In the anarchy following Hurricane Katrina, a group of heavily armed New Orleans police officers were dispatched to the Danziger Bridge in response to an emergency call reporting shots being fired at police. There, amid chaos, they shot and killed two unarmed men, one of them developmentally disabled, and wounded four other unarmed civilians. The police then allegedly orchestrated a cover-up to deny what happened. Some of those involved were tried by the state, but a mistrial was ordered. The federal government took over the prosecution and has also bungled it. Five former officers have been convicted of serious crimes and received lengthy sentences. Yet they appear in this court as Appellees, and the federal government as the Appellant, because the district court granted a new trial.
The reasons for granting a new trial are novel and extraordinary. No less than three high-ranking federal prosecutors are known to have been posting online, anonymous comments to newspaper articles about the case throughout its duration. The government makes no attempt to justify the prosecutors’ ethical lapses, which the court described as having created an “online 21st century carnival atmosphere.” Not only that, but the government inadequately investigated and substantially delayed the ferreting out of information about its in-house contributors to the anonymous postings. The district court also found that cooperating defendants called to testify by the government lied, an FBI agent overstepped, defense witnesses were intimidated from testifying, and inexplicably gross sentencing disparities re-*340suited from the government’s plea bargains and charging practices.
Like the district court, we are well aware of our duty normally to affirm convictions that are tainted only by. harmless error. In this extraordinary case, however, harmless error cannot even be evaluated because the full consequences of the federal prosecutors’ misconduct remain uncertain after less-than-defínitive DOJ internal investigations. The trial, in any event, was permeated by the cumulative effect of the additional irregularities found by the district court. We conclude that the grant of a new trial was not an abuse of the district court’s discretion.
The following discussion cannot be fully understood without reference to the district court’s lengthy, comprehensive, and careful opinions in which it evaluated the prosecutorial misconduct as it was revealed to the district court and made significant findings of fact on which we rely. This opinion summarizes only the highlights of those findings. See United States v. Bowen (December Order), No. 10-204, 2013 WL 6531577 (E.D.La. Dec. 12, 2013); United States v. Bowen (September Order), 969 F.Supp.2d 546 (E.D.La.2013); United States v. Bowen (November Order), 969 F.Supp.2d 518 (E.D.La.2012).
BACKGROUND
A federal grand jury returned a 25-count indictment against former New Orleans Police Department (“NOPD”) officers Kenneth Bowen, Robert Gisevius, Robert Faulcon, Anthony Villavaso, and Arthur “Archie” Kaufman for their roles in the Danziger Bridge shootings and ensuing alleged cover-up. The indictment charged defendants with civil rights, firearms, conspiracy, and obstruction of justice offenses; only Faulcon was indicted for actually making a fatal shot.1
Several other former police officers indicted at the same time pled guilty, and most testified at trial for the government. Despite their egregious behavior at the Danziger Bridge, the cooperating defendants received much lighter sentences because the government agreed not to charge a series of firearms offenses that carry substantial minimum required, consecutive sentences.
Emotions ran high as the prosecution progressed. Local news coverage of the impending federal indictments was punctuated by press leaks “from unnamed sources” that tended to favor the government. One cooperating defendant, Lehrmann, signed a confidential plea agreement, and a magistrate judge sealed the Information against him. One day before Lehrmann was scheduled to enter the plea in open court, the Associated Press and the New Orleans Times-Picayune, the local paper of record, published articles announcing that fact. The district court ordered the government to attempt to find the leak, but the order bore no fruit.
Concomitantly, commenters on the website for the New Orleans Times Picayune vigorously debated the significance of the case and the guilt of the individual perpetrators and the entire New Orleans Police Department. The indictments were handed down on July 12, 2010, the trial occurred over a two week period from late June to early July 2011, and the defendants were found guilty on nearly all counts. There is no dispute that the district court conducted a thorough and con*341scientious jury voir dire based on the information known at the time.2
During the interim between the verdict and sentencing, events reflecting shocking breaches of prosecutorial ethics were revealed and then compounded by further breaches. To make a very long story short, the district court was led on a “legal odyssey” by the government that began in March 2012 when another target of federal investigation in New Orleans discovered that a high-ranking Assistant United States Attorney, Senior Trial Counsel Sal Perricone, had been posting comments to Nola.com under multiple assumed names.3 Perricone’s comments frequently involved other matters pending in the United States Attorney’s Office (“USAO”) and were inflammatory, highly opinionated, and pro-prosecution. Perricone’s comments were soon also tied to the Danziger Bridge prosecutions and were shown to have begun well before the indictments and continued through trial. He castigated the defendants and their lawyers and repeatedly chastised the NOPD as a fish “rotten from the head down.” Within ten days of the March 2012 revelation of Perricone’s comments, he resigned as an AUSA, and then-United States Attorney Jim Let-ten issued a press release attempting to confine any online misconduct to Perricone alone.
Prompted by the revelation of Perricone’s comments, the defendants moved for a new trial based on the commenting and repeated press leaks that, they contended, had inflamed public opinion against them. In addition to the prejudicial atmosphere, they charged that the government induced coerced guilty pleas and procured false testimony to secure convictions at any cost.
The district court’s first hearing on these allegations occurred in June 2012. At the hearing, United States Attorney Letten was flanked by his First Assistant United States Attorney and Chief of the Office’s Criminal Division Jan Mann as he promised “gospel truth” that no one else had commented on stories related to pending cases.4 Additionally, the DOJ’s chief prosecutor in this case, Barbara Bernstein, represented to the district court that no member .of “the trial team” had commented online. The district court acutely observed that its concern about leaks and publicity was not limited to the “team” but extended to all of the federal government. The district court repeatedly expressed skepticism that a new trial would be required, but it ordered the USAO to conduct two investigations. The first asked the government to reveal who had leaked the Lohman guilty plea to the press. The second, responsive to the defendants’ claims about online activity, was to verify that only Perricone had publicly commented about the case and to catalog the comments chronologically.
Jan Mann was tasked by Letten to conduct the investigations within the New Or*342leans office. She reported back to the district court with assurances that Perricone was the sole culprit in the USAO and that the defense was likely responsible for press leaks.5
These incomplete initial reports failed to overcome the district court’s concerns. The district court’s opinion explains several deficiencies, November Order, 969 F.Supp.2d at 533-38, but it suffices here to note that Perricone, now a private citizen, had not been questioned under oath by Jan Mann. The catalog of Perricone’s comments obtained by the district court reflected those he composed as “HenryL.Mencken_1951” but did not include comments he submitted under alternative monikers.
Surprisingly, in the first week of August, while the district court’s inquiries were still being pursued, an extensive interview of Perricone was published in New Orleans Magazine. The interview was both revelatory and self-serving. Revelatory, inter alia, was the information that Perricone had actually posted comments about pending USAO matters under not one but several assumed names. Self-servingly, he stated his commenting was “my secret,” that he had been motivated to post comments only to defend the practices of his office, and that no one else in the office, specifically Letten and Jan' Mann, had known of his activity.
The district court was pondering a request by the defendants to hold a hearing on their motion for a new trial. While responses from the USAO had up to this point been under seal, the district court identified several categories of emails produced by the government that the district court believed could be relevant to the extent of possible misconduct. After hearing both sides about whether those documents should be revealed to the defendants, the district court ordered all but one produced under a protective order. The district court also set a status conference at which Perricone would testify under oath.
