UNITED STATES v. THOMAS (2025)

OPINION

Deangelus Thomas was indicted on two counts of being a felon in possession under 18 U.S.C. § 922(g)(1) for his role in a shooting. Though his indictment gave him notice that he might be subject to enhanced penalties based on his criminal history, he was not formally indicted as an armed career criminal. At trial, a jury found him guilty of both felon-in-possession counts.

Throughout sentencing, Thomas maintained that he could not be sentenced under the Armed Career Criminal Act (ACCA). 18 U.S.C. § 924(e)(1). Because he had never been indicted for it, and the jury had not found the essential fact —that he had three prior violent-felony convictions committed on different “occasions,” —he claimed he could only be subject to the penalties associated with § 922(g)(1). Otherwise, his Fifth and Sixth Amendment rights would be violated.

The district judge disagreed. Following then-binding Sixth Circuit precedent, the judge found the fact of Thomas's three prior convictions by a preponderance of evidence at sentencing, and imposed an enhanced, 432-month sentence on Thomas. We affirmed. But the Supreme Court later decided Erlinger v. United States, 602 U.S. 821, 144 S.Ct. 1840, 219 L.Ed.2d 451 (2024), which held that the Fifth and Sixth Amendments require a jury to find the three-occasions element of an ACCA conviction. Now, Thomas is back before us.

He argues that Erlinger is a form of structural error and requires automatic reversal. As a remedy, he argues for a remand to the district judge for resentencing based on the crime he was convicted of by a jury—the unenhanced § 922(g)(1). Because our circuit has already decided that Erlinger errors are subject to harmless error, we affirm his sentence because the failure to charge the jury was harmless.

I.

Thomas thought his girlfriend was cheating on him. Jealous, Thomas pulled out a handgun, chambered a round, and threatened her with her life if she didn't confess. Once she admitted to seeing another man on the side, Thomas punched her twice and pistol-whipped her. Still, Thomas wanted revenge on her lover. So he forced his girlfriend to set up a meeting so Thomas could confront him.

Thomas and his girlfriend drove together to a predetermined location where Thomas forced her into the driver seat while he hid in the backseat to ambush her lover. When he approached the car, Thomas shot him twice with a rifle. Thomas and his girlfriend drove off to his cousin's house, leaving the lover to fend for himself.

His cousin called the police, advising them of Thomas's assault on his girlfriend. Memphis Police Officers, with consent from his cousin, searched the area around Thomas while he slept and recovered a DB9R rifle loaded with five rounds of ammo and a Sig Sauer P239 pistol loaded with eight rounds. Because Thomas has been convicted of several violent felonies, his possession of a firearm was prohibited.

On January 19, 2021, Thomas was arrested, and officers confirmed through witnesses and their own investigation that Thomas was responsible for his girlfriend's injuries and for shooting her lover. For these events, Thomas was indicted on two counts of being a felon in possession in violation of 18 U.S.C. § 922(g)(1). Thomas went to trial and the jury found him guilty of both.

A garden-variety § 922(g)(1) conviction has a maximum sentence of 15 years. 18 U.S.C. § 924(a)(8). But if a defendant is an “armed career criminal, the penalty ratchets up to a mandatory minimum of 15 years if he has “three previous convictions” for a “violent felony” that he “committed on occasions different from one another.” Id. § 924(e)(1).

Between the time Thomas was indicted and sentenced, the Supreme Court decided Wooden v. United States, 595 U.S. 360, 369, 142 S.Ct. 1063, 212 L.Ed.2d 187 (2022), which concluded that the occasions-different inquiry was fact-intensive and “multi-factored in nature.” By implication Thomas argued, the three-occasions requirement was an element of the enhanced offense that needed to be charged in the indictment, proven to a jury, and found beyond a reasonable doubt. And the failure to do so had violated his Fifth and Sixth Amendment rights. After a hearing on the question, the judge overruled Thomas's objections and the government's request for a sentencing jury. Because then-binding Sixth Circuit precedent had determined that Wooden left untouched the conclusion that a judge could find the facts necessary to apply the ACCA enhancement, the judge made the necessary factual findings and imposed the enhanced sentence.

The judge, relying on the facts of the Presentence Report (PSR), found the three predicate offenses necessary for the enhancement. Thomas's first predicate offense was an aggravated robbery from October 2000. Second, was an aggravated assault from December 2007. Finally, was a conviction for attempted second-degree murder from August 2010. Based on these findings, and the judge's consideration of the 18 U.S.C. § 3553 factors, the judge found Thomas was an armed career criminal and calculated the advisory-Guidelines range: thirty years to life. He imposed a 432-month term of imprisonment and a five-year term of supervised release.

Thomas appealed, arguing that the ACCA enhancement could not be imposed without it first being indicted and found by a jury beyond a reasonable doubt. United States v. Thomas, No. 22-6067, 2023 WL 5535124, at *2 (6th Cir. Aug. 8, 2023). Because then-binding Sixth Circuit precedent held to the contrary, we affirmed. So Thomas petitioned for, and was granted, certiorari on the Fifth and Sixth Amendment questions raised by imposing an ACCA enhancement without a jury finding. Thomas v. United States, ––– U.S. ––––, 144 S. Ct. 2710, 219 L.Ed.2d 1315 (2024). The Supreme Court then remanded for further consideration in light of Erlinger, 602 U.S. 821, 144 S.Ct. 1840. Now, we must determine how Erlinger impacts Thomas's sentence.

II.

Thomas challenges his enhanced sentence on three grounds. First, he claims Erlinger errors are structural. Second, he argues that, even if Erlinger errors are subject to harmless error, the government has not met its burden of proof because the Shepard documents are unreliable for facts unrelated to the essential elements of a prior conviction. And finally, he claims that jeopardy attached when he was tried for the unenhanced offense, and so his case must be remanded for resentencing based on the jury's conviction.

The government concedes that after Wooden and Erlinger it was error not to charge in the indictment the occasions-different element, not to have the jury be instructed on the same, and not to have them find the facts beyond a reasonable doubt. Still, it maintains that these errors were harmless. And because a previous panel has resolved that Erlinger errors are not structural, and because the Supreme Court has reiterated that Shepard documents may be used for the limited purpose of harmless-error review, we hold that the Erlinger error was harmless in his case. Nor does the Double Jeopardy Clause pose a problem.

A.

Thomas argues that Erlinger is a form of structural error—it renders the proceeding “fundamentally unfair or an unreliable vehicle for determining guilt or innocence” and so his sentence is subject to automatic reversal. Neder v. United States, 527 U.S. 1, 9, 119 S.Ct. 1827, 144 L.Ed.2d 35 (1999). In support of this view, he raises three primary arguments. First, he claims that omitting this element from the jury's purview operates as a wholesale abridgement of his jury trial right, and under Sullivan v. Louisiana, constitutes a structural error. 508 U.S. 275, 278, 113 S.Ct. 2078, 124 L.Ed.2d 182 (1993). Next, he argues that harmless error presupposes a trial. So applying harmless error when the indictment did not allege the three occasions, and when no evidence was admitted at trial, would effectively have the court act as a hypothetical jury, impermissibly intruding on the Sixth Amendment. And finally, he asks us to read the tea leaves and based on the Erlinger majority's unwillingness to answer whether the error was harmless, we should view this as evidence that the error is structural.

