Dissenting opinion filed by Senior Circuit Judge RANDOLPH.
TATEL, Circuit Judge:
This appeal, sealed due to appellant’s failed attempt to cooperate with the government, concerns the application of the United States Sentencing Guidelines’ career offender provision. After appellant pled guilty to possession of cocaine base with intent to distribute, the district court sentenced him as a career offender based on its finding that he had two prior felony convictions for crimes of violence — robbery and armed robbery. Although the parties agree that we must remand to allow the district court to consider a non-Guidelines sentence, appellant argues that he does not qualify as a career offender because his two prior convictions should have been counted as one, and in any event the robbery conviction did not qualify as a crime of violence. We agree with the district court’s decision to count appellant’s prior convictions separately, but we find that the government failed to establish that appellant’s plea to robbery, as defined in the District of Columbia Code, necessarily admitted the elements of a crime of violence.
I.
Appellant pled guilty to a one-count information charging him with possessing five or more grams of cocaine base with intent to distribute in violation of 21 U.S.C. § 841(a)(1), (b)(l)(B)(iii). The district court sentenced appellant as a career offender under U.S.S.G. § 4B1.1, a provision that imposes a substantially higher Guidelines range on those who commit certain offenses after being convicted of two prior drug crimes or crimes of violence.
Although the record documenting appellant’s prior convictions is thin, several facts are undisputed. On October 1, 1986, police officers arrested appellant for an armed robbery he committed on September 17, 1986. Three months later, while appellant remained in custody, the prosecutor filed a second complaint charging him with committing a robbery on August 6, 1986. Appellant pled guilty to both criminal complaints on February 6, 1987, and the same Superior Court judge sentenced him in both cases on March 20, 1987.
In sentencing appellant on the federal drug charge, the district court found that the prior D.C. convictions were unrelated and thus properly counted as two convictions, and determined that they both qualified as crimes of violence. Accordingly, it concluded that appellant was a career offender and sentenced him to a 188-month [1087]*1087term of imprisonment, within the 60- to 480-month statutory range provided by section 841 (b)(1)(B)(iii).
Appealing his sentence, appellant argues that the district court erred in finding both that his two prior convictions were unrelated and that one — the robbery conviction— qualified as a crime of violence. He also argues that the district court misapprehended its authority to sentence outside the Guidelines range under Kimbrough v. United States, — U.S. -, 128 S.Ct. 558, 575, 169 L.Ed.2d 481 (2007) (holding that district judges may conclude that the Guidelines’ disparate treatment of cocaine and cocaine base warrants a non-Guidelines sentence), and imposed a sentence “greater than necessary” to achieve the goals of punishment in violation of 18 U.S.C. § 3553(a)’s parsimony provision. Because the government concedes that remand for resentencing is appropriate in light of Kimbrough, we have no occasion to consider either that issue or the parsimony challenge. The only issue we need resolve is whether the district court properly adjudicated appellant a career offender under section 4B1.1.
II.
Section 4B1.1 of the November 1, 2005 Sentencing Guidelines enhances the offense level for “career offenders.” The Guidelines define a “career offender” as an individual who (1) is convicted of a felony crime of violence or controlled substance offense, (2) was at least eighteen years old at the time of that offense, and (3) has “at least two prior felony convictions of either a crime of violence or a controlled substance offense.” § 4Bl.l(a). The parties agree that these Guidelines govern appellant’s July 26, 2006 sentencing, as well as his resentencing on remand from this appeal. See 18 U.S.C. § 3742(g)(1). They also agree that on remand the district judge may consider the current Guidelines, under which appellant would not be deemed a career offender, as a factor relevant to the imposition of sentence. They disagree about whether, under the 2005 Guidelines, appellant’s convictions may be counted separately and whether both may be counted as crimes of violence. We consider each issue in turn.
Relatedness of Prior Convictions
Under section 4B1.1, two prior felony convictions are treated as one if “related” within the meaning of U.S.S.G. § 4A1.2(a)(2). See § 4B1.2(e)(2). Regardless of any factual similarities or differences between them, two convictions are “related” in this technical sense if they are for offenses that were not separated by an intervening arrest and “(A) occurred on the same occasion, (B) were part of a single common scheme or plan, or (C) were consolidated for trial or sentencing.” § 4A1.2 cmt. n. 3 (2005). As it is undisputed that appellant’s crimes were neither separated by an intervening arrest nor committed on the same occasion or as part of a common scheme or plan, the question boils down to whether the sentences were for offenses that were consolidated for trial or sentencing.
