Thompson v. Runnels

657 F.3d 784, 2011 U.S. App. LEXIS 11945, 2011 WL 2279451

• Court: Court of Appeals for the Ninth Circuit
• Decided: June 9, 2011
• Docket: No. 08-16186
• Status: Published

Opinions

Order; Dissent to Order by Judge CALLAHAN; Opinion by Judge BERZON; Dissent by Judge IKUTA.

ORDER

Judge Goodwin and Judge Berzon voted to deny the petition for rehearing. Judge Berzon voted to deny the petition for rehearing en banc, and Judge Goodwin so recommended. Judge Ikuta voted to grant the petition for rehearing and the petition for rehearing en banc.

The full court was advised of the petition for rehearing en banc. A judge requested a vote on whether to rehear the matter en banc, and the matter failed to receive a majority of the votes of the nonrecused active judges in favor of en banc consideration. Fed. RApp. P. 35.

The majority opinion and dissent filed in this case on September 8, 2010 are withdrawn. The opinion and dissent filed with this order replace the withdrawn opinion and dissent.

The petition for rehearing and the petition for rehearing en banc are DENIED.

CALLAHAN, Circuit Judge, with whom O’SCANNLAIN, GOULD, TALLMAN, BYBEE, BEA, and IKUTA, Circuit Judges, join, dissenting from the denial of rehearing en banc:

Antwion Thompson killed his girlfriend and he admitted that he had done so both before and after he was given Miranda warnings. Indeed, after being given the Miranda warnings, he voluntarily participated in a videotaped reenactment of the crime at his girlfriend’s house. The California state courts consistently denied his challenges to the admission of his post-Miranda statements as did the federal district court. However, a majority of the three-judge panel, over Judge Ikuta’s insightful dissent, applies the wrong federal law, improperly reviews the case de novo, and based on its interpretation of the underlying facts, orders Thompson’s conviction vacated.

We should have reheard this case en banc for two reasons. First, the panel erred in concluding that Missouri v. Seibert, 542 U.S. 600, 124 S.Ct. 2601, 159 L.Ed.2d 643 (2004), which was not decided until five months after the California Court of Appeal decision, was the “clearly established federal law” that the state court should have applied. Second, the majority fails to adhere to our limited role in reviewing state criminal convictions under the Antiterrorism and Effective Death Penalty Act of 1996 (“AEDPA”), 28 U.S.C. § 2254 by reviewing the case de novo. See Harrington v. Richter, 562 U.S.-, 131 S.Ct. 770, 785, 178 L.Ed.2d 624 (2011). In doing so, the panel majority misreads the Supreme Court’s opinion in Seibert so as to eviscerate its opinion in Oregon v. Elstad, 470 U.S. 298, 105 S.Ct. 1285, 84 L.Ed.2d 222 (1985).

1. Although the extent to which California asserted that Elstad, rather than Seiberi, was the clearly established Federal law for the purpose of reviewing this case under AEDPA is admittedly arguable, there can be no question that the panel held that Seibert was clearly established Federal law. Its opinion states:

The California Court of Appeals, in the last reasoned state court decision on Thompson’s Miranda claim, did not apply the rule announced in Seibert. Rather, based on its reading of Oregon v. Elstad, 470 U.S. 298, 105 S.Ct. 1285, 84 L.Ed.2d 222 (1985), the state appellate court ruled that “so long as the earlier [unwarned confession] was not involuntary due to police coercion, [a] subsequent voluntary, warned statement is admissible.” The state court assumed, therefore, that deliberately delayed Miranda warnings are always effective absent actual police coercion. Because this “ ‘rule ... contradicts the governing law set forth [by the Supreme Court]’ ” in Seibert, see Rios v. Garcia, 390 F.3d 1082, 1084 (9th Cir.2004) (quoting Williams v. Taylor, 529 U.S. 362, 405-06, 120 S.Ct. 1495, 146 L.Ed.2d 389 (2000)), the state court’s decision was “contrary to ... clearly established Federal law, as determined by the Supreme Court of the United States.” 28 U.S.C. § 2254(d)(1).

Thompson v. Runnel, 621 F.3d 1007, 1016 (9th Cir.2010) (parallel citations omitted).¹

The Supreme Court, however, has explained that “clearly established Federal law, as determined by the Supreme Court of the United States” means “the holdings, as opposed to the dicta, of [the Supreme] Court’s decisions as of the time of the relevant state court decision.” Williams, 529 U.S. at 412, 120 S.Ct. 1495 (O’Connor, J, writing for the Court). We have subsequently recognized that “[e]learly established Federal law under § 2254(d)(1) is the governing legal principle or principles set forth by the Supreme Court at the time the state court renders its decision.” Moses v. Payne, 555 F.3d 742, 751 (9th Cir.2009) (quoting Lockyer v. Andrade, 538 U.S. 63, 75, 123 S.Ct. 1166, 155 L.Ed.2d 144 (2003)). Elstad had been decided when the California Court of Appeal issued its decision, but Seibert had not.²

Arguably, Justice Stevens’ opinion in Williams is a potential source of confusion as to the proper temporal cutoff because it announces a rule different than the rule announced by Justice O’Connor in the same opinion.³ Early in Section III of his opinion, he states that “[t]he threshold question under AEDPA is whether Williams seeks to apply a rule of law that was clearly established at the time his state-court conviction became final.” Williams, 529 U.S. at 390, 120 S.Ct. 1495.

Whatever confusion that might have been born out of Williams should have been laid to rest by the Supreme Court’s subsequent opinions. Justice O’Connor’s “as of the time of the relevant state-court decision” version has been used by the Supreme Court every time it has stated the rule since Williams was decided. See Carey v. Musladin, 549 U.S. 70, 74, 127 S.Ct. 649, 166 L.Ed.2d 482 (2006) (“ ‘clearly established Federal law’ in § 2254(d)(1) ‘refers to the holdings, as opposed to the dicta, of this Court’s decisions as of the time of the relevant state-court decision’ ”); Yarborough v. Alvarado, 541 U.S. 652, 660-61, 124 S.Ct. 2140, 158 L.Ed.2d 938 (2004) (“clearly established law as determined by this Court ‘refers to the holdings, as opposed to the dicta, of this Court’s decisions as of the time of the relevant state-court decision’ ”); Andrade, 538 U.S. at 71, 123 S.Ct. 1166 (“Section 2254(d)(l)’s ‘clearly established’ phrase ‘refers to the holdings, as opposed to the dicta, of this Court’s decisions as of the time of the relevant state-court decision.’ ”). Indeed, Andrade goes even further, explaining “[i]n other words, ‘clearly established Federal law’ under § 2254(d)(1) is the governing legal principle or principles set forth by the Supreme Court at the time the state court renders its decision.” Id. at 71-72, 123 S.Ct. 1166.

If there was any life left in the majority’s perspective as to the proper temporal cutoff after those cases, the Court put a dagger through its heart in Cullen v. Pinholster, — U.S.-, 131 S.Ct. 1388, 179 L.Ed.2d 557 (2011). There, the Court reversed an en banc panel of the Ninth Circuit which had approved the district court’s use of an evidentiary hearing to supplement the record on habeas review. Id. at 1397. The Court explained:

We now hold that review under § 2254(d)(1) is limited to the record that was before the state court that adjudicated the claim on the merits. Section 2254(d)(1) refers, in the past tense, to a state-court adjudication that “resulted in” a decision that was contrary to, or “involved” an unreasonable application of, established law. This backward-looking language requires an examination of the state-court decision at the time it was made. It follows that the record under review is limited to the record in existence at that same time i.e., the record before the state court.

