United States v. Tyrone Davis

 FOR PUBLICATION UNITED STATES COURT OF APPEALS FOR THE NINTH CIRCUIT UNITED STATES OF AMERICA, No. 13-30133 Plaintiff-Appellee, D.C. No. v. 3:04-cr-05350-RBL-2 TYRONE DAVIS, Defendant-Appellant. OPINION Appeal from the United States District Court for the Western District of Washington Ronald B. Leighton, District Judge, Presiding Argued and Submitted En Banc September 10, 2015 San Francisco, California Filed June 13, 2016 Before: Sidney R. Thomas, Chief Judge and William A. Fletcher, Richard A. Paez, Richard C. Tallman, Johnnie B. Rawlinson, Consuelo M. Callahan, Carlos T. Bea, Morgan Christen, Jacqueline H. Nguyen, Andrew D. Hurwitz and John B. Owens, Circuit Judges. Opinion by Judge Paez; Concurrence by Judge Christen; Dissent by Judge Bea 2 UNITED STATES V. DAVIS SUMMARY* Criminal Law The en banc court reversed the district court’s determination that Tyrone Davis is not eligible for a sentence reduction under 18 U.S.C. § 3582(c)(2), and remanded for reconsideration of whether Davis should receive a reduction under § 3582(c)(2) and the Guidelines’ related policy statements. Revisiting Freeman v. United States, 564 U.S. 522 (2011), which addressed whether a defendant sentenced pursuant to a Federal Rule of Criminal Procedure 11(c)(1)(C) plea agreement is eligible for a sentence reduction under § 3582(c)(2), and this court’s application of Marks v. United States, 430 U.S. 188 (1977), to fractured Supreme Court opinions, the en banc court held that where no rationale common to a majority of the Justices can be identified, only the result is binding. In so holding, the en banc court joined the D.C. Circuit, which concluded that Freeman lacks a controlling opinion “because the plurality and concurring opinions do not share common reasoning whereby one analysis is a logical subset of the other.” Bound only by Freeman’s specific result, the en banc court adopted the plurality opinion’s approach, which holds that “[e]ven when a defendant enters into an 11(c)(1)(C) agreement, the judge’s decision to accept the plea and impose the recommended sentence is likely to be based on the Guidelines; and when it is, the defendant should be eligible to seek § 3582(c)(2) * This summary constitutes no part of the opinion of the court. It has been prepared by court staff for the convenience of the reader. UNITED STATES V. DAVIS 3 relief.” Adopting and applying the plurality approach, the en banc court overruled United States v. Austin, 676 F.3d 924 (9th Cir. 2012), which held that Justice Sotomayor’s concurring opinion was controlling, and concluded that Davis is eligible to seek a sentence reduction under § 3582(c)(2). Judge Christen, joined by Chief Judge Thomas, and Judges Tallman, Nguyen, and Hurwitz, concurred. She joined in the en banc court’s holding as far as it goes, but disagreed with the majority’s assumption that a court might be free to take dissenting opinions into account in future Marks analyses of what binding rule, if any, emerges from a fractured Supreme Court decision. Dissenting, Judge Bea wrote that the majority’s “logical subset” requirement finds no support in Marks or any other Supreme Court precedent; that even if there were such a requirement, the majority misreads Justice Kennedy’s plurality opinion to the extent it concludes that there are circumstances in which Justice Sotomayor would permit sentence modification but the Kennedy plurality would not; and that the majority’s adoption of the Kennedy plurality’s approach violates stare decisis because five Justices in Freeman agreed that a court looks to the plea agreement itself to determine whether a plea was “based on” since-modified sentencing Guidelines. 4 UNITED STATES V. DAVIS COUNSEL Nathaniel Garrett (argued), San Francisco, California; Anna M. Tolin, Kirkland, Washington, for Defendant-Appellant. Michael S. Morgan (argued), Assistant United States Attorney; Jenny A. Durkan, United States Attorney; United States Attorney’s Office, Western District of Washington, Seattle, Washington; for Plaintiff-Appellee. Nancy L. Talner, ACLU-WA Foundation, Seattle, Washington; Michael Filipovic, Federal Public Defender for the Western District of Washington, Seattle, Washington; Theresa M. DeMonte and Andrew R.W. Hughes, Calfo Harrigan Leyh & Eakes LLP, Seattle, Washington; Suzanne Lee Elliott, Co-Chair, WACDL Amicus Committee, Seattle, Washington; for Amici Curiae ACLU-WA, Federal Public Defender for the Western District of Washington, and WACDL. UNITED STATES V. DAVIS 5 OPINION PAEZ, Circuit Judge: In this case, we must consider how to interpret the Supreme Court’s fractured opinion in Freeman v. United States, which addressed whether a defendant sentenced pursuant to a Federal Rule of Criminal Procedure 11(c)(1)(C) plea agreement is eligible for a sentence reduction under 18 U.S.C. § 3582(c)(2). 564 U.S. 522 (2011). Although a majority of the Court held that in such cases defendants remain eligible for relief under § 3582(c)(2), the plurality and concurring opinions did not agree on a single rationale. Citing Marks v. United States, 430 U.S. 188 (1977), which provides guidance on interpreting fractured Supreme Court opinions, we held in United States v. Austin that Justice Sotomayor’s concurring opinion was controlling, as it represented the narrowest grounds on which a majority of the justices agreed. See 676 F.3d 924 (9th Cir. 2012). Applying Justice Sotomayor’s approach, the district court denied Appellant Tyrone Davis’s (“Davis”) motion for a sentence reduction, ruling that his sentence was based on a Rule 11(c)(1)(C) plea agreement and not a “sentencing range that has subsequently been lowered by the Sentencing Commission” as required by § 3582(c)(2). Revisiting Freeman and our application of Marks to fractured Supreme Court opinions, we hold that where we can identify no rationale common to a majority of the Justices, we are bound only by the result. In so holding, we join the D.C. Circuit, which concluded that Freeman lacks a controlling opinion “because the plurality and concurring opinions do not share common reasoning whereby one analysis is a logical subset of the other.” United States v. Epps, 707 F.3d 337, 6 UNITED STATES V. DAVIS 350 (D.C. Cir. 2013) (internal quotation marks and citation omitted). Bound only by Freeman’s specific result, the D.C. Circuit adopted the plurality opinion’s approach, which holds that “[e]ven when a defendant enters into an 11(c)(1)(C) agreement, the judge’s decision to accept the plea and impose the recommended sentence is likely to be based on the Guidelines; and when it is, the defendant should be eligible to seek § 3582(c)(2) relief.” Freeman, 564 U.S. at 534 (plurality opinion). We also adopt this approach and overrule Austin. Applying the plurality approach, we conclude that Davis is eligible to seek a sentence reduction under § 3582(c)(2). We therefore reverse the district court’s denial of Davis’s motion and remand for a determination of whether Davis should receive a reduction in his sentence. I. A. In 2005, pursuant to a plea agreement entered into under Rule 11(c)(1)(C), Davis pled guilty to a series of counts related to distribution of cocaine base, or “crack cocaine.”1 In his plea agreement, Davis admitted his offense conduct involved at least 170.5 grams of crack cocaine. Given the quantity of crack cocaine for which Davis accepted direct responsibility, the parties agreed that Davis’s base offense level was 34 under United States Sentencing Guidelines 1 Davis pled guilty to the following offenses alleged in the Third Superseding Indictment: conspiracy to distribute cocaine base in violation of 21 U.S.C. §§ 841(a)(1), 841(b)(1)(A), and 846 (Count 1); distribution of cocaine base in violation of 21 U.S.C. §§ 841(a)(1), 841(b)(1)(A), and 860 (Counts 10 and 11); and possession with intent to distribute cocaine base in violation of 21 U.S.C. §§ 841(a)(1) and 841(b)(1)(C) (Counts 12 and 13). UNITED STATES V. DAVIS 7 (“USSG”) § 2D1.1(c)(3) (Nov. 2005) and that the appropriate prison sentence would be eighteen years. The plea agreement acknowledged that under Rule 11(c)(1)(C), the district court had discretion to accept or reject the proposed agreement and recommended sentence. If the district court accepted the agreement, then under Rule 11(c)(1)(C) the recommended sentence would be binding on the court. On the other hand, if the court rejected the recommended sentence, the parties could withdraw from the agreement. At sentencing in 2006, the district court calculated Davis’s total offense level as 37 with a Criminal History Category II, resulting in a Guidelines range of 235 to 293 months. The court accepted the Rule 11(c)(1)(C) plea agreement and imposed the recommended eighteen-year (216-month) sentence. We reversed and remanded, holding that the district court had erred in its determination of Davis’s criminal history category and in its imposition of an “organizer or leader” enhancement. United States v. Davis, 312 F. App’x 909, 911–14 (9th Cir. 2009). At Davis’s resentencing in 2009, the district court calculated his total offense level as 36 with a Criminal History Category I, resulting in a Guidelines range of 188 to 235 months. The court reimposed the plea agreement’s recommended eighteen- year (216-month) sentence, finding it “fair and reasonable” under the Guidelines. We affirmed. United States v. Davis, 389 F. App’x 616 (9th Cir. 2010). B. When Davis pled guilty, the Guidelines punished defendants far more harshly for crack cocaine offenses than for powder cocaine offenses. A defendant responsible for one gram of crack cocaine faced the same Guidelines sentence as 8 UNITED STATES V. DAVIS a defendant responsible for one hundred grams of powder cocaine. This 100:1 ratio was roundly criticized for its racially disparate effects. See, e.g., Kimbrough v. United States, 552 U.S. 85, 98 (2007) (citing the Sentencing Commission’s finding that the “severe sentences required by the 100-to-1 ratio are imposed primarily upon black offenders”) (internal quotation marks omitted). As a 2006 survey by the Substance Abuse and Mental Health Services Administration showed, whites formed the biggest group of crack cocaine users in absolute numbers,2 but African Americans were disproportionately arrested and convicted for crack cocaine offenses. A 2007 report by the Sentencing Commission documented that when Davis was sentenced in 2006, 81.8% of federal crack cocaine offenders were African American. U.S. Sentencing Commission, Report to Congress: Cocaine and Federal Sentencing Policy at 15 (2007) (“2007 Report”). Moreover, in a series of reports to Congress, the Sentencing Commission warned that the “data no longer support” the assumption that crack cocaine is more harmful than powder cocaine. Kimbrough, 552 U.S. at 97–98 (quoting U.S. Sentencing Commission, Report to Congress: Cocaine and Federal Sentencing Policy at 96 (2002); also citing the 2007 Report); see also United States v. Baptist, 2 Results from the Substance Abuse and Mental Health Services Administration’s National Survey on Drug Use and Health from 2006 show that 3.3% of whites report ever using crack cocaine compared with 5.4% of African Americans. Thus, in absolute numbers, white crack cocaine users far outnumber African-American crack cocaine users. Quick Table: Ever Used Crack BY Race and Ethnicity, National Survey on Drug Use and Health, 2006, available at https://www.icpsr.umich.edu/icpsrweb/NAHDAP/series/00064/studies (follow “National Survey on Drug Use and Health, 2006”; then follow “Quick Tables, Drug Use: Entire Sample”; then follow “Crack Use”’; then select “Race and Ethnicity”; and generate table). UNITED STATES V. DAVIS 9 646 F.3d 1225, 1226, 1228 n.1 (9th Cir. 2011) (per curiam). Citing the “urgent and compelling” problems raised by the overly punitive crack sentencing scheme, the Sentencing Commission repeatedly called on Congress to reduce the 100:1 ratio. See, e.g., 2007 Report at 8–9. Federal judges and Department of Justice officials likewise joined the chorus of voices demanding reform. See, e.g., United States v. Then, 56 F.3d 464, 467 (2d Cir. 1995) (Calabresi, J., concurring) (“The unfavorable and disproportionate impact that the 100- to-1 crack/cocaine sentencing ratio has on members of minority groups is deeply troubling.”); Remarks of Attorney General Eric Holder, D.C. Court of Appeals Judicial Conference (June 19, 2009), available at http://www.justice.gov/opa/speech/attorney-general-eric- holder-dc-court-appeals-judicial-conference (“It is the view of this Administration that the 100-to-1 crack-powder sentencing ratio is simply wrong. It is plainly unjust to hand down wildly disparate prison sentences for materially similar crimes.”). In 2010, Congress responded by passing the Fair Sentencing Act, Pub. L. No. 111-220, 124 Stat. 2372 (2010), which dramatically reduced the sentencing ratio to 18:1. The Fair Sentencing Act also gave the Sentencing Commission “emergency authority” to “make such conforming amendments to the Federal sentencing guidelines.” Id. § 8. The Sentencing Commission responded by issuing amended Guidelines reflecting the new 18:1 ratio3 and made the 3 See U.S. Sentencing Guidelines Manual app. C, Amend. 748 (U.S. Sentencing Comm’n 2010) (adjusting Guidelines temporarily); id. at Amend. 750 (2011) (making adjustment permanent). 10 UNITED STATES V. DAVIS changes retroactive4 for all defendants who have “been sentenced to a term of imprisonment based on a sentencing range that has subsequently been lowered by the Sentencing Commission.” 18 U.S.C. § 3582(c)(2). C. In 2012, Davis filed a pro se motion under § 3582(c)(2) seeking a retroactive reduction of his sentence in light of the amended Guidelines.5 Before the Fair Sentencing Act, the stipulated amount of 170.5 grams of crack cocaine in Davis’s Rule 11(c)(1)(C) plea agreement resulted in a base offense level of 34. After the dramatic reduction in the sentencing ratio, that same amount yielded a base offense level of 28. USSG § 2D1.1(c)(6) (2011). Davis argued that he was eligible for a sentence modification under § 3582(c)(2) because his sentence was “based on” the Guidelines. 4 Id. at Amend. 759 (making Amendment 750 retroactive). 5 Section 3582(c)(2) permits a district court to modify a term of imprisonment . . . in the case of a defendant who has been sentenced to a term of imprisonment based on a sentencing range that has subsequently been lowered by the Sentencing Commission pursuant to 28 U.S.C. 994(o), upon motion of the defendant or the Director of the Bureau of Prisons, or on its own motion, the court may reduce the term of imprisonment, after considering the factors set forth in section 3553(a) to the extent that they are applicable, if such a reduction is consistent with applicable policy statements issued by the Sentencing Commission. UNITED STATES V. DAVIS 11 The district court denied the motion, holding that it lacked jurisdiction to modify Davis’s sentence because it was “based on” the Rule 11(c)(1)(C) plea agreement, not the Guidelines. In so ruling, the district court concluded that it was bound by Justice Sotomayor’s concurring opinion in Freeman. Davis appealed for a third time. A three-judge panel affirmed, relying on Austin. United States v. Davis, 776 F.3d 1088 (9th Cir. 2015). We granted rehearing en banc. United States v. Davis, 795 F.3d 1188 (9th Cir. 2015).6 II. A. In Freeman v. United States, the Supreme Court considered whether a defendant sentenced under a Rule 11(c)(1)(C) agreement may be eligible for a sentence reduction under § 3582(c)(2). 564 U.S. 522 (2011). Five justices ultimately agreed that Freeman was eligible for a reduction, but no rationale commanded a majority of the Court. A four-justice plurality held that “[e]ven when a defendant enters into an 11(c)(1)(C) agreement, the judge’s decision to accept the plea and impose the recommended sentence is likely to be based on the Guidelines; and when it is, the defendant should be eligible to seek § 3582(c)(2) relief.” Id. at 534 (plurality opinion). Writing for the plurality, Justice Kennedy explained that this approach was 6 Whether a district court has jurisdiction to modify a defendant’s sentence under 18 U.S.C. § 3582(c)(2) is a legal question that we review de novo. United States v. Paulk, 569 F.3d 1094, 1095 (9th Cir. 2009) (per curiam). 12 UNITED STATES V. DAVIS consistent with the “text and purpose” of the Sentencing Reform Act, Rule 11(c)(1)(C), and the binding Guidelines policy statements in sections 6B1.2 and 1B1.10 relating to Rule 11(c)(1)(C) and § 3582(c)(2). Id. at 530. In a dissenting opinion by Chief Justice Roberts, four Justices took the contrary position that a sentence imposed under a Rule 11(c)(1)(C) agreement is never “based on” the Guidelines because the agreement itself serves as the foundation for the sentence imposed. Id. at 544 (Roberts, C.J., dissenting). Concurring only in the judgment, Justice Sotomayor staked out yet a third position. Justice Sotomayor argued that a sentence imposed under a Rule 11(c)(1)(C) agreement is generally based on the agreement, not the Guidelines, but rejected the dissent’s categorical bar to relief for all defendants sentenced under such agreements. Id. at 534 (Sotomayor, J., concurring in the judgment). Justice Sotomayor concluded that district courts have jurisdiction to consider a sentence reduction in at least two circumstances: when the Rule 11(c)(1)(C) agreement either 1) “call[s] for the defendant to be sentenced within a particular Guidelines sentencing range,” or 2) “make[s] clear that the basis for the specified term is a Guidelines sentencing range applicable to the offense to which the defendant pleaded guilty” and “that sentencing range is evident from the agreement itself.” Id. at 538–39. To say that Freeman divided the Court would be an understatement. Not only did the plurality and dissenting opinions take opposite positions, but both also strongly criticized Justice Sotomayor’s concurrence. The plurality warned that the “consequences of [the concurrence’s] UNITED STATES V. DAVIS 13 erroneous rule would be significant,” id. at 533 (plurality opinion), while the dissent complained that Justice Sotomayor’s approach would “foster confusion in an area in need of clarity,” id. at 550 (Roberts, C.J., dissenting). The dissenting opinion accurately stated that the plurality and concurrence “agree on very little except the judgment.” Id. at 544 (Roberts, C.J., dissenting). Thus, the 4-1-4 Freeman Court did not articulate a clear path forward for analysis of sentence-reduction requests by defendants sentenced under Rule 11(c)(1)(C) agreements. B. 1. In Marks v. United States, the Supreme Court explained that “[w]hen a fragmented Court decides a case and no single rationale explaining the result enjoys the assent of five Justices, the holding of the Court may be viewed as that position taken by those Members who concurred in the judgments on the narrowest grounds.” 430 U.S. 188, 193 (1977) (internal quotation marks and citation omitted). In the nearly forty years since Marks, lower courts have struggled to divine what the Supreme Court meant by “the narrowest grounds.” Indeed, the Court has acknowledged that the Marks inquiry at times has “baffled and divided the lower courts that have considered it,” Nichols v. United States, 511 U.S. 738, 746 (1994), and that the “test is more easily stated than applied.” Grutter v. Bollinger, 539 U.S. 306, 325 (2003) (quoting Nichols, 511 U.S. at 745–46). In the face of this confusion, two main approaches have emerged: one focusing on the reasoning of the various opinions and the other on the ultimate results. 14 UNITED STATES V. DAVIS The D.C. Circuit has offered a clear example of the first approach. In King v. Palmer, the court explained: Marks is workable—one opinion can be meaningfully regarded as “narrower” than another—only when one opinion is a logical subset of other, broader opinions. In essence, the narrowest opinion must represent a common denominator of the Court’s reasoning; it must embody a position implicitly approved by at least five Justices who support the judgment. 950 F.2d 771, 781 (D.C. Cir. 1991) (en banc). The D.C. Circuit reaffirmed this approach in Epps, describing Marks as applicable only when “the concurrence posits a narrow test to which the plurality must necessarily agree as a logical consequence of its own, broader position.” 707 F.3d at 348 (emphasis omitted) (quoting King, 950 F.2d at 782). The second approach looks to results rather than reasoning. It defines the narrowest ground as the rule that “would necessarily produce results with which a majority of the Justices from the controlling case would agree.” See, e.g., Planned Parenthood of Se. Pa. v. Casey, 947 F.2d 682, 694–97 (3d Cir. 1991) (finding that Justice O’Connor’s concurring opinions controlled the fractured decisions in Webster v. Reproductive Health Services, 492 U.S. 490 (1989), and Hodgson v. Minnesota, 497 U.S. 417 (1990), because a majority of justices in each case would have agreed with her result), aff’d in part, rev’d in part, 505 U.S. 833 (1992). UNITED STATES V. DAVIS 15 2. Our cases interpreting Marks have not been a model of clarity. On one occasion, we cited the “results” language described above. See United States v. Williams, 435 F.3d 1148, 1157 n.9 (9th Cir. 2006) (explaining that a concurrence is controlling under Marks if it “would affect a narrower range of cases than that of the plurality”). Nonetheless, in Williams and other decisions applying Marks to a fractured Supreme Court decision, we analyzed whether the reasoning of a narrower opinion fit entirely into the circle drawn by a broader opinion in order to derive a rule. Our most recent decision to address Marks explicitly employed the “reasoning” approach. Lair v. Bullock, 697 F.3d 1200 (9th Cir. 2012). In Lair, we approvingly cited King and held that the Marks standard applies only “where one opinion can be meaningfully regarded as narrower than another and can represent a common denominator of the Court’s reasoning.” Id. at 1205 (quoting United States v. Rodriguez–Preciado, 399 F.3d 1118, 1140 (9th Cir.), amended by 416 F.3d 939 (9th Cir. 2005)). Unless “the narrowest opinion is actually the logical subset of other broader opinions, . . . the only binding aspect of a splintered decision is its specific result.” Id. (internal quotation marks and citation omitted). To foster clarity, we explicitly adopt the reasoning-based approach to applying Marks. This approach is not only consistent with our most recent caselaw, see Lair, 697 F.3d 1200, but also makes the most sense.7 A fractured Supreme 7 This approach is not “fundamentally inconsistent with Marks itself.” Dissent at 41, 42 n.7. Marks never defined the “narrowest grounds,” and the dissent identifies no subsequent Supreme Court case that has offered an explanation or clarification of Marks as requiring an unwavering focus 16 UNITED STATES V. DAVIS Court decision should only bind the federal courts of appeal when a majority of the Justices agree upon a single underlying rationale and one opinion can reasonably be described as a logical subset of the other. When no single rationale commands a majority of the Court, only the specific result is binding on lower federal courts. on results. The difficult task of interpreting Marks has been left to the courts of appeal. See, e.g., King, 950 F.2d at 781; Lair, 697 F.3d at 1205; United States v. Johnson, 467 F.3d 56, 62–64 (1st Cir. 2006). More importantly, Marks cannot be viewed in isolation. In subsequent cases interpreting fractured Supreme Court decisions, the Court has frequently focused on reasoning, rather than results. Indeed, the dissent recognizes as much when it argues that Marks requires consideration of dissenting opinions. Dissent at 52–57, 52–53 n.9 (describing the various opinions in the fractured Tidewater National Mutual Insurance Co. v. Tidewater Transfer Co., 377 U.S. 582 (1949), decision). As the dissent states, “[s]ince Tidewater, courts have universally accepted that Congress may not expand the scope of subject- matter jurisdiction conferred by Article III through passage of a Congressional Act,” even though that rule could only be derived by combining the “views” or “rationale[s]” of Tidewater’s concurrence and dissent. Dissent at 52–54; 52–53 n.9. Thus, even the dissent acknowledges that the Supreme Court and lower courts have employed a reasoning-based approach to analyzing prior fractured Supreme Court decisions. A results approach cannot explain the governing rule that emerged from Tidewater. Similarly, the dissent mischaracterizes United States v. Jacobsen, 466 U.S. 109 (1984), as employing a results-based approach. Dissent 42 n.7. In our view, the rule that “the legality of the government search must be tested by the scope of the antecedent private search,” id. at 115–16, represents a common reasoning shared by the dissenting and plurality opinions, not a result or outcome on which they agree. The dissent’s own cases therefore contradict its assertion that a reasoning-based approach is an “invention” of the Ninth and D.C. Circuits. Dissent at 34–35, 44–45. UNITED STATES V. DAVIS 17 III. Applying Marks, as clarified above, to Freeman, we overrule our holding in Austin that Justice Sotomayor’s concurrence controls.8 Instead, we adopt the analysis of the D.C. Circuit in Epps that there was no common denominator in Freeman “because the plurality and concurring opinions do not share common reasoning whereby one analysis is a logical subset of the other.” Epps, 707 F.3d at 350 (internal quotation marks and citation omitted). 8 Following the adoption of the Fair Sentencing Act, we have issued, in addition to Austin, two opinions that discuss sentence modifications under § 3582(c)(2) and that bear mentioning here. First, in United States v. Bride, we held that a defendant who was sentenced under a Rule 11(c)(1)(C) agreement could not seek a sentence reduction because his sentence was not “based on a sentencing range that had been subsequently lowered by the Sentencing Commission.” 