Our sister circuits disagree about whether formal consolidation, i.e., an actual consolidation order, is required, or whether offenses may still be found “functionally consolidated” without such an order. Compare, e.g., United States v. Adams, 509 F.3d 929, 933 (8th Cir.2007) (requiring a formal consolidation order), with United States v. Best, 250 F.3d 1084, 1095 (7th Cir.2001) (allowing functional consolidation when there is an indication that the trial court considered the cases sufficiently related to be treated as one). This circuit has yet to rule on the issue, which turns out to have little prospective importance given that the current Guidelines jettison the concept of consolidation. [1088]*1088U.S.S.G. § 4A1.2(a)(2) (2007). Nor need we answer the question for purposes of this case. Whether consolidation is assessed formally or functionally, the district court was correct in finding that appellant’s convictions were never consolidated. As the Superior Court entered no formal order of consolidation, appellant’s convictions were unquestionably never formally consolidated. Nor would appellant fare any better under a functional consolidation standard. Although appellant offers several reasons why he thinks he would prevail if we adopted such a standard, none is persuasive.
Appellant first argues that the district court erred by applying a formal test. To be sure, were we to adopt a functional consolidation standard, it would have been error for the district court to have applied a formal test. But we see no evidence that it did. At the first sentencing hearing, the district court noted an out-of-circuit precedent that it took to apply a formal consolidation test and adjourned the sentencing to consider this authority. Sent’g Tr. at 16-17 (Feb. 24, 2006). In doing so, the district court cautioned that it was unlikely to adopt a formal consolidation test because it thought “the reality of what happens, the substance of what happens,” should govern consolidation findings. Id. at 17. At the beginning of the second hearing, the district court said it needed no further argument to determine whether appellant was a career offender, stating, “The Court finds that [appellant] is a career offender because he has been convicted in the past, and the evidence of that is sufficient. He has been convicted in the past twice of committing crimes of violence. And the evidence is sufficient, in this Court’s view, to show that.” Sent’g Tr. at 3 (July 26, 2006). Pressed by defense counsel, the court elaborated: “[The convictions] are completely unrelated, in my view, I so find. I made that — I made — I made that finding as a matter of law based upon, frankly, what is undisputed by you or anybody else.” Id. at 4.
Appellant insists that the district court’s “matter of law” comment signals that it adopted a formal test after all, but we read the remark differently. The lengthy dialogue surrounding the statement makes clear that the district court based its finding on the facts the defendant himself set forth in his papers and in court — the dates of appellant’s offenses, separated by over a month — not the simple absence of a formal order. See id. at 9-10 (agreement between the court and defense counsel on the relevant dates). This exchange, combined with the district court’s finding that the convictions were “completely unrelated,” id. at 4, and its previous rejection of a formal test, demonstrates that the district court used a functional standard.
Appellant next argues that even if it used a functional test, the district court failed to make a finding that the offenses were not functionally consolidated. We disagree. Though terse, the district court’s statement that the convictions were “completely unrelated” amounts to a factual finding. In arguing otherwise, appellant ignores the fact that his trial counsel, following a lengthy dialogue on precisely this point, essentially admitted that the court had satisfactorily explained the basis for its finding. Id. at 10.
Appellant claims that he would prevail under a functional consolidation standard for a third reason: even if the district court found that the offenses were not functionally consolidated, it clearly erred in doing so. Reviewing the district court’s finding deferentially, we see no grounds for disturbing it. See Buford v. United States, 532 U.S. 59, 66, 121 S.Ct. 1276, 149 L.Ed.2d 197 (2001) (holding that functional consolidation findings are subject to deferential review on appeal without deciding whether functional consolidation is the [1089]*1089proper standard). Given the undisputed fact that the two convictions were for offenses involving different victims on different dates and the absence of any affirmative evidence that the Superior Court treated the two convictions as one, we are not “left with the definite and firm conviction that a mistake has been committed,” United States v. Seiler, 348 F.3d 265, 268 (D.C.Cir.2003) (internal quotation marks omitted).