Id. at 1398 (emphasis added). It also follows that the law on review is limited to the federal law in existence at that time. To drive the point home further, the Court explained that its “cases emphasize that review under § 2254(d)(1) focuses on what a state court knew and did. State-court decisions are measured against this Court’s precedents as of ‘the time the state court renders its decision.’ ” Id. at 1399(quoting Andrade, 538 U.S. at 71-72, 123 S.Ct. 1166). Indeed, “[i]t would be strange to ask federal courts to analyze whether a state court’s adjudication resulted in a decision that unreasonably applied federal law to facts not before the state court.” Id. It would be no less strange to ask federal courts to analyze whether a state court’s adjudication resulted in a decision that unreasonably applied federal law that was not yet in existence. Moreover, the Court explained:

What makes the consideration of new evidence strange is not how “different” the task would be, but rather the notion that a state court can be deemed to have unreasonably applied federal law to evidence it did not even know existed. We cannot comprehend how exactly a state court would have any control over its application of law to matters beyond its knowledge.

Id. at 1399 n. 3. Similarly, it is incomprehensible how exactly a state court could have any control over its application of Supreme Court decisions that have not yet been handed down. In other words, the state court need not be clairvoyant to withstand AEDPA review.

Contrary to this clear weight of Supreme Court authority, the majority held the state court accountable for not applying Seibert, a Supreme Court opinion that issued almost five months after the state appellate court ruled. The majority attempts to justify its position by citing Spisak, 130 S.Ct. at 681, which it said casts doubt as to the applicable rule. Thompson, 621 F.3d at 1016, n. 7. Spisak noted the different standards set forth in Williams, but stated it would “assume” that Mills v. Maryland, 486 U.S. 367, 108 S.Ct. 1860, 100 L.Ed.2d 384 (1988), was the clearly established law because all the parties and the Court of Appeals assumed it was. Spisak, 130 S.Ct. at 681. Mills was decided after the state court issued its decision denying Spisak relief, but before his conviction became final under the Teague rule. However, the Court affirmed Spisak’s conviction over his claim that it violated the standard set forth in Mills. In other words, the Court determined that even assuming that Mills, the more favorable standard for Spisak, applied, he was not entitled to relief.

Here, the majority erred by assuming Seibert applied and looking to Spisak for support. Unlike the parties in Spisak, the state did, in one part of its brief, assert that Elstad was the clearly established Federal law because Seibert had not been decided at the time the state court ruled. Rather than confront the state’s assertion that Elstad rather than Seibert was “the clearly established Supreme Court law at the time of the last reasoned state court decision,” the panel majority chose to ignore it. The majority is able to grant Thompson relief only by applying Seibert rather than Elstad. We should have taken this case en banc to clarify that “clearly established Federal law” is determined at the time the state court issues its decision.⁴

2. Furthermore, even if Seibert is applicable to this case, by failing to recognize the extent to which Seibert reaffirms Elstad, the majority lost sight of the AEDPA standard that relief is only available when the state appellate court decision was “contrary to, or involved an unreasonable application of, clearly established Federal law, as determined by the Supreme Court.” 28 U.S.C. § 2254(d)(1).

The California Court of Appeal, applying Elstad, ruled that “so long as the earlier [unwarned confession] was not involuntary due to police coercion, [a] subsequent voluntary, warned statement is admissible.” Thompson, 621 F.3d at 1016. The panel majority, however, by misreading Seibert, concludes that the state court’s reliance on Elstad contradicts Seibert, and therefore, that this court is entitled to “review the substantive constitutionality of the state custody de novo.” Id. (quoting Frantz v. Hazey, 533 F.3d 724, 737 (9th Cir.2008) (en banc)).⁵

But contrary to the majority’s assertion, Seibert is really a continuation or refinement of Elstad; Seibert did not overrule Elstad. Justice Kennedy, whose concurrence provided the fifth vote for the plurality in Seibert,⁶ stated:

In my view, Elstad was correct in its reasoning and its result. Elstad reflects a balanced and pragmatic approach to enforcement of the Miranda warning. An officer may not realize that a suspect is in custody and warnings are required. The officer may not plan to question the suspect or may be waiting for a more appropriate time. Skilled investigators often interview suspects multiple times, and good police work may involve referring to prior statements to test their veracity or to refresh recollection. In light of these realities it would be extravagant to treat the presence of one statement that cannot be admitted under Miranda as sufficient reason to prohibit subsequent statements preceded by a proper warning. See Elstad, 470 U.S. at 309 [105 S.Ct. 1286](“It is an unwarranted extension of Miranda to hold that a simple failure to administer the warnings ... so taints the investigatory process that a subsequent voluntary and informed waiver is ineffective for some indeterminate period”). That approach would serve “neither the general goal of deterring improper police conduct nor the Fifth Amendment goal of assuring trustworthy evidence would be served by suppression of the ... testimony.” Id. at 308 [105 S.Ct. 1285].

542 U.S. at 620, 124 S.Ct. 2601 (parallel citations omitted). He noted that Seibert presented different considerations because the “police used a two-step questioning technique based on a deliberate violation of Miranda.” Id. He asserted that “[wjhen an interrogator uses this deliberate, two-step strategy, predicated upon violating Miranda during an extended interview, post-warning statements that are related to the substance of pre-warning statements must be excluded absent specific, curative steps.” Id. at 621, 124 S.Ct. 2601 (emphasis added). In sum, Seibert is not contrary to Elstad, but merely recognizes that when the officers use a deliberate two-step strategy designed to violate Miranda, the balance of the competing interests set forth in Elstad require the suppression of a post-Miranda, admission.

Thus, even assuming that the Supreme Court’s opinion in Seibert may be applied to the California Court of Appeal’s decision, the state court’s use of Elstad was not contrary to “clearly established Federal law.” The state appellate court applied the standard set forth by the Supreme Court in Elstad. Indeed, the majority admits as much. The majority, however, asserts that it may undertake a de novo review because it thinks the state appellate court contradicted Seibert. 621 F.3d at 1016. But this is the type of “improper understanding” of AEDPA that caused the Supreme Court to reverse us in Richter:

The Court of Appeals appears to have treated the unreasonableness question as a test of its confidence in the result it would reach under de novo review: Because the Court of Appeals had little doubt that Richter’s Strickland claim had merit, the Court of Appeals concluded the state court must have been unreasonable in rejecting it. This analysis overlooks arguments that would otherwise justify the state court’s result and ignores further limitations of § 2254(d), including its requirement that the state court’s decision be evaluated according to the precedents of this Court. See Renico v. Lett, — U.S.-, 130 S.Ct. 1855, 1866, 176 L.Ed.2d 678 (2010). It bears repeating that even a strong case for relief does not mean the state court’s contrary conclusion was unreasonable. See Lockyer, [538 U.S.] at 75, 123 S.Ct. 1166.

Richter, 131 S.Ct. at 786 (parallel citations omitted). We should have taken this case en banc to correct the majority’s failure to appreciate that the California Court of Appeal opinion is not “an unreasonable application of, clearly established Federal law.” 28 U.S.C. § 2254(d)(1).

Since the state court properly applied clearly established Federal law, Thompson would only be entitled to relief on his federal habeas petition if he could show that the state court’s opinion was “based on an unreasonable determination of the facts in light of the evidence.” 28 U.S.C. § 2254(d)(2). The majority, again failing to recognize the constraints of AEDPA, in effect wrongly determines that the state court opinion was an unreasonable determination of the facts.

