dissenting in part.
With the exception of subsection II.B.3, I agree with the majority’s opinion in its entirety. I cannot agree, however, with the majority’s conclusion that the Miranda warning administered to Rodríguez-Pre-ciado on the night of June 26 was still effective the following afternoon.1 By the time of the interrogation in question, Rod-ríguez-Preciado had been moved twice since first being advised of his rights, and was then incarcerated at the local county jail. The combination of the lapse in time and the change in custodial circumstances is sufficient that a suspect could well question whether the same rights still obtained.
I therefore believe that we must address the midstream Miranda warning given to Rodríguez-Preciado during his interrogation on June 27. The propriety of those warnings — and of the interrogation elicited thereafter — turns on what rule, if any, the fractured Supreme Court handed down in Missouri v. Seibert, — U.S. -, 124 S.Ct. 2601, 159 L.Ed.2d 643 (2004). As I explain below, because I would follow the reasoning of the Seibert plurality, I would hold that the district court erred in not suppressing the statements Rodríguez-Preciado made on June 27 and the evidence discovered as a result of those statements.
The district court’s error in this regard may well have been harmless beyond a reasonable doubt. ; The government, however, has not pressed that position here. At least one previous panel of this circuit has suggested, albeit in passing, that the failure to raise harmless error constitutes waiver. See United States v. Vallejo, 237 F.3d 1008, 1026 (9th Cir.2001). Though some of our sister circuits have recognized limited circumstances in which harmless error may be considered sua sponte, see, e.g., United States v. Rodriguez Cortes, 949 F.2d 532, 542-43 (1st Cir.1991). I do not believe that this is such a case, for reasons I elaborate upon below. I would therefore reverse Rodriguez-Preciado’s conviction and remand for • a • new trial.
I.
As the majority recognizes, “[t]he courts have generally rejected a per se rule as to when a suspect must be readvised of his rights after the passage of time or a *1134change in questioners.” United States v. Andaverde, 64 F.3d 1305, 1312 (9th Cir.1995); see also United States v. Ross, 123 F.3d 1181, 1188 (9th Cir.1997) (“The case-law does not delineate how long Miranda warnings protect a defendant or at what point that protection evaporates.”). Instead, we are charged with looking at the totality of the circumstances in each individual case. See Wyrick v. Fields, 459 U.S. 42, 48-49, 103 S.Ct. 394, 74 L.Ed.2d 214 (1982) (per curiam).
In framing our totality-of-the-circumstances inquiry, the operative question is whether a reasonable defendant in Rodriguez-Preciado’s position would consider that the Miranda warning given to him the night before at the Satellite Motel still applied to anything he said the following day, after he had been booked and lodged at the county jail. In this case, the majority and I agree that the issue is whether Rodriguez-Preciado could have reasonably believed that the Miranda rights — including the right to remain silent and to consult with counsel — of which he was apprised the night before did not still obtain on June 27, in light of the changed circumstances. We disagree, however, over the controlling import of Rodriguez-Preciado’s statement that he “thought he had” received the Miranda warnings the night before. The majority holds that Rodriguez-Preciado’s statement “indicates that he did, in fact, understand that his rights had not materially changed notwithstanding the change in circumstances.” Ante at 1130. In contrast, I believe that the statement says nothing about whether Rodriguez-Preciado understood the earlier warnings to apply to the current circumstances. And, in my view, the totality of the other circumstances weigh in favor of the conclusion that the police should have re-advised Rodriguez-Preciado before beginning questioning on June 27.
The majority relies heavily on our decision in Andaverde, where we held that a one-day time gap was not enough, on its own, to vitiate the continuing effectiveness of the first Miranda warning, see 64 F.3d at 1313, and on Guam v. Dela Pena, 72 F.3d 767 (9th Cir.1995), in which we held that a fifteen-hour interval between a Miranda warning and a custodial interrogation was acceptable, id. at 769-70.2 In Dela Pena, however, the court emphasized that the relevant time difference was not the fifteen hours that elapsed between the Miranda warning and the interrogation, but the far shorter time period that elapsed between the end of the first, noncustodial interview (at the beginning of which the Miranda warning had been given) and the custodial interrogation. See id. at 769. Moreover, neither precedent involved other changed circumstances to the degree presented here.
