waters under AEDPA/Chapman before turning to Brecht’s more demanding inquiry. See
    Stewart v. Trierweiler, 
    867 F.3d 633
    , 636–38 (6th Cir. 2017).10
    The dissent’s misunderstanding of this point is yet again illustrated by the case it holds up
    as “particularly instructive.”          Dissent at 37.        In Malone v. Carpenter, the Tenth Circuit
    interpreted Ayala’s instruction that “although a federal court reviewing a state [court merits
    decision under Chapman] need not ‘formally apply both Brecht and AEDPA,’ AEDPA still ‘sets
    forth a precondition to the grant of habeas relief.’” 911 F.3d at 1030 (quoting Ayala, 
    135 S. Ct. at 2198
    ).      The Malone Court continued: “as we understand the Court, satisfaction of the
    AEDPA/Chapman standard is a necessary condition for relief (that is, failure to satisfy the
    standard requires denial of relief), but satisfaction of the standard is not a sufficient condition for
    relief because Brecht must also be satisfied.” 
    Id.
     (emphasis in original)).
    Malone did not say, nor could it, that satisfaction of the Brecht standard (which
    incorporates AEDPA/Chapman) is not a sufficient condition for relief. It is. The Supreme Court
    has said so. Ayala, 
    135 S. Ct. at
    2198–99.                  Though courts may choose to apply both, see
    9Hammonds    also noted: “[t]he majority of the federal courts of appeals directly apply the Brecht test rather
    than first determining whether a petitioner meets the AEDPA standard.” Hammonds v. Comm’r, Alabama Dep’t of
    Corr., 712 F. App’x 841, 850 (11th Cir. 2017), cert. denied sub nom. Hammonds v. Dunn, 
    139 S. Ct. 106
     (2018).
    10The  dissent cites Stewart as an emblem of proper harmless error analysis at the habeas stage. But there
    the petitioner failed to get past AEDPA/Chapman and we affirmed on that issue; there was no reason to reach
    Brecht. 
    Id. at 638
    . The Brecht-only approach taken here fully accords with Stewart.
    No. 17-2267                        Davenport v. MacLaren                               Page 18
    Malone, 911 F.3d at 1037; Ayala, 
    135 S. Ct. at 2208
    , it is not necessary. 
    Id.
     at 2198–99. And
    since Malone, the Tenth Circuit has evidenced its agreement with that principle by applying a
    Brecht-only framework. In Harmon v. Sharp, for example, it held: “[o]n habeas review, we may
    only hold that a constitutional error was not harmless if, after applying de novo review, we
    determine that the error ‘had substantial and injurious effect or influence in determining the
    jury’s verdict,’” 
    936 F.3d 1044
    , 1081 (10th Cir. 2019) (quoting Brecht, 
    507 U.S. at 637
    ; citing
    Ayala, 
    135 S. Ct. at 2198
    ). And contrary to the position taken by the dissent, Harmon did not
    otherwise apply an AEDPA/Chapman test to the harmlessness analysis performed by the state
    court. Id.; see also Harmon v. State, 
    248 P.3d 918
    , 933 (Okla. Crim. App. 2011) (applying
    Chapman).      In Coddington v. Sharp, the Tenth Circuit ratified its conclusion that Brecht
    necessarily includes an assessment of whether the state court’s “harmlessness determination was
    itself unreasonable.” 
    959 F.3d 947
    , 953, 957 (10th Cir. 2020) (quoting Ayala, 
    135 S. Ct. at 2199
    ).
    The dissent does not—and cannot—cite a single court of appeals case where a habeas
    petitioner would have prevailed under Brecht but then lost under AEDPA/Chapman. The circuit
    court cases it cites merely reveal how the application of Brecht satisfies AEDPA/Chapman.
    Some of the cases cited by the dissent begin with AEDPA/Chapman, and only get to Brecht if
    necessary; others on its list—as specifically authorized by the Supreme Court—go straight to
    Brecht. No case on the dissent’s list completes Brecht’s harmless error inquiry only to repeat the
    analysis under a different test.
    Applying the Brecht standard set out by the Supreme Court and employed in our circuit,
    we examine the strength of the evidence against Davenport.
    C. Substantial and Injurious Effect or Influence
    1. Evidence of Guilt
    The State argues that the evidence Davenport committed some degree of murder was
    overwhelming. Davenport conceded that he killed White but argued that he did so in self-
    defense. Yet unrebutted expert testimony explaining that strangulation requires several minutes
    to kill is inconsistent with his self-defense claim. On the other hand, the evidence showing the
    No. 17-2267                             Davenport v. MacLaren                                       Page 19
    premeditation and deliberation required for first-degree murder is considerably weaker. To
    resolve the core dispute in this case, it is necessary to delve briefly into the Michigan caselaw
    that governs.
    Michigan law draws a sharp distinction between first-degree and second-degree murder.
    “First-degree and second-degree murder are separate offenses, carrying vastly different penalties,
    distinguished only by the requirement that a homicide punishable as first-degree murder be
    committed with premeditation and deliberation.” People v. Morrin, 
    187 N.W.2d 434
    , 448–49
    (Mich. Ct. App. 1971); see also Mich. Crim. J.I. 16.6 (comparing the elements of first-degree
    and second-degree murder).11 For this reason, “premeditation and deliberation must be given
    independent meaning in a prosecution for first-degree murder.” Morrin, 
    187 N.W.2d at 449
    .
    The “legislative classification of murder into two degrees would be meaningless if ‘deliberation’
    and ‘premeditation’ were construed as requiring no more reflection than may be involved in the
    mere formation of a specific intent to kill.” People v. Hoffmeister, 
    229 N.W.2d 305
    , 307 (Mich.
    1975) (quoting People v. Anderson, 
    447 P.2d 942
    , 948 (Cal. 1968)). On the contrary, the use of
    the terms “deliberate” and “premeditated” denotes that first-degree murder requires
    “substantially more reflection on and comprehension of the nature of the act than the mere
    amount of thought necessary to form the intent to kill.” People v. Plummer, 
    581 N.W.2d 753
    ,
    757 (Mich. Ct. App. 1998).
    Michigan caselaw reveals two prerequisites for a finding of premeditation and
    deliberation. First, “[w]hile the minimum time necessary to exercise this process is incapable of
    exact determination, the interval between initial thought and ultimate action should be long
    enough to afford a reasonable man time to subject the nature of his response to a second look.”
    People v. Tilley, 
    273 N.W.2d 471
    , 473–74 (Mich. 1979) (quoting People v. Vail, 
    227 N.W.2d 535
     (1975)). And, second, there must be evidence that the defendant had—and took—a moment
    for calm reflection before the murder: “[W]hen a homicide occurs during a sudden affray . . . it
    would be ‘a perversion of terms to apply the term deliberate to any act which is done on a sudden
    11Compare     
    Mich. Comp. Laws Ann. § 750.316
     (first-degree murder), with 
    id.
     § 750.317 (second-degree
    murder). Though there are other types of first-degree murder under Michigan law, such as felony murder or murder
    of a law enforcement officer, Davenport was accused only of first-degree murder involving a “deliberate, and
    premeditated killing.” Id. § 750.316(a).
    No. 17-2267                              Davenport v. MacLaren                                         Page 20
    impulse.’” Id. at 473 (quoting Nye v. People, 
    35 Mich. 16
    , 19 (1876)). Thus, “[w]hen the
    evidence establishes a fight and then a killing, there must be a showing of ‘a thought process
    undisturbed by hot blood’ . . . . The critical inquiry is not only whether the defendant had the
    time to premeditate, but also whether he had the capacity to do so.” Plummer, 
    581 N.W.2d at 757
     (quoting Morrin, 
    187 N.W.2d at 449
    ); accord 3A Gillespie Mich. Crim. L. & Proc. § 91:12
    (2d ed. 2020). For these reasons, the Michigan Supreme Court has held that the mere fact that a
    defendant repeatedly stabbed the decedent, inflicting “a great many wounds,” was insufficient
    evidence of premeditation and deliberation to support a conviction for first-degree murder.
    Hoffmeister, 229 N.W.2d at 307–08. The court concluded: “There is no basis on this record for
    an inference that between the successive, potentially lethal blows the killer calmly, in a cool state
    of mind, ‘measured and evaluated’ and subjected ‘the nature of his response to a second look.’”
    Id. (quoting Morrin, 
    187 N.W.2d at 449
    ).12
    In this case, the amount of time the strangling must have taken is the only evidence of
    premeditation and deliberation the prosecution pointed to in its closing argument. Certainly, the
    duration of the strangling demonstrates the intent to kill—or at least the intent to cause great
    bodily harm—required for second-degree murder. See Mich. Crim. J.I. 16.5. But it is not
    definitive proof of premeditation or deliberation. “[E]vidence of manual strangulation can be
    used as evidence that a defendant had an opportunity to take a ‘second look.’” People v.