The majority’s factual determination fails to appreciate that Seibert defines an extreme end of the balancing standard set forth in Elstad. The four justices in dissent,⁷ as well as Justice Kennedy in his concurring opinion, agreed that Elstad rejected “the argument that the lingering compulsion inherent in a defendant having let the ‘cat out of the bag’ required suppression.”⁸ Seibert, 542 U.S. at 627, 124 (O’Connor, J. dissenting). Justice Kennedy set forth “a narrower test applicable only in the infrequent case, such as we have here in which the two-step interrogation technique was used in a calculated way to undermine the Miranda warning.” Id. at 622, 124 S.Ct. 2601(Kennedy, J. concurring). As noted by Justice O’Connor in her dissent, Justice Kennedy’s approach “would extend Miranda’s exclusionary rule to any case in which the use of the ‘two-step interrogation technique’ was ‘deliberate’ or ‘calculated.’ ” 542 U.S. at 626-27, 124 S.Ct. 2601.

Here, contrary to the majority’s opinion, the district court did make a finding that the officers did not employ a deliberate two-step strategy. It found that there was “no evidence in the record concerning an official police policy of deliberately withholding Miranda warnings until a suspect had confessed.” The district court further found — as the majority admits — that Thompson did “not cite to the record or present any evidence in support of his assertion that the inspectors in this case deliberately withheld their Miranda advisement until Petitioner had incriminated himself.” 621 F.3d at 1012 n. 5. This should have been the end of the case. The district court’s finding on deliberateness of a two-step interrogation is a factual finding that can only be disturbed if clearly erroneous. United States v. Narvaez-Gomez, 489 F.3d 970, 974 (9th Cir.2007). There being no evidence of “an official police policy of deliberately withholding Miranda warnings” and Thompson having failed to present any evidence that a deliberate two-step interrogation was used against him, the state court’s decision cannot be “an unreasonable determination of the facts in light of the evidence.” 28 U.S.C. § 2254(d)(2). Nor was the district court’s determination that the officers did not employ a deliberate two-step interrogation strategy clear error.

The panel majority, nonetheless, insists on reweighing the evidence de novo. This ignores the deference owed to the district court’s factual finding and’violates the intent of § 2254 “to confirm that state courts are the principal forum for asserting constitutional challenges to state convictions.” Richter, 131 S.Ct. at 787. “[I]f the state court denies the claim on the merits, the claim is barred in federal court unless one of the exceptions to § 2254(d) set out in §§ 2254(d)(1) and (2) applies.” Id. ’Even accepting that reasonable minds might differ on the voluntariness of Thompson’s post-Miranda admissions, the panel majority, like our opinion reversed in Richter:

gave § 2254(d) no operation or function in its reasoning. Its analysis illustrates a lack of deference to the state court’s determination and an improper intervention in state criminal processes, contrary to the purpose and mandate of AEDPA and to the now well-settled meaning and function of habeas corpus in the federal system.

Id.

Thompson murdered his girlfriend in 1998, and his state murder conviction became final under the Teague rule in 2004. The panel majority’s failure in 2010 to appreciate that Seibert is not contrary to Elstad, and its re-weighing of the inferences to be drawn from the circumstances, undermine the state’s significant interest in finality and society’s right to punish an admitted offender. See Harris v. Reed, 489 U.S. 255, 282, 109 S.Ct. 1038, 103 L.Ed.2d 308 (1989) (Kennedy, J., dissenting). We should have taken this case en banc and affirmed the district court’s denial of the habeas petition. ■

OPINION

BERZON, Circuit Judge:

Before he was given Miranda warnings, Antwion Thompson confessed to killing his girlfriend. He then confessed again once he was properly advised of his rights. He was convicted by a California jury of first-degree murder, mayhem, and personal deadly weapon use. Before the California courts and in this federal habeas proceeding he maintains that the admission at trial of his confession violated the privilege against self-incrimination, because the investigating officers deliberately withheld Miranda warnings until after he had confessed to the crime. The district court denied the petition. We reverse.

I.

Arie Bivins, Thompson’s sometime girlfriend, was • murdered between 1:30 and 4:30 p.m. on June 22, 1998. Bivins was seventeen, Thompson eighteen., In the preceding days and months, Bivins had attempted to break up with Thompson, prompting violent reactions from him.

On the day of the murder, Thompson’s father saw Thompson and Bivins talking outside his house at 1:30 p.m. Thompson left his father’s house at 2:00, not saying where he was going. At about 3:00, a dog in the yard next to Bivins’ house barked ferociously. Thompson returned home at 4:00, told his father he was worried about Bivins, and had his father drive him to Bivins’ home. There, Thompson found Bivins’ front door unlocked and her dead body just inside the door.

When the police arrived, Thompson appeared distraught. Officer Solzman approached Thompson, who said he did not feel well. Solzman offered to let Thompson lie down in the air-conditioned police car, and Thompson agreed. Later, homicide detective Conaty woke Thompson to ask, him to go to the police station to talk about finding the body. Thompson responded that he wanted to go home and sleep. When Conaty explained that Thompson’s assistance could be critical to the investigation, Thompson agreed to go to the station. Thompson was not placed under arrest at that time.

When Thompsoh arrived at the station he was placed in a break room, where he waited approximately six hours. Officer Solzman sat outside the break room doing paperwork. Thompson’s father testified that he asked to speak to his son but was refused; a police witness denied that there was any such request.

Around 11:00 p.m., Inspectors Conaty and Giacomelli moved Thompson into an interview room containing three chairs and no other furniture. Thompson was not handcuffed and did not ask to go home, but, by then, Conaty considered him “the primary suspect.” The ensuing two-hour interview was videotaped.

At the outset, Conaty told Thompson that the interview could be conducted another time in the event Thompson was too tired to do it. No Miranda warnings were given. Thompson agreed to talk about the incident and gave an initial account of his activities that day with little prompting by the officers. Thompson insisted that he did not go to Bivins’ house between 10:30 a.m. and 4:00 p.m.

The tone of the interrogation then became more confrontational: The officers invented an eyewitness account that put Thompson at Bivins’ house around 2:30 p.m. and pressed Thompson to explain the apparent contradiction. Thompson suggested that the witness got the time wrong, but Conaty forcefully disagreed: “No, no, no bro. Eight hours we’ve been up there talking to these people. I’ve been very clear with them about what we’re talking about.... Now you’ve got to help me out with this thing.” As Inspector Conaty testified at trial, this fabricated eyewitness account was one of several techniques that he and Giacomelli employed for the purpose of “keeping] the interview going” and “hav[ing] the defendant place [himself] at the location.”

The breakthrough occurred when the officers tried again to get Thompson to admit that he had been to the house in the early afternoon, this time suggesting that Thompson had lied earlier because he was scared, “understandable,” they said, in light of his youth.¹ Thompson thereupon broke into tears and said he went alone to Bivins’ house around 2:00 p.m. where he found her dead.² Thompson told the officers he was scared and wanted to kill himself.

No Miranda warnings had yet been administered, but the interrogation continued. The officers told Thompson — again, falsely — that they had found “high-velocity blood spatter” on a brown shirt left in his bedroom and his fingerprint in blood on a chair in Bivins’ living room. Citing this “evidence” as proof that Thompson was at the scene and that a fight occurred, the officers told Thompson, “What makes or breaks this thing for how it comes out for you is to tell us what the circumstances were.... [T]his is your one chance to do that.”

Taking the bait, Thompson abandoned his story that Bivins was already dead when he arrived at her house in the afternoon.³ He admitted to finding her alive and to stabbing her in the chest during an altercation, although he insisted that he did so accidentally. In response to further questions, Thompson then elaborated upon the details of the altercation and the location of the murder weapon and his bloodied clothing. When Conaty asked Thompson whether he felt better after “getting it all off [his] chest,” Thompson repeated that he wanted to, and intended to, commit suicide.

At this point, Conaty told Thompson that the decision about what would happen next to Thompson would be up to the District Attorney. Asked after that for more details about the incident — still with no Miranda warnings — Thompson gave a yet more detailed account of the altercation in Bivins’ living room. In response to specific questioning about who held the knife, Thompson admitted that Bivins never wielded it. Recounting the altercation once more, he admitted to stabbing her and slitting her throat after she had collapsed on the floor. The officers asked several more questions about Thompson’s intent in doing so and about his trip home afterwards.

