BARRON

Raul S. Mariani–Franco, for appellant. Francisco A. Besosa–Martínez, Assistant United States Attorney, with whom Rosa Emilia Rodríguez–Vélez, United States Attorney, and Nelson Pérez–Sosa, Assistant United States Attorney, were on brief, for appellee.
Before THOMPSON, LIPEZ, and BARRON, Circuit Judges.

BARRON, Circuit Judge.

Wilson Serrano–Mercado contends the District Court made two mistakes in sentencing him for a federal gun crime. First, he argues the District Court erred in counting more than one of his prior convictions for Puerto Rico criminal offenses as a conviction for a “crime of violence” under the Sentencing Guidelines. Second, he contends the District Court gave too much significance under those same guidelines to the existence of an obliterated serial number on the frame of the firearm he was convicted of possessing, when the serial number on the slide was unaltered. We hold the District Court did not commit reversible error in either respect and thus affirm the sentence imposed.

I.

In District Court, Serrano pled guilty to being a felon in knowing possession of a firearm—a 9mm pistol. 18 U.S.C. §§ 922(g)(1), 924(a)(2). The Sentencing Guidelines specify a suggested sentencing range for such a conviction. U.S.S.G. §§ 2K2.1, 5A. Serrano rests his challenge to his sentence on the two errors that he claims the District Court made in identifying the proper range. And thus, it is helpful to provide some background about how, in general, such ranges are identified, and then how, in particular, the range was identified here.

Under the guidelines, two variables provide the basis for the sentencing range. The first variable is called the offense level. It is expressed in terms of a point score. Id. § 5A. The score is a function, initially, of what is known as the base offense level. Id. § 2 introductory cmt. The base offense level is generally calculated with reference to the nature of the crime of conviction. The guidelines then add points to or subtract points from the base offense level for various enhancing or mitigating factors that may or may not be present in a defendant's case. The result is the total offense level.

The second variable is a defendant's criminal history category. Id. § 5A. The guidelines assign criminal sentences certain point values. Id. § 4A1.1. These points are then translated into one of six criminal history categories, represented by the use of a Roman numeral from I to VI. Id. § 5A. The more severe the criminal history a defendant has on the basis of the points assigned, the higher the category.

On the basis of these two variables, the guidelines then set forth suggested sentencing ranges in a chart. Id. One axis of the chart lists possible total offense levels. The other axis lists possible criminal history categories. At the intersection of every possible value for these two variables, the chart sets forth a suggested range of sentences.

Before actually imposing a sentence, a district court often receives input from various actors about how to calculate the defendant's guidelines sentencing range. If there is a plea agreement, as there was here, the agreement will often recommend a range. And, in setting forth that recommendation, the agreement will often set forth certain facts that bear on the calculation of the base offense level, the total offense level, and the criminal history category. SeeFed.R.Crim.P. 11(c)(1).

The district court will also have the benefit—as, again, was true here—of a probation officer's pre-sentence report, which is based on that officer's investigation. That report, too, will set forth facts bearing on the sentencing guidelines calculation. And that report may, in light of those facts, suggest a calculation different from the plea agreement. SeeFed.R.Crim.P. 32(d).

The district court need not accept the calculations in the plea agreement or the pre-sentence report. Nor must the district court choose a sentence that falls within the range the district court's own guidelines calculation yields, though the sentence must comply with additional substantive and procedural limitations. See18 U.S.C. § 3553; United States v. Booker, 543 U.S. 220, 245, 261, 125 S.Ct. 738, 160 L.Ed.2d 621 (2005). But if the district court errs in making the guidelines calculation, the sentence may be reversed even though that calculation does not directly compel the sentence. See Gall v. United States, 552 U.S. 38, 51, 128 S.Ct. 586, 169 L.Ed.2d 445 (2007); United States v. Tavares, 705 F.3d 4, 25 (1st Cir.2013). And that is what Serrano argues must happen here.

In this case, the plea agreement recommended a sentencing range tied to a base offense level of 22. The agreement made that calculation because it stated that Serrano had been convicted of one prior felony for a “crime of violence” at the time of his unlawful firearm possession. U.S.S.G. §§ 2K2.1(a)(3), 4B1.2(a). The plea agreement did not identify any of Serrano's prior convictions. The plea agreement thus did not specify which one qualified as the crime of violence that warranted that base offense level of 22. The plea agreement's calculation also did not include a four-point increase under the guidelines' enhancement that applies when the firearm involved in a felon-in-possession charge has “an altered or obliterated serial number.” Id. § 2K2.1(b)(4).

The probation officer's pre-sentence report, as amended, departed from the plea agreement's guidelines calculation. And it did so in two respects.

First, the amended pre-sentence report suggested a base offense level of 24, rather than 22. The report used that higher base offense level because it stated that Serrano actually had more than one prior felony conviction for a “crime of violence.” Id. §§ 2K2.1(a)(2), 4B1.2(a). The report did not expressly identify which of Serrano's prior convictions qualified as a crime of violence. The report thus did not identify the ones the report relied upon in setting the base offense level at 24.

The report did list, however, a number of prior convictions for Serrano. These convictions included a 2006 Puerto Rico conviction for assault that the parties both appear to agree does qualify as a conviction for a crime of violence. These convictions also included a 2005 Puerto Rico conviction under Article 3.1 of Law 54, Puerto Rico's Domestic Abuse Prevention and Intervention Act, P.R. Laws Ann. tit. 8, § 631, which the government on appeal now contends also qualifies but which Serrano argues does not. And, finally, the list included an earlier 2004 conviction that the government does not argue qualifies.

The second respect in which the pre-sentence report differed from the plea agreement concerned the serial-number enhancement. Unlike the plea agreement, the report concluded the enhancement did apply. The report thus increased its calculation of the total offense level by four points. U.S.S.G. § 2K2.1(b)(4).

The District Court adopted the pre-sentence report's recommendations regarding the guidelines calculation. The District Court stated Serrano had “two domestic violence convictions and one assault conviction which meet the guidelines criteria for crimes of violence.” The District Court thus started from a base offense level of 24 because it had found, contrary to the representation in the plea agreement, that Serrano had been convicted of more than one offense that qualified as a crime of violence. The District Court then applied the four-point serial-number enhancement. Finally, and consistent with the plea agreement and the pre-sentence report, the District Court subtracted three points for the defendant's acceptance of responsibility, U.S.S.G. § 3E1.1 cmt. 3, due to the plea.

The District Court thus arrived at a total offense level of 25. The District Court also determined Serrano had a criminal history category of V. These calculations then combined to set Serrano's guidelines sentencing range between 100 and 125 months. The District Court imposed a sentence at the lower bound of that range: 100 months.

On appeal, Serrano argues for the first time that his base offense level should have been 22, not 24.

He contends that the lower base offense level is the right one because his 2006 felony conviction for assault is the only one of his prior convictions that qualifies as a crime of violence under the guidelines. Serrano also argues, as he did below, that the serial-number enhancement cannot apply because even though one serial number on the gun's frame was obliterated, another serial number on the slide remained unaltered. For that reason, he contends the District Court erred in adding four points to his total offense level.

