Defendant-appellant Ricardo Montañez-Quiñones seeks to set aside his 109-month sentence for possession of child pornography. In support, he both reproves the government for allegedly violating the plea agreement through its overzealous advocacy at sentencing and reproves the district court for enhancing his offense level through an allegedly erroneous finding that he knowingly distributed child pornography. Concluding, as we do, that neither claim of error withstands scrutiny, we affirm the challenged sentence.
I. BACKGROUND
We briefly rehearse the facts and travel of the case. Because this appeal follows a guilty plea, we draw our account from the plea agreement, the undisputed portions of the pre-sentence investigation report (PSI Report), and the transcripts of the change of plea and sentencing hearings. See *62United States v. Coleman, 884 F.3d 67, 69 (1st Cir.2018).
On September 20, 2015, as part of an investigation of pornography sharing on Ares (a peer-to-peer file-sharing network), a computer forensic laboratory associated with the Department of Homeland Security (DHS) successfully downloaded a seven-minute video that depicted a sexual encounter between a young girl (approximately eight to ten years of age) and an adult man. DHS agents traced the file to the residence of the defendant in Gurabo, Puerto Rico, and executed a search warrant for that address. The agents seized two devices: a laptop computer and a desktop computer.
The seized computers collectively housed 1,072 child sex abuse images. Those images showed boys and girls between four and fourteen years of age performing oral sex on adult men and being vaginally and anally penetrated by adult men. The agents' analysis also revealed an additional 3,613 child sex abuse files, which had either been downloaded and erased or were incomplete downloads, 89 child sex abuse files being shared on Ares, and at least 48 search terms related to child sex abuse.
In due course, a federal grand jury sitting in the District of Puerto Rico handed up an indictment charging the defendant with two counts of transportation of child pornography and one count of possession of child pornography (including images of prepubescent minors engaged in sexually explicit conduct). 18 U.S.C. § 2252A(a)(5)(B), (b)(2). Although the defendant originally maintained his innocence, he eventually executed a non-binding plea agreement, see Fed. R. Crim. P. 11(c)(1)(B), and entered a guilty plea to the charge of possession of child pornography. In exchange for the defendant's plea, the government agreed to dismiss the remaining two counts.
In the plea agreement (the Agreement), the parties agreed to a total offense level of 28, which included a two-level enhancement for distribution, see USSG § 2G2.2(b)(3)(F), and a three-level enhancement premised on a stipulation that the offense of conviction involved between 150 and 300 offending images, see USSG § 2G2.2(b)(7)(B). These stipulations were not intended to bind the sentencing court, see Fed. R. Crim. P. 11(c)(1)(B), and the Agreement contained no stipulation as to the defendant's criminal history category (CHC). The parties nonetheless agreed that, with a CHC of I, the guideline sentencing range would be 78-97 months; that the defendant could argue for a sentence at the low end of that hypothetical range; and that the government could argue for a sentence up to 87 months (the mid-point of the hypothetical range).1
The probation officer offered a slightly different assessment. The PSI Report calculated the defendant's total offense level at 30 based on a finding that the defendant possessed 600 or more offending images. With a CHC of I, the applicable guideline sentencing range would be 97-121 months. In his objections to the PSI Report, the defendant took issue with its inclusion of the two-level enhancement for knowing distribution. Although the same enhancement had been contemplated by the Agreement, the defendant argued that there was a critical distinction: since executing the Agreement, USSG § 2G2.2(b)(3)(F) had been amended to include a mens rea requirement. See *63USSG App. C, Amend. 801 (effective Nov. 1, 2016). The defendant argued that there was too little evidence to satisfy this new requirement. Specifically, he asserted that in order to prove knowing distribution, the government was obliged to introduce "evidence concerning the operation of the specific file sharing program used in the present case" and that it had failed to do so.
The district court was not persuaded that so precise an evidentiary showing was necessary to ground the enhancement. It overruled the defendant's objection based on its determination that "the evidence on record showed that defendant knew of the file-sharing properties of the 'Ares' program." In this regard, the court noted that the defendant was a "sophisticated and long-time computer user." This background, which included the defendant's degrees in computer science and computer networks and his statements that he was skilled in computers and would like to pursue an advanced degree in computer networks, was sufficient to infer the requisite knowledge. To cinch matters, the defendant had stored a portion of his downloaded child sex abuse files to a "shared" folder, indicating that he had curated "the particular contraband that he wanted to exchange through the 'Ares' file-sharing program."
After the court upheld the propriety of the knowing distribution enhancement, the disposition hearing proceeded. In accordance with the Agreement, the defendant argued for a sentence of 78 months (the low end of the hypothetical guideline range stipulated to by the parties). The government argued for a sentence of 87 months (the mid-point of the hypothetical range). In support of his argument, the defendant emphasized his difficult childhood and a history of abuse. The government countered that the defendant's conduct had helped to support an industry that "feeds on the sexual abuse and torture of children."
