MEMORANDUM
MCHUGH, J.This case arises out of a 15-year mandatory sentence under the Armed Career Criminal Act (ACCA) that was imposed before the Supreme Court’s decision in Mathis v. United States, — U.S. —, 136 S.Ct. 2243, 195 L.Ed.2d 604 (2016). The predicate “violent felony” offenses for that sentence were convictions for first-degree robbery under 18 Pa. Cons. Stat. § 3701(a) (1) (i) — (iii), a statute that defines robbery in such a way as to allow for. the possibility of conviction for conduct that, after Johnson v. United States, — U.S. —, 135 S.Ct. 2551, 192 L.Ed.2d 569 (2015), no longer qualifies as an ACCA-covered violent felony. Because Pennsylvania defines first-degree robbery so broadly, I am persuaded that Mathis forbids using a conviction for it as the basis for an enhanced sentence unless the sentencing court can determine with certainty that the elements of that conviction fit within ACCA’s definition of violent felony. As that certainty cannot be achieved here, Defendant Antoine Singleton is entitled to a resentencing.
I. Procedural Background
Five years ago, Singleton pled guilty under 18 U.S.C. § 922(g) to possession of a firearm and ammunition by a convicted felon. ACCA mandates a minimum prison sentence of 15 years for anyone who violates § 922(g) and has three prior .convictions for a “violent felony or a serious drug offense.” 18 U.S.C. § 924(e). As relevant here, ACCA defines “violent felony” to include any felony that has “as an element the use, attempted use, or threatened use of physical force” (the force clause) or. that “otherwise involves conduct that presents a serious risk of physical injury to another” (the residual clause). § 924(e)(2)(B)(i)-(ii). Because Singleton had four prior Pennsylvania first-degree robbery convictions (plus a prior drug conviction), the sentencing court concluded he had at least three convictions for a violent felony or serious drug offense and gave him the mandatory sentence of 15 years,
Two years ago, however, in Johnson, the Supreme Court struck down ACCA’s residual clause as-unconstitutionally vague. Then last year, in Welch v. United States, — U.S. —, 136 S.Ct. 1257, 194 L.Ed.2d 387 (2016), the Court held- that Johnson applied retroactively to cases on collateral review. Singleton now.moves under 28, U.S.C. § 2255 to vacate his sentence,, arguing that, -with the residual clause invalid, his prior robbery .convictions do not have the required element of force necessary to fit within ACCA’s force clause, rendering his mandatory sentence unlawful, For the *427following reasons, I am compelled to agree.
II. Legal Background
The specific question presented by this case ⅛' whether Singleton’s four prior convictions for Pennsylvania first-degree robbery under 18 Pa. Cons. Stat. § 3701(a)(l)(i) — (iii)1 have “as an' element the use, attempted use, or threatened use of physical force.” To answer that specific question, however, requires broader consideration of the proper way to determine whether a prior conviction qualifies as a violent felony under ACCA. And the governing law is not simply daunting in its complexity, but also counterintuitive, in that it largely requires sentencing courts to ignore the underlying facts of prior convictions.
A. The Categorical Approach
The Supreme Court has long held that in evaluating most prior convictions, ACCA “mandates a formal categorical approach.” Taylor v. United States, 495 U.S. 575, 600, 110 S.Ct. 2143, 109 L.Ed.2d 607 (1990). This approach' requires court's “to look only to the fact of conviction and the statutory definition of the prior offense” — “not to the particular facts underlying th[at] conviction[ ].” Id. at 600, 602, 110 S.Ct. 2143. Among the reasons for this elements-based approach are two of particular relevance here:' Sixth Amendment concerns and procedural fairness. Mathis, 136 S.Ct. at 2252-53. First, the Sixth Amendment requires that “[o]ther than the fact of a prior conviction, any fact that increases the penalty for a crime beyond the prescribed statutory maximum must be submitted to a jury, and proved beyond a reasonable doubt.” Apprendi v. New Jersey, 530 U.S. 466, 490, 120 S.Ct. 2348, 147 L.Ed.2d 435 (2000). Since that maximum penalty is increased whenever a court later uses a prior conviction to trigger an ACCA enhancement, it would'“raise serious Sixth Amendment concerns” if the court went “beyond identifying the crime of conviction to explore the manner in which the defendant committed that offense.” Mathis, 136 S.Ct. at 2252. Second, “an elements-focus avoids unfairness to- defendánts.” Id. at 2253. “Statements of ‘non-elemental fact’ in the records of prior convictions are prone to error precisely because their proof is unnecessary,” which means “[a]t trial, and still more at plea hearings, a defendant may have no incentive to contest” them. Id. (citation omitted). These types of factual inaccuracies “should not come back to haunt the defendant many years down the road by triggering a lengthy mandatory sentence.” Id.