At that October 10, 2012 status conference, attended by DOJ representatives, Jan Mann, and defense counsel, Perricone testified extensively. From the district court’s perspective, Perricone raised further questions about the possible involvement of the local FBI in press leaks and about online monikers that he affirmed he did not use, specifically those of “eweman” and certain variations on “campstblue.” Postings under these additional names had come to the district court’s attention because their content implied they might have been written by insiders to the prosecution. Perricone admitted to using several monikers, but he could not recall if he had used others. He also stated that he posted generally at nights and on weekends, although a handful were written in his office at work. A colloquy toward the end of this hearing led the district court shortly afterward to write Jan Mann seeking information on whether any personnel associated with the federal courts might also have been posting comments online. Jan Mann responded on October 19 with a letter that included the following statements:
Prior to the Perricone incident, I was not a follower of nola.com postings and had no real sense of what was happening there.... In trying to express these thoughts about human failings and flaws, about hypocrisy and hidden agen*343das, I did not intend to suggest that anyone else in particular was posting. If I was so perceived, I regret it.
Before it could take testimony from another former AUSA,6 as the district court put it “Another Shoe Drops/Another ‘Secret’ — Friday November 2, 2012,” a lawsuit was filed alleging that Jan Mann had been commenting, as “eweman.” About forty inappropriate comments under this moniker had appeared on the Nola.com website from November 2011 until Perrieone was exposed in March 2012. The suit alleged that “approximately 63 percent of the posts by [First AUSA Mann] appear with comments posted by Perricone, and they frequently reply to express consistency with the points of view expressed by the other.” November Order, 909 F.Supp.2d at 530. It took four days, plus a specific request by the district court, before Letten informed the district court that indeed, Jan Mann had “much to his surprise” admitted her activity on Nola.com.
The district court’s hearing to take testimony from Magner occurred on November 7, with Letten (standing in for Jan Mann), representatives of the DOJ, and defense counsel. Magner yielded insights into the possible knowledge of other AUSAs and office personnel about Perricone’s comments. He specifically insinuated that Mann and her husband, AUSA Jim Mann, as friends of Perricone, knew about the commenting. Bernstein from DOJ stated her “surprise” at the original allegations against Perricone and that she was “flabbergasted” and incredulous about Mann’s involvement. Bernstein assured the district court that, having interviewed current and former members of “the prosecution team,” she was told that none of them had been posting comments. She contended that because the postings were anonymous, the district court had conducted a thorough jury voir dire, and Jan Mann’s postings on the Danziger Bridge case apparently post-dated the trial, “the conduct at issue had no effect on the validity” of the verdict against the defendants. Id. at 533.
The district court’s November 2012 opinion and order summarizing developments to this point culminates with two overlapping questions: (1) what is the full extent of the misconduct, and what are its institutional ramifications; and (2) how does such misconduct, both that which has occurred and what the defendants believe to have occurred, affect the validity of the verdicts under Federal Rule of Criminal Procedure 33? Id.
The district court was as yet unable to rule on the motion for a new trial because it was “unfortunately hampered by the inability of the DOJ to provide reliable investigatory answers.” Id. at 537. The district court’s indecision stemmed from an. inadequate investigation into the leak of the Lohman guilty plea. Also contributing was Jan Mann’s conflicted and untrustworthy role in the investigation. She had stood by in open court while Letten proclaimed “gospel truth” that no one else in his office had posted comments and was the sole AUSA responsible for investigating the extent of online commenting. All the while, the district court surmised, she had to be trying to protect herself from Perricone’s fate. While still indicating skepticism that the government’s errors *344could justify a new trial, the district court made significant prehminary findings:
(1) Certain members of the USAO monitored and reviewed Nola.com articles, in particular the “comment” postings, and shared them with other members of the office;
(2) Some members of the USAO determined that the posts suspiciously seemed to contain confidential, privileged, or sensitive information about a variety of cases in which the office was involved;
(3) Certain members of the USAO commented to each other on their suspicions, particularly concerning commented named “legacyusa” and “HenryL.Mencken_1951,” and linked those posts possibly to Perricone;
(4) Two individuals in the office emailed each other only a week before the Lohman plea and four months before the Danziger Bridge indictments, and indicated that comments concerning the Danziger Bridge incident were written by people “who know a little bit too much about our office ... ”; and
(5) Jan Mann supervised the responses to the district court’s attempts to ascertain the extent of online commenting within the office to that point.
The district court quoted a number of the Perricone comments directly relating to the Danziger Bridge prosecution and was worried, particularly in light of the belated identification of Jan Mann as “ewe-man,” whether these were the only unauthorized comments from within the office. The district court reiterated its concern, previously expressed, that at least one cooperating defendant felt coerced, into pleading guilty, that the sentences meted out to defendants were shockingly disparate, that FBI Agent William Bezak had used coercive tactics against a defense witness, and that the defense was deprived of live testimony by at least three witnesses who refused to testify at trial, when DOJ targeted them for possible perjury charges. Three years after trial, however, not one of those people had actually been charged with a crime.
In sum, the district court ordered DOJ to recommence investigation of both the Lohman leak and the commenting and to finally answer the questions the court had raised. The district court strongly urged DOJ to appoint impartial investigators. This order was entered on November 26, 2012.
Would that this were the end of the story. Two weeks after this order, Letten resigned as United States Attorney, and both Jan Mann and her husband Jim retired with their panoply of federal benefits intact.7 The DOJ acted on the district court’s suggestion by appointing two attorneys — FAUSA John Horn and AUSA Charysse Alexander — from the Northern District of Georgia to conduct an investigation ranging from the New Orleans office of the United States Attorney to the Criminal Division of DOJ.8 From the end of January until July 2013, Horn and Alexander issued an initial report, followed by *345four supplemental reports and voluminous supporting material, which are best described as a whirlpool cycling toward and gradually reaching its drainage outlet. From a starting point of generality and vagueness about the misconduct within the USAO and DOJ, the district court painstakingly pried more details and startling information by asking questions two and three times.9
For example, the district court ascertained that neither Jan Mann nor her husband had ever been placed: under oath when being interviewed by Horn and OPR, and each had refused to execute affidavits.10 The district court finally gained access to Jan Mann’s interview, well after it occurred, and found it incomplete.11 During her OPR interview, she claimed for instance, that she had essentially told Let-ten about her own online comments back in March 2012, and that she assumed that his conduct and carefully worded statements thereafter were influenced by her confession. Letten, in his interview, denied having been so informed by Ms. Mann. As the district court later queried, who is to be believed?
The district court’s curiosity was further piqued by a carefully worded reference in the initial Horn report to comments about the trial posted by a DOJ Civil Rights Division employee (pseudonym “Dipsos”) “who had first-hand knowledge of the Danziger Bridge case but was not a member of the prosecution team.” Answering the district court’s inquiry, the,First Supplemental report, filed in late March, stated that this employee was, inter alia, “walled off from the prosecution team and was prohibited from having any substantive discussion about the investigation with any member of the prosecution team or any supervisor over the prosecution team.” The investigators remained obviously evasive about this person’s identity until pointedly questioned by the district court. Not until a May 15 in-chambers hearing with Horn and Alexander was “Dipsos” revealed to be Ms. Karla Dobinski, a decades-long Civil Rights Division attorney, who had actually testified before Judge Engelhardt in her capacity as the head of the DOJ’s internal “taint team” in the Danziger Bridge case. Her responsibility in the course of the prosecution was to protect indicted police officers’ civil rights.12 In the October 2012 hearing, Bernstein had not mentioned Dobinski as one of the “prosecution team” whom she queried about online commenting. The district court makes no accusation against Bernstein; we note, however, that her denial of online commenting by the team proved misleading.