After briefing occurred in this case, another panel of this court responded to these exact arguments, dismissed each, and concluded that Erlinger errors are subject to harmless-error review. United States v. Campbell, 122 F.4th 624, 629–31 (6th Cir. 2024), denying reh'g en banc, No. 22-5567 (6th Cir. Feb. 19, 2025) (order). Because we are bound by the decisions of another panel unless there is intervening Supreme Court authority that requires modification, or we sit en banc to overrule the panel's decision, we need not rehash the same arguments. Salmi v. Sec'y of Health & Hum. Servs., 774 F.2d 685, 689 (6th Cir. 1985); see also United States v. Cogdill, 130 F.4th 523, 527–31 (6th Cir. 2025) (applying harmless error to an Erlinger error based on Campbell, 122 F.4th 624).

Campbell’s logic is two-fold. First, most constitutional errors are not structural, and the Supreme Court has applied harmless-error review when an element is omitted from the jury instructions, so by analogy, a district judge's factfinding should also be subject to harmless error. Campbell, 122 F.4th at 630 (citing Washington v. Recuenco, 548 U.S. 212, 218, 222, 126 S.Ct. 2546, 165 L.Ed.2d 466 (2006) and Neder, 527 U.S. at 18, 119 S.Ct. 1827). Second, because Erlinger was “nearly on all fours” with Apprendi v. New Jersey, 530 U.S. 466, 120 S.Ct. 2348, 147 L.Ed.2d 435 (2000), and Alleyne v. United States, 570 U.S. 99, 133 S.Ct. 2151, 186 L.Ed.2d 314 (2013), and because we have consistently reviewed such errors for harmlessness, Erlinger’s remedy should be the same. Campbell, 122 F.4th at 629–30 (quoting Erlinger, 602 U.S. at 835, 144 S.Ct. 1840).

Even if we could overrule a panel decision, we see no reason to disagree. So for the reasons outlined in Campbell, we proceed to determine whether the error in this case was harmless.

B.

Thomas claims that even if Erlinger errors aren't structural, this error was harmful. We will not set aside an “otherwise valid conviction,” if we can “confidently say, on the whole record, that the constitutional error was harmless beyond a reasonable doubt.” Rose v. Clark, 478 U.S. 570, 576, 106 S.Ct. 3101, 92 L.Ed.2d 460 (1986) (quoting Delaware v. Van Arsdall, 475 U.S. 673, 681, 106 S.Ct. 1431, 89 L.Ed.2d 674 (1986)). The government bears the burden to show that an error was harmless. Reiner v. Woods, 955 F.3d 549, 555 (6th Cir. 2020) (citing Arizona v. Fulminante, 499 U.S. 279, 296, 111 S.Ct. 1246, 113 L.Ed.2d 302 (1991)). So we ask whether the “record contains evidence that could rationally lead to a contrary finding with respect to the omitted element.” Neder, 527 U.S. at 19, 119 S.Ct. 1827.

Wooden acknowledges that, typically, the inquiry for the three occasions is “multifactored in nature.” 595 U.S. at 369, 142 S.Ct. 1063. Because a “range of circumstances may be relevant to identifying episodes of criminal activity”—that is, just because two crimes are committed at different times, does not automatically mean they are on different occasions. Id. Wooden itself recognizes that “most cases” won't involve “extra-ordinary work”; the factors, though reflective of the fact-based nature of the inquiry, are “straightforward and intuitive.” Id. at 369–70, 142 S.Ct. 1063. So even if there may be “some hard cases in between,” the single factor of time or place will often “decisively differentiate occasions.” Id. at 370, 142 S.Ct. 1063. Thomas's is such a straightforward case.

The government makes two points to satisfy its harmless-error burden: the dates of each offense and the absence of any common scheme or purpose. Based on a review of the Shepard documents described in the PSR, the district judge's error of finding the occasions-different element by a preponderance of the evidence was harmless beyond a reasonable doubt.

The three predicate offenses took place years apart—in fact, a decade separates the first and third offenses. October 2000, December 2007, and August 2010. They also took place in three different places. For example, the first was at a residence, the second in a parking lot. Each involved a different victim. Each involved a different crime. The first an aggravated robbery, the second an aggravated assault, and the third, attempted-second-degree murder. The only unifying theme is their inherent violence and Thomas's proclivity for violence with a firearm. Neither of which suggests that Thomas was engaged in one continuous violent felony in the way Wooden describes.

Thomas fights this conclusion by challenging our use of Shepard documents. On his view, they are “prone to error” and so cannot be relied on for non-elemental information they contain. R.25, Appellant Supp. Br., p.14 (quoting Mathis v. United States, 579 U.S. 500, 512, 136 S.Ct. 2243, 195 L.Ed.2d 604 (2016)). He claims Erlinger cast Shepard documents as “inherently unreliable.” Id. at 14. But Campbell addressed this too.

It's true that Erlinger has some harsh words about Shepard documents. Erlinger, 602 U.S. at 839–42, 144 S.Ct. 1840 (discussing Shepard v. United States, 544 U.S. 13, 125 S.Ct. 1254, 161 L.Ed.2d 205 (2005)). As Campbell notes, the fact that such documents are prone to error is in part why Erlinger determined that the occasions-different inquiry should be submitted to the jury. Campbell, 122 F.4th at 632. That conclusion, however, doesn't preclude our reliance on these documents for our harmless-error review. With caution, appellate courts must look to “all ‘relevant and reliable information’ in the ‘entire record,’ which in turn is gauged through ‘case-by-case adjudication rather than through a categorical bar against’ certain categories of evidence.” Id. at 632–33 (quoting Greer v. United States, 593 U.S. 503, 511, 141 S.Ct. 2090, 210 L.Ed.2d 121 (2021)). And that includes the information in the PSR. Erlinger, 602 U.S. at 841, 144 S.Ct. 1840; Greer, 593 U.S. at 510–11, 141 S.Ct. 2090.

That doesn't mean that Shepard documents are automatically valid. Where there is reason to, a defendant can challenge the validity of the PSR and its contents. This includes instances where “adversarial testing was ‘unnecessary’ ” in the proceeding where the facts were originally developed. Erlinger, 602 U.S. at 841, 144 S.Ct. 1840 (quoting Mathis, 579 U.S. at 512, 136 S.Ct. 2243). In that way, the defendant can undermine the government's theory of harmless error. But without such an objection, and absent a reason to question the reliability of the PSR's findings, we can rely on its contents for our harmless-error inquiry.

Below, Thomas only challenged the constitutional nature of the occasions-different inquiry. So because he didn't contest the contents of the Shepard documents or the PSR, we have no reason to doubt the information synthesized by it and we can rely on its contents to affirm his sentence.

Thomas raises a final challenge to the harmless-error analysis. He claims that in many other cases, juries have acquitted on the three-occasions question; essentially theorizing that the possibility of a jury acquittal is enough to undermine a finding of harmless error. But again, we rejected this attack in Campbell: “[I]f only a theoretical possibility of acquittal were sufficient to defeat demonstrated harmless error, harmless error could not exist.” Campbell, 122 F.4th at 633.

At bottom, the government has satisfied its burden to show that its failure to submit the occasions-different question to the jury was harmless. The Shepard documents—as aggregated by the PSR—provide the necessary information to answer the occasions-different inquiry. Finding that an error was harmless does not constitute a “denigration” of Thomas's Fifth and Sixth Amendment jury trial rights. Neder, 527 U.S. at 19, 119 S.Ct. 1827 (quoting Rose, 478 U.S. at 577, 106 S.Ct. 3101). Instead, it reflects the “very useful purpose” of ensuring that convictions are not set aside for “small errors or defects” that have “little, if any, likelihood of having changed the result of the trial.” Id. (quoting Chapman, 386 U.S. at 22, 87 S.Ct. 824). 1

C.

Thomas's final argument is that double-jeopardy principles prevent the harmless-error conclusion in these circumstances. He argues that because jeopardy attached during his trial for the unenhanced § 922(g)(1) offense, the government cannot “punish him as if he had pled [guilty] to it.” R.25, Appellant Supp. Br., p.6–7.

