In January 1988 Aaron Lindh marched into the City-County Building of Madison, Wisconsin, and shot three strangers for no apparent reason. Two died. Lindh pleaded guilty to carrying and using a firearm in a public building; a jury convicted him of two murders and one attempted murder. During the second phase of a bifurcated proceeding, Lindh argued that he was insane at the time of the shootings, which under Wisconsin law would alter the place of his confinement (from a prison to a prison-hospital) and entitle him to release if at some future time he should be deemed “recovered.” But the jury found that Lindh did not have a mental disease when he pulled the trigger, and the judge sentenced him to life plus 35 years in prison.
Lindh’s principal contention on appeal was that the judge unduly restricted his cross-examination of Dr. Leigh Roberts, a psychiatrist who interviewed him on the day of the shootings (and several times thereafter) and testified for the prosecution during the second phase of the trial. During March 1988 Roberts learned that he was under investigation for engaging in improper sexual conduct with a female patient; in May 1988 Roberts learned that the Medical Examining Board was looking into allegations made by three female patients. By the time of trial in September 1988 a criminal investigation was ongoing — conducted by the Milwaukee County District Attorney, acting as a special prosecutor after the Dane County District Attorney (whose jurisdiction includes Madison) recused himself. Lindh’s attorney sought to explore the allegations made against Roberts, contending that the pending investigation would give Roberts a reason to slant his testimony in the prosecution’s favor. Lindh did not argue that Roberts had entered into discussions with the prosecutor about the sexual misconduct allegations; any plea agreement or negotiations in progress would have been subjects of cross-examination under Giglio v. United States, 405 U.S. 150, 92 S.Ct. 763, 31 L.Ed.2d 104 (1972). Instead Lindh argued that even if the Dane County prosecutor could not provide a quid pro quo by forgoing charges or reducing their severity, he might put in a good word with the Milwaukee prosecutor; and whether or not lenity was in the cards, Roberts might believe that he had something to gain from helping to convict Lindh, which could color his analysis and testimony. Lindh argued that both state law and the Confrontation Clause of the Sixth Amendment, applied to the states by the due process clause of the Fourteenth, entitled him to cross-examine Roberts about all potential sources of bias. See Olden v. Kentucky, 488 U.S. 227, 109 S.Ct. 480, 102 L.Ed.2d 513 (1988); Delaware v. Van Arsdall, 475 U.S. 673, 106 S.Ct. 1431, 89 L.Ed.2d 674 (1986); United States v. Abel, 469 U.S. 45, 105 S.Ct. 465, 83 L.Ed.2d 450 (1984); Davis v. Alaska, 415 U.S. 308, 94 S.Ct. 1105, 39 L.Ed.2d 347 (1974).
The trial judge foreclosed inquiry into this subject (although he permitted extensive cross-examination on other matters), the court of appeals reversed, State v. Lindh, 156 Wis.2d 768, 457 N.W.2d 564 (Wis.App.1990), and the Supreme Court of Wisconsin reversed in turn, reinstating the sentences, State v. Lindh, 161 Wis.2d 324, 468 N.W.2d 168 (1991). The state’s highest court concluded that the possibility of bias was so remote, given the appointment of a special prosecutor, and the prospect of diverting attention to the sexual encounters (a subject of no relevance to Lindh’s sanity) sufficiently great, that the trial judge did not abuse the discretion he possessed under both state and federal law. Justice Abrahamson dissented on state-law grounds. 468 N.W.2d at 185-89. Lindh then commenced a collateral attack in federal court. The district court denied the petition, writing that it “agrees wholeheartedly with the analysis of’ the state’s Supreme Court. Lindh appealed to this court.
Fifteen days after a panel heard oral argument, the President signed the Antiterrorism and Effective Death Penalty Act of 1996, Pub.L. 104-132, 110 Stat. 1214. Section 104 of the new statute amends 28 U.S.C. § 2254, the law under which Lindh seeks relief. *861Unaffected by the 1996 Act, § 2254(a) provides that a writ of habeas corpus may issue “on the ground that [the petitioner] is in custody in violation of the Constitution or laws or treaties of the United States.” It does not further elaborate. Federal courts exercising their authority under § 2254(a) disregarded the state courts’ legal conclusions and reached independent judgments on issues presented to them. Brown v. Allen, 344 U.S. 443, 458, 73 S.Ct. 397, 407-08, 97 L.Ed. 469 (1953). Section 104(2) of the 1996 Act redesignates the former § 2254(d), which deals with state courts’ findings of fact, as § 2254(e); § 104(3) of the 1996 Act, 110 Stat. 1219, adds a new § 2254(d) that for the first time specifies the appropriate treatment of legal determinations by state courts:
An application for a writ of habeas corpus on behalf of a person in custody pursuant to the judgment of a State court shall not be granted with respect to any claim that was adjudicated on the merits in State court proceedings unless the adjudication of the claim—
(1) resulted in a decision that was contrary to, or involved an unreasonable application of, clearly established Federal law, as determined by the Supreme Court of the United States; or
(2) resulted in a decision that was based on an unreasonable determination of the facts in light of the evidence presented in the State court proceeding.
We set this case for reargument before the full court in order to decide whether the new provision applies to pending cases and, if it does, how it affects them. We invited and received supplemental briefs from the parties, and we have had the benefit of briefs from the American Bar Association and a group of former federal judges as amici curiae.
I
Landgraf v. USI Film Products, 511 U.S. 244, 114 S.Ct. 1483, 128 L.Ed.2d 229 (1994), lays out a sequence of issues. First we must decide whether Congress has decided to which cases a new law applies; if it has, the only task is to follow the statute. If Congress has not provided one way or the other, we must apply the law in force at the time of decision — which is to say, the new § 2254(d) — unless “the new provision attaches new legal consequences to events completed before its enactment.” 511 U.S. at -, 114 S.Ct. at 1499. But what is a “new legal consequence”? Landgraf says that not every change in outcome counts; it matters whether the party adversely affected by the change has legitimate reliance interests in the operation of the former law.
A
Has Congress answered the question at hand? Not directly. Section 104 of the 1996 Act lacks an effective-date provision. Lindh contends that Congress addressed the subject indirectly, by providing that the new Chapter 154 of Title 28 (28 U.S.C. §§ 2261-66), captioned Special Habeas Corpus Procedures in Capital Cases and contained in § 107(a) of the 1996 Act, applies “to cases pending on or after the date of enactment of this Act.” Section 107(e), 110 Stat. 1226. This establishes, Lindh submits, that §§ 101 to 106 of the statute do not apply to pending cases.
“Establishes” is too strong a word. Sections 101 to 106 amend Chapter 153 of Title 28. Nothing in the 1996 Act provides one way or another for the temporal extent of the changes. Do they govern collateral attacks arising out of crimes committed after April 24, 1996? Convictions after that date? Appellate decisions after that date? Collateral attacks filed after that date? The statute is silent. Congress addressed those issues for Chapter 154, but not Chapter 153. There is at most a negative implication. Should we draw it? A competing inference is that Congress could not agree on an effective-date provision for the amendments to Chapter 153, leaving the subject to judicial resolution. Perhaps instead Congress overlooked the subject when drafting §§ 101-106 (the changes to Chapter 153 and the new Chapter 154 originated in different Houses of Congress at different times) — or recognized its importance but thought the answer so clear that express provision was unnecessary. Which understanding is superior?
*862Potential negative implications of effective-date provisions have been urged before, most recently in Landgraf. The Civil Rights Act of 1991 provides that “[e]xcept as otherwise specifically provided, this Act and the amendments made by this Act shall take effect upon enactment.” The Court held that this language “does not even arguably suggest that [the Act] has any application to conduct that occurred at an earlier date.” 511 U.S. at -, 114 S.Ct. at 1493 (emphasis added). Two other provisions of the 1991 Act are more explicit. One says that the statute does not apply to a particular ongoing case; the other says that the extension of the civil rights laws to overseas employers “shall not apply with respect to conduct occurring before the date of the enactment of this Act.” Landgraf argued that these two anti-retroac-tivity provisions established that the rest of the 1991 Act applied to conduct preceding its enactment. Otherwise the two anti-retroac-tivity provisions would be irrelevant, and courts try to interpret laws to avoid both irrelevance and redundancy. Landgraf, 511 U.S. at -, 114 S.Ct. at 1494. The Court conceded this but held nonetheless that the 1991 Act is effectively silent on the question. Id. at - - -, 114 S.Ct. at 1494-96. It was unwilling to draw a negative implication from two provisions that may have been inserted just to make double sure. Although “a majority of the 1991 Congress [may have] favored retroactive application, even the will of the majority does not become law unless it follows the path charted in Article I, § 7, el. 2 of the Constitution” (511 U.S. at -, 114 S.Ct. at 1496) — that is, unless it yields a text agreed on by both Houses and signed by the President.
