Patrick James Jeffries v. Tana Wood, Superintendent

KOZINSKI, Circuit Judge,

with whom Judges GOODWIN, BRUNETTI, T.G. NELSON and HAWKINS join, dissenting.

I

If you listen closely, perhaps you too can hear what the majority hears: the sounds of silence. Congress said nothing about section 2254(d)’s temporal scope anywhere in Chapter 153 of Title I of the Antiterrorism and Effective Death Penalty Act of 1996, Pub.L. No. 104-132, 1996 U.S.C.C.A.N. (110 Stat.) 1214. Under Landgraf v. USI Film Prods., 511 U.S. 244, 280 , 114 S.Ct. 1483, 1505 , 128 L.Ed.2d 229 (1994), we must therefore inquire whether section 2254(d) has a retroactive effect; if it does not, it must be applied to pending cases. The majority struggles to avoid this inquiry because it leads to the inescapable conclusion that section 2254(d) applies to this case. See, e.g., Drinkard v. Johnson, 97 F.3d 751, 764-66 (5th Cir.1996), cert. denied, — U.S.-, 117 S.Ct. 1114 , 137 L.Ed.2d 315 (1997); Lindh v. Murphy, 96 F.3d 856, 861-67 (7th Cir.1996) (en banc), cert. granted, — U.S. -, 117 S.Ct. 726 , 136 L.Ed.2d 643 (1997). Thus, the majority must make the unusual argument that Congress expressed itself on the retroactivity question, even though it said nothing at all about it. This oxymoronic conclusion is wrong for two reasons. First, it’s based on a misreading of Landgraf . Second, it’s just not true; there’s no evidence Congress intended anything.

A

The cornerstone of the majority’s argument is its assertion that “Landgraf altered the legal landscape so that prospective application has unquestionably become the default rule.” Maj. op. at 1494. According to the majority, when Congress did not address the retroactivity question in passing Chapter 153, it did so with this default rule in mind, and thus meant to endorse prospective application only. But this is not what Landgraf *1502 said, so it’s hardly what Congress could have had in mind. 1

In fact, Landgraf was designed to preclude this precise argument. In Landgraf the Supreme Court confronted a perceived conflict between two judicially-created rules for interpreting statutes that do not specify a temporal reach: the presumption against retroactive application and the contrary presumption that each case is governed by the law in effect at the time of decision. See 511 U.S. at 264 , 114 S.Ct. at 1496 . Courts often broke this inferential equipoise by following one of the presumptions and ignoring the other, much like the majority does today. Landgraf tried to fix this problem by explaining that the presumptions apply to different types of laws, and thus don’t really conflict. Courts must first figure out which kind of law is at issue, and then apply the appropriate presumption.

The key to understanding Landgraf therefore is grasping the difference between a “retroactive law” and, what our court has called, a “retrospective law.” See United States ex rel. Lindenthal v. General Dynamics Corp., 61 F.3d 1402, 1407-08 (9th Cir.1995), ce rt. denied, — U.S.-, 116 S.Ct. 1319 , 134 L.Ed.2d 472 (1996). Contrary to popular misconception, not all laws that apply to pending cases are “retroactive.” See Landgraf, 511 U.S. at 269-70 , 114 S.Ct. at 1499 . Only a law that would have a “retroactive effect,” i.e., that would impair substantive rights on which a party relied when it acted or that would upset settled expectations, is deemed retroactive. General Dynamics, 61 F.3d at 1407-08 . Laws that have no “retroactive effect” when applied to pending cases are deemed, in our circuit, “retrospective” laws. Id. at 1408 . The presumption against retroactive laws is only triggered by “the class of new statutes that would have genuinely ‘retroactive’ effect.” Landgraf, 511 U.S. at 277 , 114 S.Ct. at 1503 . This makes not only semantic sense, but common sense as well: Courts are leery of retroactive laws because they can be very unfair, so the presumption is triggered only when a “retroactive effect” is threatened. Id. at 272 , 114 S.Ct. at 1500 . If a statute has no retroactive effect, the opposite presumption kicks in: The law in effect at the time of decision applies, and thus retrospective laws are presumed to apply to pending cases. Id. at 277 , 114 S.Ct. at 1503 ; see also Bradley v. Richmond Sch. Bd., 416 U.S. 696, 711 , 94 S.Ct. 2006, 2016 , 40 L.Ed.2d 476 (1974). As a result, before courts apply judicial default rules, they must determine whether a statute would have a retroactive effect. See, e.g., General Dynamics, 61 F.3d at 1407-08 ; United States v. $814,254.76 in U.S. Currency, 51 F.3d 207, 211-12 (9th Cir.1995).

So, does Section 2254(d) have a retroactive effect? The answer is plainly no. While we won’t repeat the arguments made by the Eleventh, Fifth, and Seventh Circuits, see Hunter v. United States, 101 F.3d 1565, 1571-73 (11th Cir.1996) (en banc), cert. denied, — U.S. -, 117 S.Ct. 1695 , 137 L.Ed.2d 822 (1997); Drinkard, 97 F.3d at 766 ; Lindh, 96 F.3d at 863-68 , we note that section 2254(d) is paradigmatic of laws lacking any retroactive effects. Landgraf singled out three types of laws that generally have no retroactive effect: laws that (1) “authorize[ ] or affect[ ] the propriety of prospective relief’; (2) “confer[ ] or oust[ ] jurisdiction”; and (3) make “[cjhanges in procedural rules.” Landgraf, 511 U.S. at 273-75 , 114 S.Ct. at 1501-02 . Based on those categories, section 2254(d) scores a hat-trick: It’s (1) a procedural change, (2) jurisdictional in nature, that (3) merely regulates the power of federal courts to grant prospective relief-a writ of habeas corpus. See Lindh, 96 F.3d at 863-68 . The majority’s arguments to the contrary are fully answered by Lindh. Compare maj. op. at 1497-99 with Lindh, 96 F.3d at 863-68 . Were section 2254(d) to attach new legal consequences to past con *1503 duct, presumably Jeffries could point to something that he would have done differently had he known about those new consequences. See Landgraf, 511 U.S. at 282 , 114 S.Ct. at 1506 (a law has a retroactive effect if it’s “the type of legal change that would have an impact on private parties’ planning”). Obviously, Jeffries hasn’t argued that he killed the Skiffs relying on the state of federal habeas law as it existed before the 1996 Act. He also hasn’t pointed to anything he would have done differently in litigating his claim in the state courts had he known federal habeas review would be curtailed; he presented this exact juror misconduct claim to the state courts. See State v. Jeffries, 722 P.2d 99 , 100-01 (Wash.1986); see also Personal Restraint Petition, Washington v. Jeffries, No. 52323-1 (Wash. Dec.17,1985). The majority seems to admit that there is nothing that Jeffries would have done differently when it states that “the primary conduct implicated by federal habeas procedure is the state court proceeding.” Maj. op. 1499. But, in determining whether a law has a retroactive effect, petitioner must point to some aspect of his conduct that he undertook in reliance on the prior law.

