Harmless-Error Review of § 924(c) Alternative-Predicate Instruction Errors on § 2255 Is Record-Based, Not Categorical

I. Introduction

Case: Shea v. United States (First Circuit, Feb. 23, 2026).
Parties: Anthony M. Shea (petitioner/appellant; defendant/appellant) vs. United States (respondent/appellee; appellee).
Posture: Consolidated appeals from (1) a partial denial of Shea’s motion under 28 U.S.C. § 2255 and (2) an amended judgment entered after resentencing.

Shea was convicted in 1997 of fourteen federal offenses tied to two robbery episodes (Seabrook and Hudson, New Hampshire), including two convictions under 18 U.S.C. § 924(c) (using/carrying a firearm “during and in relation to” a “crime of violence”). At trial, the jury was instructed that each § 924(c) count could be predicated on multiple alternative “crimes of violence,” including both substantive robbery offenses and conspiracy-to-rob offenses. The jury returned general verdicts.

After the Supreme Court’s void-for-vagueness decisions culminating in United States v. Davis, Shea filed a second/successive § 2255 motion. He argued that conspiracy to commit robbery was no longer a valid § 924(c) predicate and that his § 924(c) convictions must be vacated because the general verdicts might have rested on invalid predicates. He also sought resentencing based on a conceded error in his “career offender” designation under U.S.S.G. § 4B1.1, an issue previously discussed in Shea v. United States, 976 F.3d 63 (1st Cir. 2020).

Core issue on appeal: When a § 924(c) jury instruction listed both valid and invalid predicate “crimes of violence,” must a court, in deciding harmlessness on collateral review, apply the “categorical approach” (ignoring case-specific facts), or may it examine the record to assess whether the error substantially influenced the verdict?

II. Summary of the Opinion

The First Circuit affirmed both challenged rulings. It held that although Davis rendered the conspiracy predicates legally invalid and thus made the jury instructions erroneous, Shea was not entitled to § 2255 relief because he failed to show that the instructional error had a “substantial and injurious effect or influence” on the jury’s verdict under Brecht v. Abrahamson.

Critically, the court rejected Shea’s argument that harmless-error review must “incorporate” the categorical approach. Instead, it held that harmlessness is assessed by examining the record and the “setting of a particular case,” consistent with Chapman v. California and O’Neal v. McAninch. The court also rejected Shea’s invocation of the rule of lenity and found certain district court cases he cited inapposite.

Because Shea’s only challenge to the amended judgment was derivative of his claim that the § 924(c) convictions were invalid, the resentencing appeal failed as well.

III. Analysis

A. Precedents Cited

1. The “crime of violence” framework and the categorical approach

  • Delligatti v. United States, 604 U.S. 423 (2025): Cited for the definition of the § 924(c) “force clause”/“elements clause” and the requirement that courts use the categorical approach to determine whether an offense always involves the use/attempted use/threatened use of force. The First Circuit used Delligatti to distinguish “categorical approach” classification (what counts as a predicate) from “harmless error” analysis (whether an instructional error mattered to the verdict).
  • United States v. Cruz-Rivera, 904 F.3d 63 (1st Cir. 2018): Cited for terminology (“force clause”) and for describing the categorical approach’s role in deciding whether a predicate qualifies under § 924(c). The court emphasized that this classification function is different from harmless-error review.
  • Mathis v. United States, 579 U.S. 500 (2016): Shea relied on Mathis for the proposition that the categorical approach requires courts to ignore the facts of the case. The First Circuit accepted that point as to predicate classification, but rejected importing that fact-blind methodology into harmless-error analysis.

2. Void-for-vagueness decisions (why the instructions became erroneous)

  • Johnson v. United States, 576 U.S. 591 (2015): Held ACCA’s residual clause unconstitutionally vague. The decision supplied the logic later applied to other residual clauses.
  • Welch v. United States, 578 U.S. 120 (2016): Held Johnson announced a retroactive substantive rule on collateral review, enabling § 2255 litigants to bring vagueness challenges.
  • United States v. Davis, 588 U.S. 445 (2019): Extended Johnson to § 924(c) by holding § 924(c)(3)(B)’s residual clause void for vagueness. Here, Davis supplied the premise that conspiracy-to-rob predicates were invalid “crimes of violence,” making the jury’s alternative-predicate instructions erroneous.