Just so with the 1996 Act. And the argument by negative implication is weaker for the 1996 Act than for the 1991 Act. The Supreme Court’s holding in Landgraf made two sections of the statute irrelevant. Nothing we could hold, one way or the other, about §§ 101-106 could deprive § 107(c) of independent meaning. True enough, 28 U.S.C. § 2264(b), in Chapter 154, refers to § 2254(d), so § 107(e) requires the application of the amended § 2254(d) to capital cases within the scope of Chapter 154. But we think that Lindh misunderstands Chapter 154 when contending that § 2264(b) has “no function other than to cause certain of the otherwise inapplicable Chapter 153 revisions to apply retroactively” (Supp. Reply Br. 3). Section 2264(a) provides that, when Chapter 154 applies, a court shall consider only “a claim or claims that have been raised and decided on the merits in the State courts” unless failure to raise a claim was caused by one of three identified circumstances. Section 2264(b) adds: “Following review subject to subsections (a), (d), and (e) of section 2254, the court shall rule on the claims properly before it.” Why only “subsections (a), (d), and (e) of section 2254”? Surely not because these are to be retroactive, while subsections (b), (e), (f), (g), (h), and (i) are not. If as Lindh believes the only reason to include § 2254(d) in § 2264(b) is to apply it to pending cases, then this must also be the sole reason for referring to § 2254(a) — but § 2254(a) is not amended by the 1996 Act, while many subsections omitted from the list in § 2264(b) were amended or added by the new statute.
Section 2264(b) conveys its meaning only when read together with § 2264(a). Subsections (b) and (e) of § 2254 have been omitted from the list because § 2264(a) replaces their rule with one more favorable to the states. Section 2254(b) and (c) codify the exhaustion requirement, which § 2264(a) displaces for Chapter 154 cases. The remaining subsections of § 2254 likewise don’t fit the Chapter 154 scheme. For example, § 2254(i), added by the 1996 Act, says that “ineffectiveness or incompetence of counsel during Federal or State collateral post-conviction proceedings shall not be a ground for relief in a proceeding arising under section 2254.” 110 Stat. 1219. Chapter 154 addresses this question directly in the new 28 U.S.C. § 2261(e), 110 Stat. 1222, making incorporation of § 2254(i) inapt. Section 2264(b) tells us not “when,” but “which.” Lindh’s argument therefore misses the mark: this section neither addresses any other section’s temporal effectiveness nor is made irrelevant by applying § 2254(d) to pending eases.
Chapter 154 comprises many rules that Congress evidently wanted to apply forthwith no matter what the courts made of *863§§ 101-106. For example, the new 28 U.S.C. § 2262(a) provides for automatic stays of execution during initial collateral attacks covered by Chapter 154. Prisoners on death row received stays on April 24, 1996, even if courts had denied stays under prior law. Having blocked executions while litigation continues, Congress curtailed the time federal courts may take to act — for example, a eourt of appeals must decide a capital appeal within 120 days after the filing of the reply brief. 28 U.S.C. § 2266(c)(1)(A). Section 107(c) applies that limit to pending capital cases, a step that serves quite a different function from a decision one way or the other about the application of § 2254(d) to non-capital cases. Congress obviously wanted to ensure that both the stays of execution and the time limits — provisions of Chapter 154 without counterparts in Chapter 153 — extended to as many cases as possible. Nothing about that decision has any implications for Chapter 153. We conclude that § 107(c) does not govern the question at hand. Sections 101-106 lack an effective-date provision. We must decide what to do when the legislation is silent.
B
Landgraf reiterates two long-established rules that govern when the legislature is silent: first, courts normally apply the law in force at the time of decision; second, the court does not use the new law if application would be “retroactive.” Each side appeals to one of these propositions. Wisconsin contends that § 2254(d) is today’s rule of decision and should be applied. Lindh rejoins that § 2254(d) would be “retroactive” to the extent it reduces his chance of prevailing. He also contends that he is entitled to a writ of habeas corpus whether or not § 2254(d) applies, but he uses the possibility of an adverse effect as the definition of retroactivity. As a definition, it will not do — for it would obliterate the first rule, turning it into something like “a court applies the law in force at the time of its decision whenever it is the same as the law in force all along.” Landgraf was explicit that “[e]ven uneontroversially prospective statutes may unsettle expectations and impose burdens on past conduct”. 511 U.S. at - n. 24, 114 S.Ct. at 1499 n. 24; see also id. at - - -, 114 S.Ct. at 1501-04.
“A statute does not operate ‘retrospectively’ merely because it is applied in a case arising from conduct antedating the statute’s enactment, or upsets expectations based on prior law. Rather, the court must ask whether the new provision attaches new legal consequences to events completed before its enactment.” Landgraf, 511 U.S. at -, 114 S.Ct. at 1499 (citation and footnote omitted). Section 2254(d) does not attach new legal consequences to the filing of petitions for habeas corpus, although some other parts of the 1996 Act may do this. We take it that under Landgraf a certificate of probable cause to appeal issued before April 24, 1996, authorizes an appeal, although after April 24 appeal depends- on a “certificate of appealability” under § 102 of the 1996 Act (amending 28 U.S.C. § 2253(e)), which not only changes the name but also requires the court to identify the appealable issue. Similarly, a second or successive petition already pending on April 24 does not require prior approval of the court of appeals under § 106 (amending 28 U.S.C. § 2244(b)), see Williams v. Calderon, 83 F.3d 281, 285-86 (9th Cir.1996), although the application of the substantive standards in the new § 2244(b) to determine who is eligible for relief is a more difficult question, which the court addresses separately today in Burris v. Parke, 95 F.3d 465 (7th Cir.1996).
For a provision such as § 2254(d), which affects the relation between federal and state courts, rather than regulating the details of litigation, the date on which the petition was filed can’t be any more important than the date on which the suits were filed in Landgraf and Mojica v. Gannett Co., 7 F.3d 552 (7th Cir.1993) (en banc). In Landgraf the Court asked whether the Civil Rights Act of 1991 could be applied to employment decisions that preceded the statute. A lawsuit concerning 1990 conduct is governed by 1990 law, no matter when the suit began, Land-graf held (as we had held in Mojica, 7 F.3d at 558-59). See also Rivers v. Roadway Express, Inc., 511 U.S. 298, 114 S.Ct. 1510, 128 L.Ed.2d 274 (1994). The parallel ques*864tion is whether the 1996 Act may be applied when the crime or state court decision preceded April 24,1996.
Understandably, Lindh does not argue that, when deciding whether to commit a crime (and, if so, which one), he relied on the availability of the version of § 2254 that was on the books in 1988. Under the Ex Post Facto clauses people charged with crime have powerful, and enforceable, expectations about the legal rules that apply to their conduct. But the 1996 Act does not change any of the rules defining or penalizing crime. Moreover, having lost in the state’s judicial system before the amendment, Lindh cannot successfully argue that the change of law affected the judicial process. By any standards, the former version of § 2254 was more favorable to prisoners than the current one, and the prospect of more intensive review by federal courts may have made state courts more attentive to claims under federal law. Lindh wants us to cast a darker light on state judges — to assume that the prospect of plenary federal review made them inattentive to federal rights, because then they could appear tough on crime (to improve reelection prospects) while knowing that the people they confine in prison would eventually be let go. He does not offer any empirical support for believing that state judges wrongfully imprison people to further their own careers, and in Wisconsin judges who proclaim willingness to give defendants extra rights have had no difficulty being retained. See Shirley S. Abrahamson, The Emergence of State Constitutional Law, 63 Tex.L.Rev. 1141 (1985). It would not be appropriate to presume that the Justices of the Supreme Court of Wisconsin treated Lindh poorly because they anticipated independent federal review.