If the District Court had used a base offense level of 22 and had not applied the serial-number enhancement, then, after the deduction for acceptance of responsibility, Serrano's total offense level would have been 19. With his criminal history category of V, his guidelines sentencing range would have been 57 to 71 months in prison. U.S.S.G. § 5A. Under the District Court's actual guideline calculation, by contrast, the range was 100– to 125–months.

II.

Serrano's first challenge is to the District Court's conclusion that his base offense level was 24 because he had two prior felony convictions that counted under the guidelines as convictions for a “crime of violence.” We start by describing how we usually decide whether a prior conviction is for a crime of violence. We then explain the problem with using that same approach here, given Serrano's failure to preserve the argument by properly raising it below.

A.

Ordinarily, we use what the precedents call a “categorical approach” to decide if a defendant's prior felony conviction was for a crime of violence. United States v. Jonas, 689 F.3d 83, 86 (1st Cir.2012). Under this approach, the conviction counts as one for a crime of violence if the elements of the conviction fit the guidelines' definition of a crime of violence. Id. at 86–87. Otherwise, the conviction does not count, no matter what the facts show the defendant actually did in committing the crime—even, that is, if those facts show he acted violently. Id. at 86.

This focus on the elements of the conviction—rather than the underlying conduct—fits with the text of the Sentencing Guidelines, which makes the base offense level for the felon-in-possession offense turn on prior “ convictions of ... a crime of violence,” not on prior conduct.U.S.S.G. 2K2.1(a)(2), (3) (emphasis added); see Descamps v. United States, ––– U.S. ––––, 133 S.Ct. 2276, 2287, 186 L.Ed.2d 438 (2013). And this approach also ensures present sentences are not based on documents that could be quite old, might be uncertain or disputed, and may contain factual allegations the defendant did not contest at the time for any of a number of reasons unrelated to the accuracy of the allegations. Descamps, 133 S.Ct. at 2289.

In some cases, though, this categorical approach runs into a potential obstacle. That obstacle arises when the conviction is for a crime set forth in a statute that is “divisible.” A divisible statute is one that “sets out one or more elements of the offense in the alternative—for example, stating that burglary involves entry into a building or an automobile.” Id. at 2281. The problem such a statute poses is that these alternative elements may create distinct offenses, each of which may or may not itself be a crime of violence.

To deal with this wrinkle, we employ what the precedents call—not surprisingly—a “modified categorical approach.” Under this approach, we look to limited materials, often called Shepard documents, from the convicting court, such as charging documents, plea agreements, plea colloquies, and jury instructions. Id. at 2281, 2284 (relying on Shepard v. United States, 544 U.S. 13, 125 S.Ct. 1254, 161 L.Ed.2d 205 (2005)). We do so not to determine the conduct the defendant engaged in while committing an offense, as such conduct is of no relevance. We instead inspect these materials in order to identify (if such identification is possible) the actual offense of conviction from among the distinct offenses set forth in a divisible statute. Id. at 2281.

Once we identify the distinct offense of conviction by consulting the materials, we then return to the categorical approach. We consider whether the elements of that distinct offense meet the definition of a “crime of violence.”

All of which brings us to the final stage in this process: the analysis of how the elements of the offense of conviction match up with the guidelines' definition of a “crime of violence.” A conviction for an offense qualifies as a conviction for a crime of violence if the elements of the underlying offense satisfy either (or both) of two clauses set forth in the relevant guideline and that offense is punishable by more than a year in prison. U.S.S.G. § 4B1.2(a); see alsoU.S.S.G. § 2K2.1 cmt. 1 (cross-referencing the definition in § 4B1.2 to determine the base offense level of the felon-in-possession crime).

The guideline's first clause provides that a crime of violence is “any offense under federal or state law ... that ... has as an element the use, attempted use, or threatened use of physical force against the person of another.” U.S.S.G. § 4B1.2(a)(1). This so-called “force clause” requires that the offense of conviction include as an element “ violent force,” that is, “force capable of causing physical pain or injury to another person.” Johnson v. United States, 559 U.S. 133, 140, 130 S.Ct. 1265, 176 L.Ed.2d 1 (2010). If the offense of conviction does not involve the “use, attempted use, or threatened use” of such violent physical force—as may be the case with an offense of common-law battery, whose force element can “be satisfied by even the slightest offensive touching”—then that offense does not meet the requirements of the force clause. Id. at 139, 130 S.Ct. 1265.

The guideline's second clause provides that a prior felony conviction qualifies as a crime of violence if it is for “any offense under federal or state law ... that ... is burglary of a dwelling, arson, or extortion, involves use of explosives, or otherwise involves conduct that presents a serious potential risk of physical injury to another.” U.S.S.G. § 4B1.2(a)(2). Even if an offense does not fall under the force clause, therefore, that offense qualifies as a crime of violence if it matches one of these enumerated crimes or otherwise satisfies the requirements of the guideline's so-called “residual clause.”

B.

In applying this framework, we begin by noting the parties agree that Serrano's 2006 conviction for assault under Puerto Rico law does count as a conviction for a crime of violence. We also note that Serrano does not dispute that the District Court counted the 2005 conviction for domestic violence under Article 3.1 in finding that Serrano had more than one conviction for a crime of violence. Serrano's challenge to the District Court's use of the base offense level of 24 can succeed, therefore, only if Serrano can show the District Court erred in counting that Article 3.1 conviction. Otherwise, there would be at least two such qualifying convictions. We thus now turn to the propriety of the District Court's finding on that point.

The first thing to note is that Article 3.1 is a divisible statute. It covers “[a]ny person who employs physical force or psychological abuse, intimidation or persecution against the person of [a domestic partner] ... to cause physical harm to the person, the property held in esteem by him/her, ... or to another's person, or to cause grave emotional harm....” P.R. Laws Ann. tit. 8, § 631 (emphasis added). The statute thus sets out multiple constellations of elements in the alternative. One set of elements requires the use or threat of “physical force.” The others require “psychological abuse, intimidation or persecution.”

Faced with such a statute, we ordinarily would apply the modified categorical approach. Using that approach, we first would try to determine, from the relevant documents, whether Serrano's prior conviction under Article 3.1 was for an offense predicated on the “physical force” element or instead for an offense predicated on the other elements set forth in that statute. Then, after having identified the actual offense of conviction, we would determine whether that offense met the guideline's requirements for a crime of violence.

But we are frustrated in doing so here. Serrano made no specific challenge to the pre-sentence report's contention that the list of his prior convictions included two felonies that were for a crime of violence. That was so even though that list included a conviction under Article 3.1 but did not specify further the particular offense under that law that had resulted in that conviction. At sentencing, moreover, the District Court simply identified as qualifying convictions the one for assault and the two for domestic violence. Yet Serrano did not complain that the District Court, in so finding, did not consult the limited set of documents from the court of conviction that would have helped it determine the distinct elements of the offense that provided the basis for Serrano's actual 2005 conviction under Article 3.1. In consequence, we have no such documents to review as part of the record on appeal.

As a legal matter, moreover, Serrano's failure to object in the District Court affects the standard of review. Rather than reviewing de novo whether the conviction under Article 3.1 counts as a conviction for a crime of violence, see Jonas, 689 F.3d at 86, we may review only for plain error, United States v. Ríos–Hernández, 645 F.3d 456, 462 (1st Cir.2011). And that standard is strict. Serrano can satisfy it “if, and only if, [he] succeeds in showing ‘(1) that an error occurred (2) which was clear or obvious and which not only (3) affected the defendant's substantial rights, but also (4) seriously impaired the fairness, integrity, or public reputation of judicial proceedings.’ ” United States v. Padilla, 415 F.3d 211, 218 (1st Cir.2005) (quoting United States v. Duarte, 246 F.3d 56, 60 (1st Cir.2001)).