In Dela Pena, the central issue before the court was whether Miranda warnings given to a suspect before he was “in custody” needed to be repeated once the police began a custodial interrogation. Two facts were central to the court’s decision that they did not: First, the police reminded Déla Pena of his waiver of his Miranda rights before the custodial interrogation began. See id. The court held that “it *1135was not necessary to repeat the earlier Miranda warnings,” id. at 769 n. 1 (emphasis added), yet it stressed several times the extent to which the officers reminded Dela Pena of his earlier waiver. See, e.g., id. at 770.
Second, the lapse of time was the only significant changed circumstance. Although Dela Pena was not in custody at the time he was first' given his Miranda warnings, he was in the police station, the same place where the custodial interrogation eventually took place. .Id. at 770 (“Other than this passage of time, Dela Pena points to nothing to suggest the effectiveness of the earlier Miranda warnings was diminished.”). The same interrogator questioned him in the same room, and, though fifteen hours elapsed from the initial provision of the Miranda warnings to the beginning of the custodial interrogation, only six-and-a-half hours elapsed between the two interviews. See id. at 769 (noting that questioning ended at 4:00 a.m., and recommenced at 10:35 that same morning). Here, roughly fourteen hours elapsed between the end of the first interrogation and the beginning of the second.3
Similarly, in Andaverde, the defendant was first questioned by a.police officer, who properly Mirandized him. Shortly thereafter, he was questioned by a probation officer, who did not re-advise Andav-erde of his Miranda rights. The same probation officer interviewed Andaverde again the following day, and, though the opinion is not lucid on this point, it appears that both interviews took place in the same location.4 Relying — somewhat erroneously — on Jarrell v. Balkcom, 735 F.2d 1242, 1254 (11th Cir.1984),5 the court held that the probation officer was not required to re-advise Andaverde of his Miranda rights initially, since “the two interrogations were an uninterrupted sequence of events,” Andaverde, 64 F.3d at 1312, and that the orié-day6 interval, standing alone, was insufficient to render the original warning ineffective, id. at 1313.
In Andaverde, then, as in Dela Pena, the only material consideration detracting from the continued effectiveness of the Miranda warnings was the time between the interviews. Even at their broadest, Andaverde and Dela Pena, taken together, thus hold that an interval of up to one day between questioning does not necessarily require officers to re-advise suspects of their Miranda rights. At the same time, neither Andaverde nor Dela Pena embraces the opposite holding — that an interval of up to one day necessarily validates a *1136later interview, regardless of other circumstances.
By contrast, at least two district courts have suggested that re-administration of Miranda would have been necessary in a case presenting facts analogous to those here. Of perhaps most interest is the Middle District of Pennsylvania’s decision in United States v. Vasquez, 889 F.Supp. 171 (M.D.Pa.1995), in which the court applied a five-factor test to determine whether a statement made after a significant delay since Miranda warnings were given should be admissible. Borrowing from the Pennsylvania Supreme Court’s decision in Commonwealth v. Hughes, 521 Pa. 423, 555 A.2d 1264 (1989), Vasquez spelled out five considerations:
“(1) the time lapse between the last Miranda warnings and the appellant’s statement; (2) interruptions in the continuity of the interrogation; (3) whether there was a change of location between the place where the last Miranda warnings were given and the place where the appellant’s statement was made; (4) whether the same officer who gave the warnings also conducted the interrogation resulting in the appellant’s statement; and (5) whether the statement elicited during the complained-of interrogation differed significantly from other statements which had been preceded by Miranda warnings.”
Vasquez, 889 F.Supp. at 177 (quoting Hughes, 555 A.2d at 1276); see also id. at 178 (“[W]e believe [this list] provides an excellent barometer against which to at least begin the review of the circumstances in a particular case.”).7
A host of district courts have since relied on the Vasquez factors. In particular, the Eastern District of Michigan applied Vasquez to a case factually similar to this one, concluding in United States v. Jones, 147 F.Supp.2d 752 (E.D.Mich.2001), that statements from a subsequent interrogation eighteen hours after the Miranda warnings, by a different law enforcement officer, and in a different place, should not be admissible: *1137Jones, 147 F.Supp.2d at 761-62 (citations omitted).
*1136More than eighteen hours had passed ... between the Miranda warnings that Sgt. Payer gave Defendant on February 14 and Defendant’s interview with Agent Kendall on February 15. During that period, there was an interruption in the continuity of the interrogation, as Sgt. Payer had ceased his questioning at 1:13 a.m. on February 15 and Agent Kendall did not begin his interview earlier than 6:00 p.m. There was a change in location between the interview that Sgt. Payer conducted and the interrogation done by Agent Kendall. The former was at the police station; the latter was at the federal building. Agent Kendall, who conducted the interview of Defendant on February 15, was not the same officer who administered the warnings on February 14. All of these factors militate toward the conclusion that Defendant could not fully appreciate a waiver of his Miranda rights when he spoke to Agent Kendall on February 15 and that his statements to Agent Kendall must therefore be suppressed. The Court apprehends no factors militating toward the opposite conclusion.