1.

The Double Jeopardy Clause provides that no person shall “be subject for the same offence to be twice put in jeopardy of life or limb.” U.S. Const. amend. V. It offers three distinct but related protections. A defendant may not be subject to (1) a “second prosecution for the same offense after acquittal,” (2) a “second prosecution for the same offense after conviction,” or (3) “multiple punishments for the same offense.” Brown v. Ohio, 432 U.S. 161, 165, 97 S.Ct. 2221, 53 L.Ed.2d 187 (1977) (quoting North Carolina v. Pearce, 395 U.S. 711, 717, 89 S.Ct. 2072, 23 L.Ed.2d 656 (1969)).

Double jeopardy is unlike other protections and is not concerned with finality. Instead, jeopardy relates to the court's power “to determine [a defendant's] guilt or innocence,” Soto v. Siefker, 79 F.4th 715, 719 (6th Cir. 2023), so it does not hinge on a final judgment, Swisher v. Brady, 438 U.S. 204, 214–15, 98 S.Ct. 2699, 57 L.Ed.2d 705 (1978). Jeopardy questions, then, are a matter of both timing and substance—that is, they require answering whether jeopardy has attached and whether two crimes constitute the “same offense.”

In a jury trial, jeopardy attaches when the “jury is empaneled and sworn.” Martinez v. Illinois, 572 U.S. 833, 834, 134 S.Ct. 2070, 188 L.Ed.2d 1112 (2014) (quoting Crist v. Bretz, 437 U.S. 28, 35, 98 S.Ct. 2156, 57 L.Ed.2d 24 (1978)). And whether jeopardy precludes a second prosecution based on the “same offense” is answered by the Blockburger test—if the same “act” violates two “distinct statutory provisions,” the question is whether “each provision requires proof of a fact which the other does not.” Blockburger v. United States, 284 U.S. 299, 304, 52 S.Ct. 180, 76 L.Ed. 306 (1932). Unless two statutes require proof of an “additional fact,” double jeopardy forbids a second or successive prosecution. Brown, 432 U.S. at 166, 97 S.Ct. 2221 (internal quotations marks omitted).

So lesser and greater offenses raise potential double-jeopardy concerns because of the relationship between the elements necessary to prove each. A lesser offense requires “no proof beyond that which is required for conviction of the greater” offense and therefore the greater offense is “by definition” the same as the lesser offense for purposes of double jeopardy. Id. at 168, 97 S.Ct. 2221. And the double-jeopardy bar applies regardless of the order of prosecution—a second prosecution is prohibited (except for certain exceptions inapplicable here) no matter if the prosecution for the lesser or greater offense came first. Id.

2.

Because Erlinger is an outgrowth of Apprendi, the conclusion that a jury must find the occasions-different element effectively confirms that § 922(g)(1) and § 924(e)(1) convictions are related, lesser and greater offenses, respectively. Without a finding of three occasions, a felon in possession is subject only to the lesser offense in § 922(g)(1) and the corresponding penalty. But if the government proves both the lesser § 922(g)(1) offense and that the defendant committed three prior violent felonies on different occasions in § 924(e)(1), the defendant is guilty of the greater offense and the enhanced penalties.

Controlled substance prosecutions illustrate why this is. It is unlawful for any person to “knowingly or intentionally” “manufacture, distribute, or dispense, or possess with intent to manufacture, distribute, or dispense, a controlled substance.” 21 U.S.C. § 841(a)(1). But if death results from one of these crimes, then the defendant is subject to an enhanced mandatory minimum and maximum. Id. § 841(b)(1)(A), (b)(1)(B), (b)(1)(C). The Supreme Court explained—relying on Apprendi and Alleyne—that “[b]ecause the ‘death results’ enhancement increased the minimum and maximum sentences to which [the defendant is] exposed, it is an element that must be submitted to the jury and found beyond a reasonable doubt.” Burrage v. United States, 571 U.S. 204, 210, 134 S.Ct. 881, 187 L.Ed.2d 715 (2014).

So after Burrage, a violation of § 841(a)(1) is a “lesser included offense.” Id. at 210 n.3, 134 S.Ct. 881. By contrast, the death-resulting charge is the greater offense and requires the government to prove “two principal elements: (i) knowing or intentional distribution of heroin, § 841(a)(1), and (ii) death caused by (‘resulting from’) the use of that drug.” Id. at 210, 134 S.Ct. 881 (footnote and citation omitted) (emphasis added); see also Alleyne, 570 U.S. at 104–05, 133 S.Ct. 2151 (“The Sixth Amendment provides that those ‘accused’ of a ‘crime’ have the right to a trial ‘by an impartial jury.’ This right, in conjunction with the Due Process Clause, requires that each element of a crime be proved to the jury beyond a reasonable doubt.” (citations omitted)). And thus after Erlinger, the same conclusion applies to § 922(g)(1) and § 924(e)(1)—that is, they are lesser and greater offenses.

* * *

Altogether then, three principles relate to Thomas's conviction. First, jeopardy attached as soon as the jury was empaneled for his § 922(g)(1) trial. Second, that jeopardy was offense specific, attaching to the § 922(g)(1) offense he was indicted for. Third, his indictment, trial, and jury findings all concerned the lesser included § 922(g)(1) offense rather than the enhanced § 924(e)(1) greater offense.

3.

Reasoning from these conclusions, Thomas claims he cannot be subject to an enhanced-ACCA penalty without having pled to it or having been indicted for it. For support, he points only to United States v. Bell, 37 F.4th 1190 (6th Cir. 2022).

Thomas roots his argument in the Double Jeopardy Clause's prohibition on successive prosecutions. And on his view, Bell holds that the Double Jeopardy Clause precludes a defendant from being tried for a greater offense once he has pleaded guilty to a lesser-included offense. So to allow Thomas to be sentenced for the “unindicted, greater offense in these circumstances perpetuates the same kind of Double Jeopardy harm.” R.25, Appellant Supp. Br., p.7. But Bell contemplated a unique factual circumstance arising out of a Federal Rule of Criminal Procedure 11(c)(1)(C) plea agreement and does not sweep as broadly as Thomas suggests.

Take Bell’s facts. There, a grand jury returned a two-count indictment: Count 1 alleged distribution of controlled substances with death resulting (in violation of 21 U.S.C. § 841(b)(1)(C)) and Count 2 charged possession with intent to distribute. Bell, 37 F.4th at 1194. He pleaded guilty via a 11(c)(1)(C)-plea to a “lesser included offense in Count 1” (a distribution charge) even though it was not described in the indictment. Id. And under the proposed plea agreement, the parties had agreed to a 168-month sentence. Id. The judge accepted Bell's plea but rejected the plea agreement; there was a “vast discrepancy” between the agreed-to 168 months and the advisory-Guidelines range (six to twelve months). Id. The government objected and asked the judge to set a trial date on the greater, death-resulting offense. Overruling this request, the judge proceeded to sentencing and imposed a thirty-month sentence. Id.

The government appealed and asked the Sixth Circuit to vacate the conviction on the lesser-included offense and remand for proceedings on the greater offense. Id. For support, it relied on a catch-all provision in the agreement. It explained that if the district judge rejected the agreement, the plea would be void, neither party would be bound to it, and the government could then “pursue any and all charges supported by the facts and law and, upon conviction, pursue imposition of any sentence supported by the facts and law.” Id. at 1197.

