concurring in the judgment:
The panel does not reach the merits of Brice’s claim that the district judge’s “impartiality might reasonably be questioned,” a claim that if correct would have required the district judge to recuse herself. See 28 U.S.C. § 455(a). It rests on two propositions: first, the rule created in United States v. Barrett, 111 F.3d 947, 951 (D.C.Cir.1997), that failure to call for recu-sal “within a reasonable time after the grounds for it are known” waives any claim under § 455(a); second, the “general rule of appellate procedure” that where an argument could have been raised on an earlier appeal, it is inappropriate to consid*1291er the argument in a later appeal following a remand, at least absent “exceptional circumstances,” United States v. Henry, 472 F.3d 910, 913 (D.C.Cir.2007). I agree that we are bound by Barrett, though I can find no logic behind its silent choice that this particular omission of counsel must be classified as waiver (entailing no review at all), rather than forfeiture (allowing review for “plain error”). And I believe the circumstances are exceptional enough that counsel’s omission on the prior appeal should not prevent review of the current claim.
The substantive claim here is one of bias, and its facts are surely exceptional. The key event was the initial appearance of a prosecution witness, an appearance designed by judge and prosecutor — in an ex parte sidebar — to generate both pathos and sympathy for the witness. The witness, known as K.H., was among the women that the defendant evidently controlled in the course of the prostitution offenses for which he was ultimately found guilty. She and two others had been held as material witnesses, after material-witness proceedings conducted by the district judge who handled the trial. Because of concern over KH.’s mental health, she was to be voir dired outside the presence of the jury. But the Assistant U.S. Attorney spied a chance for more impact, as he explained in the ex parte sidebar:
I was hoping that there was some way that the jury could see [the witness] come in to see the defendant for the first time because I anticipate that there’s going to be a reaction because she’s so in love with him and when she saw the photo spread, she sobbed, I was there.
But it sounds like if you were going to do the voir dire that will be beforehand outside of the jury and the defendant present.
Tr. of Feb. 21, 2006, at 203. The district court obliged, helpfully suggesting the following: “You could call her as a witness, she could enter the courtroom. Have me excuse the jury and do the voir dire. Then we can just do it in that order.” Id. at 204.
The district court then set out to execute the plan. The court called the marshal over and said that the witness, who was then detained, should be treated “more like a victim than a criminal,” id., and then laid out the proposal agreed on with the prosecutor. The court deputy marshal responded by noting that to bring a detained witness through the front door was a policy deviation that would require the approval of his chief. He proposed an alternative, but the judge insisted on the original plan: “It’s important that she come in and that the jury see her and the defendant the first time that they see each other.” Id. at 206. To be sure that it came off, she said she would speak to the supervisor. Evidently she did so, and the staged entry proceeded just as the court and prosecutor had planned.
The entry evidently did not strike defense counsel as odd enough to trigger an inquiry or objection. Even the government doesn’t claim that counsel’s inaction at trial precludes review here; rather it rests on the fact that in the first appeal (not handled by trial counsel), the transcripts available to counsel included the text of the ex parte sidebar.
Besides arranging the presentation of K.H., the district court on other occasions showed some hostility to the defendant. Standing alone, these expressions might not amount to much. In dismissing a defense contention during one pre-trial hearing, the judge referred to Brice as “the criminal,” but then instantly corrected herself — “or alleged, forgive me, alleged criminal.” Tr. of Feb. 14, 2006, at 26. The *1292next day, in a discussion of the material witness’s mental health and drug use, the defense observed that the witness had a history of drug use. To this, the district court responded, “Well, you should ask perhaps Mr. Brice about that,” Tr. of Feb. 15, 2006, at 8, suggesting that any drug use by the witness could be chalked up entirely to Brice, a government claim that the defense contested. The thread of hostile pretrial comments, together with other concerns no longer pressed, prompted the defense to move to recuse the district judge, Tr. of Feb 17, 2006, at 6, a motion later withdrawn after the judge offered assurances of impartiality, id. at 16-17. (Counsel may have had in mind Machiavelli’s famous caution — “Never strike at a king except to kill.”)
The doctrinal obstacle to our consideration of Brice’s current claim is Barrett’s rule that a party waives any objection to the judge’s appearance of bias if he fails to raise the issue “within a reasonable time after the grounds for [disqualification] are known.” Ill F.3d at 951. As the Barrett court expressly ruled against Barrett’s bias claim on the merits, the waiver theory was quite unnecessary, as Judge Tatel noted in his concurring opinion. Id. at 954. But because the court appeared to rest the outcome in part on the waiver theory, we are obliged to treat it as an alternative holding rather than mere dictum. Woods v. Interstate Realty Co., 337 U.S. 535, 537, 69 S.Ct. 1235, 93 L.Ed. 1524 (1949) (“[Wjhere a decision rests on two or more grounds, none can be relegated to the category of obiter dictum.”).