Crime of Violence
This brings us to the second question: whether appellant’s robbery conviction qualifies as a crime of violence. Section 4B1.2 defines a crime of violence as, among other things, any felony 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.” § 4B1.2(a)(l). As this language mirrors the definition of a “violent felony” under the Armed Career Criminal Act (ACCA), 18 U.S.C. § 924(e)(2)(B)®, we apply the ACCA standard to determine whether an offense qualifies as a crime of violence under section 4B1.2. See, e.g., United States v. Andrews, 479 F.3d 894, 897 (D.C.Cir.2007) (applying ACCA cases directly to section 4B1.2 “crime of violence” inquiry).
Under the ACCA standard, we examine not the defendant’s actual conduct, but rather the elements of the offense of conviction. See James v. United States, 550 U.S. 192, 127 S.Ct. 1586, 1594, 167 L.Ed.2d 532 (2007). Ordinarily a straightforward inquiry, this question becomes more difficult where, as here, a single statutory provision “covers both violent and non-violent crimes.” Andrews, 479 F.3d at 897. The District of Columbia robbery statute in effect at the time of appellant’s offense provided that:
Whoever by force or violence, whether against resistance or by sudden or stealthy seizure or snatching, or by putting in fear, shall take from the person or immediate actual possession of another anything of value, is guilty of robbery, and any person convicted thereof shall suffer imprisonment for not less than 2 years nor more than 15 years.
D.C.Code § 22-2901 (1987). As we explained in United States v. Mathis, by defining “force or violence” to include the minimal level of force necessary to obtain property “by sudden or stealthy seizure or snatching,” the statute covers offenses that fail to qualify as crimes of violence under section 4B1.2. 963 F.2d 399, 408-09 (D.C.Cir.1992).
Despite the fact that the D.C. statute covers both violent and nonviolent crimes, we still look to the nearest possible analogues to the elements of the offense of conviction, rather than to the defendant’s conduct. Specifically, when the conviction results from a trial, we look to the statutory definition as well as “the indictment or information and jury instruction” to see if “the jury necessarily had to find” the elements that would qualify the offense as a crime of violence. Taylor v. United States, 495 U.S. 575, 602, 110 S.Ct. 2143, 109 L.Ed.2d 607 (1990) (emphasis added). Where, as here, the conviction results from a guilty plea, the Supreme Court’s decision in Shepard v. United States requires that we look to the “statutory definition, charging document, written plea agreement, transcript of the plea colloquy, and any explicit factual finding by the trial judge to which the defendant assented,” 544 U.S. 13, 16, 125 S.Ct. 1254, 161 L.Ed.2d 205 (2005); see also id. at 20, 125 S.Ct. 1254 (specifying that relevant factual findings are those to which the defendant assents “upon entering the plea”), in order to determine whether the defendant “necessarily admitted” the qualifying elements, id. at 26, 125 S.Ct. 1254 (emphasis added). We [1090]*1090may look to no other materials, even if they might shed light on what the defendant probably did or probably admitted to, or what the jury probably found. See, e.g., id. at 26, 125 S.Ct. 1254 (holding that courts considering whether a defendant necessarily admitted to a qualifying offense can look only to the charging document, the plea agreement or colloquy, or comparable judicial records of the factual basis of the plea).
Applying these principles, we held in Mathis that in order for a conviction under the D.C. robbery statute to qualify as a violent felony under ACCA, the court had to determine that, in cases where the defendant went to trial, “the jury necessarily found” (that is, was “required to find” in order to convict) that the defendant had used sufficient force to qualify the offense as an ACCA predicate. 963 F.2d at 409 (emphasis added). There we made clear that the inquiry was not into what the defendant probably did or what the jury probably found:
If the record of the previous conviction is ambiguous — i.e., if it is possible, on the basis of the indictment or jury instructions, to conclude that the element of force was satisfied solely by a jury finding that the defendant had lifted money out of a purse — then the conviction may not be used as a [violent felony under ACCA],
Id. at 410. We employ the same approach in cases involving guilty pleas. See Shepard, 544 U.S. at 19, 125 S.Ct. 1254 (“Taylor’s reasoning controls the identification of generic [i.e., ACCA-qualifying] convictions following pleas, as well as convictions on verdicts, in States with nongeneric offenses”).
