PAMELA HARRIS, Circuit Judge

ARGUED: Jorgelina E. Araneda, ARANEDA LAW FIRM, Raleigh, North Carolina, for Appellant. Barbara Dickerson Kocher, OFFICE OF THE UNITED STATES ATTORNEY, Raleigh, North Carolina, for Appellee. ON BRIEF: John Stuart Bruce, Acting United States Attorney, Jennifer P. May-Parker, First Assistant United States Attorney, OFFICE OF THE UNITED STATES ATTORNEY, Raleigh, North Carolina, for Appellee.

Before TRAXLER, FLOYD, and HARRIS, Circuit Judges.

Vacated and remanded by published opinion. Judge Harris wrote the opinion, in which Judge Traxler and Judge Floyd joined. PAMELA HARRIS, Circuit Judge:

A jury convicted Hemza Menade Lefsih, an Algerian native who entered the United States through the Diversity Immigrant Visa Program, of immigration fraud. During Lefsih's trial, the district court interjected numerous times, expressing skepticism of the Diversity Immigrant Visa Program and a negative impression of individuals who participate in the program. We find that this judicial intervention was improper and denied Lefsih the opportunity for a fair and impartial trial, and therefore vacate the judgment of conviction.

I.

A.

Hemza Menade Lefsih immigrated to the United States through the Diversity Immigrant Visa Program ("Diversity Program"). The Diversity Program awards permanent residence immigration visas—"diversity visas"—to individuals from countries with historically low immigration numbers, on the basis of a random lottery system. Lefsih, from Algeria, won a Diversity Program lottery and received a diversity visa. As allowed under the Diversity Program, he then sought United States citizenship, submitting an N-400 form—the application form for naturalization—five years after entering the country.

Questions 22 through 28 of the N-400 relate to an applicant's criminal history. Specifically, Question 22 inquires whether an applicant has "ever committed ... a crime or offense for which [he or she was] not arrested"; Question 24, whether an applicant has "ever been charged with committing ... a crime or offense"; and Question 25, whether an applicant has "ever been convicted of a crime or offense[.]" S.A. 375 (emphases in original). And between those questions is Question 23—the question at issue here—asking whether an applicant has "ever been arrested, cited, or detained by any law enforcement officer ... for any reason." Id. (emphasis in original).

Lefsih answered "no" to Question 23. Id. In fact, however, Lefsih had been "cited" by several law enforcement officers, receiving a total of 11 traffic citations while working as a cab driver in North Carolina. Lefsih later would testify that he understood Question 23 as referring only to serious criminal offenses that resulted in arrests or detentions, and not to traffic tickets, and so believed that he was answering the question truthfully. But Lefsih concedes that in actuality, his assertion that he never had been "cited" was false.

Because Lefsih failed to acknowledge his traffic tickets in answer to Question 23, the government charged Lefsih with two counts of making a false statement on a naturalization form, see 18 U.S.C. § 1015(a), and two counts of immigration fraud, see 18 U.S.C. § 1546(a). In order to obtain a conviction under either provision, the government was required to prove Lefsih's state of mind: that contrary to his account, Lefsih knowingly provided a false answer to Question 23. See 18 U.S.C. §§ 1015(a), 1546(a).

B.

Lefsih's two-day trial began on April 27, 2016. The government could present no direct evidence that Lefsih knew, at the time he filled out his N-400 form, that his answer to Question 23 was false. Instead, as is common in establishing a defendant's state of mind, the government relied on circumstantial evidence. See United States v. Santos , 553 U.S. 507, 521, 128 S.Ct. 2020, 170 L.Ed.2d 912 (2008) (government customarily proves knowledge with circumstantial evidence).

Through its first witness, Special Agent Tony Bell of the Immigration and Customs Enforcement division of Homeland Security Investigations, the government sought to establish that Lefsih was fully capable of correctly understanding Question 23. Bell, who had investigated and interviewed Lefsih prior to Lefsih's indictment, testified that Lefsih was a proficient English speaker. He also reviewed Lefsih's educational background, including work toward a master's degree in physics at a Paris school; excellent performance in classes at Wake Tech Community College; and high grades on English proficiency and placement tests.