To frame properly, the issue in this ease requires preliminary recognition of an important point. Besides its force and residual clauses, the ACCA also defines violent felony to include certain listed offenses, such 'as “burglary.” § 924(e)(2)(B)(ii). The Supreme Court has held that any offense so listed refers to the “generic” version of the crime — “i.e., the offense as commonly understood.” Descamps v. United States, 570 U.S. 254, 133 S.Ct. 2276, 2281, 186 L.Ed.2d 438 (2013). With listed offenses, then, a sentencing court applying the categorical approach is limited to “comparing] the elements of the statute forming the basis of the defendant’s conviction with the elements of the ‘generic’ crime.” Id. The “prior conviction qualifies as an ACCA predicate only if the statute’s elements are the same as, or narrower than, those of the generic offense.” Id. Consider, for example, a bur*428glary statute that criminalizes the entry of certain locations with the intent to commit certain crimes, like theft, regardless whether that entry was lawful or unlawful. See id. at 2282 (citing Cal. Penal Code § 459 (West 2010)). Generic burglary, however, is defined more narrowly, requiring an unlawful entry. Taylor, 495 U.S. at 598, 110 S.Ct. 2143. As a result, such a burglary statute criminalizes conduct that generic burglary does not — for example, shoplifting. In such a case, the categorical approach forbids a court from using a conviction under the statute to enhance a sentence under ACCA. See Descamps, 133 S.Ct. at 2282-83. And an appreciation of that principle is important here because, although most of the governing decisions analyzing the categorical approach have focused upon the offenses listed within ACCA, the same limiting principles apply equally “to the other clauses in ... ACCA’s ‘violent felony’ definition” — including the force clause. See United States v. Robinson, 844 F.3d 137, 142-43 (3d Cir. 2016).
B. The Modified Approach
In a typical case, then, a sentencing court performing a categorical-approach analysis “cannot go beyond identifying the crime of conviction” and reviewing the statutory definition. Mathis, 136 S.Ct. at 2252. But there is an exception. “[I]n a narrow range of cases,” the “categorical approach ... may permit the sentencing court to go beyond the mere fact of conviction” and the statutory text, Taylor, 495 U.S. at 602, 110 S.Ct. 2143, and also consult the “charging document, written plea agreement, transcript of plea colloquy, and any explicit factual finding by the trial judge to which the defendant assented,” Shepard v. United States, 544 U.S. 13, 16, 125 S.Ct. 1254, 161 L.Ed.2d 205 (2005). This limited universe of court records, which has been extended to include jury instructions, see Descamps, 133 S.Ct. at 2281, has been termed “Shepard documents.” The purpose of this “modified” approach, and its accompanying license to review any available Shepard documents, is to provide sentencing judges with a tool to determine reliably not what specific conduct was the basis for a particular crime as a factual matter, but rather what type of conduct is generally represented by an underlying conviction as a legal matter. But because the modified approach is accordingly intended only to “serve[ ] the ■ limited function” of “helping] effectuate the categorical analysis,” id., it retains the categorical approach’s single purpose: to figure out the precise elements of the crime of conviction. Shepard, 544 U.S. at 26, 125 S.Ct. 1254.
In line with that limited function, the modified approach also has limited applicability: a sentencing court may use it only if the underlying conviction was under a so-called “divisible” statute. Descamps, 133 S.Ct. at 2281-82. A divisible statute “sets out one or more elements of the offense in the alternative,” id. at 2281, and in the “typical case brought under the statute, the prosecutor charges one of those alternatives, and the judge instructs the jury accordingly,” id. at 2284. Returning to the example of a state burglary statute encompassing both lawful and unlawful entry, consider a situation where, instead of merely sweeping more broadly than generic burglary, the statute was specifically phrased to prohibit “the lawful entry or the unlawful entry” of a premises with intent to steal. See Mathis, 136 S.Ct. at 2249. As with the earlier statute, a conviction under this one would qualify as an ACCA offense if it involved unlawful entry, but would not so qualify if it involved lawful entry. Id. But unlike the earlier statute, the “alternative phrasing” of this one permits (indeed, requires) a sentencing court) to use the modified ap*429proach to “figur[e] out which of the alternative elements listed — lawful entry or unlawful entry — was integral to the defendant’s conviction.” Id. The sentencing court accomplishes this by reviewing the Shepard documents, to “determine what crime, with what elements, the defendant was convicted of.” Id. Stated differently, the question is whether the court can isolate the conviction to qualifying violent elements alone, to the exclusion of lesser, non-qualifying conduct.
The modified approach thus only has any possible application to “alternatively phrased statute[s].” Id. at 2256. So, when a sentencing court is faced with alternative wording in a statute, before going beyond the law’s text and turning to Shepard documents to glean more information about the nature of a conviction, the court must first decide whether the statute is, in fact, divisible. That analysis, in turn, depends on whether the statute’s components (“listed items,” per Mathis) “are elements or means.” Id. Elements are prerequisites to a conviction; means, by contrast, are not essential, but merely represent ways in which a statute might be violated. If a statute’s components merely describe different means, “the court has no call to decide which of the statutory alternatives was at issue in the earlier prosecution.” Id. A conviction under such a statute thus cannot be analyzed using the modified approach and — assuming the statute criminalizes conduct outside ACCA’s reach — the conviction may not be used to enhance a sentence. Id. Only if a statute’s components are elements may a court review any available Shepard documents to try to isolate which alternative element formed the basis for the conviction. Id.
In determining whether listed items are elements or means, the Supreme Court has authorized four possible approaches, the last of which is permissible only if the first three fail to yield an answer. See id. at 2256-57. First, a “state court decision [may] definitively answer[ ] the question.” Id. at 2256. In Mathis, for example, the Iowa Supreme Court had expressly held that the Iowa burglary statute at issue listed “alternative method[s]” — not elements — rendering the modified approach impermissible and disqualifying any conviction under the statute from ACCA consideration. Second, “the statute on its face may resolve the issue.” Id. If, for example, the statute provides different penalties for violations of its different alternatives, it is considered to list elements; if instead it is “drafted to offer ‘illustrative examples,’ ” it lists means. Id. Third, the “statute may itself identify which things must be charged (and so are elements) and which need not be (and so are means).” Id. Finally, if none of those provides a “clear answer[ ],” the court may conduct a limited review of the Shepard documents — which the Court has described as a “peek” — “for ‘the sole and limited purpose of determining whether [the listed items are] element[s] of the offense.’ ” Id. at 2256-57 (citation omitted).