*346Other gaps in the investigation were evident. Neither Perricone nor Jan Mann has acknowledged that the monikers so far -discovered to have been used by them are the only ones under which they posted comments. Dobinski is disturbingly vague in her OPR interview about how many other people in her department were aware of her commenting and whether “Dipsos” was her only moniker.13 DOJ simply refused to follow up with the newspaper reporters who had written articles referencing “two people familiar with the investigation” and “a source close to the probe” concerning the Lohman plea leak. And, “[i]n a truly disappointing and unsettling crucial development, the Second Supplemental Report also indicates that the DOJ could not forensically recover computer data from the USAO’s Internet portals for years 2010 and 2011 (prior to December 19, 2011) because ‘it did not retain data for the period before that.’ ” The district court’s attempt to discover other online prosecutorial misconduct was thus undermined.
Because of the indeterminacy about the extent of prosecutorial misconduct, the district court was faced with holding a public hearing on the new trial motion in which it would sift live testimony about the Lohman guilty plea leak and prosecutorial remarks about the Danziger Bridge case. Critical testimony would have been developed from the conflicting statements by prosecutors themselves, including Letten, on these subjects. The district court desisted from this potentially embarrassing course of action in part because of the additional delay, but more so based on its conclusion that a critical mass of unethical and unprofessional deeds supported a new trial.14 In ninety pages, the district court develops the grounds' for granting a new trial.
*347Factually, the order relies on a number of events. Foremost are the online comments of Perricone and Dobinski, which in tandem tended to create an “On-Line 21st Century ‘Carnival Atmosphere’,” as graphically depicted in the district court’s quotation of all of the comments posted by these two. September Order, 969 F.Supp.2d at 588-603. Second is Jan Mann’s testimony, which includes the implications that Letten was aware of her online posting activity in March 2012 (when Perricone’s postings were first ‘outed’); that she suspected and believed Letten had reported up the line to DOJ about the posting; she “believe[d] in her heart” that other AUSAs were also commenting on Nola.com; and she believes that the USAO and DOJ purposefully avoided directly asking its personnel about their online posting while denying that any organized propaganda campaign was occurring in the Danziger Bridge case. Third, the district court evinces doubt about the credibility and tactics of FBI Special Agent Bezak, who at one point appeared to have defied a court order prohibiting contact with defendant Villavaso without his attorneys’ presence. After the government produced a corrected timeline of events, Bezak appeared to be cleared of direct misconduct, but his critical chronological error in a significant matter troubled the district court. Additionally, Bezak did himself no credit by attempting to excuse the highly questionable testimony of the government’s cooperating defendant Lohman by saying that, “It is Mike Lohman’s truth,” and “Every person has their own memory, recollection interpretation of events.” Bezak had also threatened a potential defense witness with being separated from her three children “for lying,” yet this witness was never charged.
The government’s tactics extended to the presentation of testimony by cooperating defendant Hunter that was inconsistent and incredible to the point that the district court dismissed Count 10, the only count that depended on Hunter’s testimony, and the government has not appealed. That the government was familiar with “problems” in Hunter’s testimony is infer-able from the district court’s finding that the testimony deviated significantly from Agent Bezak’s handwritten notes of his previous interview with Hunter. Not only this judge heard testimony from Hunter: Chief Judge Sarah Vance heard his testimony and strongly doubted its credibility and his sincerity in accepting responsibility. Denying the government’s request for more lenient treatment, she sentenced him to the maximum eight years imprisonment on charges under a favorable plea agreement for obstructing justice and misprision of a felony. See id. at 612.
The disparity between the punishment meted out to cooperating defendants and those who went to trial is stark. The cooperating defendants’ participation in the incident and cover-up seems comparable, yet the government threw the book at those who went to trial by stacking firearms charges. As a result of the charging disparity, those who went to trial have been sentenced from thirty-eight to sixty-five years in prison, while the cooperating defendants garnered from five to eight years. In addition to Hunter,15 another cooperating defendant, Hills, had fired at people on the bridge. Hills pled guilty to misprision/obstruction charges and was sentenced to six and a half years in prison, but he had earlier denied his guilt to his supervisor. He explained that he had to take the plea deal as “the best [he] could get.” A third cooperating defendant, Bar*348rios, was the only defendant who pled out but was not presented as a witness by the government. Barrios had been present at the bridge and changed his statements about whether he fired on the civilians. Barrios received a five-year sentence. When he testified for the defense, he had to contradict his wife’s statements that he had been forced to admit guilt despite his innocence.
Also troubling was the saga of cooperating defendant Jeffrey Lehrmann, who received only a three-year sentence for misprision of a felony. Yet Lehrmann worked hand in glove with Kaufman, who was charged with multiple felonies and received a six-year sentence. In crafting false evidence, Lehrmann went so far as to create a fictitious witness, “Lakeisha Smith,” to fortify the defendant officers’ stories about the shooting. Lehrmann also falsely charged crimes against Lance Madison, the brother of a murdered, disabled victim. Lehrmann received his favorable plea deal even after he had lied to the federal grand jury. Even more surprising, Lehrmann had been hired as a federal ICE agent during the pendency of the Danziger Bridge investigation and worked as a federal agent for nearly four years. His federal employment terminated several months after his formal guilty plea.
Finally, the district court pointed to its understanding that at least three potential defense witnesses refused to testify following prosecution threats to bring perjury charges against them. As the district court explained, since these witnesses had earlier testified to the grand jury, then-transcripts could be offered at trial, but transcripts are never as powerful as live witnesses. In any event, not one of those people was later charged.
In granting the defendants’ Rule 33 motion, the district court principally relied on footnote nine of Brecht v. Abrahamson, which reserves the possibility that a new trial can in some egregious circumstances be mandated for certain “trial-type” errors even without a showing of prejudice to the defendants. 507 U.S. 619, 638 n. 9, 113 S.Ct. 1710, 1722 n. 9, 123 L.Ed.2d 353 (1993). The court also concluded that the defendants were in fact prejudiced.
In autumn 2011, well before it ordered a new trial, the district court dismissed, for insufficient evidence, and alternatively granted a new trial on Count 10 against Bowen only, Count 12 against Bowen, Gisevius, Faulcon and Villavaso, and Count 13 against Bowen and Gisevius. The government did not oppose, and has not appealed, dismissal of Count 10. Count 10 charged that Bowen kicked Ronald Madison as he lay dying and was supported essentially by the testimony of the government’s discredited cooperating defendant Hunter. Counsel for the parties agreed during oral argument that if we affirm the grant of a new trial on all counts other than Count 10, we need not discuss the dismissal of Counts 12 and 13.
The district court issued a subsequent order on December 12, 2013 that, in the course of deciding which documents to unseal for public view, restated and bolstered the district court’s new trial findings.16 In January 2014, the district court stayed further proceedings pending this appeal.
DISCUSSION
The government’s appeal, in pertinent part, challenges the district court’s new trial decision with arguments why “anonymous online postings by government attor*349neys do not warrant a new trial in this case.” As subsidiary points,.the government challenges the ruling that other aspects of the prosecution cumulatively harmed the defendants and supported the new trial grant, and it seeks removal of the district judge from further proceedings. We discuss each issue in turn.