581 F.3d 888, 889 (9th Cir. 2009). We explicitly stated, however, that we did not “reach the issue of whether § 3582(c)(2) relief is necessarily precluded when the district court imposes a sentence pursuant to a Rule 11(c)(1)(C) plea agreement.” Id. at 891 n.5. Austin distinguished Bride on this basis. Austin, 676 F.3d at 927 n.1. Nonetheless, to the extent Bride conflicts with our opinion today, it is overruled. Second, in United States v. Pleasant, we cited Austin for the proposition that Justice Sotomayor’s Freeman concurrence controls, and held that the defendant was eligible for a sentence reduction because his Rule 11(c)(1)(C) plea agreement fell under one of her two exceptions. 704 F.3d 808, 811 (9th Cir. 2013). Although the thrust of the Pleasant opinion focused on the separate question of whether a sentence reduction would be “consistent” with the Guidelines, we determined that Pleasant satisfied the threshold eligibility determination. Id. at 811–12. As with Bride, to the extent Pleasant’s interpretation of Freeman is inconsistent with this opinion, it is overruled. 18 UNITED STATES V. DAVIS Justice Sotomayor’s concurrence cannot reasonably be described as a logical subset of Justice Kennedy’s plurality opinion. The Freeman plurality explicitly rejected the concurrence’s reasoning, in particular its underlying premise that a sentence imposed under a Rule 11(c)(1)(C) agreement is “based on” the parties’ agreement, not the Guidelines. 564 U.S. at 529 (plurality opinion); Id. at 535–36 (Sotomayor, J., concurring in the judgment). Even in setting out the circumstances in which she would find a defendant sentenced under a Rule 11(c)(1)(C) agreement eligible for relief, Justice Sotomayor focused on the role the parties’ Guidelines calculations play in crafting a Rule 11(c)(1)(C) agreement. Id. at 538–39. By contrast, the plurality focuses on the role of the judge’s Guidelines calculations in deciding whether to accept or reject the agreement. Id. at 529 (plurality opinion). This fundamental divergence in reasoning is enough to demonstrate that Justice Sotomayor’s rationale is not controlling Supreme Court law. Although in Freeman these divergent approaches led to the same result, the D.C. Circuit properly recognized that “the set of cases where the defendant prevails under the concurrence is not always nestled within the set of cases where the defendant prevails under the plurality . . . .” Epps, 707 F.3d at 351. Two examples from Epps are instructive.9 First consider 9 The dissent criticizes these examples, dissent at 47–49, but its analysis is oversimplified in suggesting that the plurality would always allow a sentence modification in a Rule 11(c)(1)(C) agreement. If the plurality intended such a rule, it could easily have explicitly said so. Instead, Justice Kennedy wrote “[e]ven when a defendant enters into an 11(c)(1)(C) agreement, the judge’s decision to accept the plea and impose the recommended sentence is likely to be based on the Guidelines.” Freeman, 564 U.S. at 534 (plurality opinion) (emphasis added). Likely is not the same as always. Similarly, the plurality wrote that a “district UNITED STATES V. DAVIS 19 the following scenario: [T]he parties may state in the plea agreement that a particular range applies and agree to a sentence at the bottom of that range, but the district court may not agree that the range determined by the parties applies, finding for example that the career offender range is applicable instead, but notwithstanding this finding accept the plea because it is to a term that is acceptable to the court for reasons unrelated to the guideline range determined by the parties. Id. at 350 n.8. Justice Sotomayor would allow a sentence reduction in this example because the agreement explicitly “call[s] for the defendant to be sentenced within a particular Guidelines sentencing range.” Freeman, 564 U.S. at 538 (Sotomayor, J., concurring in the judgment). The plurality, on the other hand, “would find [Freeman] ineligible because the range that the parties agreed to played no role in the judge’s decision to impose a sentence may therefore be based on the Guidelines even if the defendant agrees to plead guilty under Rule 11(c)(1)(C).” Id. at 526 (emphasis added). That is different than saying that a Rule 11(c)(1)(C) sentence is always based on the Guidelines. Nor did Justice Kennedy explicitly reject the idea “that his approach would limit relief to only a ‘subset of defendants,”’ as the dissent claims. Dissent at 49. Rather, in the passage the dissent cites, the plurality rejected Justice Sotomoyor’s approach because of the arbitrariness of allowing sentence reductions only for those defendants whose plea agreements refer to the Guidelines. Freeman, 564 U.S. at 532–33 (plurality opinion). Justice Kennedy never said he disagreed with Justice Sotomayor because her approach would “fail to permit resentencing in all cases.” Dissent at 49. 20 UNITED STATES V. DAVIS court’s determination that this was an appropriate sentence, despite the fact that the court imposed the agreed-upon term of imprisonment.” Epps, 707 F.3d at 350 n.8. Thus, the plurality opinion is actually the narrower one in certain respects. A second example produces a similar result: The sentencing court . . . might consider and reject the guideline range used by the parties, not because the court finds that a different guidelines range (such as the career offender range) applies, but because, having considered the applicable guidelines range, the court rejects it as a matter of policy and selects its sentence without regard to it. Id. Here again, if the court decides “for reasons unrelated to the guidelines range to impose the sentence the parties agreed upon,” the defendant would be eligible for a reduction under Justice Sotomayor’s approach but not under the plurality’s. Id. These examples make clear that the plurality and concurring opinions cannot be explained by a diagram in which a circle representing the reasoning of Justice Sotomayor’s opinion sits neatly within a circle representing the reasoning of the plurality opinion. Because both opinions would allow sentence reductions in situations where the other UNITED STATES V. DAVIS 21 would not, Justice Sotomayor’s concurrence is not the “narrowest grounds” envisioned by Marks.10 We recognize that, with the exception of the D.C. Circuit, every other circuit that has considered the issue has adopted Justice Sotomayor’s concurrence as the controlling opinion in Freeman. But we do not find those opinions convincing. Most engage with Marks only superficially, quoting its language with no analysis. See, e.g., United States v. Graham, 704 F.3d 1275, 1278 (10th Cir. 2013); United States v. Browne, 698 F.3d 1042, 1045–46 (8th Cir. 2012); United States v. Dixon, 687 F.3d 356, 359–60 (7th Cir. 2012); United States v. Smith, 658 F.3d 608, 611 (6th Cir. 2011).11 Given their lack of meaningful analysis, these opinions lack persuasive force. Those few cases that do discuss how Marks should apply to Freeman mistakenly conclude that although the “gap between the plurality and the concurrence is wide, [ ] it is still possible to tease out a common denominator.” United States 10 We emphasize that this results-oriented approach is used only to highlight the lack of a shared reasoning between Freeman’s plurality and concurring opinions. Our primary focus remains on the text of the two opinions, rather than on their application to hypothetical cases. 11 For example, the Tenth Circuit quoted the Marks “narrowest grounds” test and then summarily agreed with the district court that “[a]pplying this rule,” Justice Sotomayor’s concurrence controls. Graham, 704 F.3d at 1278. Similarly, the Sixth Circuit concluded that “Justice Sotomayor’s opinion is the narrowest ground for the Court’s decision and thus represents the Court’s holding in Freeman” without citing Marks. Smith, 658 F.3d at 611. The Second Circuit has considered the issue only in an unpublished summary order. United States v. White, 429 F. App’x 43, 47 (2d Cir. 2011) (finding that Justice Sotomayor’s concurrence is controlling). 22 UNITED STATES V. DAVIS v. Rivera-Martinez, 665 F.3d 344, 348 (1st Cir. 2011); see also United States v. Thompson, 682 F.3d 285, 289–90 (3d Cir. 2012). Not so. As the examples above demonstrate, there are some circumstances where defendants would be eligible for relief under Justice Sotomayor’s approach but not under the plurality’s. We therefore cannot agree with the First Circuit’s assertion that the “plurality would surely agree that in every case in which a defendant’s C-type plea agreement satisfies the criteria for Justice Sotomayor’s exception . . . the sentencing judge’s decision to accept that sentence is based on the guidelines.” Rivera-Martinez, 665 F.3d at 348. A more nuanced reading of both opinions leads us to conclude that “there is no controlling opinion in Freeman because the plurality and concurring opinions do not share common reasoning whereby one analysis is a logical subset of the other.” Epps, 707 F.3d at 350 (internal quotation marks and citation omitted). Marks instructs us to consider the opinions only of “those Members who concurred in the judgments on the narrowest grounds” when deriving a rule from a fractured Supreme Court decision. Marks, 430 U.S. at 193 (quoting Gregg v. Georgia, 428 U.S. 153, 169 n.15 (1976) (opinion of Stewart, Powell, & Stevens, JJ.) (emphasis added)). Nonetheless, we acknowledge that the Supreme Court and our sister circuits have considered dissenting opinions when interpreting fragmented Supreme Court decisions. See, e.g., United States v. Jacobsen, 466 U.S. 109, 115–17 (1984) (relying on a dissenting opinion to derive the rule in Walter v. United States, 447 U.S. 649 (1980)); Moses H. Cone Mem’l Hosp. v. Mercury Constr. Corp., 460 U.S. 1, 16–17 (1983) (holding that Will v. Calvert Fire Insurance did not overrule the “Colorado River test” because Will’s four dissenting Justices agreed with the concurring opinion that the test remained in UNITED STATES V. DAVIS 23 effect); United States v. Donovan, 661 F.3d 174, 182 (3d Cir. 2011) (explaining that the court “looked to the votes of dissenting Justices if they, combined with votes from plurality or concurring opinions, establish a majority view on the relevant issue”). Here, we assume but do not decide that dissenting opinions may be considered in a Marks analysis.12 Considering Chief Justice Roberts’s dissent would not change our conclusion because we cannot derive any common denominator by combining Freeman’s dissenting opinion with either the plurality or concurring opinion. First, no rule can be derived from the Freeman plurality and dissenting opinions, as neither is a “logical subset” of the other. Indeed, the plurality holding that “[e]ven when a defendant enters into an 11(c)(1)(C) agreement, the judge’s decision to accept the plea and impose the recommended sentence is likely to be based on the Guidelines,” Freeman, 564 U.S. at 534 (plurality opinion) is diametrically opposed to the dissent’s position that a sentence imposed under a Rule 11(c)(1)(C) agreement is never “based on” the Guidelines because the agreement itself serves as the foundation for the sentence imposed, id. at 544–45 (Roberts, C.J., dissenting). In practical terms, this divergence means the dissent would categorically find all defendants sentenced under a Rule 11(c)(1)(C) agreement ineligible for a sentence reduction, while the plurality would permit a reduction in most cases. Thus, the plurality and dissent “do not share common reasoning whereby one analysis is a logical subset of the 12 We note that in King, the D.C. Circuit explicitly stated that it was not “free to combine a dissent with a concurrence to form a Marks majority.” King, 950 F.2d at 783. We emphasize here, however, that we do not decide that issue. 24 UNITED STATES V. DAVIS other.” Epps, 707 F.3d at 350 (internal quotation marks and citation omitted). Second, Justice Sotomayor’s concurring opinion is not a logical subset of the dissenting opinion, or vice versa. Again, the dissent would categorically bar defendants sentenced under Rule 11(c)(1)(C) agreements from seeking relief under § 3582(c)(2). Freeman, 564 U.S. at 544–45 (Roberts, C.J., dissenting). Justice Sotomayor, by contrast, would permit sentence reductions in limited circumstances: if the Rule 11(c)(1)(C) agreement either 1) “call[s] for the defendant to be sentenced within a particular Guidelines sentencing range,” or 2) “make[s] clear that the basis for the specified term is a Guidelines sentencing range applicable to the offense to which the defendant pleaded guilty” and “that sentencing range is evident from the agreement itself.” Id. at 538–39 (Sotomayor, J., concurring in the judgment). Despite the dissent’s arguments to the contrary, these approaches cannot be reconciled because Justice Sotomayor explicitly “reject[ed] the categorical rule advanced by the Government and endorsed by the dissent, which artificially divorces a [Rule 11(c)(1)(C)] agreement from its express terms.” Id. at 539. Indeed, in evaluating Freeman’s case, Justice Sotomayor stated that “contrary to the dissent’s curious suggestion that ‘there is no way of knowing what th[e] sentence was based on,’” the basis for Freeman’s sentence was evident from the Rule 11(c)(1)(C) agreement itself. Id. at 542–43. The Freeman dissent is similarly critical of Justice Sotomayor, describing her view that certain Rule 11(c)(1)(C) defendants are eligible for relief as just “as mistaken as the position of the plurality—and basically for the same reasons.” Id. at 545 (Roberts, C.J., dissenting). Chief Justice Roberts UNITED STATES V. DAVIS 25 characterized her approach as “head-scratching,” id. at 547, and likely to “foster confusion in an area in need of clarity,” id. at 550. Neither opinion lays out the “common denominator of the Court’s reasoning” because Justice Sotomayor and the dissenters disagree on the fundamental question of whether a defendant sentenced under a Rule 11(c)(1)(C) agreement can ever be eligible for a sentence reduction under § 3582(c)(2). Simply put, no combination of Freeman’s dissenting and concurring opinions yields a binding rule that we must follow. IV. A. Given that no opinion in Freeman controls, we consider which of the rationales set forth in the varying opinions is most persuasive. Epps, 707 F.3d at 351. In so doing, we are restricted only by the ultimate result in Freeman: that defendants sentenced under Rule 11(c)(1)(C) agreements are not categorically barred from seeking a sentence reduction under § 3582(c)(2). We join the D.C. Circuit and adopt the plurality’s rule: “Even when a defendant enters into an 11(c)(1)(C) agreement, the judge’s decision to accept the plea and impose the recommended sentence is likely to be based on the Guidelines; and when it is, the defendant should be eligible to seek § 3582(c)(2) relief.” Freeman, 564 U.S. at 534 (plurality opinion). As the plurality explained, three critical sources support this approach. First, “[f]ederal sentencing law requires the district judge” to impose sentences that comply with “the purposes of federal sentencing, in light of the Guidelines and other § 3553(a) factors.” Id. at 529. Thus, by statute, a 26 UNITED STATES V. DAVIS sentencing judge’s discretion is always framed by the Guidelines. Second, Justice Kennedy looked to the district court’s authority under Rule 11(c)(1)(C). Although the Rule “permits the defendant and the prosecutor to agree on a specific sentence,” it preserves “the district court’s independent obligation to exercise its discretion” and review the proposed sentence. Id. Because judges “use the Guidelines range as the starting point,” they serve in a “real sense [as] a basis for the sentence,” “[e]ven where the judge varies from the recommended range.” Id. Third, the Guidelines policy statements that apply to Rule 11(c)(1)(C) plea agreements and § 3582(c)(2) motions support the plurality’s approach. Once a district court accepts a Rule 11(c)(1)(C) plea agreement, the parties’ recommended sentence is binding on the court. As the Freeman plurality noted, however, the applicable Guidelines policy statement “forbids the district judge to accept an 11(c)(1)(C) agreement without first evaluating the recommended sentence” under the Guidelines. Id.; USSG § 6B1.2(c). Indeed, as the plurality further noted, the commentary to the policy statement instructs a sentencing court to accept the recommended sentence only if it is an appropriate sentence within the applicable Guidelines range or “departs . . . for justifiable reasons.” Freeman, 564 U.S. at 529 (plurality opinion). The Guidelines policy statement that applies to § 3582(c)(2) motions likewise supports the plurality opinion. See USSG § 1B1.10(b)(1). As the plurality explained, § 1B1.10(b)(1) directs the district judge “in modifying a sentence to substitute only the retroactive amendment and then leave all original Guidelines determinations in place.” Freeman, 564 U.S. at 530 (plurality opinion). The goal of the sentence UNITED STATES V. DAVIS 27 modification is to “isolate” the effect of the amended Guideline while leaving undisturbed the other factors that determined the sentence imposed. Id.13 This suggests that a defendant should be eligible for a sentence reduction when one factor in a defendant’s sentence was a “since-rejected Guideline.” Id. Not only does the plurality approach best conform with these relevant sources, but a “contrary focus on the parties’ intentions would contribute to the unwarranted disparity that the [Sentencing Reform Act] was designed to reduce.” Epps, 707 F.3d at 351. In reducing the crack cocaine sentencing range, Congress and the Sentencing Commission sought to address “the urgent and compelling problem of crack-cocaine sentences.” Freeman, 564 U.S. at 533 (plurality opinion) (internal quotation marks omitted); supra at I.B. “Section 3582(c)(2) empowers district judges to correct sentences that depend on frameworks,” like the one for crack cocaine, “that later prove unjustified.” Freeman, 564 U.S. at 526 (plurality opinion). Justice Sotomayor’s approach would “extend the benefit of the Commission’s judgment only to an arbitrary subset of defendants whose agreed sentences were accepted in light of a since-rejected Guidelines range based on whether their plea agreements refer to the Guidelines.” Id. at 533–34. 13 By preserving all Guidelines calculations other than the one that was retroactively reduced, the policy statements in section 1B1.10(b) substantially limit district court discretion when ruling on § 3582(c)(2) motions. Freeman, 454 U.S. at 531–32 (plurality opinion). Further, district courts cannot vary below the Guidelines in a § 3582(c)(2) proceeding, as they can in an initial sentencing proceeding. Id.; USSG § 1B1.10(b)(2)(A). And, given the availability of appellate review, any concern that the plurality’s approach will “upset the bargain struck between prosecutor and defendant” is overstated. Freeman, 564 U.S. at 531 (plurality opinion). 28 UNITED STATES V. DAVIS Thus, adoption of the concurring opinion would “undercut a systemic solution” to a “systemic injustice.” Id. at 534. For all these reasons, we adopt the approach of the Freeman plurality opinion. B. Applying the plurality’s approach, we hold that Davis is eligible for relief under § 3582(c)(2) because the district court’s “decision to accept the plea and impose the recommended sentence” was “based on the Guidelines.” Freeman, 564 U.S. at 534 (plurality opinion). Davis’s Rule 11(c)(1)(C) plea agreement was clearly rooted in the Guidelines. First, it required the district judge to “determine Defendant’s applicable Sentencing Guidelines range at the time of the sentencing.” Second, the agreement stated that the amount of crack cocaine for which Davis admitted direct responsibility would yield a base offense level of 34 under Guidelines § 2D1.1(c)(3). Third, Davis’s agreement explained that he qualified for a Guidelines increase under § 2D1.2 for proximity to a school zone and a Guidelines reduction under § 3E1.1 for acceptance of responsibility. The district judge’s decision to reimpose the eighteen- year sentence was also based on the Guidelines. During the resentencing hearing, the district court recalculated Davis’s total offense level at 36 and a Guidelines range of 188 to 235 months. Then, reflecting on all the evidence presented, the court determined that the original eighteen-year sentence—which, at 216 months, fell within the calculated range—was “fair and reasonable” under the Guidelines. Taken together, the text of Davis’s plea agreement and the judge’s statements during the sentencing hearing leave no UNITED STATES V. DAVIS 29 doubt that the sentence imposed was “based on” the Guidelines. Thus, Davis is eligible for a sentence reduction and all that remains for the district court is to make the discretionary determination whether Davis should actually receive a reduction under § 3582(c)(2) and the Guidelines’ related policy statements in section 1B1.10.14 Conclusion In sum, when applying Marks to a fractured Supreme Court decision, we look to those opinions that concurred in the judgment and determine whether one of those opinions sets forth a rationale that is the logical subset of other, broader opinions. When, however, no “common denominator of the Court’s reasoning” exists, we are bound only by the “specific result.” Applying that framework to Freeman, we conclude that, contrary to our prior decision in Austin, Justice Sotomayor’s concurrence is not the logical subset of the plurality opinion. Nor can we extract a shared reasoning by including the dissent in our analysis. Thus, we overrule Austin and adopt the Freeman plurality approach as the most persuasive means of analyzing sentence reductions in the context of Rule 11(c)(1)(C) plea agreements. Accordingly, we reverse the district court’s determination that Davis is not eligible for a sentence reduction and remand for reconsideration of whether 14 We emphasize that our decision merely removes the jurisdictional hurdle that led the district court to deny Davis a resentencing hearing. On remand, “[i]f the district court, based on its experience and informed judgement, concludes the [Rule 11(c)(1)(C)] agreement led to a more lenient sentence than would otherwise have been imposed, it can deny the motion, for the statute permits but does not require the court to reduce a sentence.” Freeman, 564 U.S. at 532 (plurality opinion). 30 UNITED STATES V. DAVIS Davis should receive a sentence reduction under § 3582(c)(2) and the Guidelines’ related policy statements. REVERSED and REMANDED. CHRISTEN, Circuit Judge, joined by THOMAS, Chief Judge, and TALLMAN, NGUYEN, and HURWITZ, Circuit Judges, concurring: Freeman v. United States, 564 U.S. 522 (2011), addressed an issue of grave importance to Davis and to countless other prisoners in his position. The opinion issued today corrects an error in our circuit’s interpretation of Freeman, but it also represents a missed opportunity to straighten out our circuit’s inconsistent applications of Marks v. United States, 430 U.S. 188 (1977). Marks specifically directs lower courts how to interpret splintered Supreme Court decisions. Its rule tends to crop up in the most contentious cases where, as here, the stakes are significant. Freeman is important, but Marks has even broader application to the wide spectrum of issues we decide. I join in the court’s holding—as far as it goes—but it is regrettable that our court articulates an incomplete interpretation of Marks. Leaving this work unfinished will surely result in continued uneven application of Marks within our circuit. The rule announced in Marks appears simple at first glance but it has proven to be confounding. See Grutter v. Bollinger, 539 U.S. 306, 325 (2003) (noting Marks has “baffled and divided the lower courts that have considered it” UNITED STATES V. DAVIS 31 (quoting Nichols v. United States, 511 U.S. 738, 746 (1994))). The opinion issued today untangles part of the problem because it decisively adopts a reasoning-based approach to determine when splintered decisions produce binding precedent. Under this approach, I agree that no binding rule emerges from Freeman. I also agree that Justice Kennedy’s plurality opinion is the best framework for analyzing motions for sentence reduction in the context of Rule 11(c)(1)(C) plea agreements. Unfortunately, we leave unanswered whether our court will take into account dissenting opinions when applying Marks. I join the majority because its holding is entirely consistent with Marks: “[W]e look to those opinions that concurred in the judgment and determine whether one of those opinions sets forth a rationale that is the logical subset of other, broader opinions.” But I disagree with the majority’s assumption that we might be free to take dissenting opinions into account in future Marks analyses. Marks leaves some questions unanswered, but it plainly limits our review to the opinions of “those Members [of the Court] who concurred in the judgments.” Marks, 430 U.S. at 193 (emphasis added). Because I do not see that this language leaves any room for our court to consider dissenting opinions, I would go further than the majority does and expressly state that dissents play no role in a Marks analysis. This is not to say that dissents serve no purpose. They can and should be read to provide context and a deeper understanding of the Court’s decisions, but they do not inform our analysis of what binding rule, if any, emerges from a fractured decision. The dissent points to National Mutual Insurance Co. v. Tidewater Transfer Co., 337 U.S. 582 (1949), as support for its view that dissenting opinions should be considered. 