The fact that applying section 2254(d) to Jeffries’s case may alter the outcome to his detriment doesn’t alone make it retroactive, as almost every change in law will have an outcome determinative effect-else why bother passing a law at all? Compare Boria v. Keane, 90 F.3d 36, 37 (2d Cir.) (“Assuming ... that the new statute would require a different outcome, application of the new statute to these circumstances would be retroactive.”), petition for cert. filed, 65 U.S.L.W. 3342 (Oct. 11, 1996) (No. 96-628) with Landgraf, 511 U.S. at 269 , 114 S.Ct. at 1499 (a law doesn’t have a retroactive effect “merely because it ... upsets expectations based in prior law”). That the legal consequence at stake here is the death penalty makes no difference, unless our ‘sound instincts’” tell us that death is also different when it comes to retroactivity jurisprudence. But see maj. op. at 1497-98.

We also note that the majority’s retroactive effect argument is incompatible with two recent Supreme Court decisions. According to the majority, at the moment an unconstitutional state judgment is entered, a prisoner’s right to habeas relief vests, and it would be unfair to take away this right because the law changed in the interim. See maj. op. at 1498. The majority therefore argues that the 1996 Act would have a retroactive effect if applied to cases where the state proceedings were completed before April 24, 1996. If the majority’s argument is sound, then the Court committed a grievous error in Felker v. Turpin, — U.S.-, 116 S.Ct. 2333 , 135 L.Ed.2d 827 (1996), in which it applied the 1996 Act’s bar on successive petitions to a petitioner whose state court proceedings were completed long before April 24, 1996. See id. at-, 116 S.Ct. at 2336 . Moreover, in Lockhart v. Fretwell, 506 U.S. 364 , 113 S.Ct. 838 , 122 L.Ed.2d 180 (1993), the Supreme Court rejected an argument similar to the majority’s claim that a prisoner has “settled expectations” in the state-court judgment subjecting him to incarceration. See maj. op. 1498. In Fretwell , the Court held that “[a] federal habeas petitioner has no interest in the finality of the state-court judgment under which he is incarcerated.” 506 U.S. at 373 , 113 S.Ct. at 844-45 . 2 Because section 2254(d) has no retroactive effect, it’s presumed to apply to Jeffries’s ease. See Bradley, 416 U.S. at 711 , 94 S.Ct. at 2016 .

Perhaps aware of the weakness of its argument that this is a retroactive law, the majority goes to great lengths in arguing that Congress expressly resolved the retroactivity question. See maj. op. at 1497 (“[Wje find the first stage of the Landgraf analysis con- *1504 elusive and need not reach the subsequent stages of the examination.”). But Congress said nothing about retroactivity here; the majority therefore comes up with a legal misnomer, the express implied statement. See maj. op. at 1494-97. How a statement can be both express and implied is beyond me, but the fact remains that Congress said nothing on the subject, and this is clearly insufficient under Landgraf to avoid the retroactive effect analysis; only an explicit statutory command will do:

When a ease implicates a federal statute enacted after the events in suit, the court’s first task is to determine whether Congress has expressly prescribed the statute’s proper reach____ When ... the statute contains no such express command, the court must determine whether the new statute would have retroactive effect____

511 U.S. at 280 , 114 S.Ct. at 1505 (emphasis added).

Unlike the majority, we cannot read this requirement as being satisfied by anything short of a clear statement in the statute. Perhaps the majority’s interpretation would be linguistically plausible if the Supreme Court had omitted the words “express” and “expressly” from this passage. Our responsibility then would be to determine whether Congress had prescribed the statute’s temporal reach by using the normal means of statutory construction: canons, legislative history, inference. But Justice Stevens inserted the words “express” and “expressly” in the Landgraf opinion and they must be given some meaning. In our dictionary, the adjective “express” means “clearly indicated; explicit”-whieh in turn precludes the type of inferential analysis the majority engages in at length. Since the statute contains no express-! e., no explicit-temporal command, we must proceed to determine whether the statute has a retroactive effect. 3

Assuming, however, that something short of an explicit statutory command will suffice to signal congressional intent regarding retroactivity, the evidence of congressional intent the majority cites is very thin. The majority rightly points out that Chapter 153’s statutory companion-Chapter 154, which creates special “opt-in” capital habeas procedures-was made immediately applicable to pending cases by an express statutory command. Pub.L. No. 104-132, § 107 (c), 1996 U.S.C.C.A.N. (110 Stat.) 1226. According to the majority, by explicitly making Chapter 154 retroactive but saying nothing about Chapter 153, Congress must have intended Chapter 153 to apply prospectively. Were we writing on a clean slate, this might be a pretty good argument. But Landgraf specifically rejected this sort of negative inference; in fact, our court made a very similar argument which the Court rejected. See Estate of Reynolds v. Martin, 985 F.2d 470, 474 (9th Cir.1993). We have little doubt it will do so again. 4

*1505 The majority makes much of the legislative history, which it claims supports its negative inference. But there’s nothing in the legislative record on the question of Chapter 153’s temporal scope. In comparison, Landgraf decided that the 1991 Civil Rights Act’s legislative history, which was full of many conflicting opinions about whether the Act was intended to apply to pending cases, lacked a crucial ingredient to make the negative inference persuasive: any evidence that “an agreement had been tacitly struck.” 511 U.S. at 262-63 , 114 S.Ct. at 1495-96 . We’re at a loss to figure out how a completely barren record can reveal that “an agreement had been tacitly struck” as to Chapter 153’s temporal scope.