We review de novo the district court’s determination that appellant’s robbery conviction qualified as a crime of violence. United States v. Hill, 131 F.3d 1056, 1062 n. 5 (D.C.Cir.1997). Because the government failed to produce the transcript of appellant’s Superior Court plea colloquy, the district court had no record of what facts were necessary to appellant’s plea. For this reason, it examined an arrest warrant affidavit alleging that appellant committed the robbery at gunpoint; a detention order in the armed robbery case describing the robbery offense as an armed robbery as well; the judgment recording appellant’s robbery plea; and the criminal information charging appellant with robbery. The district court found that each of these documents demonstrates that appellant was convicted of a crime of violence. Sent’g Tr. at 119 (July 26, 2006).
The parties agree that under Shepard the district court should have examined neither the warrant affidavit nor the detention order. Because the judgment records only that appellant pled guilty to robbery, it sheds no light on whether he pled guilty to a crime of violence. The dispute on appeal thus centers on the probative force of the criminal information, which states:
On or about August 9, 1986, within the District of Columbia, [appellant], by force and violence, against resistance and by putting in fear, stole and took from the person and from the immediate actual possession of Caroline Kaplan, property of value belonging to Caroline Kaplan, consisting of money.
Appellant’s App. 109.
While no one doubts that under the D.C.Code the “force and violence” language could have been satisfied by a minimal level of force that would not qualify the offense as a crime of violence, the parties disagree over the significance of the “against resistance and by putting in fear” language. Appellant contends that the disputed language represents mere surplusage — an articulation of the govern[1091]*1091ment’s “theory of the case,” Appellant’s Opening Br. 35 — and imposed no limit on the offense to which appellant could have pled guilty. Specifically, appellant argues that if he had allocated only to nonviolent robbery by snatching while insisting he never put the victim in fear or overcame her resistance, the court could have accepted the plea without requiring the government to produce a new information. The government fails to dispute this claim in its brief. Instead, it argues that we have no reason to believe anything of this sort actually happened.
The government is right that the record nowhere shows that appellant pled guilty only to robbery by snatching and not by violence. But by disputing only the likelihood — and not the possibility — that appellant pled guilty to a non-qualifying offense, the government essentially concedes the critical point. That is, since the government’s brief nowhere disputes the possibility that appellant pled to a non-qualifying offense, it necessarily admits that “the record of the previous conviction is ambiguous” within the meaning of Mathis, 963 F.2d at 410. After all, under Shepard the question is not what appellant probably pled to, but what he necessarily pled to.
That being the case, we cannot treat appellant’s conviction as one for a crime of violence. Perhaps the transcript of the plea colloquy would have shed light on the matter. Had appellant admitted to robbery at gunpoint, the trial court would have been precluded from accepting his plea based on a finding that appellant committed the robbery merely by stealth, and thus the element of violent compulsion would have been necessary to his conviction. On the other hand, had appellant insisted that he merely snatched the property, it would have been clear that the finding of violence was unnecessary to his guilty plea. But the government had the burden of proving that the elements of a crime of violence were necessary components of his conviction, and it failed to produce the transcript. “It is the responsibility of the government to produce such documents as are necessary to establish that a prior offense can be properly designated a ‘crime of violence.’ ” United States v. Richardson, 161 F.3d 728, 738 (D.C.Cir.1998) (internal quotation marks and brackets omitted).
At oral argument, the government initially suggested that had appellant attempted to plead only to snatching, the Superior Court would have had to reject the plea. According to this reasoning, the information’s “against resistance and by putting in fear” language limits the type of “force and violence” charged, so that if appellant had denied that he put the victim in fear or overcame her resistance, the Superior Court would have been unable to accept the plea. On this interpretation, the information would amount to a “generically limited charging document.” Shepard, 544 U.S. at 21, 125 S.Ct. 1254. That is, it would be limited to the “generic” (i.e., qualifying) version of robbery and could not support a plea to the nonqualifying version.