As to this fourth approach, the Court illustrated with an example:
Suppose ... an indictment and correlative jury instructions charge a defendant with burgling a “building, structure or vehicle” .... '. That is as clear an indication as any that each alternative is only a possible means of commission, not an element that the prosecutor must prove to a jury beyond a reasonable doubt.
Id. at 2257. But if those documents, by contrast, “referenc[e] one alternative term to the exclusion of all others, ... the statute contains a list of elements.” Id.2
*430Ultimately, however, while the categorical approach and its modified variant involve many moving parts, they retain one overarching purpose: giving a sentencing court a way to “satisfy ‘Taylor’s demand for certainty” ” as to this underlying question: “[W]hat crime, with what elements,” was the defendant convicted of? Id. at 2249, 2257 (citation omitted).
III. Discussion
Applying that legal background to Singleton’s four Pennsylvania first-degree robbery convictions, I conclude that his convictions do. not qualify as violent felonies under ACCA. My analysis proceeds in four parts. First, I analyze whether, under Mathis’s, first three methods of determining divisibility, the Pennsylvania first-degree robbery statute is divisible. I conclude that when, a defendant is charged under all three subsections of the statute — which is the manner in which it is commonly applied by Pennsylvania courts and prosecutors — then the statute is not divisible. Second, I apply Mathis’s residual method, which permits me to review any available Shepard documents to determine whether, even if the statute might generally be indivisible, Singleton’s particular convictions can be isolated to offenses with violent elements fitting within ACCA’s force clause. But after -review of those materials I am permitted to consider, I cannot so isolate Singleton’s convictions. Third, I am forced to conclude that the fact-based approach the Third Circuit employed in United States v. Blair, 734 F.3d 218 (3d Cir. 2013), to analyze the first-degree robbery statute is no longer valid after Mathis. And finally, I reject the government’s proposal that courts should impose heightened barriers to relief when a sentence under ACCA is attacked collaterally. For these reasons, Singleton must be resentenced.
A. Common Practice Shows That The Pennsylvania First-Degree Robbery Statute Is Generally Applied In An Indivisible Way.
Pennsylvania’s first-degree robbery statute, 18 Pa. Cons. Stat. § 3701(a)(l)(i)-(iii), provides:
(a) Offense defined.—
(1) A person is guilty of robbery if, in the course of committing a theft, he:
(i) inflicts serious bodily injury upon another;
(ii) threatens another with or intentionally puts him in fear of immediate serious bodily injury; [or]
(iii) commits or threatens immediately to commit any felony of the first or second degree[.]
Because the statute is alternatively phrased, I must decide whether it is divisible. Here that inquiry is potentially dispos-itive, for the government concedes (as it must) that § 3701(a)(l)(iii) does .not require force, because some first- and second-degree felonies are .nonviolent — say, certain types of identity theft, see 18 Pa. Cons. Stat. § 4120(c). As a result, if § 3701(a)(1)(i) — (iii).’s alternatives-are typically treated by Pennsylvania courts and prosecutors as merely means, as opposed to separate elements that a jury must And to sustain a conviction or a defendant must admit when he pleads guilty, a Pennsylvania first-degree robbery conviction cannot serve as an ACCA predicate offense.
*431Mathis provides the divisibility framework. To start, no Pennsylvania court decision that I am aware of definitively resolves the issue, but the case law that does exist lends support' for the view that § 3701(a)(1) (i) — (iii) is commonly treated as indivisible. In Commonwealth v. Payne, 868 A.2d 1257 (Pa. Super. Ct. 2005), for instance, the court, reviewing afirst-de- ■ gree robbery conviction, noted “ambiguity in the jury’s verdict” as to whether the conviction was under subsection (i) or (ii), “because the jurors were instructed on both variants, and returned only” a verdict of guilty of robbery generally. Id. at 1262. What Payne demonstrates is that if a robbery is charged under more than one subsection, jurors need not agree on which subsection applies — so long as they agree that at least .one does.3 Given the bedrock concept that elements “are what the jury must find beyond a reasonable doubt to convict the defendant,” or “what the defendant necessarily admits when he pleads guilty,” Mathis, 136 S.Ct. at 2248 (emphases added), Payne provides strong evidence that subsections (i)-(iii) are means.
In response, the government contends (Br. 10) that Commonwealth v. Neal, 274 Pa.Super. 225, 418 A.2d 378 (1980), establishes that § 3701(a)(1)(h) and (iii) “define different offenses,” with • the result ■ that subsections (ii) and (iii) state different elements. Neal, however, involved a very limited question: whether a defendant who was indicted only under subsection (iii) was prejudiced when the trial judge, recognizing that the proof at trial only supported a conviction under subsection (ii), then proceeded to submit the case. to the jury under that subsection instead. In-finding prejudice, the Superior Court noted that “the original and amended indictments cannot be said to have charged the same particular offense, as each indictment would require the proof of different elements.” Id. at 380.