32 UNITED STATES V. DAVIS Tidewater, of course, says nothing about how to interpret fractured Supreme Court decisions, though it was a fractured decision itself. In Tidewater, two concurring justices and four dissenting justices relied on the rule that Congress lacks authority to expand federal court subject matter jurisdiction beyond that provided in Article III. See id. at 604–46. Our dissenting colleague is correct that courts have universally accepted this rule, but doing so does not require looking to Tidewater’s dissenting opinions. Indeed, as recognized in Verlinden B.V. v. Central Bank of Nigeria, 461 U.S. 480 (1983), this rule pre-dates Tidewater by a long shot. See id. at 491 (“This Court’s cases firmly establish that Congress may not expand the jurisdiction of the federal courts beyond the bounds established by the Constitution.” (citing Hodgson v. Bowerbank, 3 L. Ed. 108 (1809); Kline v. Burke Constr. Co., 260 U.S. 226, 234 (1923))); see also Seminole Tribe of Fla. v. Florida, 517 U.S. 44, 65 (1996) (describing as “fundamental” that “Congress could not expand the jurisdiction of the federal courts beyond the bounds of Article III” (citing Marbury v. Madison, 2 L. Ed. 60 (1803))). The Supreme Court at times looks to dissenting opinions when interpreting its own splintered decisions. See United States v. Jacobsen, 466 U.S. 109, 115 (1984); Moses H. Cone Mem’l Hosp. v. Mercury Const. Corp., 460 U.S. 1, 16–17 (1983). From this, the majority and dissent infer that we might be free to do the same. But the way the Supreme Court treats its own precedent says nothing about how lower courts must apply it. Marks, not Jacobsen or Moses H. Cone, is the Supreme Court authority that sets out the rule for lower courts UNITED STATES V. DAVIS 33 to follow.1 In my view, until the Supreme Court says otherwise, Marks precludes us from considering dissenting opinions. BEA, Circuit Judge, dissenting: From its very first sentence, the majority unjustifiably departs from not only our own, but also well-established Supreme Court precedent. We correctly and squarely resolved the questions presented by this case in United States v. Austin, 676 F.3d 924 (9th Cir. 2011) until it was overruled by today’s majority. In Austin, we considered whether a judge had jurisdiction to modify a prisoner’s sentence under 18 U.S.C. § 3582(c)(2) (permitting modification of a term of imprisonment where that term was “based on” a sentencing range which was later reduced) if the prisoner was sentenced pursuant to a Federal Rules of Criminal Procedure (“F.R.C.P.”) 11(c)(1)(C) plea agreement (a “Rule 11(c)(1)(C) agreement”) that did not expressly incorporate the since- 1 The dissent responds by pointing to the Supreme Court’s language in Moses H. Cone: “[T]he [Fourth Circuit] Court of Appeals correctly recognized that the four dissenting Justices and Justice BLACKMUN [in Will v. Calvert Fire Insurance Co., 437 U.S. 655 (1978)] formed a majority to require application of the Colorado River test.” Moses H. Cone, 480 U.S. at 17. This was merely a recognition that, in Will, there were not enough votes to undermine Colorado River Water Conservation District v. United States, 424 U.S. 800 (1976). Thus, the controlling rule the Fourth Circuit applied came from Colorado River, not Will. Moses H. Cone does not direct lower courts to look to dissenting opinions when divining a controlling rule from a fractured Supreme Court decision. 34 UNITED STATES V. DAVIS amended sentencing Guidelines.1 We noted that the Supreme Court had spoken to this very question in Freeman v. United States, 564 U.S. 522 (2011), where five Justices had voted to permit a prisoner to seek sentence modification, but where no single rationale had commanded a five-Justice majority. A four-Justice plurality of those five Justices would always permit a prisoner to seek sentence modification, under the rationale that a judge’s approval of the plea agreement is required to start with, and is necessarily “based on,” the since-amended Guidelines. Justice Sotomayor, writing only for herself, agreed that Freeman was entitled to seek resentencing, but only because his plea agreement had expressly incorporated the since-modified sentencing Guidelines. A four-Justice dissent penned by Chief Justice Roberts would find that Rule 11(c)(1)(C) agreements are always purely contractual in nature and therefore never “based on” the sentencing Guidelines. Given this 4-1-4 split, we held in Austin that we were bound (under Marks v. United States, 430 U.S. 188, 193 (1977)) to treat Justice Sotomayor’s concurrence as the “holding” of the Freeman Court because it was the “narrowest grounds” upon which to reach the disposition that commanded a majority of the Court. See Austin, 676 F.3d at 927–28. Our interpretation in Austin has garnered the support of eight out of the nine Circuits which have interpreted Freeman. See, e.g., United States v. Graham, 704 F.3d 1275, 1278 (10th Cir. 2013); United States v. Thompson, 714 F.3d 946, 949 (6th Cir. 2013); United States v. Browne, 698 F.3d 1042, 1045–47 (8th Cir. 2012); United States v. 1 A Rule 11(c)(1)(C) agreement may or may not specifically reference applicable sentencing guidelines as the basis for the government’s sentencing recommendation. See Fed. R. Crim. P. 11. UNITED STATES V. DAVIS 35 Weatherspoon, 696 F.3d 416, 422 (3rd Cir. 2012); United States v. Dixon, 687 F.3d 356, 359–60 (7th Cir. 2012); United States v. Austin, 676 F.3d 924, 927 (9th Cir. 2012); United States v. Rivera-Martinez, 665 F.3d 344, 345 (1st Cir. 2011); United States v. Brown, 653 F.3d 337 (4th Cir. 2011).2 The sole outlier circuit: the D.C. Circuit in United States v. Epps, 707 F.3d 337 (D.C. Cir. 2013). Thus, the Majority today rejects a widely accepted interpretation of Freeman in favor of a highly criticized, outlier approach, thus accentuating a Circuit split. I cannot subscribe to the Majority’s view. To start, the Majority’s “logical subset” requirement is an invention of the D.C. Circuit that finds no support in Marks or any other Supreme Court precedent. The Majority’s “logical subset” invention permits a concurring opinion to become the precedential decision of the Court if, and only if, its reasoning shares all points in common with another, broader opinion that also reaches the majority result. Of course, the concurring opinion may have fewer elements of the decision than does the plurality opinion, but it may not have any elements in conflict. See Maj. Op. at 15–16. This notion is an invention in our circuit’s jurisprudence, though the Majority tips its hat to the D.C. Circuit. See Maj. Op. at 5–6, 14. And even if there were a “logical subset” requirement as defined by the Majority, the Majority misreads Justice Kennedy’s plurality opinion to the extent it concludes that there are circumstances in which Justice Sotomayor would permit sentence modification but the Kennedy plurality 2 See also United States v. Banks, 770 F.3d 346, 351 (5th Cir. 2014) (citing to Justice Sotomayor’s concurrence in Freeman and suggesting that, were that case applicable, the Fifth Circuit would be bound by Justice Sotomayor’s concurrence). 36 UNITED STATES V. DAVIS would not. Finally, the Majority’s adoption of the Kennedy plurality’s approach violates stare decisis because five Justices in Freeman (a majority), all agreed that we look to the plea agreement itself to determine whether a plea was “based on” the since-modified sentencing Guidelines. Under cases like National Mutual Insurance Co. v. Tidewater Transfer Co., 377 U.S. 582 (1949), we are bound by holdings that garner the support of a “majority” of the nine Justices on the entire Court, even if that agreement derives in part from votes from the dissent. Thus, the Majority flouts not only Freeman, but also Supreme Court jurisprudence relating to the binding effect of splintered Supreme Court opinions, as well as this Court’s structural role as a federal intermediary court. I. Facts In 2005, Tyrone Davis pleaded guilty to possession with intent to distribute crack cocaine pursuant to a Rule 11(c)(1)(C) plea agreement. In Davis’s case, the plea agreement did not specifically mention any particular sentencing Guideline. Nor did the plea agreement itself calculate (or even contain sufficient facts with which to calculate) Davis’s Guidelines range. True, it contained some of the factors that would enable a Guidelines calculation. For example, the parties stipulated to a base offense level of 34. But the agreement failed to list a criminal history category or adjustment determinations—both of which are essential to calculate a sentencing range under the Guidelines. After successive appeals to this Court on grounds no longer relevant, the district court calculated a Guidelines range of 188–235 and approved the 216-month sentence in Davis’s UNITED STATES V. DAVIS 37 plea agreement.3 We affirmed. See United States v. Davis, 389 F. App’x 616 (9th Cir. 2010) (unpublished). Congress thereafter passed the Fair Sentencing Act of 2010, which increased the threshold amount of cocaine base necessary to trigger an enhanced Guidelines range. Pub L. 111-220, §2(a), 124 Stat. 2372. Under the new Guidelines, the amount of cocaine base that contributed to Davis’s convictions would produce a Guidelines range of only 97–121 months (after inserting the calculations made by the district court at Davis’s resentencing). Because Davis’ 216- month sentence now falls much above that range, Davis moved in September 2012 for resentencing under 18 U.S.C. § 3582(c)(2), relying on the amended Guidelines. Section 3582(c)(2) provides: 3 At Davis’s sentencing in May 2006, the district court calculated a Guidelines range of 235–293 months (later reduced, on remand, to 188–235 months, as described below), relying in part on its own determinations that Davis’s criminal history category was II and that Davis deserved a 4-level leadership enhancement for his particular role in the offenses. Although Davis’s stipulated sentence of 216 months (18 years) fell below the low end of the Guidelines range (235–293 months), the district court accepted the sentence. David appealed, arguing that the district court miscalculated the Guidelines range, and our panel reversed and remanded for resentencing. United States v. Davis, 312 F. App’x 909, 912–13 (9th Cir. 2009) (unpublished). On remand, the district court held an evidentiary hearing about Davis’s role in the offense, and then calculated a new (lower) Guidelines range of 188–235 months. The court then reimposed the same 216-month sentence as stipulated in the Rule 11(c)(1)(C) plea agreement. On a second appeal, we affirmed. See United States v. Davis, 389 F. App’x 616 (9th Cir. 2010) (unpublished). 38 UNITED STATES V. DAVIS In the case of a defendant who has been sentenced to a term of imprisonment based on a sentencing range that has subsequently been lowered by the Sentencing Commission . . . the court may reduce the term of imprisonment, after considering the factors set forth in section 3553(a) to the extent that they are applicable, if such a reduction is consistent with applicable policy statements issued by the Sentencing Commission. Id. (emphasis added). The district court properly denied the motion, ruling that Davis’s 216-month sentence was “based on” his plea agreement, not on the Guidelines range that had since been lowered. The district court relied on our decision in United States v. Austin, 676 F.3d 924, 927–28 (9th Cir. 2011), to hold that it lacked jurisdiction to review Davis’ sentence, which was adopted pursuant to a Rule 11(c)(1)(C) agreement that did not reference any Guidelines range, and therefore did not meet either of Justice Sotomayor’s exceptions in Freeman. The original panel affirmed, citing Austin as the controlling law of the circuit. United States v. Davis, No. 13- 30133, slip op. at 4–6.4 This case was successfully called en banc to reconsider our prior determination that, under the 4 The panel engaged in a straight-forward application of Justice Sotomayor’s binding concurrence in Freeman: First, the agreement did not provide that Davis be sentenced within a particular Guideline range. Id. at 6. Second, it did not expressly use a Guideline range that “was evident from the agreement itself” to arrive at the 216-month term of imprisonment. Id. Judge Berzon concurred in judgment, but urged us to overrule Austin as wrongly decided. She viewed the decision to have misapplied Marks to Freeman. Id. at 8–10 (Berzon, J., concurring). UNITED STATES V. DAVIS 39 methodology prescribed by Marks, Justice Sotomayor’s concurrence in Freeman constitutes the binding “holding” of that case. II. Legal Analysis A. In Marks v. United States, the Supreme Court made clear that even splintered determinations of our highest court are binding on lower federal courts: “When a fragmented Court decides a case and no single rationale explaining the result enjoys the assent of five Justices, ‘the holding of the Court may be viewed as that position taken by those Members who concurred in the judgment on the narrowest grounds . . . .’” Marks v. United States, 430 U.S. 188, 193 (1977) (reversing the Sixth Circuit’s determination that Memoirs v. Massachusetts, 383 U.S. 413 (1966), had no binding precedential effect because it was a plurality opinion). At issue in Marks was the precedential effect of Memoirs—an earlier, splintered Supreme Court opinion. In Memoirs, a three-Justice plurality had held that sexually explicit literature was constitutionally protected unless it met the three-part definition of obscenity set forth in Roth v. United States, 354 U.S. 476 (1957).5 See Marks, 430 U.S. at 193 (citing 5 The proper test, as enumerated by the three-Justice plurality, was “whether to the average person, applying contemporary community standards, the dominant theme of the material taken as a whole appeals to prurient interest.’ . . . . Under this definition . . . three elements must coalesce: it must be established that (a) the dominant theme of the material taken as a whole appeals to a prurient interest in sex; (b) the material is patently offensive because it affronts contemporary community standards relating to the description or representation of sexual matters; and (c) the material is utterly without redeeming social value.” A Book Named “John 40 UNITED STATES V. DAVIS Memoirs, 383 U.S. at 421). Justices Black and Douglas (both writing separately) concurred in Memoirs on the broader grounds that the First Amendment prohibits government censorship of any “obscene” material. See Memoirs, 383 U.S. at 421, 424–33. Finally, Justice Stewart concurred based on his somewhat different view that only “hard-core pornography” may constitutionally be suppressed. Id. at 421 (citing his dissenting opinion in Mishkin v. State of N.Y., 383 U.S. 502, 518 (1966)). In sum, six Justices agreed that the material at issue in Memoirs was protected by the First Amendment, but no five Justices agreed about the scope of First Amendment protection for sexually explicit material—nor about the proper reasoning to be employed to reach that result. In this circumstance, Marks explained, the three-Justice Memoirs plurality opinion, which applied the Roth tests, “constituted the holding of the Court and provided the governing standards,” because it was the “narrowest grounds” for finding First Amendment protection. Marks, 430 U.S. at 193–94. The Majority today rejects the only application of Marks to Freeman that is consistent with stare decisis in favor of a widely-criticized approach endorsed by one panel in the D.C. Circuit in United States v. Epps, 707 F.3d 337, 351 (D.C. Cir. 2013).6 The Majority holds that Marks produces a Cleland’s Memoirs of a Woman of Pleasure” v. Attorney Gen. of Com. of Mass., 383 U.S. 413, 418 (1966). 6 In addition to being widely rejected by our sister circuits, Epps has been criticized even within the D.C. Circuit. As Judge Kavanaugh contended in United States v. Duvall, 740 F.3d 604 (D.C. Cir. 2013): Following Justice Sotomayor’s opinion with regard to the “based on” issue would produce results with which UNITED STATES V. DAVIS 41 “controlling” opinion only when the “narrowest grounds” in a splintered opinion is “represent[s] a common denominator of the Court’s reasoning,” meaning “the reasoning of a narrower opinion fit[s] entirely into the circle drawn by a broader opinion.” Maj. Op. at 15. The Majority’s adoption of a reasoning-based, “common denominator” approach is fundamentally inconsistent with Marks itself. In Memoirs, none of Justices Stewart, Black, or Douglas agreed with the rule enumerated by the Justice Brennan plurality (that the Court should apply the Roth test to determine whether speech is constitutionally protected). Thus, there was no “common denominator” in Memoirs, as that term is defined by the Majority today. Yet, Marks applied the three-Justice plurality opinion authored by Justice Brennan. It is true, of course, that in every circumstances in which the Brennan plurality would find speech protected under the Roth obscenity test, Justices Black and Douglas (who ascribe to the broader view that obscene speech is always constitutionally protected) would agree with the plurality’s result. But there is a distinction between a majority of the Supreme Court in Freeman would agree because—to put it in simple terms—“sometimes” is a middle ground between “always” and “never.” In other words, when Justice Sotomayor concludes that a plea agreement was based on the Guidelines, she would agree with the result reached under Justice Kennedy’s opinion for four Justices. When she concludes that a plea agreement was not based on the Guidelines, she would agree with the result reached under Chief Justice Roberts’s opinion for four Justices. But unlike every other court of appeals, Epps did not follow this commonsense approach to interpreting Freeman. Id. at 612 (emphasis added). 42 UNITED STATES V. DAVIS agreement with a result and agreement with the reasoning that leads one to adopt that result that appears to be lost on my colleagues in the Majority. Indeed, the Majority’s own reasoning supports only a votes-based reading of Marks: Justice Brennan’s plurality opinion was the “narrowest grounds” for the Court’s holding in Memoirs because it would always produce a result with which at least five Justices would agree.7 7 The Majority argues that their approach is not “fundamentally inconsistent with Marks itself.” Maj. Op. at 15–16, n.7. This, because I have not identified a subsequent Supreme Court case that has unequivocally stated that Marks requires an unwavering focus on results. But, as explained above, the Majority overlooks Marks itself. Under the rule the Majority advances today, we would be unable to derive a controlling rule from Memoirs, the earlier Supreme Court case with respect to which the Court in Marks was called upon to give binding effect. Yet that would be directly contrary to Marks’ holding that we can derive a controlling rule from Memoirs. By the same token, the Majority’s rule would preclude us from deriving a binding rule from Walter v. United States, 447 U.S. 649 (1980), where five Justices voted that the government’s warrantless seizure of contraband films and viewing of those films on a projector violated the defendant’s Fourth Amendment right. Id. at 652, 657, 660–62. Two Justices voted for this result on the grounds that the government’s act of viewing the films expanded the scope of a private party’s earlier search, which had consisted only of opening the package that contained the films. Id. at 657. Three Justices voted for this result for a completely different reason: that the government had exceeded its authority under the “plain view” doctrine. Id. at 660–62. Thus, neither the approach adopted by the plurality nor the approach advanced by the concurrence was a “logical subset” of the other; and under the rule the Majority announces today, we could discern no “controlling” rule in Walter. Yet such a conclusion is inconsistent with the Court’s holding, only three years later in Jacobsen, that Walter did set forth a controlling rule: the rule advanced by the two-Justice plurality. See United States v. UNITED STATES V. DAVIS 43 This reading is also consistent with Marks’ dictate that “the holding of the Court . . . [is] the position taken by those Members who concurred in the judgment on the narrowest grounds . . . .” Marks, 430 U.S. at 193 (emphasis added). Marks’ emphasis on the Court’s “judgment” demonstrates that it is the ultimate “vote” of five Justices that is important in determining the binding effect of a splintered Supreme Court opinion. That is, Marks requires us to find a “legal standard which, when applied, will necessarily produce results with which a majority of the Court from that case would agree.” United States v. Williams, 435 F.3d 1148, 1157 n.9 (9th Cir. 2006) (emphasis added). That is not to say, of course, that the respective rationales of a splintered Supreme Court decision are irrelevant. Consideration of competing rationales is necessary to determine which would consistently produce a result with which a majority of the Court would agree. But the Majority’s reasoning-based approach must be incorrect, because there was no common reasoning in Memoirs, yet the Marks Court was nonetheless able to derive from Memoirs a binding rule, which it applied in Marks. Jacobsen, 466 U.S. 109 (1984). The plurality’s view was the “controlling” rule of Walter, because four Justices in dissent had also voted for the plurality’s rule. Id. at 115–16 (“[A] majority” in Walter “agree[d]” that “the legality of the governmental search must be tested by the scope of the antecedent private search.”). Like in Marks, it was the vote of a majority of Justices that counted. I need not belabor the point, discussed more thoroughly infra, at pp. 52–56, here. Suffice it to say that, as a federal intermediary court, we are not free to adopt a reasoning-based approach to Marks—an approach plainly inconsistent with the facts and holdings of Marks and later Supreme Court cases interpreting splintered opinions—simply because the Court has not yet had occasion expressly to reject that reasoning-based approach. 44 UNITED STATES V. DAVIS The Majority’s adoption of a “logical subset” precondition to Marks applicability is plagued by the same logical fallacy. King v. Palmer, the D.C. Circuit case on which the Majority relied, justified its invention of a logical subset requirement on the grounds that: Marks is problematic[] [i]f applied in situations where the various opinions supporting the judgment are mutually exclusive [because] Marks will turn a single opinion that lacks majority support into national law. When eight of nine Justices do not subscribe to a given approach to a legal question, it surely cannot be proper to endow that approach with controlling force . . . .” King v. Palmer, 950 F.2d 771, 782 (D.C. Cir. 1991). Even accepting, arguendo, the King court’s premise that a concurring opinion should be given stare decisis effect only when it consistently produces a result with which a majority of the Court would agree, that, again, would support the adoption of a rule that it is each Justice’s vote, and not his reasoning, that counts under Marks. The King court’s conclusion that Marks “works” only when a majority of Justices “subscribe to a given approach to a legal question,” such that “one opinion supporting the judgment . . . fit[s] entirely within a broader circle drawn by the others,” id. at 782, simply does not follow from that court’s premise—or from the many Supreme Court precedents interpreting and applying Marks to splintered opinions over the last four UNITED STATES V. DAVIS 45 decades.8 Indeed, to require complete overlap between both the result and the reasoning of Justices in the majority before a binding rule can be discerned renders Marks a virtual nullity. Agreement as to both the Court’s reasoning and its result does not produce a concurring opinion—it produces a “join.” And even if a reasoning-based approach to Marks were not fundamentally incompatible with Marks itself, the idea that a court’s holding, adopted by a majority of judges, must have a rationale common throughout the majority is a novelty to any branch of our government, including the judiciary. Even our courts adhere to that most democratic of principles: as to how to decide this case, the majority rules. People, including legislators and judges, may vote for a result for a variety of different reasons. That is how coalitions are achieved and compromises made to reach results although the distinct motives and thinking which produced the majority’s result remain quite distinct. 8 The Supreme Court has never adopted a “logical subset” requirement in its numerous applications of Marks over the past four decades. The Supreme Court has on numerous occasions applied Marks to its own decisions. See, e.g., Glossip v. Gross, 135 S. Ct. 2726, 2738 n.2 (2015) (holding that a three-Justice plurality opinion constituted the “holding” of the Court in Baze v. Rees, 553 U.S. 35 (2008), because Justices Scalia and Thomas had concurred in the result reached by the plurality but on “broader” grounds); Panetti v. Quarterman, 551 U.S. 930, 949 (2007) (citing Marks and holding simply that “Justice Powell’s concurrence [in Ford v. Wainwright, 477 U.S. 399 (1986)], . . . offered a more limited holding. When there is no majority opinion, the narrower holding controls.”). In neither of these cases did the Supreme Court state that the controlling opinion must be a “logical subset” of the broader view that produces the same result—it has only reiterated that the controlling opinion is the one that relies on “narrower” grounds. 