Concurring and Dissenting.—Four years ago, in People v. Gamache (2010) 48 Cal.4th 347 [106 Cal.Rptr.3d 771, 227 P.3d 342], this court upheld a death verdict for Richard Gamache despite the fact that the jury, during its penalty phase deliberations, had inadvertently been allowed to view a videotape never admitted into evidence. (Id. at pp. 395-403.) The videotape, which the jury watched two times before sentencing Gamache to death, showed a detailed confession he had given to the police on the day of the crime. (Id. at pp. 395, 400-402.) We held that “[introduction of the . . . videotape into the jury room was indisputably error . . . .” (Id. at p. 396.) But we went on to conclude that the error was harmless beyond a reasonable doubt under Chapman v. California (1967) 386 U.S. 18, 24 [17 L.Ed.2d 705, 87 S.Ct. 824] (Chapman). (People v. Gamache, at p. 399, fn. 22; id. at p. 403 [“[T]here is *775no reasonable possibility Gamache would have received a more favorable outcome had the . . . videotape not been erroneously placed in the jury room.”].)
The United States Supreme Court denied review. (See Gamache v. California (2010) 562 U.S. 1083 [178 L.Ed.2d 514, 131 S.Ct. 591] (statement of Sotomayor, J.) (Gamache).) However, Justice Sotomayor, joined by Justice Ginsburg, Justice Breyer, and Justice Kagan, filed a statement respecting the denial of the petition for writ of certiorari. In that statement, Justice Sotomayor agreed that the error was harmless beyond a reasonable doubt, but she observed that this court, in reaching that conclusion, had said, “ ‘[I]n the absence of misconduct, the burden remains with the defendant to demonstrate prejudice under the usual standard for ordinary trial error.’ 48 Cal. 4th, at 397, 227 P.3d, at 387 (emphasis added).” (Id. at p. 1084 [131 S.Ct. at p. 592] (statement of Sotomayor, J.).) Justice Sotomayor said: “It is not clear what the court intended in allocating the burden to the defendant to demonstrate prejudice, but if it meant to convey that the defendant bore the burden of persuasion, that would contravene Chapman. See 386 U. S., at 24 (noting that the ‘original common-law harmless-error rule put the burden on the beneficiary of the error either to prove that there was no injury or to suffer a reversal of his erroneously obtained judgment’); cf. O’Neal v. McAninch, 513 U. S. 432, 438-139 [130 L.Ed.2d 947, 115 S.Ct. 992] (1995) (describing Chapman as ‘placing the risk of doubt’ about harmlessness on the State).” (Ibid, (statement of Sotomayor, J.).) The four-justice statement concluded with a warning: “With all that is at stake in capital cases, [citation], in future cases the California courts should take care to ensure that their burden allocation conforms to the commands of Chapman.” (Id. at p. 1085 [131 S.Ct. at p. 593] (statement of Sotomayor, J.).)
Today’s decision neglects to heed this warning.
Defendant Jonathan Keith Jackson was convicted of murder and sentenced to death while being forced, without lawful justification, to wear a REACT stun belt throughout his trial. REACT stands for “remote electronically activated control technology.” Neither this colorless moniker nor anything in today’s opinion provides an inkling of the power of this electric shock device. Here is how Chief Justice George, writing for a six-justice majority, described it in People v. Mar (2002) 28 Cal.4th 1201 [124 Cal.Rptr.2d 161, 52 P.3d 95] (Mar): “ ‘The type of stun belt which is used while a prisoner is in the courtroom consists of a four-inch-wide elastic band, which is worn underneath the prisoner’s clothing. This band wraps around the prisoner’s waist and is secured by a Velcro fastener. The belt is powered by two 9-volt batteries connected to prongs which are attached to the wearer over the left kidney region. . . . [f] The stun belt will deliver an eight-second, 50,000-volt *776electric shock if activated by a remote transmitter which is controlled by an attending officer. The shock contains enough amperage to immobilize a person temporarily and cause muscular weakness for approximately 30 to 45 minutes. The wearer is generally knocked to the ground by the shock and shakes uncontrollably. Activation may also cause Immediate and uncontrolled defecation and urination, and the belt’s metal prongs may leave welts on the wearer’s skin requiring as long as six months to heal. An electrical jolt of this magnitude causes temporary debilitating pain and may cause some wearers to suffer heartbeat irregularities or seizures. [Citations.]’ ” (Id. at pp. 1214-1215.)
An article we quoted in Mar described a brochure of the belt’s manufacturer, Stun Tech, as follows: “One of the great advantages, the company says, is its capacity to humiliate the wearer. ‘After all, if you were wearing a contraption around your waist that by the mere push of a button in someone else’s hand could make you defecate or urinate yourself,’ the brochure asks, ‘what would that do to you from the psychological standpoint?’ And if the shock ever has to be administered? ‘One word—,’ brags the brochure, ‘DEVASTATION! ’ ” (Schulz, Cruel & Unusual Punishment (Apr. 24, 1997) 44 N.Y. Review of Books 51; see Mar, supra, 28 Cal.4th at p. 1227, fn. 8.) The Indiana Supreme Court has banned the use of stun belts in the courtrooms of that state altogether, concluding that other forms of restraint “can do the job without inflicting the mental anguish that results from simply wearing the stun belt and the physical pain that results if the belt is activated.” (Wrinkles v. State (Ind. 2001) 749 N.E.2d 1179, 1195.)
While mindful that trial judges must have broad discretion to deal with “disruptive, contumacious, stubbornly defiant defendants” (Illinois v. Allen (1970) 397 U.S. 337, 343 [25 L.Ed.2d 353 90 S.Ct. 1057]), we observed in Mar that the use of a stun belt to maintain control over the accused during trial “has not been without problems or controversy” (Mar, supra, 28 Cal.4th at p. 1205). The belt has been abused, as when one California trial judge ordered a defendant to be shocked for verbally interrupting her. (Id. at p. 1223, fn. 6, citing Hawkins v. Comparet-Cassani (9th Cir. 2001) 251 F.3d 1230.) There have been “a disturbing number of accidental activations,” including at least one instance in a California courtroom “resulting in the defendant’s hospitalization.” (Mar, at pp. 1228, 1229, fn. 9.) Most importantly, the “ ‘ “total psychological supremacy” ’ ” that the stun belt is designed to achieve “may impair the defendant’s ability to think clearly, concentrate on the testimony, communicate with counsel at trial, and maintain a positive demeanor before the jury.” (Id. at p. 1226.)
Against this backdrop, the court today concludes that any error in requiring defendant to wear a stun belt throughout this capital trial was harmless *777beyond a reasonable doubt. I respectfully disagree with this conclusion and, in particular, with two aspects of the court’s reasoning.
First, even though it is undisputed by the parties that the trial court erred in forcing defendant to wear the stun belt, today’s opinion resolves defendant’s claim on the basis of harmless error without actually finding that the trial court erred. The court instead offers the vague locution that “[i]n light of the People’s concession at oral argument that the trial court erred under Mar, supra, 28 Cal.4th 1201, we proceed directly to the issue of prejudice.” (Maj. opn., ante, at p. 744.) Does this mean the court accepts the Attorney General’s concession on the merits? Or does it mean the court is merely assuming error, as we often do, in order to resolve the issue on harmless error grounds? The court does not say. By tiptoeing around the issue, today’s opinion fails to confront the seriousness of the error and may be read to suggest that there is some question as to whether error occurred at all—even though, as both parties recognize and as I explain herein, the error in this case is at least as apparent as similar errors in other cases finding inadequate justification for physical restraints during trial. The court’s refusal to find error on this record sows unnecessary doubt about the stringent standards that govern the use of restraints in the courtroom.
Second, the court concludes that “the People have satisfied their burden under Chapman, supra, 386 U.S. 18, to show that any federal errors are harmless beyond a reasonable doubt.” (Maj. opn., ante, at p. 748.) I agree the error was harmless at the guilt phase, but I cannot agree it was harmless at the penalty retrial. Wearing a stun belt carries a substantial risk of altering a defendant’s demeanor, and a defendant’s demeanor is often one of the most important considerations for the jury in deciding whether a capital defendant deserves to live or die. (See Riggins v. Nevada (1992) 504 U.S. 127, 143-144 [118 L.Ed.2d 479, 112 S.Ct. 1810] (conc. opn. of Kennedy, J.) (Riggins).) As discussed below, the record here provides reason to worry about this risk. Moreover, the penalty determination in this case was a close issue. The first jury, which saw and heard the entirety of the guilt phase evidence, hung at the penalty phase, with nine jurors willing to go either way and one or two jurors opposed to death. The second jury was impaneled only for the penalty retrial. We cannot say with any confidence what verdict it would have reached absent the stun belt error.
Today’s ruling simply cannot be squared with Chapman’s principle that the burden lies with “the beneficiary of a constitutional error to prove beyond a reasonable doubt that the error complained of did not contribute to the verdict obtained.” (Chapman, supra, 386 U.S. at p. 24.) The court holds that “reversal of a judgment is unwarranted when the record on appeal is devoid of evidence that the unjustified use of ... a stun belt had any adverse effect.” *778(Maj. opn., ante, at p. 740.) But under Chapman, reversal is unwarranted not when the record is devoid of evidence that the error had an adverse effect, but only when the state has shown beyond a reasonable doubt that the error did not have an adverse effect. Today’s opinion effectively puts the burden on defendant to demonstrate that the erroneous use of the stun belt was prejudicial. The court even goes so far as to supply, on the state’s behalf, an argument for harmless error that the state itself never made. (See post, at pp. 802-803.) In essence, the court inverts Chapman’s burden allocation in the very manner that four justices of the high court warned against in Gamache. (See Gamache, supra, 562 U.S. at p. 1085 [131 S.Ct. at p. 593] (statement of Sotomayor, J.).) At most, the court’s analysis of the record in this case establishes that it is uncertain whether defendant was prejudiced at the penalty retrial, not that the state has shown beyond a reasonable doubt that defendant was not prejudiced. To conclude otherwise, as the court does, confirms that “the allocation of the burden of proving harmlessness can be outcome determinative in some cases.” (Ibid. (statement of Sotomayor, J.).)
To observers of this court, it is no mystery that harmless error doctrine has played a major role in the outcomes of California capital appeals. Unfortunately, today’s decision is not unique in our jurisprudence; it is only the most recent instance in which this court has misapplied Chapman. (See post, at pp. 805-807.) Here, a man was unlawfully forced to have a 50,000-volt electric shock device strapped around his waist while a jury sat in judgment as to whether he should live or die for his crimes. It is reasonably possible that the stun belt adversely affected his demeanor, and we cannot know what decision the jury would have made absent this error. Because the court’s refusal to find prejudicial error in this case erodes the applicable standards for both error and prejudice, I respectfully dissent.
I.
When used for courtroom security, a stun belt “ ‘is worn underneath the prisoner’s clothing’ ” (Mar, supra, 28 Cal.4th at p. 1214), so it may be presumed to be out of the jury’s sight. Nevertheless, the legal standards governing the use of stun belts have their origins in concerns applicable to the use of visible restraints such as shackles. I begin by reviewing the development of those standards in state and federal law, and then apply them to the facts of this case.
A.