*1137What is notable about Vasquez and Jones is that in both cases, the court looked at' the various factors from the perspective of the suspect, centering on whether he could reasonably believe that the first warning did not still have force— that is, on whether the original warning was still fully effective. As discussed in more detail below; this is the core principle enunciated, albeit -in a different context, by the Supreme Court plurality in Seibert. Interpreting this case through the reasonableness lens, and applying the Vasquez factors, I conclude that a suspect could reasonably believe that the rights of which he was earlier advised no longer obtained.
First, fourteen hours passed between the end of the second June 26 interview by Officer Haseall and the beginning of Rodriguez-Preciado’s June 27 interview by Sergeant Romanaggi. During this period, there was a significant interruption in the interrogation, as Officer Haseall had concluded his two separate interrogations of Rodriguez-Preciado the night before. There was a change in location,8 as the June 27 interview was conducted at the county jail, where Rodriguez-Preciado had been booked and lodged the night before. And the interrogating officer, Sergeant Romanaggi, was not the officer who had provided Miranda warnings the night before, nor was he in the motel room when the warnings were given.
On these facts alone, Vasquez and Jones suggest that the Miranda warning given to Rodriguez-Preciado at the Satellite Motel on the night of June 26 was no longer effective on the afternoon of June 27. Add to this Rodriguez-Preciado’s fundamental change in status’ from a suspect arrested in his own motel room to an incarcerated ■defendant taken from his cell in a county jail for interrogation, and the case for suppressing his June 27 statements is all the stronger.
Once a suspect is subjected to pretrial detention — as opposed, for example, to arrested but released while awaiting trial— he is obligated to follovy the directions and orders he is given, declining to do so at his peril. See Bell v. Wolfish, 441 U.S. 520, 537-40, 99 S.Ct. 1861, 60 L.Ed.2d 447 (1979). Under those circumstances, a suspect in Rodriguez-Preciado’s ' position could reasonably doubt the efficacy of the original warning, positing that the same rules might not apply after being booked and incarcerated as applied before he became, in effect, a prisoner. While an individual learned in the law would know otherwise, Miranda warnings ássume that the suspect is a tabula rasa, lacking knowledge of his or her legal rights. To that end, Miranda provides prophylactic warnings só as to avoid the neéd for later inquiry into subjective voluntariness. That being the case, it is inconsistent with the preventative nature of the Miranda requirement to assume that a suspect is aware of the diverse circumstances in which the rights covered by Miranda warnings obtain.
I conclude that, taking all of the pertinent factors together, as Wyrick instructs, a reasonable person receiving the Miranda warning issued by Officer Haseall on the night of June 26 in the motel room would not necessarily assume that the same ground rules applied regarding interrogation conducted the next day, under quite different custodial circumstances and by different interrogators. I would therefore hold that all statements .made during *1138the June 27 interrogation before Romanag-gi re-advised Rodriguez-Preciado of his Miranda rights should have been suppressed.
II.
The question remains whether the statements made after Romanaggi delivered the Miranda warning are admissible. Their admissibility is an open question in this circuit.
We are no longer bound by United States v. Orso, 266 F.3d 1030 (9th Cir.2001) (en banc), as the Supreme Court explicitly overruled that decision in Missouri v. Seibert, — U.S. at -, 124 S.Ct. at 2607,9 last Term. Seibert did not, however, create a per se rule barring the admissibility of all statements taken subsequent to mid-interrogation Miranda warnings.
Instead, a plurality of the Court carved out an objective exception to Oregon v. Elstad, 470 U.S. 298, 105 S.Ct. 1285, 84 L.Ed.2d 222 (1985), which had held that statements taken subsequent to midstream Miranda warnings should only be suppressed if they were not knowingly or voluntarily made. Id. at 309, 105 S.Ct. 1285. For the Seibert plurality, the admissibility of statements taken post-Miranda in such a case turns on whether a midstream warning was effective under the circumstances.