Of course, if it were clear or obvious that none of Serrano's prior felony convictions—save for the 2006 one for assault—could qualify as one for a crime of violence, then the defendant's task on appeal might not be so daunting, despite the strict standard of review. But because Serrano was convicted under Article 3.1, and Article 3.1 is a divisible statute, we could come to that conclusion only if we were confident that none of the distinct offenses set forth in that law would so qualify. And, as we now explain, we are not of that view, given how we interpret one portion of Article 3.1.

C.

The case for concluding that at least one offense under Article 3.1 qualifies as a crime of violence is strong. Among the divisible offenses set forth in that statute is one that covers “[a]ny person who employs physical force ... to cause physical harm” to a protected person. P.R. Laws Ann. tit. 8, § 631.

In making physical force an element, the text of Article 3.1 suggests that something more than a mere non-consensual touching is required to satisfy that element. Instead, the text requires the physical force be intended to “cause physical harm.” The Puerto Rico Supreme Court has also interpreted the physical-force element of Article 3.1. And consistent with the text, that court has construed that element to “prohibit[ ] ... physical abuse,” Pueblo v. Ayala García, 186 P.R. Dec. 196, 213 (2012) (translation provided by stipulation of parties through letter under Federal Rule of Appellate Procedure 28(j)), and stated that “any degree of force is sufficient to configure the offense if ... employed with the intention of causing some damage,” id.; see also Pueblo v. Roldán López, 158 P.R. Dec. 54, 61 (2002).

Taken together, the text of Article 3.1 and the Puerto Rico Supreme Court's interpretation of it strongly suggest the statute's physical-force element involves the kind of violent force “capable of causing physical pain or injury to another person.” Johnson, 559 U.S. at 140, 130 S.Ct. 1265.

And that is the kind of force required by the crime of violence sentencing guidelines' force clause. Id.

To the extent any uncertainty remains, moreover, we do not believe it is so great as to make it clear or obvious that the physical-force offense set forth in Article 3.1 could not qualify as a crime of violence under the guideline. Yet it is just such a clear or obvious exclusion from the guideline that Serrano must demonstrate given that our review is for plain error.

Serrano argues, however, that he still should win because it is at least possible he was convicted of an offense under Article 3.1 that does not qualify as a crime of violence. And that is because, he contends, that statute is divisible and the elements of “psychological abuse, intimidation or persecution” plainly do not set forth an offense that is a crime of violence.

Serrano rests that fall-back contention on more than his assertion that those particular elements, by their plain terms, do not require “physical force” or a threat of such force. He also argues those elements establish distinct offenses that are too unlike the other crimes enumerated in the residual clause of the crime of violence guideline to be swept up by it. Cf. Begay v. United States, 553 U.S. 137, 142, 128 S.Ct. 1581, 170 L.Ed.2d 490 (2008) (holding that the enumerated crimes preceding the residual clause “illustrate the kinds of crimes that fall within the statute's scope” and “indicate[ ] that the statute covers only similar crimes”). Serrano then closes out this argument by contending that, without documents that show which elements in Article 3.1 supported his actual conviction under that law, there is no way to know whether that conviction qualifies as one for a crime of violence. And, in the face of that claimed uncertainty, he argues, it is plain error to hold that he was convicted of such a qualifying crime.

The government responds by arguing that uncertainty about what such documents might show is beside the point. The government argues that, in fact, all offenses described in Article 3.1 are crimes of violence, or, at least, that we should view them as such on review for plain error. And the government bases that contention on the residual clause of the crime of violence guideline, which, the government contends, encompasses all of those offenses. Or, at least, the government contends, the residual clause of the guideline does not clearly or obviously exclude them, whether they include the physical-force element or not.

But we do not need to resolve this dispute over how to characterize all parts of Article 3.1. Because our review is only for plain error, it is enough that we have determined that a conviction under the physical-force element of Article 3.1 would likely qualify as a crime of violence. For as we next explain, our precedents show that Serrano may not benefit from having left us completely in the dark (through his failure to object below) about what the documents relating to the conviction under Article 3.1 would reveal about whether he was convicted of an offense that contains the physical-force element or instead some other offense that does not require proof of that element.

D.

We confronted a situation very much like this in United States v. Turbides–Leonardo, 468 F.3d 34 (1st Cir.2006). There, the defendant also challenged his sentence on appeal because it rested in part on a conviction under a divisible statute, one portion of which contained elements that qualified for a guideline enhancement—there, for drug trafficking—and another of which did not. Id. at 37. And there, too, the defendant had not challenged either the pre-sentence report's characterization that the conviction was for an enhancement-qualifying offense, or the district court's guideline calculation that tracked the pre-sentence report. As a result, there were no records available on appeal to show which of the divisible statute's distinct offenses was in fact the offense of conviction. Id. at 40.

After finding the defendant's failure to object below, in context, actually constituted waiver, id. at 38—a claim that the government does not advance here—we went on to consider in dicta whether the application of the drug-trafficking guideline enhancement should be reversed under the plain error standard, id. at 38–40. And we concluded it should not. Id. We explained the District Court committed no error in accepting the unchallenged characterization, but that, even if the District Court had erred in doing so, reversal was still not justified. Id.

In consequence of the defendant's failure to object below, we explained, “we [we]re left to guess” the “unknown variable” of “the contents of the record of the prior conviction.” Id. at 40. And because we were left to guess, “there [wa]s no way for the appellant to show a reasonable probability that he would be better off from a sentencing standpoint had the district court not committed the claimed ... error.” Id. For that reason, we concluded the defendant could not meet the heightened prejudice showing plain error review requires. Id.

We then relied on Turbides–Leonardo's reasoning in holding there to be no prejudice in United States v. Davis, 676 F.3d 3 (1st Cir.2012), our last binding precedent on the issue. In Davis, the defendant challenged his sentence as relying on a prior conviction under a divisible assault statute, one portion of which defined a crime of violence and another portion of which did not. Id. at 7–8. Davis did not object when the prosecutor and the pre-sentence report characterized his conviction as qualifying as a crime of violence, nor did he object when the District Court characterized the conviction similarly and relied on it in crafting the sentence. Id. at 5–6. And so we reviewed only for plain error.

We held that, whether or not the District Court clearly erred by not demanding the documents of conviction before making the crime-of-violence determination, the defendant bore the burden of showing “a reasonable probability that, but for the error, the district court would have imposed a different, more favorable sentence.” Id. at 10 (quoting Turbides–Leonardo, 468 F.3d at 39). We then held, relying expressly on Turbides–Leonardo's reasoning about the need to show prejudice, that the defendant did not satisfy that burden because he failed to point to any reason to conclude that an examination of the documents would indicate the conviction was for an offense that does not qualify as a crime of violence. Id.

Here, just like in Davis, the District Court had before it a pre-sentence report that claimed the defendant had a second prior conviction that qualified for the guideline enhancement. And yet, again, like in Davis, the defendant did not contest that representation, even though the defendant informed the judge through counsel that he had reviewed the pre-sentence report containing that information.