The panel reasoned that the plea did not offer the government any cover. The provision did not indicate that the government's consent to the plea on the lesser offense was “contingent” on the judge accepting the agreement. Id. Rather, if charges had remained after the plea, or the defendant had withdrawn his plea, this provision gave the government the ability to try those charges that remained. Id. Because the defendant had pleaded guilty to the lesser offense—which the judge had accepted—there was nothing left to do but impose a sentence based on that plea. Id. at 1198. Under those circumstances, we affirmed because any “subsequent reindictment for the greater included offense implicates double jeopardy concerns under the Fifth Amendment.” Id.

In a vacuum, Bell might be read to suggest that the government is stuck with the sentence for the lesser-included § 922(g)(1) conviction in Thomas's case. But the double-jeopardy conclusion in Bell was tied to the government's argument that Ohio v. Johnson allowed the second prosecution. 467 U.S. 493, 502, 104 S.Ct. 2536, 81 L.Ed.2d 425 (1984). There, the Supreme Court said the Double Jeopardy Clause did not prohibit the state from prosecuting a defendant for murder and aggravated robbery even though the defendant had pleaded guilty to their lesser included offenses—involuntary manslaughter and grand theft, respectively. That meant Johnson still had substantive charges left for the government to prosecute after he had pleaded guilty—only “part of the charges against him” had been resolved. Id. at 501, 104 S.Ct. 2536. By contrast, Bell sat in a different place factually—his guilty plea fully resolved the charges against him. And since the government had not made its consent to the plea on the lesser-included offense contingent on the court's acceptance of the agreement, and given the unique circumstances that arise when the government offers a (c)(1)(C)-plea, we affirmed. Because Bell concerned the unique factual circumstance when the district judge accepts the defendant's guilty plea but rejects the (c)(1)(C)- agreement, its conclusions are not applicable here.

All in all, in the years following Apprendi and Alleyne, appellate courts have consistently applied harmless-error review to these types of Sixth Amendment injuries. We are skeptical that after more than 50 years of applying Chapman’s harmless-error inquiry, that all this time we have been encroaching on defendants double-jeopardy rights. 2

Because we hold that Erlinger errors are subject to harmlessness review, and because Thomas's arguments about double jeopardy do not undermine that conclusion, we AFFIRM his enhanced sentence based on his three prior violent felonies.

CONCURRENCE

A jury convicted Deangelus Thomas of being a felon in possession of a firearm in violation of 18 U.S.C. § 922(g)(1). Deciding—without a jury determination—that Thomas's three previous convictions were for violent felonies committed on different occasions, the district court sentenced him pursuant to the Armed Career Criminal Act (ACCA), 18 U.S.C. § 924(e), a designation which substantially increased his statutory penalty and guideline range.

The Supreme Court has since made clear that the district court erred in failing to submit the ACCA different-occasions inquiry to a jury, Erlinger v. United States, 602 U.S. 821, 835, 144 S.Ct. 1840, 219 L.Ed.2d 451 (2024), and our decision in United States v. Campbell, 122 F.4th 624, 630–31 (6th Cir. 2024) instructs us to conduct harmless error review. Subsequent panels have relied upon Campbell to do so. See, e.g., United States v. Cogdill, 130 F.4th 523, 527 (6th Cir. 2025); United States v. Robinson, 133 F.4th 712, 723–725 (6th Cir. 2025); United States v. Kimbrough, 138 F.4th 473, 475 (6th Cir. 2025). Nonetheless, members of this court have questioned whether Campbell “contravenes the Supreme Court's holding in Erlinger.” See Cogdill, 130 F.4th at 535 (Clay, J., dissenting). I share that concern. But because we are bound by Campbell, I concur with my colleagues that the error was harmless based upon this record. I write separately to caution against overextending Campbell so far that it encroaches upon the Supreme Court's holding in Erlinger.

I.

The Fifth and Sixth Amendments require that a “unanimous jury” determine “beyond a reasonable doubt” that a defendant's past offenses were committed on separate occasions under the ACCA. Erlinger, 602 U.S. at 836, 144 S.Ct. 1840. That requirement stems from the well-established principle that “[o]nly a jury may find ‘facts that increase the prescribed range of penalties to which a criminal defendant is exposed.’ ” Id. at 833, 144 S.Ct. 1840 (quoting Apprendi v. New Jersey, 530 U.S. 466, 490, 120 S.Ct. 2348, 147 L.Ed.2d 435 (2000)). The ACCA indisputably imposes such increased penalties.

To qualify for an ACCA enhancement, a jury must determine that a defendant has “three previous convictions” for a “violent felony or a serious drug offense, or both, committed on occasions different from one another.” 18 U.S.C. § 924(e)(1); Erlinger, 602 U.S. at 835, 144 S.Ct. 1840. Concluding that offenses occurred on different occasions requires examining a range of factors, including timing, proximity, and the character and relationship of the offenses. See Cogdill, 130 F.4th at 528 (citing Wooden v. United States, 595 U.S. 360, 369, 142 S.Ct. 1063, 212 L.Ed.2d 187 (2022)). When the district court fails to submit this inquiry to a jury, the government must prove “beyond a reasonable doubt—through ‘relevant and reliable information’ in the record—that, absent the error, any reasonable jury would have found that [the defendant] committed the prior offenses on different occasions.” Id. (quoting Campbell, 122 F.4th at 630–33).

Here, the district court relied on facts contained in the Presentence Report (PSR) to conclude that Thomas was convicted of the necessary three predicate offenses. Thomas challenges the district court's use of the PSR and other “Shepard documents”—documents that include judicial records, plea agreements, and colloquies between a judge and a defendant—which he argues are prone to error and therefore unreliable. See Shepard v. United States, 544 U.S. 13, 26, 125 S.Ct. 1254, 161 L.Ed.2d 205 (2005).

Relying on Campbell, the majority opinion readily dismisses Thomas's argument. Indeed, Campbell sets out that our “harmless error review is based on an assessment of all ‘relevant and reliable information’ in the ‘entire record,’ which in turn is gauged through ‘case-by-case adjudication rather than through a categorical bar against’ certain categories of evidence.” 122 F.4th at 632–33 (quoting Greer v. United States, 593 U.S. 503, 510–11, 141 S.Ct. 2090, 210 L.Ed.2d 121 (2021)). The “entire record” includes information contained in Shepard documents like the PSR. Id. at 633. We cannot, however, discount Erlinger’s guidance when reviewing for harmless error. Rather, Erlinger must inform our view of Shepard documents and the extent of their permissible use.

A.

Erlinger makes the Supreme Court's guidance with respect to the use of Shepard documents for the different-occasions inquiry plain: courts may use Shepard documents in a limited manner and, when they do, they must exercise caution. Erlinger, 602 U.S. at 839–42, 144 S.Ct. 1840. First, with respect to their “limited function,” the Court explains that “[t]o ensure compliance with the Fifth and Sixth Amendments,” judges may use Shepard documents to determine “the fact of a prior conviction and the then-existing elements of that offense.” Id. at 839, 144 S.Ct. 1840 (citation omitted). “No more is allowed.” Id. (citation omitted). Judges therefore “may not use information in Shepard documents to decide what the defendant actually did, or the means or manner in which he committed his offense[.]” Id. at 839–40, 144 S.Ct. 1840. (cleaned up). “To sanction that practice would be to allow a sentencing court to do exactly what the Fifth and Sixth Amendments forbid.” Id. at 840, 144 S.Ct. 1840.

Second, even when using Shepard documents for this limited purpose, courts must exercise caution. Erlinger expands on the various reasons a court should not wholly rely upon a PSR and other judicial records:

Not only are Shepard documents of limited utility, they can be prone to error. The risk of error may be especially grave when it comes to facts recounted in Shepard documents on which adversarial testing was unnecessary in the prior proceeding. As we have recognized, at trial, and still more at plea hearings, a defendant may have no incentive to contest what does not matter to his conviction at the time ․

As a matter of fair notice alone, old recorded details, prone to error, sometimes untested, often inessential, and the consequences of which a defendant may not have appreciated at the time, should not come back to haunt him many years down the road by triggering a lengthy mandatory sentence.