In addition, Bell's testimony called into question Lefsih's motives in entering the country through the Diversity Program. Bell testified that in his experience, it was unusual that someone like Lefsih would apply only for a diversity visa through the Diversity Program lottery—with low odds of success—and not for a student visa. According to Bell, Lefsih explained this decision as turning on the "better class of entry" offered by a diversity visa. J.A. 114. Bell understood Lefsih to be referring to the fact that students are admitted only for the purpose of attending school and "tracked" while they are in the country, J.A. 115, whereas diversity-visa holders enter as legal permanent residents and without similar restrictions.

The government's next witness was Gary Freitas, a senior officer with the United States Citizenship and Immigration Services, who testified about the Diversity Program and the naturalization application process. Freitas began by explaining that the Diversity Program was established by Congress so that people from countries with historically low immigration rates would have an opportunity to live permanently in the United States. Upon hearing that, the district court asked Freitas a series of pointed questions about the Program:

District court: You're saying that Congress has set up a law that your agency enforces that invites people to come to America from places where they don't normally come to America?

Freitas: Yes.

District court: That's a shorthand way of saying it?

Freitas: Yes, it is.

District court: That's incredible. And the reason that they don't come to America is because they haven't tried to come to America? Is that it?

Freitas: Usually because of—they may not have family members here from those countries or employment opportunities.

....

District court: Okay. Do you think anybody in America knows about this, other than the Committee that sent it through Congress? Probably not.

Freitas: I didn't know it before I started—

District court: And it's your job. Don't you love Congress? I mean, unbelievable, unbelievable. I'm sitting here 32 years, first time I ever heard this.

J.A. 146–47, 149. The district court continued, now focused on the particular nations covered by the Diversity Program:

District court: And you're talking about the hundred countries that nobody could name if they had a list of 180 countries in the world?

Freitas: Yes.

District court: The bottom hundred.

Freitas: Yes.

District court: Like Mauritania.

Freitas: Exactly.

District court: Moldavia or something like that.

Freitas: Correct.

J.A. 147–48.

Finally, the district court turned its questioning to the individuals—like Lefsih—who participate in the Program:

District court: And Congress is aggressively trying to bring those people to America by creating a lottery where they have special treatment?

Freitas: The [sic] set up the lottery. It's—I'm not sure what the percentage overall of between total immigrants that immigrant [sic] into the United States per year.

District court: Aren't there quotas on people coming from countries that send a lot of people here, and you have to show you're a doctor, an engineer or a rocket scientist or someone who is going to contribute to the well-being of the United States of America and make it a better place to live because of your skill or personal characteristics?

Freitas: Yes, there are limits on those.

District court: But if you're in the bottom hundred countries in the world, just come on.

Freitas: Well, they can apply for the lottery. They get a chance.

District court: But they don't have to be a back surgeon or anything?

Freitas: No, they just need minimum qualifications equivalent to a high school education.

J.A. 148–49. Shortly thereafter, the district court returned to a discussion of the kind of people who enter the United States with diversity visas:

District court: So if you get lucky and win the lottery and get a card to come to America you can drag along your ten kids and four wives or what?

Freitas: Well, your spouse.

District court: Your spouse and your kids and your uncle and your brother?

Freitas: No, no, just immediate.

J.A. 152.

At no point did Lefsih object to any of the district court's questions or comments. At the close of the government's case-in-chief, Lefsih moved for a judgment of acquittal under Federal Rule of Criminal Procedure 29, arguing that the government had not proved beyond a reasonable doubt that he knew he was answering falsely when he replied "no" to Question 23. The court took the motion under advisement.