46 UNITED STATES V. DAVIS It is the result produced by majority vote that determines the stare decisis effect of the judgment. That is because whether the majority vote is produced by the adoption of one rationale or two, the rule of law made—the decision—is based on a rationale or rationales expected to remain the same and produce the same result in the next applicable case. After all, “stare decisis” means “to stand by things decided.” B. A simple application of Marks’ methodology to Freeman compels a finding that Justice Sotomayor’s concurrence is the “holding” of Freeman. See United States v. Austin, 676 F.3d 924, 927–28 (9th Cir. 2012). Five members of the Court agreed that Freeman—who had been sentenced pursuant to a Rule 11(c)(1)(C) agreement—was eligible for sentencing modification under 18 U.S.C. § 3582(c)(2), because his plea agreement had been “based on” a subsequently modified sentencing Guidelines range. Freeman v. U.S., 564 U.S. 522, 534, 544 (2011). Writing for a four-member plurality, Justice Kennedy reasoned that a plea agreement is “based on” applicable Guidelines whenever the sentencing judge at least consulted those guidelines before approving the proposed sentence—which, Justice Kennedy explained, the judge is statutorily “required” to do in “every case.” See Freeman, 564 U.S. at 529 (plurality opinion). Concurring in result, Justice Sotomayor, a former district court judge experienced in actual sentencing, reasoned that plea agreements are sometimes based on sentencing guidelines, but only when the agreement itself “expressly uses a Guidelines sentencing range applicable to the charged offense to establish the term of imprisonment,” or the sentencing range is otherwise “evident from the agreement UNITED STATES V. DAVIS 47 itself.” Id. at 534, 539 (Sotomayor, J., concurring) (emphasis added). Chief Justice Roberts, writing for the four dissenting Justices, “agree[d] with Justice Sotomayor that ‘the term of imprisonment imposed pursuant to a [Rule 11(c)(1)(C)] agreement is . . . based on the agreement itself.’” Id. at 544 (Roberts, C.J., dissenting) (emphasis added) (internal quotation marks omitted). However, the dissent would find that plea agreements are a matter of contract and thus never “based on” the sentencing Guidelines. Id. at 544–51 (Roberts, C.J., dissenting). Justice Sotomayor’s opinion is controlling because “‘sometimes’ is the middle ground between ‘always’ and ‘never.’” See United States v. Duvall, 740 F.3d 604, 612 (D.C. Cir. 2013) (Kavanaugh, J., concurring in the denial of rehearing en banc); see also supra, n.6. In circumstances in which Justice Sotomayor would permit reduction of a prior sentence, so too would the plurality (resulting in a five- Justice majority). Where Justice Sotomayor’s criterion are not met, she would find agreement in the four-Justice dissent that the prisoner’s sentence is not “based on” the Guidelines (which would also result in a five-Justice majority). Justice Sotomayor’s approach therefore constitutes the “narrowest grounds” for reaching a result that, in any circumstance, will be consistent with the result that a majority of the Supreme Court would reach under Freeman. Under Justice Sotomayor’s framework, Davis cannot seek resentencing, because his plea agreement does not meet either of her exceptions. It neither expressly cites, nor otherwise manifests that it is predicated upon, any particular Guidelines range. In fact, it omits several details (such as criminal history, and adjustments) necessary even to calculate a Guidelines range. Davis’s sentence is therefore not subject to 48 UNITED STATES V. DAVIS modification under § 3582(c)(2). The district court correctly determined that it lacked jurisdiction to resentence Davis, and the panel should affirm on that basis. The Majority rejects this straight-forward approach on the grounds that circumstances could arise in which Justice Sotomayor would find a plea “based on” sentencing guidelines, but the Kennedy plurality would not. The Majority posits two hypotheticals, both of which assume express agreement in a plea bargain that a particular sentencing range applies (such that Justice Sotomayor would find the plea agreement “based on” the sentencing Guidelines, and subject to § 3582(c)(2) resentencing). See Maj. Op. at 18–20. Both hypotheticals then posit that the “sentencing court . . . might consider and reject the guideline range used by the parties”—in one scenario because the judge believed another range should apply, and, in the other, for “policy” reasons. Id. The Majority suggests that in either of these circumstances, the Freeman plurality would not find the plea agreement “based on” the sentencing guidelines, and thus would not grant relief. The Majority is simply incorrect. The very fact that the sentencing judge in the Majority’s hypotheticals must reject the Guidelines range recommended by the parties necessarily presupposes that the judge’s analysis started with a consideration of the Guidelines range recommended in the plea agreement. Under the Kennedy plurality’s approach, this consideration, at the inception of the sentencing, is enough to entitle a defendant to seek resentencing—regardless of the judge’s ultimate reasons for approving the plea agreement. See Freeman, 564 U.S. at 529–30 (plurality opinion) (“[I]f the judge uses the sentencing range as the beginning point to explain the decision to deviate from it, then the Guidelines are UNITED STATES V. DAVIS 49 in a real sense a basis for the sentence.” (emphases added)). The Majority recognizes as much on page 26 of its opinion, wherein it quotes Justice Kennedy’s statement that “the applicable Guidelines policy statement ‘forbids the district judge to accept an 11(c)(1)(C) agreement without first evaluating the recommended sentence’ under the Guidelines.” Maj. Op. at 26 (quoting Freeman, 564 U.S. at 529, and USSG § 6B1.2(c)). The Majority criticizes my reading of Justice Kennedy’s plurality opinion—a reading adopted by an overwhelming majority of circuits—as “oversimplified.” In support of its more limited reading, the Majority relies solely on Justice Kennedy’s statement that a “recommended sentence is likely to be based on the Guidelines.” Maj. Op. at 18 n.9 (quoting Freeman, 564 U.S. at 534). But Justice Kennedy’s use of the word “likely” in one sentence cannot be read in isolation. In the immediately preceding paragraph, Justice Kennedy in fact rejects the notion—advanced by the Majority—that his approach would limit relief to only a “subset of defendants.” Freeman, 564 U.S. at 533–34 (“[When] [t]he Commission determine[s] that [the] Guidelines [are] flawed, and therefore that sentences that relied on them ought to be reexamined[,] [t]here is no good reason to extend the benefit of the Commission’s judgment only to an arbitrary subset of defendants whose sentences were accepted in light of a since- rejected Guidelines range based on whether their plea agreements refer to the Guidelines.”). Indeed, Justice Kennedy criticizes Justice Sotomayor’s approach because it would fail to permit resentencing in all cases: “Congress enacted § 3582(c)(2) to remedy systemic injustice, and the approach outlined in [Justice Sotomayor’s] opinion 50 UNITED STATES V. DAVIS concurring in the judgment would undercut a systemic solution.” Id. at 534 (emphasis added). Nothing about Justice Kennedy’s opinion suggests any exceptions. He notes that the “Guidelines require the district judge to give due consideration to the relevant sentencing range, even if the defendant and prosecutor recommend a specific sentence as a condition of the guilty plea.” Id. at 530 (emphases added). He further reasons that “[f]ederal sentencing law requires” the sentencing judge to look to the Guidelines as “a framework or starting point” in “every case.” Id. at 529. Thus, notwithstanding his use of the word “likely” in one sentence, Justice Kennedy’s opinion is most reasonably read as endorsing an approach under which a defendant may “always” seek resentencing on the basis of amended Guidelines. Certainly, for the reasons already stated above, Justice Kennedy would permit a defendant to seek resentencing in the examples given by the Majority. The Majority imbues far more meaning into Justice Kennedy’s single use of the word “likely” than the rest of Justice Kennedy’s plurality opinion can bear. It may be that Justice Kennedy simply did not want to speak in absolutes. That is, he declined to say, as a matter of empirical fact, that a judge always consults the sentencing Guidelines, because there is always the possibility that a judge could make a mistake or fail to follow the law. But one thing is for sure: Justice Kennedy does not even hint at a case in which the sentencing judge could lawfully start sentencing with any consideration other than the Guidelines, and the Majority has not suggested any either. See 18 U.S.C. § 3553 (directing that a “court, in determining the particular sentence to be imposed, shall consider . . . the sentencing range established for . . . the applicable category of offense committed by the UNITED STATES V. DAVIS 51 applicable category of defendant as set forth in the guidelines” (emphasis added)); see also Gall v. United States, 552 U.S. 38, 49 (2007) (“[A] district court should begin all sentencing proceedings by correctly calculating the applicable Guidelines range. As a matter of administration and to secure nationwide consistency, the Guidelines should be the starting point and the initial benchmark.” (citing Rita v. United States, 551 U.S. 338, 347–48 (2007))). Indeed, the failure of a sentencing judge to start the calculation of a sentence by considering the applicable sentencing Guidelines is in itself grounds for reversal for resentencing. See Gall, 552 U.S. at 51 (instructing that appellate courts must “first ensure that the district court committed no significant procedural error, such as failing to calculate . . . the Guidelines range”); United States v. Denton, 611 F.3d 646, 651 (9th Cir. 2010) (explaining that “[a] failure to calculate the correct advisory range constitutes procedural error” justifying reversal and remand for resentencing); United States v. Hammons, 558 F.3d 1100, 1106 (9th Cir. 2009) (holding that a sentencing court “committed plain error by failing to . . . calculate the app[licable] guideline range” and vacating and remanding for resentencing). The Majority’s contrary analysis appears to substitute the Freeman plurality’s requirement that a trial judge “consider” the Guidelines with its own innovation—that the trial judge must base his ultimate acceptance of the plea agreement on the Guidelines in order for a defendant to be entitled to seek resentencing. See Maj. Op. at 18–20. But the latter is not the test enumerated by Justice Kennedy in Freeman. Properly read, Justice Kennedy’s opinion would unquestionably permit resentencing in the hypotheticals offered by the Majority. See id. Thus, even if the Majority were correct that Marks applies 52 UNITED STATES V. DAVIS only when one opinion is a “logical subset” of another, that precondition would be met here. C. But even putting aside the “logical subset” issue, the Majority still cannot reach its result consistent with basic principles of stare decisis for the independent reason that we, as a federal intermediate court, are bound by holdings upon which five Justices of the Court agree—even if that agreement derives in part from dissenting Justices. The Supreme Court’s fragmented decision in National Mutual Insurance Co. v. Tidewater Transfer Co., 337 U.S. 582 (1949), is a famous illustration of this principle.9 Since 9 The question in Tidewater was whether Congress’s amendment to 28 U.S.C. § 1332 to permit citizens of the District of Columbia to be characterized as “citizens of a state” for purposes of diversity jurisdiction was constitutional under Article III. The problem in that case was that the Supreme Court had previously addressed that same question (albeit in the absence of a Congressional statute) and had interpreted Article III’s reference to “citizens of a state” as not encompassing citizens of the District of Columbia. See Hepburn & Dundas v. Ellzey, 6 U.S. (2 Cranch.) 44 (1805); see also U.S. Const. art. III, § 2, cl. 1 (“Federal courts will have jurisdiction over: . . . citizens of a state . . . .”). Nevertheless, by a vote of 5 to 4, the Tidewater Court upheld the constitutionality of a Congressional grant of diversity jurisdiction to citizens of Washington D.C., though no five justices agreed on a rationale. Writing for a three-justice plurality, Justice Jackson voted to uphold the statute under the rationale that Congress has the power to expand Article III by statute and thereby to confer subject-matter jurisdiction on bases not specified in Article III. See id. at 583–603. That is, Justice Jackson would not disturb Hepburn’s interpretation of Article III; he would simply hold that Congress may simply add a new basis for jurisdiction that does not exist in Article III. In a concurrence, Justice Rutledge, joined by Justice Murphy, strenuously disagreed that Congress UNITED STATES V. DAVIS 53 Tidewater, courts have universally accepted that Congress may not expand the scope of subject-matter jurisdiction conferred by Article III through passage of a Congressional Act.10 Yet this rule can be divined only by combining a two- had the power to expand Article III jurisdiction beyond the bases enumerated in the Constitution. See id. at 604–17. Nevertheless, Justice Rutledge would overrule Hepburn’s interpretation of Article III and would reinterpret Article III’s reference to “citizen[s] of a state” as including citizens of Washington D.C. See id. at 617–626. Justice Rutledge reasoned that “nothing but naked precedent, . . . and the prestige of [Justice] Marshall’s name, supports . . . [the] unjust and discriminatory exclusion of District citizens from the federal courts. All of the reasons of justice, convenience, and practicality . . . point to the conclusion that [citizens of the District of Columbia] should enter freely and fully as other citizens and even aliens do.” Id. at 617. The four remaining Justices would have declined to overrule Hepburn; but they—like Justice Rutledge—also vehemently rejected Justice Jackson’s suggestion that Congress had the power to create subject-matter jurisdiction not conferred by Article III. Thus, based on the rationale that the Court was bound by Justice Marshall’s interpretation of Article III in Hepburn, and that Congress lacked authority to expand Article III, the dissenting Justices would have found the statute unconstitutional. See id. at 626–46. In sum, six Justices agreed that Congress could not expand the scope of subject-matter jurisdiction of Article III courts beyond that provided by the Constitution. This view—which finds majority support only by combining the views of two concurring and four dissenting justices—is the governing rule. Tidewater therefore demonstrates that a “majority” view of the Court that binds us as intermediate courts may be comprised of dissenting and concurring justices, regardless of the quite obvious lack of any “logical subset” between the views expressed by the concurring and dissenting Justices. 10 See, e.g., Seminole Tribe of Florida v. Florida, 517 U.S. 44, 65 (1996) (describing as “fundamental” the notion “that Congress could not expand the jurisdiction of the federal courts beyond the bounds of Article III” and overruling Pennsylvania v. Union Gas Co., 491 U.S. 1 (1989) to the extent it suggested otherwise); Rosmer v. Pfizer Inc., 263 F.3d 110, 120 n.5 (4th Cir. 2001) (citing Tidewater for the proposition that “Congress cannot 54 UNITED STATES V. DAVIS Justice concurrence in Tidewater with a four-Justice dissent. See id. at 604–46. United States v. Jacobsen, 466 U.S. 109 (1984), provides a more recent example of the same rule. The question in Jacobsen was the government’s authority to conduct a warrantless search on the heels of a private search that identified potential contraband. The Jacobsen Court extracted the “controlling” legal standard from its prior precedent in Walter v. United States, 447 U.S. 649 (1980), by combining the opinion of the Walter Court (which garnered only two votes) with the opinion of four dissenting Justices, which it described as the “standard . . . adopted by the majority of the Court in Walter . . . .” Id. at 116–17 & n.12 (emphasis added). In Walter, a private party had opened a package containing films that, from the descriptions on the packaging, the private party concluded were contraband. Walter v. United States, 447 U.S. 649, 651–52 (1980). The government seized the films and, without obtaining a warrant, screened them from a projector. Id. at 652. Delivering the two-Justice opinion of the Court, Justice Stevens reasoned that the government had violated the defendant’s Fourth Amendment rights by actually watching films because the private party’s search had consisted only of opening the package that contained the films. Id. at 657 (The FBI’s subsequent confer jurisdiction on Article III courts by statute when Article III does not authorize that jurisdiction.”); Lo Duca v. United States, 93 F.3d 1100, 1108 (2d Cir. 1996) (explaining that “[i]n Tidewater, . . . six Justices reaffirmed the traditional view that federal courts are courts of limited jurisdiction whose judicial powers are bounded by Article III,” a notion that dates “as far back as Marbury v. Madison . . . .”). UNITED STATES V. DAVIS 55 screening of such films constituted an “expansion of the search that had been conducted previously by the private party.”); see also Jacobsen, 446 U.S. at 115–16 (quoting Walter, 447 U.S. at 657 (Opinion of Stevens, J., joined by Stewart, J.)). Three Justices in Walter concurred in the judgment on the grounds that the government had exceeded its authority under the “plain view” doctrine, but expressly rejected the notion that the scope of one’s Fourth Amendment right could be tethered to the scope of an antecedent private search. Walter, 447 U.S. at 660–62 (White, J., concurring). A four-Justice dissent agreed with Justice Stevens that the legality of a governmental search depended on the scope of the private party’s antecedent search, but would have found no constitutional violation because “the FBI’s subsequent viewing of the movies on a projector did not ‘change the nature of the search’ and [thus] was not an additional search subject to the warrant requirement.” Id. at 663–64 (Blackmum, J., dissenting); see also Jacobsen, 446 U.S. at 116. Presented with these competing views in Walter, the Jacobsen Court (in a six-Justice opinion of the Court) held that “a majority [in Walter] did agree on the appropriate analysis of a governmental search which follows on the heels of a private one. Two Justices [referring to Justices Stevens and Stewart] . . . . [and] [f]our additional Justices [referring to the dissent] were . . . of the view that the legality of the governmental search must be tested by the scope of the antecedent private search.” Jacobsen, 446 U.S. at 115–16. The Majority opinion here simply cannot be squared with the Court’s reading of Walter in Jacobsen. Jacobsen recognized that the rule adopted by the two-Justice plurality in Walter was the precedential holding of the Walter Court, because it garnered the approval of six Justices (a majority) of the 56 UNITED STATES V. DAVIS Court. This was so even though the three-Justice concurrence specifically rejected the plurality’s rationale, and thus neither the plurality opinion nor the concurrence was a “logical subset” of the other. Moses H. Cone Mem’l Hosp. v. Mercury Const. Corp., 460 U.S. 1 (1983) [hereinafter, “Memorial Hospital”] provides yet another example. There, the Court considered whether a lower court was bound to apply the Colorado River test,11 notwithstanding that a four-Justice plurality in Will v. Calvert Fire Insurance Co., 437 U.S. 655 (1978), had purported to overrule it. Id. at 17. The Memorial Hospital Court affirmed that the “Court of Appeals [had] correctly recognized that the four dissenting Justices and Justice Blackmum [who concurred in judgment in Will] formed a majority to require application of the Colorado River test.” Id. (emphases added). By holding that the Fourth Circuit had 11 Colorado River held that federal courts have discretion in “exceptional” circumstances to stay federal court proceedings pending the resolution of a parallel state court proceeding. See Colorado River Water Conservation Dist. v. United States, 424 U.S. 800, 813, 817–18 (1976) (recognizing, however, that “[a]bstention from the exercise of federal jurisdiction is the exception, not the rule”). The “Colorado River test” refers to the list of factors courts consider in determining whether to invoke this exceptional prudential abstention doctrine. These factors include: (a) the extent to which the federal legislation pursuant to which the federal suit is brought favors the state versus federal forum; (b) which forum offers the “greatest experience and expertise” in the particular subject-matter; (c) the “absence of any substantial progress in the federal- court litigation”; (d) the extent to which the suit involves questions or “rights governed by state law”; (e) “the geographical inconvenience of the federal forum”; (f) “the desirability of avoiding piecemeal litigation”; (g) “the order in which jurisdiction was obtained by the concurrent forums”; and (h) “the Government’s previous willingness to litigate similar suits in state court.” Memorial Hospital, 460 U.S. at 16. UNITED STATES V. DAVIS 57 “correctly recognized” that it was “require[d]” to apply the Colorado River test by virtue of a five-Justice majority comprised of four dissenting Justices and one concurring Justice, id., the Supreme Court in Memorial Hospital confirmed that its precedents relating to the consideration of dissenting opinions do bind us as a federal intermediate court (contrary to the suggestion of my concurring colleagues). The Majority is correct that the Justices in Freeman did not agree on much. But a five-Justice majority (Justice Sotomayor, plus the four dissenting Justices) did agree on one point—that a sentence imposed under a Rule 11(c)(1)(C) plea agreement is always “based on the [plea] agreement” itself. See, e.g., Freeman, 564 U.S. at 534 (Sotomayor, J., concurring) (“[T]he term of imprisonment imposed by a district court pursuant to an agreement authorized by Federal Rule of Criminal Procedure 11(c)(1)(C) . . . agreement[] is ‘based on’ the agreement itself, not on the judge’s calculation of the Sentencing Guidelines.”); see also id. at 544 (Roberts, C.J., dissenting) (“I agree with Justice SOTOMAYOR that ‘the term of imprisonment imposed pursuant to a [Rule 11(c)(1)(C)] agreement is, for purposes of § 3582(c)(2), “based on” the agreement itself.’” (quoting Justice Sotomayor’s concurrence, id. at 534)). This was a holding that received the vote of five Justices (a majority) of the Court. Like the Fourth Circuit in Memorial Hospital, we are bound by that holding.12 12 The Majority incorrectly suggests that Tidewater and its progeny somehow support a reasoning-based approach to Marks. Maj. Op. at 15–16, n.7. But quite the opposite is true. In Tidewater, Justice Rutledge (joined by Justice Murphy) held that Congress’ power to confer Article III jurisdiction was limited to the bases enumerated in the Constitution. See discussion supra, n.9. And four dissenting Justices expressly agreed with that holding. Of course, Justice Rutledge and the four dissenting Justices 58 UNITED STATES V. DAVIS ultimately disagreed about whether Article III’s reference to “citizen[s] of a state” should be understood as encompassing District of Columbia citizens. This disagreement led the two factions of Justices to vote for different case results. But all six Justices voted to hold that the Constitution provided the starting point for the Court’s analysis; Congress had no authority to add new bases for Article III jurisdiction by statute. Tidewater and its progeny hold that we are bound by holdings from splintered Court opinions that garner the five or more votes from the Court. Consideration of dissenting opinions to derive the “narrowest grounds” does not focus on the various reasonings as determinative of results. Indeed, it is just the opposite. Consideration of dissenting opinions is done not for the purpose of combining the rationales—an impossible task, since they are contradictory—but for predicting the vote (the result) which the dissenting opinions would add to the plurality opinion’s votes for the next analogous case. It is not the contradictory rationales that combine in Tidewater to result in a rule that “Congress may not expand the scope of subject-matter jurisdiction conferred by Article III through passage of a Congressional Act.” See supra, at pp. 52–54. It is the combined results of Justice Rutledge’s and Justice Murphy’s votes in favor of such a rule, plus the similar votes of the four dissenting Justices on that same issue that established the rule. Seen from the other side of the case, it is the combination of the result of the votes of the 3-member plurality that Congress had the power to so expand subject-matter jurisdiction, with the 2-member concurrence, which vehemently rejected such power, but found that Art. III itself was originally intended to include D.C. citizens for purposes of establishing diversity of citizenship jurisdiction, that established the rule. It was not a rule “derived by combining the ‘views’ or ‘rationales[s]’ of Tidewater’s concurrence and dissent.” Majority Op. 16, n.7. Just the opposite. It was a rule derived from the votes of the Justices, notwithstanding contradictory views or rationales used to explain the votes. UNITED STATES V. DAVIS 59 The Majority blatantly ignores Chief Justice Roberts’ express agreement with Justice Sotomayor and focuses only on the disagreements between them. But of course there are points on which they disagree; that is why there is both a concurrence and a dissent in Freeman (just as there was ample disagreement between the concurring and dissenting Justices in Tidewater). But those disagreements do not negate the fact that there are no sentence reductions which Justice Sotomayor would deny that the four dissenting Justices would not also deny. Where, as here, a plea agreement contains no mention of either the sentencing Guidelines or the criteria necessary to calculate the applicable Guidelines range (Justice Sotomayor’s two exceptions), five Justices in Freeman (Justices Sotomayor, Roberts, Scalia, Thomas, and Alito) would always vote to deny the defendant’s petition to seek resentencing. Under Marks, we are bound by this result. In sum, the Majority makes a good case that “federal sentencing law,” Rule 11(c)(1)(C), and the Guidelines’ policy statements all support the view adopted by the Justice Kennedy plurality in Freeman. See Maj. Op. at 25–26. And these arguments may well be the basis for a future Supreme Court opinion abrogating Freeman and adopting outright the plurality opinion of Justice Kennedy. But that is the Court’s province, not ours. As an intermediate federal court, we are not free to disregard binding Supreme Court precedent simply because we can think of a rule we like better. The purpose of determining a “holding” is to apply stare decisis in decisions I note the Majority’s “logical subset” also cannot be squared with Tidewater, as neither opinion in Tidewater was a logical subset of the other, and yet we have derived a binding holding from that splintered decision. 60 UNITED STATES V. DAVIS by intermediate appellate courts. It is only by intermediate courts following the holdings of the Supreme Court that one can hope to have predictability of law—the Rule of Law—from intermediate courts of appeal. While I may not agree with Justice Sotomayor’s approach, I think Marks constrains our discretion. The Majority today defies stare decisis by adopting a contrary approach and result. * * * For all of these reasons, we had it right in Austin, and I respectfully dissent. 