Section 2254(d) does not regulate primary conduct. Instead it makes the initial judgment more stable. ■ Do prisoners have legitimate expectations about the scope of collateral review? This is a question the Supreme Court addressed in Lockhart v. Fretwell, 506 U.S. 364, 113 S.Ct. 838, 122 L.Ed.2d 180 (1993). A prisoner called his lawyer ineffective for failing to take full advantage of an appellate opinion that was overruled after the prisoner commenced his collateral attack. The Supreme Court held that the prisoner could not rely on the overruled decision (or obtain relief because his lawyer failed to do so, when he could), even though its disappearance reversed the outcome of his case. The Court’s discussion is worth quoting:
The dissent contends that this holding is inconsistent with the retroactivity rule announced in Teague v. Lane, 489 U.S. 288, 310 [109 S.Ct. 1060, 1075, 103 L.Ed.2d 334] (1989), but we think otherwise. Teague stands for the proposition that new constitutional rules of criminal procedure will not be announced or applied on collateral review. As the dissent acknowledges, this retroactivity rule was motivated by a respect for the States’ strong interest in the finality of criminal convictions, and the recognition that a State should not be penalized for relying on “the constitutional standards that prevailed at the time the original' proceedings took place.” “The ‘new rule’ principle therefore validates reasonable, good-faith interpretations of existing precedents made by state courts even though they are shown to be contrary to later decisions.” Butler v. McKellar, 494 U.S. 407, 414 [110 S.Ct. 1212, 1217, 108 L.Ed.2d 347] (1990).
A federal habeas petitioner has no interest in the finality of the state court judgment under which he is incarcerated: indeed, the very purpose of his habeas petition is to overturn that judgment. Nor does such a petitioner ordinarily have any claim of reliance on past judicial precedent as a basis for his actions that corresponds to the State’s interest described in the quotation from Butler, swpra. The result of these differences is that the State will benefit from our Teague decision in some federal habeas cases, while the habeas petitioner will not. This result is not, as the dissent would have it, a “windfall” for the State, but instead is a perfectly logical limitation of Teague to the circumstances which gave rise to it. Cessante ratione legis, cessat et ipsa lex.
506 U.S. at 372-73, 113 S.Ct. at 844 (citations omitted). In other words, legal changes that *865reduce the willingness of federal courts to set aside judgments presumptively apply to existing judgments.
This is the historical practice. The Supreme Court consistently applies statutory changes in the law of collateral attack to pending cases, and perforce to newly filed cases that seek relief from judgments entered before the statute’s enactment. In Felker v. Turpin, - U.S. -, 116 S.Ct. 2333, 135 L.Ed.2d 827 (1996), the Court decided a case under the 1996 Act’s standards, even though the conviction preceded the amendments. (The petition for habeas corpus in Felker was filed after April 24, 1996, but we have already explained why the date of the petition cannot be dispositive.) Thirty years ago Congress made extensive amendments to Chapter 153, which the Court applied to cases under review without so much as remarking on the date of enactment. Smith v. Yeager, 393 U.S. 122, 124-25, 89 S.Ct. 277, 278-79, 21 L.Ed.2d 246 (1968); Carafas v. LaVallee, 391 U.S. 234, 239, 88 S.Ct. 1556, 1560, 20 L.Ed.2d 554 (1968). Earlier amendments to the rules of collateral attack likewise were brought to bear in pending cases. Gusik v. Schilder, 340 U.S. 128, 131-33 & n. 4, 71 S.Ct. 149, 151-52 & n. 4, 95 L.Ed. 146 (1950), implements the exhaustion requirement introduced by the 1948 amendment to § 2254. Accord, Darr v. Burford, 339 U.S. 200, 70 S.Ct. 587, 94 L.Ed. 761 (1950).
The language of the amendments likewise implies immediate application. Recall that § 2254(d) does not empower a court to issue the writ. Instead it forbids issuance of the writ — the application “shall not be granted unless” certain conditions have been met. The prohibition is prospective: a court must not issue the writ tomorrow, unless.... The difference between an authorization and a prohibition is one we must respect. See Ivey v. Harney, 47 F.3d 181 (7th Cir.1995). It is the sort of difference that the Supreme Court has relied on when holding that changes in the law affecting judicial ability to supply prospective relief govern pending cases. E.g., Hall v. Beals, 396 U.S. 45, 48, 90 S.Ct. 200, 201-02, 24 L.Ed.2d 214 (1969); American Steel Foundries v. Tri-City Central Trades Council, 257 U.S. 184, 201-02, 42 S.Ct. 72, 75-76, 66 L.Ed. 189 (1921); Duplex Printing Press Co. v. Deering, 254 U.S. 443, 464, 41 S.Ct. 172, 175, 65 L.Ed. 349 (1921), all discussed favorably in Landgraf, 511 U.S. at -, 114 S.Ct. at 1501. Current law normally governs when statutes “speak to the power of the court rather than to the rights or obligations of the parties,” Republic National Bank, of Miami v. United States, 506 U.S. 80, 99-100, 113 S.Ct. 554, 565, 121 L.Ed.2d 474 (1992) (Thomas, J., concurring) (quoted with approval in Landgraf, 511 U.S. at-, -, 114 S.Ct. at 1499, 1502). We do not suggest that § 2254(d) reduces the power of the court in the sense of subject-matter jurisdiction, which survives intact in 28 U.S.C. § 2241; a state may waive the benefits of § 2254(d), although other provisions of the 1996 Act, such as the revised exhaustion requirement in § 2254(b), suggest that counsel’s inattention to the niceties of federal practice does not forfeit the benefits the law bestows on the state as an entity. See also Eaglin v. Welborn, 57 F.3d 496, 498-99 (7th Cir.1995) (en bane) (holding that a court may decide a case on the basis of Teague even if the state’s lawyers overlooked that possibility). Our point, rather, is that the amended § 2254(d) is designed to curtail collateral review and augment the finality of judgments, which strongly implies application to existing judgments.
C
Lindh and the State of Wisconsin litigated in the state’s forums the constitutional claim he presents to us. To leave the ensuing judgment in force — and to leave all of the rules governing primary conduct intact — is not “retroactive” in the sense Landgraf uses that term, and the amended § 2254(d) is therefore presumptively applicable. We respectfully disagree with the contrary conclusions of Edens v. Hannigan, 87 F.3d 1109, 1111 n. 1 (10th Cir.1996), and Boria v. Keane, 90 F.3d 36 (2d Cir.1996).
There remains a possibility that application of the 1996 Act could “attach new legal consequences” to events other than the crime and the state’s legal processes and judgment. Consider, for example, how the filing dead*866line in § 101 of the 1996 Act applies to existing judgments. Under the new statute, 28 U.S.C. § 2244(d), a petitioner has only one year from the completion of direct review to commence a collateral attack; there was no corresponding limit in prior law, although Rule 9(a) of the Rules Governing Section 2254 Cases in the United States District Courts established an amorphous standard linked to prejudice to the state. A prisoner could wait a decade without transgressing Rule 9(a). See Lonchar v. Thomas, - U.S. -, 116 S.Ct. 1293, 134 L.Ed.2d 440 (1996). Lindh got his collateral attack under way less than a year after the Supreme Court of Wisconsin decided his case. But suppose he had waited two years, filing in 1993 rather than 1992. Would we now apply § 2244(d) and dismiss the petition? Not at all. A prisoner’s decision to defer filing — perhaps while doing legal research, or while waiting for the Supreme Court to decide a case that could influence the selection of issues — is the sort of event to which the amended statute would “attach new legal consequences.” Courts treat a reduction in the statute of limitations as a rule for new eases only. See Landgraf, 511 U.S. at-n. 29, 114 S.Ct. at 1502 n. 29. And although no decision of the Supreme Court addresses the question directly, we do not doubt that the Court would give a plaintiff who files after the enactment a reasonable post-amendment time to get litigation underway. A new statute is unlike a judicial construction of an existing statute, which after Lampf, Pleva, Lipkind, Prupis & Petigrow v. Gilbertson, 501 U.S. 350, 111 S.Ct. 2773, 115 L.Ed.2d 321 (1991), and Jaimes B. Beam Distilling Co. v. Georgia, 501 U.S. 529, 111 S.Ct. 2439, 115 L.Ed.2d 481 (1991), governs all pending cases. Section 2244(d) is short enough that the “reasonable time” after April 24, 1996, and the one-year statutory period coalesce; reliance interests lead us to conclude that no collateral attack filed by April 23, 1997, may be dismissed under § 2244(d) and the parallel provision added to 28 U.S.C. § 2255 by § 105 of the 1996 Act.