As Justice Souter wrote for the four-member plurality,
The threshold issue when interrogators question first and warn later is ... whether it would be reasonable to find that in these circumstances the warnings could function “effectively” as Miranda requires. Could the warnings effectively advise the suspect that he had a real choice about giving an admissible statement at that juncture? Could they reasonably convey that he could choose to stop talking even if he had talked earlier? For unless the warnings could place a suspect who has just been interrogated in a position to make such an informed choice, there is no practical justification for accepting the formal warnings as compliance with Miranda, or for treating the second stage of interrogation as distinct from the first, unwarned and inadmissible segment.
Seibert, — U.S. at -, 124 S.Ct. at 2610 (plurality opinion) (emphasis added); see also id. at 2611 (“When [Miranda ] warnings are inserted in the midst of coordinated and continuing interrogation, they are likely to mislead and ‘deprive a defendant of knowledge essential to his ability to understand the nature of his rights and the consequences of abandoning them.’ ” (quoting Moran v. Burbine, 475 U.S. 412, 424, 106 S.Ct. 1135, 89 L.Ed.2d 410 (1986))). Whether statements given after midstream Miranda warnings should be admissible, the plurality concluded, turned entirely on “whether Miranda warnings delivered midstream could be effective enough to accomplish their object.” Id. at 2612. Effectiveness, the plurality suggested, was a question of what the suspect reasonably believed. Depending on the circumstances, midstream warnings can “be seen as challenging the comprehensibility and efficacy of the Miranda warnings to the point that a reasonable person in the suspect’s shoes would not have un*1139derstood them to convey a message that she retained a choice about continuing to talk.” Id. at 2613.
Concurring in the judgment, however, Justice Kennedy concluded that such a reasonableness test, which “envisions an objective inquiry from the perspective of the suspect, and applies in the case of both intentional and unintentional two-stage interrogations .... 'cuts too broadly.” Id. at 2615-16 (Kennedy, J., concurring in the judgment). Justice Kennedy believed that Elstad should govern absent a showing that the law enforcement officers deliberately attempted an end-run around Miranda, as in Seibert. See id.
That no opinion in Seibert commanded the agreement of a majority of the Justices creates a difficulty in determining which rule to apply here. Generally, where there is no majority opinion, the narrowest opinion adhered to by at least five Justices controls. See Marks v. United States, 430 U.S. 188, 193, 97 S.Ct. 990, 51 L.Ed.2d 260 (1977); see also Townsend v. Quasim, 328 F.3d 511, 519 n. 3 (9th Cir.2003) (citing Smith v. Univ. of Wash., Law Sch., 233 F.3d 1188, 1199 (9th Cir.2000)). Applying the Marks rule to Seibert, however, is not a straightforward endeavor.
Justice Kennedy concurred in Seibert on a ground arguably narrower10 than that relied upon by the plurality. He stated that deliberateness on the part of the police — or the láck thereof — should guide the inquiry, not the objective effectiveness factors outlined in Justice Souter’s plurality opinion.11 But three of the four Justices in the plurality and the four dissenters decisively rejected any subjective good faith consideration, based on deliberateness on the part of the police.12 In dissent, Justice O’Connor, joined by the Chief Justice and Justices Scalia and Thomas, repeatedly agreed with the plurality that the subjective intent of the interrogator cannot control. See, e.g., Seibert, — U.S. at -, 124 S.Ct. at 2616 (O’Connor, J., dissenting) (“[T]he plurality correctly declines to focus its analysis on the subjective intent of the interrogating officer.”); id. at 2617 (“The plurality’s rejection of an intent-based test is also, in my view, correct.”); id. (“Because voluntariness is a matter of the suspect’s state of mind, we focus our analysis on the way in which suspects experience interrogation.... Thoughts kept inside a *1140police officer’s head cannot affect that experience.”); id. at 2618 (“[Rjecognizing an exception to Elstad for intentional violations would require focusing constitutional analysis on a police officer’s subjective intent, an unattractive proposition that we all but uniformly avoid.”). Most definitive is Justice O’Connor’s statement at the end of Part I of her dissent: “[T]he approach espoused by Justice KENNEDY is ill advised .... This approach untethers the analysis from facts knowable to, and therefore having any potential directly to affect, the suspect.” Id. at 2618-19.
The dissenters went on to disagree with the plurality over the force of Elstad: The plurality, along with Justice Kennedy, favored creating an exception to Elstad, although the opinions differed fundamentally as to the nature of the exception. The dissent, in contrast, took issue with the extent to which the plurality “devour[ed]” Elstad. Id. at 2616. Under Elstad, the dissent suggested, “if [the defendant’s] first statement is shown to have been involuntary, the court must examine whether the taint dissipated through the passing of time or a change in circumstances....” Id. at 2619. In so maintaining, however, the dissent also necessarily disagreed with the plurality that the relevant standard should be the objective effectiveness of the warnings. Instead, Justice O’Connor suggested that question-first interrogations should be analyzed “under the voluntariness standards central to the Fifth Amendment and reiterated in Elstad.” Id.