The majority briefly cites the selective incorporation of 28 U.S.C. §§ 2254 (a),(d) & (e) of Chapter 153 into Chapter 154 as evidence that Congress intended Chapter 153 to be prospective. See maj. op. at 1495; 28 U.S.C. § 2264 (b). 5 The majority, however, never bothers to respond to Judge Easterbrook’s well-reasoned explanation of this selective incorporation: that it was designed to specify “not “when,’ but ‘which’ ” provisions of section 2254 would apply to opt-in cases. See Lindh, 96 F.3d at 862 ; see also Hunter, 101 F.3d at 1570 n. 3. Petitioner attempts to counter Judge Easterbrook’s argument by pointing out that it would mean the exhaustion rules in opt-in cases would be different from those in traditional habeas proceedings: sections 2254(b) & (c) (the traditional exhaustion rules) weren’t among those provisions incorporated into Chapter 154. The answer, it seems to me, is that Congress made a deliberate choice to create new and different exhaustion rules for opt-in cases, see 28 U.S.C. § 2264 (a), and we must respect that decision. That may put states at a disadvantage in some cases, see id. § 2264(a)(3), but that may be a price Congress intended states to pay in return for an expedited process.

B

Applying section 2254(d) to Jeffries’s case, it’s clear we cannot second guess the Washington Supreme Court’s adjudication of the juror misconduct claim. That court resolved the claim on the merits, applied the correct law and reasonably applied the law to the facts before it. See 28 U.S.C. § 2254 (d). The majority asserts the authority to disregard the Washington Supreme Court’s opinion because that court allegedly applied the wrong harmless error standard and relied on an erroneous fact. According to the majority, these mistakes strip the opinion of the protection granted by section 2254(d)’s deferential standard of review. Ironically, this holding is itself based on serious factual and legal errors.

The factual error: The majority chastises the Washington Supreme Court for failing to apply the Chapman v. California, 386 U.S. 18 , 87 S.Ct. 824 , 17 L.Ed.2d 705 (1967), harmless error standard on collateral review. See maj. op. at 1500-01. Citing Rose v. Clark, 478 U.S. 570 , 106 S.Ct. 3101 , 92 L.Ed.2d 460 (1986), the majority claims that Chapman was the clearly established collateral review standard as announced by the United States Supreme Court in 1986, the time of the Washington Supreme Court’s decision. Although Rose was a habeas case, and the opinion does refer to the Chapman standard, id. at 576, 106 S.Ct. at 3105 , the Court in Rose did not establish the correct harmless error standard for collateral review. How do we know? The Supreme Court told us so in Brecht v. Abrahamson, 507 U.S. 619 , 113 S.Ct. 1710 , 123 L.Ed.2d 353 (1993): “Although we have applied the Chapman standard in a handful of federal habeas cases, see, e.g., ... Rose v. Clark, 478 U.S. 570 [, 106 S.Ct. 3101 , 92 L.Ed.2d 460 ] (1986) ..., we have yet squarely to address its applicability *1506 on collateral review.” Id. at 630 ,113 S.Ct. at 1718 (citations omitted). Although ahead of its time, the Washington Supreme Court actually applied the correct harmless error standard. Compare Brecht, 507 U.S. at 637 , 113 S.Ct. at 1721-22 (habeas petitioner must show “substantial and injurious effect,” i.e., “actual prejudice”) with Jeffries, 722 P.2d at 100 (“petitioner must show ‘actual and substantial prejudice’ ”).

The legal error: Section 2254(d)(2) provides that federal courts need give no deference to a state court adjudication “that was based on an unreasonable determination of the facts in light of the evidence presented in the State court proceeding.” 28 U.S.C. § 2254 (d)(2) (emphasis added). Significantly, section 2254(d)(2) doesn’t require that the state court discover the ultimate truth; rather it requires the state court to do its best given the information in the record before it.

Here, the Washington Supreme Court’s observation that “[ojver two and a half years passed between the trial and the time the alleged misconduct was discovered” reflected the evidence presented in Jeffries’s collateral proceeding. See Jeffries, 722 P.2d at 100. In fact, it appears to be based on Jeffries’s own assertion:

Mr. Tyszko came forward later after being troubled by his conscience. Until he did these issues were hidden from everybody.

Reply Brief to Personal Restraint Petition, Washington v. Jeffries, No. 52323-1, at 7 n. 1 (Wash. Mar.3, 1986). Earlier in the same brief, Jeffries states that Tyszko “came forward almost two years later and told defense counsel what happened.” Id. at 4. 6 It’s therefore not surprising that the Washington Supreme Court based its ruling on the assumption that over two years passed before “the alleged misconduct was discovered.” Even the dissenting justices had the same understanding of the facts. See Jeffries, 722 P.2d at 101 (Utter, J., dissenting) (“Two years [after Jeffries’s trial], two jurors ... filed affidavits____”).

Of course, we know now that this statement was incorrect, but the evidence to the contrary wasn’t presented to the Washington Supreme Court when it was adjudicating the claim of juror misconduct. Kathleen Sims’s November 16, 1983, affidavit wasn’t part of the record in the collateral proceedings. As the majority notes, it was buried in the substantial record on direct appeal, where Jeffries hadn’t raised the juror misconduct claim. See maj. op. at 1500 n. 20. This evidence therefore cannot be considered for purposes of section 2254(d)(2), which limits federal courts to evaluating a state court’s factual determinations “in light of the evidence presented in the State court proceeding” in which “the claim” at issue was adjudicated. 28 U.S.C. § 2254 (d)(2). 7

The majority ignores this important limitation and, in so doing, threatens to render all of section 2254(d) a nullity. 8 Recall that section 2254(d)(2) creates an exception to the rule of deference established by section 2254(d). If the exception weren’t narrowly limited, petitioners could easily circumvent section 2254(d) by planting factual errors in state court dispositions. This case is a per- *1507 feet blueprint of how to do it. Not only was the Sims affidavit included in the record for a different proceeding than the one at issue here, but it was never mentioned in any of Jeffries’s briefs before the Washington Supreme Court, in neither the direct nor collateral proceedings. See maj. op. at 1500 n. 20. And, as noted above, Jeffries himself planted the factual mistake in the minds of the Washington Supreme Court Justices. See p. 1506 supra. Moreover, once the Washington Supreme Court issued its opinion rejecting his personal restraint petition, Jeffries didn’t move for rehearing in an attempt to alert the court to the mistake. If the Washington Supreme Court’s treatment of this issue in light of petitioner’s presentation is unreasonable, Congress surely wasted its time in passing section 2254(d)(2). 9

While the application of the 1996 Act ends this case for us, we go on to address the majority’s arguments on the merits because they amount to a serious misapplication of various important jurisprudential doctrines.