But Lindh lacks any reliance interest nearly that strong. His able counsel identifies only two steps he took (rather, omitted) in supposed reliance on the prior state of the law: he did not ask the Supreme Court of the United States to review the decision of the Supreme Court of Wisconsin, and he did not file a collateral attack in state court. Let us start with the omission of a petition for certiorari. Lindh observes that the Supreme Court, engaged in direct review under 28 U.S.C. § 1257, would not have been (and still is not) affected by § 2254(d). That is true enough, but even under the former text of § 2254 petitioners faced obstacles (such as the exhaustion requirement, the special protection of state courts’ findings of fact, and the Teague doctrine) that did not apply on direct review under § 1257. A court exercising appellate jurisdiction has now, and had in 1991, more authority to correct errors than a court hearing a collateral attack. Defendants in state cases bypass their opportunity to seek certiorari not because they expect favorable treatment under § 2254, but because they recognize that the Supreme Court of the United States rarely grants such requests. “A petition for a writ of certiorari will be granted only for compelling reasons.” S.Ct. R. 10. As a practical matter, to obtain direct review Lindh would have had to establish not only the national importance of the question at hand, but also that the decision of the Supreme Court of Wisconsin interpreted the Constitution in a way disagreeing with the interpretation of the highest court of another state, or of a United States court of appeals. S.Ct. R. 10(b). Lindh does not argue that his case presents such an abstract legal question; instead he argues that the state’s courts misapplied settled law to the facts of his case. Such a contention had no prospect in 1991 of receiving an audience in the Supreme Court of the United States, and Lindh gave up nothing of value by deciding not to seek certiorari. Lightning can strike a petition for certiorari, but Lindh’s would not have been a good lightning rod.
As for the omission of a collateral attack in state court: Wisconsin would not have entertained one, had Lindh filed it. Wisconsin does not permit an inferior court to review a decision of the state’s highest court. Wis. Stat. § 974.06. A prisoner who can establish an unusual reason, such as an
*867intervening change of law, may wage a collateral attack in Wisconsin even after the state’s Supreme Court has addressed the merits, see State v. Escalona-Naranjo, 185 Wis.2d 168, 517 N.W.2d 157 (1994), but Lindh does not qualify for this exception — and, if he does, the state courts remain open to him. Lindh contends that the Supreme Court of Wisconsin misunderstood one factual question: whether Roberts’ privileges at the University of Wisconsin Hospital had been suspended by the time of trial. The court wrote that the record did not show such a suspension, 468 N.W.2d at 173, and Lindh says that this is wrong. If so, and if this is important, Lindh could have sought rehearing in the state court or could have introduced factual support for his position in the federal district court. He did neither, raising this contention for the first time only in supplemental briefs filed after the case had been set for rehearing en banc. Issues raised so belatedly are forfeited; we deny the motion to supplement the record. Lindh’s delay cannot be attributed to the 1996 Act, as he had ample opportunity to make his version of events known under both former and current § 2254 (the amended § 2254(e)(1), 110 Stat. 1219, authorizes a petitioner to dispute a factual determination by a state court, although proof of error requires clear and convincing evidence). Once again, therefore, Lindh’s litigating decisions were not affected by the difference between the versions of § 2254. The 1996 Act does not “impair rights a party possessed when he acted, increase a party’s liability for past conduct, or impose new duties with respect to transactions already completed.” Landgraf, 511 U.S. at-, 114 S.Ct. at 1505. The new law therefore governs our consideration of Lindh’s contentions.
D
Almost as an afterthought, Lindh (with the support of the American Bar Association, as amicus curiae) asserts that any alteration in the scope of collateral review after a prisoner has filed a petition under § 2254 violates Art. I, § 9, cl. 2 of the Constitution: “The Privilege of the Writ of Habeas Corpus shall not be suspended, unless when in Cases of Rebellion or Invasion the public Safety may require it.” Lindh does not cite any authority for this proposition, and we think the contention feckless. If the new § 2254(d) “suspends” the Great Writ, it does so no less for cases filed on April 25, 1996, than for cases pending on April 24. Yet to alter the standards on which writs issue is not to “suspend” the privilege of the writ. Felker so holds for another amendment made by the 1996 Act.
The writ known in 1789 was the pre-trial contest to the executive’s power to hold a person captive, the device that prevents arbitrary detention without trial. Ex parte Bollman & Swartwout, 8 U.S. (4 Cranch) 75, 2 L.Ed. 554 (1807); Ex parte McCardle, 73 U.S. (6 Wall.) 318, 18 L.Ed. 816 (1868); Ex parte Royall, 117 U.S. 241, 6 S.Ct. 734, 29 L.Ed. 868 (1886). See McCleskey v. Zant, 499 U.S. 467, 478, 111 S.Ct. 1454, 1461-62, 113 L.Ed.2d 517 (1991); Schlup v. Delo, — U.S. -, -, 115 S.Ct. 851, 862, 130 L.Ed.2d 808 (1995). The power thus enshrined did not include the ability to reexamine judgments rendered by courts possessing jurisdiction. Ex parte Kearney, 20 U.S. (7 Wheat.) 38, 5 L.Ed. 391 (1822); see also Dallin H. Oaks, Habeas Corpus in the States — 1776-1865, 32 U. Chi. L.Rev. 243, 244-45 (1965). Under the original practice, “a judgment of conviction rendered by a court of general criminal jurisdiction was conclusive proof that confinement was legal ... [and] prevented issuance of a writ”. United States v. Hayman, 342 U.S. 205, 211, 72 S.Ct. 263, 268, 96 L.Ed. 232 (1952); see also Ex parte Yerger, 75 U.S. (8 Wall.) 85, 101, 19 L.Ed. 332 (1869).
The founding-era historical evidence suggests a prevailing view that state courts were adequate fora for protecting federal ' rights. See Max Farrand, 1 Records of the Federal Convention of 1787 124-25 (1937). Based on this assumption, there was (and is) no constitutionally enshrined right to mount a collateral attack on a state court’s judgment in the inferior Article III courts and, a fortiori, no mandate that state court judgments embracing questionable (or even erroneous) interpretations of the federal Consti*868tution be reviewed by the inferior Article III courts.
Collateral review of judgments accordingly is subject to legislative control, as the first Congress, dominated by those who wrote and ratified the Constitution, recognized. See Ex parte Watkins, 28 U.S. (3 Pet.) 193, 7 L.Ed. 650 (1830). Section 14 of Judiciary Act of 1789 prohibited any inquiry by the federal courts into the propriety of state custody. 1 Stat. 82; see Ex parte Dorr, 44 U.S. (3 How.) 103, 11 L.Ed. 514 (1845). Congress first created a general power of collateral review in 1867, 14 Stat. 385, and then repealed it the next year, 15 Stat. 44, a step sustained in McCardle. (Limited power's creatéd in 1833, 4 Stat. 634-35, to protect federal officers from state interference, and in 1842, 5 Stat. 539^10, to protect foreign nationals, did not affect the bulk of criminal cases.) The general power was not reestablished until 1885, 23 Stat. 437, and has been subject to frequent revision by Congress — and reinterpretation by the Supreme Court — ever since. Collateral review of judgments entered after full opportunity for litigation is the work of the 20th Century. Whether the first occasion came in 1953 with Brown v. Allen, or that honor belongs instead to Waley v. Johnston, 316 U.S. 101, 62 S.Ct. 964, 86 L.Ed. 1302 (1942), or eyen Frank v. Mangum, 237 U.S. 309, 35 S.Ct. 582, 59 L.Ed. 969 (1915), which stretched the rule that a writ may issue when the convicting court lacks “jurisdiction” (see Ex parte Lange, 85 U.S. (18 Wall.) 163, 21 L.Ed. 872 (1874)), does not matter. Justices debated the question in Wright v. West, 505 U.S. 277, 112 S.Ct. 2482, 120 L.Ed.2d 225 (1992), as a matter of statutory interpretation. “Judgments about the proper scope of the writ are ‘normally for Congress to make.”’ Felker, - U.S. at -, 116 S.Ct. at 2340, quoting from Lonchar, - U.S. at-, 116 S.Ct. at 1298. Any suggestion that the Suspension Clause forbids every contraction of the powers bestowed by Congress in 1885, and expanded by the 1948 and 1966 amendments to § 2254, is untenable. The Suspension Clause is not a ratchet.