In reviewing whether the district court abused its discretion in the grant of a new trial, we review questions of law de novo, but the district court’s findings of . fact must be upheld unless they are clearly erroneous. United States v. Mann, 161 F.3d 840, 860 (5th Cir.1998). The district court’s lengthy opinions embody hundreds of subsidiary findings, few of which are challenged by the government. With only one exception, the government fails to challenge the district court’s findings on the prosecutors’ credibility.17
The motion for new trial here was granted because “the interest of justice so requires,” Fed.R.Crim.P. 33(a), and the motion was specifically based on newly discovered evidence. Fed.R.Crim.P. 33(b)(1). Newly discovered evidence need not relate only to guilt or innocence, but may be relevant to any controlling issue of law. C. Wright & S. Welling, 3 Fed. Practice & Proc. § 588, at 448 (2011). If a court finds “that a miscarriage of justice may have occurred at trial, ... this is classified as such an ‘exceptional ease’ as to warrant granting a new trial in the interest of justice.” United States v. Robertson, 110 F.3d 1113, 1120 n. 11 (5th Cir.1997) (citations and internal quotation omitted). A miscarriage of justice harms the substantial rights of a defendant, and it may consist of errors and omissions considered for their cumulative effect on the trial proceedings. United States v. Barrett, 496 F.3d 1079, 1121 (10th Cir.2007); see also United States v. Sipe, 388 F.3d 471, 492 (5th Cir.2004).
The dissent seems to disconnect Rule 33(b)(1), governing the timing of a new trial for newly discovered evidence, from the overarching principle that for any new trial motion, “the interest of justice” must be considered. Rule 33(a). We disagree with the dissent in two respects. First, the dissent would confine the instant analysis to the strictures on newly discovered evidence stated in the case law, which ordinarily require a demonstration of prejudice to the verdict. See, e.g., United States v. Bowler, 252 F.3d 741, 747 (5th Cir.2001) (per curiam). We are aware of these strictures, and we note that the other four criteria stated in the case law— that the evidence of illicit government online posting was newly discovered and unknown at the time of trial; the defendants did not lack diligence in discovering the evidence; the evidence is not merely cumulative or impeaching; and the evidence is material — are not challenged by the dissent. Id. The problem with this reasoning is that this court has never had occasion to consider how to respond to the unique set of events presented in this case, that is, to online commenting discovered post-verdict and to the inability of the trial court and this court to know even at this point the extent of the prejudicial commenting because of the government’s dilatory conduct.
*350The Brecht analysis, as will be seen, fits this unprecedented scenario and is but a gloss on current Rule 33 precedent; surely the Federal Rules of Criminal Procedure have to accommodate Supreme Court decisions concerning the fundamental fairness of a trial. It is inexplicable hyperbole to predict that this decision renders the Bowler standard “meaningless.”
Second, the dissent artificially confines the scope of newly discovered evidence to “the identities of the commenters,” as if (a) Perricone, Jan Mann and Dobinski were the only commenters, and (b) the “identities” were separable from the inflammatory comments themselves, and therefore (c) the impact of the comments on the voir dire process, trial jurors, witnesses, and defendants are quantifiable. From the state of this record, however, neither we nor the defendants can know who all the commenters were, how many online comments are attributable to biased and vindictive federal government employees acting outside the bounds of their ethical duties, and thus the full impact of the misconduct. What is known, at a minimum, is that seven of twelve seated jurors were familiar with nola.com postings.
I. The Consequences of Anonymous Online Postings
That three supervisory-level prosecutors committed misconduct in connection with the Danziger Bridge prosecution is beyond dispute. Perricone’s comments spanned the entire prosecution and went directly to the guilt of the defendants, the collective guilt of NOPD, and the relative competence and integrity of defense counsel versus the USAO. Dobinski’s comments stirred the pot by encouraging commenters who were plainly familiar with the trial proceedings, one of whom was Perricone, to keep doing a “public service” with their biased reports. Mann’s comments, posted during post-trial sentencing proceedings, displayed partiality toward the prosecution and denigrated the district court and defense counsel in another Danziger Bridge case.
The government acknowledges significant, repeated misconduct by Perricone and Jan Mann and, to a lesser extent, Dobinski. The government concedes that Perricone “intentionally committed professional misconduct” violating (a) federal regulations restricting extrajudicial statements by DOJ personnel relating to civil and criminal proceedings,18 (b) DOJ policies 19 and (c) court and state bar rules of professional conduct.20 The government acknowledges that besides his postings in this case, Perricone posted “thousands” of anonymous comments on various topics over the course of several years. As to Jan Mann, the government admits that her postings on Nola.com of “anonymous comments about Department of Justice matters” violated the same rules, although the results of her postings relating to the Danziger Bridge prosecution are mitigated because they allegedly occurred after the trial had concluded. The government also admits that Mann acted dishonestly during the new trial proceedings when she misrepresented facts and allowed them to be misrepresented to the district court. The government rejects Perricone’s and Mann’s repeated assertions that their pri*351vate, anonymous online commenting could be separated from their professional public duties. As the “taint” team leader for the prosecution, the government observes, Dobinski was prohibited from participating anonymously in a public forum discussing the case, because “several sources of authority broadly prohibit Department attorneys from making any extrajudicial statements regarding a pending matter.”21 Contending, however, that her comments were innocuous and “not intentional” misconduct, the government acknowledges only that Dobinski exercised “poor judgment” in posting comments during the trial.
What the government nowhere confronts is the incomplete, dilatory, and evasive nature of its efforts to respond to the district court’s inquiries about the full extent of online activity by government employees and the source of the Lohman plea leak. The district court was stymied and frustrated for more than twelve months (June 2012-July 2013) by the government’s tactics, while private sources like a local magazine and an individual under federal investigation repeatedly leapfrogged government admissions of official misbehavior. The district court doggedly pursued the truth about these matters, but to this day the government has never fully answered the district court’s legitimate questions.
Ignoring the procedural deficiencies of its misconduct investigation, the government’s defense against a new trial consists of two essential premises: first, only a finding of specific prejudice to the verdict will suffice to support a new trial; and second, prejudice was not shown on the record developed in the district court. We disagree that specific prejudice is a necessary prerequisite to a new trial in this sui generis case. But even if required, prejudice here was shown both from this pattern of misconduct and evasion and from other abusive prosecutorial actions.
A. Brecht
The district court concluded that the government’s protracted misconduct required a new trial under Brecht.22 In Brecht, the Supreme Court held that to obtain relief on collateral review, a habeas petitioner must establish that the constitutional trial error had a “substantial and injurious effect or influence in determining the jury’s verdict.” Id. at 637-38, 113 S.Ct. at 1722 (quoting Kotteakos v. United States, 328 U.S. 750, 776, 66 S.Ct. 1239, 1253, 90 L.Ed. 1557 (1946)). In other words, the habeas petitioner must establish actual prejudice. The Court distinguished such “trial errors,” which occur during the presentation of a case to the jury, from “structural defects” in the prosecution, like denial of counsel, that require automatic reversal of a conviction. 507 U.S. at 629-30, 113 S.Ct. at 1717. The Court also stated that its holding:
does not foreclose the possibility that in an unusual case, a deliberate and especially egregious error of the trial type, or one that is combined with a pattern of prosecutorial misconduct, might so infect the integrity of the proceeding as to warrant the grant of habeas relief, even if it did not substantially influence the jury’s verdict.
Id. at 638 n. 9, 113 S.Ct. at 1722 n. 9 (emphasis added).
The district court found that the government’s pervasive misconduct so contami*352nated every phase of the prosecution that this case, unique “in nature [and] in scope,” fit “squarely within” Brecht’s prejudice exception. September Order, 969 F.Supp.2d at 619.
The government argues that (1) the Supreme Court did not create an exception to the prejudice requirement in Brecht but merely reserved the possibility for a future decision and (2) thus, prejudice cannot be presumed but must'be proven. These contentions assume that the misconduct in this case — online commenting, guilty plea leak, and the incomplete investigation of misconduct — was not sui generis. Yet neither the government’s efforts nor our additional research reveals cases on point or closely analogous.
