Dissent

For more than four decades, the post-conviction architecture of federal sentencing has rested on a clear division of labor. Section 2255 of Title 28, enacted in 1948 and progressively calibrated by the Antiterrorism and Effective Death Penalty Act of 1996, is the vehicle through which a federal prisoner may challenge the legal validity of a sentence — its constitutionality, its statutory authority, its jurisdictional foundation. AEDPA wrote into that vehicle a careful structure of limits: a one-year statute of limitations, § 2255(f); a second-or-successive bar, § 2255(h); a certificate-of-appealability requirement, 28 U. S. C. § 2253(c); and the substantive standards governing procedural default and actual innocence that this Court developed in Bousley v. United States, 523 U. S. 614 (1998), McCleskey v. Zant, 499 U. S. 467 (1991), and Teague v. Lane, 489 U. S. 288 (1989). Section 3582(c)(1)(A), enacted within the Sentencing Reform Act of 1984, is the vehicle for discretionary reduction of an otherwise valid sentence in "extraordinary and compelling" circumstances. Each statute does what it was designed to do.

The Court today holds that those two vehicles may share cargo. A reason a prisoner cannot raise under § 2255 — whether because the limitations period has run, or because the second-or-successive bar applies, or because the prisoner has previously raised and lost the claim, or because the claim is one this Court has held is not cognizable under § 2255 at all — may, the Court tells us, nevertheless be advanced as part of an "extraordinary and compelling reasons" combination in a § 3582 motion. The result is that what AEDPA forbids on the front door may be smuggled in through the side door, so long as the prisoner can identify other reasons of any kind that, packaged together, persuade a district judge that the combined record is "extraordinary and compelling." We would not so unsettle the work Congress did.

We would affirm. We proceed in five parts. Part I addresses the text Congress wrote, which contemplates two distinct mechanisms operating on two distinct planes. Part II addresses the habeas-channeling line, which the majority confines to its narrowest formulation. Part III addresses the practical consequence the Court invites: the systematic relitigation of completed convictions under a more lenient discretionary standard than § 2255 supplies. Part IV addresses the institutional and historical record. Part V addresses the case before us — the particular ground on which the District Court actually granted relief — and explains why, even on the majority's reading, the judgment below should have been affirmed.

I

The majority's textual analysis begins and ends with the dictionary. Ante, at 7–8. "Extraordinary" and "compelling," the Court tells us, are "terms of degree" that "do their own work." They name a threshold. They do not name a category. Whatever class of reason a prisoner advances, the relevant question is whether that reason, in combination with others, is sufficiently extraordinary and sufficiently compelling to clear the bar.

That reading treats two evocative adjectives as the entirety of what § 3582(c)(1)(A)(i) says. But the statute does more than name a threshold. It frames the entire compassionate-release inquiry around the limited proposition that a district court "may reduce the term of imprisonment . . . if it finds that . . . extraordinary and compelling reasons warrant such a reduction." The verb is "reduce." The object is "the term of imprisonment." A term of imprisonment is reduced from one number to a lower one; it is not adjudicated unlawful. See Berman v. United States, 302 U. S. 211, 212 (1937) ("The sentence is the judgment."). The very text the statute uses presupposes a valid sentence that the court is empowered to shorten. See Resp. Br. 24. A claim that the sentence was unlawfully imposed in the first place — that the court was without jurisdiction, that the constitutional rules of trial were violated, that the verdict was infected by structural error — is not a claim that the sentence should be reduced. It is a claim that the sentence should be vacated. Vacatur is the work of § 2255, not § 3582.