II
Section 2254(d)(1) provides that a writ may not issue unless the state court’s decision “was contrary to, or involved an unreasonable application of, clearly established Federal law, as determined by the Supreme Court of the United States”. What does this mean? Is it congruent with the judicial role established by Article III of the Constitution? And how does it affect Lindh’s contentions?
A
The Conference Report on the 1996 Act says that § 2254(d) “requires deference to the determinations of state courts that are neither ‘contrary to,’ nor an ‘unreasonable application of,’ clearly established federal law.” H.R.Conf.Rep. 104r-518, 104th Cong., 2d Sess. Ill (1996). This passage, together with similar comments on the floor, has led to an unproductive debate among the parties and the amici about what “deference” entails. It is a wonderful illustration why legislative history so often misleads — for the word “deference” does not appear in the statute. It does not tell us to “defer” to state decisions, as if the Constitution means one thing in Wisconsin and another in Indiana. Nor does it tell us to treat state courts the way we treat federal administrative agencies. Deference after the fashion of Chevron U.S.A Inc. v. Natural Resources Defense Council, Inc., 467 U.S. 837, 104 S.Ct. 2778, 81 L.Ed.2d 694 (1984), depends on delegation. See Adams Fruit Co. v. Barrett, 494 U.S. 638, 110 S.Ct. 1384, 108 L.Ed.2d 585 (1990). Congress did not delegate either interpretive or executive power to the state courts. They exercise powers under their domestic law, constrained by the Constitution of the United States. “Deference” to the jurisdictions bound by those constraints is not sensible. We see no need to pursue this subject, however, because our task is to construe the enacted statute, not to construe its legislative history.
The first phrase' of § 2254(d)(1)— authorizing a federal court to issue the writ when the state court’s decision is “contrary to ... clearly established Federal law, as determined by the Supreme Court of the United States”, preserves rather than under*869mines federal courts’ independent interpretive power. Federal courts are free to express an independent opinion on all legal issues in the ease. So if, for example, the Supreme Court of Wisconsin had held that the Confrontation Clause does not entitle defendants to cross-examine witnesses to establish bias, then Davis v. Alaska would show that the decision is “contrary to ... clearly established Federal law, as determined by the Supreme Court of the United States”, even if decent arguments could be constructed that Davis misunderstood the Confrontation Clause. Section 2254(d) requires us to give state courts’ opinions a respectful reading, and to listen carefully to their conclusions, but when the state court addresses a legal question, it is the law “as determined by the Supreme Court of the United States” that prevails.
This is a retrenchment from former practice, which allowed the United States courts of appeals to rely on their own jurisprudence in addition to that of the Supreme Court. The novelty in this portion of § 2254(d)(1) is not the “contrary to” part but the reference to “Federal law, as determined by the Supreme Court of the United State s” (emphasis added). This extends the principle of Teague by limiting the source of doctrine on which a federal court may rely in addressing the application for a writ. It does not, however, purport to limit the federal courts’ independent interpretive authority with respect to federal questions.
An application for a writ is only an invitation to a federal court to grant the requested relief upon its review of a criminal conviction and confinement under state law. See Felker, - U.S. at-, 116 S.Ct. at 2339. The authority to issue the writ is conferred on courts by statute. Ex parte Bollman & Swartwout, supra. It is doetri-nally distinct from the statutory grant of jurisdiction to decide a case or controversy, e.g., 28 U.S.C. § 1331. Yet just as Congress may restrict the jurisdiction of the inferior Article III courts, so it may prescribe limits on the granting of the extraordinary relief provided by the writ of habeas corpus.
A conflict between decisions of the Seventh Circuit and the Supreme Court of Wisconsin would not authorize issuance of a writ under § 2254(d)(1). State courts must knuckle under to decisions of the Supreme Court, but not of this court. So Lindh must be able to point-to an authoritative decision of the Supreme Court in order to secure a writ. He has Davis and Van Arsdall. We therefore need not shoulder the potentially difficult task of determining when an appellate gloss on a decision of the Supreme Court has so far departed from its wellsprings as to be the “real” source of law, or when the chain of reasoning has stretched so long that the rule cannot be called “clearly established” as of the time the state court rendered its decision. These are the sorts of questions presented by the Teague standard, e.g., Gray v. Nether-land, - U.S. -,-, 116 S.Ct. 2074, 2083-85, 135 L.Ed.2d 457 (1996); Sawyer v. Smith, 497 U.S. 227, 236, 110 S.Ct. 2822, 2828, 111 L.Ed.2d 193 (1990); Saffle v. Parks, 494 U.S. 484, 488, 110 S.Ct. 1257, 1260, 108 L.Ed.2d 415 (1990) (“compelled by existing precedent”), and the “clearly established” language is unlikely to pose a different kind of interpretive challenge.
The new § 2254(d)(1) adds a twist to the complexities of the Teague inquiry, however, because it explicitly identifies only the Supreme Court as the font of “clearly established” rules. This language could be understood to adopt an unfortunate view that applies the unitary model of the executive department to the judicial department as well. Such a view of the judiciary, which runs counter to the explicit textual differences between the vesting clauses of Articles II and III, assigns to the inferior Article III courts a role analogous to that of an executive agency. From this flawed perspective, the inferior courts are seen as agencies of the judicial department with authority derived from the power vested in the Supreme Court.
The Constitution vests the judicial power “in one supreme Court, and in such inferior Courts as the Congress may from time to time ordain and establish.” Art. Ill § 1. The vesting clause of Article III is quite different from the vesting clause of Article II, which says that the “executive Power shall be vested in a President of the United *870States of America.” Art. II § 1 cl. 1. By virtue of the grant in Article II to a single person, agencies of the executive department exercise executive authority on behalf of the President. See Myers v. United States, 272 U.S. 52, 47 S.Ct. 21, 71 L.Ed. 160 (1926).
In contrast, the judicial power is given by the Constitution to each of the inferior Article III courts. Accordingly, when these courts act within their jurisdictions, they exercise the judicial power directly, though subject to the limits of revisionary jurisdiction of a court superior within the Article III hierarchy. The inferior courts do not rely on the Supreme Court for their authority to exercise the judicial power, for this power is given them directly by Article III once they are ordained, established, and provided with jurisdiction by Congress.
Lindh would like us to stop here, to announce that § 2254(d) has not changed the law in any way material to his case, and to proceed to decision. We cannot do that, because one ingredient of Lindh’s claim is not well established. The Confrontation Clause (and therefore the principles of Davis and Van Arsdall) governs at trial of the merits, but not at sentencing. Is the dispositional phase of a bifurcated trial closer to the merits, or to sentencing? We return to that question in Part II.C. Moreover, we do not read the “contrary to ...” language to authorize issuance of a writ whenever a court errs, as Lindh and the amici prefer; that would vitiate the second (“unreasonable application”) clause of § 2254(d)(1). The two clauses have separate functions. We must ensure that the Supreme Court of Wisconsin adheres to legal principles articulated by the Supreme Court of the United States. But when the dispute lies not in the meaning of the Constitution, but in its application to a particular set of facts — when it is, in the standard phrase, a “mixed question of law and fact” — sec. 2254(d)(1) restricts the grant of collateral relief to cases in which the state’s decision reflects “an unreasonable application of’ the law.
The reference to an “unreasonable application of Federal law” sounds like the official-immunity question in constitutional-tort litigation, but we doubt that this is quite the right analogy. Principles of immunity are designed to draw the line between prospective relief and damages from a governmental body, on the one hand, and payment from the pocket of a public employee, on the other. Section 2254(d)(1) serves an entirely different function. Perhaps, then, the unreasonable-application language should be understood as another variation on Teague. Recall the quotation in Lockhart from Butler: Teag-ue “validates reasonable, good-faith interpretations of existing precedents made by state courts even though they are shown to be contrary to later decisions.” Section 2254(d)(1) generalizes this principle of respect by withdrawing the “later decisions” proviso.