Indeed, although the defendant made a general objection to the probation office regarding the total offense level used in the first version of the pre-sentence report, the record does not indicate that Serrano raised a more specific objection to the probation office regarding the base offense level and the number of his prior convictions for a crime of violence. And, the record further shows, he failed to do so even after the office clearly explained its view that Serrano had two such prior convictions.

Nor did the defendant raise an objection in his sentencing memorandum, or inform the District Court at sentencing that it believed it had erred in concluding—as it plainly stated in announcing the sentence—that, in addition to the 2006 conviction for assault, there was another qualifying conviction that was for domestic violence. The District Court thus had no Shepard documents before it—nor any request that it obtain and review such documents—that might cast doubt on either the pre-sentence report's assertion that the enhancement applied or on the defendant's apparent agreement with that assertion. Accordingly, we have no such Shepard documents before us now. And thus, as Davis—by incorporating Turbides–Leonardo's reasoning—instructs, we have no basis for concluding it is reasonably probable that those documents would show Serrano was convicted of an offense under Article 3.1 that would not qualify as a crime of violence.

In fact, even now, on appeal, Serrano still does not assert he was not convicted under Article 3.1 of the offense involving physical force, nor does he request to supplement the record to include the appropriate documents of conviction on the ground that they would redound to his benefit. See United States v. Zubia–Torres, 550 F.3d 1202, 1209 n. 3 (10th Cir.2008) (declining to consider “the effect if counsel had proffered the relevant documents on appeal”). He contends only that it cannot be certain on this record whether he was so convicted and that, in any event, the “physical force” offense clearly or obviously does not qualify—a contention we have already rejected.

Therefore, as in Turbides–Leonardo and Davis, we conclude Serrano has not shown the necessary prejudice, even assuming the District Court erred in not independently seeking out the records of conviction.

This conclusion comports with the decisions of several sister circuits in similar plain-error cases. See Zubia–Torres, 550 F.3d at 1208–10; United States v. Williams, 358 F.3d 956, 966–67 (D.C.Cir.2004); United States v. Ransom, 502 Fed.Appx. 196, 198–201 (3d Cir.2012) (unpublished). And while we are aware that other circuits have vacated sentences and remanded after finding plain error in arguably analogous circumstances, they did not, in so doing, address the lack-of-prejudice argument that the other circuits just mentioned have relied upon and that Davis requires us to find determinative here. See United States v. Reyes, 691 F.3d 453, 460 (2nd Cir.2012) (concluding, without explanation, that the district court's failure to sua sponte investigate the documents of conviction led to an erroneously elevated offense level); United States v. Castillo–Marin, 684 F.3d 914, 927 (9th Cir.2012) (same); United States v. Pearson, 553 F.3d 1183, 1186 (8th Cir.2009) (same), partially overruled on other grounds by United States v. Tucker, 740 F.3d 1177, 1184 (8th Cir.2014); United States v. Bonilla–Mungia, 422 F.3d 316, 321 (5th Cir.2005) (vacating and remanding without discussing prejudice).

We do not say, however, that there are no circumstances in which reversal in a related case, involving different facts, might be warranted. In United States v. Torres–Rosario, 658 F.3d 110 (1st Cir.2011), we held the District Court committed prejudicial plain error in characterizing the conviction at issue in that case as a crime of violence, id. at 116. But there, under First Circuit precedent, binding at the time of sentencing, it was clear from the charging documents in the record that the conviction qualified categorically as a crime of violence. Id. at 115. The defendant thus understandably did not contest the characterization or assert that the other documents of conviction would be relevant to whether the conviction was in fact qualifying, and indeed stipulated that his convictions qualified. Id. at 115–16.

By the time of the appeal, however, the First Circuit had changed course in response to a recent case from the Supreme Court. We had made clear that the type of conviction at issue did not necessarily qualify categorically as a crime of violence and, therefore, that further inquiry into the documents of conviction under the modified categorical approach would be appropriate. Id. at 115. For that reason, the un-objected-to characterization of the conviction in Torres–Rosario could not have been understood as an unchallenged agreement to a factual characterization of the conviction.

Here, by contrast, as in Turbides–Leonardo, no First Circuit precedent, later overruled, established at the time of sentencing that the conviction for the underlying offense categorically qualified as a crime of violence. And so the defendant's failure to contest the pre-sentence report's and the District Court's characterization of those prior convictions is, as Davis held in applying Turbides–Leonardo, key to our assessment that he has not met his burden of showing prejudice. And while Davis is itself a case with facts like Torres–Rosario, that does not make its express adoption of Turbides–Leonardo's prejudice analysis any less controlling in a case like this one, which mirrors the facts in Turbides–Leonardo rather the facts in Torres–Rosario. We thus do not address how Davis and Torres–Rosario's analysis of the prejudice issue should be reconciled in a case presenting the distinct facts presented in those cases.

III.

Under the Sentencing Guidelines, the offense level increases by four points if the firearm involved in a felon-in-possession conviction “had an altered or obliterated serial number.” U.S.S.G. § 2K2.1(b)(4)(B). Serrano's pistol had an obliterated serial number on the frame and an unaltered serial number on the slide. The District Court therefore applied the four-point serial-number enhancement.

Serrano argues, however, that the District Court erred because the serial number, though obliterated in one place, remained unaltered elsewhere on the gun. He contends that the guideline could not have been intended to apply in such circumstance because the serial number itself remains perfectly visible, albeit in only one place rather than two.

Whether Guideline § 2K2.1(b)(4)(B)'s four-point serial-number enhancement may apply in this type of case is a question of law (and, apparently, a question of first impression). Because Serrano properly preserved this argument below, our review is de novo. See United States v. Maldonado, 614 F.3d 14, 17 n. 1 (1st Cir.2010) (“Abstract legal issues under the guidelines are reviewed de novo....”).

Like the District Court, we conclude the enhancement does apply in Serrano's case. The text of the guideline requires only “ an altered or obliterated serial number,” U.S.S.G. § 2K2.1(b)(4)(B) (emphasis added). The guideline's text does not require that all of the gun's serial numbers be so affected. And here, the complete defacement of the serial number on the frame of the firearm resulted in the required obliteration.

Moreover, this plain reading of the text—that the obliteration of “a [ ]” serial number is enough—accords with the intent of Guideline § 2K2.1(b)(4), which is “to ‘discourag[e] the use of untraceable weaponry.’ ” United States v. Carter, 421 F.3d 909, 914 (9th Cir.2005) (alteration in original) (quoting United States v. Seesing, 234 F.3d 456, 460 (9th Cir.2001)). Applying an enhancement for firearms that have a single totally obscured serial number may serve as a deterrent to tampering, even when incomplete. And, relatedly, the single-obliteration rule could facilitate tracking each component that bears a serial number, given that various parts of firearms may be severable.

And precedent is not to the contrary. We have held the mere alteration of a serial number violates 18 U.S.C. § 922(k), a related criminal statute, without regard to whether such alteration is severe enough to prevent that same serial number from being read, United States v. Adams, 305 F.3d 30, 34 (1st Cir.2002); see also Carter, 421 F.3d at 915–16 (applying Adams to interpret Guideline § 2K2.1(b)(4)(B)). So, too, we conclude the text of this guideline is best construed—consistent with the plain meaning of its words—to trigger the enhancement when the serial number on the frame of a firearm is obliterated even if other serial numbers on the firearm, like the one left intact on the slide of this weapon, are unaltered.