Only then did the interrogating officers provide the warnings that Miranda specifies. See Miranda v. Arizona, 384 U.S. 436, 444-45, 86 S.Ct. 1602, 16 L.Ed.2d 694 (1966). Having done so, they took Thompson back through the day’s events. When Thompson reported that he slit Bivins’ throat to prevent her from suffering, Conaty corrected him based on a pre-Miranda warning admission: “That, and you didn’t want her to necessarily survive and tell on you, isn’t that right?” The officers repeatedly referred back to the previous conversation as Thompson recapitulated his account.

Some time after 1:00 a.m., after receiving the Miranda warnings, Thompson asked to end the interview, saying that he was sleepy and needed to lie down. But the interview continued with a few more questions. The officers then handcuffed Thompson, without telling him that he was under arrest, and, around 2:00 a.m., took him to look for the murder weapon and clothing he had burned. Only after that excursion was Thompson booked into jail. He spent the rest of the night shackled to the floor in a safety cell, on suicide watch. Stripped to his underwear and without a bed or blankets, Thompson was unable to sleep.

At the jail the next morning Inspectors Conaty and Giacomelli re-advised Thompson of his Miranda rights. After lunch, Thompson participated in a videotaped reenactment of the crime at Bivins’ house.

Before trial, Thompson moved to suppress all of the statements he had made during the interrogations and reenactment on June 22 and 23. After a brief evidentiary hearing, the state trial court first addressed the “custodial” prerequisite to the Miranda requirement,⁴ determining that Thompson was not in custody at the outset of the interrogation but that the interrogation became custodial sometime after Thompson admitted to visiting Bivins’ house alone but before he conceded that he found her there alive. The trial court suppressed the statements made after the interrogation became custodial and before Miranda warnings were administered. But the court admitted Thompson’s post-Miranda confession and the videotaped reenactment of the crime.

At trial, Thompson was convicted of first-degree murder, CahPenal Code § 187, mayhem, § 203, and personal deadly weapon use, § 12022(b)(1). His sentence was twenty-six years to life in prison.

The state appellate court, deciding Thompson’s appeal on February 3, 2004, relied on Oregon v. Elstad, 470 U.S. 298, 105 S.Ct. 1285, 84 L.Ed.2d 222 (1985), to conclude that Thompson made a knowing and voluntary waiver of his rights once he was given Miranda warnings, even though he had first confessed during a non-Mirandized custodial interrogation. The California Supreme Court summarily denied review on April 21, 2004. The United States District Court for the Northern District of California denied Thompson’s habeas petition, affirming the state court determination that all of Thompson’s statements were voluntary and citing a lack of evidence that the use of deliberate “two-step” interrogations was an official policy of the police department.⁵ This timely appeal followed.

II.

Because Thompson’s petition was filed after the effective date of the Anti-Terrorism and Effective Death Penalty Act (“AEDPA”) and his claims were rejected by the state courts in a decision on the merits, we may grant relief only if the last reasoned state decision was “ ‘based on an unreasonable determination of the facts in light of the evidence,’ ” or on a legal determination that was “ ‘contrary to, or involved an unreasonable application of, clearly established Federal law, as determined by the Supreme Court of the United States.’” Kennedy v. Lockyer, 379 F.3d 1041, 1046 (9th Cir.2004) (quoting 28 U.S.C. § 2254).

We review the district court’s denial of a habeas petition de novo, except that the district court’s findings of fact are reviewed for clear error. McClure v. Thompson, 323 F.3d 1233, 1240 (9th Cir. 2003).

A.

Unless adequately exhausted in the California state courts, Thompson’s challenge to the admission at trial of his post-Miranda statements is not cognizable on habeas. 28 U.S.C. § 2254(b)(1). California argues that Thompson’s challenge to the admission of his confession is unexhausted and so we may not entertain it.

A petitioner satisfies “the exhaustion requirement if ... he has ‘fairly presented’ his federal claim to the highest state court with jurisdiction to consider it....” Johnson v. Zenon, 88 F.3d 828, 829 (9th Cir. 1996) (quoting Anderson v. Harless, 459 U.S. 4, 6, 103 S.Ct. 276, 74 L.Ed.2d 3 (1982)). To do so, Thompson was required to “include reference to a specific federal constitutional guarantee, as well as a statement of the facts that entitlefd him] to relief.” Gray v. Netherlands, 518 U.S. 152, 162-63, 116 S.Ct. 2074, 135 L.Ed.2d 457 (1996).

Thompson satisfied this requirement. In his petition for review to the California Supreme Court, Thompson argued that his postwarning statements were inadmissible under Miranda v. Arizona, 384 U.S. 436, 86 S.Ct. 1602,16 L.Ed.2d 694 (1966), “even though he made them after the inspectors belatedly advised him of his rights.... ” Thompson attempted to distinguish Oregon v. Elstad, 470 U.S. 298, 105 S.Ct. 1285, 84 L.Ed.2d 222 (1985) — which had approved the admission of a warned confession even though the Miranda advisement was preceded by an unwarned confession — on the ground that deliberate police misconduct in his case warranted a stricter approach. Specifically, he noted that the U.S. Supreme Court had recently heard argument in Missouri v. Seibert, 542 U.S. 600, 124 S.Ct. 2601, 159 L.Ed.2d 643 (2004). In that case, Thompson pointed out, the Missouri Supreme Court had suppressed postwarning statements because “police had purposefully withheld Miranda warnings to obtain the defendant’s breakthrough admission.” Thompson argued that his was a “very similar” case to Seibert: In both, he maintained, “the detectives deliberately withheld Miranda advisements until after appellant admitted he was holding the knife when the victim was stabbed.” (Emphasis added.) Thompson’s petition concluded by urging the California Supreme Court to “grant review to clarify the application of the fruit of the poisonous tree doctrine in the Miranda context under the forthcoming opinion by the U.S. Supreme Court, or transfer appellant’s case to the [California] Court of Appeal with directions to apply the new rule to be stated by the U.S. Supreme Court in its forthcoming opinion.”

Thompson thus fairly presented to the California Supreme Court not only the substance of his claim — that admission of his statements violated the Fifth Amendment — but the very same argument that he now makes on federal habeas. The only difference is that, as will appear, Thompson’s position has now been adopted by the U.S. Supreme Court.

Despite the clarity of Thompson’s position in the California Supreme Court, California nonetheless contends that the emergence of new Supreme Court authority— Missouri v. Seibert, 542 U.S. 600, 609, 124 5.Ct. 2601, 159 L.Ed.2d 643 (2004) — in support of Thompson’s position renders his claim unexhausted. According to the state, once Seibert was decided and even though he expressly asked the California Supreme Court to apply the precedent from the pending Seibert case, Thompson was required to file another petition or lose the right to come to federal court on the issue on habeas.

For this unlikely proposition California relies on Blair v. California, 340 F.2d 741, 745 (9th Cir.1965). In Blair, the habeas petitioner argued to the California Supreme Court in his 1961 petition for review that the deprivation of counsel on direct appeal violated the federal Constitution, and the petition was denied. In 1963, the U.S. Supreme Court endowed the right to counsel on appeal with federal constitutional protection for the first time in Douglas v. California, 372 U.S. 353, 364, 83 S.Ct. 814, 9 L.Ed.2d 811 (1963). Blair sought federal habeas relief in light of Douglas. In response to Blair’s federal petition, California argued then, as it does now, that the claim was unexhausted. We agreed, stating,

The general issue as to Blair’s constitutional right to counsel on his appeal to the California District Court of Appeal was presented to that court and decided adversely to him. But the particularized issue which Blair now raises is whether that ruling is correct in the light of the subsequently-decided Douglas v. California.