II

If the vague comments referring to Jeffries’s prior conviction made in the juryroom were harmful in this case, there is no case where such an error could be harmless. The majority thus has created what in practice is a per se rule of reversal for cases where information adverse to the defendant is uttered by or to one of the jurors outside the courtroom. In doing so, the majority gives birth to a mutant breed of trial errors that are “inherently prejudicial” and thus indistinguishable from structural errors. This end-run around Brecht v. Abrahamson was attempted once before in Thompson v. Barg, 74 F.3d 1571, 1577 (9th Cir.) (Reinhardt, J., dissenting), cert. denied, — U.S.-, 117 S.Ct. 227 , 136 L.Ed.2d 159 (1996), but was rebuffed. See Thompson, 74 F.3d at 1575 n. 1 (the dissent’s reasoning is “contrary to the particularized case-by-case analysis of prejudice required by Brecht”). We should do the same here. Our failure to faithfully apply the Brecht standard will likely force the Supreme Court to send us another wake-up call. Cf. California v. Roy, — U.S.-, -, 117 S.Ct. 337, 339 , 136 L.Ed.2d 266 (1996) (9th Circuit applied the wrong harmless error standard by modifying Brecht), rev’g Roy v. Gomez, 81 F.3d 863 (9th Cir.1996) (en banc).

Of course, the majority can’t announce its new creation out loud because a per se rule of reversal for non-structural errors would squarely contradict Arizona v. Fulminante, 499 U.S. 279, 307-10 , 111 S.Ct. 1246, 1263-65 , 113 L.Ed.2d 302 (1991), which holds that only structural errors merit automatic reversal on collateral review; any other error must be “quantitatively assessed in the context of other evidence presented in order to determine whether [it] was harmless----” Id. at 308 , 111 S.Ct. at 1264 . Instead, the majority reaches its goal through the back door by purporting to conduct a harmless error analysis. Sure, the majority enumerates a number of factors that could mitigate the harmful effect of the juror misconduct, but there’s no evidence the majority actually considers them. All that seems to matter is that the information was “intrinsically prejudicial.” See maj. op. at 1491.

A genuine harmless error analysis would lead the majority to the inescapable conclu *1508 sion that the extrinsic evidence here didn’t have a substantial injurious impact on the verdict. See Brecht, 507 U.S. at 623 , 113 S.Ct. at 1713-14 . First, there was “overwhelming evidence” of Jeffries’s guilt. Jeffries, 722 P.2d at 101; see Hughes v. Borg, 898 F.2d 695, 700 (9th Cir.1990). The jury deliberations were lightning quick: Guilt phase deliberations took no more than ten hours; 10 penalty phase deliberations took five hours. Moreover, the extrinsic evidence was irrelevant as to one theory of aggravation. Recall that to convict Jeffries of aggravated murder, the jury had to find that he killed in order to conceal another crime. Wash.Rev.Code § 10.95.020(7) (1983). Evidence of Jeffries’s prior robbery conviction might arguably have affected a finding that he murdered the Skiffs to conceal his robbery, but it couldn’t possibly have affected the State’s other theory, that Jeffries’s killed Mrs. Skiff to prevent her from exposing Jeffries’s earlier murder of her husband. See State v. Jeffries, 105 Wash.2d 398 , 717 P.2d 722, 741 (1986); R.T. at 3503, 3514.

Second, the substance of the information communicated wasn’t nearly as damaging as the majority claims. The district judge merely found that a juror stated something “to the effect that Jeffries either had a record or had been convicted of robbery.” Order on Remand, Jeffries v. Wood, No. C90925D, at 1-2 (Feb. 17,1995). 11 This information was extremely vague, see Thompson, 74 F.3d at 1576 , and related to a prior conviction of a veiy different sort than that charged-it’s not very likely that jurors would conclude that, because a defendant had committed a robbery, he would be predisposed to commit murder, see United States v. Field, 625 F.2d 862, 872 (9th Cir.1980) (“[T]he conviction was not for a crime sufficiently similar to [the charged crime] to raise the spectre of ‘he did it before, he could do it again.’ ”).

In fact, Jeffries’s own actions belie the assertion that information regarding his criminal record was “intrinsically prejudicial.” Although Jeffries tried to keep this information from the jury, he allowed one juror-the same juror who made the first misconduct allegations-to take a seat in the jury box even though she knew about his record. See Voir Dire of Kathleen Sims, Oct. 13, 1983, at 36-38. We must infer that Jeffries himself didn’t think the information about his prior record was so “intrinsically prejudicial” that any exposure to it would automatically taint a juror’s judgment. 12

Third, the context in which the remark was made couldn’t have been less damaging. We know this in part because no one is quite sure when and what was said: Seven of the ten jurors who gave statements couldn’t remember any comment about Jeffries’s record, one thought he had a recollection but suspected it might be imaginary, and the two that actually recall a comment gave vague and somewhat contradictory accounts. See Jeffries, 771 F.Supp. at 1538. Whatever was said apparently wasn’t very memorable. At most, the district court found the remark made during the guilt phase was “made only in passing and was not repeated,” “was neither considered nor dwelled upon by the *1509 jury,” and was not made during the deliberations-the most critical time. Id. at 1539. There are even fewer details about the comment allegedly made during the penalty phase. See id. at 1538.

Fourth, it’s hard to imagine more effective curative steps than were taken here. Not only did the trial judge instruct jurors “to disregard any evidence which ... was not admitted,” but it also appears the jurors themselves took steps to cure the error by immediately informing whoever made the comment about Jeffries’s prior record that it wasn’t supposed to be considered. See Jeffries, 771 F.Supp. at 1538. 13

It’s therefore understandable that the district judge, who held an evidentiary hearing and heard the jurors testify in person, twice concluded the error here was harmless. In 1991, the district judge concluded that “there is no reasonable possibility that the introduction of extrinsic evidence affected the jury verdict____” Id. at 1539 (emphasis added). And upon remand from Jeffries III, the district judge in apparent exasperation declared:

This Court remains convinced that there was no prejudicial error from statements by a juror as to Jeffries’ past record or past conviction for armed robbery. However, this Court’s review is tightly circumscribed by the Court of Appeals.