“A trial court has broad power to maintain courtroom security and orderly proceedings.” (People v. Hayes (1999) 21 Cal.4th 1211, 1269 [91 Cal.Rptr.2d 211, 989 P.2d 645]; see Illinois v. Allen, supra, 397 U.S. at p. 343.) But this *779power is not unlimited. The constitutional prohibition on forcing a criminal defendant to wear visible physical restraints during trial without special justification “has deep roots in the common law.” (Deck v. Missouri (2005) 544 U.S. 622, 626 [161 L.Ed.2d 953, 125 S.Ct. 2007] (Deck) [citing Blackstone and other early authorities]; see People v. Duran (1976) 16 Cal.3d 282, 288 [127 Cal.Rptr. 618, 545 P.2d 1322] (Duran) [same].) Although originally motivated by concern for the physical suffering caused by chains or other restraints, the constitutional rule today reflects “three fundamental legal principles.” (Deck, at p. 630.)
“First, the criminal process presumes that the defendant is innocent until proved guilty. [Citation.] Visible shackling undermines the presumption of innocence and the related fairness of the factfinding process.” (Deck, supra, 544 U.S. at p. 630.) “Second, the Constitution, in order to help the accused secure a meaningful defense, provides him with a right to counsel. [Citations.] The use of physical restraints diminishes that right Shackles can interfere with the accused’s ‘ability to communicate’ with his lawyer. [Citation.] Indeed, they can interfere with a defendant’s ability to participate in his own defense . . . .” (Id. at p. 631.) “Third, judges must seek to maintain a judicial process that is a dignified process. The courtroom’s formal dignity, which includes the respectful treatment of defendants, reflects the importance of the matter at issue, guilt or innocence, and the gravity with which Americans consider any deprivation of an individual’s liberty through criminal punishment. And it reflects a seriousness of purpose that helps to explain the judicial system’s power to inspire the confidence and to affect the behavior of a general public whose demands for justice our courts seek to serve. The routine use of shackles in the presence of juries would undermine these symbolic yet concrete objectives.” (Ibid.)
The first concern, the presumption of innocence, is not a factor during the penalty phase of a capital case. But in light of the second and third concerns, along with “the ‘ “severity” ’ and ‘ “finality” ’ of the sanction” and “the ‘acute need’ for reliable decisionmaking when the death penalty is at issue,” courts may not place defendants in visible restraints during either the penalty phase or the guilt phase of a capital trial without a “case[-] specific” determination that “reflects] particular concerns, say, special security needs or escape risks, related to the defendant on trial.” (Deck, supra, 544 U.S. at pp. 632-633.)
1.
Consistent with Deck, California case law has long recognized limitations on the use of physical restraints on criminal defendants during trial. (See Mar, supra, 28 Cal.4th at pp. 1216-1218.) In People v. Harrington (1871) 42 Cal. 165 (Harrington), this court said that “any order or action of the Court which, *780without evident necessity, imposes physical burdens, pains and restraints upon a prisoner during the progress of his trial, inevitably tends to confuse and embarrass his mental faculties, and thereby materially to abridge and prejudicially affect his constitutional rights of defense . . . .” (Id. at p. 168.) There, finding “no pretense of necessity for the manacles and chains upon the defendants during their trial,” we reversed the convictions. (Id. at p. 169.)
In Duran, we declared “our continued adherence to the Harrington rule” in light of the “possible prejudice in the minds of the jurors, the affront to human dignity, the disrespect for the entire judicial system which is incident to unjustifiable use of physical restraints, as well as the effect such restraints have upon a defendant’s decision to take the stand.” (Duran, supra, 16 Cal.3d at p. 290.) Duran “reaffirm[ed] the rule that a defendant cannot be subjected to physical restraints of any kind in the courtroom while in the jury’s presence, unless there is a showing of a manifest need for such restraints.” (Id. at pp. 290-291, fn. omitted.) We further concluded that “in any case where physical restraints are used those restraints should be as unobtrusive as possible, although as effective as necessary under the circumstances.” (Id. at p. 291.)
Duran went on to specify that “[t]he showing of nonconforming behavior in support of the court’s determination to impose physical restraints must appear as a matter of record and, except where the defendant engages in threatening or violent conduct in the presence of the jurors, must otherwise be made out of the jury’s presence. The imposition of physical restraints in the absence of a record showing of violence or a threat of violence or other nonconforming conduct will be deemed to constitute an abuse of discretion.” (Duran, supra, 16 Cal.3d at p. 291.) Applying these principles in Duran, we found an abuse of discretion: “No reasons for shackling the defendant appear on the record. There is no showing that defendant threatened to escape or behaved violently before coming to court or while in court. The fact that defendant was a state prison inmate who had been convicted of robbery and was charged with a violent crime did not, without more, justify the use of physical restraints.” (Id. at p. 293.)
In subsequent cases, we have emphasized that in deciding whether to impose restraints, “the court is obligated to base its determination on facts, not rumor and innuendo . . . .” (People v. Cox (1991) 53 Cal.3d 618, 652 [280 Cal.Rptr. 692, 809 P.2d 351] (Cox); see ibid, [finding shackling unwarranted where the record “does not contain a single substantiation of violence or the threat of violence on the part of the accused”].) We have also stressed that “it is the trial court, not law enforcement personnel, that must make the decision an accused be physically restrained in the courtroom. A trial court abuses its discretion if it abdicates this decisionmaking authority to security personnel *781or law enforcement. [Citations.]” (People v. Hill (1998) 17 Cal.4th 800, 841 [72 Cal.Rptr.2d 656, 952 P.2d 673], fn. omitted (Hilly see id. at p. 842 [finding shackling unjustified where the trial court “deferred to the sheriff s department’s decision that shackles were necessary” and “fail[ed] to determine independently whether, in its view, there existed a manifest need to place defendant in restraints”].) Likewise, “ ‘[t]he imposition of restraints in a proper case is normally a judicial function in which the prosecutor plays no necessary part. . . . [I]t is the function of the court, not the prosecutor, to initiate whatever procedures the court deems sufficient in order that it might make a due process determination of record that restraints are necessary.’ (Duran, supra, 16 Cal.3d at p. 293, fn. 12.)” (Id. at pp. 841-842.)
In Mar, we had our “first occasion ... to address the use of a stun belt in courtrooms in California.” (Mar, supra, 28 Cal.4th at p. 1205.) Because “the stun belt posed an imminent risk of both serious injury and humiliation . . . ,” we held that the principles set forth in Duran governing the use of visible restraints apply equally to the use of a stun belt worn underneath a defendant’s clothing. (Mar, at p. 1219.) We explained: “In light of the nature of the device and its effect upon the wearer when activated, requiring an unwilling defendant to wear a stun belt during trial may have significant psychological consequences that may impair a defendant’s capacity to concentrate on the events of the trial, interfere with the defendant’s ability to assist his or her counsel, and adversely affect his or her demeanor in the presence of the jury. In addition, past cases both in California and in other jurisdictions disclose that in a troubling number of instances the stun belt has activated accidentally, inflicting a potentially injurious high-voltage electric shock on a defendant without any justification. The potential for accidental activation provides a strong reason to proceed with great caution in approving the use of this device.” (Id. at p. 1205.) Accordingly, we concluded that a defendant may not be forced to wear a stun belt unless there is an on-the-record showing of manifest need based on the trial court’s own determination of the facts and independent judgment under the circumstances. (Id. at pp. 1219-1223.) Because “the trial court [in Mar] never made, nor purported to make, a finding or determination that there was a ‘manifest need’ to impose the stun belt upon defendant because he posed a serious security threat in the courtroom,” we found an abuse of discretion. (Id. at p. 1222; see ibid, [use of the stun belt resulted from an “apparently unilateral decision” by the bailiff or jail officials].)
We recently summarized the governing principles for use of a stun belt as follows: “(1) there must be a showing of manifest need for the stun belt; (2) the defendant’s threatening or violent conduct must be established as a matter of record; and (3) it is the function of the court to initiate whatever procedures it deems necessary to make a determination on the record that the stun belt is necessary. The court must make an independent determination *782based on facts, not rumor or innuendo, and must not merely rely on the judgment of jail or court security personnel. [Citations.]” (People v. Howard (2010) 51 Cal.4th 15, 28 [118 Cal.Rptr.3d 678, 243 P.3d 972] (Howard).) Although the trial in the case now before us predates Mar, we said in Mar and in subsequent cases that the manifest need standard for use of a stun belt applies to trials conducted before Mar. (Mar, supra, 28 Cal.4th at pp. 1222-1223; see, e.g., People v. Virgil (2011) 51 Cal.4th 1210, 1270-1271 [126 Cal.Rptr.3d 465, 253 P.3d 553]; People v. Lomax (2010) 49 Cal.4th 530, 561 [112 Cal.Rptr.3d 96, 234 P.3d 377]; People v. Gamache, supra, 48 Cal.4th at pp. 366-367.) For trials conducted after Mar, the requirements are even more stringent, as trial courts must take into account several additional considerations specific to stun belts before forcing a defendant to wear one. (See Mar, at pp. 1226-1230.)
2.
In light of Mar, the standards governing the use of a stun belt during trial are settled as a matter of California law. The issue is not entirely settled, however, as a matter of federal constitutional law.
Many courts have adopted federal constitutional standards similar to those we set forth in Mar. In U.S. v. Durham (11th Cir. 2002) 287 F.3d 1297 (Durham), the court observed that “stun belts plainly pose many of the same constitutional concerns as do other physical restraints, though in somewhat different ways.” (Id. at p. 1306.) First, a stun belt “seemingly poses a far more substantial risk of interfering with a defendant’s Sixth Amendment right to confer with counsel than do leg shackles. The fear of receiving a painful and humiliating shock for any gesture that could be perceived as threatening likely chills a defendant’s inclination to make any movements during trial— including those movements necessary for effective communication with counsel.” (Id. at p. 1305.) Second, a stun belt could have an “adverse impact” on a “defendant’s Sixth Amendment and due process rights to be present at trial and to participate in his defense. Wearing a stun belt is a considerable impediment to a defendant’s ability to follow the proceedings and take an active interest in the presentation of his case. It is reasonable to assume that much of a defendant’s focus and attention when wearing one of these devices is occupied by anxiety over the possible triggering of the belt. A defendant is likely to concentrate on doing everything he can to prevent the belt from being activated, and is thus less likely to participate fully in his defense at trial.” (Id. at pp. 1305-1306, fn. omitted.) Finally, “stun belts have the potential to be highly detrimental to the dignified administration of criminal justice. . . . Shackles are a minor threat to the dignity of the courtroom when compared with the discharge of a stun belt, which could cause the defendant to lose control of his limbs, collapse to the floor, and defecate on himself.” (Id. *783at p. 1306.) The Eleventh Circuit adopted substantive and procedural standards almost identical to those we established in Mar and held that the district court in Durham “abused its discretion” in ordering a stun belt because it “did not, on the record, consider any less restrictive alternatives” and did not articulate “on the record” a “rationale for imposing this highly intrusive method of restraint.” (Durham, at p. 1308; see id. at p. 1312 (conc, opn. of Tjoflat, J.).)