[by Paez]

 OPINION PAEZ, Circuit Judge: In this case, we must consider how to interpret the Supreme Court’s fractured opinion in Freeman v. United States, which addressed whether a defendant sentenced pursuant to a Federal Rule of Criminal Procedure 11(c)(1)(C) plea agreement is eligible for a sentence reduction under 18 U.S.C. § 3582 (c)(2). 564 U.S. 522 , 131 S.Ct. 2685 , 180 L.Ed.2d 519 (2011). Although a majority of the Court held that in such cases defendants remain eligible for relief under § 3582(c)(2), the plurality and concurring opinions did not agree on a single rationale. Citing Marks v. United States, 430 U.S. 188 , 97 S.Ct. 990 , 51 L.Ed.2d 260 (1977), which provides guidance on interpreting fractured Supreme Court opinions, we held in United States v. Austin that Justice Sotomayor’s concurring opinion was controlling, as it represented the narrowest grounds on which a majority of the justices agreed. See 676 F.3d 924 (9th Cir. 2012). Applying Justice Sotomayor’s approach, the district court denied Appellant Tyrone Davis’s (“Davis”) motion for a sentence reduction, ruling that his sentence was based on a Rule 11(e)(1)(C) plea agreement and not a “sentencing range that has subsequently been lowered by the Sentencing Commission” as required by § 3582(c)(2). Revisiting Freeman and our application of Marks to fractured Supreme Court opinions, we hold that where we can identify no rationale common to a majority of the Justices, we are bound only by the result. In so holding, we join the D.C. Circuit, which concluded that Freeman lacks a controlling opinion “because the plurality and concurring opinions do not share common reasoning whereby one analysis is a logical subset of the other.” United States v. Epps, 707 F.3d 337, 350 (D.C. Cir. 2013) (internal quotation marks and citation omitted). Bound only by Free *1017 man’s specific result, the D.C. Circuit adopted the plurality opinion’s approach, which holds that “[e]ven when a defendant enters into an 11(c)(1)(C) agreement, the judge’s decision to accept the plea and impose the recommended sentence is likely to be based on the Guidelines; and when it is, the defendant should be eligible to seek § 3582(c)(2) relief.” Freeman, 564 U.S. at 534 , 131 S.Ct. 2685 (plurality opinion). We also adopt this approach and overrule Austin. Applying the plurality approach, we conclude that Davis is eligible to seek a sentence reduction under § 3582(c)(2). We therefore reverse the district court’s denial of Davis’s motion and remand for a determination of whether Davis should receive a reduction in his sentence. I. A. In 2005, pursuant to a plea agreement entered into under Rule 11(c)(1)(C), Davis pled guilty to a series of counts related to distribution of cocaine base, or “crack cocaine.” 1 In his plea agreement, Davis admitted his offense conduct involved at least 170.5 grams of crack cocaine. Given the quantity of crack cocaine for which Davis accepted direct responsibility, the parties agreed that Davis’s base offense level was 34 under United States Sentencing Guidelines (“USSG”) § 2D1.1(c)(3) (Nov. 2005) and that the appropriate prison sentence would be eighteen years. The plea agreement acknowledged that under Rule 11(c)(1)(C), the district court had discretion to accept or reject the proposed agreement and recommended sentence. If the district court accepted the agreement, then under Rule 11(c)(1)(C) the recommended sentence would be binding on the court. On the other hand, if the court rejected the recommended sentence, the parties could withdraw from the agreement. At sentencing in 2006, the district court calculated Davis’s total offense level as 37 with a Criminal History Category II, resulting in a Guidelines range of 235 to 293 months. The court accepted the Rule 11(c)(1)(C) plea agreement and imposed the recommended eighteen-year (216-month) sentence. We reversed and remanded, holding that the district court had erred in its determination of Davis’s criminal history category and in its imposition of an “organizer or leader” enhancement. United States v. Davis, 312 Fed.Appx. 909, 911-14 (9th Cir. 2009). At Davis’s resen-tencing in 2009, the district court calculated his total offense level as 36 with a Criminal History Category I, resulting in a Guidelines range of 188 to 235 months. The court reimposed the plea agreement’s recommended eighteen-year (216-month) sentence, finding it “fair and reasonable” under the Guidelines. We affirmed. United States v. Davis, 389 Fed.Appx. 616 (9th Cir. 2010). B. When Davis pled guilty, the Guidelines punished defendants far more harshly for crack cocaine offenses than for powder cocaine offenses. A defendant responsible for one gram of crack cocaine faced the same Guidelines sentence as a defendant responsible for one hundred grams of powder cocaine. This 100:1 ratio was roundly criticized for its racially disparate effects. See, e.g., Kimbrough v. United States, 552 *1018 U.S. 85, 98 , 128 S.Ct. 558 , 169 L.Ed.2d 481 (2007) (citing the Sentencing Commission’s finding that the “severe sentences required by the 100-to-l ratio are imposed primarily upon black offenders”) (internal quotation marks omitted). As a 2006 survey by the Substance Abuse and Mental Health Services Administration showed, whites formed the biggest group of crack cocaine users in absolute numbers, 2 but African Americans were disproportionately arrested and convicted for crack cocaine offenses. A 2007 report by the Sentencing Commission documented that when Davis was sentenced in 2006, 81.8% of federal crack cocaine offenders were African American. U.S. Sentencing Commission, Report to Congress: Cocaine and Federal Sentencing Policy at 15 (2007) (“2007 Report”). Moreover, in a series of reports to Congress, the Sentencing Commission warned that the “data no longer support” the assumption that crack cocaine is more harmful than powder cocaine. Kimbrough, 552 U.S. at 97-98 , 128 S.Ct. 558 (quoting U.S. Sentencing Commission, Report to Congress: Cocaine and Federal Sentencing Policy at 96 (2002); also citing the 2007 Report); see also United States v. Baptist, 646 F.3d 1225, 1226 , 1228 n. 1 (9th Cir. 2011) (per curiam). Citing the “urgent and compelling” problems raised by the overly punitive crack sentencing scheme, the Sentencing Commission repeatedly called on Congress to reduce the 100:1 ratio. See, e.g., 2007 Report at 8-9. Federal judges and Department of Justice officials likewise joined the chorus of voices demanding reform. See, e.g., United States v. Then, 56 F.3d 464, 467 (2d Cir. 1995) (Calabresi, J., concurring) (“The unfavorable and disproportionate impact that the 100-to-l crack/cocaine sentencing ratio has on members of minority groups is deeply troubling.”); Remarks of Attorney General Eric Holder, D.C. Court of Appeals Judicial Conference (June 19, 2009), available at http://www.justice.gov/opa/speech/ attorney-general-eric-holder-dc-court-appeals-judicial-conferenee (“It is the view of this Administration that the 100-to-l crack-powder sentencing ratio is simply wrong. It is plainly unjust to hand down wildly disparate prison sentences for materially similar crimes.”). In 2010, Congress responded by passing the Fair Sentencing Act, Pub. L. No. 111-220, 124 Stat. 2372 (2010), which dramatically reduced the sentencing ratio to 18:1. The Fair Sentencing Act also gave the Sentencing Commission “emergency authority” to “make such conforming amendments to the Federal sentencing guidelines.” Id. § 8. The Sentencing Commission responded by issuing amended Guidelines reflecting the new 18:1 ratio 3 and made the changes retroactive 4 for all defendants who have “been sentenced to a term of imprisonment based on a sentencing range that has subsequently been lowered by the Sentencing Commission.” 18 U.S.C. § 3582 (c)(2). *1019 C. In 2012, Davis filed a pro se motion under § 3582(c)(2) seeking a retroactive reduction of his sentence in light of the amended Guidelines. 5 Before the Fair Sentencing Act, the stipulated amount of 170.5 grams of crack cocaine in Davis’s Rule 11(c)(1)(C) plea agreement resulted in a base offense level of 34. After the dramatic reduction in the sentencing ratio, that same amount yielded a base offense level of 28. USSG § 2D1.1(c)(6) (2011). Davis argued that he was eligible for a sentence modification under § 3582(c)(2) because his sentence was “based on” the Guidelines. The district court denied the motion, holding that it lacked jurisdiction to modify Davis’s sentence because it was “based on” the Rule 11(c)(1)(C) plea agreement, not the Guidelines. In so ruling, the district court concluded that it was bound by Justice Sotomayor’s concurring opinion in Freeman . Davis appealed for a third time. A three-judge panel affirmed, relying on Austin. United States v. Davis, 776 F.3d 1088 (9th Cir. 2015). We granted rehearing en banc. United States v. Davis, 795 F.3d 1188 (9th Cir. 2015). 6 II. A. In Freeman v. United States, the Supreme Court considered whether a defendant sentenced under a Rule 11(c)(1)(C) agreement may be eligible for a sentence reduction under § 3582(c)(2). 564 U.S. 522 , 131 S.Ct. 2685 , 180 L.Ed.2d 519 (2011). Five justices ultimately agreed that Freeman was . eligible for a reduction, but no rationale commanded a majority of the Court. A four-justice plurality held that “[e]ven when a defendant enters into an 11(c)(1)(C) agreement, the judge’s decision to accept the plea and impose the recommended sentence is likely to be based on the Guidelines; and when it is, the defendant should be eligible to seek § 3582(c)(2) relief.” Id. at 534, 131 S.Ct. 2685 (plurality opinion). Writing for the plurality, Justice Kennedy explained that this approach was consistent with the “text and purpose” of the Sentencing Reform Act, Rule 11(c)(1)(C), and the binding Guidelines policy statements in sections 6B1.2 and 1B1.10 relating to Rule 11(c)(1)(C) and § 3582(c)(2). Id. at 530, 131 S.Ct. 2685 . In a dissenting opinion by Chief Justice Roberts, four Justices took the contrary position that a sentence imposed under a Rule 11(c)(1)(C) agreement is never “based on” the Guidelines because the agreement itself serves as the foundation for the sentence imposed. Id. at 544 , 131 S.Ct. 2685 (Roberts, C.J., dissenting). Concurring only in the judgment, Justice Sotomayor staked out yet a third position. Justice Sotomayor argued that a sentence imposed under a Rule 11(c)(1)(C) *1020 agreement is generally based on the agreement, not the Guidelines, but rejected the dissent’s categorical bar to relief for all defendants sentenced under such agreements. Id. at 534 , 131 S.Ct. 2685 (Sotomayor, J., concurring in the judgment). Justice Sotomayor concluded that district courts have jurisdiction to consider a sentence reduction in at least two circumstances: when the Rule 11(c)(1)(C) agreement either 1) “call[s] for the defendant to be sentenced within a particular Guidelines sentencing range,” or 2) “make[s] clear that the basis for the specified term is a Guidelines sentencing range applicable to the offense to which the defendant pleaded guilty” and “that sentencing range is evident from the agreement itself.” Id. at 538-39 , 131 S.Ct. 2685 . To say that Freeman divided the Court would be an understatement. Not only did the plurality and dissenting opinions take opposite positions, but both also strongly criticized Justice Sotomayor’s concurrence. The plurality warned that the “consequences of [the concurrence’s] erroneous rule would be significant,” id. at 533 , 131 S.Ct. 2685 (plurality opinion), while the dissent complained that Justice Sotoma-yor’s approach would “foster confusion in an area in need of clarity,” id. at 550 , 131 S.Ct. 2685 (Roberts, C.J., dissenting). The dissenting opinion accurately stated that the plurality and concurrence “agree on very little except the judgment.” Id. at 544 , 131 S.Ct. 2685 (Roberts, C.J., dissenting). Thus, the 4-1-4 Freeman Court did not articulate a clear path forward for analysis of sentence-reduction requests by defendants sentenced under Rule 11(c)(1)(C) agreements. B. 1. In Marks v. United States, the Supreme Court explained that “[w]hen a fragmented Court decides a ease and no single rationale explaining the result enjoys the assent of five Justices, the holding of the Court may be viewed as that position taken by those Members who concurred in the judgments on the narrowest grounds.” 430 U.S. 188 , 193, 97 S.Ct. 990 , 51 L.Ed.2d 260 (1977) (internal quotation marks and citation omitted). In the nearly forty years since Marks , lower courts have struggled to divine what the Supreme Court meant by “the narrowest grounds.” Indeed, the Court has acknowledged that the Marks inquiry at times has “baffled and divided the lower courts that have considered it,” Nichols v. United States, 511 U.S. 738, 746 , 114 S.Ct. 1921 , 128 L.Ed.2d 745 (1994), and that the “test is more easily stated than applied.” Grutter v. Bollinger, 539 U.S. 306, 325 , 123 S.Ct. 2325 , 156 L.Ed.2d 304 (2003) (quoting Nichols, 511 U.S. at 745-46 , 114 S.Ct. 1921 ). In the face of this confusion, two main approaches have emerged: one focusing on the reasoning of the various opinions and the other on the ultimate results. The D.C. Circuit has offered a clear example of the first approach. In King v. Palmer, the court explained: Marks is workable' — one opinion can be meaningfully regarded as “narrower” than another — only when one opinion is a logical subset of other, broader opinions. In essence, the narrowest opinion must represent a common denominator of the Court’s reasoning; it must embody a position implicitly approved by at least five Justices who support the judgment. 950 F.2d 771, 781 (D.C. Cir. 1991) (en banc). The D.C. Circuit reaffirmed this approach in Epps , describing Marks as applicable only when “the concurrence posits a narrow test to which the plurality must necessarily agree as a logical consequence of its own, broader position.” 707 F.3d at 348 (emphasis omitted) (quoting King, 950 F.2d at 782 ). *1021 The second approach looks to results rather than reasoning. It defines the narrowest ground as the rule that “would necessarily produce results with which a majority of the Justices from the controlling case would agree.” See, e.g., Planned Parenthood of Se. Pa. v. Casey, 947 F.2d 682 , 694-97 (3d Cir. 1991) (finding that Justice O’Connor’s concurring opinions controlled the fractured decisions in Webster v. Reproductive Health Services, 492 U.S. 490 , 109 S.Ct. 3040 , 106 L.Ed.2d 410 (1989), and Hodgson v. Minnesota, 497 U.S. 417 , 110 S.Ct. 2926 , 111 L.Ed.2d 344 (1990), because a majority of justices in each case would have agreed with her result), aff'd in part, rev’d in part, 505 U.S. 833 , 112 S.Ct. 2791 , 120 L.Ed.2d 674 (1992). 2. Our cases interpreting Marks have not been a model of clarity. On one occasion, we cited the “results” language described above. See United States v. Williams, 435 F.3d 1148 , 1157 n. 9 (9th Cir. 2006) (explaining that a concurrence is controlling under Marks if it “would affect a narrower range of cases than that of the plurality”). Nonetheless, in Williams and other decisions applying Marks to a fractured Supreme Court decision, we analyzed whether the reasoning of a narrower opinion fit entirely into the circle drawn by a broader opinion in order to derive a rule. Our most recent decision to address Marks explicitly employed the “reasoning” approach. Lair v. Bullock, 697 F.3d 1200 (9th Cir. 2012). In Lair , we approvingly cited King and held that the Marks standard applies only “where one opinion can be meaningfully regarded as narrower than another and can represent a common denominator of the Court’s reasoning.” Id. at 1205 (quoting United States v. Rodriguez-Preciado, 399 F.3d 1118, 1140 (9th Cir.), amended by 416 F.3d 939 (9th Cir. 2005)). Unless “the narrowest opinion is actually the logical subset of other broader opinions, ... the only binding aspect of a splintered decision is its specific result.” Id. (internal quotation marks and citation omitted). To foster clarity, we explicitly adopt the reasoning-based approach to applying Marks . This approach is not only consistent with our most recent caselaw, see Lair, 697 F.3d 1200 , but also makes the most sense. 7 A fractured Supreme *1022 Court decision should only bind the federal courts of appeal when a majority of the Justices agree upon a single underlying rationale and one opinion can reasonably be described as a logical subset of the other. When no single rationale commands a majority of the Court, only the specific result is binding on lower federal courts. III. Applying Marks , as clarified above, to Freeman , we overrule our holding in Austin that Justice Sotomayor’s concurrence controls. 8 Instead, we adopt the analysis of the D.C. Circuit in Epps that there was no common denominator in Freeman “because the plurality and concurring opinions do not share common reasoning whereby one analysis is a logical subset of the other.” Epps, 707 F.3d at 350 (internal quotation marks and citation omitted). Justice Sotomayor’s concurrence cannot reasonably be described as a logical subset of Justice Kennedy’s plurality opinion. The Freeman plurality explicitly rejected the concurrence’s reasoning, in particular its underlying premise that a sentence imposed under a Rule 11(c)(1)(C) agreement is “based on” the parties’ agreement, not the Guidelines. 564 U.S. at 529 , 131 S.Ct. 2685 (plurality opinion); Id. at 535-36 , 131 S.Ct. 2685 (Sotomayor, J., concurring in the judgment). Even in setting out the circumstances in which she would find a defendant sentenced under a Rule 11(c)(1)(C) agreement eligible for relief, Justice Sotomayor focused on the role the parties’ Guidelines calculations play in crafting a Rule 11(c)(1)(C) agreement. Id. at 538-39 , 131 S.Ct. 2685 . By contrast, the plurality focuses on the role of the judge’s Guidelines calculations in deciding whether to accept or reject the agreement. Id. at 529 , 131 S.Ct. 2685 (plurality opinion). This fundamental divergence in reasoning is enough to demonstrate that Justice Soto-mayor’s rationale is not controlling Supreme Court law. Although in Freeman these divergent approaches led to the same result, the D.C. Circuit properly recognized that “the set of cases where the *1023 defendant prevails under the concurrence is not always nestled within the set of cases where the defendant prevails under the plurality....” Epps, 707 F.3d at 351 . Two examples from Epps are instructive. 9 First consider the following scenario: [T]he parties may state in the plea agreement that a particular range applies and agree to a sentence at the bottom of that range, but the district court may not agree that the range determined by the parties applies, finding for example that the career offender range is applicable instead, but notwithstanding this finding accept the plea because it is to a term that is acceptable to the court for reasons unrelated to the guideline range determined by the parties. Id. at 350 n. 8. Justice Sotomayor would allow a sentence reduction in this example because the agreement explicitly “call[s] for the defendant to be sentenced within a particular Guidelines sentencing range.” Freeman, 564 U.S. at 538 , 131 S.Ct. 2685 (Sotomayor, J., concurring in the judgment). The plurality, on the other hand, “would find [Freeman] ineligible because the range that the parties agreed to played no role in the court’s determination that this was an appropriate sentence, despite the fact that the court imposed the agreed-upon term of imprisonment.” Epps, 707 F.3d at 350 n. 8. Thus, the plurality opinion is actually the narrower one in certain respects. A second example produces a similar result: The sentencing court ... might consider and reject the guideline range used by the parties, not because the court finds that a different guidelines range (such as the career offender range) applies, but because, having considered the applicable guidelines range, the court rejects it as a matter of policy and selects its sentence without regard to it. Id. Here again, if the court decides “for reasons unrelated to the guidelines range to impose the sentence the parties agreed upon,” the defendant would be eligible for a reduction under Justice Sotomayor’s approach but not under the plurality’s. Id. These examples make clear that the plurality and concurring opinions cannot be explained by a diagram in which a circle representing the reasoning of Justice Soto-mayor’s opinion sits neatly within a circle representing the reasoning of the plurality opinion. Because both opinions would allow sentence reductions in situations where the other would not, Justice Sotomayor’s con *1024 currence is not the “narrowest grounds” envisioned by Marks . 10 We recognize that, with the exception of the D.C. Circuit, every other circuit that has considered the issue has adopted Justice Sotomayor’s concurrence as the controlling opinion in Freeman . But we do not find those opinions convincing. Most engage with Marks only superficially, quoting its language with no analysis. See, e.g., United States v. Graham, 704 F.3d 1275, 1278 (10th Cir. 2013); United States v. Browne, 698 F.3d 1042, 1045-46 (8th Cir. 2012); United States v. Dixon, 687 F.3d 356, 359-60 (7th Cir. 2012); United States v. Smith, 658 F.3d 608, 611 (6th Cir. 2011). 11 Given their lack of meaningful analysis, these opinions lack persuasive force. Those few cases that do discuss how Marks should apply to Freeman mistakenly conclude that although the “gap between the plurality and the concurrence is wide, [] it is still possible to tease out a common denominator.” United States v. Rivera-Martinez, 665 F.3d 344, 348 (1st Cir. 2011); see also United States v. Thompson, 682 F.3d 285, 289-90 (3d Cir. 2012). Not so. As the examples above demonstrate, there are some circumstances where defendants would be eligible for relief under Justice Sotomayor’s approach but not under the plurality’s. We therefore cannot agree with the First Circuit’s assertion that the “plurality would surely agree that in every case in which a defendant’s C-type plea agreement satisfies the criteria for Justice Sotomayor’s exception ... the sentencing judge’s decision to accept that sentence is based on the guidelines.” Rivera-Martinez, 665 F.3d at 348 . A more nuanced reading of both opinions leads us to conclude that “there is no controlling opinion in Freeman because the plurality and concurring opinions do not share common reasoning whereby one analysis is a logical subset of the other.” Epps, 707 F.3d at 350 (internal quotation marks and citation omitted). Marks instructs us to consider the opinions only of “those Members who concurred in the judgments on the narrowest grounds” when deriving a rule from a fractured Supreme Court decision. Marks, 430 U.S. at 193 , 97 S.Ct. 990 (quoting Gregg v. Georgia, 428 U.S. 153 , 169 n. 15, 96 S.Ct. 2909 , 49 L.Ed.2d 859 (1976) (opinion of Stewart, Powell, & Stevens, JJ.) (emphasis added)). Nonetheless, we acknowledge that the Supreme Court and our sister circuits have considered dissenting opinions when interpreting fragmented Supreme Court decisions. See, e.g., United States v. Jacobsen, 466 U.S. 109, 115-17 , 104 S.Ct. 1652 , 80 L.Ed.2d 85 (1984) (relying on a dissenting opinion to derive the rule in Walter v. United States, 447 U.S. 649 , 100 S.Ct. 2395 , 65 L.Ed.2d 410 (1980)); Moses H. Cone Mem’l Hosp. v. Mercury Constr. Corp., 460 U.S. 1, 16-17 , 103 S.Ct. 927 , 74 L.Ed.2d 765 (1983) (holding that Will v. Calvert Fire Insurance did not overrule *1025 the “Colorado River test” because Will’s four dissenting Justices agreed with the concurring opinion that the test remained in effect); United States v. Donovan, 661 F.3d 174, 182 (3d Cir. 2011) (explaining that the court “looked to the votes of dissenting Justices if they, combined with votes from plurality or concurring opinions, establish a majority view on the relevant issue”). Here, we assume but do not decide that dissenting opinions may be considered in a Marks analysis. 