    Johnson, 
    597 N.W.2d 73
    , 79 (Mich. 1999) (quoting People v. Furman, 
    404 N.W.2d 246
    , 249–50
    (Mich. Ct. App. 1987)). Yet, by itself, it is not conclusive: “[N]either the brutal nature of a
    killing nor manual strangulation alone is sufficient to show premeditation . . . .” 
    Id.
     That is
    especially true when, as here, there is substantial evidence that the strangulation occurred in the
    course of a fight and thus “the homicide occurred during an affray whose nature would not
    12The   Michigan Supreme Court recently characterized Hoffmeister as a case where “there was no basis for
    the jury to conclude that the defendant had adequate time for a ‘second look’” because “the only evidence presented
    was the number of stab wounds.” People v. Oros, 
    917 N.W.2d 559
    , 567 (Mich. 2018). In Oros, the court held that
    the evidence of deliberate and premeditated first-degree murder was legally sufficient because there was additional
    evidence besides the number of stab wounds. 
    Id.
     at 567–70. Oros distinguished Hoffmeister; it did not overrule it.
    See 
    id. at 570
     (“Our holding is consistent with Hoffmeister as we do not hold today that the sheer number of stab
    wounds alone established the elements of premeditation and deliberation.”). And without regard to Oros,
    Hoffmeister was good law at the relevant time—Davenport’s trial was in 2008.
    No. 17-2267                          Davenport v. MacLaren                              Page 21
    permit cool and orderly reflection.” Plummer, 
    581 N.W.2d at 757
     (quoting Morrin, 
    187 N.W.2d at 450
    ).
    The record contains substantial evidence that a fight took place. One witness testified
    that he asked White to leave his house at about 2:30 a.m. because she was “agitated” and
    “getting crazy” after smoking some crack cocaine. Other witnesses testified that White would
    get angry when she smoked crack cocaine and that she was “a spitfire” who had a reputation for
    fighting. Even the evidence that the State introduced of Davenport’s admission that he killed
    White is consistent with this story: A prosecution witness testified that Davenport told him
    White “kept coming back at him and it just got out of hand, and that’s when he offed her.”
    Based on the evidence presented, the only time Davenport could have engaged in the
    requisite period of calm reflection “undisturbed by hot blood” would have been while he was
    strangling White. And the only evidence of premeditation and deliberation the prosecution
    pointed to in its closing was the time that strangulation would have taken. Yet, under Michigan
    law, evidence of manual strangulation alone is not enough to prove premeditation. Johnson, 597
    N.W.2d at 79. The jury easily could have found that this was second-degree murder, not first-
    degree murder, because “the homicide occurred during an affray whose nature would not permit
    cool and orderly reflection.” Plummer, 
    581 N.W.2d at 757
     (quoting Morrin, 
    187 N.W.2d at 450
    ). The evidence of premeditation and deliberation was therefore not overwhelming.
    The closeness of the case is further demonstrated by the duration of the jury’s
    deliberations. In Ruimveld, we observed that the case is not “open-and-shut,” giving as partial
    support that “the jury deliberated for over three hours despite the simple facts.” 
    404 F.3d at 1016
    . This case was arguably even simpler; given that Davenport admitted killing White, the
    only disputed fact at trial was his state of mind, the critical component of first-degree murder.
    Yet the jury still deliberated for approximately six hours, through one afternoon and into the next
    morning.
    As to the merits, the dissent dismisses the analysis of applicable Michigan cases and law
    as “generalities.”   Dissent at 45.    Instead, the dissent wholly frames its argument on the
    “strikingly similar” case of People vs. Johnson, dissent at 27, which it says results in “the
    No. 17-2267                        Davenport v. MacLaren                                 Page 22
    inescapable conclusion” that Johnson brands Davenport’s conduct as premeditated murder,
    dissent at 44. As noted above, Johnson does have applicability as part of the full examination of
    Michigan law. But employing Johnson requires acknowledging that review by the Michigan
    Supreme Court was granted solely to determine whether it was error to deny Johnson’s motion
    for directed verdict. The question before that Court was whether there was “sufficient evidence
    to justify a rational tryer of fact in finding guilt beyond a reasonable doubt.”          Johnson,
    597 N.W.2d at 75–76.
    That is not the question we ask here. Under Brecht, we ask whether the constitutional
    violation of shackling Davenport throughout the trial had a “substantial and injurious effect or
    influence in determining the jury’s verdict.” Brecht, 
    507 U.S. at 637
    . It is, of course, possible
    that a trial record contains evidence sufficient to convict a defendant of a crime but the jury,
    considering that evidence and making the credibility determinations entrusted to it as the trial’s
    factfinder, votes to acquit. We are aware of no case in which a habeas petitioner claiming a
    constitutional violation at trial has been required to satisfy a directed verdict standard of review
    in order to find that the violation was not harmless.      We have held that we will not find a
    constitutional violation to be harmless simply because the record reflected enough evidence to
    support the jury’s verdict. McCarley, 801 F.3d at 665. In fact, “[t]ime and again, the Supreme
    Court has emphasized that a harmless-error inquiry is not the same as a review for whether there
    was sufficient evidence at trial to support a verdict.” Jensen v. Clements, 
    800 F.3d 892
    , 902 (7th
    Cir. 2015) (collecting cases). Even if there was “‘ample evidence’ to support the conviction,”
    that would be “irrelevant” because “the question before us is not whether there was sufficient
    evidence to support the jury’s conclusions, but rather whether the evidence is so strong that a
    reviewing court can be assured that the shackling did not affect the jury’s conclusions.”
    Ruimveld, 
    404 F.3d at
    1017 n.5. Even if we were to assume that Johnson’s holding dictates a
    finding that the State introduced sufficient evidence to support conviction of Davenport, that is
    not sufficient to show that the unconstitutional shackling was harmless. Thus, Johnson provides
    no “inescapable conclusion” that controls this case. Dissent at 44.
    No. 17-2267                              Davenport v. MacLaren                                           Page 23
    2. Juror Testimony
    In addition to arguing that this is a case of overwhelming evidence, the State tries to meet
    its burden of showing that the shackling did not have a substantial and injurious effect or
    influence on the jury’s verdict by pointing to the testimony of jurors that the shackling did not
    affect their deliberations. The state courts also relied on this testimony in denying Davenport
    relief after the evidentiary hearing. But, as Davenport argues, the Supreme Court has made clear
    that jurors’ subjective testimony about the effect shackling had on them bears little weight. If a
    practice “‘involves such a probability that prejudice will result that it is deemed inherently
    lacking in due process,’” like shackling a defendant without case-specific reasons, “little stock
    need be placed in jurors’ claims to the contrary. Even though a practice may be inherently
    prejudicial, jurors will not necessarily be fully conscious of the effect it will have on their
    attitude toward the accused.” Holbrook, 
    475 U.S. at 570
     (quoting Estes v. Texas, 
    381 U.S. 532
    ,
    542–43 (1965) (citations omitted). Since Holbrook was decided, a voluminous body of social-
    science research has demonstrated support for Holbrook’s conclusion.13 The dissent finds this
    footnoted evidence to be improper on the basis that our review must be guided only by
    established Supreme Court law, “not abstract sociology.” Dissent at 42. But it was the Supreme
    Court in Holbrook that stated the danger of relying on after-the-fact juror conclusions regarding
    “inherently prejudicial” actions such as shackling because jurors may be not fully aware of how
    such effects “their attitude toward the accused.” Holbrook, 
    475 U.S. at 570
    . This scientific
    evidence merely provides further support for the Supreme Court’s determination.
    Moreover, two aspects of the jurors’ factual testimony at the evidentiary hearing suggest
    that Davenport’s shackling may have prejudiced his trial. First, when questioned three years
    after the trial, a majority of jurors still remembered that they either saw his restraints or heard
    13See generally Jennifer L. Eberhardt, Biased: Uncovering the Hidden Prejudice that Shapes What We See,
    Think, and Do (2019). This research suggests that the shackling of Davenport, a 6’5” tall black man weighing
    approximately 300 pounds, would tend to “prime” racialized presumptions of dangerousness and guilt. See, e.g.,
    Mark W. Bennett & Victoria C. Plaut, Looking Criminal and the Presumption of Dangerousness: Afrocentric Facial
    Features, Skin Tone, and Criminal Justice, 
    51 U.C. Davis L. Rev. 745
    , 785 (2018) (“Repeated studies indicate
    Blacks with darker skin tones and stronger Afrocentric facial features ‘activate automatic associations with negative
    behavioral stereotypes of Black men, such as aggression, violence, and criminality.’” (citations omitted)); Justin D.
    Levinson et al., Guilty by Implicit Racial Bias: The Guilty/Not Guilty Implicit Association Test, 
    8 Ohio St. J. Crim. L. 187
    , 207 (2010) (“[W]e found that participants held implicit associations between Black and Guilty. . . . [T]hese
    implicit associations were meaningful—they predicted judgments of the probative value of evidence.”).