A number of other jurisdictions, applying federal constitutional law, have reached similar results. (See U.S. v. Wardell (10th Cir. 2009) 591 F.3d 1279, 1293-1294 [holding that “ ‘[t]he use of stun belts, depending somewhat on their method of deployment, raises all of the traditional concerns about the imposition of physical restraints’ ” and that a district court must make “a defendant-specific determination of necessity resulting from security concerns”]; Wrinkles v. Buss (7th Cir. 2008) 537 F.3d 804, 813-815 [finding defense counsel’s failure to object to stun belt where there was no “individualized justification” for imposing it to be deficient performance under first prong of Strickland v. Washington (1984) 466 U.S. 668 [80 L.Ed.2d 674, 104 S.Ct. 2052]]; U.S. v. Miller (6th Cir. 2008) 531 F.3d 340, 344-346 (Miller) [holding that district court erred “[b]y deferring to the Marshals’ judgment” instead of making “individualized determinations or specific findings” to justify use of stun belt]; Gonzalez v. Pliler (9th Cir. 2003) 341 F.3d 897, 901-902 (Gonzalez) [finding stun belt error where the “decision was apparently made by the bailiff, not the trial judge,” the record was devoid of evidence that the defendant posed a security problem, and the trial court failed to consider less restrictive alternatives]; Mobley v. State (Tenn. 2013) 397 S.W.3d 70, 101 [requiring “particularized findings” that include “whether there is a less onerous but adequate means of providing security”]; Illinois v. Allen (2006) 222 Ill.2d 340 [305 Ill.Dec. 544, 856 N.E.2d 349, 353-354] (Allen) [holding that “the use of electronic stun belts in the courts of this state is warranted only where there has been a showing of manifest need for the restraint” and finding that the trial court erred because “it simply deferred to the judgment of the sheriff’]; Hymon v. Nevada (2005) 121 Nev. 200 [111 P.3d 1092, 1098-1099] [adopting the standards of Durham, Gonzalez, and Mar as a matter of federal constitutional law]; Weaver v. State (Fla. 2004) 894 So.2d 178, 195 [stun belt must be “reasonably necessary to ensure order and safety in the courtroom,” and trial court must consider whether “alternative forms of restraint... are less prejudicial or viable”].)
Notwithstanding this substantial body of authority, the United States Supreme Court has yet to provide definitive guidance on the use of stun belts, and as a consequence, a number of courts have upheld their use under less stringent standards on habeas corpus or plain error review (see, e.g., Earhart v. Konteh (6th Cir. 2009) 589 F.3d 337, 349; U.S. v. Fields (5th Cir. 2007) 483 F.3d 313, 356-357; Mungo v. U.S. (D.C. 2010) 987 A.2d 1145, 1149-1150; *784State v. Bowen (2006) 340 Ore. 487 [135 P.3d 272, 279]; Reynolds v. State (Ala.Crim.App. 2010) 114 So.3d 61, 82) or as a matter of state law (see State v. Benson (2013) 232 Ariz. 452 [307 P.3d 19, 28]; Young v. State (1998) 269 Ga. 478 [499 S.E.2d 60, 61]). The high court last considered the issue of physical restraints during trial almost a decade ago in Deck, supra, 544 U.S. 622, which set forth rigorous standards for the use of visible restraints. Since then, courts have divided on the applicability of Deck to a stun belt worn underneath a defendant’s clothing and shielded from the jury’s view. (Compare Miller, supra, 531 F.3d at p. 345 [concluding on plain error review that a concealed stun belt raises the “ ‘same fundamental issues’ ” as those considered in Deck] and Allen, supra, 856 N.E.2d at p. 352 [“the Deck Court’s stated reasons which prompt due process scrutiny in visible restraint cases . .. may be applied with like force to stun belts which are not necessarily visible to the jury”] with Mungo, supra, 987 A.2d at pp. 1149-1150 [concluding on plain error review that Deck does not entail that “a stun belt qualifies as a type of physical restraint whose use is subject to [Deck’s] strictures . . .” or that “on-the-record findings are required before stun belts may be used”] and Benson, supra, 307 P.3d at pp. 28-29 [distinguishing Deck where “the record reflects that the stun belt . . . [was] not visible to the jury”].) For the reasons stated in Durham and Mar, the better view is that a requirement of individualized on-the-record justification applies not only to visible restraints but also to stun belts as a matter of federal constitutional law.
B.
In this case, two trial judges entered two separate orders requiring defendant to wear a stun belt: the first for the duration of the guilt phase and the first penalty phase, and the second for the penalty phase retrial. On appeal, defendant contends that each order violated his rights under the California Constitution and the United States Constitution. The record clearly indicates that neither order was supported by a proper finding of manifest need.
1.
Before the guilt phase, defendant submitted a motion objecting to physical shackling. In response, the prosecutor requested a stun belt as an alternative to shackling, arguing that a stun belt was not a physical restraint within the meaning of Duran. (See People v. Garcia (1997) 56 Cal.App.4th 1349, 1356 [66 Cal.Rptr.2d 350] (Garcia), disapproved in Mar, supra, 28 Cal.4th at p. 1219.) At a hearing on March 30, 1999, the prosecutor said, “I feel very strongly about this defendant having a react belt during this trial.” The trial court inquired whether defendant’s objection would apply to a stun belt. Defense counsel replied, “My position would be that Mr. Jackson has demonstrated his ability to be responsible in the courtroom and that we don’t need anything.”
*785The court then ruled as follows: “I tell you what we will do, if we have two deputies [,] and they try to give two deputies on murder cases, we will not have the react belt. ... I have evidence in the Statement of Aggravation that [defendant] did make a threat, he has engaged in violence in custody, and there is some—I don’t have a clear record of it but there is certainly hints from the probation report that he had when he was in [the California Youth Authority] and there is also information apparently from the CDC 115s that he has posed some threat to other people, [f] And it seems to me in a courtroom context if things don’t go the right way there might be a problem with him. . . . [f] I think you ought to tell the sergeant it would be my strong preference to have a second deputy sitting back there so we can avoid this problem. Tell them if it’s not, then I will have a react belt, and it’s my understanding you have to have a second deputy then anyway.” The courtroom deputy confirmed that if the court ordered defendant to wear the stun belt, a second deputy would be provided to monitor the belt. The court then told the deputy, “Tell the sergeant if you can, please, that if he forces me to order the react belt in order to get a second deputy here I will do that.”
The prosecutor objected to the court’s ruling on the ground that “there is really no showing required for a react belt, because it’s not a restraint. It’s been held they are not even a restraint on the defendant.” The court nevertheless confirmed its decision, commenting: “If there is [sic] problems going on with intimidation of witnesses or something else is occurring we will deal with it at that point. But at this point, given Deputy Young’s confidence that he can handle the situation with a second deputy, I think it would be inappropriate to have the react belt.”
The trial court continued to explain its reasoning, stating: “This is not an easy case ... to make a decision on because I have allegations—of course, I am at somewhat of a loss because these things haven’t been proved, but I think I have to assume from the starting point that there may be some basis for these things.” The court continued: “Sure he isn’t causing a problem now because he is always handcuffed and dressed in orange . . . .” “[H]e is in a situation where he really can’t cause much problems. [|] On the other hand, ... I haven’t had a problem with him in court. He has been cooperative. He hasn’t mouthed off in any way. And the cases are clear that if I could avoid shackling him or handcuffing him or anything like that I should do that. So I don’t intend to do that.”
Finally, the court restated its decision: “[W]hat I’m going to do is call Captain Moreland, who is the person over in the jail, and tell him that I need to have a second deputy, if you can’t provide it I will need to order the react belt, which then will require a second deputy because there is always a deputy that monitors the react belt.”
*786A week later, on April 6, 1999, the court announced that defendant would be required to wear a stun belt for the duration of trial. The court explained, “I could not be assured of a second officer without having the REACT Belt.”
On this record, it is clear that the trial court’s order was an abuse of discretion. The trial court made no determination that the stun belt was necessary for courtroom security. While noting that defendant “has engaged in violence in custody,” the court acknowledged that there was no “clear record” defendant posed a security threat, that defendant “has been cooperative” and “hasn’t mouthed off in any way,” that “. . . I haven’t had a problem with him in court,” and that the contrary indications consisted of “allegations” that “haven’t been proved.” Moreover, the court said “it would be inappropriate to have the react belt” in light of its judgment that any threat posed by defendant could be handled by the presence of a second deputy in the courtroom. The trial court ultimately decided to order the stun belt not because it had made a finding that the belt was necessary to restrain defendant, but because it “could not be assured of a second officer without having the REACT Belt.” As a result, defendant was guarded by two deputies and forced to wear a stun belt during trial.
When ordering restraints, “[a] trial court abuses its discretion if it abdicates [its] decisionmaking responsibility to security personnel or law enforcement.” (Hill, supra, 17 Cal.4th at p. 841; see Miller, supra, 531 F.3d at pp. 344-346; Gonzalez, supra, 341 F.3d at pp. 901-902; Allen, supra, 856 N.E.2d at pp. 353-354.) From this principle, it follows that a trial court abuses its discretion when it permits such personnel to make their presence contingent on the court’s decision to order the use of restraints. It is well established that “a court should impose the least restrictive measure that will satisfy the court’s legitimate security concerns . . . .” (Mar, supra, 28 Cal.4th at p. 1206; see Durham, supra, 287 F.3d at p. 1308.) Here, the trial court determined that having a second officer in the courtroom would have been adequate. Because the court’s decision to order the stun belt was driven by the sheriff’s department’s staffing policy rather than by any finding of manifest need, it was an abuse of discretion.
2.
Before the penalty retrial, defendant was again ordered to wear a stun belt, this time by a second judge who had just begun to preside over the case. At a hearing on October 21, 1999, the prosecutor initiated discussion of the issue and asked the court to order “[t]he zapping belt, the react belt.” In response, the court observed, “I’ve been informed that there was a need for such by the sheriff’s department and my deputy.” At that point, defense counsel said: “Well, I would object to the belt, not only as I believe it’s unnecessary, but *787it’s extremely uncomfortable for the defendant. I would ask that the Court use the leg brace on him. It fits underneath the clothing. I think that’s ample restriction on his movement and more than an adequate means of securing him in this courtroom. He hasn’t done anything in the past to exhibit any kind of unruly behavior or any disrespect to the Court or any staff. I just think it’s unnecessary.”
The court responded; “It’s my understanding that although there has not been any issue in the courtroom, there has been sufficient issues out of the courtroom that give rise to considerable concern for the safety of the public and the safety of deputies who would be called upon to attend to the security of the defendant and the people in attendance. . . . Unfortunately, my bailiff who brought the subject to me originally is not with us this morning.” The court continued: “He knew something of the background of the defendant, and I didn’t ask him to go into any detail when he broached the subject the day before yesterday.”
The prosecutor proceeded to argue, as she had at the guilt phase, that a stun belt was unlike other physical restraints and that the court was not required to make any special finding before ordering defendant to wear it. When the trial court said it was not aware that the belt had been uncomfortable to wear, defense counsel responded: “The trial days are long, and what my concern is is that the belt is uncomfortable. It’s a large object that’s placed on the kidney area, the liver area on his back. It makes it difficult to sit back. And I believe that because of the fact this is [sz'c] capital case, every caution and concern should be made to protect the rights of the defendant. I’m worried about any unconscious grimacing or exhibitions of discomfort that Mr. Jackson might make that might be misconstrued by the jury as a reaction to witness testimony or anything like that. But we did in the last trial try to alleviate that discomfort by giving him a pillow to kind of even out the back so when he sat back, he wouldn’t be putting all the pressure onto that react unit that’s strapped to him. That caused a little bit of relief, but I don’t think significant.”