Order on Remand, at 2. Likewise, our colleague Judge Fernandez has insisted throughout this case’s tortured history that the juror misconduct could only be described as harmful if one allows the nature of the information to “sweep[ ] away all other considerations.” Jeffries III, 5 F.3d at 1198 (Fernandez, J., dissenting); see also Jeffries v. Blodgett, 988 F.2d 923, 929 (9th Cir.1993) (Fernandez, J., dissenting from order granting rehearing).

While in many eases judges can disagree in good faith about what constitutes harmless error under Brecht , there are other eases where there’s only one reasonable conclusion; this is surely such a case. The only way the majority can find this error harmful is to disregard Brecht by holding that a finding of actual prejudice isn’t necessary: “[T]he communication by its nature was intrinsically prejudicial and necessarily had a substantial and injurious influence on the verdict.” Maj. op. at 1491 (emphasis added). This holding disregards Supreme Court caselaw.

Ill

We also disagree with the majority’s application of the law of the case doctrine. The doctrine heretofore has been a somewhat benign discretionary rule which empowers courts to swat away unwelcome requests from disappointed litigants who pester them to reverse earlier rulings; it has never before been thought to prevent a court from reconsidering its own earlier ruling in a pending case when it decides it made a mistake.

Law of the ease is a prudential doctrine; it’s a courteous and efficient way for a court to say “enough’s enough” when litigants seek reconsideration of prior interlocutory decisions. See, e.g., Remington v. Central Pacific R.R. Co., 198 U.S. 95, 100 , 25 S.Ct. 577, 579 , 49 L.Ed. 959 (1905) (“However stringent may be the practice in refusing to reconsider what has been done, it still is but practice, not want of jurisdiction, that makes the rule.”); Arizona v. California, 460 U.S. 605, 618 , 103 S.Ct. 1382, 1391 , 75 L.Ed.2d 318 (1983) (“Law of the case directs a court’s discretion, it does not limit the tribunal’s power.”); Messenger v. Anderson, 225 U.S. 436, 444 , 32 S.Ct. 739, 740 , 56 L.Ed. 1152 (1912) (“[L]aw of the case ... merely expresses the practice of courts generally to refuse to reopen what has been decided, not a limit to their power.”); Russell v. Commissioner, 678 F.2d 782, 785 (9th Cir.1982) (the law of the case is “a voluntary limitation”). Some such doctrine is necessary to help the court bring the case to a close where a party insists on coming up with new arguments. The law of the ease doctrine thus enables the *1510 court to treat its own interlocutory rulings as final and go on with the business of deciding the rest of the case. See Lockert v. United States Dept. of Labor, 867 F.2d 513, 518 (9th Cir.1989) (“Law of the case doctrine is purely judge-made ... to help manage efficiently their own affairs.”). This serves the interests of finality, efficiency and sound judicial administration. See, e.g., United States Office of Personnel Management v. FLRA, 905 F.2d 430 , 434 (D.C.Cir.1990); Lehrman v. Gulf Oil Corp., 500 F.2d 659, 662 (5th Cir.1974). These are very important interests, of course, but they can never stand in the way of a court’s first duty-to decide the case right. Loumar, Inc. v. Smith, 698 F.2d 759, 762 (5th Cir.1983) (“The law of the case doctrine is not ... a barrier to correction of judicial error.”). If, for whatever reason, a court develops doubts about one of its own interlocutory rulings, it need not weigh any factors or consider any countervailing considerations before it may reconsider its own questionable ruling. See 18 Charles Alan Wright et al., Federal Practice and Procedure, § 4478, at 789 (1981) (“Although courts are often eager to avoid reconsideration of questions once decided in the same proceeding, it is clear that all federal courts retain power to reconsider if they wish.”). 14 The court may-nay it must-go ahead and correct its own error, for no judicial sin is worse than issuing a judgment the judge knows to be wrong on the facts or law.

So what then of all those factors courts must consider in applying the law of the ease doctrine-e.y., whether there is mtervening authority, whether the prior decision was clearly erroneous or would cause a manifest injustice, or whether substantially new evidence was revealed in the interim? See, e.g., Leslie Salt Co. v. United States, 55 F.3d 1388, 1393 (9th Cir.), cert. denied, — U.S. -, 116 S.Ct. 407 , 133 L.Ed.2d 325 (1995); Hegler v. Borg, 50 F.3d 1472, 1475 (9th Cir.), cert. denied, — U.S.-, 116 S.Ct. 675 , 133 L.Ed.2d 524 (1995); Merritt v. Mackey, 932 F.2d 1317, 1320 (9th Cir.1991); United States v. Houser, 804 F.2d 565, 568 (9th Cir.1986). Those factors matter only where the court would otherwise choose not to reconsider its own earlier ruling in reliance on the law of the case doctrine. If, for example, a litigant who lost a summary judgment motion comes up with an intervening Supreme Court or circuit opinion that goes the other way, a district judge would not have the unfettered discretion to not reconsider its erroneous ruling. 15

*1511 The action of the panel in Jeffries v. Wood, 75 F.3d 491 (9th Cir.1996) (“Jeffries IV”), in reconsidering the Jeffries III harmless error ruling falls squarely within that prong of the law of the case doctrine which gives the court the authority to reconsider its own prior rulings. Insofar as the law of the ease is concerned, then, Jeffries TV was wholly free to reconsider what it had said earlier. This does not mean, of course, that Jeffries TV was free to rewrite Jeffries III, which was a published opinion and therefore had become the law of the circuit. As the law of the circuit operates in our court, no three-judge panel may reconsider a rule of law embodied in a prior published opinion; that can only be done by the court sitting en banc. See United States v. Gay, 967 F.2d 322, 327 (9th Cir.1992) (“[0]ne three-judge panel of this court cannot reconsider or overrule the decision of a prior panel.”); see also Atonio v. Wards Cove Packing Co., 810 F.2d 1477, 1479 (9th Cir.1987) (“A panel faced with [a conflict between precedent] must call for en banc review____”). The fact that it happens to be the same panel in the same case doesn’t make a difference. Having issued a published opinion which became final-which other panels of this court and other courts of this circuit justifiably relied on-the Jeffries panel could no longer simply reverse its earlier ruling. It was bound just like every other panel-not because its earlier ruling was the law of the case but because it was the law of the circuit. 16