Blair, 340 F.2d at 744.⁶

The requirement of re-exhaustion imposed in Blair was expressly premised, in part, on the fact that the state “judgment ... became final prior to the decision in Douglas.” Id. at 744. We reasoned that, in that situation, the state had an interest in passing in the first instance on the question whether Douglas should be applied retrospectively. Id.

Thompson’s conviction, by contrast, did not become final until after Seibert was decided, when the period for Thompson to petition the U.S. Supreme Court for certiorari elapsed. See Caspari v. Bohlen, 510 U.S. 383, 390, 114 S.Ct. 948, 127 L.Ed.2d 236 (1994); see also Clay v. United States, 537 U.S. 522, 527, 123 S.Ct. 1072, 155 L.Ed.2d 88 (2003) (adhering to the “long-recognized, clear meaning” of finality articulated in Caspari in deciding when a judgment of conviction becomes final for purposes of a different provision of AEDPA, 28 U.S.C. § 2255). More importantly, Thompson expressly requested that the California Supreme Court apply Seibert to his petition for review, either by delaying decision or by remanding to the state Court of Appeals. So, unlike in Blair, the state courts chose to forgo the opportunity to evaluate a well-articulated position in light of a then-pending, soon-to-be-decided U.S. Supreme Court case. The interests of comity and judicial efficiency underlying the exhaustion requirement on federal habeas, see Granberry v. Greer, 481 U.S. 129, 134-35, 107 S.Ct. 1671, 95 L.Ed.2d 119 (1987), are little served by requiring a petitioner to re-exhaust his claims, already fully explained to the state court, where that court elected to decide the issue without the benefit of forthcoming authority that was brought to its attention.

Moreover, although Elstad did not require the adoption of Thompson’s legal argument about the validity of confessions where Miranda warnings are deliberately delayed, nothing in Elstad precluded the California Supreme Court from accepting that position, even before the U.S. Supreme Court opinion in Seibert. The Missouri Supreme Court had already done so in Seibert itself, 93 S.W.3d at 706-07, and the U.S. Supreme Court opinions in Seibert made clear that the deliberate delay rule is fully compatible with Elstad. See 542 U.S. at 614, 124 S.Ct. 2601(plurality op.) (“[T]he argument [that Elstad approved a question-first strategy] disfigures that case.”); id. at 620, 124 S.Ct. 2601(Kennedy, J., concurring) (“Elstad was correct in its reasoning and its result.”).

Accordingly, Seibert was not such a fundamentally new rule that we should require reexhaustion of a legal position already fully presented to the state court. Given all these considerations, we decline to extend Blair’s rule to the circumstances of Thompson’s petition.

B.

We turn to the merits of Thompson’s Miranda claim, governed by the Supreme Court’s decision in Seibert.7 542 U.S. 600, 124 S.Ct. 2601. The plurality and Justice Kennedy’s concurrence in that case, read together, make clear that a deliberate two-step interrogation strategy can violate Miranda. Specifically, when police deliberately withhold warnings until after obtaining an in-custody confession, the warnings are ineffective unless the impact of the prior unwarned confession has been dissipated. Seibert, 542 U.S. at 622, 124 S.Ct. 2601 (Kennedy, J., concurring); United States v. Williams, 435 F.3d 1148, 1158 (9th Cir.2006) (“[Justice Kennedy’s] narrower test . . . represents Seibert’s holding.”).⁸

The California Court of Appeals, in the last reasoned state court decision on Thompson’s Miranda claim, did not apply the rule announced in Seibert. Rather, based on its reading of Oregon v. Elstad, 470 U.S. 298, 105 S.Ct. 1285, 84 L.Ed.2d 222 (1985), the state appellate court ruled that “so long as the earlier [unwarned confession] was not involuntary due to police coercion, [a]' subsequent voluntary, warned statement is admissible.” The state court assumed, therefore, that deliberately delayed Miranda warnings are always effective absent actual police coercion. Because this “ ‘rule ... contradicts the governing law set forth [by the Supreme Court]’ ” in Seibert, see Rios v. Garcia, 390 F.3d 1082, 1084 (9th Cir.2004) (quoting Williams v. Taylor, 529 U.S. 362, 405-06, 120 S.Ct.. 1495, 146 L.Ed.2d 389 (2000)), the state court’s decision was “contrary to ... clearly established Federal law, as determined by the Supreme Court of the United States.” 28 U.S.C. § 2254(d)(1). We therefore “review the substantive constitutionality of the state custody de novo.” Frantz v. Hazey, 533 F.3d 724, 737 (9th Cir.2008) (en banc).

1.

After Seibert, to determine whether post-confession warnings are effective, courts must first assess whether the two-step interrogation was a delibérate strategy:

[I]n determining whether the interrogator deliberately withheld the Miranda warning, courts should consider whether objective evidence and any available subjective evidence, such as an officer’s testimony, support an inference that the two-step interrogation procedure was used to undermine the Miranda warning.... Once a law enforcement officer has detained a suspect and subjects him to interrogation ... there is rarely, if ever, a legitimate reason to delay giving a Miranda warning until after the suspect has confessed. Instead, the most plausible reason for the delay is an illegitimate one, which is the interrogator’s desire to weaken the warning’s effectiveness.

Williams, 435 F.3d at 1158-59. Neither the state court nor the federal district court made a factual determination whether the warnings given to Thompson were deliberately withheld,⁹ so we examine the extensive record before us on appeal for evidence bearing on the question.¹⁰

We begin from the state court finding, which California does not contest, that Thompson’s interrogation became custodial before he admitted to any wrongdoing. By that point in the interrogation, Thompson had been at the police station for between six and seven hours. The officers had gone forward with their investigation of Thompson’s involvement, including talking to Bivins’ mother and Thompson’s father about him, showing his photo to neighbors, talking to his probation officer, and searching his home.

By the time of the interrogation, the officers regarded Thompson as the prime suspect. The officers then employed sophisticated interrogation techniques over the course of more than an hour in an admittedly purposeful attempt to “keep the interview going” and obtain incriminating statements.¹¹

Even after Thompson began to incriminate himself in the face of these techniques, the officers still did not administer warnings. Rather, they did so only after Thompson admitted to slitting Bivins’ throat.¹²

After giving the warnings, the officers used Thompson’s prior admissions to elicit further detail and hold him to his story: When Thompson claimed he slit Bivins’ throat to prevent her from suffering, Conaty corrected him based on a pre-Afiranda warning admission: “That, and you didn’t want her to necessarily survive and tell on you, isn’t that right?” Additionally,

Officer Giacomelli repeatedly referred back to Thompson’s prewarning account in framing postwarning questions.

The only reasonable inference from this interrogation sequence is that the officers deliberately withheld Miranda warnings until after obtaining a confession.

2.

Seibert directs that we proceed to determine whether the deliberately delayed warnings administered to Thompson were nonetheless effective in apprising him of his rights. 542 U.S. at 622, 124 S.Ct. 2601 (Kennedy, J., concurring); see Williams, 435 F.3d at 1160. Williams summarized the factors relevant to this determination:

(1) the completeness and detail of the prewarning interrogation, (2) the overlapping content of the two rounds of interrogation, (3) the timing and circumstances of both interrogations, (4) the continuity of police personnel, (5) the extent to which the interrogator’s questions treated the second round of interrogation as continuous with the first and (6) whether any curative measures were taken.

Id; see also Seibert, 542 U.S. at 615, 124 S.Ct. 2601 (plurality op.); id. at 622, 124 S.Ct. 2601 (Kennedy, J., concurring).