12 Considering Chief Justice Roberts’s dissent would not change our conclusion because we cannot derive any common denominator by combining Freeman’s dissenting opinion with either the plurality or concurring opinion. First, no rule can be derived from the Freeman plurality and dissenting opinions, as neither is a “logical subset” of the other. Indeed, the plurality holding that “[e]ven when a defendant enters into an 11(c)(1)(C) agreement, the judge’s decision to accept the plea and impose the recommended sentence is likely to be based on the Guidelines,” Freeman, 564 U.S. at 534 , 131 S.Ct. 2685 (plurality opinion) is diametrically opposed to the dissent’s position that a sentence imposed under a Rule 11(c)(1)(C) agreement is never “based on” the Guidelines because the agreement itself serves as the foundation for thé sentence imposed, id. at 544-45 , 131 S.Ct. 2685 (Roberts, C.J., dissenting). In practical terms, this divergence means the dissent would categorically find all defendants sentenced under a Rule 11(c)(1)(C) agreement ineligible for a sentence reduction, while the plurality would permit a reduction in most cases. Thus, the plurality and dissent “do not share common reasoning whereby one analysis is a logical subset of the other.” Epps, 707 F.3d at 350 (internal quotation marks and citation omitted). Second, Justice Sotomayor’s concurring opinion is not a logical subset of the dissenting opinion, or vice versa. Again, the dissent would categorically bar defendants sentenced under Rule 11(c)(1)(C) agreements from seeking relief under § 3582(c)(2). Freeman, 564 U.S. at 544-45 , 131 S.Ct. 2685 (Roberts, C.J.. dissenting). Justice Sotomayor, by contrast, would permit sentence reductions in limited circumstances: if the Rule 11(c)(1)(C) agreement' either 1) “call[s] for the defendant to be sentenced within a particular Guidelines sentencing range,” or 2) “make[s] clear that the basis for the specified term is a Guidelines sentencing range applicable to the offense to which the defendant pleaded guilty” and “that sentencing range is evident from the agreement itself.” Id. at 538-39 , 131 S.Ct. 2685 (Sotomayor, J., concurring in the judgment). Despite the dissent’s arguments to the contrary, these approaches cannot be reconciled because Justice Sotomayor explicitly “rejected] the categorical rule advanced by the Government and endorsed by the dissent, which artificially divorces a [Rule 11(c)(1)(C) ] agreement from its express terms.” Id. at 539 , 131 S.Ct. 2685 . Indeed, in evaluating Freeman’s case, Justice Soto-mayor stated that “contrary to the dissent’s curious suggestion that ‘there is no way of knowing what th[e] sentence was based on,’ ” the basis for Freeman’s sentence was evident from the Rule 11(c)(1)(C) agreement itself. Id. at 542-43 , 131 S.Ct. 2685 . The Freeman dissent is similarly critical of Justice Sotomayor, describing her view that certain Rule 11(c)(1)(C) defendants are eligible for relief as just “as mistaken *1026 as the position of the plurality — and basically for the same reasons.” Id. at 545 , 131 S.Ct. 2685 (Roberts, C.J., dissenting). Chief Justice Roberts characterized her approach as “head-scratching,” id. at 547 , 131 S.Ct. 2685 , and likely to “foster confusion in an area in need of clarity,” id. at 550 , 131 S.Ct. 2685 . Neither opinion lays out the “common denominator of the Court’s reasoning” because Justice Soto-mayor and the dissenters disagree on the fundamental question of whether a defendant sentenced under a Rule 11(c)(1)(C) agreement can ever be eligible for a sentence reduction under § 3582(c)(2). Simply put, no combination of Freeman’s dissenting and concurring opinions yields a binding rule that we must follow. IV. A. Given that no opinion in Freeman controls, we consider which of the rationales set forth in the varying opinions is most persuasive. Epps, 707 F.3d at 351 . In so doing, we are restricted only by the ultimate result in Freeman : that defendants sentenced under Rule 11(c)(1)(C) agreements are not categorically barred from seeking a sentence reduction under § 3582(c)(2). We join the D.C. Circuit and adopt the plurality’s rule: “Even when a defendant enters into an 11(c)(1)(C) agreement, the judge’s decision to accept the plea and impose the recommended sentence is likely to be based on the Guidelines; and when it is, the defendant should be eligible to seek § 3582(c)(2) relief.” Freeman, 564 U.S. at 534 , 131 S.Ct. 2685 (plurality opinion). As the plurality explained, three critical sources support this approach. First, “[fjederal sentencing law requires the district judge” to impose sentences that comply with “the purposes of federal sentencing, in light of the Guidelines and other § 3553(a) factors.” Id. at 529, 131 S.Ct. 2685 . Thus, by statute, a sentencing judge’s discretion is always framed by the Guidelines. Second, Justice Kennedy looked to the district court’s authority under Rule 11(c)(1)(C). Although the Rule “permits the defendant and the prosecutor to agree on a specific sentence,” it preserves “the district court’s independent obligation to exercise its discretion” and review the proposed sentence. Id. Because judges “use the Guidelines range as the starting point,” they serve in a “real sense [as] a basis for the sentence,” “[e]ven where the judge varies from the recommended range.” Id. Third, the Guidelines policy statements that apply to Rule 11(c)(1)(C) plea agreements and § 3582(c)(2) motions support the plurality’s approach. Once a district court accepts a Rule 11(c)(1)(C) plea agreement, the parties’ recommended sentence is binding on the court. As the Freeman plurality noted, however, the applicable Guidelines policy statement “forbids the district judge to accept an 11(c)(1)(C) agreement without first evaluating the recommended sentence” under the Guidelines. Id.-, USSG § 6B1.2(e). Indeed, as the plurality further noted, the commentary to the policy statement instructs a sentencing court to accept the recommended sentence only if it is an appropriate sentence within the applicable Guidelines range or “departs ... for justifiable reasons.” Freeman, 564 U.S. at 529 , 131 S.Ct. 2685 (plurality opinion). The Guidelines policy statement that applies to § 3582(c)(2) motions likewise supports the plurality opinion. See USSG § lB1.10(b)(l). As the plurality explained, § lB1.10(b)(1) directs the district judge “in modifying a sentence to substitute only the retroactive amendment and then leave all original Guidelines determinations in place.” Freeman, 564 U.S. at 530 , 131 S.Ct. 2685 (plurality opinion). The goal of *1027 the sentence modification is to “isolate” the effect of the amended Guideline while leaving undisturbed the other factors that determined the sentence imposed. Id. 13 This suggests that a defendant should be eligible for a sentence reduction when one factor in a defendant’s sentence was a “since-rejected Guideline.” Id. Not only does the plurality approach best conform with these relevant sources, but a “contrary focus on the parties’ intentions would contribute to the unwarranted disparity that the [Sentencing Reform Act] was designed to reduce.” Epps, 707 F.3d at 351 . In reducing the crack cocaine sentencing range, Congress and the Sentencing Commission sought to address “the urgent and compelling problem of crack-cocaine sentences.” Freeman, 564 U.S. at 533 , 131 S.Ct. 2685 (plurality opinion) (internal quotation marks omitted); supra at I.B. “Section 3582(c)(2) empowers district judges to correct sentences that depend on frameworks,” like the one for crack cocaine, “that later prove unjustified.” Freeman, 564 U.S. at 526 , 131 S.Ct. 2685 (plurality opinion). Justice Sotoma-yor’s approach would “extend the benefit of the Commission’s judgment only to an arbitrary subset of defendants whose agreed sentences were accepted in light of a since-rejected Guidelines range based on whether their plea agreements refer to the Guidelines.” Id. at 533-34 , 131 S.Ct. 2685 . Thus, adoption of the concurring opinion would “undercut a systemic solution” to a “systemic injustice.” Id. at 534 , 131 S.Ct. 2685 . For all these reasons, we adopt the approach of the Freeman plurality opinion. B. Applying the plurality’s approach, we hold that Davis is eligible for relief under § 3582(c)(2) because the district court’s “decision to accept the plea and impose the recommended sentence” was “based on the Guidelines.” Freeman, 564 U.S. at 534 , 131 S.Ct. 2685 (plurality opinion). Davis’s Rule 11(c)(1)(C) plea agreement was clearly rooted in the Guidelines. First, it required the district judge to “determine Defendant’s applicable Sentencing Guidelines range at the time of the sentencing.” Second, the agreement stated that the amount of crack cocaine for which Davis admitted direct responsibility would yield a base offense level of 34 under Guidelines § 2D1.1(c)(3). Third, Davis’s agreement explained that he qualified for a Guidelines increase under § 2D1.2 for proximity to a school zone and a Guidelines reduction under § 3E1.1 for acceptance of responsibility. The district judge’s decision to reimpose the eighteen-year sentence was also based on the Guidelines. During the resentencing hearing, the district court recalculated Davis’s total offense level at 36 and a Guidelines range of 188 to 235 months. Then, reflecting on all the evidence presented, the court determined that the original eighteen-year sentence—which, at 216 months, fell within the calculated range— was “fair and reasonable” under the Guidelines. Taken together, the text of Davis’s plea agreement and the judge’s statements during the sentencing hearing leave no *1028 doubt that the sentence imposed was “based on” the Guidelines. Thus, Davis is eligible for a sentence reduction and all that remains for the district court is to make the discretionary determination whether Davis should actually receive a reduction under § 3582(c)(2) and the Guidelines’ related policy statements in section 1B1.10. 14 Conclusion In sum, when applying Marks to a fractured Supreme Court decision, we look to those opinions that concurred in the judgment and determine whether one of those opinions sets forth a rationale that is the logical subset of other, broader opinions. When, however, no “common denominator of the Court’s reasoning” exists, we are bound only by the “specific result.” Applying that framework to Freeman , we conclude that, contrary to our prior decision in Austin, Justice Sotomayor’s concurrence is not the logical subset of the plurality opinion. Nor can we extract a shared reasoning by including the dissent in our analysis. Thus, we overrule Austin and adopt the Freeman plurality approach as the most persuasive means of analyzing sentence reductions in the context of Rule 11(c)(1)(C) plea agreements. Accordingly, we reverse the district court’s determination that Davis is not eligible for a sentence reduction and remand for reconsideration of whether Davis should receive a sentence reduction under § 3582(c)(2) and the Guidelines’ related policy statements. REVERSED and REMANDED. . Davis pled guilty to the following offenses alleged in the Third Superseding Indictment: conspiracy to distribute cocaine base in violation of 21 U.S.C. §§ 841 (a)(1), 841(b)(1)(A), and 846 (Count 1); distribution of cocaine base in violation of 21 U.S.C. §§ 841 (a)(1), 841(b)(1)(A), and 860 (Counts 10 and 11); and possession with intent to distribute cocaine base in violation of 21 U.S.C. §§ 841 (a)(1) and 841(b)(1)(C) (Counts 12 and 13). .Results from the Substance Abuse and Mental Health Services Administration’s National Survey on Drug Use and Health from 2006 show that 3.3% of whites report ever using crack cocaine compared with 5.4% of African Americans. Thus, in absolute numbers, white crack cocaine users far outnumber African-American crack cocaine users. Quick Table: Ever Used Crack BY Race and Ethnicity, National Survey on Drug Use and Health, 2006, available at https://www.icpsr.umich.edu/ icpsrweb/NAHDAP/series/00064/studies (follow "National Survey on Drug Use and Health, 2006”; then follow "Quick Tables, Drug Use: Entire Sample”; then follow "Crack Use”; then select "Race and Ethnicity”; and generate table). . See U.S. Sentencing Guidelines Manual app. C, Amend. 748 (U.S. Sentencing Comm'n 2010) (adjusting Guidelines temporarily); id. at Amend. 750 (2011) (making adjustment permanent). . Id. at Amend. 759 (making Amendment 750 retroactive). . Section 3582(c)(2) permits a district court to modify a term of imprisonment ... in the case of a defendant who has been sentenced to a term of imprisonment based on a sentencing range that has subsequently been lowered by the Sentencing Commission pursuant to 28 U.S.C. 994(6), upon motion of the defendant or the Director of the Bureau of Prisons, or on its own motion, the court may reduce the term of imprisonment, after considering the factors set forth in section 3553(a) to the extent that they are applicable, if such a reduction is consistent with applicable policy statements issued by the Sentencing Commission. . Whether a district court has jurisdiction to modify a defendant’s sentence under 18 U.S.C. § 3582 (c)(2) is a legal question that we review de novo. United States v. Paulk, 569 F.3d 1094, 1095 (9th Cir. 2009) (per curiam). . This approach is not “fundamentally inconsistent with Marks itself.” Dissent at 1033, 1034 n. 7. Marks never defined the "narrowest grounds,” and the dissent identifies no subsequent Supreme Court case that has offered an explanation or clarification of Marks as requiring an unwavering focus on results. The difficult task of interpreting Maries has been left to the courts of appeal. See, e.g., King, 950 F.2d at 781 ; Lair, 697 F.3d at 1205 ; United States v. Johnson, 467 F.3d 56, 62-64 (1st Cir. 2006). More importantly, Marks cannot be viewed in isolation. In subsequent cases interpreting fractured Supreme Court decisions, the Court has frequently focused on reasoning, rather than results. Indeed, the dissent recognizes as much when it argues that Marks requires consideration of dissenting opinions. Dissent at 1039-41, 1039-40 n. 9 (describing the various opinions in the fractured National Mutual Insurance Co. v. Tidewater Transfer Co., 337 U.S. 582 , 69 S.Ct. 1173 , 93 L.Ed. 1556 (1949), decision). As the dissent states, "[s]ince Tidewater, courts have universally accepted that Congress may not expand the scope of subject-matter jurisdiction conferred by Article III through passage of a Congressional Act,” even though that rule could only be derived by combining the “views” or “rationale[s]” of Tidewater’s concurrence and dissent. Dissent at 1039-40; 1039-40 n. 9. Thus, even the dissent acknowledges that the Supreme Court and lower courts have employed a reasoning-based approach to analyzing prior fractured Supreme Court decisions. A results approach cannot explain the governing rule that emerged from Tidewater. Similarly, the dissent mischaracterizes United States v. Jacobsen, 466 U.S. 109 , 104 S.Ct. 1652 , 80 L.Ed.2d 85 (1984), as employing a results-based approach. Dissent 1034 n. *1022 7. In our view, the rule that "the legality of the government search must be tested by the scope of the antecedent private search,” id. at 115-16 , 104 S.Ct. 1652 , represents a common reasoning shared by the dissenting and plurality opinions, not a result or outcome on which they agree. The dissent’s own cases therefore contradict its assertion that a reasoning-based approach is an "invention” of the Ninth and D.C. Circuits. Dissent at 1030-31, 1035-36. . Following the adoption of the Fair Sentencing Act, we have issued, in addition to Austin, two opinions that discuss sentence modifications under § 3582(c)(2) and that bear mentioning here. First, in United States v. Bride, we held that a defendant who was sentenced under a Rule 11(c)(1)(C) agreement could not seek a sentence reduction because his sentence was not "based on a sentencing range that had been subsequently lowered by the Sentencing Commission.” 581 F.3d 888, 889 (9th Cir. 2009). We explicitly stated, however, that we did not "reach the issue of whether § 3582(c)(2) relief is necessarily precluded when the district court imposes a sentence pursuant to a Rule 11(c)(1)(C) plea agreement.” Id. at 891 n. 5. Austin distinguished Bride on this basis. Austin, 676 F.3d at 927 n. 1. Nonetheless, to the extent Bride conflicts with our opinion today, it is overruled. Second, in United States v. Pleasant, we cited Austin for the proposition that Justice Sotomayor’s Freeman concurrence controls, and held that the defendant was eligible for a sentence reduction because his Rule 11(c)(1)(C) plea agreement fell under one of her two exceptions. 704 F.3d 808, 811 (9th Cir. 2013). Although the thrust of the Pleasant opinion focused on the separate question of whether a sentence reduction would be "consistent” with the Guidelines, we determined that Pleasant satisfied the threshold eligibility determination. Id. at 811-12 . As with Bride, to the extent Pleasant's interpretation of Freeman is inconsistent with this opinion, it is overruled. . The dissent criticizes these examples, dissent at 1036-38, but its analysis is oversimplified in suggesting that the plurality would always allow a sentence modification in a Rule 11(c)(1)(C) agreement. If the plurality intended such a rule, it could easily have explicitly said so. Instead, Justice Kennedy wrote "[e]ven when a defendant enters into an 11(c)(1)(C) agreement, the judge’s decision to accept the plea and impose the recommended sentence is likely to be based on the Guidelines.” Freeman, 564 U.S. at 534 , 131 S.Ct. 2685 (plurality opinion) (emphasis added). Likely is not the same as always. Similarly, the plurality wrote that a "district judge’s decision to impose a sentence may therefore be based on the Guidelines even if the defendant agrees to plead guilty under Rule 11(c)(1)(C).” Id. at 526 , 131 S.Ct. 2685 (emphasis added). That is different than saying that a Rule 11(c)(1)(C) sentence is always based on the Guidelines. Nor did Justice Kennedy explicitly reject the idea “that his approach would limit relief to only a 'subset of defendants,' ” as the dissent claims. Dissent at 1037. Rather, in the passage the dissent cites, the plurality rejected Justice Sotomoyor's approach because of the arbitrariness of allowing sentence reductions only for those defendants whose plea agreements refer to the Guidelines. Freeman, 564 U.S. at 532-33 , 131 S.Ct. 2685 (plurality opinion). Justice Kennedy never said he disagreed with Justice Sotomayor because her approach would "fail to permit resentencing in all cases.” Dissent at 1037. . We emphasize that this results-oriented approach is used only to highlight the lack of a shared reasoning between Freeman's plurality and concurring opinions. Our primary focus remains on the text of the two opinions, rather than on their application to hypothetical cases. . For example, the Tenth Circuit quoted the Marks "narrowest grounds” test and then summarily agreed with the district court that "[a]pplying this rule,” Justice Sotomayor’s concurrence controls. Graham, 704 F.3d at 1278 . Similarly, the Sixth Circuit concluded that "Justice Sotomayor’s opinion is the narrowest ground for the Court's decision and thus represents the Court’s holding in Freeman" without citing Marks. Smith, 658 F.3d at 611 . The Second Circuit has considered the issue only in an unpublished summary order. United States v. White, 429 Fed.Appx. 43, 47 (2d Cir. 2011) (finding that Justice Sotoma-yor’s concurrence is controlling). . We note that in King, the D.C. Circuit explicitly stated that it was not “free to combine a dissent with a concurrence to form a Marks majority.” King, 950 F.2d at 783 . We emphasize here, however, that we do not decide that issue. . By preserving all Guidelines calculations other than the one that was retroactively reduced, the policy statements in section IB 1.10(b) substantially limit district court discretion when ruling on § 3582(c)(2) motions. Freeman, 454 U.S. at 531-32, 102 S.Ct. 781 (plurality opinion). Further, district courts cannot vary below the Guidelines in a § 3582(c)(2) proceeding, as they can in an initial sentencing proceeding. Id.', USSG § IB 1.10(b)(2)(A). And, given the availability of appellate review, any concern that the plurality’s approach will "upset the bargain struck between prosecutor and defendant” is overstated. Freeman, 564 U.S. at 531 , 131 S.Ct. 2685 (plurality opinion). . We emphasize that our decision merely removes ’the jurisdictional hurdle that led the district court to deny Davis a resentencing hearing. On remand, “[i]f the district court, based on its experience and informed judgement, concludes the [Rule 11(c)(1)(C)] agreement led to a more lenient sentence than would otherwise have been imposed, it can deny the motion, for the statute permits but does not require the court to reduce a sentence.” Freeman, 564 U.S. at 532 , 131 S.Ct. 2685 (plurality opinion). 


[Concurrence by Christen]

 CHRISTEN, Circuit Judge, joined by THOMAS, Chief Judge, and TALLMAN, NGUYEN, and HURWITZ, Circuit Judges, concurring:. Freeman v. United States, 564 U.S. 522 , 131 S.Ct. 2685 , 180 L.Ed.2d 519 (2011), addressed an issue of grave importance to Davis and to countless other prisoners in his position. The opinion issued today corrects an error in our circuit’s interpretation of Freeman , but it also represents a missed opportunity to straighten out our circuit’s inconsistent applications of Marks v. United States, 430 U.S. 188 , 97 S.Ct. 990 , 51 L.Ed.2d 260 (1977). Marks specifically directs lower courts how to interpret splintered Supreme Court decisions. Its rule tends to crop up in the most contentious cases where, as here, the stakes are significant. Freeman is important, but Marks has even broader application to the wide spectrum of issues we decide. I join in the court’s holding— as far as it goes — but it is regrettable that our court articulates an incomplete interpretation of Marks . Leaving this work unfinished will surely result in continued uneven application of Marks within our circuit. The rule announced in Marks appears simple at first glance but it has proven to be confounding. See Grutter v. Bollinger, 539 U.S. 306, 325 , 123 S.Ct. 2325 , 156 L.Ed.2d 304 (2003) (noting Marks has “baffled and divided the lower courts that have considered it” (quoting Nichols v. United States, 511 U.S. 738, 746 , 114 S.Ct. 1921 , 128 L.Ed.2d 745 (1994))). The opinion issued today untangles part of the problem because it decisively adopts a reasoning-based approach to determine when splin *1029 tered decisions produce binding precedent. Under this approach, I agree that no binding rule emerges from Freeman . I also agree that Justice Kennedy’s plurality opinion is the best framework for analyzing motions for sentence reduction in the context of Rule 11(c)(1)(C) plea agreements. Unfortunately, we leave unanswered whether our court will take into account dissenting opinions when applying Marks . I join the majority because its holding is entirely consistent with Marks : “[W]e look to those opinions that concurred in the judgment and determine whether one of those opinions sets forth a rationale that is the logical subset of other, broader opinions.” But I disagree with the majority’s assumption that we might be free to take dissenting opinions into account in future Marks analyses. Marks leaves some questions unanswered, but it plainly limits our review to the opinions of “those Members [of the Court] who concurred in the judgments.” Marks, 430 U.S. at 193 , 97 S.Ct. 990 (emphasis added). Because I do not see that this language leaves any room for our court to consider dissenting opinions, I would go further than the majority does and expressly state that dissents play no role in a Marks analysis. This is not to say that dissents serve no purpose. They can and should be read to provide context and a deeper understanding of the Court’s decisions, but they do not inform our analysis of what binding rule, if any, emerges from a fractured decision. The dissent points to National Mutual Insurance Co. v. Tidewater Transfer Co., 337 U.S. 582 , 69 S.Ct. 1173 , 93 L.Ed. 1556 (1949), as support for its view that dissenting opinions should be considered. Tidewater, of course, says nothing about how to interpret fractured Supreme Court decisions, though it was a fractured decision itself. In Tidewater, two concurring justices and four dissenting justices relied on the rule that Congress lacks authority to expand federal court subject matter jurisdiction beyond that provided in Article III. See id. at 604-46, 69 S.Ct. 1173 . Our dissenting colleague is correct that courts have universally accepted this rule, but doing so does not require looking to Tidewaters, dissenting opinions. Indeed, as recognized in Verlinden B.V. v. Central Bank of Nigeria, 461 U.S. 480 , 103 S.Ct. 1962 , 76 L.Ed.2d 81 (1983), this rule pre-dates Tidewater by a long shot. See id. at 491 , 103 S.Ct. 1962 (“This Court’s cases firmly establish that Congress may not expand the jurisdiction of the federal courts beyond the bounds established by the Constitution.” (citing Hodgson v. Bowerbank, 9 U.S. 303 , 5 Cranch 303 , 3 L.Ed. 108 (1809); Kline v. Burke Constr. Co., 260 U.S. 226, 234 , 43 S.Ct. 79 , 67 L.Ed. 226 (1922))); see also Seminole Tribe of Fla. v. Florida, 517 U.S. 44, 65 , 116 S.Ct. 1114 , 134 L.Ed.2d 252 (1996) (describing as “fundamental” that “Congress could not expand the jurisdiction of the federal courts beyond the bounds of Article III” (citing Marbury v. Madison, 5 U.S. 137 , 1 Cranch 137 , 2 L.Ed. 60 (1803))). The Supreme Court at times looks to dissenting opinions when interpreting its own splintered decisions. See United States v. Jacobsen, 466 U.S. 109, 115 , 104 S.Ct. 1652 , 80 L.Ed.2d 85 (1984); Moses H. Cone Mem’l Hosp. v. Mercury Const. Corp., 460 U.S. 1, 16-17 , 103 S.Ct. 927 , 74 L.Ed.2d 765 (1983). From this, the majority and dissent infer that we might be free to do the same. But the way the Supreme Court treats its own precedent says nothing about how lower courts must apply it. Marks , not Jacobsen or Moses H. Cone, is the Supreme Court authority that sets out the rule for lower courts to follow. 1 In my *1030 view, until the Supreme Court says otherwise, Marks precludes us from considering dissenting opinions. . The dissent responds by pointing to the Supreme Court’s language in Moses H. Cone: "[T]he [Fourth Circuit] Court of Appeals cor *1030 rectly recognized that the four dissenting Justices and Justice BLACKMUN [in Will v. Calvert Fire Insurance Co., 437 U.S. 655 , 98 S.Ct. 2552 , 57 L.Ed.2d 504 (1978)] formed a majority to require application of the Colorado River test.” Moses H. Cone, 480 U.S. at 17, 107 S.Ct. 971 . This was merely a recognition that, in Will, there were not enough votes to undermine Colorado River Water Conservation District v. United States, 424 U.S. 800 , 96 S.Ct. 1236 , 47 L.Ed.2d 483 (1976). Thus, the controlling rule the Fourth Circuit applied came from Colorado River, not Will. Moses H. Cone does not direct lower courts to look to dissenting opinions when divining a controlling rule from a fractured Supreme Court decision. 