How much leeway does the “unreasonable application” language create? None on questions of interpretation. It does not authorize or permit state courts to deviate from the Constitution. Federal courts acting within their jurisdiction are always entitled to interpret the law independently. Section 2254(d)(1) as we read it does no more than regulate relief. It tells federal courts: Hands off, unless the judgment in place is based on an error grave enough to be called “unreasonable.” Historical practice (see Part I.D above) likewise counseled restraint in use of the writ. Other rules limiting the remedy abound. We have mentioned official immunity cases, in which federal courts retain full interpretive power, but forbear from using the remedy of damages. Consider, too, the doctrine of United States v. Leon, 468 U.S. 897, 104 S.Ct. 3405, 82 L.Ed.2d 677 (1984), under which the exclusionary rule does not apply to evidence gathered in reasonable reliance on a search warrant, even if the warrant was issued in violation of the Constitution. In such cases the court retains (but need not exercise) the power to identify the state court’s violation of the Fourth Amendment, see 468 U.S. at 925,104 S.Ct. at 3422, but the remedy of suppression (and a new trial) is withheld unless the violation is obvious, or the officer’s reliance unreasonable. The inevitable-discovery doctrine similarly specifies situations in which a constitutional violation (established by independent review in the federal courts) does not lead to *871the remedy of upsetting a criminal conviction. See Nix v. Williams, 467 U.S. 431, 104 S.Ct. 2601, 81 L.Ed.2d 377 (1984).
None of this answers the question when a departure is so great as to be “unreasonable,” for that question lacks an abstract answer, just as courts have been unable to give precise content to phrases such as “abuse of discretion.” Application of Lean therefore has required careful inquiry one case at a time, and we do not see how application of § 2254(d)(1) can be much different. For current purposes it is enough to say that when the constitutional question is a matter of degree, rather than of concrete entitlements, a “reasonable” decision by the state court must be honored. By posing the question whether the state court’s treatment was “unreasonable,” § 2254(d)(1) requires federal courts to take into account the care with which the state court considered the subject.
Questions of degree — like questions about the proper use of “discretion” — lack answers to which the labels “right” and “wrong” may be attached. When the subject is painted in shades of grey, rather than in contrasting colors, a responsible, thoughtful answer reached after a full opportunity to litigate is adequate to support the judgment. Think of the Speedy Trial Clause of the Sixth Amendment, which after Barker v. Wingo, 407 U.S. 514, 92 S.Ct. 2182, 33 L.Ed.2d 101 (1972), does not prescribe a rule for “how long is too long” but rather establishes a list of factors to consider. The Supreme Court of the United States sets the bounds of what is “reasonable”; a state decision within those limits must be respected — not because it is right, or because federal courts must abandon their independent decisionmaking, but because the grave remedy of upsetting a judgment entered by another judicial system after full litigation is reserved for grave occasions. That is the principal change effected by § 2254(d)(1).
B
The American Bar Association and a group of former federal judges, appearing as amici curiae, contend that § 2254(d)(1) is unconstitutional to the extent it requires anything less than plenary review of all contentions based on federal law. Their argument, distinct from Lindh’s invocation of the Suspension Clause, is that the “judicial Power of the United States” (Art. Ill § 1) is the power to interpret the law independently. They rely on Marburg v. Madison, 5 U.S. (1 Cranch) 137, 177, 2 L.Ed. 60 (1803) (“[i]t is emphatically the province and duty of the judicial department to say what the law is”) and modern statements such as Plaut v. Spendthrift Farm, Inc., - U.S.-,-, 115 S.Ct. 1447, 1453, 131 L.Ed.2d 328 (1995), and Gutierrez de Martinez v. Lamagno, - U.S. -, -, 115 S.Ct. 2227, 2234, 132 L.Ed.2d 375 (1995).
If by this line of argument the amici mean that federal courts must give judgment without regard to the legal views of other public actors, and without regard to the resolution of contested issues in state litigation, then their argument reaches far beyond § 2254(d). It would mean that deference in administrative law under Chevron is unconstitutional; that the respect accorded to Congress when it speaks on constitutional questions (e.g., Wisconsin v. New York City, - U.S.-, -, 116 S.Ct. 1091, 1100-01, 134 L.Ed.2d 167 (1996); Weiss v. United States, 510 U.S. 163, 177, 114 S.Ct. 752, 760-61, 127 L.Ed.2d 1 (1994); Donovan v. Dewey, 452 U.S. 594, 603, 101 S.Ct. 2534, 2540, 69 L.Ed.2d 262 (1981); United States v. Watson, 423 U.S. 411, 416, 96 S.Ct. 820, 824, 46 L.Ed.2d 598 (1976)), must be abandoned; and that the Full Faith and Credit Clause (Art. IV § 1) conflicts with Article III. This position would demolish numerous doctrines in the law of collateral attack that no one (until now) has supposed pose constitutional difficulties.
Consider Stone v. Powell, 428 U.S. 465, 482, 96 S.Ct. 3037, 3046, 49 L.Ed.2d 1067 (1976), which holds that when a “State has provided an opportunity for full and fair litigation of a Fourth Amendment claim, the Constitution does not require that a state prisoner be granted federal habeas relief on the ground that evidence obtained in an unconstitutional search and seizure was introduced at the trial.” Stone contemplates— indeed it requires — that a federal court re-*872irain from issuing a writ under § 2264 even though the court is convinced that the state judges erred on the law, and even though the error altered the outcome of the ease. On several occasions the Supreme Court has treated Stone as a doctrine that on grounds of prudence could be extended to other subjects (although the Court has not so extended it). See Reed v. Farley, 512 U.S. 339, 114 S.Ct. 2291, 129 L.Ed.2d 277 (1994) (claims under the Interstate Agreement on Detain-ers); Withrow v. Williams, 507 U.S. 680, 113 S.Ct. 1745, 123 L.Ed.2d 407 (1993) (claims based on Miranda v. Arizona, 384 U.S. 436, 86 S.Ct. 1602, 16 L.Ed.2d 694 (1966)). If the Court could adopt Stone and consider extending its scope, creating a gap between the “merits” and the obligation to issue a writ, then Congress may do the same by amending § 2254.
Teague likewise establishes a disjunction between the meaning of the Constitution and the use of habeas corpus. If a state judgment becomes final in 1992, and the Supreme Court articulates a new rule of constitutional law in 1993, then a petition for collateral review in 1994 will fail — not because the state court was “right” on the merits, but because some errors of constitutional law do not support collateral relief. Section 2254(d)(1) codifies and extends the principle of Teojgue, and if Teague is consistent with Article III, then so is § 2254(d)(1) as we have construed it.
The amici curiae neglect a basic distinction that Plant recognizes: Congress cannot tell courts how to decide a particular case, but it may make rules that affect classes of cases. See - U.S. at - - -, 115 S.Ct. at 1452-53; see also Robertson v. Seattle Audubon Society, 503 U.S. 429, 441, 112 S.Ct. 1407, 1414-15, 118 L.Ed.2d 73 (1992). Congress cannot say that a court must award Jones $35,000 for being run over by a postal truck, but it may prescribe maximum damages for categories of cases, Carter v. United States, 982 F.2d 1141 (7th Cir.1992), or provide that victims of torts by federal employees cannot receive punitive damages, see 28 U.S.C. § 2674. It may establish that if the driver was acting within the scope of his employment, the United States must be substituted as a party and the driver dismissed — even if that turns out to deprive the victim of compensation. See United States v. Smith, 499 U.S. 160, 111 S.Ct. 1180, 113 L.Ed.2d 134 (1991). Even for claims based on the Constitution, there may be rights without remedies. FDIC v. Meyer, 510 U.S. 471, 484-86, 114 S.Ct. 996, 1005-06, 127 L.Ed.2d 308 (1994); Schweiker v. Chilicky, 487 U.S. 412, 108 S.Ct. 2460, 101 L.Ed.2d 370 (1988); Bush v. Lucas, 462 U.S. 367, 103 S.Ct. 2404, 76 L.Ed.2d 648 (1983); Chappell v. Wallace, 462 U.S. 296, 103 S.Ct. 2362, 76 L.Ed.2d 586 (1983). So too with § 2254(d)(1). Congress lacks power to revise the meaning of the Constitution or to require federal judges to “defer” to the interpretations reached by state courts. Once the judicial power is brought to bear by the presentation of a justiciable case or controversy within a statutory grant of jurisdiction, the federal courts’ independent interpretive authority cannot constitutionally be impaired. Regulating relief is a far cry from limiting the interpretive power of the courts, however, and Congress has ample power to adjust the circumstances under which the remedy of the writ of habeas corpus is deployed.