When all was said and done, the district court refused to accept the parties' stipulated guideline range. Instead, it embraced the guideline calculations contained in the PSI Report, which included a higher offense level that added five levels for possession of 600 or more offending images. Using a total offense level of 30 and a CHC of I, the court adopted a guideline sentencing range of 97-121 months. It proceeded to sentence the defendant to a mid-range 109-month term of immurement. This timely appeal ensued.
II. ANALYSIS
In this venue, the defendant attacks his sentence on two fronts. First, he contends that the government breached the terms of the Agreement by failing to advocate for the bargained-for sentence. Second, he contends that the district court's finding that he knowingly distributed child pornography was in error. We examine each contention in turn.
A. Alleged Breach of Plea Agreement.
The defendant begins by asseverating that statements made by the prosecutor during the disposition hearing, along with statements that the government failed to make, comprised a breach of the Agreement. This asseveration breaks new ground, as the defendant failed to mount this claim of error below. Consequently, our review is only for plain error - "a formidable standard of appellate review." United States v. Saxena, 229 F.3d 1, 5 (1st Cir.2000) ; see United States v. Almonte-Nuñez, 771 F.3d 84, 89 (1st Cir.2014) (citing Puckett v. United States, 556 U.S. 129, 143, 129 S.Ct. 1423, 173 L.Ed.2d 266 (2009) ). Under this standard, an appellant bears the burden of showing "(1) that an error occurred (2) which was clear or obvious *64and 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. Duarte, 246 F.3d 56, 60 (1st Cir.2001). Within this rubric, an appellant's substantial rights are deemed to be affected only when an error "likely affected the outcome of the proceedings." Almonte-Nuñez, 771 F.3d at 89.
It cannot be gainsaid that "[a] plea agreement is a binding promise by the government and is an inducement for the guilty plea." United States v. Gonczy, 357 F.3d 50, 53 (1st Cir.2004) (citing Santobello v. New York, 404 U.S. 257, 262, 92 S.Ct. 495, 30 L.Ed.2d 427 (1971) ). It follows that "a failure to support that promise is a breach of the plea agreement, whether done deliberately or not." Id. Because a defendant waives a panoply of constitutional rights by entering into a plea agreement, we hold the government to "the most meticulous standards of both promise and performance." Correale v. United States, 479 F.2d 944, 947 (1st Cir.1973). Simply providing "lip service" to these solemn obligations will not suffice. Saxena, 229 F.3d at 6.
Before us, the defendant asserts that the government violated the Agreement because it did not assiduously advocate for the bargained-for sentence and made a bad situation worse by misrepresenting the number of offending images stipulated in the Agreement. Some further facts are needed to put the assertion into perspective.
The government had agreed to recommend an incarceration sentence of no more than 87 months. At the disposition hearing the prosecutor stated, consistent with this agreement, on no fewer than five occasions that the government was recommending a sentence of 87 months. The defendant views these repeated recommendations as hollow: he points out that the prosecutor did not mention the total offense level of 28 referenced in the Agreement but, rather, stated (incorrectly) that the parties had stipulated to 300 to 600 offending images. Furthermore, the defendant claims that the prosecutor "excoriated [him] and condemned his conduct in the strongest terms," thereby nullifying whatever "lip service" that the prosecutor might have given to the bargained-for sentencing recommendation.
We start our consideration of the defendant's argument with first principles: "[n]o magic formula exists for a prosecutor to comply with the agreed-upon sentence recommendation." Gonczy, 357 F.3d at 54. Having repeatedly stated the government's sentencing recommendation of 87 months to the court, the prosecutor was not required to discuss any specific aspects of the government's thinking. In assessing whether the government breached its agreement to argue for the bargained-for sentence, we look instead to whether its "overall conduct" was "reasonably consistent with making such a recommendation, rather than the reverse." Id. (quoting United States v. Canada, 960 F.2d 263, 268 (1st Cir.1992) ); see Almonte-Nuñez, 771 F.3d at 91 ("We consider the totality of the circumstances in determining whether a prosecutor engaged in impermissible tactics.").
To be sure, the defendant perceives an inconsistency between the prosecutor's limited discussion of the government's sentencing recommendation and the strong language that the prosecutor used to describe the nature of the defendant's crime. We acknowledge, of course, that "it is possible for a prosecutor to undercut a plea agreement while paying lip service to its covenants." Almonte-Nuñez, 771 F.3d at 90-91. For example, we have found (albeit under a more sympathetic standard of review)
*65that such a breach occurred when the government never affirmatively recommended the agreed-upon sentence, see Canada, 960 F.2d at 268 ; when the government effectively argued against a sentencing reduction in contravention of the plea agreement, see United States v. Clark, 55 F.3d 9, 12-13 (1st Cir.1995) ; and when the government's zealous advocacy belied its agreement to recommend the low end of the applicable guideline sentencing range, see Gonczy, 357 F.3d at 54. Those cases, though, are at a far remove from the case at hand.