The majority's response is that Fernandez did not ask for vacatur; he asked for a reduction. Ante, at 16. So he did. But the Court's substance-over-form discipline, articulated by the Court of Appeals below and shared by every circuit but one, addresses precisely the rhetorical move the majority's reading invites: the recharacterization of an invalidity claim as a request for reduction, so as to bypass the limits of § 2255. See United States v. Ferguson, 55 F. 4th 262, 270 (4th Cir. 2022); United States v. Amato, 48 F. 4th 61, 65 (2d Cir. 2022) (per curiam); United States v. Jenkins, 50 F. 4th 1185, 1202 (D. C. Cir. 2022); United States v. Wesley, 60 F. 4th 1277, 1284 (10th Cir. 2023); United States v. Hunter, 12 F. 4th 555, 567–568 (6th Cir. 2021); United States v. McCall, 56 F. 4th 1048, 1063–1064 (6th Cir. 2022) (en banc). When a sentencing judge looks at a record and concludes that the verdict has shaken his confidence — that the evidence may not have warranted conviction in the first place — and reduces a mandatory life sentence to time served on that basis, the relief is a reduction in form. In substance, the relief responds to a perceived defect in the conviction. That is the work § 2255 was enacted to do.

The majority answers that the rehabilitation-alone bar of 28 U. S. C. § 994(t) confirms its reading: an enacted limit on one category implies the exclusion of others. Ante, at 8. The negative implication is the wrong way around. Section 994(t) directs the Sentencing Commission to "describe what should be considered extraordinary and compelling reasons for sentence reduction, including the criteria to be applied and a list of specific examples." That delegation contemplates a Commission that draws lines around the kinds of reasons that qualify. It presupposes that the universe of qualifying reasons is bounded. The rehabilitation-alone bar is best read not as Congress's only substantive limit on "extraordinary and compelling reasons," but as Congress's single anticipatory clarification within a universe whose outer contours it left for the Commission to develop and for the courts to apply. The reading is consistent with the unbroken understanding of the Bureau of Prisons, codified in its 1994 regulations, that § 3582(c)(1)(A) is "compassionate release" for "particularly meritorious or unusual circumstances" — terminal illness, severe age-related disability, family catastrophe — that emerge after sentencing. See 59 Fed. Reg. 1238 (Jan. 7, 1994); 28 C. F. R. §§ 572.40(a)–(b) (1984); Resp. Br. 28–32.

The majority points to the fact that the Sentencing Commission's most recent policy statement, U. S. S. G. § 1B1.13 (Nov. 1, 2023), includes a "catch-all" for "any other circumstance or combination of circumstances . . . similar in gravity to those described in paragraphs (1) through (4)." § 1B1.13(b)(5). The policy statement also contains, however, "not a word about errors in a conviction or sentence as a basis for compassionate release." United States v. Wesley, 78 F. 4th 1221, 1222 (10th Cir. 2023) (Tymkovich, J., concurring in the denial of rehearing en banc). The Commission's silence on the very category the majority now welcomes is not a casual omission. It reflects the Commission's understanding of what § 3582(c)(1)(A) does and what it does not do.

II

The majority's habeas-channeling analysis treats Heck v. Humphrey, 512 U. S. 477 (1994), as the controlling precedent and Heck's "necessarily imply the invalidity" test as the dispositive formulation. Ante, at 13–14. The chosen formulation answers the question only by an avoidance technique that Preiser v. Rodriguez, 411 U. S. 475 (1973), expressly forbade.

Preiser identified the operative principle six years before Heck and twenty years before Wilkinson v. Dotson, 544 U. S. 74 (2005). The principle is anti-evasion. "It would wholly frustrate explicit congressional intent to hold that the respondents in the present case could evade this requirement by the simple expedient of putting a different label on their pleadings." 411 U. S., at 489–490. Preiser did not ask whether the relief sought necessarily implied invalidity in a mechanical, judgment-by-judgment sense. It asked whether the alternative vehicle could be used to obtain by indirection what the congressional scheme reserved for habeas.

That is the principle this Court has applied across more than a half-century of decisions. See Heck, 512 U. S., at 484 ("[Section] 1983 . . . is not an appropriate vehicle for challenging the validity of outstanding criminal judgments."); Edwards v. Balisok, 520 U. S. 641, 646 (1997) (extending Heck to a procedural § 1983 challenge whose success would "necessarily imply the invalidity" of the deprivation of good-time credits); Nance v. Ward, 597 U. S. 159, 167 (2022) (recognizing an implicit habeas exception within § 1983); Wilkinson, 544 U. S., at 83 ("[A] case challenging a sentence seeks a prisoner's 'release' in the only pertinent sense: It seeks invalidation (in whole or in part) of the judgment authorizing the prisoner's confinement; the fact that the State may seek a new judgment (through a new trial or a new sentencing proceeding) is beside the point."). And most recently in Jones v. Hendrix, 599 U. S. 465 (2023), the Court declined to permit a prisoner to use § 2255(e)'s saving clause to circumvent § 2255(h)'s second-or-successive bar: "the saving clause does not authorize such an end-run around AEDPA." Id., at 479.