    No. 17-2267                         Davenport v. MacLaren                                Page 24
    another juror remark on his shackles. This suggests the shackles made an impression. Second,
    several jurors testified that they thought Davenport might be dangerous when they saw his
    restraints.   The dissent asserts that there is “little reason to believe” that juror testimony
    concerning whether Davenport was dangerous was due to his partial shackling, as opposed to the
    “gruesome” killing at issue in this case. Dissent at 45.      This statement misreads the record;
    those jurors were not making a holistic assessment of whether Davenport was dangerous in light
    of all that they knew about him. Rather, the jurors were asked specifically whether Davenport’s
    shackling left them with the impression that he was dangerous at the time they observed the
    restraints.
    Juror James Vanderveen testified as follows:
    Q:      And given that you saw the restraints at some portions of the trial, did you
    think that [Davenport] might be dangerous?
    A:      Well I would assume that, yes.
    Q:      Okay. Did you think that he had done something wrong and that is why
    he was shackled?
    A:      Well it was a murder trial, correct?
    Similarly, Juror Robert Jankford stated that he noticed the shackles during the trial and
    thought that the purpose of the shackles was “[s]ecurity.” Mr. Jankford was then asked, “[d]id
    you think that he might be dangerous?” Mr. Jankford replied “[a]bsolutely.” Juror Bradley
    Lewis described how, initially, he did not notice the shackles on Davenport, but that a different
    juror pointed out the shackles to him and other jurors while the jurors were sitting in the jury box
    during the middle of the trial.
    The fact that Davenport stood charged with, and was later convicted of, a violent crime
    does not provide a reason to write off the jurors’ explanations of their impressions concerning
    Davenport’s shackles.        If anything, they underscore the due process concerns that
    unconstitutional shackling raises. Leaving jurors with an impression that a defendant has already
    been determined to be dangerous is particularly troublesome when that defendant is charged with
    a crime a jury might expect a dangerous person to commit. Thus, the charges at issue in this case
    do not excuse the error created by the unconstitutional shackling, they exacerbate it.
    No. 17-2267                         Davenport v. MacLaren                                   Page 25
    That these jurors did not attribute great significance to the shackles, does not mean the
    shackling had no effect. As the Supreme Court has observed, shackling a defendant “almost
    inevitably implies to a jury, as a matter of common sense, that court authorities consider [him] a
    danger to the community . . . . [I]t thereby inevitably undermines the jury’s ability to weigh
    accurately all relevant considerations . . . .” Deck, 
    544 U.S. at 633
     (discussing the penalty phase
    of capital trials). Thus, “common sense” teaches that a defendant in shackles is in peril of being
    presumed dangerous rather than presumed innocent.
    The Ninth Circuit relied on this logic in granting a habeas petition in a case with
    analogous facts. Rhoden v. Rowland was a case where, much like this one, “several of the jurors
    actually saw the shackles during the trial” and “[a]t least two jurors remember[ed] other jurors
    making comments to them about the shackles.” 
    172 F.3d 633
    , 637 (9th Cir. 1999). As in this
    case, the defendant “was charged with violent crimes” and the evidence was disputed—indeed,
    “the jurors deliberated for over nine hours over three days.” 
    Id.
     The court concluded, “[b]ecause
    at least some of the jurors saw the shackles and because the shackles essentially branded Rhoden
    as having a violent nature in a case where his propensity for violence was the crucial issue, the
    shackles had ‘substantial and injurious effect or influence in determining the jury’s verdict’ . . . .”
    
    Id.
     (quoting Brecht, 
    507 U.S. at 637
    ). The shackling was therefore not harmless error. 
    Id.
    Here, similarly, the shackles branded Davenport as having a violent nature in a case
    where the crucial point of contention was whether he engaged in deliberate and premeditated
    murder. Given the closeness of this question, the number of jurors who observed the restraints,
    and the inherently prejudicial nature of shackling, the State has failed to carry its burden to show
    that the shackles did not have a “substantial and injurious effect or influence in determining the
    jury’s verdict.” Brecht, 
    507 U.S. at 637
    ; see Ruimveld, 
    404 F.3d at 1018
    ; Rhoden, 
    172 F.3d at 637
    .
    III. CONCLUSION
    For the foregoing reasons, we REVERSE the district court’s denial of Davenport’s
    § 2254 petition, GRANT Davenport a conditional writ of habeas corpus that will result in his
    release from prison unless the State of Michigan commences a new trial against him within 180
    No. 17-2267                    Davenport v. MacLaren                           Page 26
    days from the date of this opinion and REMAND the case for further proceedings consistent
    with this opinion.
    No. 17-2267                        Davenport v. MacLaren                                 Page 27
    _________________
    DISSENT
    _________________
    CHAD A. READLER, Circuit Judge, dissenting. In the federal courts, it is hard to
    imagine a habeas case where AEDPA (more formally known as The Antiterrorism and Effective
    Death Penalty Act) does not help guide our review. Enacted by Congress in 1996, AEDPA
    brought meaningful change to how habeas claims are treated by the federal courts. Short of
    declaring the law unconstitutional, we are bound to enforce it. And we have, all across the
    habeas case-law landscape.
    But in today’s decision granting habeas relief, AEDPA is conspicuously absent. As we
    sit in review of the Michigan courts’ judgment that any error in Davenport’s state court
    proceedings was harmless, we would naturally apply 
    28 U.S.C. § 2254
    (d)(1), as amended by
    AEDPA. Yet the majority opinion fails to ask the fundamental question posed by § 2254(d)(1):
    Whether the Michigan courts’ determination “was contrary to, or involved an unreasonable
    application of, clearly established Federal law”? Id.
    Failing to do so puts us at odds with Davis v. Ayala, 
    135 S. Ct. 2187
     (2015). Ayala
    reminds us of the two inquiries a federal habeas court must make when assessing the impact of a
    constitutional error in a collateral state court proceeding. One, did the error have a “substantial
    and injurious effect or influence in determining the jury’s verdict,” the collateral review standard
    announced in Brecht v. Abrahamson, 
    507 U.S. 619
    , 623 (1993)? Ayala, 
    135 S. Ct. at
    2198–99
    (internal quotation marks omitted). Two, honoring AEDPA, was the state court’s “harmlessness
    determination itself” an “unreasonable application of clearly established federal law”?          
    Id.
    (emphasis in original). The latter is particularly important here, when the Supreme Court has
    neither previously found a constitutional violation in a comparable setting nor held that a state
    court may not consider post-trial juror testimony in concluding that a constitutional error was
    harmless. See Deck v. Missouri, 
    544 U.S. 622
     (2005) (holding that the routine use of physical
    restraints fully visible to the jury violates due process); Holbrook v. Flynn, 
    475 U.S. 560
     (1986)
    (addressing whether potential jurors hypothetically would be prejudiced by subsequent
    No. 17-2267                        Davenport v. MacLaren                                 Page 28
    courtroom security measures). The Brecht inquiry may “subsume” some AEDPA requirements.
    Ayala, 
    135 S. Ct. at 2198
    . But, Ayala makes clear, it does not consume them altogether.
    To my eye, the majority decision entrenches us as the lone circuit to grant habeas relief
    from a state court judgment without applying AEDPA deference to that court’s conclusion that a
    trial error was harmless. In the wake of Ayala, every other circuit to reach the question has
    agreed that a habeas petitioner, before he may be granted habeas relief, must satisfy the distinct
    requirements of both Brecht and AEDPA. We have done the same, just not today. See Stewart
    v. Trierweiler, 
    867 F.3d 633
    , 636–37, 640 (6th Cir. 2017) (applying Ayala and measuring a state
    court’s harmless error analysis against the backdrop of both Brecht and AEDPA before holding
    that the state court’s harmless error determination was not “so lacking in justification that there
    was an error well understood and comprehended in existing law beyond any possibility for
    fairminded disagreement”). It may be that a federal court can deny habeas relief by “go[ing]
    straight to Brecht.” Ruelas v. Wolfenbarger, 
    580 F.3d 403
    , 413 (6th Cir. 2009); see also
    Hollman v. Sprader, 803 F. App’x 841, 843–45 (6th Cir. 2020) (denying habeas relief on
    AEDPA grounds after observing that a federal habeas court may choose to take a “shortcut” to
    Brecht in denying habeas relief). But we stand alone as the only Circuit to award habeas relief
    without expressly applying the requirements of both Brecht and AEDPA. Cf. Reiner v. Woods,
    
    955 F.3d 549
    , 556–57 (6th Cir. 2020) (recognizing a “colorable argument” that Ruelas and its
    progeny are incorrect in light of Ayala).
    Because the majority opinion fails to employ an AEDPA analysis before granting habeas
    relief, because the majority opinion does not cite a Supreme Court decision contrary to the
    harmless error determination reached by the Michigan courts, and because the Michigan
    Supreme Court previously affirmed a strikingly similar first-degree murder conviction in People
    v. Johnson, 
    597 N.W.2d 73
     (Mich. 1999), I respectfully dissent.
    I.      AEDPA Requires Federal Courts To Show Great Deference To State Court
    Adjudications On The Merits, Including Harmless Error Determinations.