The court then said: “[F]rom the little I’ve seen, it appears that the defendant is sufficiently violent that it’s not a matter of we’re concerned about so much his escape, which would be the subject of the leg restraint. . . .” The court continued: “[H]ere we’re concerned about the violent nature of his responses to people in authority. And that just is not sufficiently addressed by a leg brace. ... It will not keep him from attacking members of the public or the attachés and his own counsel.... [f] ... I’m not about to have this [be] a matter of something that we wish we would have done in retrospect which would have avoided some serious injury.”
*788Finally, after observing that there was no reason to shackle defendant if the stun belt was an available alternative, the court ruled as follows: “[S]ince the Court is opposed to shackling and since the Court feels that the leg brace is, in the opinion of the Court, insufficient to secure the safety of the public, the attachés and so on, counsel, from violent attacks by the defendant, the use of the react belt is approved . . . .”
This order was also an abuse of discretion. The information on which the second trial judge relied was precisely the sort that case law has found insufficient. In raising a safety concern, the trial court made reference to “sufficient issues out of the courtroom” and “the violent nature of [defendant’s] responses to people in authority” without ever identifying what the specific issues or incidents were. Further, rather than “mak[ing] its own determination of the ‘manifest need’ for the use of such restraint,” the trial court “reified] solely on the judgment of jail or court security personnel in sanctioning the use of such restraints.” (Mar, supra, 28 Cal.4th at p. 1218.) The court acknowledged that “there has not been any issue in the courtroom” and did not point to any on-the-record evidence that defendant posed a security threat during trial. Instead, the trial court relied on unidentified information it had been told in off-the-record conversations with the sheriff’s department and the bailiff. By the court’s own account, those conversations were not in-depth; the bailiff “knew something of the background of the defendant,” but the court “didn’t ask him to go into any detail.” (See Cox, supra, 53 Cal.3d at p. 652 [“the court is obligated to base its determination on facts, not rumor and innuendo . . .”].) In sum, the record does not “demonstrate that the trial court independently determined on the basis of an on-the-record showing of defendant’s nonconforming conduct that ‘there existed a manifest need to place defendant in restraints.’ ” (Mar, supra, 28 Cal.4th at p. 1218; see Durham, supra, 287 F.3d at p. 1308.) Thus, I would conclude that the trial court erred under state and federal law in forcing defendant to wear a stun belt during the penalty retrial.
3.
Today’s opinion neither agrees nor disagrees with the analysis above. The court simply says: “In light of the People’s concession at oral argument that the trial court erred under Mar, supra, 28 Cal.4th 1201, we proceed directly to the issue of prejudice.” (Maj. opn., ante, at p. 744.) I am not sure what this sentence is supposed to mean, but it certainly leaves the reader with the impression that the court, by refusing to say it accepts the Attorney General’s concession, is not fully convinced that use of the stun belt was error.
Although it is not rare for our court to decline to resolve an issue on the merits when it can be resolved on the basis of harmless error, this practice is *789typically reserved for issues that present a close question, where the risk of an incorrect ruling or unintended consequences may counsel restraint. In this case, the trial court’s errors in ordering the stun belt are no less obvious than the similar errors found in Mar, Durham, Miller, Gonzalez, and Allen. (See ante, at pp. 738-740.) Indeed, after reciting defendant’s arguments on this issue (maj. opn., ante, at p. 744), today’s opinion does not even bother to state any counterarguments, for there are none that could withstand scrutiny under well-established law. Any hesitation the court may have about finding error is simply unexplained. The court’s refusal to find error under these circumstances sows unnecessary doubt about the rigorous standards that govern the use of restraints in the courtroom. Even if this were unintended, the upshot remains that a violation of the right to be free of unjustified restraints during trial has gone unrecognized. We should not allow due process of law to be silently eroded in this way.
n.
I agree with today’s decision that the stun belt was not prejudicial at the guilt phase because the belt’s potential effect on defendant’s demeanor was not relevant to the jury’s determination of the objective facts comprising his crimes and because defendant does not contend he would have testified or otherwise aided his defense at the guilt phase had the court not ordered the belt. However, I cannot agree that the stun belt error was harmless beyond a reasonable doubt at the penalty retrial, where it is reasonably possible that the stun belt adversely affected defendant’s demeanor in a way that contributed to the jury’s verdict. In reaching a contrary result, the court effectively assigns to defendant the burden of proving that the error was prejudicial, thereby inverting Chapman’s requirement that the state demonstrate lack of prejudice beyond a reasonable doubt. (Chapman, supra, 386 U.S. at p. 24; see Gamache, supra, 562 U.S. at p. 1085 [131 S.Ct. at p. 593] (statement of Sotomayor, J.).) This case illustrates that the conceptual and institutional difficulties inherent to harmless error analysis call for strict adherence to established legal standards in assessing prejudice, especially when it comes to errors in the penalty phase of a capital trial.
A.
To illuminate what is problematic about today’s decision, I begin by bringing into focus the nature of harmless error analysis and the importance of the legal standards that guide it.
At bottom, harmless error inquiry mediates a basic tension in the law. On one hand, few if any trials are entirely free from error, and an appellate court would impair the basic functioning of the criminal justice system if it were to *790reverse a conviction whenever some slight misstep occurred. This is a lesson learned from experience. Until the beginning of the last century, a rule of near-automatic reversal prevailed in criminal matters such that, “ ‘as one trial judge put it . . . [,] courts of review ‘tower[ed] above the trials of criminal cases as impregnable citadels of technicality.’ ” (Kotteakos v. United States (1946) 328 U.S. 750, 759 [90 L.Ed. 1557, 66 S.Ct. 1239].) As a result, “criminal trial[s] became a game for sowing reversible error in the record, only to have repeated the same matching of wits when a new trial had been thus obtained.” (Ibid.) It was in response to such problems that many jurisdictions, including California, adopted measures to limit reversals of judgments only to cases where substantial errors had occurred. (Id. at pp. 759-760; see Cal. Const., art. VI, § 16; People v. O’Bryan (1913) 165 Cal. 55, 63-66 [130 P. 1042].)
On the other hand, an appellate court runs two related risks when it evaluates the effect of an error to determine whether it warrants reversal. First, to say that a conviction may stand in spite of underlying error is at odds with the norm of legality that justifies the state’s imposition of criminal punishment in the first place. A declaration that an error is harmless is, in essence, a conclusion that even though a legal right has been violated, there will be no remedy for that violation. As former District of Columbia Circuit Chief Judge Harry Edwards has put it, “each time we employ the imaginary tonic of harmless error, we erode an important legal principle. When we hold errors harmless, the rights of individuals, both constitutional and otherwise, go unenforced.” (Edwards, To Err Is Human, But Not Always Harmless: When Should Legal Error Be Tolerated? (1995) 70 N.Y.U. L.Rev. 1167, 1170 (Edwards).) Divorcing legal rights from remedies in this fashion may encourage, or at least fail to discourage, future violations by undermining the “deterrent force of a reversal.” (Ibid.)
Second, when an appellate court engages in harmless error inquiry, it risks invading the province of the jury. A court trying to determine what would have happened in a counterfactual proceeding in which the error at issue did not occur may end up, consciously or not, conducting an inquiry into a defendant’s guilt or innocence, a question that our system of justice reserves for the jury. (See Apprendi v. New Jersey (2000) 530 U.S. 466, 476-484 [147 L.Ed.2d 435, 120 S.Ct. 2348]; Sullivan v. Louisiana (1993) 508 U.S. 275, 280 [124 L.Ed.2d 182, 113 S.Ct. 2078].) The risk of an appellate court usurping the jury’s role becomes especially great when harmless error analysis focuses not on whether error might have affected the jury’s decisionmaking, but on whether there was overwhelming evidence to support the result. As Chief Justice Traynor explained, “It is one thing for an appellate court to determine that a verdict was or was not affected by error. It is quite another for an appellate court to become in effect a second jury to determine whether the defendant is guilty.” (Traynor, The Riddle of Harmless Error (1970) p. 21 *791(Traynor); see Neder v. United States (1999) 527 U.S. 1, 31-32 [144 L.Ed.2d 35, 119 S.Ct. 1827] (conc. & dis. opn. of Scalia, J.); Sullivan, at p. 280; Edwards, supra, 70 N.Y.U. L.Rev. at pp. 1185-1199; Field, Assessing the Harmlessness of Federal Constitutional Error—A Process in Need of a Rationale (1976) 125 U.Pa. L.Rev. 15.)
These two risks arise in unique and potentially troubling ways when a reviewing court assesses the prejudicial impact of errors in the penalty phase of a capital trial. To state the obvious, it is no small thing for a court of law to affirm a sentence of death while holding that the defendant’s legal rights were violated in the sentencing proceeding. This court has recognized as much in holding that an exacting “reasonable possibility” standard of harmless error review—which is “ ‘the same, in substance and effect,’ as the harmless-beyond-a-reasonable-doubt standard of Chapman”—specially applies when evaluating any error that occurs during the penalty phase of a capital trial. (People v. Cowan (2010) 50 Cal.4th 401, 491 [113 Cal.Rptr.3d 850, 236 P.3d 1074]; see People v. Gonzalez (2006) 38 Cal.4th 932, 961 [44 Cal.Rptr.3d 237, 135 P.3d 649]; People v. Ashmus (1991) 54 Cal.3d 932, 990 [2 Cal.Rptr.2d 112, 820 P.2d 214]; People v. Brown (1988) 46 Cal.3d 432, 448 [250 Cal.Rptr. 604, 758 P.2d 1135] (Brown).) Our less stringent “reasonable probability” standard applicable to ordinary state law errors (People v. Watson (1956) 46 Cal.2d 818, 837 [299 P.2d 243]) is “simply insufficient to ensure ‘reliability in the determination that death is the appropriate punishment in a specific case.’ ” (Brown, at p. 448; see Deck, supra, 544 U.S. at p. 632 [“The Court has stressed the ‘acute need’ for reliable decisionmaking when the death penalty is at issue.”].)