But this argument appears nowhere in the government’s brief. In fact, the government’s brief seems to take the contrary position: it admits that appellant could have pled guilty to robbery by snatching, maintaining only that it is extremely unlikely that he in fact did so. The government’s brief merely says that appellant “speculates that he might have pled guilty to something different than indicated in the language of the charging document, but offers no proof, and offered none at sentencing, that anything of that nature occurred,” Appellee’s Br. 37, and that there is “no support in the record” for the possibility that he did so, id. at 38. To be sure, at oral argument government counsel [1092]*1092initially claimed that the district court would have been unable to accept a plea to nonviolent robbery. Oral Arg. at 18:47. Later, however, he admitted that this argument is absent from the government’s brief, seeming to retreat from this contention and maintaining only that a preponderance of the evidence demonstrated that appellant pled guilty to violent robbery. Id. at 19:50.
Indeed, without wading into a question of District of Columbia criminal procedure never briefed or argued, we note that the government may have been right to effectively concede that appellant was able to enter such a plea. This is not a case where the information included no language that could refer to nonviolent conduct-it charged appellant with robbery by “force and violence,” which under the D.C.Code includes minimal force that does not qualify an offense as a crime of violence. The information may thus permit the defendant to plead guilty to nonviolent robbery, so it may not be “generically limited.” Shepard, 544 U.S. at 21, 125 S.Ct. 1254. Given this uncertainty, the absence of briefing on the issue, and the government’s failure to contest the point, we have no occasion to rule on this hypothetical argument.
The dissent reaches the contrary conclusion only by assuming that the charging document is in fact “generically limited.” Of course if it were generically limited, we would look no further. But whether it was is exactly the question before us. And to answer that question, we must decide not whether appellant in fact pled guilty to nonviolent robbery, but whether he could have under the information. This is not a question of fact foreclosed by Shepard, but rather a question of law. Under Shepard, a charging document is only “generically limited” if it in fact limits the type of offense to which a defendant can plead guilty. Contrary to the dissent, the government nowhere argues that this charging document imposes such a limit, as counsel conceded at oral argument. Oral Arg. at 19:50.
Nothing in United States v. Hill, 131 F.3d at 1064-65, on which the government relies, requires a different result. There we held that when a defendant pleads guilty to a lesser included offense of a robbery indictment bearing essentially the same language as that present here, the court must look beyond the indictment to other documents indicating what offense the defendant pled to. Id. Although remarking in passing that the indictment’s “against resistance and by putting in fear” language “suggest[ed]” that the robbery was committed violently, we had no occasion to rule definitively on the matter because the defendant pled guilty to a lesser included offense, not to the indictment. Id. at 1063.
To the extent Hill could be read to authorize a factual inquiry into what the defendant probably admitted, Shepard has superseded it. Hill authorizes examination of presentence reports adopted by the court and other judicial findings to determine the basis of a defendant’s plea. 131 F.3d at 1065. Though not establishing what the defendant necessarily admitted, such documents would amount to reliable evidence of the defendant’s conduct. But Shepard indicates that the sentencing court’s findings of fact, if not adopted by the defendant, must be disregarded. 544 U.S. at 16, 125 S.Ct. 1254 (holding that a court determining whether a prior crime counts as an ACCA predicate “is generally limited to examining the statutory definition, charging document, -written plea agreement, transcript of plea colloquy, and any explicit factual finding by the trial judge to which the defendant assented” (emphasis added)). Indeed, the very documents at issue in Shepard were free from [1093]*1093inconsistencies, and the Court, though never questioning the government’s claim that those documents showed that the plea in that case could “only plausibly have rested” on qualifying conduct, nonetheless forbade reference to those apparently reliable documents. 544 U.S. at 21-22, 125 S.Ct. 1254. Of course if the question were what the defendant probably admitted to, we would have no reason to disregard any evidence that could reliably shed light on that issue. The Court’s rejection of otherwise reliable evidence demonstrates what Shepard explicitly says: regardless of what Hill may have suggested, we look to what the defendant “necessarily admitted,” not just probably admitted. 544 U.S. at 26, 125 S.Ct. 1254.
Because the government failed to demonstrate on this record that appellant pled guilty to a violent robbery, the district court erred in ruling that appellant’s robbery conviction qualified as a crime of violence within the meaning of section 4B1.2. We vacate the sentence and remand for resentencing in light of Kimbrough and the government’s failure to demonstrate that appellant has two prior convictions for crimes of violence.
So ordered.