One certainty is that the government presents an overly restrictive interpretation of Brecht footnote nine. Contrary to the government’s position, this court has described the errors contemplated in footnote nine as “hybrid” errors, falling somewhere on a spectrum between structural errors (prejudice presumed) and trial errors (subject to the harmless error standard). See Burgess v. Dretke, 350 F.3d 461, 471 (5th Cir.2003). In Burgess, this court held that under Brecht, “if ‘structural’ or ‘hybrid’ error occurs, harmless error review is inappropriate.” Id.; see also Cupit v. Whitley, 28 F.3d 532, 538 (5th Cir.1994). We interpreted Brecht to create a new category of errors, not simply to preserve the possibility of doing so in the future. See Burgess, 350 F.3d at 471; Cupit, 28 F.3d at 532. And we have interpreted Brecht to hold that “a federal court in habeas must generally review a state court’s decision using a strict ‘harmless error’ standard, but that cases involving ‘structural’ or ‘hybrid’ error require reversal regardless of harm.” Burgess, 350 F.3d at 471 (emphasis added).
This court’s holdings are in accord with those of the other circuits that have addressed the issue. See United States v. Harbin, 250 F.3d 532, 545 (7th Cir.2001) (trial errors described in Brecht footnote nine require automatic reversal); Hardnett v. Marshall, 25 F.3d 875, 879 (9th Cir.1994) (hybrid footnote nine error is “assimilated to structural error and declared to be incapable of redemption by actual prejudice analysis”). The Hardnett court added, “[w]e assume that the facts set out in Footnote Nine are illustrative, not exclusive, and that the key consideration is whether the integrity of the proceeding was so infected that the entire trial was unfair.” Id. In addition, the Third Circuit, while refusing to recognize footnote nine as “truly” establishing an exception to harmless error, still did not “foreclose the possibility” that such an exception could exist. Hassine v. Zimmerman, 160 F.3d 941, 959 n. 29 (3d Cir.1998) (quoting Brecht, 507 U.S. at 638 n. 9, 113 S.Ct. at 1710 n. 9).
Most decisions considering the possibility of Brecht footnote nine “hybrid” error have declined to grant relief to defendants, because most of the complaints have involved pure trial error.23 As the Court noted in Brecht, when the errors occur during the presentation of the case to the jury, they are amenable to harmless error review “because they may be quantitatively assessed in the context of the evidence as a whole, to determine the effect on the trial.” Harbin, 250 F.3d at 544 (citing Brecht, 507 U.S. at 629, 113 S.Ct. at 1710). But “[n]ot every error ... is easily shoe*353horned into one of those neat categories. The ‘nature, context, and significance of the violation,’ for instance, may determine whether automatic reversal or the harmless error analysis is appropriate.” Id. (citation omitted). Courts must therefore decide where, along the spectrum of errors, those which are not clearly trial type or structural may fall.
The “unprecedented” Harbin case, 250 F.3d at 539, is instructive. In Harbin, the court allowed the prosecutor to “save” until mid-trial a peremptory juror challenge, which he exercised to replace one juror with an alternate. The replacement juror was not shown to have been biased, nor could any impact on the verdict be proven. Nevertheless, the Seventh Circuit held that the error was precisely the type that “defies harmless error analysis.” Id. at 545. The defendants had been denied a comparable right to excise a juror during trial not only because they had previously used all of their peremptory challenges, but also because they were not even informed of the possibility of delayed exercise. Harbin held that this error affected the fundamental fairness of the trial and demanded a new trial without harmless error proof, because the imbalance in peremptory challenges “gave the prosecutor unilateral discretionary control over the composition of the jury mid-trial.”. Id. at 547-48. And it was “simply impossible as a practical matter to assess the impact on the jury of such an error.” Id. at 548.
So too here, the breadth of the government’s misconduct and continued obfuscation, as the district court found, makes this the “unusual case” contemplated by Brecht. The online commenting alone, which breached all standards of prosecutorial ethics, gave the government a surreptitious advantage in influencing public opinion, the venire panel, and the trial itself. And by degrees, the district court was led to conclude that what it had previously considered to be isolated missteps was actually evidence of a pattern of misconduct that permeated every stage of the prosecution. And because the government refused to adequately investigate its errors, covered up what it knew to be misleading omissions, and in some instances lied directly to the court, the district court could neither uncover the extent of the prosecution’s transgressions nor determine the severity of the prejudice suffered by the defendants.
The government’s unrelenting efforts thus prevented the district court from evaluating the fairness of defendants’ trial and thrust the prosecution into the rare territory of Brecht hybrid error. Trial errors can be evaluated for harmlessness precisely because the nature and extent of the harm is ascertainable from a review of the record. Brecht, 507 U.S. at 629, 113 S.Ct. at 1710. Here, the ability of trial and appellate courts to evaluate the effect of the anonymous comments has been thwarted by the government’s subsequent lack of cooperation. There is a fundamental imbalance between the knowledge of the prosecutors, on one hand, and the defendants and courts, on the other, concerning the true extent and significance of the ongoing commenting. This ease thus presents the unclassifiable and pervasive errors to which the Supreme Court referred in Brecht when it identified a category of errors capable of infecting the integrity of the prosecution to a degree warranting a new trial irrespective of prejudice.
Our conclusion is reinforced by overarching standards of prosecutorial conduct and the nature of their breach.24 Prosecu*354tors maintain the integrity, fairness and objectivity of the criminal justice system in part by refraining from speaking in public about pending and impending cases except in very limited circumstances. The government’s own list of applicable regulations and ethical rules demonstrates that the prosecutors’ obligation of silence extends beyond “confidential and grand jury matters” and beyond the “prosecution team” narrowly defined to include only those who participate in a particular case. Further, there is no dividing line between the prosecutors’ professional and private lives with respect to these duties. Had Perrieone, Mann, or Dobinski frequented a bar or habitually called in to a radio talk show and blown off steam about the Danziger Bridge prosecution in the terms they used online, their misconduct would have been the same as it is with their anonymous online commentary.
The reasons for prosecutorial self-restraint are manifest.
Although ‘[statements to the press may be an integral part of a prosecutor’s job, and ... may serve a vital public function,’ that function is strictly limited by the prosecutor’s overarching duty to do justice.’ Those who wield the power to make public statements about criminal cases must ‘be guided solely by their sense of public responsibility for the attainment of justice.’
Aversa v. United States, 99 F.3d 1200, 1216 (1st Cir.1996) (quoting Souza v. Pina, 53 F.3d 423, 427 (1st Cir.1995)). Insulating the prosecution and trial from bias, prejudice, misinformation, and evidence revealed outside the courtroom are crucial to the fairness of our processes. Equally important, the prosecutor must respect the presumption of innocence even as he seeks to bring a defendant to justice.
Justice Sutherland eloquently captured the prosecutor’s calling. A government prosecutor
is the representative not of an ordinary party to a controversy, but of a sovereignty whose obligation to govern impartially is as compelling as its obligation to govern at all; and whose interest, therefore, in a criminal prosecution is not that it shall win a case, but that justice shall be done. As such he is in a peculiar and very definite sense the servant of the law, the twofold aim of which is that guilt shall not escape or innocence suffer. He may prosecute with earnestness and vigor — indeed, he should do so. But, while he may strike hard blows, he is not at liberty to strike foul ones. It is as much his duty to refrain from improper methods calculated to produce a wrongful conviction as it is to use every legitimate means to bring about a just one.
Berger v. United States, 295 U.S. 78, 88, 55 S.Ct. 629, 79 L.Ed. 1314 (1935), overruled on other grounds by Stirone v. United States, 361 U.S. 212, 80 S.Ct. 270, 4 L.Ed.2d 252 (1960). In short, that the prosecutors’ misconduct was so incongruous with their duties buttresses our conclusion that this is the rare case involving Brecht error.