[Dissent by Bea]

 BEA, Circuit Judge, dissenting: From its very first sentence, the majority unjustifiably departs from not only our own, but also well-established Supreme Court precedent. We correctly and squarely resolved the questions presented by this' case in United States v. Austin, 676 F.3d 924 (9th Cir. 2012) until it was overruled by today’s majority. In Austin , we considered whether a judge had jurisdiction to modify a prisoner’s sentence under 18 U.S.C. § 3582 (c)(2) (permitting modification of a term of imprisonment where that term was “based on” a sentencing range which was later reduced) if the prisoner was sentenced pursuant to a Federal Rules of Criminal Procedure (“F.R.C.P.”) 11(c)(1)(C) plea agreement (a “Rule 11(c)(1)(C) agreement”) that did not expressly incorporate the since-amended sentencing Guidelines. 1 We noted that the Supreme Court had spoken to this very question in Freeman v. United States, 564 U.S. 522 , 131 S.Ct. 2685 , 180 L.Ed.2d 519 (2011), where five Justices had voted to permit a prisoner to seek sentence modification, but where no single rationale had commanded a five-Justice majority. A four-justice plurality of those five Justices would always permit a prisoner to seek sentence modification, under the rationale that a judge’s approval of the plea agreement is required to start with, and is necessarily “based on,” the since-amended Guidelines. Justice Sotomayor, writing only for herself, agreed that Freeman was entitled to seek resentencing, but only because his plea agreement had expressly incorporated the since-modified sentencing Guidelines. A four-Justiee dissent penned by Chief Justice Roberts would find that Rule 11(c)(1)(C) agreements are always purely contractual in nature and therefore never “based on” the sentencing Guidelines. Given this 4-1-4 split, we held in Austin that we were bound (under Marks v. United States, 430 U.S. 188, 193 , 97 S.Ct. 990 , 51 L.Ed.2d 260 (1977)) to treat Justice Sotomayor’s concurrence as the “holding” of the Freeman Court because it was the “narrowest grounds” upon which to reach the disposition that commanded a majority of the Court. See Austin, 676 F.3d at 927-28 . Our interpretation in Austin has garnered the support of eight out of the nine Circuits which have interpreted Freeman. See, e.g., United States v. Graham, 704 F.3d 1275, 1278 (10th Cir. 2013); United States v. Thompson, 714 F.3d 946, 949 (6th Cir. 2013); United States v. Browne, 698 F.3d 1042, 1045-47 (8th Cir. 2012); United States v. Weatherspoon, 696 F.3d 416, 422 (3rd Cir. 2012); United States v. Dixon, 687 F.3d 356, 359-60 (7th Cir. 2012); United States v. Austin, 676 F.3d 924, 927 (9th Cir. 2012); United States v. Rivera-Martinez, 665 F.3d 344, 345 (1st Cir. 2011); United States v. Brown, 653 F.3d 337 (4th *1031 Cir. 2011). 2 The sole outlier circuit: the D.C. Circuit in United States v. Epps, 707 F.3d 337 (D.C. Cir. 2013). Thus, the Majority today rejects a widely accepted interpretation of Freeman in favor of a highly criticized, outlier approach, thus accentuating a Circuit split. I cannot subscribe to the Majority’s view. To start, the Majority’s “logical subset” requirement is an invention of the D.C. Circuit that finds no support in Marks or any other Supreme Court precedent. The Majority’s “logical subset” invention permits a concurring opinion to become the precedential decision of the Court if, and only if, its reasoning shares all points in common with another, broader opinion that also reaches the majority result. Of course, the concurring opinion may have fewer elements of the decision than does the plurality opinion, but it may not have any elements in conflict. See Maj. Op. at 1021-22. This notion is an invention in our circuit’s jurisprudence, though the Majority tips its hat to the D.C. Circuit. See Maj. Op. at 1016-17, 1020. And even if there were a “logical subset” requirement as defined by the Majority, the Majority misreads Justice Kennedy’s plurality opinion to the extent it concludes that there are circumstances in which Justice Soto-mayor would permit sentence modification but the Kennedy plurality would not. Finally, the Majority’s adoption of the Kennedy plurality’s approach violates stare de-cisis because five Justices in Freeman (a majority), all agreed that we look to the plea agreement itself to determine whether a plea was “based on” the since-modified sentencing Guidelines. Under eases like National Mutual Insurance Co. v. Tidewater Transfer Co., 337 U.S. 582 , 69 S.Ct. 1173 , 93 L.Ed. 1556 (1949), we are bound by holdings that garner the support of a “majority” of the nine Justices on the entire Court, even if that agreement derives in part from votes from the dissent. Thus, the Majority flouts not only Freeman , but also Supreme Court jurisprudence relating, to the binding effect of splintered Supreme Court opinions, as well as this Court’s structural role as a federal intermediary court. I. Facts In 2005, Tyrone Davis pleaded guilty to possession with intent to distribute crack cocaine pursuant to a Rule 11(c)(1)(C) plea agreement. In Davis’s case, the plea agreement did not specifically.mention any particular sentencing Guideline. Nor did the plea agreement itself calculate (or even contain sufficient facts with which to calculate) Davis’s Guidelines range. True, it contained some of the factors that would enable a Guidelines calculation. For example, the parties stipulated to a base offense level of 34. But the agreement failed to list a criminal history category or adjustment determinations — both of which are essential to calculate a sentencing range under the Guidelines. After successive appeals to this Court on grounds no longer relevant, the district court calculated a Guidelines range of 188-235 and approved the 216-month sentence in Davis’s plea agreement. 3 We affirmed. See United States v. *1032 Davis, 389 Fed.Appx. 616 (9th Cir. 2010) (unpublished). Congress thereafter passed the Fair Sentencing Act of 2010, which increased the threshold amount of cocaine base necessary to trigger an enhanced Guidelines range. Pub L. 111-220, § 2(a), 124 Stat. 2372 . Under the new Guidelines, the amount of cocaine base that contributed to Davis’s convictions would produce a Guidelines range of only 97-121 months (after inserting the calculations made by the district court at Davis’s resentencing). Because Davis’ 216-month sentence now falls much above that range, Davis moved in September 2012 for resentencing under 18 U.S.C. § 3582 (c)(2), relying on the amended Guidelines. Section 3582(c)(2) provides: In the case of a defendant who has been sentenced to a term of imprisonment based on a sentencing range that has subsequently been lowered by the Sentencing Commission ... the court may reduce the term of imprisonment, after considering the factors set forth in section 3553(a) to the extent that they are applicable, if such a reduction is consistent with applicable policy statements issued by the Sentencing Commission. Id. (emphasis added). The district court properly denied the motion, ruling that Davis’s 216-month sentence was “based on” his plea agreement, not on the Guidelines range that had since been lowered. The district court relied on our decision in United States v. Austin, 676 F.3d 924, 927-28 (9th Cir. 2011), to hold that it lacked jurisdiction to review Davis’ sentence, which was adopted pursuant to a Rule 11(c)(1)(C) agreement that did not reference any Guidelines range, and therefore did not meet either of Justice Sotoma-yor’s exceptions in Freeman . The original panel affirmed, citing Austin as the controlling law of the circuit. United States v. Davis, No. 13-30133, op. at 1016-17. 4 This case was successfully called en banc to reconsider our prior determination that, under the methodology prescribed by Marks , Justice Sotomayor’s concurrence in Freeman constitutes the binding “holding” of that case. II. Legal Analysis A. In Marks v. United States, the Supreme Court made clear that even splintered determinations of our highest court are binding on lower federal courts: “When a fragmented Court decides a case and no single rationale explaining the result enjoys the assent of five Justices, ‘the holding of the Court may be viewed as that position taken by those Members who concurred in the judgment on the narrowest grounds....’” Marks v. United States, 430 U.S. 188, 193 , 97 S.Ct. 990 , 51 L.Ed.2d 260 (1977) (reversing the Sixth Circuit’s determination that Memoirs v. Massachu *1033 setts, 383 U.S. 413 , 86 S.Ct. 975 , 16 L.Ed.2d 1 (1966), had no binding prece-dential effect because it was a plurality opinion). At issue in Marks was the prece-dential effect of Memoirs — an earlier, splintered Supreme Court opinion. In Memoirs , a three-Justice plurality had held that sexually explicit literature was constitutionally protected unless it met the three-part definition of obscenity set forth in Roth v. United States, 354 U.S. 476 , 77 S.Ct. 1304 , 1 L.Ed.2d 1498 (1957). 5 See Marks, 430 U.S. at 193 , 97 S.Ct. 990 (citing Memoirs, 383 U.S. at 421 , 86 S.Ct. 975 ). Justices Black and Douglas (both writing separately) concurred in Memoirs on the broader grounds that the First Amendment prohibits government censorship of any “obscene” material. See Memoirs, 383 U.S. at 421, 424-33 , 86 S.Ct. 975 . Finally, Justice Stewart concurred based on his somewhat different view that only “hardcore pornography” may constitutionally be suppressed. Id. at 421 , 86 S.Ct. 975 (citing his dissenting opinion in Mishkin v. State of N.Y., 383 U.S. 502, 518 , 86 S.Ct. 958 , 16 L.Ed.2d 56 (1966)). In sum, six Justices agreed that the material at issue in Memoirs was protected by the First Amendment, but no five Justices agreed about the scope of First Amendment protection for sexually explicit material — nor about the proper reasoning to be employed to reach that result. In this circumstance, Marks explained, the three-Justice Memoirs plurality opinion, which applied the Roth tests, “constituted the holding of the Court and provided the governing standards,” because it was the “narrowest grounds” for finding First Amendment protection. Marks, 430 U.S. at 193-94 , 97 S.Ct. 990 . The Majority today rejects the only application of Marks to Freeman that is consistent with stare decisis in favor of a widely-criticized approach endorsed by one panel in the D.C. Circuit in United States v. Epps, 707 F.3d 337, 351 (D.C. Cir. 2013). 6 The Majority holds that Marks produces a “controlling” opinion only when the “narrowest grounds” in a splintered opinion is “represents] a common denominator of the Court’s reasoning,” meaning “the reasoning of a narrower opinion fit[s] entirely into the circle drawn by a broader opinion.” Maj. Op. at 1021. *1034 The Majority’s adoption of a reasoning-based, “common denominator” approach is fundamentally inconsistent with Marks itself. In Memoirs , none of Justices Stewart, Black, or Douglas agreed with the rule enumerated by the Justice Brennan plurality (that the Court should apply the Roth test to determine whether speech is constitutionally protected). Thus, there was no “common denominator” in Memoirs , as that term is defined by the Majority today. Yet, Marks applied the three-justice plurality opinion authored by Justice Brennan. It is true, of course, that in every circumstances in which the Brennan plurality would find speech protected under the Roth obscenity test, Justices Black and Douglas (who ascribe to the broader view that obscene speech is always constitutionally protected) would agree with the plurality’s result. But there is a distinction between agreement with a result and agreement with the reasoning that leads one to adopt that result that appears to be lost on my colleagues in the Majority. Indeed, the Majority’s own reasoning supports only a votes-based reading of Marks : Justice Brennan’s plurality opinion was the “narrowest grounds” for the Court’s holding in Memoirs because it would always produce a result with which at least five Justices would agree. 7 This reading is also consistent with Marks’ dictate that “the holding of the Court ... [is] the position taken by those Members who concurred in the judgment on the narrowest grounds.... ” Marks, 430 U.S. at 193 , 97 S.Ct. 990 (emphasis *1035 added). Marks’ emphasis on the Court’s “judgment” demonstrates that it is the ultimate “vote” of five Justices that is important in determining the binding effect of a splintered Supreme Court opinion. That is, Marks requires us to find a “legal standard which, when applied, will necessarily produce results with which a majority of the Court from that case would agree.” United States v. Williams, 435 F.3d 1148 , 1157 n. 9 (9th Cir. 2006) (emphasis added). That is not to say, of course, that the respective rationales of a splintered Supreme Court decision are irrelevant. Consideration of competing rationales is necessary to determine which would consistently produce a result with which a majority of the Court would agree. But the Majority’s reasoning-based approach must be incorrect, because there was no common reasoning in Memoirs , yet the Marks Court was nonetheless able to derive from Memoirs a binding rule, which it applied in Marks . The Majority’s adoption of a “logical subset” precondition to Marks applicability is plagued by the same logical fallacy. King v. Palmer, the D.C. Circuit case on which the Majority relied, justified its invention of a logical subset requirement on the grounds that: Marks is problematic ] [i]f applied in situations where the various opinions supporting the judgment are mutually exclusive [because] Marks will turn a single opinion that lacks majority support into national law. When eight of nine Justices do not subscribe to a given approach to a legal question, it surely cannot be proper to endow that approach with controlling force.... ” King v. Palmer, 950 F.2d 771, 782 (D.C. Cir. 1991). Even accepting, arguendo, the King court’s premise that a concurring opinion should be given stare decisis effect only when it consistently produces a result with which a majority of the Court would agree, that, again, would support the adoption of a rule that it is each Justice’s vote, and not his reasoning, that counts under Marks . The King court’s conclusion that Marks “works” only when a majority of Justices “subscribe to a given approach to a legal question,” such that “one opinion supporting the judgment ... fit[s] entirely within a broader circle drawn by the others,” id. at 782, simply does not follow from that court’s premise — or from the many Supreme Court precedents interpreting and applying Marks to splintered opinions over the last four decades. 8 Indeed, to require complete overlap between both the result and the reasoning of Justices in the majority before a binding rule can be discerned renders Marks a virtual nullity. Agreement as to both the Court’s reasoning and its result does not produce a concurring opinion — it produces a “join.” *1036 And even if a reasoning-based approach to Marks were not fundamentally incompatible with Marks itself, the idea that a court’s holding, adopted by a majority of judges, must have a rationale common throughout the majority is a novelty to any branch of our government, including the judiciary. Even our courts adhere to that most democratic of principles: as to how to decide this ease, the majority rules. People, including legislators and judges, may vote for a result for a variety of different reasons. That is how coalitions are achieved and compromises made to reach results although the distinct motives and thinking which produced the majority’s result remain quite distinct. It is the result produced by majority vote that determines the stare decisis effect of the judgment. That is because whether the majority voté is produced by the adoption of one rationale or two, the rule of law made — the decision' — 'is based on a rationale or rationales expected to remain the same and produce the same result in the next applicable case. After all, “stare decisis” means “to stand by things decided.” B. A simple application of Marks’ methodology to Freeman compels a finding that Justice Sotomayor’s concurrence is the “holding” of Freeman. See United States v. Austin, 676 F.3d 924, 927-28 (9th Cm. 2012). Five members of the Court agreed that Freeman — who had been sentenced pursuant to a Rule 11(c)(1)(C) agreement — was eligible for sentencing modification under 18 U.S.C. § 3582 (c)(2), because his plea agreement had been “based on” a subsequently modified sentencing Guidelines range. Freeman v. U.S., 564 U.S. 522, 534, 544 , 131 S.Ct. 2685 , 180 L.Ed.2d 519 (2011). Writing for a four-member plurality, Justice Kennedy reasoned that a plea agreement is “based on” applicable Guidelines whenever the sentencing judge at least consulted those guidelines before approving the proposed sentence — which, Justice Kennedy explained, the judge is statutorily “required” to do in “every case.” See Freeman, 564 U.S. at 529 , 131 S.Ct. 2685 (plurality opinion). Concurring in result, Justice Sotomayor, a former district court judge experienced in actual sentencing, reasoned that plea agreements are sometimes based on sentencing guidelines, but only when the agreement itself “expressly uses a Guidelines sentencing range applicable to the charged offense to establish the term of imprisonment,” or the sentencing range is otherwise “evident from the agreement itself.” Id. at 534, 539 , 131 S.Ct. 2685 (Soto-mayor, J., concurring) (emphasis added). Chief Justice Roberts, writing for the four dissenting Justices, “agree[d] with Justice Sotomayor that ‘the term of imprisonment imposed pursuant to a [Rule 11(c)(1)(C) ] agreement is ... based on the agreement itself’ ” Id. at 544, 131 S.Ct. 2685 (Roberts, C.J., dissenting) (emphasis added) (internal quotation marks omitted). However, the dissent would find that plea agreements are a matter of contract and thus never “based on” the sentencing Guidelines. Id. at 544-51 , 131 S.Ct. 2685 (Roberts, C.J., dissenting). Justice Sotomayor’s opinion is controlling because “ ‘sometimes’ is the middle ground between ‘always’ and ‘never.’ ” See United States v. Duvall, 740 F.3d 604, 612 (D.C. Cir. 2013) (Kavanaugh, J., concurring in the denial of rehearing en banc); see also supra, n.6. In circumstances in which Justice Sotomayor would permit reduction of a prior sentence, so too would the plurality (resulting in a five-Justice majority). Where Justice Sotomayor’s criterion are not met, she would find agreement in the four-Justice dissent that the prisoner’s sentence is not “based on” the *1037 Guidelines (which would also result in a five-justice majority). Justice Sotomayor’s approach therefore constitutes the “narrowest grounds” for reaching a result that, in any circumstance, will be consistent with the result that a majority of the Supreme Court would reach under Freeman . Under Justice Sotomayor’s framework, Davis cannot seek resentencing, because his plea agreement does not meet either of her exceptions. It neither expressly cites, nor otherwise manifests that it is predicated upon, any particular Guidelines range. In fact, it omits several details (such as criminal history, and adjustments) necessary even to calculate a Guidelines range. Davis’s sentence is therefore not subject to modification under § 3582(c)(2). The district court correctly determined that it lacked jurisdiction to resentence Davis, and the panel should affirm on that basis. The Majority rejects this straight-forward approach on the grounds that circumstances could arise in which Justice Sotomayor would find a plea “based on” sentencing guidelines, but the Kennedy plurality would not. The Majority posits two hypothetical, both of which assume express agreement in a plea bargain that a particular sentencing range applies (such that Justice Sotomayor would find the plea agreement “based on” the sentencing Guidelines, and subject to § 3582(c)(2) re-sentencing). See Maj. Op. at 1022-23. Both hypothetical then posit that the “sentencing court ... might consider and reject the guideline range used by the parties” — in one scenario because the judge believed another range should apply, and, in the other, for “policy” reasons. Id. The Majority suggests that in either of these circumstances, the Freeman plurality would not find the plea agreement “based on” the sentencing guidelines, and thus would not grant relief. The Majority is simply incorrect. The very fact that the sentencing judge in the Majority’s hypothetical must reject the Guidelines range recommended by the parties necessarily presupposes that the judge’s analysis started with a consideration of the Guidelines range recommended in the plea agreement. Under the Kennedy plurality’s approach, this consideration, at the inception of the sentencing, is enough to entitle a defendant to seek resentencing — regardless of the judge’s ultimate reasons for approving the plea agreement. See Freeman, 564 U.S. at 529-30 , 131 S.Ct. 2685 (plurality opinion)' (“[I]f the judge uses the sentencing range as the beginning point to explain the decision to deviate from it, then the Guidelines are in a real sense a basis for the sentence.” (emphases added)). The Majority recognizes as much on page 1026 of its opinion, wherein it quotes Justice Kennedy’s statement that “the applicable Guidelines policy statement ‘forbids the district judge to accept an 11(c)(1)(C) agreement without first evaluating the recommended sentence’ under the Guidelines.” Maj. Op. at 1026 (quoting Freeman, 564 U.S. at 529 , 131 S.Ct. 2685 , and USSG § 6B1.2(c)). The Majority criticizes my reading of Justice Kennedy’s plurality opinion — a reading adopted by an overwhelming majority of circuits — as “oversimplified.” In support of its more limited reading, the Majority relies solely on Justice Kennedy’s statement that a “recommended sentence is likely to be based on the Guidelines.” Maj. Op. at 1022 n. 9 (quoting Freeman, 564 U.S. at 534 , 131 S.Ct. 2685 ). But Justice Kennedy’s use of the word “likely” in one sentence cannot be read in isolation. In the immediately preceding paragraph, Justice Kennedy in fact rejects the notion — advanced by the Majority— that his approach would limit relief to only a “subset of defendants.” Freeman, 564 U.S. at 533-34 , 131 S.Ct. 2685 (“[When] [t]he Commission determine[s] that [the] *1038 Guidelines [are] flawed, and therefore that sentences that relied on them ought to be reexamined[,] [tjhere is no good reason to extend the benefit of the Commission’s judgment only to an arbitrary subset of defendants whose sentences were accepted in light of a since-rejected Guidelines range based on whether their plea agreements refer to the Guidelines.”). Indeed, Justice Kennedy criticizes Justice Sotoma-yor’s approach because it would fail to permit resentencing in all cases: “Congress enacted § 3582(c)(2) to remedy systemic injustice, and the approach outlined in [Justice Sotomayor’s] opinion concurring in the judgment would undercut a systemic solution.” Id. at 534, 131 S.Ct. 2685 (emphasis added). Nothing about Justice Kennedy’s opinion suggests any exceptions. He notes that the “Guidelines require the district judge to give due consideration to the relevant sentencing range, even if the defendant and prosecutor recommend a specific sentence as a condition of the guilty plea.” Id. at 530 , 131 S.Ct. 2685 (emphases added). He further reasons that “[federal sentencing law requires” the sentencing judge to look to the Guidelines as “a framework or starting point” in “every case.” Id. at 529 , 131 S.Ct. 2685 . Thus, notwithstanding his use of the word “likely” in one sentence, Justice Kennedy’s opinion is most reasonably read as endorsing an approach under which a defendant may “always” seek re-sentencing on the basis of amended Guidelines. Certainly, for the reasons already stated above, Justice Kennedy would permit a defendant to seek resentencing in the examples given by the Majority. The Majority imbues far more meaning into Justice Kennedy’s single use of the word “likely” than the rest of Justice Kennedy’s plurality opinion can bear. It may be that Justice Kennedy simply did not want to speak in absolutes. That is, he declined to say, as a matter of empirical fact, that a judge always consults the sentencing Guidelines, because there is always the possibility that a judge could make a mistake or fail to follow the law. But one thing is for sure: Justice Kennedy does not even hint at a case in which the sentencing judge could lawfully start sentencing with any consideration other than the Guidelines, and the Majority has not suggested any either. See 18 U.S.C. § 3553 (directing that a “court, in determining the particular sentence to be imposed, shall consider ... the sentencing range established for ... the applicable category of offense committed by the applicable category of defendant as set forth in the guidelines” (emphasis added)); see also Gall v. United States, 552 U.S. 38, 49 , 128 S.Ct. 586 , 169 L.Ed.2d 445 (2007) (“[A] district court should begin all sentencing proceedings by correctly calculating the applicable Guidelines range. As a matter of administration and to secure nationwide consistency, the Guidelines should be the starting point and the initial benchmark.” (citing Rita v. United States, 551 U.S. 338, 347-48 , 127 S.Ct. 2456 , 168 L.Ed.2d 203 (2007))). Indeed, the failure of a sentencing judge to start the calculation of a sentence by considering the applicable sentencing Guidelines is in itself grounds for reversal for resentencing. See Gall, 552 U.S. at 51 , 128 S.Ct. 586 (instructing that appellate courts must “first ensure that the district court committed no significant procedural error, such as failing to calculate ... the Guidelines range”); United States v. Denton, 611 F.3d 646, 651 (9th Cir. 2010) (explaining that “[a] failure to calculate the correct advisory range constitutes procedural error” justifying reversal and remand for resentencing); United States v. Hammons, 558 F.3d 1100, 1106 (9th Cir. 2009) (holding that a sentencing court “committed plain error by failing to ... calculate the applicable] guideline range” *1039 and vacating and remanding for resentenc-ing). The Majority’s contrary analysis appears to substitute the Freeman plurality’s requirement that a trial judge “consider” the Guidelines with its own innovation— that the trial judge must base his ultimate acceptance of the plea agreement on the Guidelines in order for a defendant to be entitled to seek resentencing. See Maj. Op. at 1022-23. But the latter is not the test enumerated by Justice Kennedy in Freeman . Properly read, Justice Kennedy’s opinion would unquestionably permit re-sentencing in the hypotheticals offered by the Majority. See id. Thus, even if the Majority were correct that Marks applies only when one opinion is a “logical subset” of another, that precondition would be met here. C. But even putting aside the “logical subset” issue, the Majority still cannot reach its result consistent with basic principles of stare decisis for the independent reason that we, as a federal intermediate court, are bound by holdings upon which five Justices of the Court agree — even if that agreement derives in part from dissenting Justices. The Supreme Court’s fragmented decision in National Mutual Insurance Co. v. Tidewater Transfer Co., 337 U.S. 582 , 69 S.Ct. 1173 , 93 L.Ed. 1556 (1949), is a famous illustration of this principle. 