    1. I start with a point of agreement.    In recognition of the deference owed to a state
    court’s judgment in a habeas posture, Ayala, the majority opinion, and I all agree that we employ
    the “actual prejudice” standard from Brecht in assessing the impact of a constitutional error on a
    No. 17-2267                        Davenport v. MacLaren                              Page 29
    habeas petitioner’s state proceeding. Guided by the Brecht standard, we may grant habeas relief
    only where there is “grave doubt about whether a trial error of federal law had substantial and
    injurious effect or influence in determining the jury’s verdict.” Ayala, 
    135 S. Ct. at 2198
    (quoting O’Neal v. McAninch, 
    513 U.S. 432
    , 436 (1995)) (internal quotation marks omitted). In
    the words of Ayala, there “must be more than ‘a reasonable probability’ that the error was
    harmful.” 
    Id.
     (quoting Brecht, 
    507 U.S. at 637
    ).
    But that is not all Ayala had to say. The Supreme Court emphasized that AEDPA’s
    requirements are also alive and well when it comes to collateral review of a state court’s
    harmless error analysis. AEDPA’s standards are distinct, and they are stringent. Brecht permits
    a habeas court to grant relief where any “error of federal law” had a prejudicial effect on the
    verdict. AEDPA, on the other hand, permits habeas relief only where the state court reached “a
    decision that was contrary to, or involved an unreasonable application of, clearly established
    Federal law.” 
    28 U.S.C. § 2254
    (d)(1). That is, the rare set of cases for which the “necessity to
    apply” an earlier rule recognized by the Supreme Court “[is] beyond doubt.” Yarborough v.
    Alvarado, 
    541 U.S. 652
    , 666 (2004). And that deferential standard applies both to a state court’s
    merits determination as well as its harmless error assessment. Ayala, 
    135 S. Ct. at
    2198–99.
    AEDPA could thus foreclose relief even in cases in which Brecht’s harmless error standard is
    satisfied—most notably cases in which the purported prejudice is based on a ground for relief not
    yet clearly established by the Supreme Court. Yarborough, 
    541 U.S. at 666
    . The majority fails
    to consider this critical feature of AEDPA.
    Compare Ayala to the majority opinion.        First Ayala.   There, the Supreme Court
    considered in a habeas context a state court’s determination that any constitutional error in the
    collateral state court proceeding was harmless. In so doing, the Supreme Court disagreed with
    the Ninth Circuit’s conclusion that “a state court’s harmlessness determination has no
    significance under Brecht.” 
    135 S. Ct. at 2198
    . Rather, a habeas petitioner, in addition to
    satisfying Brecht, must also satisfy AEDPA, which continues to set forth a precondition on the
    grant of habeas relief. 
    Id.
     Earlier post-AEDPA cases applying Brecht, the Supreme Court
    acknowledged, may have muddied the point. See 
    id.
     (citing Fry v. Pliler, 
    551 U.S. 112
    , 120
    (2007)). But those decisions, Ayala explained, “would have had no possible basis for holding
    No. 17-2267                         Davenport v. MacLaren                                Page 30
    . . . that Brecht somehow abrogates the limitation on federal habeas relief that § 2254(d) plainly
    sets out.” 
    135 S. Ct. at 2198
    .
    Honoring this precondition, Ayala undertook the familiar AEDPA analysis.                 The
    Supreme Court cited the standards set out in § 2254(d) and made clear that a federal court could
    not grant relief in this context unless the state court’s decision “was contrary to or involved an
    unreasonable application of clearly established federal law.” Id. It re-emphasized that the
    “highly deferential AEDPA standard applies,” and noted that, in the harmless error context, a
    federal court must determine whether a state court applied Supreme Court precedent “in an
    objectively unreasonable manner” in denying relief to the petitioner. Id. at 2198–99 (internal
    quotation marks and citations omitted). And it concluded that habeas relief may not issue “under
    § 2254 unless the harmlessness determination itself was unreasonable.” Id. at 2199 (internal
    quotation marks and citations omitted).
    Now the majority opinion.          Virtually all of these AEDPA standards are missing.
    In analyzing whether to grant habeas relief, the majority opinion barely mentions the “highly
    deferential AEDPA standard,” instead claiming that the Brecht standard does all of the landmark
    statute’s work. Invoking that AEDPA-free framework, the majority opinion ultimately pays no
    deference to the Michigan courts’ conclusion that any error in Davenport’s case was harmless.
    People v. Davenport, 
    832 N.W.2d 389
    , 390 (Mich. 2013) (“Given the substantial evidence of
    guilt presented at trial, we cannot conclude that there was an unacceptable risk of impermissible
    factors coming into play.”); People v. Davenport, No. 306868, 
    2012 WL 6217134
    , at *2 (Mich.
    Ct. App. Dec. 13, 2012) (per curiam) (finding that the state trial court properly relied on juror
    testimony regarding prejudice and concluding that “[a]ll of the evidence indicated that the
    shackling did not affect the verdict in any way”). Blatantly disregarding “principles of comity,
    finality, and federalism” in this manner is precisely what AEDPA was crafted to avoid.
    Woodford v. Garceau, 
    538 U.S. 202
    , 206 (2003); see also Williams v. Taylor, 
    529 U.S. 362
    , 386
    (2000) (Through AEDPA, “Congress intended federal judges to attend with the utmost care to
    state-court decisions . . . before concluding that those proceedings were infected by constitutional
    error sufficiently serious to warrant issuance of the writ.”).
    No. 17-2267                        Davenport v. MacLaren                                  Page 31
    Compounding those conspicuous omissions, the majority opinion then fails to ask or
    answer whether the state court’s harmless error determination was an “unreasonable application
    of clearly established Federal law.” 
    28 U.S.C. § 2254
    (d)(1). Ordinarily, we would assess
    whether the Michigan courts’ harmless error analysis “was so lacking in justification that there
    was an error well understood and comprehended in existing law beyond any possibility for
    fairminded disagreement.” Ayala, 
    135 S. Ct. at 2199
     (quoting Harrington v. Richter, 
    562 U.S. 86
    , 103 (2011)). But other than quoting the Michigan Supreme Court’s holding, the majority
    ignores the underlying state court decisions altogether—a textbook example of acting as if “a
    state court’s harmlessness determination has no significance under Brecht.” Id. at 2198.
    2. While Supreme Court precedent leads me ultimately to disagree with my friends in the
    majority, the majority opinion’s conclusion is not without its own precedent. But it is mistaken
    precedent, in my mind, especially in the aftermath of Ayala. The notion that we need pay no
    deference to a state court’s harmless error determination finds its roots in a line of our cases
    starting with Ruelas, 
    580 F.3d at 412
    . In a nutshell, Ruelas held that federal habeas courts
    reviewing state harmless error decisions may push aside AEDPA’s stringent statutory
    requirements in favor of applying only Brecht’s “substantial and injurious effect” test. Where,
    one might ask, did Ruelas find such sweeping authority to close its eyes to an act of Congress?
    From one word in Fry v. Pliler: It “makes no sense to require formal application of both tests
    (AEDPA/Chapman and Brecht) when the latter obviously subsumes the former.” 
    Id.
     (quoting
    Fry, 
    551 U.S. at 120
    ) (emphasis added). See Chapman v. California, 
    386 U.S. 18
    , 24 (1967).
    Embracing the term “subsumes,” Ruelas concluded that the Brecht test fully accounts for all of
    AEDPA’s requirements, effectively reading AEDPA out of existence on collateral harmless error
    review. Ruelas, 
    580 F.3d at 412
    .
    But Ruelas failed to consider Fry in context. Unlike here, and unlike in Ruelas, the
    collateral state court decision at issue in Fry had not passed judgment on harmless error, meaning
    the Supreme Court was not reviewing a state court’s harmless error analysis. Fry addressed a
    different question, namely, how to measure alleged prejudice resulting from a constitutional error
    in collateral proceedings when the error is first recognized on federal habeas review. 
    551 U.S. at 114
    .
    No. 17-2267                        Davenport v. MacLaren                                Page 32
    To answer that question, the Supreme Court had to harmonize its prior decisions in
    Chapman and Brecht in the aftermath of AEDPA. See 
    id.
     Chapman set forth the standard courts
    are to use in assessing constitutional error on direct review: “[B]efore a federal constitutional
    error can be held harmless, the reviewing court must be able to declare a belief that it was
    harmless beyond a reasonable doubt.” 
    386 U.S. at 24
    . The burden of meeting this standard rests
    with the government. See 
    id.
     at 23–24. Brecht, by comparison, addressed the prejudice standard
    federal courts are to employ after finding (or assuming) constitutional error in collateral
    proceedings. It held that federal habeas relief may issue only when a constitutional error actually
    prejudiced the defendant, 
    507 U.S. at
    637–38—in stark contrast to the government’s burden to
    disprove such prejudice on direct review. Chapman, 
    386 U.S. at 24
    .