In this instance, the prosecutor repeatedly stated the government's recommendation of 87 months in accordance with the Agreement. See Saxena, 229 F.3d at 7 (finding no breach where prosecutor "resolutely stood by the bottom-line recommendation that the government had committed to make"); United States v. Irizarry-Rosario, 903 F.3d 151, 155 (1st Cir.2018) (finding no breach where explanation of sentencing recommendation was "interspersed with reaffirmations of the ... sentencing recommendation"). While the prosecutor's statements to this effect were simple and straightforward, a prosecutor is not obliged to present an agreed recommendation either with ruffles and flourishes or "with any particular degree of enthusiasm." Canada, 960 F.2d at 270. Nor is the defendant entitled "to have the government sugarcoat the facts." Almonte-Nuñez, 771 F.3d at 91.
The defendant's attempt to find a breach of the plea agreement in the prosecutor's unflattering narrative about the heinous nature of the defendant's crime is unpersuasive. This argument overlooks the salient fact that, under the Agreement, the government had a right to advocate for a sentence higher than the sentence that the defendant was seeking. Thus, the prosecutor had a right (indeed, a duty) to explain to the court why the higher sentence that it was urging was more appropriate. Almonte-Nuñez illustrates this point. There, we held that where a plea agreement entitled the prosecutor to argue for the high end of a guideline range while the defendant argued for the low end of that range, the prosecutor "was within fair territory in emphasizing facts that made a sentence at the low end of that [range] inappropriate." Id.
So it is here. The prosecutor had every right to highlight the serious nature of the offense and its impact on society in order to advocate for a sentence above the sentence requested by the defendant, as well as to demonstrate the unsuitable nature of the defendant's request. To this end, the prosecutor told the court that the conduct underlying the conviction was such as to "feed[ ] a terrible industry" supported by "the sexual abuse and torture of children," and that "the defendant chose to pursue his own sexual gratification with flagrant disregard for the welfare of thousands of minor children." Such language, though harsh, coheres both with the government's decision to charge the defendant with this serious crime and with its reservation of the right to argue for an 87-month sentence. We hold, therefore, that the prosecutor's statements at sentencing did not contradict any terms of the Agreement, nor did they " 'gratuitously offer[ ] added detail garbed in implicit advocacy' that might have led the district court to rethink the government's recommendation." Irizarry-Rosario, 903 F.3d at 155 (quoting United States v. Miranda-Martinez, 790 F.3d 270, 275 (1st Cir.2015) ). When the parties agree that a defendant may argue for a particular sentence while the government may argue for a somewhat stiffer sentence, the government is not constrained to pull its punches when arguing for the stiffer sentence.
*66The defendant has a fallback position. He says that the government breached the Agreement when it "advocated for a higher number of images than stipulated in the plea agreement." The government concedes that the prosecutor misstated the number of images stipulated in the Agreement but maintains that this was a slip of the tongue. Everything in the record points toward a finding of inadvertence. At the disposition hearing, there was no contemporaneous objection and, indeed, none of the parties appear to have noticed the misstatement when it was made. The prosecutor proceeded to recommend a sentence of 87 months - a recommendation derived from a hypothetical guideline sentencing range determined in accordance with the number of images stipulated in the Agreement. The bottom line, then, is that "[t]his is not a record in which the misstep conveyed a message that the ultimate recommendation was insincere." United States v. Oppenheimer-Torres, 806 F.3d 1, 4 (1st Cir.2015).
Nor does it appear that the misstatement in any way affected the outcome of the proceedings. The record is bereft of any basis from which we might reasonably infer that the district court was misled as to the number of images stipulated to by the parties. That number was correctly described both in the Agreement and in the PSI Report - and those documents were before the district court at sentencing. And in any event, the court itself had independently determined that the offense conduct involved 600 or more images. Given the totality of the circumstances, we find no prejudice attendant to the prosecutor's lapses linguae and, thus, no merit in the defendant's claim that this misstatement heralded a breach of the plea agreement.
B. Knowing Distribution.
This brings us to the defendant's plaint that the district court erred when it included a two-level enhancement for knowing distribution in its calculation of the guideline sentencing range. This plaint has a narrow focus: while the defendant does not dispute that distribution occurred, he alleges that the government failed to provide evidence that he knew of the file-sharing properties of the program.