Neal adds nothing to the question of divisibility. The issue there was simply whether the proof conformed to the indictment. It is true, as Neal noted, that subsections, (ii) and (iii) are “not interchangeable,” id.—but that self-evident proposition is a function of the axiom that “whenever possible each word in a statutory provision is to be given meaning and not to be treated as surplusage,” Commonwealth v. Lassiter, 554 Pa. 586, 596, 722 A.2d 667, 661 (1998). Given that, if a jury is charged only under subsection (ii), it is by definition answering a different question than if it were charged only under subsection (iii). But the question here is different: whether, when a jury is charged under multiple subsections, it must necessarily and unanimously agree which subsection applies to reach a verdict of guilty. On that, Neal provides no help.4
Mathis’s second and third methods can be grouped, as each requires a close look at the robbery statute itself. The second — which checks whether the statute “on its face” provides answers — offers little guidance for two reasons: ■ first, a conviction under any of- subsections (i), (ii) or (iii) constitutes the same grade of robbery — first-degree;5 and second, though it *432may ultimately be true that the statute’s three subsections merely offer “illustrative examples” of first-degree robbery, to summarily assume so would be to sidestep the divisibility inquiry entirely.
Mathis’s third way to determine divisibility — figuring out “which things must be charged [under the statute] ... and which need not be” — is significantly more useful. As a starting point, the current version of Pennsylvania’s Suggested Standard Jury Instruction for first-degree robbery requires that the jury find “two elements”: (1) conduct falling into any of subsections (i), (ii), or (iii); and (2) that this conduct was done in the course of committing a theft. See 2 Pennsylvania Bar Institute, Pennsylvania’s Suggested Standard Criminal Jury Instructions § 15.3701A (3d ed. 2016).6 This instruction indicates that a jury charged on each of subsections (i), (ii), and (iii) need not agree which subsection applies to reach a verdict of guilty. And while suggested instructions, as their title makes plain, are nonbinding, even the government admits that they “shed light on whether a factual determination is a means or an element.” Br. 12. Post-Mathis, the Third Circuit has on multiple occasions used suggested instructions to show what a statute “require[s] the jury to unanimously agree on.” United States v. Steiner, 847 F.3d 103, 119 (3d Cir. 2017) (citation omitted) (burglary statute indivisible); accord United States v. Henderson, 841 F.3d 623, 629 (3d Cir. 2016) (controlled substance act divisible).
In this regard, the suggested first-degree robbery instruction deserves particular weight, because as Singleton points out, “Pennsylvania judges commonly instruct juries on the different subsections of § 3701(a)(1 )(i) — (iii) in the alternative.” Reply Br. 12. Singleton provides many examples of judges charging juries in precisely this way, using the suggested instruction essentially verbatim. See Dkts. 59-1 to 59-5. Even more relevant here, he also offers multiple guilty plea colloquies and agreements where defendants pled guilty to first-degree robbery generally, with no indication of a particular subsection. See Dkts. 59-6 to 59-8.
The government counters that while it might be true in general that the suggested instruction treats subsections (i), (ii), and (iii) together as one element, that “instruction plainly would not apply where a defendant is charged with violating only one subsection.” Br. 12-13. True enough: there are certainly cases where only one subsection of § 3701(a)(1)(i) — (iii) could possibly apply, and so the prosecutor only charges, and the jury is only instructed on *433(or the defendant only pleads to), that specific subsection. In such a case, it is likely true that a later federal sentencing court can be relatively certain of which “statutory alternative” that defendant was convicted.
But that is not universally the case. As Singleton amply shows (and, as his case exemplifies), Pennsylvania prosecutors often — perhaps typically — charge defendants on all three subsections of § 3701(a)(1)(i) — (iii). Consequently, in the ordinary course of things, jury charges or plea documents are often entirely unclear as to whether any particular subsection was a defendant’s precise crime of conviction. And that, the Supreme Court has emphasized, is the very essence of an indivisible statute: “As long as the statute itself requires only an indeterminate” element — here, any of subsections (i), (ii), or (iii) — “that is all the indictment must (or is likely to) allege and all the jury instructions must (or are likely to) mention. And most important, that is all the jury must find to convict[.]” Descamps, 133 S.Ct. at 2290. I conclude that when a defendant is charged under all three subsections of § 3701(a) (1) (i) — (iii), the statute is not on its face divisible.
B. The Available Shepard Documents Do Not Definitively Show That Singleton’s Convictions Were For Any Specific Subsection Of§ 3701(a)(l)(i)-(iii).
As to Singleton’s four robbery convictions, none was for the sort of robbery where only one subsection could possibly apply. This leaves as a last resort Mathis’s residual technique — a “peek” at the Shepard documents for “the sole and limited purpose of determining whether [subsections (i), (ii), or (iii) are] elementfe] of the offense.” But even this method takes me back to the same conclusion, for the charging and plea documents available here do not suffice to show that Singleton necessarily pled to ACCA-covered crimes.
To be clear, those documents do indicate that all of Singleton’s robberies were performed at gunpoint. So if the question were whether his robberies were violent crimes in the everyday sense of the term, the answer would almost certainly be yes. But the question before me is. not factual, but formalistic: Were his convictions for those crimes categorically violent felonies under ACCA? As to this, the documents are equally clear: the answer is no. And by acknowledging, both in its brief (at 5) and at oral argument (at 3:6-21), that even these documents fail to establish any particular subsection of conviction, the government effectively concedes the point.