Barrett contains several curious features. First, in calling defendant’s delay a “waiver” of defendant’s claim under 28 U.S.C. § 455, it completely overlooks what § 455 has to say about waiver. Subsection 455(e) forbids a judge to “accept” a waiver of any of the grounds set out in § 455(b), whereas for § 455(a), dealing with any instance where the judge’s impartiality might “reasonably be questioned” (the subsection relevant to our case), § 455(e) allows “waiver [to] be accepted provided it is preceded by a full disclosure on the record of the basis for disqualification.” Before us, the parties haven’t argued the matter at all. At a casual first glance, however, the requirement of “a full disclosure on the record” conjures something far more deliberate and elaborate than mere delay, coupled with counsel’s imputed awareness of the transcript of the ex parte dealings.
Other courts have recognized § 455(e)’s strictures on waiver yet gone on to insist on timeliness, i.e., to treat delay as effecting a de facto waiver. United States v. York, 888 F.2d 1050 (5th Cir.1989), offers perhaps the most extensive justification. (Barrett, treating delay and waiver as interchangeable yet not mentioning what § 455(e) had to say on waiver, evidently saw no need for reconciling its delay rule with the statute.) The York court acknowledged that § 455(e)’s bar on waiver of § 455(b) violations “suggests that Congress believed the gain in protecting against actual bias, prejudice, or conflict of interest outweighs the loss to judicial economy in prohibiting waivers.” Id. at 1055. And the court further observed that the “motivation behind a timeliness requirement is also to a large extent one of judicial economy.” Id. The “also” is a puzzler, as the language of § 455(e) on its face made the values protected by § 455(b) trumps over judicial economy. No matter. The court went on to offer this rationale for a timeliness requirement:
[T]he gains in judicial economy from a timeliness requirement are greater than those from permitting waiver. Since both parties must agree to any waiver, no new trial will be saved by waiver *1293once the outcome of trial has been determined. In fact, once any party senses that the proceedings have been favorable to it up to that point, no waiver is likely to occur. On the other hand, a timeliness requirement will proscribe motions that would have invalidated a fully completed trial.
Id. In other words, the gain in judicial economy from a timeliness requirement exceeds the hypothetical gain from allowing waiver, so it is reasonable, the court thought, to suppose that Congress was not ruling out a timeliness requirement. This is true, of course, to the extent that one focuses exclusively on waivers after the litigation outcome is known or at any rate heavily foreshadowed: in those cases, waiver achieves no judicial economy at all, as the loser, having little or no incentive to preserve the outcome, will not waive. But as waivers normally will not occur in those circumstances at all, it seems very doubtful that the scenario played any role in Congress’s resolution of the balance. Moreover, the argument does little to refute the rule’s apparent anomaly: while recognizing that deliberate waiver of § 455(b) values is impossible, it allows an easy loss of those values through mere neglect. And while deliberate waiver of § 455(a) is possible but seemingly very difficult, occurring only — so far as appears on the text of the statute — through a rather formal ceremony, the timeliness rule makes loss easy through neglect. As an absolute bar, a timeliness requirement of course entirely ignores the purpose of § 455(a), which is “to promote public confidence in the impartiality of the judicial process.” H.R.Rep. No. 93-1453, at 5 (1974), 1974 U.S.C.C.A.N. 6351, 6355.
Many courts have nonetheless accepted such a timeliness requirement. See, e.g., Kolon Indus. Inc. v. E.I. DuPont de Nemours & Co., 748 F.3d 160 (4th Cir.2014); United States v. Brinkworth, 68 F.3d 633, 639 (2d Cir.1995); United States v. Owens, 902 F.2d 1154, 1156-57 (4th Cir.1990); United States v. Nobel, 696 F.2d 231, 236-37 (3d Cir.1982); Delesdernier v. Porterie, 666 F.2d 116, 121 & n. 3 (5th Cir.1982). The 7th Circuit initially read § 455(e) limits on waiver as barring any timeliness requirement, SCA Servs., Inc. v. Morgan, 557 F.2d 110, 117 (7th Cir.1977), but then noted in dictum a readiness to rethink the matter, Union Carbide Corp. v. U.S. Cutting Senice, 782 F.2d 710, 716-17 (7th Cir.1986).
Not only Barrett but the other cases insisting on timeliness lay great stress on a concern — which to be sure is plausible— that a party might “take his chances” with a judge, and then raise the recusal issue if unhappy with the outcome. Barrett, 111 F.3d at 951 (internal quotation marks omitted). But neither Barrett nor the others explains why that risk is so great in connection with § 455 that an absolute bar is the solution, rather than, as for all other rulings or omissions not challenged until appeal, merely limiting relief to review for plain error.