Id. at 841–42, 144 S.Ct. 1840 (cleaned up). A defendant's failure to challenge the documents’ accuracy below does not per se make them reliable either:

At the time of his prior conviction, a defendant might not have cared if a judicial record contained a mistake about, say, the time or location of his offense. Back then, fine details like those might not have mattered a bit to his guilt or innocence. Contesting them needlessly, too, might have risked squandering the patience and good will of jury or the judge responsible for pronouncing a sentence. Yet, years later and faced with an ACCA charge, those kinds of details can carry with them life-altering consequences.

Id. at 841, 144 S.Ct. 1840.

The majority acknowledges that “Erlinger has some harsh words about Shepard documents.” (Maj. Op. at ––––.) That characterization understates the Supreme Court's skepticism of the use of Shepard documents to conduct the different occasions analysis. While our binding precedent holds that “Erlinger did not preclude the use of Shepard documents in reviewing an error for harmlessness,” Campbell, 122 F.4th at 632, the Supreme Court's concerns about their utility remain relevant even when we are reviewing for harmlessness.

True, Erlinger discusses these concerns in the context of sentencing judges, not appellate courts’ review for harmless error. But it is unlikely that the Supreme Court views appellate judges as less immune than trial judges to the risks presented by Shepard documents when conducting the wide-ranging factual inquiry required to establish beyond a reasonable doubt that a defendant committed his offenses on different occasions. Indeed, precedent counsels the opposite. See, e.g., Buford v. United States, 532 U.S. 59, 64–66, 121 S.Ct. 1276, 149 L.Ed.2d 197 (2001) (observing that district courts are better positioned to conduct fact-intensive inquiries); United States v. Clark, 24 F.4th 565, 578 (6th Cir. 2022) (“The district court is better suited to develop [ ] fact-intensive questions.”); United States v. Shanklin, 924 F.3d 905, 919 (6th Cir. 2019) (observing that an inquiry that is “necessarily fact-specific” is “better examined by the district court”) (citation omitted).

To proceed with harmless error review without accounting for Erlinger’s cautions risks reproducing the same infringements on a defendant's constitutional rights the Supreme Court sought to guard against. Even if our post-Erlinger precedent requires us to conduct harmless error review, we should not sanction wholesale reliance on Shepard documents. Although Shepard documents may form part of the record we can examine when reviewing for harmlessness, our examination of them must be colored by Erlinger. And where our use or the district court's use of Shepard documents must exceed Erlinger’s boundaries to achieve the multi-factored analysis required for the ACCA's different occasions inquiry, the error cannot be harmless. See Erlinger, 602 U.S. at 840, 144 S.Ct. 1840 (“Often, [ ] Shepard documents will not contain all the information needed to conduct a sensible ACCA occasions inquiry[.]”).

B.

This case presents that rare situation where our necessarily limited and skeptical view of Shepard documents still permits us to conclude, beyond a reasonable doubt, that a defendant committed his offenses on different occasions. 1

Here, the district court did not have to go beyond the fact of Thomas's convictions—the dates and the offense elements—to determine that they constituted qualifying offenses. See Erlinger, 602 U.S. at 839, 144 S.Ct. 1840 (citation omitted). Thomas qualified for the ACCA enhancement based on the following three predicate offenses: an aggravated robbery conviction from October 2000; an aggravated assault conviction from December 2007; and an attempted second-degree murder conviction from August 2010. 2 Each offense is separated by at least three years, with nearly ten years separating the first from the third. And each conviction involves different offense conduct, against different victims. And while “no particular lapse of time or distance between offenses automatically separates a single occasion from distinct ones,” id. at 841, 144 S.Ct. 1840, the different nature of the offenses combined with their temporal distance make it clear that “any reasonable jury would have found that [Thomas] committed the prior offenses on different occasions.” See Cogdill, 130 F.4th at 528 (quoting Campbell, 122 F.4th at 630–33).

Additionally, the majority indicates that Thomas's decision not to dispute the PSR's accuracy gives us “no reason to doubt” the reliability of the PSR's contents. (Maj. Op. at ––––.) The fact that Thomas did not do so on this appeal may support our reliance on its information here. To be clear, however, the fact that Thomas chose not to challenge the contents of the PSR during proceedings below would not support our reliance on it, for the reasons Erlinger outlines. See 602 U.S. at 839–41, 144 S.Ct. 1840.

II.

Finally, I concur with my colleagues that no Double Jeopardy problem is presented here. But that does not foreclose the potential for a Double Jeopardy problem arising if the district court's error had been harmful. As the majority opinion explains, Erlinger makes § 922(g)(1) and § 924(e)(1) lesser and greater offenses, respectively. If remanding for resentencing was required, it is not clear why empaneling a second jury to make the necessary findings to support the § 924(e)(1) enhancement would not present a Double Jeopardy problem. See United States v. Pena, 742 F.3d 508, 518–519 (1st Cir. 2014). Thomas's final § 922(g)(1) conviction would have completely resolved the charges pending against him: here, being a felon in possession of a firearm. Compare Ohio v. Johnson, 467 U.S. 493, 502, 104 S.Ct. 2536, 81 L.Ed.2d 425 (1984), with United States v. Bell, 37 F.4th 1190, 1198 (6th Cir. 2022). And jeopardy attached for that offense. As such, the prospect of empaneling a second jury to support the government's continued pursuit of an unindicted § 924(e)(1) enhancement would likely “raise[ ] a thicket of potential and thorny double jeopardy issues, into which it is wiser not to enter.” Pena, 742 F.3d at 518. But since remand is not required, we stop short of that thicket here.

CONCURRENCE

This is yet another case that challenges the application of harmless-error review to Erlinger errors under the Double Jeopardy Clause. United States v. Kimbrough, 138 F.4th 473, 477–79 (6th Cir. 2025); United States v. Cogdill, 130 F.4th 523, 532 (6th Cir. 2025); United States v. Sain, No. 22-6131, 2025 WL 801366, at *2 (6th Cir. Mar. 13, 2025). Defendants in these cases claim that if the jury didn't find the facts supporting their ACCA enhancement, then they can't be punished as if the jury had because this would constitute a successive prosecution. This theory raises an interesting question about the relationship between the Double Jeopardy Clause, Chapman v. California’s harmless-error rule, and Griffith v. Kentucky’s retroactive-application-of-new-rules principle. 386 U.S. 18, 87 S.Ct. 824, 17 L.Ed.2d 705 (1967); 479 U.S. 314, 107 S.Ct. 708, 93 L.Ed.2d 649 (1987).

Though the majority doesn't resolve the question, I write separately to explain my skepticism that, for cases pending on direct appeal, the Double Jeopardy Clause is a bar to imposing an enhanced ACCA sentence on remand after a harmful Erlinger error. As I see it, the core double-jeopardy protection against successive prosecutions is not implicated when the “error” results from a post hoc change in the law. The government is free to prove the occasions-different element to a jury on remand.

I.