It is elementary that "the government bears the burden of proving sentence-enhancing factors by a preponderance of the evidence." United States v. Nuñez, 852 F.3d 141, 144 (1st Cir.2017). We apply a clear error standard of review to the sentencing court's factfinding - a standard that extends to any findings based on inferences drawn from discerned facts. See id. This is a demanding standard, satisfied only if, "upon whole-record-review, an inquiring court 'form[s] a strong, unyielding belief that a mistake has been made.' " United States v. Cintrón-Echautegui, 604 F.3d 1, 6 (1st Cir.2010) (alteration in original) (quoting Cumpiano v. Banco Santander P.R., 902 F.2d 148, 152 (1st Cir. 1990) ).2
*67Section 2G2.2(b)(3)(F) of the sentencing guidelines was amended, effective as of November 2016, to limit the two-level enhancement to possessors of child pornography who "knowingly engaged in distribution." In incorporating a mens rea requirement, the Sentencing Commission resolved a circuit split and "generally adopt[ed] the approach of the Second, Fourth, and Seventh Circuits." USSG Supp. to App. C, Amend. 801 at 145 (2016); see United States v. Baldwin, 743 F.3d 357, 361 (2d Cir.2014) (per curiam); United States v. Robinson, 714 F.3d 466, 469-70 (7th Cir.2013) ; United States v. Layton, 564 F.3d 330, 335 (4th Cir.2009). Even as amended, though, the enhancement does not require proof that the defendant intended to distribute child pornography - "as long as he had knowledge that by using a peer-to-peer file-sharing program, his child pornography was made accessible to others." United States v. Cates, 897 F.3d 349, 359 (1st Cir.2018) (emphasis in original). In all events, "the government need not prove knowledge by direct evidence, but may prove knowledge by circumstantial evidence." Id. Viewed against this backdrop, the defendant's argument that the government failed to provide "some evidence" that he affirmatively knew of the file-sharing properties of the application "confuses a lack of direct evidence with a lack of evidence." Id.
Cates is instructive. There, we determined that the district court drew a reasonable inference that the defendant knew of the file-sharing properties of a peer-to-peer network when it relied on evidence that the defendant was "relatively sophisticated in computer matters" and had demonstrated familiarity with the program's file-sharing properties. Id. at 359-60. The findings in Cates, albeit based on a stronger evidentiary predicate, are on the same order as those of the court below, which drew an equally reasonable inference of knowledge based on uncontradicted evidence that the defendant was a "sophisticated and long-time computer user" who had selected from thousands of downloaded files a limited number to share through the file-sharing program.
On this record, the sentencing court was entitled to draw the plausible inferences that led to a finding of knowledge. Inferences based on circumstantial evidence "need not be compelled but, rather, need only be plausible." See Nuñez, 852 F.3d at 146. The court below reasonably could infer that the defendant was a sophisticated computer user based on evidence that he had acquired two degrees in computer science and computer networks. Similarly, the court reasonably could infer that the defendant selected a limited number of child sex abuse files to be shared on Ares. That conclusion was based on evidence that the defendant had downloaded thousands of child sex abuse files but that he shared only 74 and 15 child sex abuse files, respectively, on each of his two computers.
Surely, other plausible inferences could be drawn from this evidence. But that is not the test: the decisive consideration is that, on the record before it, the court below plausibly could infer that the disparity between files downloaded and files shared was a result of the defendant's desire to share only some files. And it is apodictic that "[w]here the raw facts are susceptible to competing inferences," a district court's "choice between those inferences cannot be clearly erroneous." United States v. McCormick, 773 F.3d 357, 359 (1st Cir.2014).
The defendant challenges the sufficiency of these findings. He submits that the government was required to furnish evidence concerning the operation of the particular file-sharing program at issue. We *68previously have called such an argument a "red herring," holding that the sentencing court drew a reasonable inference of knowledge without the benefit of evidence that files downloaded through the program were automatically accessible for others to download. Id. The argument has not changed its color in the short time that has elapsed since Cates was decided.
Let us be perfectly clear. We do not hold that such evidence is irrelevant to the issue of knowing distribution. Simply using a program (like Ares) that automatically steers downloaded files into a shared folder may well be insufficient, standing alone, to support an inference of knowledge, particularly if the government has not provided evidence that the defendant knew of this mechanism or otherwise possessed the technological proficiency to understand that it was in place. See, e.g., United States v. Carroll, 886 F.3d 1347, 1354 & n.4 (11th Cir.2018) (holding that government was required to "put forth evidence that [defendant] had some advanced technological proficiency" to support finding of knowing distribution by means of file-sharing program that did not notify users of automatic sharing); Robinson, 714 F.3d at 470 (concluding that computer novice who "had never seen a file-sharing program before might not realize" that "shared files are accessible automatically to other persons online"). Conversely, concerns about automatic file-sharing have been allayed where - as in Cates- courts have found that the defendant possessed "advanced computer knowledge" or used the program in a manner that indicated an understanding of how the program worked. See United States v. Alpizar, 743 Fed.Appx. 928, 935 (11th Cir.2018) ; United States v. Nordin, 701 F. App'x 545, 546 (8th Cir.2017) (per curiam).