That omission leads directly to another frailty of Barrett: it completely disregards the distinction between waiver and forfeiture drawn by the Supreme Court’s decision in United States v. Olano, 507 U.S. 725, 113 S.Ct. 1770, 123 L.Ed.2d 508 (1993). Because waiver is “ ‘the intentional relinquishment or abandonment of a known right,’ ” id. at 733, 113 S.Ct. 1770 (citing Johnson v. Zerbst, 304 U.S. 458, 464, 58 S.Ct. 1019, 82 L.Ed. 1461 (1938)), a waiver (if valid and effective) “extin-guishfes]” the erroneous character of the relevant ruling. Id. By contrast, mere forfeiture has no such effect; an error, despite the absence of a timely objection, remains an error for purposes of Rule 52’s provision for review of “plain error.” Id. *1294at 733-34, 113 S.Ct. 1770; accord United States v. Laslie, 716 F.3d 612, 614 (D.C.Cir.2013); In re Sealed Case, 356 F.3d 313, 317 (D.C.Cir.2004). Barrett neither recognizes that plain error review is an option, nor offers an explanation as to why it is not enough of a gaming deterrent in the context of § 455(a). The same objection of course applies to the decisions in other circuits that insist on timeliness but do not call it waiver.
In fact, many courts apply plain error review to § 455(a) claims unchallenged at trial. See, e.g., United States v. Berger, 375 F.3d 1223, 1227 (11th Cir.2004) (reviewing under plain error after failure to seek recusal below); United States v. Kimball, 73 F.3d 269, 273 (10th Cir.1995) (same); United States v. Franklin, 197 F.3d 266, 270 (7th Cir.1999) (noting the “specter of ‘sand bagging’ ” and applying plain error review as a result); Baldwin Hardware Corp. v. FrankSu Enter. Corp., 78 F.3d 550, 557 (Fed.Cir.1996) (applying plain error); United States v. Schreiber, 599 F.2d 534, 535 (3d Cir.1979) (noting the concern about gaming and applying plain error standard as a result); see also Noli v. Comm’r, 860 F.2d 1521, 1527 (9th Cir. 1988).
Indeed, even the government may have no faith in the waiver theory, or even waiver in timeliness’s clothing. In its briefing here it makes no claim of either version and offers no citation to Barrett. It rather argues, in alignment with the many circuits applying more standard remedies for an omission by counsel, merely that we resolve the issue under a plain error standard. Resp. Br. at 17; see also Brief for United States at 20-21, United States v. Lang, 364 F.3d 1210 (10th Cir. 2004) (No. 02-4075).
Were it not for Barrett, we would almost certainly regard Brice’s failure to raise the issue earlier as forfeiture, not waiver, and we would review under plain error (absent other obstacles). On the facts of this case, I believe the impartiality of the judge “might reasonably be questioned,” § 455(a), even if reviewed under the plain error standard. As Brice doesn’t challenge the trial outcome but asks only for resentencing, it is governed by our rule cutting more slack for assertions of plain error when only sentencing is at stake. United States v. Saro, 24 F.3d 283, 287-88 (D.C.Cir.1994).
The panel opinion also cites a second, independent reason for withholding review. Under conventional appellate procedures, we do not consider arguments raised for the first time on a second appeal if they might have been raised on initial appeal. Though generally true, this is “a prudential rule rather than a jurisdictional one,” Crocker v. Piedmont Aviation, Inc., 49 F.3d 735, 739-40 (D.C.Cir.1995), motivated by a “practical concern for judicial economy,” id. at 740. Accordingly, we “always possess! ] discretion to reach” issues not raised on initial appeal, though this discretion “is normally exercised only in exceptional circumstances, where injustice might otherwise result,” id. (internal quotation marks omitted); cf. U.S. National Bank of Oregon v. Independent Ins. Agents of America, Inc., 508 U.S. 439, 447, 113 S.Ct. 2173, 124 L.Ed.2d 402 (1993). The ex parte cooperation of the court and prosecutor in this case certainly strikes me not only as “exceptional” but also as creating exceptional circumstances. See Yesudian ex rel. United States v. Howard Univ., 270 F.3d 969, 971 (D.C.Cir.2001) (exercising such discretion, and noting that the bar presented by a “failure to raise an issue in an initial appeal is far from absolute”).
So, contrary to the majority I do not view our appellate procedures as controlling the outcome of this case. Rather, we *1295have a straightforward application of Barrett, and are therefore bound to follow it, however much it may be in tension with 28 U.S.C. § 455, with Olano, and with our standard treatment of claimed errors not raised in district court.