*1512 The majority conflates the two concepts, to the detriment of both. On the one hand, it applies law of the case to prevent a court from correcting its * own erroneous ruling and, as we explain below, with disastrous consequences for petitioner. On the other hand, the majority weakens the law of the circuit doctrine by holding that a panel does, in fact, have discretion to overrule its own published opinions, so long as the law of the case balancing is satisfied. See maj. op. at 1492 (“When [subsequent panels rely on the initial decision], a panel must be exceedingly careful in altering the law of its earlier [published] opinion.”). This is a power we have never before granted to a single panel of the court and it’s a troubling weakening of the rule that only an en banc court may overrule published circuit precedent.

But it is the other aspect of the majority’s holding that is the most troublesome-the rule that a court is bound by its own erroneous interlocutory ruling, even when it comes to the realization that it has made a big mistake. Is this just window dressing-a lot of fancy language that we can expect will be applied sensibly in the future? Not at all. The majority’s application of this doctrine in our case is nothing short of procrustean.

The majority, it will be remembered, has concluded that the jury deliberations were impermissibly tainted because some of the jurors were aware of Jeffries’s prior robbery conviction. On this basis it reverses petitioner’s death sentence. But precisely the same analysis applies to Jeffries’s murder conviction-only more so. While the jury might well have given Jeffries the death penalty even if it did not believe he had participated in the robbery, 17 commission of the robbery was at the heart of the prosecution’s murder case. 18 It would seem to follow, therefore, that if the aggravated murder conviction was tainted, so was the underlying first-degree murder conviction.

The majority seems to concede as much but nonetheless leaves the murder conviction intact-and defendant subject to a sentence of life imprisonment. See Wash.Rev.Code § 9A.32.040. Why? Strange as it may seem, because this is “true to our law of the case holding.” Maj. op. at 1493. If the law of the case doctrine precludes reconsideration of a first degree murder conviction we now know is tainted, we can’t imagine the ease where a panel is ever justified in defying the law of the case doctrine by reconsidering an earlier ruling.

Of course, we do not agree that either the sentence or the conviction was tainted, so we are not troubled about the substantive outcome of this part of the majority’s ruling. But the majority’s willingness to let stand a conviction that, by its own analysis, was unconstitutionally obtained-and let petitioner rot in prison for the rest of his life-just points out how powerful a doctrine law of the case is in the majority’s mind. This lesson will not be lost on future panels, and on other courts in this circuit, when they consider going back to correct their own errors. The result will be a fair number of more erroneous rulings and useless appeals to correct mistakes that could have been taken care of by the court that made the error in the first place. 19

. The majority's claim that Landgraf announced a default rule of prospective application is based on a single sentence whose meaning the majority distorts by ripping it out of context. Landgraf, 511 U.S. at 272 , 114 S.Ct. at 1500-01 . On the very next page from the one on which the majority’s key sentence appears, the Court declares: "Even absent specific legislative authorization, application of new statutes passed after the events in suit is unquestionably proper in many situations." Id. at 273 , 114 S.Ct. at 1501 (emphasis added). How can this be squared with the majority’s claim that "prospective application has unquestionably become the default rule”? See Maj. op. at 1494.

. In Fretwell , petitioner had a colorable ineffective assistance of counsel claim based on precedent existing at the time his state proceedings were completed. See Fretwell, 506 U.S. at 367 , 113 S.Ct. at 841 . However, by the time his federal habeas proceedings got underway, the precedent had been overruled. Id. at 368 , 113 S.Ct. at 841-42. Applying the intervening law, the Court rejected his ineffective assistance claim. Id. at 373, 113 S.Ct. at 844-45. In doing so, the Court rejected the argument that applying the intervening precedent would be unfair because it would upset the petitioner’s interest in habeas relief created upon entry of the final state-court judgment. See id. at 372-73, 113 S.Ct. at 844-45. We must reject the same argument here.

. The majority believes Landgraf can't have meant what it says because this would impose a big burden on courts: Anytime Congress doesn't explicitly provide a temporal provision, courts would have to inquire whether the statute has a retroactive effect. See maj. op. at 1496. We fail to see how the "retroactive effect” inquiry is any more burdensome than trying to figure out congressional intent in the absence of an explicit statutory command.

The majority also objects that courts will have to engage in a provision-by-provision "retroactive effect” analysis. Maj. op. at 1496. This may be undesirable, but it's precisely what Landgraf calls for. See 511 U.S. at 280 , 114 S.Ct. at 1505 ("ITIhere is no special reason to think that all the diverse provisions of the Act must be treated uniformly ... [so] courts should evaluate each provision of the Act in light of ordinary judicial principles concerning the application of new rules to pending cases and preenactment conduct.”). In fact, the very next case after Landgraf in the United States Reports concerns whether a different provision of the same law at issue in Landgraf is retroactive. See Rivers v. Roadway Express, Inc., 511 U.S. 298 , 114 S.Ct. 1510 , 128 L.Ed.2d 274 (1994).

. Actually, it is a bit odd that Congress inserted a temporal provision into Chapter 154, but didn’t do the same in Chapter 153, but it’s not so odd that we are left with the inescapable conclusion that Congress meant to attach any significance to the difference. Congress may have focused on Chapter 154's temporal scope but overlooked Chapter 153's because of a significant difference between the two chapters: Chapter 153 merely amends existing procedures while Chapter 154 creates a brand new habeas track. Congress may have assumed that the amendments to existing procedures would go into effect immediately, but wasn't so sure about the brand new procedures, especially since these new procedures were contingent on another event occurring in the future-a state qualifying as an "opt-in.” See 28 U.S.C. § 2261 ("opt-in” state must provide *1505 competent post-conviction counsel to defendants); Leavitt v. Arave, 927 F.Supp. 394, 398 (D.Idaho 1996) (Chapter 154 needed an effective date provision to prevent the inference from the opt-in requirements that it only applied to future cases).