Moreover, we have long “recognized a fundamental difference between review of a jury’s objective guilt phase verdict, and its normative, discretionary penalty phase determination” in a capital trial. (Brown, supra, 46 Cal.3d at p. 447.) The nature of the task assigned to the jury at the penalty phase of a capital trial makes it particularly difficult to analyze identified error for harmlessness. “A capital penalty jury ... is charged with a responsibility different in kind from . . . guilt phase decisions: its role is not merely to find facts, but also—and most important—to render an individualized, normative determination about the penalty appropriate for the particular defendant. . . .” (Id. at p. 448; see Woodson v. North Carolina (1976) 428 U.S. 280 [49 L.Ed.2d 944, 96 S.Ct. 2978].) In light of the jury’s “vast discretion” in making this decision and the endless variety of factors that may affect its judgment {Brown, at p. 447), a counterfactual assessment of whether an error did not influence the jury can be “extremely speculative or impossible” (Clemons v. Mississippi (1990) 494 U.S. 738, 754 [108 L.Ed.2d 725, 110 S.Ct. 1441]; cf. Traynor, supra, at p. 73 [“[A]n appellate court cannot possibly determine what factors influenced a jury to impose the death penalty. Any error, unless it related only to the proof of some fact otherwise indisputably established, *792might have tipped the scales against the defendant.”]; Carter, Harmless Error in the Penalty Phase of a Capital Case: A Doctrine Misunderstood and Misapplied (1995) 28 Ga. L.Rev. 125, 126, 149-150). It bears emphasis that “this court is [not] entitled legally to make the determination of whether appellant is to live or be put to death. The determination of what penalty shall be imposed rests exclusively with the jury.” (People v. Hamilton (1963) 60 Cal.2d 105, 138 [32 Cal.Rptr. 4, 383 P.2d 412]; cf. Ring v. Arizona (2002) 536 U.S. 584, 614 [153 L.Ed.2d 556, 122 S.Ct. 2428] (conc. opn. of Breyer, J.) [“the Eighth Amendment requires that a jury, not a judge, make the decision to sentence a defendant to death”].)
B.
There is no easy solution for navigating these tensions either in the mine run of harmless error cases or in cases involving error in the penalty phase of a capital trial. But appellate courts can mitigate the risks associated with harmless error review by complying strictly with the standards established by law to guide such review. Rigorous adherence to these standards serves to maintain the crucial role of appellate review in promoting adherence to the law, to restrain reviewing courts from invading the province of the jury, and to preserve jury verdicts that may be deemed untainted by error based on disciplined application of a legally specified confidence level rather than an appellate judge’s ad hoc intuitions of guilt, innocence, or just punishment.
As today’s opinion acknowledges, the Chapman harmless error standard applies to defendant’s federal constitutional claims. (Maj. opn., ante, at p. 748.) Chapman has two principal features. First, a finding of harmless error requires a showing “beyond a reasonable doubt” that the error did not affect the jury’s verdict. (Chapman, supra, 386 U.S. at p. 24.) The reasonable doubt standard has long been understood to indicate “the very high level of probability required by the Constitution” to deprive an individual of life or liberty. (Victor v. Nebraska (1994) 511 U.S. 1, 14 [127 L.Ed.2d 583, 114 S.Ct. 1239].) The requirement is not one of absolute certainty. (See id. at p. 17 [“A fanciful doubt is not a reasonable doubt.”]; Pen. Code, § 1096 [reasonable doubt “ ‘is not a mere possible doubt’ ”].) But the standard is, and is intended to be, very stringent; it is not satisfied so long as there is a doubt “based upon ‘reason.’ ” (Jackson v. Virginia (1979) 443 U.S. 307, 317 [61 L.Ed.2d 560, 99 S.Ct. 2781].) The stringency of the standard reflects not only its protective function but also its amenability to principled application. Under Chapman, a reviewing court need not calibrate its certitude to some vaguely specified probability; instead, the court must be convinced the error was harmless to the maximal level of certainty within the realm of reason, a level that admits no reasonable doubt.
*793Second, the burden of proving beyond a reasonable doubt that the error did not affect the jury’s verdict lies with “the beneficiary of the error,” namely, the state. (Chapman, supra, 386 U.S. at p. 24; see Deck, supra, 544 U.S. at p. 635; United States v. Dominguez Benitez (2004) 542 U.S. 74, 81, fn. 7 [159 L.Ed.2d 157, 124 S.Ct. 2333]; Arizona v. Fulminante (1991) 499 U.S. 279, 295-296 [113 L.Ed.2d 302, 111 S.Ct. 1246]; Satterwhite v. Texas (1988) 486 U.S. 249, 256 [100 L.Ed.2d 284, 108 S.Ct. 1792].) Under Chapman, it is not the defendant’s burden to show that the error did have adverse effects; it is the state’s burden to show that the error did not have adverse effects. Because it may be difficult to determine whether a particular error contributed to the jury’s verdict given the counterfactual nature of the inquiry, “the allocation of the burden of proving harmlessness can be outcome determinative in some cases.” (Gamache, 562 U.S. at p. 1085 [131 S.Ct. at p. 593] (statement of Sotomayor, J.).)
The significance of this burden allocation has been confirmed by high court precedent since Chapman. In Riggins, the high court reversed a capital conviction where the defendant was unconstitutionally forced to take antipsychotic drugs during the course of his trial. (Riggins, supra, 504 U.S. at p. 138.) In deciding that the error was prejudicial, the high court made no finding that the medication actually affected the defendant’s outward appearance, testimony, ability to follow the proceedings, or communication with counsel; the court simply said such effects were “clearly possible.” (Id. at p. 137.) While acknowledging that “the precise consequences of forcing antipsychotic medication upon Riggins cannot be shown from a trial transcript,” the high court held that Riggins need not “demonstrate how the trial would have proceeded differently” in order to prevail. (Ibid.) Riggins reversed the conviction upon observing that “[ejfforts to prove or disprove actual prejudice from the record before us would be futile, and guesses whether the outcome of the trial might have been different if [the defendant had not been unlawfully medicated] would be purely speculative.” (Ibid.)
The high court’s decision in O’Neal v. McAninch (1995) 513 U.S. 432 [130 L.Ed.2d 947, 115 S.Ct. 992] (O’Neal) likewise underscores that the state bears the burden of demonstrating the absence of prejudice and that this burden allocation can be dispositive. In O’Neal, the defendant had been convicted in state court of murder and other crimes, but the Sixth Circuit on habeas corpus review held that the trial court had given a misleading jury instruction in violation of the Constitution. The question before the high court was whether an error may be deemed harmless in “the special circumstance in which record review leaves the conscientious judge in grave doubt about the likely effect of an error on the jury’s verdict. (By ‘grave doubt,’ we mean that, in the judge’s mind, the matter is so evenly balanced that he feels himself in virtual equipoise as to the harmlessness of the error.)” (Id. at p. 435.) O’Neal held that in such cases “the petitioner must win” (id. at *794p. 437), relying significantly on Chapman’s rule that “ ‘constitutional error .. . casts on someone other than the person prejudiced by it a burden to show that it was harmless’ ” (id. at p. 438, quoting Chapman, supra, 386 U.S. at p. 24). O’Neal thus confirms that an error cannot be found harmless under Chapman even when a reviewing court is not convinced but is genuinely unsure that there is a reasonable possibility that the error affected the verdict. The burden is on the state, not the defendant, to dispel the uncertainty.
C.
In the present case, a straightforward application of Chapman leads to the conclusion that the death judgment must be reversed. Not only did defendant expressly complain about the stun belt; the record also gives rise to reasonable concern about the belt’s effect on defendant’s demeanor, an important consideration for the jury in a capital penalty proceeding. Further, the prosecution’s case for death at the penalty retrial was not overwhelming, and the first penalty phase, which set forth no less persuasive a case for death, resulted in a hung jury. These considerations are more than sufficient to prevent us from concluding beyond a reasonable doubt that the stun belt error did not affect the penalty verdict.
As context for examining the particulars of this case, it bears mention that less than five years ago, in People v. Stevens (2009) 47 Cal.4th 625 [101 Cal.Rptr.3d 14, 218 P.3d 272] (Stevens), we described the use of stun belts for purposes of courtroom security as an “exceptional practice[]” that is “inherently prejudicial.” (Id. at pp. 632, 644.) Stevens held that “the stationing of a courtroom deputy next to a testifying defendant is not an inherently prejudicial practice that must be justified by a showing of manifest need.” (Id. at p. 629.) In reaching this conclusion, we drew a contrast between the presence of security officers in the courtroom and “visible shackling, stun belts, or other affronts to human dignity” that are “[i]nherently prejudicial practices.” (Id. at p. 644.) In this context, the term “inherently prejudicial” does not mean that erroneous use of a physical restraint requires automatic reversal. But it does mean that the use of a stun belt carries a “substantial risk of prejudice” and “an inordinate risk of infringing upon a criminal defendant’s right to a fair trial.” (Id. at p. 632; accord, Durham, supra, 287 F.3d at p. 1305.) This substantial risk of prejudice underscores the heavy burden the state must meet to demonstrate that the stun belt error was harmless beyond a reasonable doubt.
As we said in Mar, “requiring an unwilling defendant to wear a stun belt during trial may have significant psychological consequences that may . . . adversely affect his or her demeanor in the presence of the jury.” (Mar, supra, 28 Cal.4th at p. 1205.) This concern is borne out in the record here. From the *795beginning, defendant repeatedly complained about the stun belt. On April 6, 1999, when defendant was first fitted with the device at the guilt phase, defense counsel observed that it prevented his client from leaning back and that “it’s going to be very uncomfortable.” Defendant said, “It’s punching up in my side.” Defendant was also worried about setting off the belt inadvertently. As the attending deputy stated in the course of trying to reposition the device, “I think he’s concerned it’s going to activate it, but just by leaning on it, it won’t activate it.” The deputy said that “[t]he belt is designed to be worn around the waist tightly” and that according to the belt’s instructions, “[i]t needs to be near the kidney.” Eventually, defendant was fitted with a pillow to cushion the belt, and he said, “That’s a little better.” Still, defendant continued to complain about the tightness of the belt, and the trial court observed, “It really is bulky. I can see how it would be uncomfortable.”
Defense counsel reiterated these concerns on October 21, 1999, before the penalty phase retrial, stating: “Well, I would object to the belt, not only as I believe it’s unnecessary, but it’s extremely uncomfortable for the defendant.” When the trial judge (who was new to the case) said, “I have not been aware . . . that there has been that much of an uncomfortable feeling for the defendant by reason of the react belt,” defense counsel pushed back and expressed concern about the stun belt’s potential effect on defendant’s demeanor: “The trial days are long, and what my concern is is that the belt is uncomfortable. It’s a large object that’s placed on the kidney area, the liver area on his back. It makes it difficult to sit back. . . .I’m worried about any unconscious grimacing or exhibitions of discomfort that Mr. Jackson might make that might be misconstrued by the jury as a reaction to witness testimony or anything like that. But we did in the last trial try to alleviate that discomfort by giving him a pillow to kind of even out the back so when he sat back, he wouldn’t be putting all the pressure onto that react unit that’s strapped to him. That caused a little bit of relief, but I don’t think significant.”
In addition to defense counsel’s statement, the record elsewhere reveals reasonable grounds for concern that the stun belt may have “impair[ed] the defendant’s ability to . . . maintain a positive demeanor before the jury.” (Mar, supra, 28 Cal.4th at p. 1226.) On May 20, 1999, during the first penalty phase, the defense moved to preclude the prosecution from commenting on defendant’s courtroom demeanor. In granting the motion, the trial court said: “The only real issue is lack of emotion at this point.” Today’s opinion says this observation “read in context” reflects no concern related to the belt. (Maj. opn., ante, at p. 746.) But when one reads the full transcript, one finds nothing to suggest that defendant’s lack of emotion was, as the court speculates, merely an impassive response to the testimony of the prosecution’s witnesses rather than a foreseeable effect of wearing the stun belt. It is hardly unreasonable to think that lack of emotion may be one effect of being *796forced to wear an electric shock device that is susceptible to accidental activation and designed to achieve “ ‘ “total psychological supremacy” ’ ” through the threat of “ ‘debilitating pain.’ ” (Mar, at pp. 1226, 1215.) I suspect most people, with such a device strapped around the waist, would sit as still and impassively as they possibly could in order to avoid activating it. (I would hesitate to even sneeze.)