The failure of law enforcement to take any curative measures may be dispositive of the inquiry into the effectiveness of delayed warnings. Justice Kennedy adopted that rule in his concurring opinion in Seibert: “When an interrogator uses this deliberate, two-step strategy, predicated upon violating Miranda during an extended interview, postwarning statements that are related to the substance of prewarning statements must be excluded absent specific, curative steps.”¹³ Seibert, 542 U.S. at 621, 124 S.Ct. 2601 (Kennedy, J., concurring); see also United States v. Reyes-Bosque, 596 F.3d 1017, 1031(9th Cir.2010) (so characterizing the holding of Seibert), petition for cert. filed, (Jun. 28, 2010) (No. 10-5140) and petition for cert. filed, (Jul. 28, 2010) (No. 10-5718). And nothing in the plurality opinion suggested that, once it is determined that police have employed that deliberate tactic, a post-Miranda confession should be admitted ab-

sent curative measures.¹⁴ Indeed, Justice Breyer, who joined the plurality opinion in full, was of the view that the fruits of a deliberate two-step strategy should always be suppressed. Seibert, 542 U.S. at 617, 124 S.Ct. 2601.

We need not decide in this case the precise relationship among the Williams factors. Here, every factor weighs in favor of suppression of Thompson’s first postwarning confession.

The prewarning interrogation was highly confrontational and detailed; the two sessions took place in the same small interrogation room, back-to-back, with no break at all; the police personnel were exactly the same; and, as described above, the officers’ questioning treated the two sessions as continuous and drew, in one instance, on Thompson’s pre-Miranda statement during the second session to ensure that the earlier inculpatory material was reiterated after the requisite warnings were given. And the police took no curative measures whatsoever. The post-confession Miranda warnings could not have been effective in meaningfully apprising Thompson of his rights and enabling him to invoke them.

The second set of warnings, administered the next morning at the jail, before the videotaped reenactment of the crime, presents a closer question. Still, after careful consideration, we are convinced that all of the factors continue to point to the conclusion that it too was ineffective. The completeness and detail of the prewarning interrogation remained unchanged from the time of the first, ineffective, warnings. If anything, Thompson would have perceived the invocation of his rights as even more futile the next morning, having in the interim confessed to murder a second time and shown the inspectors — in the early morning hours after the completion of the late-night interrogation at the station — the place where he tried to dispose of the evidence. In addition, there was almost complete overlap in content between Thompson’s first two confessions and the reenactment he was to conduct at Bivins’ house later that day. Indeed, the inspectors consistently treated the reenactment as continuous with the previous night’s interrogation, making clear to Thompson before allowing him to go to sleep the night before that he would need to participate in the reenactment the next day¹⁵ and telling him immediately before the reenactment, “[A]ll we’re gonna do is what we talked about yesterday, is go through what happened.”

The timing and circumstances of the second set of warnings, particularly the break in time and change in location, were somewhat more conducive to a knowing and intelligent waiver than in the case of the first warnings. But on balance, this factor does not support the conclusion that the warnings were effective either. At the conclusion of the previous night’s interrogation at around 2:00 a.m., Thompson accompanied the police to search for the murder weapon and his bloodied clothing. Afterwards, still distraught and suicidal, he spent the rest of the night shackled to the floor of a suicide-watch room at the main detention facility in Martinez. Stripped to his underwear and deprived of blankets or a bed, Thompson was too cold to sleep.

It was there, at the main detention facility, that Inspectors Conaty and Giacomelli administered the second set of warnings the next morning. Thompson thus spent the night “isolated in an ‘unfamiliar,’ ‘police-dominated atmosphere,’ Miranda, 384 U.S. [at 456-57], 86 S.Ct. 1602, where his captors ‘appeared] to control [his] fate,’ Illinois v. Perkins, 496 U.S. 292, 297, 110 S.Ct. 2394, 110 L.Ed.2d 243 [(1990)].” Maryland v. Shatzer, — U.S. -, 130 S.Ct. 1213, 1220-21, 175 L.Ed.2d 1045 (2010). Under the circumstances, the short break in time and minor change in location did not provide an opportunity for “further deliberation in familiar surroundings,” see id. at 1221, and do not weigh in favor of finding the warnings effective.

Moreover, there was complete continuity of police personnel during the first confession, the first warning, the second confession, and the second warning. Just as was so the night before, Thompson was alone with Inspectors Conaty and Giacomelli in a jailhouse room when he received these warnings. Faced with the same two people to whom he had repeatedly confessed, Thompson would have found absurd the suggestion that he retained a meaningful right to “remain silent.”

Finally, the inspectors failed once more to take any curative measures at all. Particularly after Thompson had already incriminated himself in several unwarned or improperly warned interactions with the inspectors, it was incumbent upon them to give “an additional warning that explained] the likely inadmissibility of the prewarning custodial statement].” Seibert, 542 U.S. at 622, 124 S.Ct. 2601 (Kennedy, J., concurring).

In light of all these circumstances, we have little difficulty concluding on de novo review that the officers’ deliberate two-step interrogation strategy rendered ineffective the Miranda warnings administered to Thompson.¹⁶ The admission of Thompson’s inculpatory statements at trial was reversible error unless harmless.

C.

Because the “[e]rroneous admission of a confession does not constitute structural error,” Williams, 435 F.3d at 1162, such admission is harmless on collateral review unless it had a substantial and injurious effect or influence in determining the jury’s verdict. Brecht v. Abrahamson, 507 U.S. 619, 637-38, 113 S.Ct. 1710, 123 L.Ed.2d 353 (1993). A confession, however, is ordinarily the most persuasive evidence that can be admitted against a criminal defendant. See Arizona v. Fulminante, 499 U.S. 279, 296, 111 S.Ct. 1246, 113 L.Ed.2d 302 (1991). When a jury considers a full confession that “discloses the motive for and means of the crime,” there is a high probability that the jury will rely on that evidence alone in reaching its decision. Fulminante, 499 U.S. at 296, 111 S.Ct. 1246. Admission of such confessions will seldom be harmless. Williams, 435 F.3d at 1162.

Here, Thompson’s confession was the heart of the prosecution’s case. The jury watched video recordings of Thompson’s thorough confession and of his vivid reenactment of the brutal crime. While other evidence at trial suggested that Thompson had a motive to kill Bivins, no other evidence identified him as the killer. Without the confession, the prosecution had at best a weak circumstantial case, based on motive and opportunity. The erroneous admission of Thompson’s inculpatory statements was clearly prejudicial.

III.

In affirming the admission of Thompson’s confessions, the California Court of Appeals applied a rule contrary to that announced by the Supreme Court in Missouri v. Seibert, 542 U.S. 600, 124 S.Ct. 2601, 159 L.Ed.2d 643 (2004). On de novo review, we conclude that police officers’ deliberate withholding of Miranda warnings until after Thompson confessed rendered the belated warnings ineffective. Because the introduction at trial of Thompson’s confession was both constitutionally infirm and highly prejudicial, we reverse the district court’s denial of Thompson’s petition for a writ of habeas corpus.

REVERSED.

1 . In her dissent, Judge Ikuta notes:

The majority chooses to ignore this principle on the ground that, on appeal, the government did not argue that Seibert was inapplicable under AEDPA. Thompson, 621 F.3d at 1015-16 n. 7. But this does not change § 2254(d)(l)’s statutory mandate that the federal court cannot grant habeas relief unless the state court's decision is contrary to or an unreasonable application of clearly established Supreme Court precedent, which the Supreme Court defines as “the governing legal principle or principles set forth by the Supreme Court at the time the state court renders its decision.” Andrade, 538 U.S. at 71-72, 123 S.Ct. 1166. Although we may deny a petitioner's writ in light of Supreme Court precedent decided after the relevant date, see Smith v. Spisak, -U.S. -, 130 S.Ct. 676, 680-81, 175 L.Ed.2d 595 (2010); Danforth v. Minnesota, 552 U.S. 264, 288, 290-91, 128 S.Ct. 1029, 169 L.Ed.2d 859 (2008), the Supreme Court has never held that a reviewing court may ignore the statutory command of § 2254(d) and grant habeas relief when the state court’s decision was consistent with then-applicable Supreme Court precedent.