    And how did AEDPA, enacted only three years after the decision in Brecht, impact the
    interplay between Chapman and Brecht? The petitioner in Fry asserted that, in the aftermath of
    AEDPA, federal courts were to apply in habeas proceedings the direct review standard set forth
    in Chapman through the lens of AEDPA. In other words, the petitioner argued, AEDPA in
    essence replaced the actual prejudice standard under Brecht with a joint AEDPA/Chapman
    standard for claims under § 2254. Adopting that joint standard, however, would have lowered
    the bar for state habeas petitioners in a sense by allowing federal courts to grant habeas relief
    from state court judgments without finding any actual prejudice to the petitioner. That was so
    because Chapman places the burden on the government to disprove prejudice, Chapman, 
    386 U.S. at 24
    , rather than requiring an affirmative showing of prejudice, Brecht, 
    507 U.S. at 637
    .
    What the Fry Court confronted, then, was the counterintuitive notion that AEDPA made
    it easier rather than harder for a petitioner to obtain habeas relief when measuring prejudice
    arising from a purported error. Rejecting that odd result, the Supreme Court applied the more
    restrictive Brecht standard. To hold otherwise would have allowed federal courts to “[overturn]
    final and presumptively correct convictions on collateral review because the State cannot prove
    that an error is harmless under Chapman.” 
    Id.
     In other words, Fry was merely heeding Brecht’s
    warning about expanding collateral review beyond carefully circumscribed limits.
    But that does not mean, contrary to the understanding in Ruelas and today’s majority
    opinion, that if the AEDPA/Chapman prejudice standard is less stringent than Brecht, AEDPA
    No. 17-2267                        Davenport v. MacLaren                                 Page 33
    does nothing at all in the harmless error context. AEDPA, remember, does not simply articulate
    a prejudice standard. It also cabins federal habeas review by preventing habeas courts from
    extending grounds for relief beyond those explicitly required by Supreme Court precedent,
    independent of any prejudice those errors may have caused. In other words, relief under AEDPA
    requires more than prejudice. It also requires habeas courts to extract concrete legal rules from
    Supreme Court precedent, to apply them to the letter, and not to expand them as we might on
    direct review. See Yarborough, 
    541 U.S. at 666
    .
    This fundamental AEDPA principle was front-and-center in Yarborough. There, the
    Supreme Court reviewed the Ninth Circuit’s grant of § 2254 relief based upon a violation of
    Miranda v. Arizona, 
    384 U.S. 436
     (1966). In awarding habeas relief, the Ninth Circuit faulted
    the state court for failing to consider the petitioner’s age. But age was not a factor the Supreme
    Court had explicitly required courts to consider in that context. By requiring the state court to do
    so, the Ninth Circuit extended—rather than applied—existing law, a practice wholly inconsistent
    with AEDPA: “Section 2254(d)(1) would be undermined if [federal] courts introduced rules not
    clearly established under the guise of extensions to existing law.” Yarborough, 
    541 U.S. at 666
    .
    “Evaluating whether a rule application was unreasonable requires considering the rule’s
    specificity,” the Supreme Court explained. 
    Id. at 664
    . The more general the rule, for example,
    “the more leeway courts have in reaching outcomes in case-by-case determinations.” 
    Id.
     But in
    all events, in cases warranting habeas relief, “the necessity to apply the earlier rule will be
    beyond doubt.” 
    Id. at 666
    .
    We followed that tailored approach in the analogous setting of Mendoza v. Berghuis,
    
    544 F.3d 650
     (6th Cir. 2008). Like here, at issue there was a defendant’s shackling at trial. 
    Id. at 655
    . And then, as today, the relevant Supreme Court authority, for purposes of § 2254(d)(1),
    was Deck, 
    544 U.S. at 622
    . In view of the key factual distinctions between Mendoza and Deck,
    “it [was] not obvious that Deck should be extended to the particular facts present” in Mendoza.
    
    544 F.3d at 655
    . AEDPA, moreover, forbids “breaking new ground on unsettled legal issues or
    interpreting existing caselaw to decide an open question in our jurisprudence.” Dewald v.
    Wriggelsworth, 
    748 F.3d 295
    , 300 (6th Cir. 2014) (internal quotations and citations omitted).
    Accordingly, we denied habeas relief in Mendoza. 
    544 F.3d at 655
    .
    No. 17-2267                        Davenport v. MacLaren                               Page 34
    The Brecht standard does not capture this critical feature of AEDPA. AEDPA requires a
    federal habeas court to assess whether Supreme Court precedent put a state court on notice of
    precise constitutional limitations. See Yarborough, 
    541 U.S. at 665
    . Brecht, on the other hand,
    writes largely on a clean slate. Unchecked by then-existing Supreme Court precedent, Brecht
    simply asks a federal habeas court to assess the prejudice arising from an alleged error. And that
    distinction can make all the difference. A habeas claim alleging a deeply prejudicial trial error
    may easily clear Brecht’s “actual prejudice” bar. But the claim may nonetheless fail AEDPA’s
    comity-inspired requirements if the reviewing court must create new law or extend existing
    Supreme Court precedent to find underlying legal error, or that the error was not harmless. 
    Id. at 666
    .
    It follows that before awarding habeas relief today, we must explain why no fairminded
    jurist could find that the differences between this case and the Supreme Court’s holdings in Deck
    (addressing shackling) and Holbrook (addressing whether potential jurors can fairly predict
    whether they will be improperly influenced by courtroom security measures), could justify a
    different outcome. See Yarborough, 
    541 U.S. at
    663–66. Yet the majority opinion—tellingly, to
    my mind—simply refuses to do so. Rather than engaging in this demanding AEDPA analysis,
    the majority opinion brushes it aside, concluding that the AEDPA standard is “subsumed” by
    Brecht. Which begs the question: If, as the majority opinion posits, Davenport’s claim passes
    the “significantly harder” Brecht test, why does the majority opinion not show its work, as the
    state asked us to do, in finding that AEDPA is also satisfied?
    3. Any lingering confusion over whether AEDPA also applies alongside Brecht in the
    context of reviewing a state court’s harmless error determination was put to rest by Ayala. As
    explained above, Ayala repeatedly referenced AEDPA’s standards.            It then applied those
    standards, in addition to applying the Brecht standard, in collaterally reviewing a state court’s
    harmlessness determination. See Ayala, 
    135 S. Ct. at
    2198–99 (noting that habeas relief may not
    issue “under § 2254 unless the harmlessness determination itself was unreasonable”) (internal
    quotation marks and citations omitted); see also id. at 2207 (“The most that Ayala can establish
    is that reasonable minds can disagree about whether the prosecution’s fears were well founded,
    No. 17-2267                        Davenport v. MacLaren                                 Page 35
    but this does not come close to establishing ‘actual prejudice’ under Brecht. Nor does it meet the
    AEDPA standard.”) (emphasis added).
    While Ruelas, decided six years before Ayala, might be excused for believing that habeas
    courts can dispense with AEDPA in the harmless error context, our cases that follow Ayala
    cannot. That starts with McCarley v. Kelley, 
    801 F.3d 652
    , 665 (6th Cir. 2015), and extends
    through O’Neal v. Balcarcel, 
    933 F.3d 618
    , 625 (6th Cir. 2019), and Reiner, 955 F.3d at 556–57,
    our most recent published opinions to follow Ruelas. The majority opinion holds out Reiner in
    particular as a beacon of light in the continuing post-Ayala march to vindicate Ruelas. But to do
    so, the majority opinion must first rewrite Reiner. The majority opinion quotes Reiner as
    follows: “The state argues that the Supreme Court’s subsequent decision in Davis v. Ayala
    changed this dynamic . . . [t]he problem for the state is that our precedent forecloses this
    approach.” 955 F.3d at 556–57. What, one might wonder, has the majority opinion omitted
    through its use of ellipsis? Only Reiner’s Ayala-inspired recognition that the state’s position is a
    “colorable argument,” but foreclosed by our Circuit precedent.         Id.   In that sense, Reiner
    confesses itself more a prisoner to our past mistakes than a proponent of them.
    Confined by earlier flawed precedents, Reiner, like McCarley and O’Neal before it, thus
    merely continued our earlier error in Ruelas, citing that decision for the proposition that federal
    courts need not engage in an AEDPA analysis of a state court’s harmless error conclusion
    because the Brecht test does the job by itself. That collective conclusion, it bears repeating,
    overlooks the fact that Fry—from where much of this misunderstanding emanates—was not
    reviewing a state court’s harmless error determination, to which AEDPA would plainly apply.