This distinction between rights and remedies is fundamental. Every day, courts decline to disturb judgments that they know are wrong. This is the principal function of the law of judgments. Suppose A and B are plaintiffs in the same lawsuit, which they lose; A appeals and wins, while B does not appeal. It has now been established that the judgment against B is wrong. May B file a new suit to obtain the benefit of A’s victory? The answer is “no.” Federated Department Stores, Inc. v. Moitie, 452 U.S. 394, 101 S.Ct. 2424, 69 L.Ed.2d 103 (1981). The Full Faith and Credit Clause is designed to make this result a matter of constitutional entitlement. And although today we think of claim preclusion (res judicata) as a specialty of civil law, it is only in this century that courts have treated civil and criminal judgments differently. Recall the discussion in Part I.D above. See also Felker, - U.S. at -, 116 S.Ct. at 2340; Paul M. Bator, Finality in Criminal Law and Federal Habeas Corpus for State Prisoners, 76 Harv.L.Rev. 441, 465-99 (1963) (tracing the history).
*873In suits under 42 U.S.C. § 1983 and Bivens v. Six Unknown Named Agents, 403 U.S. 388, 91 S.Ct. 1999, 29 L.Ed.2d 619 (1971), courts frequently rule for the defendant even though the plaintiff may be right on the merits. Public employees receive the benefit of the doubt on legal questions and must pay damages only when the legal right has been sufficiently well established and particularized that a reasonable official would have understood that what he is doing violates that right. See Anderson v. Creighton, 483 U.S. 635, 640, 107 S.Ct. 3034, 3039, 97 L.Ed.2d 523 (1987); Auriemma v. Rice, 910 F.2d 1449 (7th Cir.1990) (en bane). Section 2254(d)(1) creates a related approach and is no less consistent with Article III. Even in criminal cases, courts sometimes enforce decisions they would not have made in the first instance. Stone, Teague, Leon, Nix, and the harmless-error cases, e.g., Brecht v. Abrahamson, 507 U.S. 619, 113 S.Ct. 1710, 123 L.Ed.2d 353 (1993), are among many illustrations of the gap between having a good legal argument and winning release from custody. X and Y are indicted for a joint crime. X pleads guilty; Y pleads double jeopardy and wins, whereupon X claims that his sentence is illegal and asks for relief. United States v. Broce, 488 U.S. 563, 109 S.Ct. 757, 102 L.Ed.2d 927 (1989), holds that X must serve his sentence, because the plea of guilty waives even arguments that in retrospect are known to be correct.
Section 2254(d)(l)’s requirement that judges apply “Federal law, as determined by the Supreme Court of the United States”, rather than their own understanding of the law, is consistent with the hierarchical nature of the federal judiciary. Judges of the inferi- or courts must implement the views of their superiors, from which it follows that many decisions of the lower courts will be inconsistent with the conclusions their judges would have reached, if unfettered by precedent. Applying, even predicting, the work of other judges, rather than reaching independent conclusions, makes up the bulk of the work of a federal judge — not only when interpreting the decisions of the Supreme Court, but also when deciding cases under the diversity jurisdiction, see Erie R.R. v. Tompkins, 304 U.S. 64, 58 S.Ct. 817, 82 L.Ed. 1188 (1938), and when coping with federal law that is not uniform geographically, see Eckstein v. Balcor Film Investors, 8 F.3d 1121, 1126-27 (7th Cir.1993); Olcott v. Delaware Flood Co., 76 F.3d 1538, 1544-48 (10th Cir.1996). See also Lehman Brothers v. Schein, 416 U.S. 386, 390-91, 94 S.Ct. 1741, 1743-44, 40 L.Ed.2d 215 (1974); Charles Alan Wright, Arthur R. Miller & Edward H. Cooper, 17A Federal Practice and Procedure § 4248 (2d ed. 1988), discussing a circumstance under which a state court’s decision is directly binding in federal litigation: the certification of a question of law to a state court. Article III does not establish a system under which judges of the inferior federal courts always must render judgment without regard to the conclusions of other courts.
Shortly before Brown v. Allen changed the rules, Learned Hand could write with confidence that
upon habeas corpus a federal court does not in any sense review the decision in the state courts. Here, for example, the District Court could not properly have issued the writ, no matter how erroneous the judge had thought the state judge’s conclusion that the evidence did not make out a prima facie case of the deliberate use of perjured testimony. The writ was limited to the assertion of the relator’s rights under the Fourteenth Amendment; and due process of law does not mean infallible process of law. If the state courts have honestly applied the pertinent doctrines to the best of their ability, they have accorded to an accused his constitutional rights.
Schechtman v. Foster, 172 F.2d 339, 341 (2d Cir.1949). This expression of the longstanding distinction between unlawful custody, which supported a writ of habeas corpus, and unlawful procedure in the course of a trial, which did not, reflected the law of the United States until 1953. Congress has elected to move back in that direction — but hardly very far, preserving independent federal review on pure questions of law, and subjecting mixed questions of law and fact to review for reasonableness. We would have to cast history to the winds to say that this decision, which respects fully-litigated judgments un*874less the state court has gone seriously wrong, transgresses constitutional limitations.
The American Bar Association’s further argument (which Lindh joins) that the Due Process Clause of the Fifth Amendment protects § 2254 from amendment is unpersuasive. The ABA contends that because federal “courts traditionally have conducted a plenary review of [state] courts’ rulingfs] on legal issues”, the entitlement to such a review has become a “fundamental right” protected by the substantive component of due process analysis. To establish this “traditional” role, the ABA cites no case before Norris v. Alabama, 294 U.S. 587, 55 S.Ct. 579, 79 L.Ed. 1074 (1935), a decision that is hardly pertinent. (Norris was not a collateral attack but arose on certiorari to a state court under what is now § 1257.) The summary of collateral review in Part I.D of this opinion, and Learned Hand’s assessment of the state of the law in 1949, should disabuse any observer of the belief that plenary federal review of state criminal judgments is so firmly rooted in American tradition that any alteration has passed beyond legislative power. The Constitution contains an express limit on the power of Congress over the writ of habeas corpus — the Suspension Clause. It is not an appropriate exercise of the “judicial Power” to supplement the Suspension Clause with a rule that enhancements of the 20th Century cannot be altered.
C
At last we reach Lindh’s arguments about the restrictions the trial judge placed on his lawyer’s cross-examination of Dr. Roberts. Lindh insists that he was entitled to cross-examine Roberts about the allegations of sexual misconduct in order to show, if not actual bias, a reason Roberts may have had to make his assessment more favorable to the prosecutor. Wisconsin replies with two arguments: first, that the decision is not “contrary to ... clearly established Federal law” because the Supreme Court has never held that witnesses during the dispositional phase of bifurcated trials are subject to cross-examination; second, that if the Confrontation Clause applies, the decision to limit cross examination was not “an unreasonable application of’ federal law, because under the Confrontation Clause “trial judges retain wide latitude ... to impose reasonable limits on such cross-examination based on concerns about, among other things, harassment, prejudice, confusion of the issues, the witness’ safety, or interrogation that is repetitive or only marginally relevant.” Van Arsdall, 475 U.S. at 679, 106 S.Ct. at 1435.
Lindh rejoins that the first of these arguments is untenable because the Supreme Court of Wisconsin did not decide the case on that ground. Recall that the amended § 2254(d) begins by saying that a petition must not be granted “with respect to any claim that was adjudicated on the merits in State court proceedings” unless subsection (1) or (2) authorizes that step. Because the Supreme Court of Wisconsin addressed only the propriety of restricting the scope of cross-examination, it did not adjudicate on the merits the state’s “claim” about the scope of the confrontation right. Lindh has this backwards. Wisconsin is not making any “claim” based on the Confrontation Clause. It is defending against Lindh’s claim that the Confrontation Clause entitled him to cross-examine Roberts about the criminal investigation of the sexual-misconduct allegations. To prevail on this contention, Lindh must establish two propositions: that the Confrontation Clause applies at the dispositional phase of a bifurcated trial, and that the limitations on cross-examination were excessive. The Supreme Court of Wisconsin resolved this claim adversely to Lindh on the merits, by deciding the second proposition against him. Nothing in § 2254(d) calls on state courts to fill their opinions with discussions that by their lights are unnecessary, as the price of avoiding de novo review in federal court. There is a fundamental difference in federal practice between a “claim” and a legal theory. A “claim” is a demand for relief from an identified injury, which may be supported (or defeated) by many different theories. See Christianson v. Colt Industries Operating Corp., 486 U.S. 800, 807-10, 108 S.Ct. 2166, 2173-75, 100 L.Ed.2d 811 (1988); Herrmann v. Cencom Cable Associates, Inc., 999 F.2d 223 (7th Cir.1993). The injury here is the restriction of cross-exami*875nation; the relief, a new trial. That claim has been resolved on the merits.