621 F.3d at 1023-24 (footnote and parallel citations omitted).

2 . Seibert was decided after the California Court of Appeal issued its decision and after the California Supreme Court denied Thompson's direct appeal, but before the time elapsed for Thompson to file a petition for certiorari to the United States Supreme Court directly appealing his conviction. Thus, under the rule of Teague v. Lane, 489 U.S. 288, 109 S.Ct. 1060, 103 L.Ed.2d 334 (1989), which limits the applicability of new Constitutional rules of criminal procedure on de novo habeas review, his conviction was not yet "final.” See Caspari v. Bohlen, 510 U.S. 383, 390, 114 S.Ct. 948, 127 L.Ed.2d 236 (1994) (explaining that a state court conviction becomes "final” for Teague purposes when “the availability of direct appeal to the state courts has been exhausted and the time for filing a petition for a writ of certiorari has elapsed or a timely filed petition has been finally denied”).

3 . To add to the confusion, both Justice Sevens and Justice O’Connor were writing for the Court in the respective parts of their opinions.

4 . At the very least, we should have delayed the decision in this case to allow the Supreme Court to answer this question directly. The issue is now squarely before the Court in Greene v. Fisher, - U.S. -, 131 S.Ct. 1813, 179 L.Ed.2d 772 (mem) (2011) and, in fact, is the only issue in that appeal.

5 . The majority reaches this conclusion by first characterizing the state court's opinion as assuming "that deliberately delayed Miranda warnings are always effective absent actual police coercion.” 621 F.3d at 1016. This does not appear to be a fair reading of the California Court of Appeal's application of Elstad to this case. Rather, it presages the majority’s improper factual finding that the officers engaged in a deliberate two-step interrogation strategy. 621 F.3d at 1018, 1021.

6 . In Marks v. United States, 430 U.S. 188, 193, 97 S.Ct. 990, 51 L.Ed.2d 260 (1976), the Supreme Court reiterated that "[w]hen a fragmented Court decides a case and no single rationale explaining the result enjoys the assent of .five Justices, the holding of the Court may be viewed as that position taken by those Members who concurred in the judgments on the narrowest grounds.” (internal quotation marks and citation omitted).

7 . Justice O’Connor’s dissent was joined by Chief Justice Rehnquist, Justice Scalia, and Justice Thomas.

8 . In his concurring opinion, Justice Kennedy stated:

The plurality concludes that whenever a two-stage interview occurs, admissibility of the postwarning statement should depend on "whether [the] Miranda warnings delivered midstream could have been effective enough to accomplish their object” given the specific facts of the case. This test envisions an objective inquiry from the perspective of the suspect, and applies in the case of both intentional and unintentional two-stage interrogations. • In my view, this test cuts too broadly.

Seibert, 542 U.S. at 621-22, 124 S.Ct. 2601 (citations omitted).

1 . This too, the officers later admitted, was a technique employed during the pr e-Miranda phase of the interrogation to induce Thompson to incriminate himself — minimizing the moral blameworthiness of Thompson's conduct.

2 . When, by this technique, the officers succeeded in inducing Thompson to admit that he had gone alone to Bivins’ house, they had their first “breakthrough.” As the Missouri Supreme Court observed in State v. Seibert, "[[Interrogators are taught that procuring the first admission” — "no matter how small”'— "usually leads to a full confession.” 93 S.W.3d 700, 704 (Mo.2002) (citing Arthur S. Aubry, Jr. & Rudolph R. Caputo, Criminal Interrogation 290 (1980)).

3 .By this point, the state trial court found, the interrogation had become custodial.

4 . See Stansbury v. California, 511 U.S. 318, 322, 114 S.Ct. 1526, 128 L.Ed.2d 293 (1994).

5 . The district court stopped short of deciding whether the two-step interrogation was a deliberate tactic in this instance. Instead, the court reasoned as follows:

This Court notes that Petitioner does not cite to the record or present any evidence in support of his assertion that the inspectors in this case deliberately withheld their Miranda advisement until Petitioner had incriminated himself. Moreover, unlike the defendant in Seibert, Petitioner was not under arrest at the time of questioning by the inspectors. There is no evidence in the record concerning an official police policy of deliberately withholding Miranda warnings until a suspect has confessed. In contrast to the police officers in Seibert, the inspectors here did not testify that' they withheld Miranda warnings until after Petitioner confessed. Under these circumstances, the state appellate court's analysis under Elstad was not an unreasonable application of the law.

6 . Subsequent decisions have narrowed Blair. Pope v. Harper rejected its application to a situation similar to Thompson’s: The state courts in Pope conducted a harmless error analysis without the benefit of Chapman v. California, 386 U.S. 18, 87 S.Ct. 824, 17 L.Ed.2d 705 (1967), decided after the California Supreme Court denied Harper's petition for review. 407 F.2d 1303 (9th Cir.1969). The federal district court ordered Harper to return to state court to allow for reconsideration in light of Chapman. Id. at 1304-05. We reversed, concluding that Blair is inapplicable and re-exhaustion unnecessary when any intervening change in federal law concerns only a “narrow issue controlled by established federal principles.” Id. at 1305; see also Hudson v. Rushen, 686 F.2d 826, 830 (9th Cir. 1982) ("[Ejxhaustion of remedies does not require that the state have had the opportunity to pass on the claim under the particular authorities advanced in the federal habeas court.”).

The Supreme Court, after Pope, characterized Blair as a narrow rule that applies to the rare case where "an intervening change in federal law cast the legal issue in a fundamentally different light....” Picard v. Connor, 404 U.S. 270, 276, 92 S.Ct. 509, 30 L.Ed.2d 438 (1971).

There is, indeed, some question whether the Blair rule remains viable at all after Teague v. Lane, 489 U.S. 288, 109 S.Ct. 1060, 103 L.Ed.2d 334 (1989). Butler v. Curry declared that "[ajfter Teague, an intervening change in federal law that casts the legal issue in a fundamentally different light is a 'new rule' that cannot be applied on collateral review under any circumstances, regardless of whether the petitioner has exhausted his state court remedies. In other words, after Teague, [Blair] no longer serves any function.” 528 F.3d 624, 639 (9th Cir.2008). This entombment of Blair after Teague is a bit of an overstatement. As this case illustrates, there is a small set of cases in which Teague is not triggered because the case is still pending on direct appeal but a pertinent Supreme Court decision issues after the state courts have completed their consideration of the case. Still, the number of such instances is likely to be exceedingly small, indicating that the justification for the Blair rule was indeed largely obviated by Teague.

7 . We note that California does not argue that Thompson's reliance on Seibert, decided after the California Supreme Court denied Thompson's petition for review but before the period to file a petition for certiorari with the U.S. Supreme Court had expired, is precluded by a rule of nonretroactivity under AEDPA. The State acknowledged as much in its Petition for Rehearing with Suggestion for Rehearing En Banc at 9 & n.3: "We acknowledged at oral argument we did not raise this issue.... We argued that because the state courts never had the opportunity to address Seibert, it was unexhausted.” The dissent agrees, acknowledging that "[i]nstead of arguing that Seibert was not 'clearly established' Supreme Court precedent relevant to Thompson’s claim under AEDPA, the government argued that Thompson failed to exhaust his claim that the state court’s decision was inconsistent with Seibert.” Dissent at n. 1.