The majority distinguishes Jones as concerning the internal architecture of § 2255 alone. Ante, at 14. The distinction is too thin. Jones's reasoning was structural: "Basic principles of statutory interpretation require that we construe the saving clause and § 2255(h) in harmony, not set them at cross-purposes." 599 U. S., at 478. The same construction applies, with equal force, to § 3582(c)(1)(A) and § 2255. To allow § 3582(c)(1)(A) to do what § 2255 does only under more restrictive conditions is to set the two statutes at cross-purposes — exactly what Jones counseled against.

The majority's reliance on § 3582(b) deserves separate response. Ante, at 14. Section 3582(b) provides that a judgment of conviction subject to modification under § 3582(c) "constitutes a final judgment for all other purposes." The majority reads that provision to mean that a § 3582 grant cannot "necessarily imply" the invalidity of the conviction because the conviction is statutorily declared to remain valid. Ibid. But that reading proves too much. Section 3582(b)'s "final judgment for all other purposes" clause exists to confirm that a sentence modification under § 3582(c) does not unsettle the judgment — for purposes of res judicata, of double-jeopardy attachment, of collateral estoppel — not to convert the substantive nature of the reason for modification. A district court that reduces a sentence because of accumulated doubts about a witness's credibility has, in substance, granted relief on a ground that goes to the validity of the conviction; the statute's declaration that the underlying judgment retains formal finality does not cleanse the substance of what occurred. Wilkinson's "seek invalidation (in whole or in part) of the judgment authorizing the prisoner's confinement" framing, 544 U. S., at 83, applies on the substance of the relief, not on the formal status of the underlying judgment.

The majority responds that Wilkinson's "necessarily demonstrate" formulation is the operative one and that, because a § 3582 grant does not formally invalidate the conviction, the rule does not apply. Ante, at 14. But the same opinion contains the language we have just quoted, and the language describes how a "case challenging a sentence" works. The majority chooses one formulation and discards the other. We would read the cases as a whole.

III

The practical consequence of the rule the Court adopts today deserves direct attention.

Congress legislated AEDPA's calibrations in 1996 against the background of a federal post-conviction docket whose volume — and whose institutional cost — Congress concluded required structural management. See Jones, 599 U. S., at 470–472. The one-year statute of limitations, the second-or-successive bar, the procedural-default rules of Bousley, the abuse-of-the-writ doctrine of McCleskey — each addressed a particular cost of unlimited collateral attack. The provisions worked together to mark a defendant's claim either as cognizable, with the procedural conditions Congress imposed, or not cognizable.

The rule the majority announces today permits the architecture to be circumvented. A defendant whose § 2255 motion is barred by AEDPA's limitations period may, the Court holds, raise the substance of the same claim under § 3582(c)(1)(A)(i) so long as he can point to other "reasons" of any kind that, in combination with the claim of error, meet the "extraordinary and compelling" threshold. A defendant who has previously raised and lost a claim under § 2255 may raise it again under § 3582, packaged with whatever additional reasons may then be available. A defendant whose claim of freestanding actual innocence cannot be brought under § 2255 at all — Herrera v. Collins, 506 U. S. 390, 400 (1993) — may bring the substance of it under § 3582 instead.