. The argument seems to be that this incorporation would have been superfluous-another section of Chapter 154 already declares that section 2254 will apply to opt-in cases-unless it was intended to make the retroactivity provision in Chapter 154 apply to sub-sections 2254(a),(d) and (e) when used in opt-in cases. This would only be necessary if section 2254 wasn't retroactive already, which could only be true if Chapter 153 weren't retroactive.

. In fact, Jeffries attempted to raise this issue again in two subsequent personal restraint petitions, and his briefs filed in those cases also contain the same factual misstatement. See Personal Restraint Petition Reply Br., In re Patrick James Jeffries, No. 53397-1, at 7 (Apr. 30, 1987); Brief of Petitioner, Personal Restraint Petition, In re Patrick James Jeffries, No. 56153-2, at 23 (June 15, 1989). In both cases, the Washington Supreme Court refused to reconsider its earlier ruling.

. We're confused by the majority’s argument that use of the plural “proceedings” in the preamble to section 2254(d) makes some difference, especially since the applicable subsection clearly uses the singular. Compare maj. op. at 1500 n. 19 with 28 U.S.C. § 2254 (d)(2).

.The majority seems to think we would require petitioners to file voluminous trial court records in every state appellate proceeding. See maj. op. 1500 n. 19. Not at all. It's not where the evidence is located that matters; Jeffries could have incorporated the Sims affidavit by reference. What matters is that he never brought the evidence to the court’s attention; he never mentioned the Sims affidavit in any of his state appeals and misled the court by making contrary representations. Under such circumstances, a state court cannot then be said to have made an “unreasonable determination of the facts in light of the evidence presented.” 28 U.S.C. § 2254 (d)(2) (emphasis added).

. The majority also claims that new section 2254(d) doesn't apply to Jeffries because the writ was granted before April 24, 1996. See maj. op. at 1501 n. 21. The majority seems to believe that new section 2254(d)'s standard of review applies only at the moment when the initial decision to issue the writ is made. The majority therefore argues that, "[bjecause we are not considering whether to order the district court to issue a writ, but are merely affirming the prior issuance of the writ ..., the Act does not preclude relief:” Id. This argument proves too much: If it were true, courts of appeals would never be bound by section 2254(d) in reviewing a district court's decision to grant the writ. Moreover, this argument ignores the traditional rule that a “complaint ha[s] no vested right in the decree of the District Court while it [is] subject to review.” American Steel Foundries v. Tri-City Council, 257 U.S. 184, 201 , 42 S.Ct. 72, 75 , 66 L.Ed. 189 (1921). Moreover, the majority's claim that it's merely affirming the district court's grant of the writ is belied by the opinion's mandate, which reverses the grant of the writ as to the underlying first-degree murder conviction but orders its issuance as to the aggravated murder conviction and resentencing. See maj. op. at 1501.

. The record reflects that the jury began deliberating at 3:00 p.m. on November 4, 1983, adjourned for the night at 9:35 p.m., was ordered to return at 9:00 a.m. the next morning, and reached a verdict by 12:31 p.m. that afternoon.

. The district court made two sets of factual findings. After the evidentiary hearing in 1991, the court determined the circumstances surrounding the making of the alleged "remark”i.e., when, where and how it was made-but did not actually determine the substance of the remark. See Jeffries v. Blodgett, 771 F.Supp. 1520, 1539 (W.D.Wash.1991). Jeffries v. Blodgett, 5 F.3d 1180, 1191 (9th Cir.1993) ("Jeffries III"), cert. denied, 510 U.S. 1191 , 114 S.Ct. 1294 , 127 L.Ed.2d 647 (1994), remanded the case so the district court could determine the substance of the remark; the entirety of the district court's finding on remand is quoted in the text above. These findings will only be second-guessed if clearly erroneous. See Bonin v. Calderon, 59 F.3d 815, 823 (9th Cir.1995), cert. denied, - U.S. -, 116 S.Ct. 718 , 133 L.Ed.2d 671 (1996).

. By comparison, if Jeffries had confessed, but the confession had been suppressed-or if he had pleaded guilty, but withdrawn the plea-and a prospective juror said he knew about it, there’s no doubt that Jeffries would have challenged the juror. The same would be true if Jeffries's prior conviction had been for an inflammatory crime, like child molestation. Instead, here Jeffries was willing to rely on Sims's promise not to consider the information; a promise all the jurors made.

. For example, one juror gave the following account of the incident in the jury room:

[One] of the jurors asked, “Didn’t Jeffries have a record?” I remember saying, "I think so, but that has no bearing on this case." There was general agreement. The discussion went no further at that time.

Affidavit of Kathleen Sims, Feb. 27, 1986, attached letter.

. See, e.g., Rent-A-Center, Inc. v. Canyon Television & Appliance Rental, Inc., 944 F.2d 597, 602 (9th Cir.1991) ("|T|he law of the case is sin equitable doctrine that should not be applied if it would be unfair.”); United States v. Miller, 822 F.2d 828, 832 (9th Cir.1987) ("Law of the case should not be applied woodenly in a way inconsistent with substantial justice.").

. We realize that there are opinions that characterize the law of the case doctrine somewhat differently. But the language they use doesn’t appear to be particularly well-thought out. See Joan Steinman, Law of the Case: A Judicial Puzzle in Consolidated and Transferred Cases and in Multidistrict Litigation, 135 U.Pa.L.Rev. 595, 613 (1987) ("The Supreme Court has not considered the intricacies of law of the case doctrine at length or in decisions having broad precedential value.”); Miller, 822 F.2d at 832 (“|T|he law of the case has been described by the Supreme Court as ‘an amorphous concept.’ " (quoting Arizona v. California, 460 U.S. at 618 , 103 S.Ct. at 1391 )).

Christianson v. Colt Indus. Operating Corp., 486 U.S. 800, 817 , 108 S.Ct. 2166, 2178 , 100 L.Ed.2d 811 (1988), appears to be the one exception. After the Federal Circuit found that it lacked jurisdiction over a case and transferred it to the Seventh Circuit, the Seventh Circuit reconsidered the jurisdictional issue, held that the Federal Circuit was "clearly wrong” and retransferred the case. The Supreme Court reversed the Seventh Circuit for failing to abide by the law of the case, suggesting that the law of the case should be strictly adhered to. See id. at 819 , 108 S.Ct. at 2179 .