This case is of the latter stripe. The court below reasonably inferred knowledge both from its well-supported finding that the defendant was "a sophisticated and long-time computer user" and from the defendant's storage of select files in his shared folder. No more was exigible to render the court's findings adequate as a foundation for a reasonable inference of knowledge, regardless of whether downloaded files were automatically available for distribution to others. Accordingly, we discern no clear error in the court's imposition of a two-level enhancement for knowing distribution of child pornography.
III. CONCLUSION
We need go no further. For the reasons elucidated above, the defendant's sentence is
Affirmed .
Although I concur with the majority's conclusion regarding the government's alleged violation of the plea agreement, I respectfully disagree with its conclusion regarding the sentencing enhancement that was applied to increase appellant's sentence. That enhancement is not supported by the record before the district court. Therefore, I would hold that the court clearly erred in determining that the government proved appellant's knowledge of distribution by a preponderance of the evidence. Before explaining my reasoning, I must provide some context for my assessment of the enhancement and augment the majority's description of the factual record.
I.
A. Peer-to-Peer File-Sharing Programs
In recent years, " 'peer-to-peer' ... file-sharing via the Internet has resulted in significant changes in the manner in which *69[child pornography] offenses are committed." U.S. Sentencing Comm'n, Report to the Congress: Federal Child Pornography Offenses (Dec. 2012), at 5. Peer-to-peer file-sharing networks " 'allow[ ] users to download files from the computers of other users. Unlike other means of acquiring files over the Internet, such as in a chat room or using e-mail[,] ... no personalized contact is required between the provider and receiver.' " United States v. R.V., 157 F.Supp.3d 207, 235 (E.D.N.Y. 2016) (quoting Maggie Meuthing, Inactive Distribution: How the Federal Sentencing Guidelines for Distribution of Child Pornography Fail to Effectively Account for Peer-to-Peer Networks, 73 Ohio St. L.J. 1485, 1488 (2012) ). In addition to allowing a user to download files, file-sharing programs also make files on a user's computer accessible for download by other users. Notably,
[a] crucial aspect of peer-to-peer file-sharing is that the default setting for these networks is that downloaded files are placed in the user's "shared" folder, which allows others in the network to access the files. A user must affirmatively change his network setting to disable this sharing feature.
Id. (quoting Audrey Rogers, From Peer-to-Peer Networks to Cloud Computing: How Technology is Redefining Child Pornography Laws, 87 St. John's L. Rev. 1013, 1031 (2013) ).
When first downloaded, Ares, the file-sharing program used by appellant, "sets up a shared folder on the computer where, by default, it automatically places all subsequent [Ares] downloads. Once a file is [automatically] placed in the shared folder, it is immediately available for further dissemination." United States v. Carroll, 886 F.3d 1347, 1350 (11th Cir.2018). That is, "[u]nless an Ares user changes the default settings or deliberately moves files out of the shared folder, downloaded files [from Ares] will remain freely accessible to anyone else on the Ares network." Id.
B. The "Knowing" Distribution Guideline Enhancement
In general, due to the pervasive use of file-sharing programs to access child pornography, the sentencing guideline enhancements for the non-commercial distribution of child pornography may be applied to the majority of non-production child pornography offenders. See U.S. Sentencing Comm'n, Report to the Congress, at 149-50, 154-55. Until the end of 2016, the sentencing guidelines provided for a two-level enhancement in child pornography cases "[i]f the offense involved ... [d]istribution." Compare U.S. Sentencing Guidelines Manual § 2G2.2(b)(3)(F)with U.S. Sentencing Guidelines Manual § 2G2.2(b)(3)(F) (as amended Nov. 2016).3 Courts generally agreed that a user of a peer-to-peer file-sharing network need not take affirmative steps to share files with other users in order to have "distributed" child pornography. See, e.g., United States v. Chiaradio, 684 F.3d 265, 282 (1st Cir.2012) (accepting the analogy of a peer-to-peer file-sharing program user to a self-serve gas station owner in holding that a person may "passive[ly]" distribute files by making them available for download by other users).