3. Alternative-theory instructional error and harmlessness standards

  • Hedgpeth v. Pulido, 555 U.S. 57 (2008) (per curiam) & Skilling v. United States, 561 U.S. 358 (2010): Establish that constitutional error occurs when a jury is instructed on alternative theories and returns a general verdict that may rest on a legally invalid theory. The government did not dispute this framework.
  • United States v. Latorre-Cacho, 874 F.3d 299 (1st Cir. 2017): Quoted for the proposition (drawing on Pulido and Skilling) that such alternative-theory general-verdict errors are constitutional errors.
  • Brecht v. Abrahamson, 507 U.S. 619 (1993): Supplies the collateral-review harmlessness standard: relief is warranted only if the error had a “substantial and injurious effect or influence” on the verdict. The First Circuit treated Brecht as the controlling prejudice inquiry on § 2255 collateral review.
  • O’Neal v. McAninch, 513 U.S. 432 (1995): Elaborates that a judge asks whether there is “grave doubt” that an error substantially influenced the jury’s decision—reinforcing that the inquiry is a legal judgment grounded in the record of the particular case.
  • Chapman v. California, 386 U.S. 18 (1967): Cited for the “setting of a particular case” framing of harmlessness—used here to justify considering the trial record rather than ignoring facts.
  • Neder v. United States, 527 U.S. 1 (1999): Used for the general principle that harmless-error review asks whether the jury verdict would have been the same absent the error.

4. First Circuit and sister-circuit approaches to record-based harmlessness in this context

  • United States v. Serrano-Delgado, 29 F.4th 16 (1st Cir. 2022): Demonstrates the First Circuit’s practice of examining “this record” to assess whether a jury relied on a valid predicate.
  • United States v. Rodríguez-Santos, 56 F.4th 206 (1st Cir. 2022): On plain-error review, considered whether valid and invalid predicates were “inextricably intertwined,” again reflecting a record-sensitive approach.
  • Granda v. United States, 990 F.3d 1272 (11th Cir. 2021): Rejected “extend[ing] the categorical approach” to determining which predicate a jury’s general verdict relied on in a collateral harmlessness inquiry.
  • United States v. Reed, 48 F.4th 1082 (9th Cir. 2022): Explained there is “no need to employ the categorical approach” once offenses are already categorized; the question becomes the gravity of the instructional error.
  • Nicholson v. United States, 78 F.4th 870 (6th Cir. 2023): Stated the categorical approach does not govern how courts address jury-instruction error on § 2255, which is reviewed for harmless error.
  • Stone v. United States, 37 F.4th 825 (2d Cir. 2022): Declined to import the categorical method into a prejudice determination for a § 924(c) conviction partially predicated on an invalid crime of violence.

5. Authorities discussed and distinguished

  • Pulido v. Chrones, 629 F.3d 1007 (9th Cir. 2010), cert. denied, 565 U.S. 918 (2011), abrogated on other grounds by Sansing v. Ryan, 41 F.4th 1039 (9th Cir. 2022): Cited to support the general notion that alternative-theory harmless-error review is record-based. The “abrogated on other grounds” note signals the First Circuit used it for a limited proposition.
  • United States v. Berry, No. 09-cv-00019, 2020 WL 591569 (W.D. Va. Feb. 6, 2020) (unpublished) & United States v. White, 510 F. Supp. 3d 443 (W.D. Tex. 2020): Rejected as inapposite because they involved jury instructions with a single predicate that might be invalid, rather than multiple alternative predicates with a general verdict.

B. Legal Reasoning

1. Identifying the error (but keeping the remedy question separate)

The court accepted that Davis invalidated § 924(c)(3)(B)’s residual clause, and that conspiracy to commit robbery no longer qualified as a “crime of violence” for § 924(c) purposes. Because the jury instructions permitted conviction if the firearm was used during a conspiracy-to-rob predicate, the instructions were “rendered retroactively erroneous.”

But the existence of constitutional instructional error under Hedgpeth v. Pulido and Skilling v. United States did not end the matter: on § 2255 collateral review, Shea had to satisfy Brecht v. Abrahamson.

2. The central holding: the categorical approach does not control harmless-error review

Shea’s core argument was methodological: because courts use the categorical approach to decide whether an offense is a “crime of violence,” he claimed that harmlessness analysis must also ignore the particular facts and thus could not rely on record findings that valid and invalid predicates were “interrelated and coextensive.”

The First Circuit rejected that move by drawing a functional boundary:

  • Categorical approach: a classification tool used to decide whether an offense qualifies as a predicate under the elements/force clause (e.g., per Delligatti v. United States and United States v. Cruz-Rivera).
  • Harmless-error analysis: a verdict-effect tool used to decide whether an identified legal mistake actually mattered to the jury’s decision in the “setting of a particular case” (Chapman v. California) and whether the court has “grave doubt” it influenced the verdict (O’Neal v. McAninch), applying Brecht v. Abrahamson.