The majority assures that the "demanding 'extraordinary and compelling' threshold does its own work" and that, in any event, the empirical record from the First Circuit (where the rule the majority adopts has operated under United States v. Trenkler, 47 F. 4th 42 (1st Cir. 2022)) has not produced a flood of motions. Ante, at 19. We doubt both reassurances. The threshold cannot do its own work when its operative criterion is whether a particular district judge finds the combined record sufficient — a judgment to which appellate review is deferential and on which circuit-level guidance is, by the majority's design, minimal. And the empirical record is a pre-decision empirical record. The Government's projection that a Supreme Court ruling in petitioner's favor will "super-charge" the use of § 3582, Tr. of Oral Arg. 29, is not a "scare tactic," Tr. of Oral Arg. 70; it is a recognition that the absence of a national rule has, until today, constrained the universe of motions. The grant in this case removes that constraint.

The majority counters that procedural default is not itself "extraordinary," ante, at 20, and that the AEDPA limits will continue to operate where § 2255 is the operative vehicle. But the majority's rule does not require procedural default to do any work in § 3582 motions; it requires only that the prisoner package whatever underlying claim he has with other reasons. A defendant whose actual-innocence claim has been adjudicated "plainly meritless" under § 2255 may, on the majority's reading, raise the substance of the same claim under § 3582 packaged with sentencing disparity, with rehabilitation, with conditions of confinement — and may obtain relief from a district judge whose discretion is broad and whose conclusions on the "extraordinary and compelling" threshold are reviewed for abuse. The defendant has not, in any meaningful sense, been bound by his earlier defeat under § 2255.

What the majority's rule does, in short, is replace AEDPA's structural calibrations with a single discretionary criterion lodged in the district courts. Whether that is wise as a matter of policy is not for us. Whether it is what Congress wrote is. We do not think it is.

IV

The historical record reinforces our reading.

The Sentencing Reform Act of 1984 abolished parole prospectively, see Mistretta v. United States, 488 U. S. 361, 367 (1989), and rewrote Rule 35 to eliminate the prior 120-day "any reason" reduction authority. See Resp. Br. 24–26. The SRA enacted § 3582(c)(1)(A) as a discrete, BOP-initiated mechanism. Until 2018, the only party who could file a § 3582(c)(1)(A) motion was the Director of the Bureau of Prisons. The Senate Report described § 3582(c)(1)(A) as providing authority "similar to" the pre-SRA § 4205(g) — a provision whose implementing regulations specified "particularly meritorious or unusual circumstances," "such as an extraordinary change in the inmate's personal or family situation or severe illness." S. Rep. No. 98-225, p. 121 n. 298 (1983); 28 C. F. R. §§ 572.40(a)–(b) (1984).

That history does not bear out the proposition that § 3582(c)(1)(A) was designed to address legal-error-style reasons. The BOP, the agency Congress entrusted with the front-end gatekeeping function for thirty-four years, was not expected to evaluate the legal adequacy of trial proceedings. See Tr. of Oral Arg. 9–10. Its competence was the prison: medical conditions, age-related disability, family emergencies, post-sentencing developments observable from within the facility. The structural argument matches: a statute that lodged front-end filing authority in an institution with no legal-error competence is not best read to require that institution to evaluate legal-error reasons.

The First Step Act of 2018 added the prisoner-initiated filing mechanism but did not rewrite the substantive standard. Pub. L. No. 115–391, § 603(b), 132 Stat. 5239. Section 994(t) remains as it was. The Senate sponsors of the FSA emphasized in their amicus brief that the FSA "did not alter the meaning of 'extraordinary and compelling reasons.'" Senators Br. 2. The majority agrees, ante, at 5, but draws from that agreement a conclusion the FSA's own structure cannot bear: that the pre-2018 statute, BOP-administered for thirty-four years on the assumption that "extraordinary and compelling reasons" meant personal-circumstance reasons, contained all along an unenacted authorization to consider every category of reason a § 2255 motion could raise. The reading is in tension with the text Congress wrote, with the practice that text inhabited for three decades, and with the architecture the FSA preserved.

The pre-SRA cases that surfaced at oral argument — Banks v. United States, 428 F. Supp. 1088 (D. N. J. 1977), and United States v. Diaco, see Tr. of Oral Arg. 6–7, 45–47 — do not support a broader reading either. Banks turned on rehabilitation as a factor — a category Congress addressed expressly in § 994(t). Diaco, as the United States explained at argument, predated the SRA and is one the Government regards as wrongly decided. Tr. of Oral Arg. 46–47. Neither case stands for the proposition that pre-SRA practice considered the kind of validity-attacking reasons that the majority now reads § 3582(c)(1)(A) to accommodate.