The government rejects that conclusion and contends that prejudice must be proven here, citing cases concerning pretrial publicity, jury selection and extraneous influence on impaneled jurors. See, e.g., Skilling v. United States, 561 U.S. 358,130 S.Ct. 2896, 177 L.Ed.2d 619 (2010); Sheppard v. Maxwell, 384 U.S. 333, 86 S.Ct. 1507, 16 L.Ed.2d 600 (1966); Rideau v. Louisiana, 373 U.S. 723, 83 S.Ct. 1417, 10 L.Ed.2d 663 (1963). Initially, we note that under Fed. R.Crim. Proc. 52(a), the gov-*355eminent bore the burden of proving no prejudice. Moreover, the cases cited by the government are of dubious relevance factually and legally. First, the government cannot obstruct the inquiry into online activity and then claim that there is insufficient proof to support the district court’s findings. Cf. United States v. Derrick, 163 F.3d 799, 809 (4th Cir.1998) (where prosecutorial misconduct largely consisted of discovery violations, and violations had been corrected, any prejudice that might have existed was fully remedied by a new trial order). Second, the true extent of the online misconduct has not yet been fully revealed, and there are strong implications in testimony by Magner and the interviews of Jan Mann and Dobinski and from OPR materials that the online commenting was widely known within the New Orleans USAO and known within DOJ.25 Third, the DOJ’s sluggish approach to uncovering and revealing the extent of postings to the district court suggest a prosecution insensitive — at best — to what it now acknowledges are strict' governing legal and ethical standards. Fourth, contrary to the government’s assertions, the anonymous nature of the comments does not reduce but increases their pernicious influence. Fifth, cases that concern outside pretrial publicity perpetrated by the press, e.g. Sheppard v. Maxwell and Rideau v. Louisiana, present neither prosecutorial misconduct nor unquantifiable influences on the proceedings. Sixth, the online comments could have affected not only the venire panel and actual jurors but also cooperating defendants and defense witnesses. Finally, the inevitable impression left by the government’s misconduct and ongoing pettifoggery is of a prosecution determined to convict these defendants by any means.
In sum, while a demonstration of prejudice is ordinarily a prerequisite for the grant of a new trial, the Supreme Court specifically identified the type of extraordinary errors that will dispense with this burden.26 Such errors occurred here. Prosecutorial misconduct commenced even before indictments were handed down and continued throughout trial and into the post-trial proceedings, and that misconduct affected the prosecution and trial in ways that cannot be fully evaluated due to the government’s mishandling — to put it politely — of the investigation into eyberbullying. The online anonymous postings, whether the product of lone wolf commenters or an informal propaganda campaign, gave the prosecution a tool for public castigation of the defendants that it could not have used against them otherwise, and in so doing deprived them of a fair trial. The district court’s steady drip of discoveries of misconduct infecting every stage of this prosecution, combined with the government’s continued obfuscation and deceit, renders this the rare case in which imposition of the Brecht remedy is necessary.
B. Prejudice Proven
Alternatively, even if a finding of prejudice is required here, the district court did not err in finding that the gov*356ernment’s misconduct in this prosecution prejudiced the defendants.
Although defendants frequently seek mistrials alleging prosecutorial misconduct, their motions are rarely granted. Even when a district court finds that misconduct occurred, it must also normally find that the misconduct in question actually prejudiced the defense. United States v. Bowler, 252 F.3d 741, 747 (5th Cir.2001). Prejudice, in turn depends on the extent to which the particular misconduct contributed to a guilty verdict. Id. The online activities here were viewed by Perrieone as a “public service,” designed among other things to influence the community and put pressure on the various defendants who were being pursued by the AUSO. Dobinski also characterized her encouragement of other commenters about the ongoing trial as a public service. The district court was unable to capture the extent of prejudice during jury selection or trial because the tainted source of the comments had not yet been revealed. On reviewing jury questionnaires for the new trial motion, however, the district court found that seven of twelve seated jurors had visited the Nola.com website in the months preceding trial. Further, according to the district court’s review, jurors who visited the website appeared to have a lower opinion of NOPD officers’ honesty. Cf. Harbin, 250 F.3d at 545 (prejudice presumed despite no proof of bias by replacement juror). From a practical standpoint, the defendants were prejudiced because the district court found its investigation of jury selection process could not be effectively pursued years later. See, e.g., District Court’s Order and Reasons, December 12, 2013, at 22-23 (explaining why an inquiry into juror bias so long after trial, is unworkable).
Additionally, the district court believed that the government pressured cooperating defendants to seek plea deals and then to shade their testimony against the others. The district court reiterated that government. threats of perjury charges against defense witnesses, which had never materialized, prompted several not to testify at the trial. If prosecutorial misconduct must spawn prejudice that is shown to be outcome-determinative, then the congeries of overbearing activities in this trial might not meet that goal.- On the other hand, the facts that the government engaged in misconduct, which took place off the record but in public, and that the misconduct was directed at the public but defies investigation because of the government’s tactics, should tip toward a finding of prejudice. On-the-record misconduct can be easily evaluated; the misconduct here cannot. Furtive misconduct should not escape remedy simply because it was furtive. See Sipe, 388 F.3d at 477 (affirming new trial grant for cumulative prosecutorial errors and omissions).
A prosecutor’s status, moreover, enhances the credibility of public comments and magnifies the adverse consequences of the commenter’s inappropriate remarks. The prosecutor’s comments implicate his or her inside knowledge of prosecutorial activity as they explain the significance of particular events during a case. Bias or vindictiveness in the prosecutor’s comments, reflected repeatedly in Perricone’s postings, cast doubt on the integrity of the process, as did Jan’s Mann’s online questioning of the district court’s motives in a related Danziger Bridge prosecution. Dobinski’s contributions encouraged and approved one-sided reports about the trial. All of these experienced, high-level prosecutors were well aware that they were forbidden, legally and ethically, from making in public the statements they communicated online. They all knew that em*357ployment sanctions should be imposed for their activities if undertaken publicly.
'Although the government does not deny the impropriety of online anonymous comments about pending cases, it downplays their prejudicial effects in several ways. First, the government argues that anonymity diminishes the cloak of authority that would otherwise surround the prosecutor’s pronouncements. Because the online community does not know that a prosecutor is speaking, it cannot be adversely influenced by his inflammatory opinions. Second, the extent of the publicity surrounding the anonymous comments is uncertain because no one knows how many people read online comments to the newspaper of record. Third, these comments amounted to no more than voices in a chorus of public opinion on the Danziger Bridge trial and were no more likely to exert an influence than those of any other chorister. These arguments are not insubstantial, but they are outweighed by the insidious nature of prosecutorial anonymity, the growing influence of online communications to mold public opinion in our society, and the danger of mob reactions.
Anonymity provokes irresponsibility in the speaker. A prosecutor may attempt to comment anonymously in a pending case, whether in a bar, on a talk radio show, or online. It is hard to cloak one’s experiences, however, and listeners can easily infer, as a number of readers within the New Orleans USAO evidently did, that someone with “insider knowledge” is making the comments. The speaker thus trades on his air of self-importance and his special knowledge, while imparting a biased and dramatic flair to his anonymous commentary.