The Double Jeopardy Clause's language is straightforward, but it has proven difficult to apply in some cases. Still, its motivating principle—as reflected in its text—guides its application through most complex issues, including this one. The Double Jeopardy Clause shields against prosecutorial overreach by preventing the government from getting two bites at the proverbial apple. Garrett v. United States, 471 U.S. 773, 795, 105 S.Ct. 2407, 85 L.Ed.2d 764 (1985) (O'Connor, J., concurring) (collecting cases); see also Ex Parte Lange, 85 U.S. 163, 169–70, 18 Wall. 163, 21 L.Ed. 872 (1873) (discussing the common law origins of the Double Jeopardy Clause's prohibition on successive prosecutions); id. (“[T]he law abhors a multiplicity of suits, it is yet more watchful in criminal cases that the crown shall not oppress the subject, or the government the citizen, by unreasonable prosecutions.”). The Court has reiterated that the bar on successive prosecutions ensures that the State,

with all its resources and power[,] should not be allowed to make repeated attempts to convict an individual for an alleged offense, thereby subjecting him to embarrassment, expense and ordeal and compelling him to live in a continuing state of anxiety and insecurity, as well as enhancing the possibility that even though innocent he may be found guilty.

Green v. United States, 355 U.S. 184, 187–88, 78 S.Ct. 221, 2 L.Ed.2d 199 (1957); see also Kepner v. United States, 195 U.S. 100, 120–21, 24 S.Ct. 797, 49 L.Ed. 114 (1904) (similar).

So the core protection of the Double Jeopardy Clause ensures that the government does not get a second chance to “supply evidence which it failed to muster in the first proceeding.” Tibbs v. Florida, 457 U.S. 31, 41, 102 S.Ct. 2211, 72 L.Ed.2d 652 (1982) (quoting Burks v. United States, 437 U.S. 1, 11, 98 S.Ct. 2141, 57 L.Ed.2d 1 (1978)). But that leaves unanswered whether the Double Jeopardy Clause forbids a retrial after a defendant successfully challenges his conviction on appeal. To address the Clause's application in specific cases, the Court has considered both the defendant's interest in the finality of his judgment and the government's interest in vindicating the social good of prosecuting those who violate the law. Id. at 40, 102 S.Ct. 2211; United States v. Tateo, 377 U.S. 463, 466, 84 S.Ct. 1587, 12 L.Ed.2d 448 (1964).

Based on these interests, as a general rule, if a defendant successfully appeals his conviction, the government may retry him “for the same offense of which he had been convicted.” Tibbs, 457 U.S. at 39–40, 102 S.Ct. 2211 (quoting Ball v. United States, 163 U.S. 662, 671–72, 16 S.Ct. 1192, 41 L.Ed. 300 (1896)). This is for two reasons. First, “society would pay too high a price ‘were every accused granted immunity from punishment because of any defect sufficient to constitute reversible error in the proceedings leading to conviction.’ ” Id. at 40, 102 S.Ct. 2211 (quoting Tateo, 377 U.S. at 466, 84 S.Ct. 1587). And second, retrial in these circumstances doesn't implicate the “governmental oppression targeted by the Double Jeopardy Clause.” Id. Most errors that a defendant appeals don't implicate whether the government failed to convince the jury of his guilt, and instead highlight a problem with the judicial process. So these errors “impl[y] nothing with respect to the guilt or innocence of the defendant.” Burks, 437 U.S. at 15, 98 S.Ct. 2141. In effect, the successful defendant “nullifies” his original conviction so that he can challenge the charges against him in a proceeding that satisfies the Constitution and the rules of procedure. Chaffin v. Stynchcombe, 412 U.S. 17, 24, 93 S.Ct. 1977, 36 L.Ed.2d 714 (1973) (quoting North Carolina v. Pearce, 395 U.S. 711, 721, 89 S.Ct. 2072, 23 L.Ed.2d 656 (1969)).

Originally, this general rule permitting retrial applied in every case, even when a reviewing court held that insufficient evidence supported a jury's verdict. Bryan v. United States, 338 U.S. 552, 560, 70 S.Ct. 317, 94 L.Ed. 335 (1950), overruled by Burks, 437 U.S. at 18, 98 S.Ct. 2141. That is, for double-jeopardy purposes, the Court viewed a reversal for insufficient evidence as no different than more traditional trial errors like the erroneous admission of evidence or problematic jury instructions.

But the Court later reversed course, holding that a reviewing court's finding of insufficient evidence is a “narrow exception” to the general rule that a defendant can be retried when he prevails on appeal. Tibbs, 457 U.S. at 40, 102 S.Ct. 2211; Lockhart v. Nelson, 488 U.S. 33, 39, 109 S.Ct. 285, 102 L.Ed.2d 265 (1988) (same). This makes sense because when a reviewing court sets aside a defendant's conviction for insufficient evidence, this is “in effect a determination that the government's case against the defendant was so lacking” that it never should have been submitted to the jury. Lockhart, 488 U.S. at 39, 109 S.Ct. 285. In other words, a sufficiency reversal goes directly to guilt or innocence. So whether it's by a jury acquittal, a trial court's grant of a motion to acquit, or an appellate court's insufficiency reversal, see Fed. R. Crim. P. 29, the defendant has been acquitted of guilt because the government has failed to prove its case. Importantly, this exception captures the intuitive principle—and the “core” of the Clause's protection—that the government does not get a second chance to “supply evidence which it failed to muster in the first proceeding.” Tibbs, 457 U.S. at 41, 102 S.Ct. 2211 (quoting Burks, 437 U.S. at 11, 98 S.Ct. 2141).

But subject only to Burks’s narrow exception, the general rule is straightforward: a defendant who succeeds in proving on appeal that his judicial process was deficient because of a harmful error can be retried on remand.

II.

With this background, the question here is whether, after Erlinger, the jury's failure to find the occasions-different element is really a failure to muster sufficient evidence, or just a (potentially harmful) trial error. If the latter, then retrial poses no double-jeopardy concern. But if it's really an insufficiency problem, then retrial is prohibited even in cases pending on direct review.

I think Burks’s logic shows that the Double Jeopardy Clause doesn't thwart retrial here. The relevant error is the jury's failure to find (what is now) an element of the crime—as required by the Supreme Court's Apprendi line of cases. Alleyne v. United States, 570 U.S. 99, 114, 133 S.Ct. 2151, 186 L.Ed.2d 314 (2013). And to be fair, Apprendi errors manifest elsewhere in the criminal process: the element is omitted from the indictment, so the government fails to prove it, and the jury instructions don't address it. But all these errors occurred only because the law changed while the case was pending on appeal. In other words, these are “errors” that only became errors after the fact—at trial, the parties had no reason to know they should treat the occasions-different fact as an element to be proven by the government and found by a jury.

Strong support for this outcome comes from Lockhart, 488 U.S. at 38, 109 S.Ct. 285. There, the state appellate courts had rejected the defendant's argument that a piece of evidence to support his sentencing enhancement was erroneously admitted. Id. at 34–37, 109 S.Ct. 285. On habeas review, the district court found that the defendant was right. The district court also found that the remaining evidence couldn't sustain the government's burden of proof. And the court said that, in these circumstances, Burks prohibited a retrial. Id. at 37, 109 S.Ct. 285. The Eighth Circuit agreed.

But the Court reversed, finding the Burks exception inapplicable. The relevant error for the double-jeopardy question was the erroneous admission of a particular piece of evidence, not the second-order problem that the remaining evidence was insufficient. Indeed, with the disputed evidence, there was sufficient evidence—and the error in admitting the evidence was about its “legal effect” not its “probative value.” Id. at 40, 109 S.Ct. 285. Viewed this way, there was no sufficiency problem. In fact, if the defendant had successfully excluded the relevant evidence at trial, the prosecution could have sought to rework its case or offer alternative evidence. If so, then reversing for a new trial simply placed the parties in the position they would have been in if the trial court had properly excluded the evidence in the first place. Id. at 42, 109 S.Ct. 285.