9 *1040 Since Tidewater, courts have universally accepted that Congress may not expand the scope of subject-matter jurisdiction conferred by Article III through passage of a Congressional Act. 10 Yet this rule can be divined only by combining a two-Justice concurrence in Tidewater with a four-justice dissent. See id. at 604-46, 69 S.Ct. 1173 . United States v. Jacobsen, 466 U.S. 109 , 104 S.Ct. 1652 , 80 L.Ed.2d 85 (1984), provides a more recent example of the same rule. The question in Jacobsen was the government’s authority to conduct a war-rantless search on the heels of a private search that identified potential contraband. The Jacobsen Court extracted the “controlling” legal standard from its prior precedent in Walter v. United States, 447 U.S. 649 , 100 S.Ct. 2395 , 65 L.Ed.2d 410 (1980), by combining the opinion of the Walter Court (which garnered only two votes) with the opinion of four dissenting Justices, which it described as the “standard ... adopted by the majority of the Court in Walter....” Id. at 116-17, 104 S.Ct. 1652 & n. 12 (emphasis added). In Walter , a private party had opened a package containing films that, from the descriptions on the packaging, the private party concluded were contraband. Walter v. United States, 447 U.S. 649, 651-52 , 100 S.Ct. 2395 , 65 L.Ed.2d 410 (1980). The government seized the films and, without obtaining a warrant, screened them from a projector. Id. at 652 , 100 S.Ct. 2395 . Delivering the two-Justice opinion of the Court, Justice Stevens reasoned that the government had violated the defendant’s Fourth Amendment rights by actually watching films because the private party’s search had consisted only of opening the package that contained the films. Id. at 657 , 100 S.Ct. 2395 (The FBI’s subsequent screening of such films constituted an “expansion of the search that had been conducted previously by the private party.”); see also Jacobsen, 466 U.S. at 115-16 , 104 S.Ct. 1652 (quoting Walter, 447 U.S. at 657 , 100 S.Ct. 2395 (Opinion of Stevens, J., joined by Stewart, J.)). Three Justices in Walter concurred in the judgment on the grounds that the government had exceeded its authority under the “plain view” doctrine, but expressly rejected the notion that the scope of one’s Fourth Amendment right could be tethered to the scope of an antecedent private search. Walter, 447 U.S. at 660-62 , 100 S.Ct. 2395 (White, J., concurring). A four-justice dissent agreed with Justice Stevens that the legality of a governmental search depended on the scope of the private party’s antecedent search, but would have found no constitutional violation because “the FBI’s subsequent viewing of the movies on a.projector did not ‘change the nature of the search’ and [thus] was not an additional search subject to the warrant requirement.” Id. at 663-64 , 100 S.Ct. 2395 (Blackmun, J., dissenting); *1041 see also Jacobsen, 466 U.S. at 116 , 104 S.Ct. 1652 . Presented with these competing views in Walter , the Jacobsen Court (in a six-Justice opinion of the Court) held that “a majority [in Walter] did agree on the appropriate analysis of a governmental search which follows on the heels of a private one. Two Justices [referring to Justices Stevens and Stewart].... [and] [f]our additional Justices [referring to the dissent] were ... of the view that the legality of the governmental search must be tested by the scope of the antecedent private search.” Jacobsen , 466 U.S. at 115— 16, 104 S.Ct. 1652 . The Majority opinion here simply cannot be squared with the Court’s reading of Walter in Jacobsen. Ja-cobsen recognized that the rule adopted by the two-Justice plurality in Walter was the precedential holding of the Walter Court, because it garnered the approval of six Justices (a majority) of the Court. This was so even though the three-Justice concurrence specifically rejected the plurality’s rationale, and thus neither the plurality opinion nor the concurrence was a “logical subset” of the other. Moses H. Cone Mem’l Hosp. v. Mercury Const. Corp., 460 U.S. 1 , 103 S.Ct. 927 , 74 L.Ed.2d 765 (1983) [hereinafter, “Memorial Hospital”] provides yet another example. There, the Court considered whether a lower court was bound to apply the Colorado River test, 11 notwithstanding that a four-justice plurality in Will v. Calvert Fire Insurance Co., 437 U.S. 655 , 98 S.Ct. 2552 , 57 L.Ed.2d 504 (1978), had purported to overrule it. Id. at 17, 103 S.Ct. 927 . The Memorial Hospital Court affirmed that the “Court of Appeals [had] correctly recognized that the four dissenting Justices and Justice Blackmun [who concurred in judgment in Will] formed a majority to require application of the Colorado River test.” Id. (emphases added). By holding that the Fourth Circuit had “correctly recognized” that it was “require[d]” to apply the Colorado River test by virtue of a five-Justice majority comprised of four dissenting Justices and one concurring Justice, id. the Supreme Court in Memorial Hospital confirmed that its precedents relating to the consideration of dissenting opinions do bind us as a federal intermediate court (contrary to the suggestion of my concurring colleagues). The Majority is correct that the Justices in Freeman did not agree on much. But a five-Justice majority (Justice Sotomayor, plus the four dissenting Justices) did agree on one point — that a sentence imposed under a Rule 11(c)(1)(C) plea agreement is always “based on the [plea] agreement” itself. See, e.g., Freeman, 564 U.S. at 534 , 131 S.Ct. 2685 (Sotomayor, J., concurring) (“[T]he term of imprisonment imposed by a district court pursuant to an agreement *1042 authorized by Federal Rule of Criminal Procedure 11(c)(1)(C) ... agreement ] is ‘based ori the agreement itself, not on the judge’s calculation of the Sentencing Guidelines.”); see also id. at 544 , 131 S.Ct. 2685 (Roberts, C.J., dissenting) (“I agree with Justice SOTOMAYOR that ‘the term of imprisonment imposed pursuant to a [Rule 11(c)(1)(C)] agreement is, for purposes of § 3582(c)(2), “based on” the agreement itself.’ ” (quoting Justice Soto-mayor’s concurrence, id. at 534, 131 S.Ct. 2685 )). This was a holding that received the vote of five Justices (a majority) of the Court. Like the Fourth Circuit in Memorial Hospital, we are bound by that holding. 12 The Majority blatantly ignores Chief Justice Roberts’ express agreement with Justice Sotomayor and focuses only on the disagreements between them. But of course there are points on which they disagree; that is why there is both a concurrence and a dissent in Freeman (just as there was ample disagreement between the concurring and dissenting Justices in Tidewater). But those disagreements do not negate the fact that there are no sentence reductions which Justice Sotomayor would deny that the four dissenting Justices would not also deny. Where, as here, a plea agreement contains no mention of either the sentencing Guidelines or the criteria'necessary to calculate the applicable Guidelines range (Justice Sotomayor’s *1043 two exceptions), five Justices in Freeman (Justices Sotomayor, Roberts, Scalia, Thomas, and Alito) would always vote to deny the defendant’s petition to seek re-sentencing. Under Marks , we are bound by this result. In sum, the Majority makes a good case that “federal sentencing law,” Rule 11(c)(1)(C), and the Guidelines’ policy statements all support the view adopted by the Justice Kennedy plurality in Freeman. See Maj. Op. at 1026-27. And these arguments may well be the basis for a future Supreme Court opinion abrogating Freeman and adopting outright the plurality opinion of Justice Kennedy. Rut that is the Court’s province, not ours. As an intermediate federal court, we are not free to disregard binding Supreme Court precedent simply because we can think of a rule we like better. The purpose of determining a “holding” is to apply stare decisis in decisions by intermediate appellate courts. It is only by intermediate courts following the holdings of the Supreme Court that one can hope to have predictability of law — the Rule of Law — from intermediate courts of appeal. While I may not agree with Justice Sotomayor’s approach, I think Marks constrains our discretion. The Majority today defies stare decisis by adopting a contrary approach and result. For all of these reasons, we had it right in Austin , and I respectfully dissent. . A Rule 11(c)(1)(C) agreement may or may not specifically reference applicable sentencing guidelines as the basis for the government's sentencing recommendation. See Fed. R. Crim. P. 11. . See also United States v. Banks, 770 F.3d 346, 351 (5th Cir. 2014) (citing to Justice Sotomayor's concurrence in Freeman and suggesting that, were that case applicable, the Fifth Circuit would be bound by Justice Soto-mayor’s concurrence). . At Davis’s sentencing in May 2006, the district court calculated a Guidelines range of 235-293 months (later reduced, on remand, to 188-235 months, as described below), relying in part on its own determinations that Davis’s criminal history category was II and that Davis deserved a 4-level leadership enhancement for his particular role in the offenses. Although Davis's stipulated sentence of 216 months (18 years) fell below the low end of the Guidelines range (235-293 months), the district court accepted the sentence. *1032 David appealed, arguing that the district court miscalculated the Guidelines range, and our panel reversed and remanded for resen-tencing. United States v. Davis, 312 Fed.Appx. 909, 912-13 (9th Cir. 2009) (unpublished). On remand, the district court held an evidentiary hearing about Davis’s role in the offense, and then calculated a new (lower) Guidelines range of 188-235 months. The court then reimposed the same 216-month sentence as stipulated in the Rule 11(c)(1)(C) plea agreement. On a second appeal, we affirmed. See United States v. Davis, 389 Fed.Appx. 616 (9th Cir. 2010) (unpublished). . The panel engaged in a straight-forward application of Justice Sotomayor’s binding concurrence in Freeman-. First, the agreement did not provide that Davis be sentenced within a particular Guideline range. Id. at 6. Second, it did not expressly use a Guideline range that “was evident from the agreement itself' to arrive at the 216-month term of imprisonment. Id. Judge Berzon concurred in judgment, but urged us to overrule Austin as wrongly decided. She viewed the decision to have misapplied Marks to Freeman. Id. at 8-10 (Berzon, J., concurring). . The proper test, as enumerated by the three-Justice plurality, was “ 'whether to the average person, applying contemporary community standards, the dominant theme of the material taken as a whole appeals to prurient interest.’.... Under this definition ... three elements must coalesce: it must be established that (a) the dominant theme of the material taken as a whole appeals to a prurient interest in sex; (b) the material is patently offensive because it affronts contemporary community standards relating to the description or representation of sexual matters; and (c) the material is utterly without redeeming social value.” A Book Named “John Cleland’s Memoirs of a Woman of Pleasure’’ v. Attorney Gen. of Com. of Mass., 383 U.S. 413, 418 , 86 S.Ct. 975 , 16 L.Ed.2d 1 (1966). . In addition to being widely rejected by our sister circuits, Epps has been criticized even within the D.C. Circuit. As Judge Kavanaugh contended in United States v. Duvall, 740 F.3d 604 (D.C. Cir. 2013): Following Justice Sotomayor’s opinion with regard to the "based on” issue would produce results with which a majority of the Supreme Court in Freeman would agree because — to put it in simple terms — "sometimes” is a middle ground between “always” and “never.” In other words, when Justice Sotomayor concludes that a plea agreement was based on the Guidelines, she would agree with the result reached under Justice Kennedy's opinion for four Justices. When she concludes that a plea agreement was not based on the Guidelines, she would agree with the result reached under Chief Justice Roberts's opinion for four Justices. But unlike every other court of appeals, Epps did not follow this commonsense approach to interpreting Freeman. Id. at 612 (emphasis added). . The Majority argues that their approach is not "fundamentally inconsistent with Maries itself.” Maj. Op. at 1021-22, n. 7. This, because I have not identified a subsequent Supreme Court case that has unequivocally stated that Marks requires an unwavering focus on results. But, as explained above, the Majority overlooks Marks itself. Under the rule the Majority advances today, we would be unable to derive a controlling rule from Memoirs , the earlier Supreme Court case with respect to which the Court in Maries was called upon to give binding effect. Yet that would be directly contrary to Marks’ holding that we can derive a controlling rule from Memoirs . By the same token, the Majority's rule would preclude us from deriving a binding rule from Walter v. United States, 447 U.S. 649 , 100 S.Ct. 2395 , 65 L.Ed.2d 410 (1980), where five Justices voted that the government’s warrantless seizure of contraband films and viewing of those films on a projector violated the defendant’s Fourth Amendment right. Id. at 652, 657, 660-62 , 100 S.Ct. 2395 . Two Justices voted for this result on the grounds that the government’s act of viewing the films expanded the scope of a private party’s earlier search, which had consisted only of opening the package that contained the films. Id. at 657 , 100 S.Ct. 2395 . Three Justices voted for this result for a completely different reason: that the government had exceeded its authority under the "plain view” doctrine. Id. at 660-62 , 100 S.Ct. 2395 . Thus, neither the approach adopted by the plurality nor the approach advanced by the concurrence was a "logical subset” of the other; and under the rule the Majority announces today, we could discern no "controlling” rule in Walter . Yet such a conclusion is inconsistent with the Court's holding, only three years later in Jacobsen , that Walter did set forth a controlling rule: the rule advanced by the two-Justice plurality. See United States v. Jacobsen, 466 U.S. 109 , 104 S.Ct. 1652 , 80 L.Ed.2d 85 (1984). The plurality’s view was the "controlling” rule of Walter , because four Justices in dissent had also voted for the plurality’s rule. Id. at 115-16, 104 S.Ct. 1652 ("[A] majority” in Walter "agree[dj” that "the legality of the governmental search must be tested by the scope of the antecedent private search.”). Like in Maries, it was the vote of a majority of Justices that counted. I need not belabor the point, discussed more thoroughly infra, at pp. 1039-41, here. Suffice it to say that, as a federal intermediary court, we are not free to adopt a reasoning-based approach to Marks— an approach plainly inconsistent with the facts and holdings of Marks and later Supreme Court cases interpreting splintered opinions — simply because the Court has not yet had occasion expressly to reject that reasoning-based approach. . The Supreme Court has never adopted a "logical subset” requirement in its numerous applications of Marks over the past four decades. The Supreme Court has on numerous occasions applied Marks to its own decisions. See, e.g., Glossip v. Gross, -U.S.-, 135 S.Ct. 2726 , 2738 n. 2, 192 L.Ed.2d 761 (2015) (holding that a three-Justice plurality opinion constituted the "holding” of the Court in Baze v. Rees, 553 U.S. 35 , 128 S.Ct. 1520 , 170 L.Ed.2d 420 (2008), because Justices Scalia and Thomas had concurred in the result reached by the plurality but on "broader” grounds); Panetti v. Quarterman, 551 U.S. 930, 949 , 127 S.Ct. 2842 , 168 L.Ed.2d 662 (2007) (citing Marks and holding simply that "Justice Powell’s concurrence [in Ford v. Wainwright, 477 U.S. 399 , 106 S.Ct. 2595 , 91 L.Ed.2d 335 (1986) ], ... offered a more limited holding. When there is no majority opinion, the narrower holding controls.”). In neither of these cases did the Supreme Court state that the controlling opinion must be a “logical subset” of the broader view that produces the same result — it has only reiterated that the controlling opinion is the one that relies on "narrower” grounds. . The question in Tidewater was whether Congress's amendment to 28 U.S.C. § 1332 to permit citizens of the District of Columbia to be characterized as "citizens of a state” for purposes of diversity jurisdiction was constitutional under Article III. The problem in that case was that the Supreme Court had previously addressed that same question (albeit in the absence of a Congressional statute) and had interpreted Article Ill’s reference to "citizens of a state” as not encompassing citizens of the District of Columbia. See Hepburn & Dundas v. Ellzey, 6 U.S. (2 Cranch.) 445, 2 L.Ed. 332 (1805); see also U.S. Const. art. III, § 2, cl. 1 ("Federal courts will have jurisdiction over: ... citizens of a state....”). Nevertheless, by a vote of 5 to 4, the Tidewater Court upheld the constitutionality of a Congressional grant of diversity jurisdiction to citizens of Washington D.C., though no five justices agreed on a rationale. Writing for a three-justice plurality, Justice Jackson voted to uphold the statute under the rationale that Congress has the power to expand Article III by statute and thereby to confer subject-matter jurisdiction on bases not specified in Article III. See id. at 583-603 , 69 S.Ct. 1173 . That is, Justice Jackson would not disturb Hepburn’s interpretation of Article III; he would simply hold that Congress may simply add a new basis for jurisdiction that does not exist in Article III. In a concurrence, Justice Rutledge, joined by Justice Murphy, strenuously disagreed that Congress - had the power to expand Article III jurisdiction beyond the bases enumerated in the Constitution. See id. at 604-17 , 69 S.Ct. 1173 . Nevertheless, Justice Rutledge would overrule Hepburn’s interpretation of Article III and would reinterpret Article Ill’s reference to "citizen[s] of a state” as including citizens of Washington D.C. See id. at 617-626 , 69 S.Ct. 1173 . Justice Rutledge reasoned that "nothing but naked precedent, ... and the prestige of [Justice] Marshall’s name, supports ... [the] unjust and discriminatory exclusion of District citizens from the federal courts. All of the reasons of justice, convenience, and practicality ... point to the conclusion that [citizens of the District of Columbia] should enter freely and fully as other citizens and even aliens do.” Id. at 617 , 69 S.Ct. 1173 . The four remaining Justices would have declined to overrule Hepburn-, but they — like Justice Rutledge — also vehemently rejected Justice Jackson’s suggestion that Congress had the power to create subject-matter jurisdiction not conferred by Article III. Thus, based on the rationale that the Court was bound by Justice Marshall’s interpretation of Article III in Hep-bum, and that Congress lacked authority to expand Article III, the dissenting Justices would have found the statute unconstitutional. See id. at 626-46 , 69 S.Ct. 1173 . In sum, six Justices agreed that Congress could not expand the scope of subject-matter jurisdiction of Article III courts beyond that provided by the Constitution. This view — which finds majority support only by combining the views of two concurring and four dissenting justices — is the governing rule. Tidewater therefore demonstrates that a "majority” view of *1040 the Court that binds us as intermediate courts may be comprised of dissenting and concurring justices, regardless of the quite obvious lack of any "logical subset” between the views expressed by the concurring and dissenting Justices. . See, e.g., Seminole Tribe of Florida v. Florida, 517 U.S. 44, 65 , 116 S.Ct. 1114 , 134 L.Ed.2d 252 (1996) (describing as "fundamental” the notion "that Congress could not expand the jurisdiction of the federal courts beyond the bounds of Article III” and overruling Pennsylvania v. Union Gas Co., 491 U.S. 1 , 109 S.Ct. 2273 , 105 L.Ed.2d 1 (1989) to the extent it suggested otherwise); Rosmer v. Pfizer Inc., 263 F.3d 110 , 120 n. 5 (4th Cir. 2001) (citing Tidewater for the proposition that “Congress cannot confer jurisdiction on Article III courts by statute when Article III does not authorize that jurisdiction.”); Lo Duca v. United States, 93 F.3d 1100, 1108 (2d Cir. 1996) (explaining that “[i]n Tidewater, ... six Justices reaffirmed the traditional view that federal courts are courts of limited jurisdiction whose judicial powers are bounded by Article III,” a notion that dates “as far back as Marbury v. Madison.... ”). . Colorado River held that federal courts have discretion in “exceptional” circumstances to stay federal court proceedings pending the resolution of a parallel state court proceeding. See Colorado River Water Conservation Dist. v. United States, 424 U.S. 800, 813, 817-18 , 96 S.Ct. 1236 , 47 L.Ed.2d 483 (1976) (recognizing, however, that "[a]b-stention from the exercise of federal jurisdiction is the exception, not the rule”). The "Colorado River test” refers to the list of factors courts consider in determining whether to invoke this exceptional prudential abstention doctrine. These factors include: (a) the extent to which the federal legislation pursuant to which the federal suit is brought favors the state versus federal forum; (b) which forum offers the "greatest experience and expertise” in the particular subject-matter; (c) the “absence of any substantial progress in the federal-court litigation”; (d) the extent to which the suit involves questions or “rights governed by state law”; (e) "the geographical inconvenience of the federal forum”; (f) "the desirability of avoiding piecemeal litigation”; (g) "the order in which jurisdiction was obtained by ihe concurrent forums”; and (h) "the Government’s previous willingness to litigate similar suits in state court.” Memorial Hospital, 460 U.S. at 16 , 103 S.Ct. 927 . . The Majority incorrectly suggests that Tidewater and its progeny somehow support a reasoning-based approach to Maries. Maj. Op. at 1021-22, n. 7. But quite the opposite is true. In Tidewater, Justice Rutledge (joined by Justice Murphy) held that Congress’ power to confer Article III jurisdiction was limited to the bases enumerated in the Constitution. See discussion supra, n.9. And four dissenting Justices expressly agreed with that holding. Of course, Justice Rutledge and the four dissenting Justices ultimately disagreed about whether Article Ill’s reference to ”citizen[s] of a state” should be understood as encompassing District of Columbia citizens. This disagreement led the two factions of Justices to vote for different case results. But all six Justices voted to hold that the Constitution provided the starting point for the Court’s analysis; Congress had no authority to add new bases for Article III jurisdiction by statute. Tidewater and its progeny hold that we are bound by holdings from splintered Court opinions that garner the five or more votes from the Court. Consideration of dissenting opinions to derive the "narrowest grounds” does not focus on the various reasonings as determinative of results. Indeed, it is just the opposite. Consideration of dissenting opinions is done not for the purpose of combining the rationales — an impossible task, since they are contradictory — but for predicting the vote (the result) which the dissenting opinions would add to the plurality opinion’s votes for the next analogous case. It is not the contradictory rationales that combine in Tidewater to result in a rule that "Congress may not expand the scope of subject-matter jurisdiction conferred by Article III through passage of a Congressional Act.” See supra, at pp. 1039-40. It is the combined results of Justice Rutledge’s and Justice Murphy's votes in favor of such a rule, plus the similar votes of the four dissenting Justices on that same issue that established the rule. Seen from the other side of the case, it is the combination of the result of the votes of the 3-member plurality that Congress had the power to so expand subject-matter jurisdiction, with the 2-member concurrence, which vehemently rejected such power, but found that Art. Ill itself was originally intended to include D.C. citizens for purposes of establishing diversity of citizenship jurisdiction, that established the rule. It was not a rule “derived by combining the 'views’ or ’rationales[s]’ of Tidewater’s concurrence and dissent.” Majority Op. 1022, n. 7. Just the opposite. It was a rule derived from the votes of the Justices, notwithstanding contradictory views or rationales used to explain the votes. I note the Majority’s "logical subset” also cannot be squared with Tidewater, as neither opinion in Tidewater was a logical subset of the other, and yet we have derived a binding holding from that splintered decision.