We thus do not believe the District Court erred in applying the four-point enhancement.

IV.

For these reasons, we affirm the District Court's sentence against the challenges raised in this appeal.

LIPEZ, Circuit Judge, concurring.

A defendant whose sentence is enhanced because of violent crimes he committed in the past will face substantially more time in prison than someone without a record of violence. Although I do not question sentencing enhancements for defendants with violent criminal histories, we must ensure that aggravated penalties are imposed only when the criminal histories justify them. Here, appellant challenges the district court's unsupported assumption that his conviction under a “divisible” statute was in fact for a crime of violence. I reluctantly agree with my colleagues that First Circuit precedent requires us to reject appellant's claim. However, our case law on how to evaluate plain error in this context is inconsistent, and it cannot be reconciled with the Supreme Court's decision in Shepard v. United States, 544 U.S. 13, 125 S.Ct. 1254, 161 L.Ed.2d 205 (2005). I therefore urge our court to rehear this case en banc so that we may closely examine,and fairly resolve, an important and complex question of law: how does the government's burden to establish that a conviction under a divisible statute qualifies as a predicate offense intersect with a defendant's burden to show plain error?

Under the analysis described in the majority opinion, appellant can satisfy the prejudice prong of the plain error test only if he proves that, but for the sentencing court's improper reliance on his Article 3.1 conviction, it is reasonably probable that he would have received a lesser sentence. As a practical matter, that approach switches to defendants the obligation the Supreme Court imposed on the government to produce specific court records proving that a conviction under a divisible statute qualifies as a predicate offense. In so doing, the approach creates a real risk of longer prison terms than are justified by defendants' criminal histories. As other circuits have recognized, however, that potential harm can be easily avoided, with minimal burden on the sentencing court. When the court erroneously relies on a conviction whose character cannot be determined without Shepard-approved documents, the defendant's sentence must be vacated and the case remanded for resentencing. The government will then ordinarily have the opportunity to substantiate that the conviction was for an offense that qualifies as a predicate for enhancement. If the government cannot do so, the enhancement is impermissible.

As I explain below, this modest relief follows as a matter of logic and fairness from correct application of the plain error test in this context. Indeed, with a full understanding of the underlying principles, one can only conclude that the prejudice analysis articulated in our precedent—requiring the defendant to disprove his eligibility for a sentence enhancement—is misguided. Our court should convene en banc to remedy this serious problem.

I.

A. Legal Background

As my colleagues explain well, when a court seeks to enhance a defendant's sentence based on a prior conviction under a “divisible” statute—i.e., where the statute criminalizes different types of conduct, only some of which may support the enhancement—the court applies the so-called modified categorical approach to determine which version of the crime underlies the defendant's conviction. Descamps v. United States, ––– U.S. ––––, 133 S.Ct. 2276, 2283–84, 186 L.Ed.2d 438 (2013); see also United States v. Ramos–González, 775 F.3d 483, 505 (1st Cir.2015) (describing a “divisible statute” as one that “sets forth one or more elements of a particular offense in the alternative”) (quoting United States v. Fish, 758 F.3d 1, 6 (1st Cir.2014)). Frequently, as in this case, the enhancement depends on whether the challenged prior conviction was for a “crime of violence.”

See, e.g., Ramos–González, 775 F.3d at 504–05. If a statute criminalizes both violent and non-violent conduct, “the sentencing court is permitted to consult a limited set of ‘approved records' to determine which ... provided the basis for the conviction.” Id. at 505 (quoting United States v. Carter, 752 F.3d 8, 19 (1st Cir.2014)). The permissible records consist primarily of charging documents, plea agreements, transcripts of plea colloquies, jury instructions, and verdict forms. Id.; see also Shepard, 544 U.S. at 26, 125 S.Ct. 1254 (describing the acceptable records, often described as “ Shepard materials”). If the records show that the defendant was not convicted of a crime containing the requisite elements of violence—or if the records do not reveal the nature of the crime—the conviction may not be used to enhance his current federal sentence. See United States v. Dávila–Félix, 667 F.3d 47, 57 (1st Cir.2011).

At sentencing, the burden to produce the documents that reveal (or not) the nature of the proffered conviction is on the government. Dávila–Félix, 667 F.3d at 55 (“The Government bears the burden of establishing that a prior conviction qualifies as a predicate offense for sentencing enhancement purposes.”). Hence, if the government does not demonstrate that the defendant's conviction was for a variant of the crime that satisfies the crime-of-violence definition, it is error for the court to treat that conviction as a predicate for sentencing enhancement purposes. This is so whether the documents show the crime to be of the nonviolent type or if the documents do not reveal the particular version of the crime underlying the conviction. See United States v. Davis, 676 F.3d 3, 8 (1st Cir.2012) (“If, after examination of these permissible documents, ‘it is impossible to tell whether the defendant was convicted of a violent or non-violent offense,’ the conviction may not serve as a predicate offense.”) (quoting United States v. Holloway, 630 F.3d 252, 257 (1st Cir.2011)).

The complexity arises if the defendant fails to challenge the sentencing court's reliance on such a conviction, and raises an objection for the first time on appeal. We treat such a claim as forfeited and give it only plain error review.

Under that standard, the defendant “bear[s] the ‘heavy burden’ of showing that the error was clear or obvious, and that it both affected his substantial rights and ‘seriously impaired the fairness, integrity, or public reputation of judicial proceedings.’ ” Ramos–González, 775 F.3d at 499 (quoting United States v. Ramos–Mejía, 721 F.3d 12, 14 (1st Cir.2013)).

To perform this inquiry, we need to identify the “error” before we can determine if it is clear or obvious, and prejudicial. Focusing on the Puerto Rico statute under which the defendant was convicted, my colleagues explain that we may find plain error only if we are “confident that none of the distinct offenses set forth in that law” would qualify as a crime of violence. Op. at 845. Otherwise, they say, an error in using the conviction as a predicate for enhancement would not be clear or obvious. Moreover, drawing on our precedents, they conclude that the defendant cannot satisfy the prejudice prong of the plain error inquiry unless he shows “ ‘a reasonable probability that, but for the error, the district court would have imposed a different, more favorable sentence.’ ” Op. at 847 (quoting Davis, 676 F.3d at 10, which in turn quoted United States v. Turbides–Leonardo, 468 F.3d 34, 39 (1st Cir.2006)). To accomplish this showing of prejudice, the defendant, in effect, is required to produce Shepard materials revealing that his conviction was for a non-violent offense.

The animating principle of the modified categorical approach, however, is that enhanced sentencing is improper unless the government proves that the defendant's criminal history justifies such severe punishment. Error occurs, therefore, whenever a sentencing court increases a term of imprisonment based on a predicate conviction under a divisible statute in the absence of Shepard-approved proof that the conviction was for a qualifying variant of the crime. For that reason, the defendant's burden in the trial court is simply to note the absence of proof, not to proffer the supporting documents to disprove his eligibility for an enhancement. Under the approach my colleagues draw from prior cases, Serrano's failure to make that simple objection to the lack of proof transferred the duty of production to him on plain error review.