    See Fry, 
    551 U.S. at 114
     (determining the applicable standard of review when the state court
    “did not review [a trial error] for harmlessness” under Chapman). It also overlooked many
    passages from Ayala that undermine Ruelas. In fact, it ignored all of those in favor of one other:
    That “a prisoner who seeks federal habeas corpus relief must satisfy Brecht, and if the state court
    adjudicated his claim on the merits, the Brecht test subsumes the limitations imposed by
    AEDPA.” Ayala, 
    135 S. Ct. at 2199
    . But that passage is not license to cast aside AEDPA. In
    fact, much the opposite. The Supreme Court there was reminding us that while the Brecht test
    always applies on collateral review, AEDPA also applies where, unlike in Fry, the state court
    No. 17-2267                         Davenport v. MacLaren                              Page 36
    reaches the question of harmless error. And where an underlying state court decision concludes
    that any error in the petitioner’s state court proceeding was harmless, the Brecht test, having
    subsumed AEDPA, takes on the additional requirements and demands in § 2254(d). Id. While
    Brecht in this sense may “subsume” the AEDPA analysis, nowhere has the Supreme Court
    declared that Brecht consumes AEDPA, rendering it null and void in the harmless error setting.
    Nor, I might add, could the Supreme Court so easily have dispensed with AEDPA’s
    requirements even had it desired to do so. AEDPA is a valid act of Congress. It has not been
    declared unconstitutional or otherwise unenforceable. It would be quite something, then, for the
    Supreme Court to nonetheless make that law disappear by “subsuming” it in the Brecht standard.
    Johnson v. Lamas, 
    850 F.3d 119
    , 133 (3d Cir. 2017) (“[T]he Fry Court did not hold—and would
    have had no possible basis for holding—that Brecht somehow abrogate[d] the limitation on
    federal habeas relief that § 2254(d) plainly sets out.”) (quoting Ayala, 135, S. Ct. at 2198); see
    also Sifuentes v. Brazelton, 
    825 F.3d 506
    , 534 (9th Cir. 2016) (same). That is especially true
    when one considers that AEDPA followed Brecht, not the other way around. Functionally,
    Brecht could not have subsumed (or consumed) the AEDPA statute when the decision was
    announced. At that point, after all, the statute was still in the mind’s eye.
    That AEDPA amplifies the Brecht standard is all the more apparent when one considers
    that Congress legislates against the backdrop of Supreme Court decisions. See Forest Grove
    Sch. Dist. v. T.A., 
    557 U.S. 230
    , 239–40 (2009) (internal quotations omitted). Noting the
    limitations on federal habeas review already in place following Brecht, Congress, through
    AEDPA, imposed “new requirements” on when habeas relief could be granted by a federal court.
    Felker v. Turpin, 
    518 U.S. 651
    , 662 (1996) (“Title I of the Act has changed the standards
    governing our consideration of habeas petitions by imposing new requirements for the granting
    of relief to state prisoners.”). Those additional requirements “limited rather than expanded the
    availability of habeas relief.” Fry, 
    551 U.S. at 119
    . Yet today, they are absent.
    Shifting blame to the state for this dubious omission, and perhaps hedging its bets, the
    majority opinion contends (in a sentence) that the state conceded away AEDPA review of the
    Michigan courts’ harmless error determination. But the state conceded no such thing. Both
    Michigan’s brief and its statements at oral argument reflect the state’s repeated contention that
    No. 17-2267                        Davenport v. MacLaren                                Page 37
    AEDPA is not merely Brecht’s afterthought. Appellee Br. at 15, 28; Oral Argument at 14:10–
    17:34 (Michigan’s counsel citing Ayala and arguing at length that the state court’s harmless error
    analysis was not counter to Supreme Court precedent); see also Reiner, 955 F.3d at 556–57
    (Michigan advocating that Ayala requires AEDPA review of a harmless error determination).
    Nor could the state concede the point away, in the context of AEDPA review.                Section
    2254(d)(1) requires that we give deference to a state court’s decision on the merits, including any
    harmless error determination. And we owe that statutorily mandated deference regardless of
    how the state or Davenport might characterize the decision. Langley v. Prince, 
    926 F.3d 145
    ,
    162–63 & n.8 (5th Cir. 2019) (en banc) (citing Brown v. Smith, 
    551 F.3d 424
    , 428 n.2 (6th Cir.
    2008), abrogated on other grounds, Cullen v. Pinholster, 
    563 U.S. 170
     (2011)).
    4. In view of this legislative and precedential backdrop, it is perhaps no surprise that
    every other federal appellate court to take up the issue post-Ayala has agreed that the standards
    articulated in both Brecht and AEDPA apply to a habeas court’s review of a state court’s
    harmless error analysis. By my count, at least seven other circuits have granted AEDPA
    deference to a state court’s determination that a constitutional error was harmless:
    •   Second Circuit. Orlando v. Nassau County District Attorney’s Office, 
    915 F.3d 113
    , 127 (2d Cir. 2019) (“When a state court makes a harmless error
    determination on direct appeal, we owe the harmlessness determination itself
    deference under [AEDPA].”) (internal quotations and citations omitted);
    •   Third Circuit. Johnson v. Lamas, 
    850 F.3d 119
    , 136 (3d Cir. 2017) (“[W]e cannot
    say     that    the  Superior    Court’s    determination    that    Slaughter’s
    statement was harmless was so lacking in justification that we should refuse to
    give it AEDPA deference.”) (internal quotations and citations omitted); Johnson
    v. Superintendent Fayette SCI, 
    949 F.3d 791
    , 804 (3d Cir. 2020) (reviewing an
    “undecided issue of harmless error de novo” under Brecht and observing that a
    “different standard of review” augmented by AEDPA applies
    under Lamas and Ayala when the state court reached the question of harmlessness
    on the merits);
    •   Seventh Circuit. Long v. Pfister, 
    874 F.3d 544
    , 547–48 (7th Cir. 2017) (en banc)
    (“The Appellate Court of Illinois ruled that any error was harmless in light of the
    other evidence inculpating Long. Davis v. Ayala . . . holds that a harmless-error
    decision is one on the merits as § 2254(d) uses that phrase.”) (internal quotations
    omitted); see also id. at 556 (Hamilton, Rovner, & Williams, JJ., dissenting)
    (finding that habeas relief was warranted where each harmless error inquiry was
    No. 17-2267                          Davenport v. MacLaren                                   Page 38
    satisfied, including whether the error was harmless under directly on-point
    Supreme Court precedent, consistent with AEDPA review);
    •     Eighth Circuit. Davis v. Grandienard, 
    828 F.3d 658
    , 666 (8th Cir. 2016) (“We
    fail to find any unreasonable application of clearly-established federal law in the
    Minnesota Supreme Court’s decision . . . explaining that any error committed by
    the state trial court when it admitted the inadmissible portion of Davis’s statement
    was harmless . . . .”);
    •     Ninth Circuit. Rademaker v. Paramo, 
    835 F.3d 1018
    , 1024 (9th Cir. 2016) (“[I]t
    was not objectively unreasonable for the state appellate court to conclude that the
    evidence supported the jury’s finding . . . thus rendering the charging error
    harmless beyond a reasonable doubt.”); see also Sifuentes, 825 F.3d at 534;
    •     Tenth Circuit. Malone v. Carpenter, 
    911 F.3d 1022
    , 1030 (10th Cir. 2018) (“So
    which standard prevails—Brecht or § 2254(d)(1)? The Supreme Court has
    answered the question by saying that both apply.”);
    •     Eleventh Circuit. Al-Amin v. Warden, 
    932 F.3d 1291
    , 1299 (11th Cir. 2019)
    (“Ultimately, for a federal court to grant habeas relief, it must be true both that the
    state court’s application of the Chapman harmless beyond a reasonable doubt
    standard was objectively unreasonable and that the error had a substantial and
    injurious effect or influence on the verdict.”) (emphasis in original) (internal
    citations and quotation marks omitted).
    Of the many decisions in this uniform line of cases, Malone is particularly instructive.
    Assessing on collateral review a state court’s conclusion that a constitutional error was harmless,
    the Tenth Circuit observed that “Brecht [] predated” AEDPA, and that AEDPA “limited rather
    than expanded the availability of habeas relief.” 911 F.3d at 1029–30 (quoting Fry, 
    551 U.S. at 119
    ). Notwithstanding the Supreme Court’s observation in Fry that the AEDPA/Chapman
    standard may be easier to satisfy than the Brecht standard in some respects, the Supreme Court
    “d[id] not exclude the application of AEDPA in the harmless-error context.” 
    Id.
     at 1030 (citing
    Ayala, 
    135 S. Ct. at 2198
    ). Accordingly, the Tenth Circuit applied AEDPA deference to the
    Oklahoma Court of Criminal Appeal’s conclusion that any error in the petitioner’s criminal
    proceeding was harmless. And because the petitioner failed to satisfy AEDPA, the Tenth Circuit
    denied relief.    Id. at 1033 (“The OCCA’s determination that the error in the voluntary-
    intoxication instruction was harmless was an eminently reasonable application of Chapman.”).
    Against the backdrop of this unbroken line of decisions, the majority opinion cites four
    cases for the proposition that other circuits in fact share our “Brecht-only” approach. Two of
    No. 17-2267                        Davenport v. MacLaren                                 Page 39
    them are unpublished, which says little when stacked next to the just-cited published decisions of
    those same courts.    Wharton v. Vaughn, 722 F. App’x 268 (3d Cir. 2018); Hammonds v.