Lefsih was the only witness for the defense. And the core of Lefsih's testimony was that his incorrect answer to Question 23 was an honest mistake, rather than a knowing falsity. According to Lefsih, he did not understand what the word "cited" meant, or that "traffic tickets are actually included in the word cited." J.A. 229. Instead, because the word "cited" appears between "arrested" and "detained" in Question 23, Lefsih assumed that it carried a similar meaning: being taken into custody. Because he had not been detained as a result of his traffic violations, Lefsih did not understand that those infractions fell within the scope of Question 23. As Lefsih testified, "It never occurred to [me] that [a] traffic ticket could be [a] criminal offense." J.A. 226.

The defense sought to bolster Lefsih's account by introducing other immigration forms that expressly clarify whether traffic tickets are within the scope of questions regarding criminal history. In light of Question 23's failure to provide this clarification,the defense argued, the government could not meet its burden of proving that Lefsih's answer was a knowing and purposeful false statement rather than an inadvertent error.

After closing arguments, the district court instructed the jury. Reiterating a point made during its preliminary instructions at the start of trial, the judge told the jurors: "My job ... is to preside over the trial. I don't have any position about the outcome of it. I'm like the referee or the umpire. You're the actual judges of the facts." J.A. 272. Later in the instructions, the court added: "[I]f during the course of the trial I made any comments or asked any questions or made any rulings, you should not interpret from that that I have any position or opinion about the outcome of the case. I simply do not." J.A. 274. After approximately 30 minutes of deliberation, the jury returned a verdict of guilty on all counts.

After the verdict, the government dismissed the two false statement counts against Lefsih in order to avoid potential double jeopardy issues. Lefsih renewed his Rule 29 motion for a judgment of acquittal on the remaining immigration fraud counts. The district court denied that motion at sentencing, and sentenced Lefsih to time served. This timely appeal followed.

II.

We begin with Lefsih's first argument: that the district court improperly denied his Rule 29 motion for a judgment of acquittal, because the government's evidence was insufficient to show that he knowingly gave a false answer to Question 23. "A defendant challenging the sufficiency of the evidence to support his conviction bears a heavy burden." United States v. Beidler , 110 F.3d 1064, 1067 (4th Cir. 1997) (internal quotation marks omitted). In reviewing the evidence, we must draw "all reasonable inferences from the facts proven to those sought to be established" by the government. United States v. Tresvant , 677 F.2d 1018, 1021 (4th Cir. 1982). And Lefsih can prevail only if no "rational trier of fact could have found the essential elements of the crime beyond a reasonable doubt." United States v. Green , 599 F.3d 360, 367 (4th Cir. 2010). We agree with the district court that Lefsih cannot meet this "high bar." J.A. 335 ("Given this high bar and heavy burden, the Court elects not to disturb the jury's verdict.") (internal quotation marks omitted).

The government started its case with a concededly false statement: Lefsih's false answer to Question 23. It then presented ample evidence from which a rational jury could conclude that there was no language barrier or other impediment that would have prevented Lefsih from understanding the correct meaning of Question 23 and applying it properly to his numerous traffic citations. The only remaining question was whether the jury would credit Lefsih's testimony that he nevertheless made an honest mistake when he answered "no" on his N-400 form. And while a reasonable jury might have believed Lefsih's account of confusion and misunderstanding, there is nothing that would have compelled it to do so. Cf. United States v. Hester , 880 F.2d 799, 803 (4th Cir. 1989) (finding sufficient evidence where jury could have inferred knowledge from "sufficiently strong" circumstantial evidence). Such credibility determinations fall squarely within the purview of the jury, and are not to be reassessed on appeal. See Green , 599 F.3d at 367 ; United States v. Roe , 606 F.3d 180, 186 (4th Cir. 2010). Because a rational jury could infer the requisite knowledge on Lefsih's part, Lefsih cannot prevail on his claim of insufficient evidence.

III.

We turn next to Lefsih's argument that the district court improperly conveyed negative views of Diversity Program immigrants to the jury, thus depriving him of a fair trial. Because Lefsih did not raise a timely objection, we review this alleged impropriety for plain error only. See United States v. Martinovich , 810 F.3d 232, 238 (4th Cir. 2016).