Singleton’s first two robberies were charged under all three subsections of § 3701(a)(l)(i) — (iii). Significantly, the bottom of each charging document repeats the words “FELONY COMMITTED OR THREATENED.” See Dkt. 53-1 at *3; Dkt. 53-2 at *2. Neither plea agreement gives any more information, other than the notation “F-l,” indicating first-degree robbery. See Dkt. 53-1 at *2; Dkt. 53-2 at *3. As to the second two robberies, these too were charged under every subsection, although unlike with the first pair, the charges for these specifically listed “assault” next to the subsection (iii) counts.7 See Dkt. 53-3 at *2-3. Singleton pled guilty to the latter two robberies together, and had virtually the same colloquy with his attorney for each:
Q: Do you understand the definition of robbery?
*434A: Yeah.
Q: Essentially that you in the course of committing a theft feloniously did inflict serious bodily injury or attempt to cause serious bodily injury upon another individual and in this incident the threat or force was by way of using a handgun. Do you understand that?
A; Yes.
Dkt. 53-3 at *18."
With these documents in hand, I have little difficulty concluding that Singleton’s first two convictions are not divisible. Indeed, the only indication as to any specific subsection for either conviction is both charging documents’ emphasis on “FELONY COMMITTED OR THREATENED,” which refers to subsection (iii). That, of course, is of little help to the government, because it is undisputed that a conviction under subsection (iii) is not categorically an ACCA predicate. I therefore find that these first two convictions were for the indivisible crime of first-degree robbery, and are not categorically violent felonies under ACCA.
As to Singleton’s final two convictions, although the question is closer, I reach the same conclusion. Initially, I acknowledge that the colloquies’ language — did Singleton “in the course of committing a theft feloniously .., inflict serious bodily injury or attempt to cause serious bodily injury upon another”? — could be read to refer to a conviction more specific than just indivisible first-degree robbery. See § 3701(a)© (“inflicts . serious bodily injury”), (ii) (“threatens ,.. with or intentionally puts him in fear of’ same). But any weight that wording might lend to finding these convictions divisible is offset by use of term “feloniously.” The statue’s only reference to “feloniously” or “felony” is in subsection (iii), which encompasses nonviolent crimes. Given that added confusion, then, I simply cannot say for sure which subsection Singleton pled to in either robbery. And because the entire point of the categorical approach and divisibility analysis is to “counsel against allowing a- sentencing court to make [these kinds of] disputed' determination[s],” Descamps, 133 S.Ct. at 2288, I must find that, as with Singleton’s first two convictions, these two are indivisible and outside ACCA’s reach.
C. The Approach Employed By The Third Circuit In Blair Does Not Survive Mathis.
As the government points out, the Third Circuit has addressed this question before, and held that “given the clearly laid out alternative elements of the Pennsylvania robbery statute, it is obviously divisible.” United States v. Blair, 734 F.3d 218, 225 (3d Cir. 2013). Blair s holding remains true where a defendant was not charged under subsection (iii) or if analysis of the Shep-~ ard documents otherwise isolates his conviction to an offense with ACCA-qualifying violent elements. Neither characteristic, however, describes this case.
In Blair, the Third Circuit, reviewing prior convictions for Pennsylvania first-degree robbery, looked to Blair’s charging documents, and, because those documents specified that Blair had committed subsection (iii) robberies whose predicate felonies were aggravated assaults, the court concluded that his robberies fit within ACCA’s definition of violent felony. Under subsection (iii), however, aggravated assault is merely a fact — it has absolutely nothing to do with the elements'oí the statute, which require no more than the commission of some first- or second-degree felony in the course of committing a theft. A defendant who pleads guilty under subsection "(iii) is only necessarily admitting he. committed some kind of first- or, second-degree felony.
*435Since Blair, the Supreme Court decided Mathis, In Mathis, the Court took pains to emphasize that the categorical approach “focus[es] solely on ... the elements of the crime of conviction ..while ignoring the particular facts''of the case.” 136 S.Ct. at 2248. Elements are what the jury must find to convict or what 'a defendant necessarily admits when he' pleads guilty. “Facts, by contrast, are mere real-world things — extraneous to the crime’s legal requirements.” Id. Under either the categorical approach or' its modified variant, a “crime cannot qualify as an ACCA predicate if its elements are broader than those of k listed generic offense,” Id. at 2251 (emphasis added). “How a given' defendant actually perpetrated the crime ... makes no difference; even if his conduct fits within the generic offense, the mismatch of elements saves the defendant from an ACCA sentence.” Id.
Returning then to Blair, the particular felonies that Blair committed to trigger each of his subsection (iii) robberies were facts, not elements. Those felonies involve how Blair “actually perpetrated the crime[s]” — and for this reason, under Mathis, they “make[ ] no difference.” I see no alternative to the conclusion that, in this respect, Blair was abrogated by Mathis?8
The Third Circuit’s recent decision in Chavez-Alvarez v. Attorney General United States, 850 F.3d 583 (3d Cir. 2017), illustrates Mathis’ s operation and further supports this conclusion, Chavez-Alvarez, a Mexican citizen who joined the U.S. Army, pled guilty to the military crime of sodomy. Id. at 585. The -Bureau of Immigration Appeals subsequently treated that conviction as a predicate “crime of moral turpitude” to order Chavez-Alvarez’s removal from the United States. Id. At the time of conviction, however, the- Uniform Code of Military Justice did not distinguish between forcible and consensual sodomy, and by the time the government sought deportation, consensual sodomy was not a crime of moral turpitude; indeed, since Lawrence v. Texas, 539 U.S. 558, 123 S.Ct. 2472, 156 L.Ed.2d 508 (2003), it has been constitutionally protected activity. Nevertheless, the BIA found that Chavez-Alvarez had committed forcible sodomy based upon a stipulation at his plea hearing, and concluded that his conviction therefore qualified as a deportable offense. 850 F.3d at 587.