The possibility of prejudice resulting from modification of a defendant’s demeanor during a capital sentencing proceeding is not insignificant. “As any trial attorney will attest, serious prejudice could result if [an extraneous influence] inhibits the defendant’s capacity to react and respond to the proceedings and to demonstrate remorse or compassion. The prejudice can be acute during the sentencing phase of the proceedings, when the sentencer must attempt to know the heart and mind of the offender and judge his character, Ms contrition or its absence, and Ms future dangerousness. In a capital sentencing proceeding, assessments of character and remorse may carry great weight and, perhaps, be determinative of whether the offender lives or dies.” (Riggins, supra, 504 U.S. at pp. 143-144 (conc. opn. of Kennedy, J.); see Sundby, The Capital Jury and Absolution: The Intersection of Trial Strategy, Remorse, and the Death Penalty (1998) 83 Cornell L.Rev. 1557.) Our own cases have repeatedly held that a defendant’s courtroom demeanor is relevant to the jury’s determination of the defendant’s character in the penalty phase of a capital trial. (See People v. Valencia (2008) 43 Cal.4th 268, 307-308 [74 Cal.Rptr.3d 605, 180 P.3d 351] [collecting cases].) And we have said it is not improper for a trial court to observe that “ ‘[t]he Defendant was emotionally calm during the entire trial’ ” in denying a motion to modify a death verdict because “[a] defendant’s demeanor may reflect remorse, or otherwise arouse sympathy in either jury or judge,” and is therefore “relevant” to the sentencing decision. (People v. Williams (1988) 44 Cal.3d 883, 970, fn. 50, 971-972 [245 Cal.Rptr. 336, 751 P.2d 395].)
The stun belt’s effect on defendant’s demeanor may have been especially significant at one crucial point during the penalty retrial. On November 17, 1999, the prosecution alerted the court that defendant had engaged in some “verbal and non-verbal communication” with members of his family. The court advised defendant that “the bailiff will take actions if you do it again.” Defendant could have reasonably interpreted tMs statement as a warning that the bailiff would activate the stun belt if defendant were to engage in any further verbal or nonverbal communication with Ms family. Immediately following tMs exchange, defendant’s mother testified. It is reasonable to believe that the jury would have found it particularly off-putting if defendant had remained expressionless while his mother recounted the extensive emotional, physical, and sexual abuse she had suffered in front of defendant, as well as the beatings that defendant had suffered in trying to protect her. As defense counsel observed in closing argument, the mother’s testimony caused *797defendant’s brother Antione, a Gulf War veteran, to “run from the courtroom crying.” If there was one moment in the penalty phase when the jury might have expected defendant to visibly demonstrate his capacity for empathy or compassion, it was during his mother’s testimony—right after the trial court had warned him that “the bailiff will take actions” if he engaged in any verbal or nonverbal communication with members of his family.
Finally, the weakness of the prosecution’s case for death makes it even more reasonable to believe that the stun belt error may have affected the jury’s verdict. The most significant and objective indicator of that weakness is the fact that the first penalty phase jury found the case to be very close and was ultimately unable to reach a verdict. A jury’s inability to reach a verdict is often indicative of weakness in the prosecution’s case and probative of an error’s prejudicial effect. (See, e.g., Kyles v. Whitley (1995) 514 U.S. 419, 454 [131 L.Ed. 2d 490, 115 S.Ct. 1555]; Krulewitch v. United States (1949) 336 U.S. 440, 445 [93 L.Ed. 790, 69 S.Ct. 716]; Kennedy v. Lockyer (9th Cir. 2004) 379 F.3d 1041, 1056, fn. 18; U.S. v. Beckman (8th Cir. 2000) 222 F.3d 512, 525; U.S. v. Tubol (2d Cir. 1999) 191 F.3d 88, 97; People v. Figueroa (1986) 41 Cal.3d 714, 722 [224 Cal.Rptr. 719, 715 P.2d 680]; People v. Ross (2007) 155 Cal.App.4th 1033, 1055 [66 Cal.Rptr.3d 438]; People v. Lee (2005) 131 Cal.App.4th 1413, 1430 [32 Cal.Rptr.3d 745].) The first jury expressly confirmed that the penalty decision was close in a note it sent to the trial court on May 27, 1999, shortly before the court declared a mistrial. That note indicated there were nine jurors who believed the issue was close enough that they were willing to go either way and one or two jurors who were firmly against imposing death.
Although the murder of Monique Cleveland (Monique) may well be the kind of crime worthy of a death sentence, the first jury’s hesitation to impose death suggests it may have been unsure whether defendant was the one who actually shot Monique. (Cf. Enmund v. Florida (1982) 458 U.S. 782, 794 [73 L.Ed.2d 1140, 102 S.Ct. 3368] [noting juries’ reluctance to impose the death penalty “for accomplice liability in felony murders”].) No one who testified at the trial had witnessed Monique’s murder. Two of the prosecution’s main witnesses, Kevin Jackson and Donald Profit, testified that defendant said he shot Monique, but neither was present at the crime. The third main witness, Robert Cleveland (Robert), “began losing consciousness and was only vaguely aware of hearing additional gunshots” after he had been shot. (Maj. opn., ante, at p. 734.) Further, both Robert’s testimony and Kevin Jackson’s indicated that one or more of defendant’s confederates had entered the Clevelands’ house before Monique was shot. The first jury’s uncertainty is reflected in another note it sent to the trial court on May 27, 1999, asking about the role that lingering doubt could play in its deliberations. In light of the evidence presented, it is reasonable to suspect that the jury had lingering *798doubt as to whether defendant shot Monique, an issue that was irrelevant at the guilt phase but obviously relevant at the penalty phase.
The evidence at the penalty retrial did not present a more compelling case for death. If anything, the prosecution’s case was weaker. The first jury heard three versions of the crimes—one from Donald Profit, another from Kevin Jackson, and a third from Robert Cleveland, a surviving victim who had suffered gunshot injuries to his face, upper back, and abdomen. The second jury heard only Kevin Jackson’s testimony. In addition to being a convicted felon and an admitted drug dealer, Kevin Jackson was not present at the shooting and, to the jury’s knowledge, was testifying pursuant to a deal with the prosecution. His testimony was based on a story defendant had told him while they smoked marijuana together a few days after the shooting.
There was little new evidence at the penalty retrial. The prosecution submitted a redacted autopsy photograph of a one-month-old embryo found in Monique’s womb. The photograph is rather clinical and “is neither gory nor particularly disturbing.” (Maj. opn., ante, at p. 757.) The prosecution also put on testimony by Correctional Sergeant Vem Nichols, who described an incident in which defendant and two other inmates refused an officer’s order to “get down.” “One of the three inmates told the officer, ‘Fuck you. We don’t have to get down,’ but Nichols was not sure whether it had been defendant.” (Id. at p. 736.) This evidence “was not especially prejudicial, particularly since the guard acknowledged that the three inmates were not a ‘direct threat’ to him, and the prosecution made no mention of the incident in its closing argument.” (Id. at p. 761.) Finally, the prosecution put on testimony by Officer Damon Aoki, who, after arresting defendant, heard defendant say he would have shot Aoki “if he had had a gun.” Although this statement was admissible to show defendant’s consciousness of guilt (id. at p. 753), defendant’s guilt was not in doubt. Officer Aoki’s testimony thus had limited value at the penalty retrial. (See id. at p. 753 [“Despite presenting this testimony to the jury, the prosecution made no mention of it during closing arguments.”].)
The witnesses for the defense, meanwhile, expanded on the testimony they had given in the first penalty phase. Defendant’s brother Antione added that their mother had abused them when they were children. Defendant’s mother added that she had forced defendant to participate in a gang-related fight and that she had been raped in front of him on multiple occasions. It was during this powerful testimony by defendant’s mother—testimony that the jury might reasonably have expected to elicit an emotional response from defendant—that the trial court’s error in forcing defendant to wear the stun belt may have been most prejudicial.
*799In sum, there is a reasonable possibility that the unlawful use of the stun belt adversely affected defendant’s demeanor and in turn influenced the jury’s decision to impose death. On this record, I am unable to conclude that the stun belt error was harmless beyond a reasonable doubt.
D.
The court today holds otherwise based on a different reading of the record. Its harmless error analysis consists of three contentions: First, any time defendant complained about the stun belt, his complaints were adequately addressed. Second, any time defendant was not complaining about the stun belt, we must infer that the stun belt did not adversely affect him. Third, the record affirmatively demonstrates the absence of prejudice. None of these contentions withstands scrutiny.
As to the first, the court would have us believe that the pillow provided to defendant and the adjustments made to the placement and tightness of the stun belt on defendant’s body were sufficient to address his complaints at the start of the guilt phase and again at the start of the penalty retrial. (Maj. opn., ante, at pp. 745, 747.) But this rendition is belied by statements at the penalty retrial that the court simply ignores, statements informed by defendant’s actual experience of wearing the belt throughout the guilt phase and first penalty phase. (See, ante, at pp. 787, 795.) I quote them again here: On October 21, 1999, defense counsel objected to the stun belt at the penalty retrial, saying, “it’s extremely uncomfortable for the defendant.” When the trial court said, “I have not been aware . . . that there has been that much of an uncomfortable feeling for the defendant by reason of the react belt,” defense counsel pressed the point: “The trial days are long, and what my concern is is that the belt is uncomfortable. It’s a large object that’s placed on the kidney area, the liver area on his back. It makes it difficult to sit back. . . . But we did in the last trial try to alleviate that discomfort by giving him a pillow to kind of even out the back so when he sat back, he wouldn’t be putting all the pressure onto that react unit that’s strapped to him. That caused a little bit of relief, but I don’t think significant.” The trial court again provided defendant with a pillow but took no other steps to address defendant’s complaints. Thus, defendant was no better off during the penalty retrial than he had been during the guilt phase and first penalty trial—as defense counsel put it, “extremely uncomfortable.”
Further, the court discounts the attending deputy’s observation on April 6, 1999, that defendant was worried about accidental activation (“I think he’s concerned [leaning back is] going to activate it . . . .”). The court says that “the defense offered no indication that defendant actually had this worry” and that “[throughout the duration of both trials, the defense never once *800stated or suggested that the threat of electric shock affected defendant’s mental state or that the belt caused him to experience mental stress or impairment.” (Maj. opn., ante, at p. 745.) Apparently the court would have us believe that the deputy’s observation of defendant’s concern about accidental activation lacks reasonable credibility because defendant did not state the concern himself or through counsel. If common sense were not enough to refute this suggestion, one need only consult our opinion in Mar, where we prominently discussed the risk of accidental activation not once, not twice, not thrice, but six different times as “a strong reason to proceed with great caution in approving the use of this device.” (Mar, supra, 28 Cal.4th at p. 1205; see id. at pp. 1206, 1219, 1226, 1228-1229; see also Durham, supra, 287 F.3d at p. 1306 [“It is reasonable to assume that much of a defendant’s focus and attention when wearing one of these devices is occupied by anxiety over the possible triggering of the belt. A defendant is likely to concentrate on doing everything he can to prevent the belt from being activated, and is thus less likely to participate fully in his defense at trial.”].)