Under that standard, our review is limited to plain errors that affect a defendant's substantial rights. Id. ; United States v. Olano , 507 U.S. 725, 734, 113 S.Ct. 1770, 123 L.Ed.2d 508 (1993). In this context, that means that "we may not intervene unless the judge's comments were so prejudicial as to deny the defendant[ ] an opportunity for a fair and impartial trial." Martinovich , 810 F.3d at 238 (quoting United States v. Smith , 452 F.3d 323, 331 (4th Cir. 2006) ). We conclude that the judicial intervention in this case was sufficiently prejudicial to meet this high threshold and "undermine confidence" in Lefsih's conviction, see id. at 242, and accordingly, we vacate the conviction.

A.

The crux of Lefsih's claim is that the district court improperly interfered with his trial through "ill-advised" questions and comments, see Martinovich , 810 F.3d at 239 (finding error in district court's "ill-advised comments and interference"), posed to the government's second witness, Gary Freitas. Specifically, Lefsih argues that the district court impermissibly conveyed to the jury, through questions and comments that otherwise were wholly extraneous, the court's skepticism of the Diversity Program through which Lefsih entered the country, as well as its negative view of the immigrants—like Lefsih—who avail themselves of the Program. We must agree.It is of course true, as the government reminds us, that federal judges need not sit silently during the presentation of evidence at trial. The Federal Rules of Evidence charge district courts with "exercis[ing] reasonable control over the mode and order of examining witnesses and presenting evidence," Fed. R. Evid. 611(a), and that authority extends to both "direct participation in the examination of witnesses" and "commenting, with proper deference to the jury, upon the evidence." Sit-Set, A.G. v. Universal Jet Exchange, Inc. , 747 F.2d 921, 925 (4th Cir. 1984). Indeed, in discharging his or her duties, it may be incumbent on a trial judge to question a witness "for the purpose of developing the facts," Hoffler v. United States , 231 F.2d 660, 661 (4th Cir. 1956), or to ensure that the "case on trial is presented in such way as to be understood by the jury," United States v. Parodi , 703 F.2d 768, 775 (4th Cir. 1983). Similarly, a judge may find it necessary to comment on the evidence presented at trial, in order to "assist the jury in understanding the facts and issues in dispute." United States v. Tello , 707 F.2d 85, 90 (4th Cir. 1983). And there is no question that in these core matters of trial management, a district court is granted broad discretion. See United States v. Godwin , 272 F.3d 659, 676 (4th Cir. 2001).

But there are limits to that discretion. See id. at 677 (with respect to judicial interventions in trial, "a judge has no ‘impregnable cloak of immunity’ " (quoting Wallace v. United States , 281 F.2d 656, 665 (4th Cir. 1960) )). And as we have made clear, those limits are breached when judicial intrusion gives rise to an "appearance of bias or partiality," Parodi , 703 F.2d at 776, creating for the jury "an impression of partiality" or "apparent [ ] favor or disfavor for one side or the other," United States v. Head , 697 F.2d 1200, 1210 (4th Cir. 1982). If a district court, through its questions of a witness, interjects a "negative impression" of the defendant into a trial, or conveys "skepticism" of the defendant or his evidence, then the court has crossed the line from active trial management to "unfairly lending the court's credibility" to the government's case. Godwin , 272 F.3d at 678, 675, 674.