However, the circuits were split on whether the enhancement required some mens rea despite the absence of language to that effect in the guideline. Several circuits *70held that the enhancement required evidence that a defendant knew about the file-sharing properties of the program he was using to obtain child pornography. See, e.g., United States v. Baldwin, 743 F.3d 357, 361 (2d Cir.2014) ; United States v. Robinson, 714 F.3d 466, 468 (7th Cir.2013) ; United States v. Layton, 564 F.3d 330, 335 (4th Cir.2009). Other circuits held that there was no knowledge requirement, or that knowledge could be presumed from a defendant's use of a file-sharing program. See, e.g., United States v. Abbring, 788 F.3d 565, 567 (6th Cir.2015) ; United States v. Creel, 783 F.3d 1357, 1360 (11th Cir.2015) ; United States v. Baker, 742 F.3d 618, 621 (5th Cir.2014) ; United States v. Ray, 704 F.3d 1307, 1311-12 (10th Cir.2013) ; United States v. Dodd, 598 F.3d 449, 451-52 (8th Cir.2010).
In late 2016, the guideline enhancement was amended to specify that it applied only where a defendant "knowingly engaged in distribution." U.S.S.G. § 2G2.2(b)(3)(F) (emphasis added). In amending the guideline, the Sentencing Commission noted that some file-sharing programs "employ a default file-sharing setting" and that a user has to " 'opt out' of automatically sharing files by changing the default setting to limit which, if any, files are available for sharing." U.S.S.G. App. C, Amend. 801 (eff. Nov. 1, 2016). The Commission acknowledged the existing uncertainty regarding mens rea and stated that it was "generally adopt[ing] the approach of the Second, Fourth, and Seventh Circuits," which all required evidence of a defendant's knowledge of a program's file-sharing properties. Id.
In codifying this approach, the Commission rejected both the approach of those circuits that did not require evidence of knowledge and the approach of those circuits that had held that knowledge of a program's file-sharing properties "may be inferred from the fact that a file-sharing program was used, absent 'concrete evidence' of ignorance," because "the whole point of a file-sharing program is to share." Id. (quoting Dodd, 598 F.3d at 452, and Abbring, 788 F.3d at 567 ). After the amendment, then, application of the enhancement requires specific evidence of a defendant's "knowledge that by using a peer-to-peer file-sharing program, his child pornography was made accessible to others." United States v. Cates, 897 F.3d 349, 359 (1st Cir.2018). The simple fact that a defendant used a file-sharing program does not constitute evidence of knowledge.4 In other words, it is not enough for the government to assert that a defendant "was using a peer-to-peer file sharing program and 'that is what it is.' " Carroll, 886 F.3d at 1353.
The question then becomes what constitutes evidence of knowledge to support the enhancement. In Cates, we described a substantial amount of evidence of knowledge. Specifically, there was evidence that the defendant (1) had used a file-sharing program to download child pornography for three years; (2) had created a "specialized configuration" "by which files downloaded from [the file-sharing program] would bypass his master hard drive and be saved automatically to the 'sharing folder' housed on a subservient drive"; and (3) had, in his interview with authorities, "demonstrated considerable familiarity with [the program]'s file-sharing properties" and acknowledged that he could turn off the program's default setting of automatic sharing. Cates, 897 F.3d at 359.
*71The Eleventh Circuit's recent treatment of the amended guideline in relation to the Ares program is also instructive. In Carroll, the court reversed a distribution conviction "because the government failed to put forth any evidence that [the defendant] knew downloaded files were automatically placed into a shared folder accessible to the Ares peer-to-peer network." Carroll, 886 F.3d at 1349 (emphasis added). The only proffered evidence of knowledge in Carroll was the defendant's use of the Ares program and the presence of files automatically being placed into, and shared from, the Ares-created folder. Id. at 1353. The court considered this to be no evidence at all of the defendant's knowledge. In a subsequent case affirming an application of the knowing distribution enhancement, the Eleventh Circuit distinguished Carroll by noting, inter alia, that the defendant in the present case "admitted to knowing how file sharing programs like A[res] worked" and had continued to share child pornography after being told by the FBI "how A[res] file sharing worked." United States v. Alpizar, 743 Fed.Appx. 928, 936 (11th Cir.2018).
In sum, Cates, Carroll, and Alpizar demonstrate the type of evidence needed to apply the "knowing" distribution enhancement in a case involving a program that automatically shares downloaded files -- that is, some specific evidence that the defendant used the program in a manner demonstrating his awareness of the program's file-sharing properties. Without this evidence, a court risks applying the enhancement based solely on a defendant's use of a file-sharing program, which is the approach explicitly rejected by the Sentencing Commission.
II.