Although my colleagues understandably follow a path set out in prior cases, this dramatic shift of responsibility is unfair and wrong. I therefore first review why I view our precedent as flawed and incompatible with Supreme Court precedent before elaborating on what I believe is the proper analysis.

B. The Varying Paths of our Prior Cases

Our cases do not present a uniform approach for analyzing plain error in the context of a claim that the district court improperly lengthened a sentence based on the defendant's prior conviction under a divisible statute. In some instances, we have held the government accountable for the absence of evidence in the record. See, e.g., Ramos–González, 775 F.3d at 506–08 (vacating sentence that included career offender status because the records submitted by the government did not show the nature of defendant's conviction under a divisible statute); Dávila–Félix, 667 F.3d at 57 (concluding that, “on the record before us, the Government has not met its burden of proving that [defendant's] prior drug conviction qualified as a career offender predicate”); United States v. Torres–Rosario, 658 F.3d 110, 117 (1st Cir.2011) (noting that, “[o]n remand, the government remains entitled to establish the [basis for the sentencing enhancement] by showing that one of the assault and battery convictions was a crime of violence” (citation omitted)). In the latter two cases, however, the courts identified reasons why the defendants understandably failed to make an earlier challenge to the depiction of their convictions as qualifying predicates, thereby articulating justifications for remanding the case for resentencing notwithstanding the defendant's heavy burden on plain error review.

In Ramos–González, the government already had had multiple opportunities to prove career-offender status, and the panel declined to give the government a third chance. 775 F.3d at 508.

In other cases involving divisible statutes, panels of this court have held the defendants accountable for the absence of supporting documents in the record despite the government's burden to produce such records. See, e.g., Davis, 676 F.3d at 9–10; Turbides–Leonardo, 468 F.3d at 39–40. In these cases, the panels bypassed explicit identification of the error and—ostensibly addressing the prejudice prong of the plain error standard—articulated the requirement relied on by my colleagues: a defendant must show that, absent the error, he probably would have received a shorter sentence. Davis, 676 F.3d at 10; Turbides–Leonardo, 468 F.3d at 39.

As I explain below, the failure to confront the nature of the error is a threshold flaw in the Turbides–Leonardo and Davis assessments of plain error, and the mistake results in a misdirected prejudice analysis. As my colleagues recognize, the plain error analysis in Turbides–Leonardo was dicta, given the panel's statement that, “[a]ll things considered, we think that what transpired here amounted to waiver.” 468 F.3d at 38.

In Davis, the panel followed the Turbides–Leonardo dicta without analyzing its legal foundation, perhaps because the defendant's conduct there manifested waiver.

Davis complained that he should not be sentenced as a career offender, but he never argued that career-offender status was improper because the district court failed to determine the nature of the pertinent predicate conviction. See 676 F.3d at 6 n. 2, 7, 10 n. 7. In fact, appellate counsel twice sought to withdraw on the ground that he “ ‘could not discern a non-frivolous basis for appeal.’ ” Id. at 6 n. 2 (quoting counsel's brief). The panel refused those requests and directed counsel to address the plain error standard. Counsel, however, did not submit briefing on plain error and, “when questioned at oral argument regarding any potential prejudice to Davis based on the district court's failure to undertake the categorical approach or to examine the character of Davis's 2006 assault and battery conviction, Davis's counsel could not point to any.” Id. at 10 n. 7.

In these circumstances, I can understand how the Davis panel came to rely on the Turbides–Leonardo approach to plain error without closely examining it or explicitly acknowledging it as dicta. Treating Davis's claim as forfeited rather than waived was generous and, given that Davis did not raise the district court's failure to apply the modified categorical approach even on appeal, the panel had no reason to probe deeply into the Turbides–Leonardo articulation of the inquiry. Here, by contrast, Serrano develops his claim that the district court erred by counting his domestic violence offense as a predicate crime of violence, asserting, inter alia, that some crimes under Article 3.1 “clearly do not involve the use of violent force.” Br. at 23. Nonetheless, because Davis applies the plain error test to a scenario it labels as forfeiture, it appears to be binding precedent on the application of the plain error test where, as here, there are no distinguishing facts like those in Ramos–González, Dávila–Félix, or Torres–Rosario.

The fact remains, however, that our cases fail to deal consistently with the government's initial burden of proof in the plain error context. Where the government was required to retain the burden to prove the nature of the defendant's conviction, the courts relied on particular circumstances—a change in the law, the convictions' non-essential role in the prior sentencing, or the government's multiple prior attempts—to explain the defendants' default or find the burden unmet. In the two instances where the burden was switched from the government to the defendant, the courts dealt explicitly or de facto with an intentional relinquishment of the defendant's rights—a waiver—and avoided the question of what error the court committed. We have not examined how, or if, these cases may be reconciled with each other and whether they achieve the objectives of the modified categorical approach. Furthermore, the uneven treatment within our own circuit is reflected in a conflict among the circuits. Compare, e.g., United States v. Dantzler, 771 F.3d 137, 149 (2d Cir.2014) (“The absence of an objection will not relieve the Government of its burden of proving through Taylor—and Shepard-approved sources that the ACCA enhancement applies.”), with, e.g., United States v. Zubia–Torres, 550 F.3d 1202, 1209 (10th Cir.2008) (“By failing to present any evidence that relevant documents would indicate his conviction was not for [a qualifying predicate offense], the defendant has failed to meet his burden under the third prong of plain error review.”).

In sum, we lack a thoughtful, uniform analysis for assessing plain error when a defendant claims that his sentencing enhancement was improperly based on an unexamined conviction under a divisible statute. Our court, en banc, should take the opportunity to develop such an analysis in this case.

C. The Correct Approach

To properly conduct the plain error inquiry, a court must have a correct understanding of the error at issue. As described above, some of our cases have sidestepped the question of error to focus on the question of prejudice. In so doing, however, those courts performed an analysis premised on a misidentification of the error, which leads them to cast aside the government's burden of proving the basis for an enhancement. In Turbides–Leonardo and Davis, the panels focus on the enhanced sentence, and consequently evaluate prejudice by asking the usual question we ask when sentences are reviewed for plain error: is it reasonably probable that, but for the error, the defendant would have received a lower sentence? The length of the sentence—though ultimately our concern—is not the “plain” error. Because the government initially bears the burden to prove that a conviction represents a crime of violence, Dávila–Félix, 667 F.3d at 55, the error occurs when the district court enhances a sentence based on a prior conviction under a divisible statute without first confirming that the conviction qualifies as a predicate offense. That confirmation may be achieved in various ways: through documentary evidence (i.e., the Shepard materials), by concession of the defendant, or by means of an interpretation of the predicate criminal statute—i.e., a legal ruling by the court—that every variant of the offense qualifies as a crime of violence.

Here, where the statute on its face appears to encompass alternatives that neither involve physical force against a person nor present a “serious potential risk of physical injury to another,” U.S.S.G. § 4B1.2(a),

the court erred by using the conviction to enhance Serrano's sentence without demanding proof from the government that the defendant's conviction was for a violent version of the divisible crime. It is possible that the enhancement is also erroneous because the conviction at issue was not, in fact, a crime of violence. But to find that the sentencing judge erred in applying the modified categorical approach, an appellate court need not reach the nature of the conviction. Error has occurred when the court relies on a conviction under a divisible statute without confirming, through approved sources provided by the government, that the conviction represents a crime of violence.