Unlike this court’s recent decision in United States v. McRae,27 that there is no “proof’ that members of the venire panel or actual jurors read or were influenced by the online comments exacerbates rather than alleviates prejudice here. Anonymous postings prevent uncovering the extent of improper influence, adding injury *358to the insult of the biased, inflammatory, and improper communications themselves. Had the comments in this case been delivered by the prosecutors without the shield of anonymity, the extent of the harm would have been quantifiable, but their actions eliminated the measurement of harm. Moreover, the government overlooks that potential harm extends not just to jurors but others involved in the case. It is well to assume that the jurors, once impaneled, followed the district court’s instructions not to obtain extrinsic information about the trial. For defendants, for the cooperating defendants who testified for the government, and for defense witnesses, however, there are no such restrictions. Only a naif would think that these people, and their families and friends, were not avidly consuming all available sources of information from the inception of the prosecution through trial. Inflammatory and biased online comments to news articles must have affected the participants’ approaches to their defense, testimony, or decisions, to testify. That there was some influence, although unquantifiable under these circumstances, seems inescapable.
Most pernicious, these attorneys’ online comments knowingly contributed to the mob mentality potentially inherent in instantaneous, unbridled, passionate online discourse. These prosecutors created an air of bullying against the defendants whose rights they, especially Dobinski, were sworn to respect. That they were several among dozens of commenters, some of whom may have disagreed with their views, does not dissipate the effect of this online cyberbullying. Just as a mob protesting outside the courthouse has the potential to intimidate parties and witnesses, so do streams of adverse online comments. The impact is felt not only by the defendants but by codefendants pressed to plead guilty or defense witnesses dissuaded from testifying. Preventing mob justice is precisely the goal of prosecutorial ethical constraints. The government here should not be able to shelter under a banner of “no prejudice proved” while the prosecutors acted no better than, and indeed tried to inflame, the public.
For all these reasons, we conclude that the district court did not err in finding that the defendants were prejudiced by the government’s misconduct. On this basis, too, the defendants are entitled to a new trial.
The government also argues that official and professional discipline were adequate to rebuke Perricone, Jan Mann, and Dobinski and should have sufficed in lieu of a new trial. Like the district court, we disagree. To begin with, whether those who committed misconduct were disciplined simply does not bear on whether the defendants received a fair trial. It is clear from Perricone, Mann, and Dobinski’s testimony, moreover, that none of them is particularly remorseful about the misconduct, and they claimed to believe their individual First Amendment rights were separable from their positions of public trust. Perricone and Jan Mann both resigned from office with benefits as far as the record shows, although they were referred for professional discipline to the State Bar of Louisiana. Dobinski remains in federal employment with only a bare reproof for her online commenting. Then-misdeeds are compounded by the government’s insouciant investigation, which leaves open only three inferences concerning this prosecutorial breakdown: the government is not serious about controlling extracurricular, employment-related online commenting by its officials; the government feared what it might uncover by a thorough and timely investigation; or the government’s investigation was incompetent. Exerting professional discipline on three individual government lawyers does *359nothing to solve the systemic problem, and it is not a sufficient answer to the miscarriage of justice in this case.
II. Remove the Judge?
The government’s final appellate point asks this court to remove Judge Engelhardt from this case, whatever its future course. Removal of a judge is a rare and disfavored order. In re McBryde, 117 F.3d 208, 229 (5th Cir.1997). This court must be persuaded that the judge’s conduct “might reasonably cause an objective observer to question [the judge’s] impartiality.” Johnson v. Sawyer, 120 F.3d 1307, 1333 (5th Cir.1997) (quoting United States v. Microsoft Corp., 56 F.3d 1448 (D.C.Cir.1995) (per curiam)); Liljeberg v. Health Servs. Acquisition Corp., 486 U.S. 847, 864-65, 108 S.Ct. 2194, 2205, 100 L.Ed.2d 855 (1988); 28 U.S.C. § 455(a); 28 U.S.C. § 455(a). Three considerations go into this decision: whether the original judge would reasonably be able on remand to have substantial difficulty in laying aside his previously expressed views or findings that have been declared erroneous; whether reassignment is advisable to preserve the appearance of justice; and whether the efficiency costs involved in reassignment outweigh the benefits to the appearance of fairness. In re DaimlerChrysler Corp., 294 F.3d 697, 700-01 (5th Cir.2002).
Here, there is no basis for removing Judge Engelhardt for conscientiously responding to these novel events as they unfolded. The district court was outright skeptical over the defendants’ initial new trial motion predicated on the Perricone revelations. Far from being a non-neutral arbiter, as the government contends, the district court was pushed into novel territory by ongoing revelations — that Perricone had spoken to the press before he had been questioned by DOJ; that Jan Mann had allowed Letten to lie in the district court’s face, and she had herself been commenting while purporting to investigate the USAO commenting; that Perricone and Jan Mann seemed not credible on salient points; and that Bernstein, chief prosecutor, kept narrowing the scope of misconduct deliberately and artificially. The district court’s failed confidence in the DOJ’s ability to reveal the impact of inappropriate commenting led to the district court’s strong suggestion for an independent DOJ review of Jan Mann’s “investigation.” This action was never challenged in the district court as an invasion of constitutional separation of powers, and the special investigator attempted to cooperate with the district court. The district court carefully preserved the ex parte nature of much of this investigation for several reasons, including the protection of DOJ employees during the pendency of a separate internal OPR inquiry. Finally, because we have not found the district court’s findings erroneous or its conclusions inapt, it would make little sense to remove Judge Engelhardt for doing what was called for under the circumstances.
That Judge Engelhardt expressed his candid views about the government’s highly disparate charges between cooperating defendants and those who exercised their right to a jury trial does not provide grounds for removal. Nor does his use of colorful language in his written opinions merit the severe professional sanction of removal from this prosecution. Judge Engelhardt’s stylistic choices were likely induced when words like “incredible” and “novel” and “unprecedented” were no longer enough to describe the ongoing revela-’ tions of the government’s misconduct and incomplete investigation. Judge Engelhardt’s style is far outweighed by his thorough, fact-driven approach. The government’s motion is meritless.
*360CONCLUSION
For the foregoing reasons, we find no abuse of discretion in the district court’s grant of a new trial to these five defendants. AFFIRMED; REMANDED FOR TRIAL.
. Another defendant's case was severed and separately tried. See United States v. Dugue, — F.3d -.
. Post-trial, the defendants filed an initial motion for new trial within a month, the issues in which are not relevant here.
. Perricone's and later First United States Attorney and Chief of the Office’s Criminal Division Jan Mann’s identities were uncovered by forensic comparison of their characteristic writing styles in the online comments and in court filings. The forensic expert in question had previously assisted the FBI in identifying the Unabomber.
.Letten averred: “In terms of Perricone, Judge, I will tell you right now on the record that ... neither I, nor Jan Mann, nor people in positions of authority in our office, to my knowledge did not have any knowledge of, nor did we authorize, nor did we procure or have any knowledge of Sal Perricone anonymously posting comments about cases or anything like that whatsoever until we learned about it in the filing. That is Gospel truth.”
. In another case being handled by the New Orleans USAO, Jan Mann as lead counsel also stood silent while the court repeated Letten’s assurances that only Perricone had cornmented unethically. United States v. Broussard, No. 2:11-CR-299, 2014 WL 3489725 (E.D.La. July 14, 2014), aff'd, 595 Fed.Appx.441 (5th Cir.2015).
. Defense counsel had identified Michael Magner. Magner, who had left the USAO, suggested that several people in the office had been posting comments or, contrary to Let-ten's statement, had either known about or strongly suspected Perricone was doing so. Magner also stated that Jan Mann's husband Jim, another AUSA, was Perricone’s closest professional friend, and their offices were next door to each other.
. At the time, Jim Mann was the supervisor of the USAO’s Financial Crimes and Computer Crimes Unit.