Against this backdrop, I turn to the record before the district court. The government's undisputed version of the facts, which was incorporated into the plea agreement, provides the only description in evidence of appellant's collection of child pornography on his two computers. It states:
[D]efendant's Sony VAIO laptop ... was found to contain 26 child sex abuse images. Additionally, it contained evidence of: a) 2,578 child sex abuse files having been downloaded and then erased; b) 71 incomplete downloads of child sex abuse files; c) 74 child sex abuse files being shared on "Ares[";] and, d) 23 child sex abuse-related search terms having been entered by the defendant.
[D]efendant's Compaq desktop computer ... was found to contain 1,046 child sex abuse images. Additionally, it contained evidence of: a) 802 child sex abuse files having been downloaded and then erased; b) 162 incomplete downloads of child sex abuse files; c) 15 child sex abuse files being shared on "Ares[";] and, d) 48 child sex abuse-related search terms having been entered by the defendant.5
In addition to this description of the child pornography that appellant possessed, there is no dispute that (1) he searched for and downloaded child pornography; (2) he downloaded the Ares file-sharing program onto his two computers; (3) a certain number of child pornography files were "being shared on Ares," likely meaning that these files were in the Ares folders on each device; and (4) a smaller number of files in the case of his laptop, and a larger number *72of files in the case of his desktop, were housed elsewhere on the computers.6 Finally, although this aspect of the Ares program is not made explicit in the record, there is no dispute that the program, when first downloaded, "sets up a shared folder on the computer where, by default, it automatically places all subsequent downloads" from Ares and that files automatically placed in this folder are freely accessible to other users. Carroll, 886 F.3d at 1350.
On this record, the district court concluded, "[t]he selection on both devices of a specific number of child sex abuse files to be shared on the 'Ares' network out of the thousands downloaded by defendant [indicates] that he applied his computer knowledge to pick and choose the particular contraband that he wanted to exchange through the 'Ares' file-sharing program." Based on this finding, plus a finding that appellant is a "sophisticated and long-time computer user," the court concluded that "all indications are that [he] used a shared folder that he knew others could access in order to download child pornography files." Although the majority states that the court "was entitled to draw the plausible inferences that led to a finding of knowledge," appellant contends that there is no evidence to support the district court's inference that he "picked and chose" certain files to share through Ares, the inference essential to the district court's finding of knowledge. I agree.
III.
Given that a certain number of child pornography files were on appellant's computers but not in the Ares folders, the district court inferred that he intentionally placed certain files in the Ares sharing folders, or kept certain files in the folders while removing others. The court further inferred that he performed this allocation because he was aware of the Ares program's file-sharing properties. As the district court implicitly saw it, there is no reason to intentionally place or keep files in the sharing folders other than to share these files with other Ares users.
There is no evidence to support this inference of allocation, however, because there is no evidence about the origin of the child pornography files on appellant's computers. The district court's inference would be supported if there was any evidence that appellant moved files between the Ares folders and other locations on his computers. Yet for all we know, appellant acquired all the files outside the Ares folders from a source other than Ares. In that case, all the files in the Ares folders could have been automatically placed there when they were downloaded through Ares and appellant would not necessarily have moved any files into or out of the Ares folders.
To be sure, there is a plethora of evidence that could have illuminated the allocation issue: for example, computer forensic examinations can readily ascertain the origin of files or how long they have been on a computer. See Sergeant Josh Moulin, What Every Prosecutor Should Know About Peer-to-Peer Investigations, Child Sexual Exploitation Program Update Volume 5, Number 1, 2010, National District Attorneys Association, National Center for Prosecution of Child Abuse (describing the detailed evidence about a defendant's use *73of a file-sharing program, manipulation of default settings, and handling of files in general that can be ascertained through a computer forensic examination). If we knew that any of the files stored outside the Ares folders were downloaded through Ares, for example, this would be evidence that appellant intentionally removed certain files from the Ares folders. Similarly, if we knew that any of the files inside the Ares folders were not originally downloaded through Ares -- if these files were obtained through another source, for example, via the sharing of files on external drives or even through a different file-sharing program -- this would be evidence that he intentionally placed certain files into the Ares folders. This type of evidence, however, is completely absent from the record. Thus, the district court's foundational inference -- that appellant intentionally allocated files between the Ares folders and other locations on his computers -- is pure speculation. Simply put, there was no evidence before the district court that appellant "used the program in a manner that indicated an understanding of how the program worked," as my colleagues contend.
We must also remember that references in our sentencing enhancement decisions to "plausible inferences" cannot obscure the requirement that the government has to prove the applicability of a sentencing enhancement by a preponderance of the evidence. See United States v. Lacouture, 835 F.3d 187, 189-90 (1st Cir.2016). Since the inference of allocation is at the heart of the district court's finding that appellant had knowledge of the file-sharing properties of the Ares program, the absence of any evidence to support that inference is even more striking. Moreover, a traditional rationale for deference to a district court's findings -- its ability to weigh credibility -- has no relevance here. We are only evaluating the district court's logic, not any assessment of credibility. Cf. United States v. Brum, 948 F.2d 817, 819 (1st Cir.1991) ("We review the challenged findings of fact for clear error, mindful of the deference to which the sentencing court's superior opportunity to assess witness credibility is entitled.").