We have applied that principle to find error when a district court becomes "overly involved," id. at 679, in one party's presentation of its case, giving rise to an appearance of partiality. Through a pattern of one-sided interruptions and questions of defense witnesses, for instance, a court may convey to the jury that it is on the side of the government. See id . Most recently, in Martinovich , we considered a court's repeated and unnecessary interruptions of defense counsel and questioning of defense witnesses, and concluded that the court had "strayed too far from convention." 810 F.3d at 240. The concern in such cases is not necessarily with the content of the court's questions or comments, but rather that the jury may infer from the very fact of repeated interventions or interruptions that the court is sympathetic to one side of the case. See Godwin , 272 F.3d at 677 ("[W]hen a judge cross-examines a defendant and his witnesses extensively and vigorously, he may present to others an appearance of partisanship [.]") (internal quotation marks omitted) (emphasis removed); United States v. Cassiagnol , 420 F.2d 868, 879 (4th Cir. 1970) ("Constant or persistent interruption of defense counsel may have the effect of contaminating the jury's verdict by indicating the judge's evaluation of the weight of the evidence and the merits of the defense.").

This is the more unusual case in which the primary problem is not the extent of judicial participation at trial, but instead the actual content of the court's questions and comments. See United States v. Fuller , 162 F.3d 256, 259 (4th Cir. 1998) (describing as "most troubling" district court statement, during jury instructions, that it did not credit part of defendant's testimony). This jury, in other words, would have no need to deduce from a pattern of interruptions or questions that the district court was skeptical of the defendant; here, the district court conveyed that skepticism directly. In the context of an immigration-fraud case—that is, with immigration front and center before the jury—the court began with a series of questions and comments suggesting a negative view of the very immigration program through which Lefsih had entered the country: "[U]nbelievable, unbelievable." J.A. 149. And contrary to the government's argument, what reasonably could have appeared to the jury as the court's disapproval did not stop with the Diversity Program itself, or the Congress that established it. Instead, the court went on to provide a negative assessment of the people—like Lefsih himself—who make use of the Diversity Program to come to the United States. According to the court, such immigrants, unlike those admitted through other programs, need not exhibit the "skill or personal characteristics" to "contribute to the well-being of the United States." J.A. 148. And the court further questioned whether these individuals—again, including Lefsih, on trial for immigration fraud—act in good faith when participating in the Program, suggesting that lottery winners could abuse the system by "drag[ging] along [their] ten kids and four wives or what[.]" J.A. 152. As we have explained in reviewing the sufficiency of the evidence, this is a case in which the credibility of the defendant was of "crucial importance," and as a result, aspersions cast by the court on the trustworthiness of Diversity Program entrants would have carried special weight. See Godwin , 272 F.3d at 678.

We need not go through a line-by-line analysis of the court's largely rhetorical questions and comments. It is enough to say that taken together, they would have conveyed to the jury the court's "negative impression," id. , of the Diversity Program and the immigrants who avail themselves of the Program, and thus of Lefsih himself. And the impropriety of the judicial intervention here is magnified by the fact that on the other side of the balance—the obligation of a court to keep "reasonable control" of the proceedings, Fed. R. Evid. 611(a), and ensure that relevant issues are comprehensible to the jury, see Hoffler , 231 F.2d at 661 —there is nothing. A level of judicial participation that might be understandable in a "multi-week trial that involved highly complex factual issues ... numerous witnesses, and several hundred exhibits," see Martinovich , 810 F.3d at 241, will be less appropriate in the context of the two-day, three-witness trial at issue here. Cf. Smith , 452 F.3d at 333 (trial court intervention "crossed no line" where it represented "attempt to cabin and control a two-week trial that featured numerous witnesses, extensive amounts of evidence, and, even on appeal, an eight-volume joint appendix totaling well over 2600 pages"). And the court's interruptions and questions in this case did nothing to elucidate important evidence that otherwise would have been difficult for the jury to understand. Cf. Colombo v. Flemings , 43 F.3d 1465, 1994 WL 708486, at *4 (4th Cir. Dec. 14, 1994) (unpublished). On the contrary, the court's critique of the Diversity Program and the individuals who utilize that program had little, if anything, to do with the actual evidence in the case against Lefsih, who was not on trial for his method of entry into the United States but only for his answer to Question 23.We have no reason to believe that the district court intended to convey a negative impression of Lefsih to the jury, or to lend the "imprimatur" of its office to the case against him. See Godwin , 272 F.3d at 678. But the court's questions and comments—sustained, one-sided, and in the context of this short and uncomplicated trial, wholly gratuitous—nevertheless had that effect. Accordingly, we find that the district court's actions were in error.