. Much of the information provided in the "Horn reports” remains under seal, but all of the information has been reviewed by this court on appeal. The Horn investigation was duplicated simultaneously by an OPR internal DOJ inquiry, which had begun in July 2012 and was completed in December 2013, following the district court’s grant of a new trial. The court’s September 2013 opinion granting a new trial refers to the otherwise confidential OPR investigation only when certain materials OPR developed had been revealed to the district court.
. Again, this recitation simply summarizes a 16-page excerpt from the district court’s new trial order, to which the reader is referred for complete details. September Order, 969 F.Supp.2d at 554-67.
. In fact, Jan Mann had not been asked, because her attorney said she would not answer, whether her answers to the original OPR investigatory questions, which had been posed to the employees of the USAO in July 2012, “would have changed,” after the revelation of her online activity.
. The interview was conducted in the presence of an FBI agent, rendering Ms. Mann potentially subject to prosecution under 18 U.S.C. § 1001, but as the district court noted, the DOJ had an inherent conflict of interest in selecting this mode of procedure, for DOJ was not likely to prosecute her while attempting to protect the Danziger Bridge verdicts.
.Officer Bowen testified before a grand jury pursuant to a court order granting him use immunity. Dobinski’s duty was to ensure that during the later federal prosecution, this order was not violated. Kastigar v. United States, 406 U.S. 441, 92 S.Ct. 1653, 32 L.Ed.2d 212 (1972); Garrity v. New Jersey, 385 U.S. 493, 87 S.Ct. 616, 17 L.Ed.2d 562 (1967).
. See the district court’s Order and Reasons dated December 12, 2013 at pp. 8-12, which quotes Dobinski's vague answers. Just one example:
Q: But is it correct that um that you’re not saying you’re a hundred percent sure there were no other postings? If we found additional postings, it is not going to contradict your position here today?
A: Right.
. Of course, were the district court’s decision to be reversed, it could revert to holding an evidentiary hearing to pursue further the details of the misconduct. In its footnote 127, the district court identifies some of the outstanding issues:
As HemyL.Mencken — 1951, Perricone made two very correct statements: "There are NO secrets in New Orleans....” [record citation], and "Perhaps the truth will surface. God knows, we need more truth in this city.” [record citation]. Yet the Court, even with the capable assistance of Mr. Horn and Ms. Alexander, remains unaware of: the identity of the source of the "leak” of the Lohman plea, in possible violation of Rule 6(e); the earlier user IDs of Perricone and Jan Mann, and how many other user IDs might be involved; the identities of ten of the eleven user IDs that were the subject of the DOJ administrative subpoena ...; the information which could have been obtained from a forensic recovery of computer data .for years 2010 and 2011 (prior to December 19, 2011) by the DOJ; whether USA Letten agrees with his former First Assistant that she disclosed her online posting activity in March 2012, and whether that information was conveyed to others in DOJ over six months prior to its public disclosure in early November 2012; whether Karla Dobinski’s posting was known to others at DOJ; whether “123ac” is an AUSA; whether other government agency employees posted comments on this case, as did government agency employee "A” and whether any of the three defense witnesses who refused to testify at trial will ever be charged with the crimes for which they were threatened prior to this tried. These are but a few of the outstanding issues which would be the subject of an evidentiary hearing. It is anticipated that surely others would surface upon learning further information at such a hearing.
September Order, 969 F.Supp.2d at 624 n. 127.
. Hunter drove the truck to the Danziger Bridge and shot at Ronald and Lance Madison, then testified as a "key” government witness.
. The court added significant excerpts from OPR materials, plus a caution, with which we agree, that OPR materials still under seal are also revelatory.
. The government alleges clear error only in regard to the district court's disbelief of taint team leader Dobinski's explanation why she resorted to Nola.com to be informed about the trial proceedings. Given the nature of her misconduct and its online effect, the district court’s finding is lagniappe. The government’s challenges to the overall significance of the online postings on jury selection, trial tactics, effect on witnesses and defendants are more on the order of mixed questions of fact and law, see United States v. Wall, 389 F.3d 457, 465 (5th Cir.2004), and are discussed as such infra.
.See 28 C.F.R. § 50.2, e.g., § 50.2(b)(6)(i), (vi): "Observations about a defendant's character” and "[a]ny opinion as to the accused's guilt” will "generally tend[] to create dangers of prejudice without serving a significant law enforcement function.”
. See United States Attorneys' Manual §§ 1-2.401(E), l-7.550(f).
. See Rules 53.2, 53.3, 53.5, Local Criminal Rules of the United States District Court for the Eastern District of Louisiana; Rule 8.2, Louisiana Rules of Professional Conduct.
. See 28 C.F.R. § 50.2(b)(5); USAM §§ 1-7.401 E, G; Local Rule 53.5.
. The district court also cited its supervisory power. This ground for relief is disputed by the government, and we neither discuss nor rely on it.
. See, e.g., Hassine, 160 F.3d at 961 (Doyle error not egregious enough to warrant footnote nine exception); Hardnett, 25 F.3d at 880-81 ("the prosecutor’s misconduct did not infect the whole trial”); Cupit, 28 F.3d at 538 (unconstitutional admission of hearsay is “classic trial error[]”).
. This is not a conclusion that our "supervisory duty” supports a new trial, but a reflection on the prosecution’s cynical minimization of its wrongdoing. Cf. Rideau, 373 U.S. *354at 728, 83 S.Ct. at 1420 (Harlan, J., dissenting).
. The court also found incredible Perricone’s and Mann’s denials that each knew about the other’s postings. Too often, each of them posted in reference to the other’s observations and unfailingly reinforced their prejudiced views.
. Although Brecht itself concerned habeas relief, "similar or even broader exceptions to the harmless error doctrine should logically apply at the district court level or on direct appeal, which do not involve the especially deferential habeas standard of review.” Sonja B. Starr, Sentence Reduction as a Remedy for Prosecutorial Misconduct, 97 Geo. LJ. 1509, 1561 (2009).
. In United States v. McRae, No. 14-30995, 795 F.3d 471, 2015 WL 4542651 (5th Cir. July 28, 2015), this court recently denied a new trial request to a former New Orleans police officer convicted of different crimes in the wake of Hurricane Katrina. The court found no actual or presumed prejudice attending online postings about the case by Sal Perricone. Id. at 481-83, 2015 WL 4542651 at *7. That decision is distinguishable for two reasons. Most important, McRae argued only "that the court should presume prejudice where deliberate and egregious government misuse of the media is combined with extensive pretrial publicity adverse to the defendant.” He relied principally on the Supreme Court’s decision in Skilling v. United States, 561 U.S. 358, 130 S.Ct. 2896, 177 L.Ed.2d 619 (2010), which we have distinguished, and not on Brecht’s identification of "hybrid error.” Indeed, the McRae court does not address Brecht.' Second, unlike in this case, the McRae court was not tasked with attempting to uncover the extent of press leaks or government online commenting, nor was it obstructed in doing so by government delays, nor was there cumulative evidence of high-handed prosecutorial tactics, nor was there evidence that members of the jury may have been exposed to the online commenting before the trial, nor were the court’s ultimate conclusions founded in grave uncertainty about the extent of government misconduct or the impact on the trial.
Finally, McRae and this case share an important characteristic: each decision affirms the trial court's exercise of its discretion to determine whether the "interest of justice” demands a new trial under Rule 33. See United States v. Wall, 389 F.3d 457, 465 (5th Cir.2004) (appellate review of district courts Rule 33 decision "is necessarily deferential to the trial court”). The steady trickle of troubling revelations about the ongoing misconduct here undermined "confidence in the jury verdict” even without an explicit connection between the comments and the jury, or between the prosecutors and the case team, that McRae viewed as indicative of prejudice.