Similarly, an Erlinger error is not a first-order sufficiency problem. As an outgrowth of Apprendi, the substance of the error is that the jury didn't resolve an element it should have. And like in Lockhart, this omission doesn't concern the government's failure in proof, it highlights a “judicial process” that was “defective.” Id. at 40, 109 S.Ct. 285 (quoting Burks, 437 U.S. at 15, 98 S.Ct. 2141). So also like Lockhart, no “acquittal” has occurred—by a jury or reviewing court—because there has been no “resolution” of “some or all of the factual elements” of the occasions-different element. Burks, 437 U.S. at 10, 98 S.Ct. 2141 (internal quotation marks omitted).

Yes, we apply Chapman’s harmless-error rule to Erlinger errors. And that analysis resembles a sufficiency inquiry. That is, assuming the defendant has not objected to the facts of the presentence report and there is no evidence of its unreliability, the court of appeals looks at the entire record to determine whether the Erlinger error would have changed the outcome. 1 Presumably that's what happened in Lockhart as well—everyone agreed that once the erroneous evidence was excluded, the error wasn't harmless. But that still didn't transform Lockhart into a Burks-like sufficiency case. Instead, the case was remanded for retrial, subjecting the error to the general rule that defendants can be retried after a successful appeal. Lockhart, 488 U.S. at 42, 109 S.Ct. 285.

Furthermore, for these cases pending on direct review, it's not that the government failed to offer evidence of an element that it knew it had an obligation to prove. This is another way Erlinger errors mirror Lockhart’s error. There, the trial court didn't exclude the evidence so the government didn't know that it would have to present more evidence—in that case proof of a different prior felony. Id. at 37, 119 S.Ct. 1827. So by remanding for retrial, the court placed the parties in the positions that they would've been in at the moment of proper exclusion. Id. at 42, 109 S.Ct. 285. Here, were we to remand for a retrial, the parties would be on notice of the now-prevailing standard, so the parties would know both that the government must offer proof of at least three prior violent felonies, and that a jury must be the one to find those facts.

This conclusion is similarly supported by cases from other circuits. Our sister circuits agree that when an “error” occurs because of a change in the law while the case is pending on direct appeal, the Double Jeopardy Clause does not pose a bar to retrial. They explain:

Remanding for retrial ․ does not give the government the opportunity to supply evidence it “failed” to muster at the first trial within the meaning of Tibbs. The government had no reason to introduce such evidence because, at the time of trial, under the law of our circuit, the government was not required to prove that [element]․ The district court at trial could have required such proof only by disregarding clear rulings by this court. The government therefore is not being given a second opportunity to prove what it should have proved earlier, and double jeopardy protections do not bar retrial.

United States v. Weems, 49 F.3d 528, 530 (9th Cir. 1995) (remand for retrial was appropriate after proof at trial was rendered insufficient because of intervening Supreme Court decision in Ratzlaf v. United States, 510 U.S. 135, 114 S.Ct. 655, 126 L.Ed.2d 615 (1994)).

In essence, the government never got its first bite at the apple to prove that element. So “a defendant cannot make out a sufficiency challenge as to offense elements that the government had no requirement to prove at trial under then-prevailing law.” United States v. Reynoso, 38 F.4th 1083, 1091 (D.C. Cir. 2022) (analyzing Double Jeopardy Clause implication on error that arose because of intervening Supreme Court decision in Rehaif v. United States, 588 U.S. 225, 139 S.Ct. 2191, 204 L.Ed.2d 594 (2019)); United States v. Davies, 942 F.3d 871, 874 (8th Cir. 2019) (remand for retrial did not offend Double Jeopardy Clause after harmful Rehaif error); United States v. Delgado-Marrero, 744 F.3d 167, 192–93 (1st Cir. 2014) (“Because the Alleyne error here was an instructional error, the Double Jeopardy Clause does not prohibit retrial.”).²

Finally, I'll note that in the context of this case, the government likely could not have offered occasions-different evidence at trial even if it had wanted to. Before Erlinger, in our circuit at least, if a defendant raised a Rule 29 motion at the close of the government's case claiming it had failed to prove the occasions-different element, the district court would have erred by granting the motion. Weems, 49 F.3d at 530. Under pre-Erlinger law, the government had no need to prove the prior convictions beyond a reasonable doubt, assuming it could even admit evidence to prove them.³ United States v. Cook, No. 22-5056, 2022 WL 4684595, at *2 (6th Cir. Oct. 3, 2022) (noting that though the government believed Wooden required the different occasions be found by a jury, binding circuit precedent foreclosed reversal on those grounds). So to say that the government failed to muster sufficient evidence of the prior occasions highlights why Rule 29—and by extension Burks’s insufficient-evidence-bars-retrial exception—is a poor fit for the type of argument Thomas raises here.

At bottom, retrial after an Erlinger error “is not the sort of governmental oppression at which the Double Jeopardy Clause is aimed”—the government never had the chance to prove the Erlinger fact at trial in the first place. See Lockhart, 488 U.S. at 42, 109 S.Ct. 285; see also Green, 355 U.S. at 187, 78 S.Ct. 221 (“[T]he State ․ should not be allowed to make repeated attempts to convict an individual for an alleged offense,” since “[t]he constitutional prohibition against ‘double jeopardy’ was designed to protect an individual from being subjected to the hazards of trial and possible conviction more than once for an alleged offense.”); Burks, 437 U.S. at 11, 98 S.Ct. 2141 (“The Double Jeopardy Clause forbids a second trial for the purpose of affording the prosecution another opportunity to supply evidence which it failed to muster in the first proceeding.”); Garrett, 471 U.S. at 789–90, 105 S.Ct. 2407 (“We do not think that the Double Jeopardy Clause may be employed to force the Government's hand in this manner․ One who insists that the music stop and the piper be paid at a particular point must at least have stopped dancing himself before he may seek such an accounting.”).

So though I recognize that some Erlinger errors are harmful, that just means the defendant was subject to an erroneous judicial process that must be corrected. Lockhart, 488 U.S. at 40, 109 S.Ct. 285; Burks, 437 U.S. at 15, 98 S.Ct. 2141. And on remand, the defendant can exercise his rights under the Due Process Clause and put the government to its burden to prove beyond a reasonable doubt the occasions-different question to the proper fact-finder—a jury of his peers. To hold otherwise would extend the Double Jeopardy Clause beyond its essential protection to shield defendants not only from a second shot, but also from the government's first shot under the now-prevailing standard.

Thus cases pending on direct review are differently situated from future cases the government will bring. If the government files a new indictment tomorrow and fails to include the different occasions, or prove them beyond a reasonable doubt at trial, all the double-jeopardy concerns discussed by the majority about greater-and lesser-included offenses would be directly at play. The essential difference is that the government is now on notice of Erlinger’s requirements. So if in the future it fails to allege the different occasions in the indictment, this omission could invite the possibility of a constructive amendment. And if it fails to prove the facts beyond a reasonable doubt, this will implicate the government's first bite at the apple, so it will be barred from a second bite under the Double Jeopardy Clause. Cf. Burks, 437 U.S. at 11 n.6, 98 S.Ct. 2141 (“[W]here the Double Jeopardy Clause is applicable, its sweep is absolute.”). But because cases like Thomas's were pending on direct review when Erlinger was decided, a retrial about the occasions-different element does not offend the Double Jeopardy Clause.