V

Finally, even on the rule the majority adopts, the judgment below should have been affirmed.

The reason the District Court actually relied on was not, on its terms, a § 2255-overlapping reason that fit comfortably into the "combination" framing the majority preserves. It was a free-standing "disquiet" about the jury's verdict — a disquiet that had no anchor in any new fact discovered after trial, in any new law, or in any constitutional rule Fernandez had not already attempted to invoke through his first § 2255 motion. The grounds the district judge listed were: the post-murder flight of the Darge brothers contrasted with Fernandez's settled life; the cooperator's motive to lie; ballistic inconsistencies; the possibility of more effective cross-examination of the cooperator; and the government's choice not to charge the getaway driver. Pet. App. 36a–37a. Each of those grounds was available to Fernandez at trial, on direct appeal, and during his first § 2255 motion. None depended on any post-conviction development. Each had been considered, in some form, by a court that affirmed the verdict and a court that found the actual-innocence assertion "plainly meritless."

A sentencing judge's accumulated unease about a verdict, on facts that have not changed, is not an "individual circumstance" of the defendant. Cf. Tr. of Oral Arg. 35–36. It is a circumstance of the judge. Whatever generosity our system extends to district judges who, in retrospect, come to doubt a verdict they themselves received and rejected post-trial motions against, the appropriate remedy is not a discretionary commutation through § 3582(c)(1)(A) on a record indistinguishable from the trial record. The appropriate remedy — to the extent any judicial remedy is available — runs through the channels Congress and this Court have provided for revisiting verdicts. Where, as here, those channels have been exhausted or have rejected the underlying claim, the only remaining mechanism is one not vested in any court at all. Cf. Herrera, 506 U. S., at 411–417 (executive clemency).

The majority does not disagree, exactly. It declines to decide whether "the considerations the District Court relied on, taken in combination, are 'extraordinary and compelling.'" Ante, at 18. It remands for the Second Circuit to decide that question in the first instance. Ibid. But the question, in the case as it actually arises, is not whether the District Court's combination of reasons meets the threshold. It is whether the rule the majority articulates — the "Heck necessarily implies" rule, paired with the rehabilitation-alone analogue — can do useful work in a case where the only "reason" the District Court relied on, when stripped of doublings, is the judge's own disquiet about the verdict.

We do not see how it can. The "disquiet" rationale does not survive Heck's "necessarily imply" test only by a verbal trick: the sentence reduction does not formally invalidate the conviction. But on the substance the District Court treated as decisive, the reduction was a response to the judge's view that the conviction may have been mistaken. Calling that response a "modification" rather than a "vacatur" does not change what the District Court was doing. Wilkinson, 544 U. S., at 83, said precisely this: the formal label of a remedy does not transmute the underlying claim.

* * *

The Court today adopts a rule whose effect is to convert a discretionary safety valve, designed for the case of the dying inmate and the catastrophic personal-circumstance change, into a parallel channel through which any prisoner with imagination can relitigate the conviction that imprisoned him. The rule leaves the "extraordinary and compelling" threshold as the only protection against the consequence the Government and the dissenting judges below predicted: a flood of motions whose volume, in time, will give the threshold considerably less work to do than the Court today claims it will. The rule disregards the architecture AEDPA built and Congress preserved when it amended § 3582(c)(1)(A) in 2018. The rule sets two statutes that Congress designed to do different work at cross-purposes — exactly what Jones v. Hendrix counseled against. And the rule, as the case actually arises, requires the Second Circuit to decide whether a sentencing judge's accumulated unease about a verdict on unchanged facts can be "extraordinary and compelling" — a question the majority is unwilling to answer directly and to which the answer, in fidelity to the statute Congress wrote, is no.

We would affirm the judgment of the United States Court of Appeals for the Second Circuit. We respectfully dissent.