Because harmlessness asks whether the verdict would have been the same absent the error (Neder v. United States), it necessarily invites examination of what the jury convicted on, what counts were found, and how predicates related in the trial record—rather than requiring a fact-blind abstraction.

3. Consistency with circuit practice and national consensus

The court reinforced its reasoning by noting:

  • Its own prior decisions had used record-based analysis in analogous contexts (e.g., United States v. Serrano-Delgado; United States v. Rodríguez-Santos).
  • Multiple sister circuits rejected extending the categorical approach into the harmlessness/prejudice inquiry (e.g., Granda v. United States; United States v. Reed; Nicholson v. United States; Stone v. United States).

4. Rule of lenity rejected

Shea invoked the rule of lenity, but the court held it had “no application” because there was no ambiguity about the breadth of a criminal statute to be resolved in his favor. The case concerned harmlessness/prejudice methodology, not uncertain statutory scope.

5. Outcome determinative narrowing in the resentencing appeal

Shea’s challenge to the amended judgment was expressly limited to the ground that it imposed sentences for § 924(c) counts that he believed invalid. Once the First Circuit affirmed the continued validity of those convictions, that derivative sentencing challenge necessarily failed.

C. Impact

  • Clarifies the post-Davis litigation playbook in the First Circuit: Defendants challenging § 924(c) convictions on § 2255 due to alternative-predicate instructions cannot force courts to ignore the record by invoking the categorical approach. The prejudice question remains a record-grounded harmlessness inquiry under Brecht v. Abrahamson.
  • Aligns the First Circuit with other circuits: By expressly relying on Granda v. United States, United States v. Reed, Nicholson v. United States, and Stone v. United States, the opinion contributes to an emerging consensus: categorical analysis determines predicate validity, while harmlessness determines remedy.
  • Raises the bar for collateral relief from mixed-predicate general verdicts: Even where an instruction is retroactively invalidated by Davis, relief turns on whether the petitioner can show substantial influence on the verdict. Where valid and invalid predicates are factually “interrelated,” “coextensive,” or “inextricably intertwined,” courts are more likely to find harmlessness.
  • Channels disputes into “record and linkage” arguments: Future litigants are incentivized to argue not that courts must ignore facts, but that the record meaningfully permits the jury to have relied on the invalid predicate alone. (The court underscored that Shea did not mount any such alternative challenge to the district court’s harmlessness analysis.)

IV. Complex Concepts Simplified

  • § 924(c) “crime of violence”: A separate firearm crime that depends on an underlying predicate offense meeting a statutory definition (here, the “force clause” vs. the now-invalid “residual clause”).
  • Force/elements clause vs. residual clause: The force clause focuses on whether the offense’s legal elements require violent force; the residual clause asked whether the offense “by its nature” involved a substantial risk of force—held unconstitutionally vague in United States v. Davis.
  • Categorical approach: A method that looks to the legal definition (elements) of an offense, not the defendant’s real-world conduct, to decide whether the offense qualifies as a “crime of violence.”
  • Alternative-theory/general verdict problem: If the jury is told it may convict under Theory A or Theory B, and later Theory B is declared legally invalid, a general guilty verdict creates uncertainty about which theory the jury relied on.
  • Harmless error on collateral review (Brecht v. Abrahamson): Even if there was error, the conviction stands unless the petitioner shows the error had a substantial and injurious effect on the verdict. Under O’Neal v. McAninch, if the judge is in “grave doubt” about the effect, the petitioner prevails.
  • Why “categorical” and “harmlessness” are different questions: “Categorical” asks: Was the predicate legally valid in the abstract? “Harmlessness” asks: Did the jury’s verdict in this trial likely depend on the invalid option?

V. Conclusion

Shea v. United States establishes a clear methodological rule for post-Davis § 2255 challenges in the First Circuit: while the categorical approach governs whether a predicate qualifies as a § 924(c) “crime of violence,” it does not govern harmless-error review of alternative-predicate jury instructions. On collateral review, courts may examine the trial record to decide whether the instructional error substantially influenced the verdict under Brecht v. Abrahamson.

The decision both narrows the path to vacating mixed-predicate § 924(c) convictions and aligns the First Circuit with a broader circuit consensus rejecting efforts to “import” categorical methodology into harmlessness and prejudice determinations.