This analysis of the Seibert opinions indicates that while Justice Kennedy’s was the crucial fifth vote for the result, and for the proposition that Elstad does not strictly govern cases with midstream Miranda warnings, Justice Kennedy’s opinion is not the narrowest opinion embodying a position supported by at least five Justices in the majority. It embodies a position supported by two Justices, at most.
The Marks rule is not helpful under these circumstances. As several circuits have convincingly explained, “the Marks rule is applicable only where ‘one opinion can be meaningfully regarded as “narrower” than another’ ” and “can ‘represent a common denominator of the Court’s reasoning.’ ” Anker Energy Corp. v. Consolidation Coal Co., 177 F.3d 161, 170 (3d Cir.1999) (quoting King v. Palmer, 950 F.2d 771, 781 (D.C.Cir.1991) (en banc)); see also United States v. Alcan Aluminum Corp., 315 F.3d 179, 189 (2d Cir.2003); A.T. Massey Coal Co., Inc. v. Massanari, 305 F.3d 226, 236 (4th Cir.2002). The D.C. Circuit further explained this point in King v. Palmer:
Marks is workable — one opinion can meaningfully be regarded as “narrower” than another — only when one opinion is a logical subset of other, broader opinions. In essence, the narrowest opinion must present a common denominator of the Court’s reasoning; it must embody a position implicitly approved by at least five Justices who support the judgment. ... When ... one opinion supporting the judgment does not fit entirely within a broader circle drawn by the others, Marks is problematic. If applied in situations where the various opinions supporting the judgment are mutually exclusive, Marks will turn a single opinion that lacks majority support into national law.
950 F.2d at 781-82.
“In such a case,” the Third Circuit has suggested, “the only binding aspect of a splintered decision is its specific result.” Anker Energy, 177 F.3d at 170. Seibert, however, is not a splintered decision in which “a fragmented Court[has] deeide[d] a case and no single rationale explaining the result enjoys the assent of five Jus*1141tices.” Marks, 430 U.S. at 193, 97 S.Ct. 990. Broken into separate holdings, all but one of the central points of Seibert enjoys the support of five Justices: The rejection of subjective intent enjoys the assent of at least seven Justices. The overruling of Orso enjoys the support of five Justices. The existence of exceptions to Elstad enjoys the support of five Justices. The only point not enjoying the assent of five Justices is the appropriate admissibility standard to apply, on which the Court is split 4-1-4.
As I read it, in agreement with the other circuits’ opinions discussed above, Marks does not prescribe the adoption as governing precedent of a position squarely rejected by seven Justices.13 Justice Kennedy’s opinion on the admissibility standard therefore cannot govern.
If Justice Kennedy’s opinion does not govern, then what does? There are three possibilities: The dissent controls; the plurality controls; or there is no controlling position, and we are free to start from scratch — with Orso no longer binding precedent.
The Seibert dissent cannot govern because the holding that Elstad does not control in a case like this one received five votes. The dissent’s position in Seibert is therefore irreconcilable with the conclusion of the majority of the Court. The plurality opinion is not binding precedent either, at least as to the admissibility standard, for no fifth vote supporting its rationale is to be found. Instead, I suggest that Seibert leaves this court in a situation where there is no binding Supreme Court or Ninth Circuit precedent as to the governing standard.
Nothing bars this court, however, from adopting the Seibert plurality’s standard as the law of the circuit. See United States v. Hearst, 563 F.2d 1331, 1345 n. 10 (9th Cir.1977) (following as “persuasive” the Supreme Court’s plurality opinion in Lanza v. New York, 370 U.S. 139, 82 S.Ct. 1218, 8 L.Ed.2d 384 (1962)). Various of our sister circuits have also adopted the decisions of Supreme Court pluralities as the law of the circuit. See, e.g., Kirsch v. Plovidba, 971 F.2d 1026, 1028-29 (3d Cir.1992) (adopting a three-Justice plurality in Scindia Steam Nav. Co. v. De Los Santos, 451 U.S. 156, 101 S.Ct. 1614, 68 L.Ed.2d 1 (1981), as the law of the circuit); accord Davis v. Portline Transportes Maritime Internacional, 16 F.3d 532, 536 n. 4 (3d Cir.1994) (recognizing Kirsch as so holding). Absent any governing law to the contrary, I would follow a similar path here and adopt the rule advocated by the Seibert plurality as the law of this circuit.