    Comm’r, Ala. Dep’t of Corr., 712 F. App’x 841 (11th Cir. 2017). And even on their own non-
    binding terms, those cases do not stand for the extraordinary proposition that a federal habeas
    court may “go straight to Brecht with full confidence that the AEDPA’s stringent standards will
    also be satisfied.” Ruelas, 
    580 F.3d at 413
    . Both Wharton and Hammonds extensively discuss
    the familiar AEDPA standard; they do not sweep it aside.
    The same is true of the Seventh Circuit’s en banc decision in Long. After confirming that
    the state court’s harmless error decision was entitled to AEDPA deference, the court went on to
    deny habeas relief on the basis that there was no threshold constitutional error in the state court
    proceeding, meaning the court need not reach the question of harmlessness. 874 F.3d at 547–48.
    The en banc Seventh Circuit majority thus said nothing about Brecht consuming AEPDA. Nor,
    in fact, did the dissenters. The dissenting opinion, which believed a constitutional error had
    occurred, acknowledged three inquiries to assess whether that error was harmless—including
    measuring the error against the clearly established constitutional law at the time, the traditional
    AEDPA standard. See id. at 556.
    That leaves the Ninth Circuit’s decision in Hall v. Haws, 
    861 F.3d 977
     (9th Cir. 2017),
    the lone published out-of-circuit decision granting habeas relief that the majority opinion cites to
    support its AEDPA-free framework. Yet even there, Judge Pregerson, writing for a fractured
    court, begrudgingly performed an AEDPA analysis, finding that the “California Court of
    Appeal’s harmless error determination was objectively unreasonable.” 
    Id. at 992
    . Nor, it bears
    adding, does Hall suggest that the Ninth Circuit’s earlier, controlling decisions in Rademaker
    and Sifuentes were wrongly decided or otherwise distinguishable.
    Other than a misguided line of cases in this Court, then, the courts of appeals have
    universally accepted the notion that, before granting habeas relief, a federal court reviewing a
    state court’s harmlessness determination must reach two separate conclusions: one, that the
    constitutional error had a substantial and injurious effect on the verdict; and two, that the state
    court’s harmlessness analysis constituted an unreasonable application of clearly established
    federal law. The majority opinion does only the first.
    No. 17-2267                        Davenport v. MacLaren                                Page 40
    *        *         *       *       *
    By my tally, today’s opinion flouts a long-standing federal statute, misapprehends
    Supreme Court precedent, and pays no respect to the independent judgment of our sister state
    courts—all in reversing a decision in which the magistrate judge and district court properly
    applied the correct statutory and precedential requirements. Davenport v. MacLaren, No. 1:14-
    cv-1012, 
    2016 WL 11262506
    , at *6 (W.D. Mich. Nov. 7, 2016). In so doing, the majority
    opinion divides our Circuit on the resolution of harmless error issues in the habeas context.
    Stewart, 867 F.3d at 636–37. And it sets us apart from every other circuit to have addressed the
    issue. This point bears repeating: No circuit, save for this one, has granted habeas relief without
    first finding that the underlying state court decision ran afoul of both Brecht and AEDPA. For
    these reasons, I cannot join the majority opinion.        And given the recurring nature of this
    important question, it is my hope that some court, either our en banc court or beyond, will clarify
    the standard we apply in this frequent setting.
    II.      Application Of The AEDPA And Brecht Standards Forecloses Relief To
    Davenport.
    Were we to apply AEDPA deference, as Ayala requires, we could not conclude that the
    Michigan courts’ determination regarding harmless error constituted an objectively unreasonable
    application of clearly established federal law. Likewise, even accepting the majority opinion’s
    conclusion that we may ignore AEDPA, the majority opinion fundamentally misapplies the
    Brecht standard in granting habeas relief to Davenport.
    A. In Faulting The Michigan Supreme Court On Collateral Review, The Majority
    Opinion Impermissibly Extends Both Deck and Holbrook.
    Because there is no Supreme Court precedent that shows “beyond doubt” that an error
    occurred in Davenport’s trial or that any such error was not harmless, Yarborough, 
    541 U.S. at 666
    , Davenport’s claim fails to clear AEDPA’s high bar. Stewart, 867 F.3d at 639.
    To assess whether the Michigan courts unreasonably applied clearly established federal
    law in affirming Davenport’s conviction, the most analogous benchmarks are Deck and
    Holbrook. In Deck, the Supreme Court held that the shackling of a criminal defendant at trial, in
    No. 17-2267                         Davenport v. MacLaren                                Page 41
    certain circumstances, is an error of constitutional magnitude subject to Chapman review.
    
    544 U.S. at 635
    . But Deck is unlike today’s case in ways that make the Michigan courts’
    decision to deny Davenport relief entirely reasonable. Consider the extreme measures employed
    against Deck during his state trial. His hands and feet were shackled throughout trial. The
    shackles were visible to the entire jury.        And Deck remained shackled even during the
    punishment phase—where imposition of the death penalty was quite likely. The state trial court,
    moreover, did not hold any kind of evidentiary hearing to probe what effect (if any) the shackles
    might have had on the verdict.
    Compare that dramatic circumstance to the facts of Davenport’s trial. A significant
    concern driving the result in Deck was Deck’s inability to communicate with his counsel. 
    Id. at 631
    . Not so for Davenport. His right hand remained unshackled throughout trial, meaning he
    could write notes to his counsel without impediment. Mendoza, 
    544 F.3d at
    654–55 (citing
    Deck, 
    544 U.S. at 630
    ). Another concern was Deck’s inability to participate in his defense by
    testifying on account of the shackles. 
    Id.
     Davenport testified in his own defense completely
    unshackled. Also unlike in Deck, there was a privacy curtain around counsel table to make
    Davenport’s shackles less apparent, undercutting Deck’s core requirement for granting relief,
    namely, that the shackles be visible to the jury. 
    Id.
     These same factual distinctions led us to
    deny the petitioner’s shackling claim in Mendoza because “it [was] not obvious that Deck should
    be extended to the particular facts present.” 
    Id. at 655
    . So too here.
    And in another respect, Davenport’s claim is an even weaker candidate for habeas relief
    than was Mendoza’s unsuccessful claim. Back to Deck. There, the state trial court failed to
    probe any possible influence Deck’s shackling had on the jury.           Not so here. Following the
    Michigan Supreme Court’s acknowledgement that Davenport’s partial shackling may have been
    an error under Deck, the trial court held an evidentiary hearing to elicit testimony from the jurors
    to assess whether shackling had any impact on the trial’s outcome. The court inquired whether
    the jurors had seen the shackles and, if so, what effect, if any, the shackles had on each juror’s
    deliberation. Less than half the jurors saw the shackles. And each juror affirmatively testified
    that the partial shackling had no effect on her verdict.
    No. 17-2267                        Davenport v. MacLaren                                Page 42
    So the majority opinion buries the lede. In summarizing the juror interviews that took
    place during the evidentiary hearing, the majority opinion starts with the least-revealing aspects.
    For instance, the majority notes that one juror asked another whether she was nervous sitting
    next to Davenport while he testified, an exchange that apparently reflected juror bias. But any
    apprehension over proximity to Davenport on the juror’s part could simply reflect the fact that
    Davenport, at the time, was a six-foot-five-inch, 300-pound man accused of violently strangling
    a young woman. Only after highlighting this and other largely inconsequential items does the
    majority opinion finally mention the conclusive bottom line: No juror’s verdict was influenced
    by the partial shackling.
    But that bottom-line account cannot be trusted, we are told, in view of the Supreme
    Court’s decision in Holbrook. Holbrook, however, is a poor vehicle for undermining this juror
    testimony. After all, Holbrook did not deal with shackling. Nor did it address juror reflections
    following trial. Rather, it addressed statements made by prospective jurors during voir dire. It
    may be the case that a potential juror’s pre-trial statement that she will not be prejudiced by
    visible security measures while later sitting as a juror during trial proceedings is inherently
    speculative.   
    475 U.S. at 570
     (“[W]hen jurors are questioned at the very beginning of
    proceedings[,] at that point, they can only speculate on how they will feel after being exposed to
    a practice daily over the course of a long trial.”). But that hardly describes today’s case.
    Davenport’s jurors were questioned only after sitting through a lengthy trial where they heard
    graphic evidence of a brutal killing. When the jurors were later asked about the effect of the
    partial shackling on their deliberations, they were relaying their actual experiences, not
    speculating about future events, a distinction aptly recognized by the Michigan Court of Appeals.
    Davenport, 
    2012 WL 6217134
    , at *2 (“[I]t was proper for the jurors to testify regarding how
    viewing the shackles affected their deliberations.”).
    Lacking a factual or legal basis to establish an unreasonable application of federal law by
    the Michigan courts, the majority opinion turns to legal commentators and public researchers.
    These sources, says the majority opinion, also undermine Davenport’s conviction.            Why?