The contrast between this case and cases like Cates and Carroll is telling. In Cates, we highlighted the veritable mountain of specific evidence indicating that the defendant was aware of a program's file-sharing properties. See 897 F.3d at 359-60. In Carroll, the Eleventh Circuit rejected an application of the enhancement that was based solely on the fact that the defendant was using Ares. See 886 F.3d at 1353-54. Here, the record is devoid of the type of evidence we highlighted in Cates. And when we scrutinize the district court's reasoning, it is clear that the court, in applying the enhancement, essentially relied on the bare fact that appellant was using Ares.
Without the unsupported inference that appellant "picked and chose" files to place in the Ares folders for sharing, all that we are left with is the district court's finding that appellant possesses a level of general computer proficiency. I agree that a defendant's "advanced computer knowledge" may be relevant to the knowledge inquiry. However, I am not aware of any authority in our case law for the proposition that some level of general computer proficiency on a defendant's part is enough, on its own, to support a finding of knowledge for purposes of the enhancement. But see United States v. Ryan, 885 F.3d 449, 453 (7th Cir.2018) (affirming a knowing distribution conviction because "[t]he government ... presented evidence of [defendant]'s sophisticated understanding of computers and software"). Even the majority does not contend that a defendant's general computer knowledge, such as a *74degree in computer science, is sufficient to support the enhancement. Yet once the unsupported inference of allocation is removed from the equation, the evidence of appellant's general computer knowledge is all that remains.7
IV.
The district court applied the "knowing" distribution enhancement based on an inference of allocation that is not supported by the record. Without any evidence about the origin of the various files on appellant's computers, there is no evidence that he intentionally moved files into or out of the Ares folders. The court's inference of allocation was thus pure speculation. Once this unsupported inference is put aside, it is apparent that the district court essentially applied the enhancement because appellant was using a file-sharing program. That is precisely the approach rejected by the Sentencing Commission. My colleagues tacitly accept this discredited approach. I would hold that the district court clearly erred in applying the enhancement.
The parties agree that the government remained bound to this ceiling even if the district court - as happened here - determined that a more onerous guideline sentencing range applied.
The dissent suggests that deference to the district court's factual findings may be lessened here because we are assessing the district court's logic on a paper record, which invites no weighing of credibility. See post at 73. What the dissent calls "logic" is nothing more or less than the drawing of inferences from the facts of record and, thus, the dissent's suggestion lacks force. See RCI Ne. Servs. Div. v. Bos. Edison Co., 822 F.2d 199, 202 (1st Cir.1987) ("[F]indings of fact do not forfeit 'clearly erroneous' deference merely because they stem from a paper record."); see also Limone v. United States, 579 F.3d 79, 94 (1st Cir.2009) ("The application of clear-error review to findings drawn from a paper record has long been the practice in this circuit.").
If the offense involves distribution in exchange for any type of payment or distribution to a minor, the guidelines provide for a greater enhancement. See U.S.S.G. § 2G2.2(b)(3)(A)-(E).
To the extent knowledge can be established by evidence of recklessness, the district court did not rely on this theory, and, in any event, my analysis would not differ if the government was pressing a recklessness theory.
I assume that there is no meaningful distinction between the government's use of "images" and "files" in this case, considering that neither my colleagues, nor the district court, nor the parties suggest any such distinction. For consistency, I refer to the child pornography on appellant's computers as "files."
To the extent there is any ambiguity in the government's description of appellant's child pornography collection, I note it follows logically that the files "contain[ed]" on appellant's computers are different from the files "being shared on 'Ares.' " For example, appellant's laptop was "found to contain" 26 files, but there were 74 files being shared on Ares, demonstrating that the files being shared were not a subset of the files "contain[ed]" on his computers.
Even if general computer proficiency could theoretically support application of the enhancement on its own, appellant's level of computer proficiency would fall short. He completed a bachelor's degree in computer science and an associate's degree in computer networks a decade ago. He further indicated he is "skilled in computers" and "expressed interest in completing a Master's Degree in Computer Networks." However, he also expressed interest in pursuing formal training as a hairstylist, and his most recent job before his arrest was as a "receiving supervisor" at a "produce packing company ... earning approximately $500 weekly." In other words, there is little to no evidence that he possessed "advanced computer knowledge" or was especially proficient in current computer technology, let alone file-sharing programs such as Ares. There is also no record evidence that he had used Ares for a significant period of time and thus had an opportunity to develop familiarity with the program.