The Third Circuit reversed. It held that under the categorical approach, courts “look to the elements of the statutory ... offense, not to the specific facts [underlying the particular offense], reading the applicable statute to ascertain the least culpable conduct necessary to sustain conviction under the statute.” Id. at 587-88 (alteration in original) (citation omitted). The court refused to go beyond the elements of the, sodomy,statute: “[WJe will not consider the facts underlying Chavez-Alvarez’s conviction (that it was committed forcibly)[.]” Id. at 589 n.4 (citing Mathis, 136 S.Ct. at 2252). So too with Singleton’s robberies: even if they were violent as a factual matter, they were not categorically violent as.a legal one.
* * *
I certainly recognize, as the Third Circuit recently put it, that because state legislatures do not draft their criminal laws with the ACCA in mind, there are many cases where the “ ‘elements or *436means’ inquiry is not quite as easy as the Supreme Court thought.” Steiner, 847 F.3d at 120. As this painstaking analysis has shown, Pennsylvania’s first-degree robbery statute is no exception. But the Supreme Court has instructed that even where “distinguishing between ‘alternative elements’ and ‘alternative means’ is difficult,” the Shepard documents will ultimately “reflect the crime’s elements.” Descamps, 133 S.Ct. at 2285 n.2. For Singleton, those documents fail to establish with the requisite certainty that any of his convictions was under a particular subsection of § 3701(a)(l)(i)-(iii). And it is beyond dispute that the elements of § 3701(a)(l)(i) — (iii) sweep broader than ACCA’s force clause. I must conclude that Singleton’s convictions cannot be isolated to subsections (i) or (ii) to the exclusion of subsection (iii), and are therefore not categorically violent felonies under ACCA, making Singleton entitled to a resentenc-ing.
D. There Is No Basis To Impose On Singleton A More Onerous Burden Just Because He Is Attacking His Sentence Collaterally.
The government has narrowed significantly the dispute before me, by conceding that if this were an original sentencing proceeding, the above analysis would hold and Singleton’s robberies would not qualify as ACCA predicates. See Tr. of Oral Arg. 3:6-21.9 Nevertheless, the government seeks to prevail by proposing that in collateral attacks on existing sentences, eourts should adopt a more onerous standard for defendants to meet as a prerequisite to relief. That proposed new standard has both procedural and substantive components. Procedurally, the government argues that habeas courts should be permitted to “look beyond those limited Shepard records,” which here means that I would consider the facts, in Singleton’s presen-tence investigation report (PSR) in deciding whether his convictions were ACCA-qualifying. Id. at 7:9-24. Substantively, the government argues that to secure a resen-tencing, a habeas petitioner like Singleton must make a threshold showing that his ACCA sentence was a “miscarriage of justice.” Id. at 4:11-7:24.
The government’s arguments are unpersuasive. Starting with procedure, it is true that a PSR “can be a Shepard document when the defendant does not object to a factual account of a crime therein.” United States v. Doe, 810 F.3d 132, 147 (3d Cir. 2015) (citing United States v. Siegel, 477 F.3d 87, 93 (3d Cir. 2007)). But again, a court may use Shepard documents “merely ... to implement the categorical approach.” Descamps, 133 S.Ct. at 2285. So I may use Singleton’s PSR for one purpose: to “discover ‘which statutory phrase[ ]’ ... ‘covered a prior conviction.’ ” Id. (emphasis added) (quoting Nijhawan v. Holder, 557 U.S. 29, 41, 129 S.Ct. 2294, 174 L.Ed.2d 22 (2009)). Even if the facts in that PSR are undisputed, I “may not ... consider th[ose] facts for additional sentencing purposes.” United States v. Dahl, 833 F.3d 345, 353 (3d Cir. 2016).10
*437At times, the government seems to concede this. For example, it says I “may consider the uncontested facts in the PSR and other court records in determining if the offense of conviction was a violent felony under ACCA’s force clause.” Br. 21 (emphasis added). Elsewhere, however, the government pulls from the PSR and those records the violent facts of Singleton’s robberies, and claims those facts show his convictions were violent felonies. See Br. 21-23. But as I have already explained, whether Singleton’s robberies were (or, from the available records, appear) violent as a colloquial matter is a question entirely different from the one here: Did his convictions for them have the element of force required to count as ACCA predicates? And, also for reasons already explained, I cannot both answer that question in the affirmative and remain in compliance with the categorical approach.
Moreover, logically speaking, I cannot identify a principled rationale for the fact-based method the government urges me to adopt. After having reviewed every available Shepard document, I have still concluded that Singleton’s robbery convictions lack the force element required to use them to impose a mandatory sentence under ACCA. The government agrees with this conclusion, but nonetheless urges me to sustain an unconstitutional sentence by relying upon the same information that the Supreme Court has expressly and repeatedly forbidden sentencing judges to consider: the specific facts of the defendant’s case. But how can the very method deemed to violate the Constitution then later be employed- on collateral review to determine whether the underlying constitutional violation occurred?