The failure to recognize this error is what led the Turbides–Leonardo panel astray. Its approach looks beyond the district court's erroneous reliance on an unelaborated conviction under a divisible statute and asks whether the defendant has shown that the proper analysis would have revealed that the conviction was erroneously used as a predicate for enhancement. Even if the district court had performed the proper analysis, however—involving the scrutiny of Shepard-approved documents—the inquiry may not have shed light on the predicate conviction. The government may not have been able to produce appropriate records of the targeted conviction—the documents may be inaccessible or no longer exist, meaning that the conviction could not be used to enhance the defendant's sentence. Hence, by focusing on the possibility that the defendant was convicted of a qualifying crime, and requiring him to prove that he was not, we unfairly leap over the threshold analytical error, i.e., the sentencing court's failure to require the government to establish the nature of the conviction through approved sources.

If that error were properly acknowledged, the plain error analysis here would unfold unequivocally in the defendant's favor. Given the broad language of Article 3.1, and the dearth of evidence indicating whether the defendant was convicted of a crime of violence, the court's error in relying on the unexamined conviction was sufficiently “plain” to satisfy the second prong. The gap in the record should have been obvious to the court. The remaining two elements are equally straightforward. A defendant inescapably suffers prejudice when he receives an extended term of imprisonment without the evidentiary support necessary to justify it,

and an unsupported,prolonged incarceration must be deemed a miscarriage of justice. See Ramos–González, 775 F.3d at 507 & n. 29; Torres–Rosario, 658 F.3d at 117.

Admittedly, this plain error analysis has the feel of allowing the defendant to escape with little disadvantage from his failure to make a timely objection. All four prongs of the plain error inquiry effectively turn on the finding that the error was plain, and the error will almost always be plain when there are no supporting documents in the record. Importantly, however, the typical remedy for a finding of prejudicial plain error in this context is simply a remand for development of the sentencing record. In many instances, the government on remand will be able to produce the necessary documents to substantiate the qualifying predicate offense, and the defendant's “victory” will be short-lived. This is the approach taken by a number of circuits. See, e.g., United States v. Reyes, 691 F.3d 453, 459–60 (2d Cir.2012) (per curiam) (finding plain error requiring remand where the district court relies on the PSR to characterize an offense as a “crime of violence,” “even where the defendant does not object to the PSR's description”); United States v. Castillo–Marin, 684 F.3d 914, 919, 927 (9th Cir.2012) (same); United States v. Boykin, 669 F.3d 467, 469–72 (4th Cir.2012) (finding plain error and remanding for resentencing where the district court relied on the PSR to conclude that the defendant had the requisite number of violent felonies for ACCA enhancement); United States v. McCann, 613 F.3d 486, 502 (5th Cir.2010) (“When a court ... relies on the PSR alone [to characterize an offense as a crime of violence], it makes an error that is clear and obvious.”).

Moreover, we must acknowledge the potentially severe consequences of using prior convictions improperly—substantially prolonged terms of incarceration.

Undoubtedly, that harsh impact underlies the Supreme Court's carefully circumscribed list of acceptable records for confirming that a conviction under a divisible statute may be used to enhance a sentence. At the same time, it is an unfortunate reality that many claims such as Serrano's come to us on plain error review. Criminal defendants often must rely on court-appointedcounsel who, faced with a myriad of trial and sentencing issues, predictably overlook some of them. The extremely high hurdle to post-conviction relief based on ineffective assistance of counsel means that such a remedy is uncertain at best.

In short, there is simply no reason to apply plain error in a way that will leave intact lengthy, possibly unjustified terms of imprisonment when the cost of ensuring fairness—a resentencing proceeding—is minimal. We should not be uncomfortable with an “easy” showing of plain error, even recognizing the high bar that the plain error standard ordinarily represents. Indeed, the fourth prong of the plain error test requires us to consider “ ‘the fairness, integrity, or public reputation of judicial proceedings.’ ” United States v. Mercado, 777 F.3d 532, 536 (1st Cir.2015) (quoting United States v. Duarte, 246 F.3d 56, 60 (1st Cir.2001)). The plain error approach we apply in this case is incompatible with those concerns.

D. The Role of the PSR

The mistaken approach to plain error adopted in Turbides–Leonardo reflects the confusion in our law about when it is appropriate to rely on an unobjected—to PSR to prove a defendant's criminal history. Courts may accept the PSR's representation of the existence of a prior conviction in the absence of objection. See, e.g., United States v. Jimenez, 512 F.3d 1, 7 (1st Cir.2007) (stating that, where an offense listed in a presentence report “is not disputed before the sentencing court, the report itself is competent evidence of the fact stated and, thus, is sufficient proof of that fact”); United States v. Brown, 510 F.3d 57, 74 (1st Cir.2007) (describing the government's burden of proving a predicate conviction for sentencing purposes as “modest,” and noting that it can be satisfied by, inter alia, “introducing a certified copy of the judgment, or by a statement in the PSR”).

However, courts are not permitted to rely on the PSR to establish the character of a conviction under a divisible statute. A decision to accept the PSR as adequate evidence of the nature of a defendant's prior crimes would conflict with the Supreme Court's directive that the particular offense committed in violation of a divisible statute be determined through examination of Shepard-approved documents. Indeed, police reports are a typical source of the facts reported in a PSR, see, e.g., Davis, 676 F.3d at 8–9 (noting that the PSR's summary of a prior crime was based on a police report), and police reports are expressly excluded from the list of approved documents, see, e.g., Ramos–González, 775 F.3d at 506 (noting that we “may not rely on the police reports related to the earlier conviction” (quoting Carter, 752 F.3d at 20 (citing Shepard, 544 U.S. at 16, 125 S.Ct. 1254))). Although the PSR will commonly include the details of the defendant's criminal conduct, it will not necessarily reveal the pertinent information for the modified categorical approach, i.e., the specific elements of the crime underlying the listed convictions. See, e.g., Descamps, 133 S.Ct. at 2283 (“The key [in determining whether a prior conviction can serve as an ACCA predicate] ... is elements, not facts.”); id. at 2289 (noting that a defendant may have pled guilty to a less serious version of the crime than reflected in factual statements “found in the old record”).

Thus, although our cases unequivocally allow a sentencing court to rely on the PSR to confirm the existence or validity of convictions in the absence of an objection, other cases properly recognize that such deference cannot extend to the question whether convictions under a divisible statute represent qualifying predicates for sentencing enhancements. See, e.g., Dávila–Félix, 667 F.3d at 56–57 (rejecting government's reliance “primarily upon the facts as recounted in the presentence investigation report,” despite the defendant's failure to object to the PSR's analysis); Jimenez, 512 F.3d at 7 (stating that sufficient proof of the two prior convictions “does not necessarily end our inquiry” because “[i]n some circumstances, the question would remain whether the underlying offenses qualify as controlled substance offenses within the meaning of the applicable sentencing guideline”).