The court’s treatment of defendant’s concern about accidental activation is symptomatic of its second thesis, that silence equals acquiescence for purposes of harmless error analysis. The court assigns great significance to the fact that “the defense made no mention at either trial of the stun belt’s potential or actual effects on defendant’s mental state” and to defendant’s “complete failure to mention the stun belt or its effects in any of his various motions for a mistrial, a new guilt trial, and a new penalty trial.” (Maj. opn., ante, at p. 747.) But why would we expect defendant to have continued complaining about the effects of the belt after his objections to wearing it had proven fruitless? By the time defendant was forced to wear the belt during the penalty retrial, the trial court had already rebuffed his objections twice— the second time notwithstanding defense counsel’s statement that it had been “extremely uncomfortable” for defendant to wear the belt throughout the guilt phase and first penalty phase. Perhaps defendant believed, reasonably, that he had already made a sufficient record. Perhaps defendant believed, reasonably, that under the only authority at the time (see Garcia, supra, 56 Cal.App.4th at p. 1356, disapproved in Mar, supra, 28 Cal.4th at p. 1219), there was no point in complaining further.
Further, the court says the record provides no indication that defendant’s lack of emotion resulted from wearing the stun belt or that he feared activation of the stun belt when the trial court warned, right before his mother’s testimony, that “the bailiff will take actions” if he engaged in verbal or nonverbal communication with his family. (Maj. opn., ante, at p. 745.) But these possibilities are surely reasonable. They are no less reasonable than the court’s own conjectures that the stun belt did not contribute to defendant’s lack of emotion and that defendant did not construe the trial court’s warning as a reference to the stun belt.
*801More fundamentally, the court’s one-sided inferences from silence in the record effectively place the burden on defendant to have made a record demonstrating prejudice from the stun belt error. Indeed, the court says as much when it declares that “reversal of a judgment is unwarranted when the record on appeal is devoid of evidence that the unjustified use of ... a stun belt had any adverse effect.” (Maj. opn., ante, at p. 740.) That is plainly not the law. In Riggins, the dissent argued that the defendant was not legally incompetent and that “[t]he record does not reveal any other form of unfairness relating to the purported side effects of [the forced medication]. Riggins has failed to allege specific facts to support his claim that he could not participate effectively in his defense. He has not stated how he would have directed his counsel to examine or cross-examine witnesses differently. He has not identified any testimony or instructions that he did not understand. The record, moreover, does not even support his assertion that [the medication] made him worse off.” (Riggins, supra, 504 U.S. at pp. 149-150 (dis. opn. of Thomas, J.).) But the high court in Riggins disclaimed any need for the defendant to prove “actual prejudice” and reversed the conviction on the ground that prejudice was “clearly possible.” (Id. at p. 137.)
Similarly, in Mar, we said, “It is not explicitly apparent from the transcript of the proceedings what effect the stun belt had on the content of defendant’s testimony or on his demeanor while testifying.” (Mar, supra, 28 Cal.4th at p. 1213.) The record showed only that the defendant was “nervous while testifying,” which “is, of course, not unusual for a defendant, or any witness,” and that the defendant at one point expressed concern about accidental activation. (Id. at p. 1224; see id. at p. 1232 (dis. opn. of Brown, J.).) Yet, in reversing the conviction, we said that “in view of the nature of a stun belt and the debilitating and humiliating consequences that such a belt can inflict,” “it is reasonable to conclude that defendant’s being required to wear the stun belt had at least some effect on his demeanor while testifying.” (Id. at pp. 1224-1225.)
The Eleventh Circuit in Durham likewise found prejudicial stun belt error on a record that contained only nonspecific assertions by defense counsel that wearing the belt would induce “ ‘extreme fear’ ” that “chills the exercise of the defendant’s trial rights, including altering his outward appearance and affecting his decision whether or not to testify, his ability to follow the proceedings, the substance of his communication with counsel, and his ability to actively cooperate with and assist counsel.’ ” (Durham, supra, 287 F.3d at pp. 1310-1311 (conc. opn. of Tjoflat, J.).) Nothing in the record indicated that any of these concerns actually came to pass during the trial. Yet the court reversed the conviction because “[t]he government ha[d] not demonstrated that Durham’s defense was not harmed by” the stun belt. (Id. at p. 1309.) Applying Chapman, the court said it is not “sufficient for the government to argue that the defendant cannot name any outcome-determinative issues or *802arguments that would have been raised had he been able to participate at trial. Not only does such an argument impermissibly transfer the burden of proof back to the defendant, but it also would eviscerate the right in all cases where there is strong proof of guilt.” (Ibid.)
As Riggins, Mar, and Durham illustrate, silence in the record with respect to actual prejudice does not dispel a reasonable possibility of prejudice, especially when the error is one that carries a “substantial risk” of prejudice. (Durham, supra, 287 F.3d at p. 1305; Stevens, supra, 47 Cal.4th at p. 632.) To say that “reversal of a judgment is unwarranted when the record on appeal is devoid of evidence that the [error] had any adverse effect” (maj. opn., ante, at p. 740) is to effectively place the burden on defendant to demonstrate actual prejudice. In this respect, today’s opinion contravenes Chapman.
The court’s third and last resort is its remarkable claim that the record “affirmatively demonstrates] the assessment of defense counsel that the jurors’ consideration of defendant’s courtroom demeanor would have a favorable effect on their penalty determination.” (Maj. opn., ante, at p. 746.) According to the court, this assessment may be inferred from defense counsel’s insistence on two special jury instructions proposed in a May 25, 1999 filing at the first penalty trial and again in a November 18, 1999 filing at the second penalty trial. Here is what the two proposed instructions say: (1) “PROPOSED PENALTY PHASE INSTRUCTION NO. 10 [][] Mitigating factors are not necessarily limited to those adduced from specific evidence offered at the sentencing hearing such as character testimony. A juror might be disposed to grant mercy based on other factors, such as a humane perception of the [defendant] developed during trial.” (2) “PROPOSED PENALTY PHASE INSTRUCTION NO. 11 [][] If a mitigating circumstance or an aspect of the defendant’s background or his character called to the attention of the jury by the evidence or its observation of the defendant arouses mercy, sympathy, empathy, or compassion such as to persuade you that death is not the appropriate penalty, you may act in response thereto and impose a sentence of life without possibility of parole.” At the May 24, 1999 and November 18, 1999 hearings on these proposed instructions, the parties and the trial court regarded them as essentially the same.
As an initial matter, this is an argument entirely of the court’s own invention, as the Attorney General nowhere relied on this aspect of the record to demonstrate harmless error. The court’s reliance on this argument seems a rather questionable way to adjudicate whether the state has carried its burden of proving the absence of prejudice beyond a reasonable doubt. (See Chapman, supra, 386 U.S. at p. 24.)
In any event, it is no wonder that the Attorney General never made this argument in her briefing or at oral argument. From the text of the two *803proposed instructions and from the transcripts of the two hearings on those instructions, it is clear that defense counsel sought to inform the jury that if its observation of defendant aroused mercy, sympathy, empathy, or compassion, then it could consider this factor in choosing life over death. Defense counsel gave no indication that he believed the jury would find defendant’s courtroom demeanor to be mitigating. As defense counsel explained at the November 18, 1999 hearing, he was simply seeking to enumerate for the jury as many potentially mitigating factors as he could. The court’s assertion that “the only reasonable inference” is that the stun belt did not negatively affect defendant’s demeanor (maj. opn., ante, at p. 746) would be more convincing if the proposed instructions had merely told the jury it was free to consider defendant’s demeanor in its sentencing decision—and left it at that. But a proposed instruction that says the jury may take demeanor into account if the jury finds it to be mitigating does not firmly indicate that defense counsel believed defendant’s demeanor would weigh in his favor. Rather, the conditional language of the proposed instruction, coupled with defense counsel’s opposition to any prosecutorial comment on defendant’s demeanor, suggests that defense counsel saw a downside risk to the jury’s consideration of demeanor. Indeed, the protective tilt of the proposed instructions was why the prosecutor opposed the instructions and why the trial court at the penalty retrial refused to give them. The court’s reliance on this sole aspect of the record as “affirmatively demonstrating” the absence of prejudice (id. at p. 746) serves only to underscore the lengths to which the court must go to reach its holding today.
Finally, the court’s harmless error analysis is laced with string cites purporting to show that “reversal of a judgment is unwarranted when the record on appeal is devoid of evidence that the unjustified use of shackles or a stun belt had any adverse effect.” (Maj. opn., ante, at p. 740, italics added; see id. at pp. 740-741, 748-748.) But this conflation of shackles and stun belts elides an important distinction.
Most of the cases cited by the court involved traditional physical restraints like shackles, leg restraints, or handcuffs. We have consistently found the use of such restraints to be harmless where the jury never saw the restraints and where no evidence suggests they affected the defendant’s decision to testify or any other aspect of the defense. (See People v. Foster (2010) 50 Cal.4th 1301, 1322 [117 Cal.Rptr.3d 658, 242 P.3d 105] (Foster); People v. Letner and Tobin (2010) 50 Cal.4th 99, 155 [112 Cal.Rptr.3d 746, 235 P.3d 62] (Letner); People v. Ervine (2009) 47 Cal.4th 745, 773-774 [102 Cal.Rptr.3d 786, 220 P.3d 820]; People v. Wallace (2008) 44 Cal.4th 1032, 1051 [81 Cal.Rptr.3d 651, 189 P.3d 911]; People v. Combs (2004) 34 Cal.4th 821, 838-839 [22 Cal.Rptr.3d 61, 101 P.3d 1007]; People v. Anderson (2001) 25 Cal.4th 543, 596 [106 Cal.Rptr.2d 575, 22 P.3d 347]; Cox, supra, 53 Cal.3d at p. 652.) These cases stand to reason because the principal vector of prejudice when it *804comes to shackles, leg restraints, or handcuffs is its visibility to the jury, which erodes the presumption of innocence.
By contrast, “the greatest danger of prejudice [from improper use of a stun belt] arises from the potential adverse psychological effect of the device upon the defendant rather than from the visibility of the device to the jury.” (Mar, supra, 28 Cal.4th at p. 1225, fn. 7.) No case law suggests that traditional restraints have the potential to achieve the “ ‘ “total psychological supremacy” ’ ” over the defendant that a stun belt is designed to achieve. (Id. at p. 1226.) Indeed, this court in Letner quoted Mar’s graphic description of what a stun belt does to a person when activated and said, “There is no evidence that the leg brace worn by Letner created a remotely comparable level of potential pain, injury, and humiliation, such that Letner’s ability to concentrate and participate in the trial proceedings similarly might have been affected.” (Letner, supra, 50 Cal.4th at p. 156.) Thus, our cases finding the invisible use of traditional restraints harmless in the absence of affirmative evidence that the defense was impaired do not suggest an equivalent rule for stun belts.