The government’s substantive argument is admittedly more nuanced: because Singleton seeks relief under Johnson not at a first-time sentencing but rather on habeas, the government claims he must show both that he was originally sentenced under the residual clause that Johnson invalidated and that his ACCA sentence had a “fundamental defect” that amounted to a “miscarriage of justice.” See Br. 19-21; Tr. of Oral Arg. 5:6 — 7:8. The government cites no direct authority for this position, essentially arguing that I should import certain principles from other contexts and deem them applicable here, I decline to do so.
To start, it is unclear to me what it would mean to require Singleton to establish that his ACCA sentence was originally imposed under the residual clause. It is undisputed that Singleton could only have been sentenced under ACCA’s residual clause or its force clause. The residual clause, the Supreme Court has held, is unconstitutional not just going forward but also retroactively, meaning it cannot legally be the basis for Singleton’s continued time in prison. And the force clause does not supply an alternative basis for all of the reasons set forth above.
I also will not follow the government’s invitation to require Singleton to establish *438that his ACCA sentence was a “miscarriage of justice.” Br. 19-20 (citing Bousley v. United States, 523 U.S. 614, 118 S.Ct. 1604, 140 L.Ed.2d 828 (1998); Schlup v. Delo, 513 U.S. 298, 115 S.Ct. 851, 130 L.Ed.2d 808 (1995); In re Moore, 830 F.3d 1268 (11th Cir. 2016) (per curiam)). Initially, those cases, the government freely admits, have almost no direct precedential value, because they involve what a petitioner is required to show either to surmount a procedural default (in Bousley and Schlup) or to bring a second or successive habeas petition (in Moore) — neither of which applies to Singleton. Nevertheless, the government contends that Bousley, being founded on the idea that “[hjabeas review is an extraordinary remedy,” 523 U.S. at 621, 118 S.Ct. 1604, “stands for a broader proposition,” and should, as a policy matter, apply generally to situations where the Supreme Court has issued and made retroactive a new interpretation of a criminal statute. Tr. of Oral Arg. 5:6-24. What is missing from the government’s argument, however, is a compelling justification for creating such a barrier to relief. Singleton does not seek to be released or granted a new trial; having established that he is serving an unlawful sentence, he seeks only to be resentenced lawfully. In no sense can this request be viewed as extraordinary. And again, given that the categorical approach is rooted in constitutional concerns — concerns the Supreme Court took pains to emphasize in Descamps and Mathis — it is a profound request to ask me to disregard the categorical approach simply because this case happens to be on collateral review. Where the Constitution applies, courts must apply it.
The “miscarriage of justice” standard, an exception to waiver, was created to “overcome various procedural defaults” that would otherwise .bar review. McQuiggin v. Perkins, 569 U.S. 383, 133 S.Ct. 1924, 1932, 185 L.Ed.2d 1019 (2013) (emphasis added). It.would be strange to employ that remedial standard in a case involving a sentence already deemed unlawful'to create an additional obstacle to relief. It is already the case that the “labyrinthine complexity of federal habeas review” makes it “one of the most difficult legal schemes ... to navigate.” In re Commonwealth’s Motion to Appoint Counsel Against or Directed to Def. Ass’n of Phila., 790 F.3d 457, 480 (3d Cir. 2015), as amended (June 16, 2015) (McKee, J.r concurring), cert. denied, — U.S. —, 136 S.Ct. 980, — L.Ed.2d — and — U.S. —, 136 S.Ct. 994, 194 L.Ed.2d 3 (2016). The government has not persuaded me that there is good cause to erect yet another barrier — particularly given the degree to which the “Great Writ” has already shrunken in size.11
*439It is a longstanding principle that, as in civil cases generally, the standard of proof in a habeas proceeding is by a preponderance of evidence. Walker v. Johnston, 312 U.S. 275, 286, 61 S.Ct. 574, 85 L.Ed. 830 (1941); Johnson v. Zerbst, 304 U.S. 458, 469, 58 S.Ct. 1019, 82 L.Ed. 1461 (1938); see also Pazden v. Maurer, 424 F.3d 303, 313 (3d Cir. 2005) (applying Zerbst’s standard). That standard applies with full force here. Because Singleton has shown by a preponderance of evidence that his ACCA sentence is legally invalid, he must" be re-sentenced — without the need to meet a more burdensome standard.
IY. Conclusion
Five years ago, Singleton was given a 15-year mandatory sentencé under ACCA based on prior convictions that no longer qualify as ACCA predicate offenses. His motion to vacate his sentence must be granted' and a resentencing ordered. An appropriate order follows.
. ' Though Singleton’s convictions were from 1992 and 1993, both he and the government agree that the current version of § 3701(a)(l)(i)-(iii) is the same as the one then in effect.
. In both Descamps and Mathis, the Court used specific examples of statutes in an attempt to explain its rulings, perhaps because it was aware that its "peek” technique— *430which would appear to permit courts to review Shepard documents for the threshold purpose of determining whether they may review ■Shepard documents — might cause confusion. The fact that the Court felt the need to grant certiorari in Mathis so soon after it ’ decided D'escamps suggests that lower courts did in fact struggle to apply its "elements-or-means” test.
. The government seems to concede as much, albeit without acknowledging its significance: “This decision [(Payne)] simply shows that the jury ... was permitted to find the defendant guilty if he violated either subsection!.]’’ Br. 11.