B.

As noted above, because Lefsih failed to object at trial to the district court's interventions, we may correct the court's error only if it is both plain and one that affects Lefsih's substantial rights by denying him the "opportunity for a fair and impartial trial." Godwin , 272 F.3d at 673, 679 (internal quotation marks omitted); see Martinovich , 810 F.3d at 238. Because "the legal principles governing judicial interference claims have been long settled," the district court's error qualifies as "plain" for purposes of plain error review. Godwin , 272 F.3d at 679. And on the facts of this case, we must also conclude that the error was sufficiently prejudicial that it denied Lefsih the "fair and impartial trial," id. , to which he was entitled and "undermine[d] confidence in [his] conviction[ ]," Martinovich , 810 F.3d at 242.

The most important factor here is the closeness of the government's case against Lefsih. Where the case against a defendant is "compelling and overwhelming," we have been prepared to infer that a jury did not convict because of a court's erroneous interventions. Godwin , 272 F.3d at 680 (erroneous intervention does not affect substantial rights where "the evidence is overwhelming and a perfect trial would reach the same result"); see Fuller , 162 F.3d at 260 (judge's expression as to defendant's guilt does not warrant reversal where "facts required for conviction were admitted by the defendant during his testimony and were not controverted by any other evidence"). In Martinovich , for example, we denied relief under the plain error standard where "[t]estimony from 28 witnesses and approximately 250 exhibits" added up to an "overwhelming" case against the defendants, so that we could not conclude that the district court's improper interruptions and questions had an impact on the trial's outcome. 810 F.3d at 242.

Here, by contrast, the government's case was substantially weaker. The government could present no direct evidence on the critical question at the heart of its case: whether Lefsih knew when he answered Question 23 of the N-400 form that his answer was false, or whether, as Lefsih testified, he honestly but mistakenly believed that traffic tickets fell outside the scope of the form's inquiry into criminal history. And while it is not unusual for the government to rely exclusively on circumstantial evidence in proving state of mind, see Santos , 553 U.S. at 521, 128 S.Ct. 2020, the government's circumstantial evidence in this case was not especially strong. It may be, as the government sought to establish, that Lefsih speaks English and excels in certain academic subjects. But there is space between speaking the language and understanding the meaning of "arrested, cited, or detained" as it pertains to traffic tickets, and indeed, after the jury's verdict, the district court here characterized as "sincere" Lefsih's account of his "unknowing mistake." J.A. 307. The evidence against Lefsih is constitutionally sufficient, but it is not "compelling and overwhelming," see Godwin , 272 F.3d at 680.

Additional factors point in the same direction. First, the timing and context of the district court's interjections could only have amplified their prejudicial effect. Lefsih's sole defense—that he did not believe Question 23 included traffic tickets—depended critically on his credibility. But before Lefsih could take the stand, the district court had negatively characterized the program through which he entered the United States and commented on the risk that individuals like him would take undue advantage of the Diversity Program. And those concerns would have been especially resonant for the jury, because the government already had cast as suspicious Lefsih's choice to rely on the Diversity Program for entry rather than seek a student visa. Given this context, the court's commentary was "potentially fatal" to Lefsih's credibility-based defense. See id. at 678.