Yet, in Turbides–Leonardo, the panel cited a single Eighth Circuit case for the proposition that a PSR “may be a permissible source of information about a prior conviction for sentence enhancement purposes” to bolster its conclusion that the district court acted “reasonabl[y]” in relying on the uncontroverted PSR to enhance the defendant's sentence based on a conviction under a divisible statute. 468 F.3d at 39 (citing United States v. Arrieta–Buendia, 372 F.3d 953, 955–56 (8th Cir.2004)). In the Eighth Circuit case, however, the defendant had admitted his conviction for a type of crime that qualifies as a predicate offense. See Arrieta–Buendia, 372 F.3d at 955 (stating that the defendant “told the district court he was not guilty of the California felony of transporting methamphetamine, but was forced to plead guilty to that crime”). Arrieta–Buendia is not only an out-of-circuit precedent, but it also is inapt where, as here, the PSR does not reveal whether a conviction under a divisible statute is an eligible predicate offense and the defendant has not waived or conceded the point.

To some extent, the panel in Davis recognized the difference between using a PSR to prove the fact of a conviction under a divisible statute and relying on the report to establish the specific elements of the crime underlying that conviction. At issue in Davis was whether a conviction for assault and battery was a predicate offense for career offender status. 676 F.3d at 7. The panel noted that the only evidence in the record indicating the violent nature of the offense was in the PSR, with details drawn from police reports. Id. at 5, 8–9. The defendant, however, did not object to the PSR's characterization of the offense as a crime of violence, and he did not contest the government's characterization of him as a career offender at the sentencing hearing. Id. at 6. On appeal, as discussed above, the panel found that the defendant had failed to satisfy the prejudice prong of the plain error inquiry: “[Defendant] has made no argument that the assault and battery was anything other than the harmful type, doing nothing, even on appeal, to question the description provided in the PSR or to argue that appropriate Shepard materials would prove that he committed a non-harmful battery.” Id. at 10.

The Davis panel, however, directly confronted the adequacy of the PSR to show the requisite violent conduct. It first quoted the assertion in Torres–Rosario that “ ‘treating a Massachusetts assault and battery conviction as a [career offender] predicate, without further evidence of violence, is now plain error.’ ” Id. at 9 (quoting Torres–Rosario, 658 F.3d at 116) (alteration and emphasis in Davis ). The Davis panel then went on to speculate that “the description in the PSR might constitute such further evidence,” and, for that reason, “this case does not neatly fall within the plain error standards we set in Torres–Rosario.” Id. In an immediately following footnote, the panel observed that “[w]e have never squarely addressed whether reliance on a PSR under these circumstances is proper,” but noted prior dicta indicating that, even though police records are not “permissible Shepard materials,”“we would approve of the use of a PSR's summary of police reports to support the characterization of a predicate offense when the defendant did not object to the PSR.” Id. at 9 n. 6.

This arguable approval in Davis of unchallenged police reports in a PSR to establish the character of a predicate offense is weakly grounded in our precedent and contrary to Shepard. The precedent cited for this proposition is Jimenez, where the panel's primary focus was on whether challenged predicate crimes listed in the PSR were adequately verified, not on the convictions' character for the modified-categorical inquiry. See Jimenez, 512 F.3d at 6 (noting appellant's argument that “the district court erred when it relied on the PSI Report for proof of these prior convictions”). Jimenez did not argue that either of the challenged convictions was “for an offense that falls outside the contemplation of the career offender provisions.” Id. at 5 n. 3; see also id. at 7 (observing that “appellant has made no argument, either in the lower court or in this court, that his prior convictions, if properly substantiated, do not qualify as convictions for controlled substance offenses”).

Davis thus contemplates disregarding the Supreme Court's explicit restriction on what documents may be consulted to determine the nature of a predicate conviction under a divisible statute, allowing reliance on materials (i.e., police reports) that have been expressly designated as unacceptable for this purpose. See Shepard, 544 U.S. at 16, 125 S.Ct. 1254. In my view, however, we are not free to depart from the Supreme Court's methodology for determining the eligibility of a predicate offense, even in the context of plain error. That methodology, designed to ensure that prolonged sentences are justified, has substantive importance. See Dantzler, 771 F.3d at 149 (stating that the defendant's failure to object did not “render the PSR's description more reliable in establishing the requisite” predicate). That is why, when a sentencing judge errs by failing to demand Shepard-approved proof that the defendant's conviction under a divisible statute was for a predicate offense, a sentencing enhancement cannot stand if its only foundation is the defendant's PSR, at least when the report is not drawn from approved sources. Accord Reyes, 691 F.3d at 459 (“We have little trouble concluding that a sentencing court may not rely on a PSR's description of a defendant's pre-arrest conduct that resulted in a prior conviction to determine that the prior offense constitutes a ‘crime of violence’ under U.S.S.G. § 4B1.2(a)(1), even where the defendant does not object to the PSR's description.”).

The Turbides–Leonardo approach, however, indirectly gives Serrano's PSR dispositive weight by rejecting his claim that the record does not support classifying his Article 3.1 offense as a crime of violence. The district court accepted the base offense level calculation recommended in the PSR, which was premised on multiple prior convictions—including under Article 3.1—for crimes of violence. In failing to require proof of the actual basis for Serrano's convictions, the district court necessarily deferred to the PSR's depiction of his offenses. By leaving the district court's reliance on the PSR undisturbed (unless the defendant comes forward with contrary evidence), we are sanctioning that deference. Yet, as I have shown, any suggestion in our cases that such deference may be permissible developed from inapplicable precedent and, more importantly, contravenes the Supreme Court's specific delineation in Shepard of the records that may substantiate the eligibility for enhancement of a conviction under a divisible statute.

In light of this analysis, the district court plainly erred in deferring to the PSR—or, as described above, in failing to demand acceptable forms of proof from the government. The defendant's failure to make a timely objection imposes on him the burden to show that he suffered from the court's error. We should conclude that his burden is easily met—and a resentencing required—if the court relied on such convictions to enhance his sentence.

II.

In examining a claim of plain error in the context of the modified categorical approach, we cannot lose sight of the courts' obligation to ensure that extended incarceration is imposed only when the government has proven that it is justified by a defendant's criminal history. We can, and should, meet this obligation by adopting the Second Circuit's (and other courts') approach that a “defendant's failure to object d[oes] not cure the Government's failure to submit the proper evidence.” Dantzler, 771 F.3d at 149 (describing the holding in Reyes, 691 F.3d at 459). As I have explained, requiring the government to retain its burden to justify a sentencing enhancement does not result in excusing the defendant's default. Nor will a finding of prejudicial plain error and the required relief—at most, a new sentencing proceeding—impose undue burden on the court. Although few sentences may be changed through this process, “the fairness, integrity, [and] public reputation of judicial proceedings” will be enhanced. Mercado, 777 F.3d at 536 (internal quotation marks omitted).

In this case, where variants of Article 3.1 do not include the requisite element of violence, we should not reject appellant's claim on the ground that some offenses under the statute would qualify as predicate crimes of violence. The district court committed plain error when it relied on Serrano's conviction under that divisible statute to justify an increased term of imprisonment in the absence of approved forms of evidence of the nature of his particular crime. The court should have insisted that the government shoulder its burden to substantiate that Serrano's conviction was in fact a qualifying predicate. Hence, on en banc review, this court should hold that Serrano is entitled to a new sentencing proceeding where the government may seek to show that his conviction was for a crime of violence.