Applying the Seibert plurality’s standard, the admissibility of statements given subsequent to midstream Miranda warnings turns on whether “a reasonable person in the suspect’s shoes could have seen the station house questioning as a new and distinct experience, [and] the Miranda warnings could have made sense as presenting a genuine choice whether to follow up on the earlier admission.” — U.S. at -, 124 S.Ct. at 2612 (plurality opinion).14 The factors the plurality identified as relevant to this inquiry are “the completeness *1142and detail of the questions and answers in the first round of interrogation, the overlapping content of the two statements, the timing and setting of the first and the second [interrogations], the continuity of police personnel, and the degree to which the interrogator’s questions treated the second round as continuous with the first.” Id.
In cases such as this one, where the warning was delivered in the midst of the same interrogation in the same place by the same officer, and where the officer goes back over the very same questions he asked before re-advising the suspect of his Miranda rights, the Seibert plurality approach results in exclusion of the statements taken after the midstream warning. I would so hold.
III.
It is possible, however, that the district court’s error in not suppressing Rodriguez-Preciado’s June 27 statements is harmless beyond a reasonable doubt. The error is certainly harmless as it pertains to Rodriguez-Preciado’s sentence, since the sentence was calculated based on drug quantity levels that, even without his June 27 statements, would still have resulted in the same base offense level under section 2D1.1(c)(2) of the U.S. Sentencing Guidelines.15
The government never suggested, however, in its briefs or at argument, that such error was harmless as to either the sentence or the underlying conviction. Whether we can review for harmless error sua sponte is, perhaps surprisingly, a question that the Ninth Circuit has never directly considered, though most circuits to reach this question have answered it in the affirmative, subject to considerations including “the length and complexity of the record, whether the harmlessness of the error or errors found is certain or debatable, and whether reversal will result in protracted, costly, and ultimately futile *1143proceedings in the district court.” United States v. Giovannetti, 928 F.2d 225, 227 (7th Cir.1991).16 The only Ninth Circuit case even remotely on point, United States v. Vallejo, tersely held that failure to raise harmless error constitutes waiver, without reaching whether a court could consider the issue sua sponte. See 237 F.3d at 1026 (“The Government does not argue that this error was harmless and thus waives that argument.”);17 see also Calvert v. Wilson, 288 F.3d 823, 835-37 (6th Cir.2002) (Cole, J., concurring in the judgment) (citing Vallejo and summarizing the extant harmless error/ waiver case law).
As the First Circuit has explained, the approach outlined in Giovannetti is sensible because, “[i]n a case of clearly harmless error it would be a waste of judicial resources to require a new trial where the result is likely to be the same. In a complex case, it would be equally wasteful of judicial resources to require the appellate bench to delve independently into a complex record without the aid of the government’s brief and the defendant’s responses to it.” Rodriguez Cortes, 949 F.2d at 543. Put another way, the touchstone of whether courts should reach harmless error sua sponte is the extent to which the harmlessness of the error is open to question.
Here, though the harmlessness is clear for Rodriguez-Preciado’s sentence, I believe that is unclear as to his conviction, particularly for the third count of the indictment, conspiracy to possess with intent to distribute. A substantial amount of the evidence introduced at trial concerning the existence of a conspiracy — and Rodriguez-Preeiado’s role therein — came from Sergeant Romanaggi’s testimony concerning his June 27 interrogation of Rodríguez-Preciado. It is entirely possible that the jury would still have found Rodríguez-Preciado guilty without that evidence. The salient point, however, is that the harmlessness of Romanaggi’s testimony with respect to the conspiracy charge is at the very least “debatable.” That is enough, under the Giovannetti line of cases, to bar sua sponte harmless error review. ■
* * * . ■
Because I would not reach whether the district court’s refusal to suppress Rodri-guezAPreciado’s June 27 statements constitutes harmless error as to his conviction, I would reverse and remand for a new trial. I therefore respectfully dissent from subsection II.B.3 of the majority’s opinion, and, because of the above harmless error discussion, from the result its analysis necessarily compels.