But Christianson doesn’t announce a general rule. Its seemingly strict version of the law of the case was motivated by unique policy considerations: The Court wanted to prevent jurisdictional ping-pong-i.e., to prevent cases from being perpetually bounced back and forth between courts that disagreed over jurisdictional questions, which the Court considered inherently subject to disagreement. See id. at 818-19 , 108 S.Ct. at 2178-79 . Thus, the Court carved out a special rule. In defining the law of the case as applied generally, the Court used terms consistent with our view:

"A court has the power to revisit prior decisions of its own or of a coordinate court in any circumstance, although as a rule courts should be loathe to do so in the absence of extraordinary circumstances____”

Id. at 817 , 108 S.Ct. at 2178 (emphasis added).

. To ensure that this mistake doesn’t recur, we would hold that, whenever law of the case and law of the circuit problems arise in the same case, the law of the circuit principle supplants any law of the case considerations. See United States v. 162.20 Acres of Land, 733 F.2d 377, 379 (5th Cir.1984) ("In this circuit, however, the law-of-the-case doctrine is supplanted by our flrai rule that one panel cannot disregard the precedent set by a prior panel even though it perceives error in the precedent.”); see also LaShawn A. v. Barry, 69 F.3d 556, 571 (D.C.Cir.1995) (Randolph, J., dissenting) (”[I]n circuits such as ours, where both doctrines are at work, the more exacting law-of-the-circuit doctrine supplants the law-of-the-case doctrine when panels hear multiple appeals from a single case.”), vacated by 87 F.3d 1389, 1395 (D.C.Cir.1996) (en banc), petition for cert. filed, 65 U.S.L.W. 3632 (Mar. 11, 1997) (No. 96-1433). Supplanting the law of the case with the law of the circuit not only eliminates the confusion engendered by this case, but also assures that the law of the circuit will remain an absolute rule. Once a panel publishes a disposition, it loses the power to reconsider the issue on a subsequent appeal; instead, it can only correct a mistake by calling sua sponte for en banc. Fortunately, such clear mistakes occur infrequently enough that our en banc resources should not be overly burdened. See IB Moore’s Federal Practice ¶ 0.404[4.-5], at 11-25 n. 11 (1996).

Many circuits have confronted the issue but decided not to resolve it. See, e.g., LaShawn A., 87 F.3d at 1395 ("Because the law-of-the-case doctrine alone precluded [reconsideration], we need not delve deeply into the interplay between the law-of-the-case and the law-of-the-circuit doctrines.”); Liberty Mut. Ins. v. Elgin Warehouse & Equip., 4 F.3d 567 , 571 n. 7 (8th Cir.1993) ("Arguably the en banc rule is incompatible with a subsequent panel deviating from the law of the case established by a prior panel decision.... However, ... we need not consider this issue ...."); cf. Lacy v. Gardino, 791 F.2d 980, 985 (1st Cir.1986) (“We need not decide, however, whether the two habeas petitions are the same or different cases requiring the application of either the law of the case doctrine or stare decisis."). But the Fifth Circuit is a notable exception: It supplants the law of the case with law of the circuit, although there are some anomalous decisions. Compare Missouri Pac. R.R. v. Railroad Comm'n, 948 F.2d 179 , 186 (5th Cir.1991) ("We see no reason to reconsider our earlier decision____ [I]t is a 'well-known and longstanding rule of decision in this circuit that one panel cannot overrule another.’ ”) with Blair v. Sealift, Inc., 91 F.3d 755 , 761 n. 26 (5th Cir.1996) (noting the "wrinkle” that prior decisions “could either be considered the law of the case or the application of a 'prior' panel decision, i.e., the binding precedent of this circuit” (emphasis in original)). The Third Circuit has similarly expressed some doubt whether a panel may depart from the law of the case even when a prior opinion was clearly erroneous, which is akin to enforcing the law of the circuit principle. See Todd & Co. v. SEC, 637 F.2d 154 , 157 & n. 4 (3d Cir.1980). .

Commentators are divided on how to resolve this issue. Although noting that the relationship between the law of the case and the law of the circuit “is not altogether clear,” Professor Moore appears to endorse supplanting the law of the case with the law of the circuit. See IB Moore’s Federal Practice ¶ 0.404[4.-5], at 11-25 & n. 11. Professor Wright, however, appears to disagree: "Whatever may be said of the need to bind all succeeding panels to the stare decisis effect of a circuit decision in separate litigation, [the law of the circuit] rule should not be followed on successive appeals in the same case.” See 18 Wright, Federal Practice and Procedure § 4478, at 692 n. 25 (Supp.1996) (discussing Missouri Pac. R.R., 948 F.2d at 186).

. The prosecution, it will be recalled, had two theories on which it sought the aggravated murder conviction: that Jeffries killed both victims to conceal the robbery and that he killed Mrs. Skiff to conceal his earlier killing of Mr. Skiff. See Jeffries, 717 P.2d at 741 ; R.T. at 3503, 3514.

. The state introduced the robbery evidence to show Jeffries's motive and to tie him to the murders. The state introduced evidence that $30,000 in Canadian money was missing from the Skiffs’ residence, that Jeffries was seen flashing large amounts of Canadian cash shortly after the murders, and that Jeffries sold or attempted to sell items stolen from the Skiffs shortly after the murders. See Jeffries v. Blodgett, 974 F.2d 1179, 1185 (9th Cir.1992). And, here's how the state explained Jeffries’s motive: “[W]hy [did defendant murder the Skiffs]? Two reasons, one for greed and the other for cover-up. The defendant wanted some property and he had to kill ... to get away with it.” R.T. at 3503. Finally, the state argued that the murderer and robber were the same person: "I don't know what precipitated the killing, but [whoever did it] was greedy and [Jeffries’s] Counsel admitted it. Whoever killed him, and I submit it was Jeffries, is greedy.” R.T. at 3532.

.We mention only in passing another curious aspect of the majority’s holding. At one point, the majority concedes that the law of the case doctrine does not apply to an en banc court. See maj. op. at 1492. And, of course, it does not-any more than our errors are binding on the Supreme Court. See Christianson, 486 U.S. at 817 , 108 S.Ct. at 2178 . Yet the majority then goes on and applies law of the case anyway. Very strange indeed.