. Separately, I add that the precedential value of Neal is extraordinarily weak. The opinion announcing the judgment of the court is brief and somewhat cursory. More importantly, it represents the opinion of only one judge; one member of the panel dissented, and the other, Judge Hoffman, concurred only in the result.
.The government correctly observes that violations of different . subsections of § 3701(a)(l)(i) — (iii) may carry different sentencing guidelines ranges, see, e.g., Payne, 868 A.2d at 1262 & n.5. The government subtly suggests that unless I deem the statute divisible, it might as currently structured violate *432Apprendi, because the range of potential sentences depends on which subsection was violated. That issue is not before me, and I cannot decide this case based on a concern about the implications my ruling might have should the statute be challenged on some separate ground. But I note that Pennsylvania’s sentencing guidelines are advisory, 204 Pa. Code § 303.1(d), and Apprendi’s "right to proof beyond a reasonable doubt does not apply to facts relevant to enhancements under an advisory Guidelines regime.” United States v. Grier, 475 F.3d 556, 565 (3d Cir. 2007) (en banc) (citing United States v. Booker, 543 U.S. 220, 125 S.Ct. 738, 160 L.Ed.2d 621 (2005)).
. Further, although under Mathis legislative history is deemed less useful than jury instructions, I find it instructive that the Pennsylvania legislature, in creating the first-degree robbery statute, “adopted almost verbatim” the Model Penal Code's definition of robbery. Commonwealth v. Maldonado, 343 Pa.Super. 154, 494 A.2d 402, 408 (1985). And the MPC definition, as its commentary stresses, "contains the single offense of robbery,” whose "elements ... focus upon three factors" — only one of which is "the nature of the special circumstances that serve to distinguish the offense from ordinary theft” (i.e., subsections (i), (ii), or (iii)). Model Penal Code § 222.1 note (Am. Law Inst. 1962).
. Under Pennsylvania law, aggravated assault is (and, at the time of Singleton’s convictions, was) a first- or second-degree felony. See 18 Pa. Cons. Stat. § 2702(b). Simple assault is of no relevance here as it neither is nor was a felony of any degree. See id. § 2701(b).
. Even prior to Mathis, the Third Circuit itself questioned Blair's specific approach, See United States v. Brown, 765 F.3d 185, 196 & n.5 (3d Cir. 2014) ("Descamps instructs that it is the elements, not the facts, of a prior conviction that matter, and the elements in § 3701(a)(l)(iii) .., include both felonies that are violent and those that are not,” (citation omitted)).
. The government made these concessions, and so the limited nature of the dispute, fully clear at oral argument. But it also made these arguments in its brief (at 19-24).
. To the extent the government asks me to rely on the Third Circuit’s categorical-approach analysis in Siegel more broadly, I think it is plain that this part of Siegel was abrogated by Descamps and Mathis. Siegel involved whether a prior conviction for indecent assault constituted a "crime of violence” under the Sentencing Guidelines, which define the term to include a “forcible sex offense[].” 477 F.3d at 90. Because the unob-jected-to facts in Siegel's PSR made clear that his indecent assault involved forcible conduct, the Third Circuit, applying the modified approach, concluded that his conviction was a crime of violence. Id. at 93.
*437Such a fact-based approach does not survive Descamps and Mathis. The indecent-assault statute in Siegel listed eight possible ways to commit the offense, some forcible and . some not. Id. at 91 (citing 18 Pa. Cons. Stat. § 3126 (2000)). Moreover, the Third Circuit not only conceded that Siegel’s records failed to indicate whether his conviction was specifically under any of those eight subsections, but went out of its way to hold that merely non-consensual indecent assault (one of the eight) would not qualify as a forcible offense. Id. at 91-92. But the court nonetheless used the PSR to hold that Siegel committed a forcible crime. In other words, even though Siegel was convicted of an offense whose elements included non-forcible conduct, the court used facts in the PSR to decide he had committed a forcible offense. This is precisely what the Supreme Court has since expressly and repeatedly held is impermissible. See Mathis, 136 S.Ct. at 2248, 2251-53; Descamps, 133 S.Ct. at 2283-85.
. At oral argument, the government identified a Middle District case as the only decision to address both Bousley and the Pennsylvania robbery statute in the wake of Mathis, See United States v. Harris, 205 F.Supp.3d 651 (M.D. Pa. 2016). Like Singleton, Harris moved under § 2255 to vacate his ACCA sentence, arguing in part that his prior Pennsylvania robbery convictions were not violent felonies. The court first addressed Bousley as a threshold matter because — unlike here — the government argued Harris’s claim was procedurally defaulted. Id. at 657-59. After finding Harris had cause for his default, the court moved to the merits, concluding it was not yet clear whether Harris’s convictions were- under any particular subsection of § 3701(a)(1). Id. at 672-75. The government then supplemented the record with Shepard documents to make its case; upon review of those documents, the court still found Harris’s convictions fell outside ACCA’s reach, because each of his robberies was charged as including "reckless” conduct, whereas ACCA’s force clause requires intentional conduct. See United States v. Harris, 2016 WL 6304440, at *2 (M.D. Pa. Oct. 26, 2016) (citing United States v. Otero, 502 F.3d 331, 335 (3d Cir. 2007)). *439Nothing in Harris is inconsistent'with'the conclusion I reach'here.