. I also disagree with the majority's assertion that the district court's determination on this point was a "factual finding" warranting clear error review. There is no case law in this circuit concerning the appropriate standard of review for the district court's determination of the continuing "effectiveness” of Miranda warnings. The voluntariness of a confession, by contrast, another area of Miranda law where we look to the totality of the circumstances, is an issue that we have reviewed de novo since the Supreme Court’s decision in Miller v. Fenton, 474 U.S. 104, 112-18, 106 S.Ct. 445, 88 L.Ed.2d 405 (1985). See, e.g., Tolbert v. Page, 182 F.3d 677, 682 n. 8 (9th Cir.1999) (en banc). To the extent that the majority suggests only that purely factual findings, such as what the defendant said, are reviewed for clear error, I do not disagree. But United States v. Andaverde, 64 F.3d 1305 (9th Cir.1995), the case on which the majority principally relies for its conclusion that the original warning remained effective in this case, treated that issue as a question of volun-tariness and reviewed it de novo. See id. at 1310. I would do the same.
. Also of relevance is our earlier decision in United States v. Nordling, 804 F.2d 1466 (9th Cir.1986), where the court found no requirement that a suspect be readvised of his Miranda rights when the interrogation was continued by different questioners and “[n]o appreciable time had elapsed.” Id. at 1471. The Nordling court did not explicitly state what it meant by “no appreciable time,” though it cited, with approval, an Eighth Circuit case holding that a five-hour gap between interviews did not require re-administration of the warnings. See id. (citing Stumes v. Solem, 752 F.2d 317 (8th Cir.1985)).
. Though the record in this case is not explicit, it appears that the Miranda warning was administered to Rodríguez-Preciado at the Satellite Motel shortly after 9:00 p.m. on June 26, and that the interrogation on June 27 began shortly after 1:15 p.m. The interrogation at the county jail conducted on the night of June 26 appears to have concluded by 11:30 p.m.
. The court in Andaverde noted that "[t]he next day, October 28, Keeth again questioned Andaverde at the jail." 64 F.3d at 1308 (emphasis added). Earlier in the opinion, the court suggests that the original questioning happened at the police station. See id. But the court's use of the word “again,” without noting any other change in circumstances, seems to indicate that the two interviews were both conducted in the same general place.
. The facts in Andaverde are unclear concerning when the interrogations took place on each day. Therefore, though the majority here reads Andaverde as approving a "one day” delay, see ante at 1129, and though An-daverde itself spoke of a period of "one day,” see 64 F.3d at 1313, it is entirely possible that the actual delay between the end of the second interrogation and the beginning of the third was closer to twelve hours, see id. at 1308 & n. 2, less than the interval in this case.
. The Vasquez court concluded that the five factors weighed in favor of admitting Vasquez's statements. Its determination turned on the short time delay, the fact that Vasquez knew he was going to be interviewed by different officers, and Vasquez’s "long experience and frequent contact with the criminal justice system.'' 889 F.Supp. at 178. At most, only one of the five factors weighed in Vasquez’s favor, and the court found that factor — the interview by different officers— negligible given Vasquez's prior knowledge that such a change was to occur.
. In point of fact, there had been two changes in location since the warnings were first administered — from the Satellite Motel to the police substation, and from the substation to the jail.
. The Court granted certiorari in Seibert to resolve a circuit split between Orso and a similar holding by the First Circuit, on the one hand, and the Eighth and D.C. Circuit decisions in conflict with Orso, on the other. See Seibert, - U.S. at -, 124 S.Ct. at 2607 (plurality opinion) (citing United States v. Gale, 952 F.2d 1412, 1418 (D.C.Cir.1992); United States v. Carter, 884 F.2d 368, 373 (8th Cir.1989); Orso, 266 F.3d at 1034-39; and United States v. Esquilin, 208 F.3d 315, 319-21 (1st Cir.2000)).
. Justice Kennedy characterized his opinion as "narrower.” See Seibert, - U.S. at -, 124 S.Ct. at 2616 (Kennedy, J., concurring in the judgment).
. Leaving aside its substantive analysis, Justice Kennedy’s opinion created a clear fifth vote for overruling our en banc decision in Orso, in which we had held that a confession after a midstream Miranda warning should be suppressed only if the pre-warning confession was “involuntary, and any taint therefrom had not dissipated by the time [the defendant] was read the Miranda warnings.” 266 F.3d at 1039. This position, echoed in Justice O'Connor's dissent in Seibert, is at odds with Justice Kennedy’s conclusion that, subject to deliberateness on the part of the police, "post-warning statements that are related to the substance of prewarning statements must be excluded absent specific, curative steps." - U.S. at -, 124 S.Ct. at 2615 (Kennedy, J., concurring in the judgment). Because Justice Kennedy agreed that the Elstad-based volun-tariness approach that Orso approved could not stand, there can be no question as to whether Orso remains the law of the circuit. The operative issue is why it does not.