    Because “social-science research has demonstrated the near-universal existence of implicit and
    unconscious bias.” That may be (although the debate can be left for another day). Suffice it to
    No. 17-2267                        Davenport v. MacLaren                                 Page 43
    say that the majority opinion, if it proves anything, proves too much. After all, if every person to
    sit on a jury implicitly is impermissibly biased, seemingly no verdict could ever stand, given the
    risk that some purported bias may have tainted the outcome, even in the face of overwhelming
    evidence. For today’s purposes, what guides our review is “clearly established Federal law, as
    determined by the Supreme Court of the United States,” not abstract sociology. 
    28 U.S.C. § 2254
    (d)(1).
    Plainly then, Deck and Holbrook’s application to this case was far from “beyond doubt.”
    Yarborough, 
    541 U.S. at 666
    . To be sure, on direct review, a court might reasonably extend
    Deck and Holbrook to cover the facts of this case. But we cannot fault the Michigan courts
    today, on collateral review, for failing to anticipate the majority opinion’s extension of Deck and
    Holbrook, when the Supreme Court has not done the same.
    B. Even If Deck And Holbrook Govern Davenport’s Claim, Davenport Was Not
    Actually Prejudiced By His Partial Shackling At Trial.
    1. Advancing “straight to Brecht” as the majority opinion does, it short-changes even
    that test. The majority opinion misconstrues the Supreme Court’s holding in Deck and our own
    precedents to invoke what it calls a “presumption of prejudice.”            It then employs that
    “presumption” to shift the usual burden on collateral review, requiring a showing of “sufficiently
    strong evidence of guilt” to sustain Davenport’s first-degree murder conviction. That unusual
    standard surely cannot be gleaned from Deck. Deck, keep in mind, came to the Supreme Court
    on direct (not collateral) review, meaning that any language there suggesting that prejudice is
    presumed in a shackling case can be chalked up to Deck’s unique procedural posture. Because
    of that unique posture, Deck’s invocation of the Chapman standard to measure prejudice is the
    inverse of what we apply here.
    Nor does a “presumption of prejudice” standard fairly find its footing in our decision in
    Ruimveld v. Birkett, 
    404 F.3d 1006
    , 1017–18 (6th Cir. 2005). In Ruimveld, a pre-Deck case
    involving shackling, we determined that shackling claims generally are subject to harmless error
    review (as Deck would later require). 
    Id. at 1013
    . And we acknowledged the common-sense
    notion that strong evidence of guilt may readily show that the defendant’s shackling was
    harmless. 
    Id. at 1016
    . But we did not hold that strong evidence was necessary to do so. In any
    No. 17-2267                       Davenport v. MacLaren                                Page 44
    event, to the extent Ruimveld is inconsistent with Deck’s subsequent holding that shackling
    errors are subject to ordinary Chapman analysis (and therefore Brecht analysis on collateral
    review), Deck carries the day.
    Not only is the majority opinion’s standard at odds with precedent, it is similarly at odds
    with the traditional understanding of habeas review. By employing a “presumption of prejudice”
    standard, the majority opinion engages in what essentially amounts to direct review of the
    Michigan Supreme Court’s decision—and a more exacting form of direct review at that—
    paradoxically making habeas relief easier to obtain than relief on direct review. Among other
    peculiarities, the majority’s approach sets shackling apart from other errors subject to Chapman
    analysis. But the Supreme Court has not afforded shackling violations unique treatment. As
    with other constitutional errors, the state, on direct review, must prove “beyond a reasonable
    doubt that the [shackling] error complained of did not contribute to the verdict obtained.” Deck,
    
    544 U.S. at 635
     (alterations and citations omitted). And when a state court concludes that the
    government has done so, we apply the ordinary Brecht analysis on collateral review. See 
    507 U.S. at 623
    .
    Following Brecht, to hold on collateral review that a criminal defendant was prejudiced
    by a constitutional error, we must have “grave doubt about whether a trial error of federal law
    had ‘substantial and injurious effect or influence in determining the jury’s verdict.’” Ayala,
    
    135 S. Ct. at 2198
     (quoting O’Neal, 
    513 U.S. at 436
    ). This standard, the majority opinion notes,
    was satisfied in Ruimveld, “a close case based on purely circumstantial evidence.” 
    404 F.3d at
    1017–18. Davenport’s case was not close. Ample evidence supported his conviction for first-
    degree murder. Davenport undisputedly strangled White until she died. To do so, a medical
    expert explained to the jury, Davenport had to continue to apply pressure to White’s airway for
    more than four minutes after she lost consciousness.
    In Michigan, Davenport’s conduct constitutes premeditated murder.          See People v.
    Johnson, 
    597 N.W.2d 73
    , 78–80 (Mich. 1999).            That is the inescapable conclusion from
    Johnson.   There, the Michigan Supreme Court affirmed a first-degree murder conviction
    involving strangulation.   In addition to the strangulation evidence implicating Johnson, the
    record also revealed that Johnson knew the victim, had defensive wounds, and moved the
    No. 17-2267                          Davenport v. MacLaren                              Page 45
    victim’s body after the killing. Collectively, this evidence satisfied the elements of first-degree
    murder. 
    Id.
     at 79–80.
    Compared to the record in Johnson, the government had evidence to spare in Davenport’s
    case. Davenport, the record reveals, had an unfortunate fondness for strangulation. In addition
    to strangling White, Davenport had also strangled another woman less than a week earlier,
    conduct consistent with what he had told others: that he would choke people if things ever got
    out of hand. With respect to whether Davenport had time to take a “second look” before
    murdering White, Davenport (like Johnson) had defensive wounds. If all of that was not enough
    to prove premeditation, the government also established why Davenport strangled White, and
    why he had little remorse for doing so. The two had a pre-existing relationship, Davenport hid
    White’s body after killing her, and he stole property from her apartment in the days after the
    murder.    And what did Davenport have to say when confronted with this evidence while
    testifying at trial? “[I]t’s not gonna help me any to tell the truth.”
    Measured against Johnson, the “substantial evidence of [Davenport’s] guilt” left the
    Michigan courts with the firm conclusion that the government established that any error in
    Davenport’s proceeding was harmless. Davenport, 832 N.W.2d at 390 (“Given the substantial
    evidence of guilt presented at trial, we cannot conclude that there was an unacceptable risk of
    impermissible factors coming into play.”); Davenport, 
    2012 WL 6217134
    , at *2 (“All of the
    evidence indicated that the shackling did not affect the verdict in any way.”). That should be all
    the more true in today’s habeas setting. After all, Ayala requires Davenport to satisfy an even
    higher standard on collateral review: that we have “grave doubt” that an error substantially and
    injuriously influenced the verdict. 
    135 S. Ct. at
    2197–98.
    The majority opinion hardly mentions Johnson.               Instead, it emphasizes general
    propositions of Michigan law, oddly elevating those generalities over the specific holding in
    Johnson. True, as a general proposition in Michigan, evidence that a murder was committed by
    manual strangulation, standing alone, is not enough to show premeditation. People v. Furman,
    
    404 N.W.2d 246
    , 249–50 (Mich. App. 1987) (internal citation omitted).               But death by
    strangulation helps support a prima facie case of premeditated murder. 
    Id.
     And that prima facie
    case, in turn, can be supplemented by other evidence to show premeditation beyond a reasonable
    No. 17-2267                         Davenport v. MacLaren                                    Page 46
    doubt, as was the case in Johnson. 597 N.W.2d at 78–80. Here, the supplemental evidence
    showing premeditation was plentiful. Indeed, Johnson was the harder case.
    2. Which brings me to the evidence regarding Davenport’s partial shackling, the lone
    potential ground for distinguishing Johnson.        The majority opinion posits that the partial
    shackling “branded Davenport as having a violent nature,” so much so that his “presumption of
    innocence [was] replaced by a presumption of dangerousness.” All of that, however, belies the
    Supreme Court’s clear command that shackling errors are subject to ordinary Chapman (and
    therefore Brecht) analysis. Deck, 
    544 U.S. at 635
    .
    In effect, the majority opinion assumes the jurors in Davenport’s case were influenced in
    their verdict by the partial shackling. But why make any assumptions about what was going
    through each juror’s mind during deliberations? We have their testimony. And it is conclusive.
    Every single juror testified that the shackling had no effect on the verdict.
    To be sure, some jurors made statements to the effect that Davenport was dangerous. Of
    course, those statements were made after the jurors had heard graphic evidence regarding
    Davenport’s strangulation of White—with his bare hands—along with other incriminating
    evidence. And, of course, after the jurors had deemed Davenport guilty of first-degree murder.
    That leaves little reason to believe the jurors’ judgment of Davenport was due to his partial
    shackling (visible to less than half the jurors) rather than his gruesome killing of White.
    The record is conclusive. The jury was presented with extensive evidence that Davenport
    strangled a woman to death, and that the crime was premeditated. Each juror testified that the
    partial shackling had no effect on her verdict. That evidence was enough to satisfy a unanimous
    Michigan Court of Appeals, and enough for a unanimous Michigan Supreme Court. It was
    enough for the district court and the magistrate judge. And it is enough for me. I respectfully
    dissent.