California’s briefing in this court stated, citing Justice Stevens’ opinion for the Court in Williams v. Taylor, 529 U.S. 362, 390, 120 S.Ct. 1495, 146 L.Ed.2d 389 (2000), that the law to be applied under AEDPA is that "clearly established at the time [Thompson’s] state court conviction became final," and counsel for the state agreed at oral argument that the state "more than didn’t raise [a non-retroactivity objection to reliance on Seibert ]; [it] assumed its nonexistence.” The U.S. Supreme Court recently noted "some uncertainty" surrounding whether clearly established law is assessed as of the date the state judgment of conviction becomes final or as of the date of the relevant state-court decision. Smith v. Spisak,-U.S.-, 130 S.Ct. 676, 681, 175 L.Ed.2d 595 (2010) (citing Williams, 529 U.S. at 390, 120 S.Ct. 1495 (Stevens, J., for the Court) and id. at 412, 120 S.Ct: 1495 (O’Connor, J., for the Court)). In Smith, however, the Court declined to address this ambiguity because in that case, as in this one, the parties had not raised the issue. See id. We follow the same course here and do not address the merits of this procedural question, which is not properly before us.

8 . Williams held that, under the logic of Marks v. United States, 430 U.S. 188, 193, 97 S.Ct. 990, 51 L.Ed.2d 260 (1977), Justice Kennedy's test is the holding of the Court in Seibert. 435 F.3d at 1157-58. Accord United States v. Narvaez-Gomez, 489 F.3d 970, 973-74 (9th Cir.2007). Because Justice Kennedy’s rule is the holding of Seibert under the Marks rule, it is clearly established Supreme Court law for purposes of 28 U.S.C. §. 2254. See Panetti v. Quarterman, 551 U.S. 930, 948-49, 127 S.Ct. 2842, 168 L.Ed.2d 662 (2007) (applying Marks to a splintered decision of the Court and concluding that the narrowest rule "constitutes clearly established law for purposes of § 2254” (internal quotation marks omitted)).

9 .The district court did determine that there was no outright admission by the officers that they were withholding warnings until Thompson confessed and that there was no evidence of a departmental policy encouraging that practice. On the basis of those determinations, the district court ruled that the state courts did not unreasonably apply Seibert. Williams made clear, however, that courts must consider circumstantial, as well as direct, evidence that the withholding of warnings is deliberate, because " 'the intent of the officer will rarely be ... candidly admitted.' " 435 F.3d at 1158-59 (quoting Seibert, 542 U.S. at 617, 124 S.Ct. 2601 (plurality opinion)). "Once a law enforcement officer has detained a suspect and subjects him to interrogation ... the most plausible reason for ... delay is an illegitimate one, which is the interrogator’s desire to weaken the warning’s effectiveness.” Williams, 435 F.3d at 1159. It was legal error for the district court to conclude that the absence of departmental policy or outright admissions of deliberate intent ends the inquiry under Seibert.

As the dissent notes, the district court also "found that Thompson ‘d[id] not cite to the record or present any evidence in support of his assertion that the inspectors in this case deliberately withheld their Miranda advisement until [Thompson] had incriminated himself.’ ” Dissent at 809. This conclusion by the district court constitutes an account of the course of proceedings in that court, but it is not a factual determination regarding whether the warnings were deliberately withheld. Moreover, if viewed as a finding of fact, the district court’s statement regarding the state of the record is clearly erroneous. As summarized in the text, there was ample evidence indicating that the officers deliberately withheld warnings until Thompson confessed, although that evidence does not include a direct admission by the officers. Viewed in this light, the district court’s statement that there was not "any evidence” reflects the same underlying legal error just discussed — namely, the understanding that there must be a policy or an outright admission to meet Seibert's deliberate intent standard.

10 . The record includes, inter alia, complete video and audio recordings of the interrogation, confession, and reenactment and testimony from the state court’s evidentiary hearing describing the warnings administered to Thompson off-camera on the morning of June 23.

11 . On direct examination at trial, Inspector Conaty and his attorney engaged in the following colloquy:

Q: And why did you, during the course of that initial interview with Mr. Thompson, lie to him?

A: Telling an interview subject lies, telling them about circumstances may not in fact be true [are] accepted interrogative techniques, it keeps the interview going.

Guilty persons will often attempt to explain these circumstances which in itself furthers the interview.

S{{ 5¡í ij! ‡

It’s human nature to, when admitting to something that is believed to be a reprehensible act, to minimize one’s conduct. If I offer up that opportunity in the very beginning, oftentimes the interview subject will immediately — lack of a better way to describe it — bite on that and will expound, will continue to minimize his own conduct, attempting to place the responsibility for whatever occurred on other persons. In and of itself, it just furthers the interview. Q: Keeps the interview going?

A: Yes. Actually, in a lot of cases, it will— even in an attempting to minimize the conduct will have the defendant place themselves at a location when, in fact, we have no evidence necessarily independently to place them there.

12 . The entire interview took place in an interrogation room at the station and was videotaped. So the delay in administering Miranda was not attributable to the need to accommodate the demands of a unique situation in the field. Cf. Elstad, 470 U.S. at 315-16, 105 S.Ct. 1285(suggesting that officers might have refrained from giving warnings during a preliminary interview in the suspect's living room to avoid causing alarm to the suspect’s mother).

Nor is this in any other respect the exceptional case in which a "legitimate reason” justified withholding warnings until after obtaining a confession. Williams, 435 F.3d at 1159. In its brief before us, California suggests that, as in Elstad, 470 U.S. at 315-16, 105 S.Ct. 1285, the delay here may be explained by the interrogating officers’ uncertainty over whether the interrogation had become custodial. (The officers themselves did not testify to any such explanation.) But, unlike in Elstad, the same interrogation circumstances that prevailed at the time the state court determined Miranda warnings should have been given persisted at the time they actually were given: the location was the same, Thompson had been at the station for many hours, he had not been handcuffed or formally arrested, and the same officers were interviewing him. And although the warnings followed shortly after Thompson gave his most detailed account of the crime, Thompson by that point had already made several highly incriminating statements that did not trigger any warnings. Thus, at the time warnings finally were given, there was no reason to think Thompson was any more or less free to leave than he was before. See United States v. Beraun-Panez, 812 F.2d 578, 580 (9th Cir.1987) (holding that the determination of custody turns on whether " 'a reasonable person in [the] circumstances would conclude ... that he or she would not be free to leave' ”) (quoting United States v. Hudgens, 798 F.2d 1234, 1236 (9th Cir.1986) (alteration omitted)), modified by 830 F.2d 127 (9th Cir. 1987). Any uncertainty regarding whether Thompson was in custody would not explain the delay in complying with Miranda.

13 . Justice Kennedy suggested in Seibert that “a substantial break in time and circumstances” may qualify as a curative measure that restores the effectiveness of a deliberately delayed warning, if the break is "designed to ensure that a reasonable person in the suspect's situation would understand the import and effect of the Miranda warning.” Id. at 622, 124 S.Ct. 2601 (Kennedy, J., concurring).

14 . Because the plurality opinion did not hinge on a finding of deliberate withholding of warnings, the Justices in the plurality had no occasion to address the question.

15 . Inspector Conaty testified that he and Thompson "reached an agreement” "during the course of the interview” and prior to Thompson's transfer to the main detention facility for the night that Thompson would participate in the reenactment.

16 . We do not reach Thompson's alternative argument that his waiver of his right to remain silent was not voluntary because it was the product of police overreaching. See Seibert, 542 U.S. at 617 n. 8, 124 S.Ct. 2601 (“Because we find that the warnings were inadequate, there is no need to assess the actual voluntariness of the statement.”).