. Arguably, Justice Breyer’s concurring opinion, though he fully concurred in (and joined) Justice Souter's opinion for the plurality, differs from the plurality on the deliberateness point. See, e.g., Seibert, - U.S. at -, 124 S.Ct. at 2614 (Breyer, J., concurring) (joining Justice Kennedy's opinion "insofar as it ... makes clear that a good-faith exception applies”). Accepting this position for the sake of argument, the tally is seven to two against the subjective-intent-of-the-interrogator position.
. A rule that Maries does not apply to cases where the result of applying it is indiscernible or illogical finds contemporary support in a brief passage of Justice O’Connor's opinion in Grutter v. Bollinger, 539 U.S. 306, 123 S.Ct. 2325, 156 L.Ed.2d 304 (2003). Commenting on whether Justice Powell’s solo opinion in Bakke controlled, the Grutter Court concluded that "it does not seem 'useful to pursue the Marks inquiry to the utmost logical possibility when it has so obviously baffled and divided the lower courts that have considered it.’ ” Id. at 325, 123 S.Ct. 2325 (quoting Nichols v. United States, 511 U.S. 738, 745-46, 114 S.Ct. 1921, 128 L.Ed.2d 745 (1994)).
. Adopting this standard would also reduce the complexity of the first issue considered in this opinion — whether the original Miranda *1142warnings were still effective at the outset of the June 27 interview. As the Seibert plurality/dissent emphasizes, the central issue is what the suspect could reasonably have believed, based on the totality of circumstances, regarding the continued effectiveness of the earlier warnings — not, as the majority would have it here, what he remembered concerning the earlier warnings.
. Rodriguez-Preciado confessed to being in possession of — and distributing — various drug quantities, including at least six pounds of methamphetamine (five or six to Glenn and one to Jim Grenfell), a half-kilo of cocaine (to Grenfell), a quarter-pound of cocaine (in another instance), and nineteen to twenty-three pounds of marijuana. The police were able independently to tie various other drug quantities to Rodriguez-Preciado, but as the Pre-sentence Report (PSR) makes clear, at least thirty-five pounds of marijuana, thirteen and one-quarter pounds of methamphetamine, and five kilograms of cocaine were attributed to Rodriguez-Preciado based solely on the statements he made to the police, nearly all of which came on June 27. Because these statements should have been suppressed, the quantities should not have figured into Rodriguez-Preciado’s sentence. As the PSR identifies, the drug quantities based entirely on Rodriguez-Preciado’s statements add up to a total of 7,025 kilograms of marijuana. Rodriguez-Preciado’s total sentence was based on a total drug quantity of 22,739 kilograms of marijuana. Therefore, absent the drug quantities that were attributed to him based on his June 27 statements, Rodriguez-Preciado’s sentence should have been based on a total quantity of 15,714 kilograms of marijuana. But this quantity nets the exact same base offense level — 36—under U.S.S.G. § 2D1.1(c)(2) as the 22,739 kilograms actually relied upon by the district court. Because recalculation of the quantity based on the admissible evidence would not lead to a lesser base offense level, the district court’s error in not suppressing Rodriguez-Preciado's June 27 statements, as it pertains to sentencing, was objectively harmless. See United States v. Alvarez, 358 F.3d 1194, 1213 (9th Cir.2004); see also United States v. Scheele, 231 F.3d 492, 499-500 & n. 4 (9th Cir.2000).
. Though none of the Giovannetti line of cases cite to them, the mandatory language contained in the corresponding provisions of the Federal Rules of Civil Procedure, the Federal Rules of Criminal Procedure, and the Judicial Code, supports the notion that harmless error may be reviewed sua sponte. Compare Fed. R. Civ. P. 61 (“The court at every stage of the proceeding must disregard any ' error or defect in the proceeding which does not affect the substantial rights of the parties.”), and Fed. R. Crim. P. 52(a) (“Any error, defect, irregularity, or variance that does not affect substantial rights must be disregarded.”), with 28 U.S.C. § 2111 ("On the hearing of any appeal or writ of certiorari in any case, the court shall give judgment after an examination of the record without regard to errors or defects which do not affect the substantial rights of the parties.”). See generally Chapman v. California, 386 U.S. 18, 21-24 & n. 5, 87 S.Ct. 824, 17 L.Ed.2d 705 (1967) (discussing the background and purpose of these provisions). 1 .
. In Vallejo, the court concluded that the error was not harmless in any event. See 237 F.3d at 1026. '