Second, factors we have relied on in other cases to mitigate the prejudicial effect of improper judicial interventions are absent here. This is not a case involving a "single comment" by a court, see United States v. Williams , 49 Fed.Appx. 420, 426 (4th Cir. 2002) ("single comment" by district court not so prejudicial as to warrant correction on plain error review), or even multiple comments spread "throughout several weeks of trial" so as to minimize their relative impact, see Martinovich , 810 F.3d at 242. Here, the improper questions and comments were "persistent and repeated," see Cassiagnol , 420 F.2d at 879, posed to one of only two government witnesses at a trial lasting for less than two full days. Nor were the court's interventions even-handed, directed at both parties alike. See Martinovich , 810 F.3d at 241 (finding no effect on substantial rights where district court interrupted and questioned both defense and government witnesses); Head , 697 F.2d at 1210 (finding no prejudice where court's "patent overinvolvement" and criticism were directed at and affected both sides). Instead, the improper questions and comments were entirely to the government's benefit, expressing disapproval only of the Diversity Program and its beneficiaries. And while a lengthy jury deliberation followed by a divided verdict may provide some assurance that a jury was not affected by improper judicial interjections, Martinovich , 810 F.3d at 242 ; see United States v. Cornell , 780 F.3d 616, 627 (4th Cir. 2015), the jury here deliberated for only 30 minutes before returning with a guilty verdict on all counts.

Against all of this, the government points to the district court's "curative instructions," Br. for Appellee at 33, given before and after trial, instructing the jury that the court has no "position about the outcome of the case" and that "nothing that I say or do during the course of the trial is intended to indicate, nor should you take it as indicating, what your verdict should be." J.A. 82. It is true, as the government argues, that proper curative instructions may "save [a] trial from reversal" where a district court has improperly intervened. Fuller , 162 F.3d at 260 ; see Smith , 452 F.3d at 333. But we also have recognized that instructions will not always be enough to "undo" the effects of judicial intervention even under plain error review, see Martinovich , 810 F.3d at 241, and the instructions here cannot bear the weight the government puts on them.

As we have explained, a curative instruction regarding a judge's improper expression of opinion should be given contemporaneously with or at least close in time to the inappropriate comment itself, so that the jury readily can connect the two. See Tello , 707 F.2d at 88 (judge "should give [the curative] instruction in sufficient proximity to his comment so that the jury will have it clearly in mind when the comment is made"). In Fuller , for instance, we relied on the effectiveness of curative "admonitions" that were "given to the jury immediately preceding and following the judge's expression of opinion" about the defendant. 162 F.3d at 260 ; see also Tello , 707 F.2d at 89–90 (relying on judge's statement at conclusion of improper comments). In this case, on the other hand, the court did not issue a curative instruction immediately after its improper comments, or even at the conclusion of Freitas's testimony, during which the court's interventions occurred. Nor, importantly, did the standard instructions ultimately given at the close of trial make any direct reference to the court's commentary on Diversity Program immigrants. Neither the timing nor the content of the "boilerplate" instructions, in other words, clearly tied those instructions to the error purportedly addressed. See Universal Jet Exchange, Inc. , 747 F.2d at 926 (finding district court's "boilerplate final instruction to the jury" after persistent judicial intervention insufficient to "remove[ ] the impression ... necessarily conveyed of judicial partiality"). As a result, and in light of all of the factors bearing on prejudice, we cannot find these instructions sufficient to "undo" the harm, see Martinovich , 810 F.3d at 241, caused by the district court's error.

Again, we do not doubt that the district court in this case acted without any intent to influence the jury improperly. And we appreciate that the defendant's failure to object at trial—which might have allowed the district court to correct its error in real time—creates an especially "high bar for appellate review." See id. at 239. But in the particular circumstances of this case, we conclude that the court's interventions were not only plainly erroneous but also "so prejudicial as to deny the defendant[ ] an opportunity for a fair and impartial trial," thus affecting Lefsih's "substantial rights." See id. at 238 (internal quotation marks omitted). And because the error here is of the kind that may seriously affect the "fairness, integrity or public reputation of judicial proceedings," we will correct it by vacating Lefsih's conviction. See Olano , 507 U.S. at 736, 113 S.Ct. 1770 (describing circumstances under which court of appeals may exercise discretion to correct a plain error affecting substantial rights).

IV.

For the foregoing reasons, we vacate the judgment of conviction and remand for proceedings consistent with this opinion.

VACATED AND REMANDED