Our cases that have found erroneous use of a stun belt to be harmless likewise do not support today’s holding. In People v. Manibusan (2013) 58 Cal.4th 40 [165 Cal.Rptr.3d 1], the defendant expressly agreed to wear the belt, so it is unsurprising we found any error to be harmless in the absence of evidence showing an adverse effect. (Id. at pp. 85-86; accord, Stanford v. State (2000) 272 Ga. 267 [528 S.E.2d 246, 249-250] [finding harmless error where the defendant did not object to wearing a stun belt and the record did not show prejudice]; State v. Johnson (2006) 112 Ohio St.3d 210 [2006 Ohio 6404, 858 N.E.2d 1144, 1179-1180] [same].) In People v. Virgil, supra, 51 Cal.4th 1210, we held that the trial court did not err in ordering the use of a stun belt, and we did not engage in harmless error analysis. (Id. at p. 1271.) In Foster, the defendant did not object to the stun belt, and we deemed the issue forfeited. (Foster, supra, 50 Cal.4th at pp. 1321-1322.)
That leaves Howard, supra, 51 Cal.4th 15, the case on which the court principally relies. (Maj. opn., ante, at p. 746.) But our finding of harmless error in that case turned “[m]ost significantly” on our observation that “at sentencing defendant made an extended statement to the court regarding his mental state during trial. He made no reference to a stun belt or its effects on his demeanor while testifying. Instead, defendant claimed his testimony had been affected by antipsychotic medication.” (Howard, at p. 29.) We concluded that “the record affirmatively dispels any notion that [the defendant] was prejudiced” because the stun belt “had no appreciable effect on him, even by his own account.” (Id. at pp. 28, 30; see id. at p. 30, fn. 6 [“[T]he concern that defendant may have been psychologically affected by a stun belt in this *805case was put to rest by his own statements to the court about his mental state during trial.”].) Because the record here does not include any fact remotely similar to the fact we found dispositive in Howard, it strains credulity to say that the record in this case, as in Howard, “affirmatively dispels any notion of prejudice.” (Maj. opn., ante, at p. 748.)
In sum, today’s opinion dismisses any possibility of prejudice at the penalty retrial as “pure conjecture.” (Maj. opn., ante, at p. 745.) Even though defendant repeatedly complained about the belt, even though he naturally worried about accidental activation, even though the trial court observed “lack of emotion” in his demeanor, even though the trial court warned right before his mother testified that “the bailiff will take actions” if he engaged in any verbal or nonverbal communication with his family, and even though the first jury hung at the penalty phase, the court says it is certain beyond a reasonable doubt that the jury would have reached the same verdict absent the stun belt error. Now that is pure conjecture—and it is conjecture based on reasoning plainly at odds with Chapman.
HI.
Observers of this court have long noted the prominent role of harmless error doctrine in our death penalty jurisprudence. One study of 215 capital cases decided by this court between 1986 and 1996 reported that the court found guilt phase error in 55 percent of the cases and penalty phase error in 74 percent of the cases, but found 93 percent of guilt phase errors and 85 percent of penalty phase errors to be harmless, producing a reversal rate of 3.8 percent for guilt verdicts and 11 percent for death verdicts. (Kamin, Harmless Error and the Rights/Remedies Split (2002) 88 Va. L.Rev. 1, 70.) By contrast, 33 percent of the death sentences meted out nationally from 1986 through 1996 have been overturned on appeal or collateral review, and the reversal rate is even higher for death judgments prior to 1986. (U.S. Dept, of Justice, Bur. of Justice Statistics, Capital Punishment, 2001—Statistical Tables (2013) p. 19, table 16, available at <http://www.bjs.gov/content/pub/ pdf/cpllst.pdf> [as of Mar. 3, 2014]; see Liebman et al., Capital Attrition: Error Rates in Capital Cases, 1973-1995 (2000) 78 Tex. L.Rev. 1839, 1850 [reporting that state and federal courts nationwide found prejudicial error in 68 percent of capital cases between 1973 and 1995].)
Unfortunately, the dubious reasoning in today’s decision is not an isolated lapse in our harmless error jurisprudence. For example, just as the court here supplies an argument on the state’s behalf that the state itself never made in confronting its burden to demonstrate harmless error (ante, at p. 760), the court in another case recently found the omission of a reasonable doubt *806instruction to be harmless under Chapman despite the state’s concession in its answer brief that the error required reversal. (People v. Aranda (2012) 55 Cal.4th 342, 379, 383 [145 Cal.Rptr.3d 855, 283 P.3d 632] (conc. & dis. opn. of Liu, J.) (Aranda).)
In addition, the court’s willingness to infer no reasonable possibility of prejudice where the record is silent or indeterminate as to actual prejudice is a recurring maneuver in our harmless error doctrine. Consider, for example, the reasoning in People v. Whitt (1990) 51 Cal.3d 620 [274 Cal.Rptr. 252, 798 P.2d 849] (Whitt). The defendant in Whitt had taken the witness stand during the penalty phase of his capital trial. In response to two questions posed by defense counsel—“ ‘[D]o you want to live?’ ” and “ ‘Why do you deserve to live?’ ”—the prosecutor objected, and the trial court sustained the objections. (Id. at p. 646.) This court held that the trial court’s rulings, by denying the defendant an opportunity to plead for his life before the jury, “violated his federal constitutional right to have the sentencer consider all relevant mitigating evidence.” {Id. at p. 647.) But the court went on to find the error harmless under Chapman: “In this case, though the question ‘Why do you deserve to live?’ might produce a significant answer, the phraseology is so inherently broad, and the range of conceivable answers so vast, that we cannot know whether defendant’s actual response might have influenced the penalty determination.” (Id. at p. 648.) Reversal was not warranted under these circumstances, this court concluded, because the defendant had not made a record of what he would have said and thus prejudice could not be assessed. (Ibid.) As to Chapman’s mandate that the burden is on the state to prove the error was harmless beyond a reasonable doubt, the court said “the ‘state-burden’ language in Chapman does not literally mean that an appellate court must reverse the judgment because the prosecution has failed to place evidence in the record showing that the error was harmless. Rather, Chapman . . . [means that] where federal constitutional error occurs and prejudice can be assessed from the record, there is a strong presumption that the defendant was prejudiced.” (Id. at p. 649, italics added.)
Three justices dissented. Justice Mosk explained that the court’s “analysis rests on an implied premise that the burden of proof as to prejudice rests on the person complaining of the error. But as the very words of Chapman demonstrate, that premise is radically unsound: ‘Certainly error, constitutional error, . . . casts on someone other than the person prejudiced by it a burden to show that it was harmless .... [T]he beneficiary of a constitutional error [is required] to prove beyond a reasonable doubt that the error complained of did not contribute to the verdict obtained.’ (386 U.S. at p. 24 .. ..)” (Whitt, supra, 51 Cal.3d at p. 666 (conc. & dis. opn. of Mosk, J.).) “The People simply fail to carry that burden. Because of the error, we cannot know what specific answers defendant might have given or what precise force those answers might have carried. The void in the record is indeed disturbing. But it is the *807People that created the void. And it is the People that must bear the responsibility.” (Ibid. (conc. & dis. opn. of Mosk, J.).) Justice Kennard similarly explained; “To condition an evaluation of federal constitutional error on a defendant’s offer of proof, as the majority does, relieves the beneficiary of the error (the prosecution) of the obligation to demonstrate that the error did not affect the verdict, and impermissibly shifts to the defendant the burden of proving prejudice.” (Id. at pp. 671-672 (conc. & dis. opn. of Kennard, J.); see id. at pp. 668-669 (conc. & dis. opn. of Broussard, J.).)
Whitt was not wrong to say that “the ‘state-burden’ language in Chapman-does not literally mean that an appellate court must reverse the judgment because the prosecution has failed to place evidence in the record showing that the error was harmless.” (Whitt, supra, 51 Cal.3d at p. 649.) The state’s burden in a harmless error inquiry is not a traditional evidentiary burden of production. As the high court has observed, harmless error inquiry “does not involve a judge who shifts a ‘burden’ to help control the presentation of evidence at a trial, but rather involves a judge who applies a legal standard (harmlessness) to a record that the presentation of evidence is no longer likely to affect.” (O’Neal, supra, 513 U.S. at p. 436.)
However, to say that the prosecution bears no burden of production is not to say that reversal is unwarranted whenever a defendant who objected to the error has not made a record from which a reviewing court may assess prejudice. Whitt found harmless error on the rationale that “we cannot know whether defendant’s actual response might have influenced the penalty determination” (Whitt, supra, 51 Cal.3d at p. 648); in other words, the trial court’s error might or might not have affected the verdict. If the infirmity of this reasoning were not already evident in light of Chapman, it is surely evident in light of O’Neal, where the high court, relying on Chapman, made clear that an error cannot be found harmless where “the matter is so evenly balanced that [the court] feels [itself] in virtual equipoise as to the harmlessness of the error.” (O’Neal, supra, 513 U.S. at p. 435.) Similarly, in Riggins, the high court reversed a capital conviction where “[e]fforts to prove or disprove actual prejudice from the record before us would be futile, and guesses whether the outcome of the trial might have been different [without the error] would be purely speculative.”,(Riggins, supra, 504 U.S. at p. 137.) Whitt misapplied Chapman and cannot survive the high court’s subsequent precedent. And yet, this court continues to cite Whitt for the proposition that no cognizable prejudice can result from the erroneous exclusion of potentially mitigating testimony at the penalty phase of a capital trial unless the defendant can show from the record that the testimony would have affected the verdict. (See, e.g., People v. Earp (1999) 20 Cal.4th 826, 891-892 [85 Cal.Rptr.2d 857, 978 P.2d 15], citing Whitt, at p. 648.) The court’s opinion in this case is cut from the same cloth.
*808IV.
Today’s decision confirms that “[a]ppellate judges ... are often reluctant to open the way to a new trial, given not only the risk of draining judicial resources but also the risk that a guilty defendant may go free” or escape just punishment. (Traynor, supra, at p. 50.) Yet “[t]he very reluctance of judges to confront such risks . . . serves to condone errors that may affect a judgment and thus engenders a still more serious risk, the risk of impairing the integrity of appellate review.” (Ibid.) The Chapman rule—both its reasonable doubt standard and its allocation of the burden of persuasion—has long endured as the proper accommodation of these competing risks. When a reviewing court fails to faithfully apply the standard, it compromises the fairness of the proceeding, usurps the jury’s function, and weakens the role of appellate review in deterring future errors.
Regrettably, today’s decision is only the most recent instance in which this court has deviated from Chapman’s mandate. The deviations have occurred in capital cases (see, e.g., Gamache, supra, 562 U.S. at p. 1084 [131 S.Ct. at p. 592] (statement of Sotomayor, J.); Whitt, supra, 51 Cal.3d at pp. 666-668 (conc. & dis. opn. of Mosk, J.)) as well as noncapital cases (see Aranda, supra, 55 Cal.4th at p. 377 (conc. & dis. opn. of Liu, J.)). Given the precedential force of these decisions, it is reasonable to worry that Chapman will continue to mean something different in the courts of California than what the high court has repeatedly said it means.
Today’s opinion signals a retreat from the rigorous standards governing the use of stun belts and similar devices. It also deviates from the degree of accuracy and reliability that is required when a jury is entrusted to make the most consequential decision allowed to be made in a court of law. For these reasons, I respectfully dissent from the court’s judgment upholding the sentence of death. In all other respects, I join the court’s opinion.
Appellant’s petition for a rehearing was denied May 14, 2014.