FOOTNOTES

1.   This Circuit's recent cases in Cogdill, 130 F.4th 523 and United States v. Kimbrough, 138 F.4th 473 (6th Cir. 2025) don't change our conclusion here. Those cases found Erlinger errors harmful based on dramatically different facts and ambiguities in the factual record that made it impossible to discern, under harmless error, whether the government satisfied its beyond-a-reasonable-doubt burden. Cogdill, 130 F.4th at 529–31; Kimbrough, 138 F.4th at 477–79. Both contemplate a defendant's prior felonies that occurred days apart, and a PSR that omitted key factual details, like location. Cogdill, 130 F.4th at 529–31; Kimbrough, 138 F.4th at 477–79. By contrast, they both distinguish Campbell because it applies harmless error in a factual context which we think governs here. Cogdill, 130 F.4th at 531; Kimbrough, 138 F.4th at 478. Given the many years separating Thomas's prior convictions, the many different victims, and the many locations where he committed these violent felonies, these circumstances show no similar scheme or plan akin to the continuous occasions described in Wooden. See Campbell, 122 F.4th at 631–32.

2.   Thomas invokes several out-of-circuit cases for the proposition that the proper remedy here is to “remand for resentencing without the ACCA enhancement.” R. 25, Appellant Supp. Br., p.18 (collecting cases). Thomas argues that this is the only way to avoid the second-order-double-jeopardy injury that arises after the Sixth Amendment error. And the government, for its part, offers no view on how double jeopardy might bear on our harmless-error analysis.But Thomas's argument assumes that there's reason to remand. It's true that, for example, in United States v. Pena, 742 F.3d 508 (1st Cir. 2014) the panel found an Alleyne error harmful and limited the scope of the remand to the unenhanced offense in the guilty plea. But essential to that conclusion was the government's concession that the error was harmful. And for that reason, Pena does not change the outcome here. See also Alleyne, 570 U.S. at 117–18, 133 S.Ct. 2151 (remanding for resentencing consistent with jury verdict because error was harmful).Thomas's limited double-jeopardy argument is based solely on his reading of Bell, which we reject for the reasons above. We leave to a future panel to address whether applying harmless error itself violates the Double Jeopardy Clause on some other theory. Cf. Cogdill, 130 F.4th at 532 (leaving for the district court to determine in the first instance whether double jeopardy prohibited government from retrying a harmful Erlinger error).

1.   Citing Wooden, the majority notes that “in most cases,” like here, the different-occasions inquiry will be straightforward. (Maj. Op. at ––––.) But even a seemingly straightforward inquiry does not require a finding of harmlessness, or even make it likely. Erlinger, which followed Wooden, counsels otherwise:We do not question [the] assessment that in many cases the occasions inquiry will be straightforward․ Often, a defendant's past offenses will be different enough and separated by enough time and space that there is little question he committed them on separate occasions. But none of that means a judge rather than a jury should make the call. There is no efficiency exception to the Fifth and Sixth Amendments. In a free society respectful of the individual, a criminal defendant enjoys the right to hold the government to the burden of proving its case beyond a reasonable doubt to a unanimous jury of his peers regardless of how overwhelming the evidence may seem to a judge.Erlinger, 602 U.S. at 842, 144 S.Ct. 1840 (cleaned up).

2.   The majority identifies a “unifying theme” of these offenses being “their inherent violence and Thomas's proclivity for violence with a firearm.” (Maj. Op. at ––––.) But no unifying theme is required. To the contrary, the focus of our inquiry is that each offense be three separate violent felony or drug offenses, not united in common scheme or purpose. See Cogdill, 130 F.4th at 528.

1.   I understand, as Judge Cole notes in his concurrence, that harmless-error review can sometimes be in tension with the Sixth Amendment injury itself: if the Sixth Amendment is designed to protect a defendant's right to have a jury of his peers resolve the facts of his case, how is three judges resolving the case a permissible remedy? Neder v. United States, 527 U.S. 1, 30, 119 S.Ct. 1827, 144 L.Ed.2d 35 (1999) (Scalia, J., concurring in part and dissenting in judgment) (“I believe that depriving a criminal defendant of the right to have the jury determine his guilt of the crime charged—which necessarily means his commission of every element of the crime charged—can never be harmless.”). But as the lead opinion notes, given our repeated application of harmless error to Apprendi-style errors, until we receive direction from the Supreme Court otherwise, we are bound to continue this path. Op. at –––– – ––––; see also United States v. Campbell, 122 F.4th 624, 631 (6th Cir. 2024).

2.   Our sister circuits, as in Weems, regularly apply these Burks-Tibbs-Lockhart principles in non-Apprendi cases to justify a remand for retrial when harmful errors occur because of a change in the law while a case is pending on direct appeal. Each time, the changed law impacted the government's case in some way—like making jury instructions unconstitutional or rendering the government's proof of guilt insufficient under the now-prevailing standard. And each time, the circuit court found remand for retrial did not offend the Double Jeopardy Clause. See e.g., United States v. Wacker, 72 F.3d 1453, 1464–65 (10th Cir. 1995) (proof at trial rendered insufficient because of intervening Supreme Court decision in Bailey v. United States, 516 U.S. 137, 116 S.Ct. 501, 133 L.Ed.2d 472 (1995), superseded by statute, Bailey Fix Act, Pub. L. No. 105-386, 112 Stat. 3469, as recognized in United States v. O'Brien, 560 U.S. 218, 233, 130 S.Ct. 2169, 176 L.Ed.2d 979 (2010)); United States v. Gonzalez, 93 F.3d 311, 323 (7th Cir. 1996) (proof at trial rendered insufficient because of intervening Supreme Court decision in Bailey, 516 U.S. 137, 116 S.Ct. 501); United States v. Ellyson, 326 F.3d 522, 533–34 (4th Cir. 2003) (jury instruction rendered unconstitutional because of intervening Supreme Court decision in Ashcroft v. Free Speech Coalition, 535 U.S. 234, 122 S.Ct. 1389, 152 L.Ed.2d 403 (2002)); United States v. Recio, 371 F.3d 1093, 1106–07 (9th Cir. 2004) (proof at trial rendered insufficient because of intervening Supreme Court decision in Ratzlaf, 510 U.S. 135, 114 S.Ct. 655); United States v. Ford, 703 F.3d 708, 710–11 (4th Cir. 2013) (proof at trial rendered insufficient because of intervening change in circuit law); United States v. Harrington, 997 F.3d 812, 815 (8th Cir. 2021) (proof at trial rendered insufficient because of intervening Supreme Court decision in Burrage v. United States, 571 U.S. 204, 134 S.Ct. 881, 187 L.Ed.2d 715 (2014)); United States v. Aiello, 118 F.4th 291, 300 (2d Cir. 2024) (proof at trial rendered insufficient because of intervening Supreme Court decision in Ciminelli v. United States, 598 U.S. 306, 143 S.Ct. 1121, 215 L.Ed.2d 294 (2023)).

3.   As well, before Erlinger, a defendant would have had strong grounds to object to the admission of his prior convictions. See Cook, 2022 WL 4684595, at *2 (affirming that Wooden didn't disrupt prior circuit precedent and three occasions could be found by judge without violating Apprendi). Without proof that the defendant's prior violent felonies had a permissible use, see Fed. R. Evid. 404(b)(2), that evidence would have been extremely prejudicial to the defendant and likely would have been excluded, see Fed. R. Evid. 404(b)(1). That's especially true if the prior violent felonies share traits with the present substantive offense. See Fed. R. Evid. 403.

NALBANDIAN, Circuit Judge.

NALBANDIAN, J., delivered the opinion of the court, in which COLE and McKEAGUE, JJ., concurred, and a separate concurring opinion (pp. –––– – ––––). COLE, J. (pp. –––– – ––––), delivered a separate concurring opinion.NALBANDIAN, J., delivered the opinion of the court, in which COLE and McKEAGUE, JJ., concurred, and a separate concurring opinion (pp. 20–27). COLE, J. (pp. 14–19), delivered a separate concurring opinion.