Lindh’s contrary position would have especially unfortunate consequences when state law leads a state court to bypass a particular legal theory. Suppose Wisconsin decides as a matter of its domestic law to extend the right of cross-examination to sentencing. A defendant then contends that he is entitled to cross-examine a given witness to uncover bias; the state’s highest court disagrees on the ground that the potential for bias is too remote to justify what might be a lengthy line of questioning. If the defendant then seeks a writ of habeas corpus, a state should be entitled to reply that there is no “clearly established Federal law” creating any right of cross-examination in non-capital sentencing — indeed, that it is clearly established that there is none, see Williams v. Oklahoma, 358 U.S. 576, 584, 79 S.Ct. 421, 426, 3 L.Ed.2d 516 (1959); Williams v. New York, 337 U.S. 241, 249-50, 69 S.Ct. 1079, 1084-85, 93 L.Ed. 1337 (1949); United States v. Wise, 976 F.2d 393, 397 (8th Cir.1992) (en bane) — and that the state court’s decision therefore cannot be “contrary to ... clearly established Federal law”. By Lindh’s argument this response is impermissible, which could transmute a rule of state law (the extension of cross-examination to sentencing) into a rule of constitutional law, at least for purposes of that case. Yet one well established rule, not altered by the amendments, is that “[a] federal court may not issue the writ on the basis of a perceived error of state law.” Pulley v. Harris, 465 U.S. 37, 41, 104 S.Ct. 871, 875, 79 L.Ed.2d 29 (1984). See also, e.g., Gilmore v. Taylor, 508 U.S. 333, 342, 344, 113 S.Ct. 2112, 2117-18, 2119, 124 L.Ed.2d 306 (1993); Estelle v. McGuire, 502 U.S. 62, 67-68, 112 S.Ct. 475, 480, 116 L.Ed.2d 385 (1991); Smith v. Phillips, 455 U.S. 209, 221, 102 S.Ct. 940, 948, 71 L.Ed.2d 78 (1982). If the Confrontation Clause does not apply to a particular proceeding, then a mistaken restriction on the scope of cross-examination is nothing but an error of state law (which is how Justice Abra-hamson’s dissenting opinion approached Lindh’s arguments). We therefore conclude that Wisconsin is entitled to make both of its arguments, and we turn to the first.
Starting from the premise that the Confrontation Clause is inapplicable to sentencing, Wisconsin contends that the dispositional phase of a bifurcated trial is more like sentencing than it is like the trial of the merits. At the first phase of Lindh’s trial, the jury determined that he deliberately killed two people and tried to kill a third; he pleaded guilty to two other crimes. Having established all elements of the crimes beyond a reasonable doubt, Wisconsin contends, it was entitled to treat everything that followed as a matter of what should be done with the criminal — whether to imprison him (and, if so, where and for how long) or whether to “treat” him. As the state sees things, the two Williams cases necessarily establish that a person who provides evidence relevant only to disposition is not a “witness against” the defendant for purposes of the Confrontation Clause (“[i]n all criminal prosecutions, the accused shall enjoy the right ... to be confronted with the witnesses against him”).
Like questions about sentencing, the consequences of an insanity finding may be removed from the jury’s purview. Shannon v. United States, 512 U.S. 573, 114 S.Ct. 2419, 129 L.Ed.2d 459 (1994). Because “lack of mental disease” is not an element of the crimes under state law, Wisconsin was free to decide that question by a preponderance of the evidence and even to assign the burden of persuasion to the defendant. A decision that Lindh was insane at the time of the shootings would not entitle him to release; it would only change the place of his confinement. The due process clause distinguishes prisons from mental institutions, see Vitek v. Jones, 445 U.S. 480, 100 S.Ct. 1254, 63 L.Ed.2d 552 (1980), but the Court held in that case that a state could choose to hold a criminal in a mental institution using the procedures for civil rather than criminal trials. And although an insanity finding would permit Lindh to seek release if he could later establish his recovery, this is not dramatically different from the opportunity to seek release from an indeterminate penal sentence on showing “rehabilitation,” a feature of the federal sentencing system until the last decade and of some state systems even today. So the state submits.
*876Lindh does not have a decision of the Supreme Court clearly establishing that this line of argument is mistaken. Neither does he contend that the historical practice in criminal prosecutions exposed psychiatric testimony to cross-examination for bias even if the state bifurcated its proceedings. Instead he invokes a different set of analogies. If Wisconsin had established a unitary system, in which insanity was a defense on the merits (or in which sanity was an element of the offense), then Roberts plainly would have been a “witness against” him; and if a psychiatrist would have been a witness against the defendant in a unitary trial, how can bifurcation change things?, Lindh wonders. Perhaps one could respond that a chemist testifying at trial that a certain white powder is cocaine is a witness against the defendant, but the same chemist testifying at sentencing about the same subject (to determine the quantity of the drug) is not; the contradiction is built into the Supreme Court’s current jurisprudence.
Not prepared to concede that point, Lindh contends that Specht v. Patterson, 386 U.S. 605, 87 S.Ct. 1209, 18 L.Ed.2d 326 (1967), establishes that psychiatrists are “witnesses against” defendants even at sentencing. If that is correct, then Specht would “clearly establish” the lesser included proposition that defendants are entitled to cross-examine psychiatrists at the dispositional phase of a bifurcated trial. Specht actually stands for a different point, however: that if a state separates issues in such a way that the main trial does not establish all elements essential to imposition of the maximum punishment to which the defendant is exposed, the protections of the Confrontation Clause (and many other provisions) extend to ensuing proceedings, whether or not they are called “sentencing.” Compare Bullington v. Missouri, 451 U.S. 430, 101 S.Ct. 1852, 68 L.Ed.2d 270 (1981), and Gardner v. Florida, 430 U.S. 349, 97 S.Ct. 1197, 51 L.Ed.2d 393 (1977), with Schiro v. Farley, 510 U.S. 222, 114 S.Ct. 783, 127 L.Ed.2d 47 (1994), and Gray v. Netherland, supra. Specht was convicted of “indecent liberties,” a crime that in Colorado carried a 10-year maximum sentence. But he was sentenced under that state’s Sex Offenders Act, which allowed the judge to impose an indeterminate term of one day to life on finding that the defendant “constitutes a threat of bodily harm to members of the public, or is an habitual offender and mentally ill.” The Court held in Specht that psychiatric testimony bearing on these issues was subject to cross-examination because it increased the maximum punishment beyond that set by law for the crime of conviction. As a practical matter, Colorado had made it a crime, punishable by life in prison, to be a mentally ill habitual sex offender. Wisconsin has not done anything of the kind. Murder is punishable by life in prison; the elements proved beyond a reasonable doubt at the first phase of Lindh’s trial therefore authorize the punishment imposed on him. The dispositional phase offered Lindh the opportunity for mitigation, quite unlike proceedings under the Sex Offenders Act in Colorado.
It may well be that, when the Supreme Court finally considers the question at issue here, it will decide that all testimony bearing on the defendant’s mental state, and its effect on his punishment, comes from a “witness against” the accused, and therefore must be subject to cross-examination, no matter how the state structures the trial and sentencing process. But it has not yet so held, and such a conclusion would be a nontrivial extension of current law. We think it impossible to say that “clearly established Federal law, as determined by the Supreme Court of the United States”, entitled Lindh to cross-examine Roberts about the allegations of sexual misconduct with patients.
Lindh therefore has to demonstrate that the state court’s decision “involved an unreasonable application of, clearly established Federal law”. Davis and Van Arsdall, which demonstrate that the Confrontation Clause entitles defendants to cross-examine adverse witnesses to expose their bias to the jury, are equally clear that the proper scope of cross-examination is a matter of degree. We quoted the pertinent language of Van Arsdall above. The Supreme Court of Wisconsin asked the legally correct question by inquiring whether the trial judge abused his discretion; this is exactly how Van Arsdall poses the issue. And the fact-specific answer can*877not be called “unreasonable” even if it is wrong (as some members of this court who join this opinion believe). Roberts knew that prosecutors were investigating allegations made against him, but, by the time he filed his final report and testified, he also knew that the Dane County prosecutor was not in charge. He may have thought that the Milwaukee County prosecutor would reward aid to the Dane County prosecutor — but if Roberts thought that, he probably also believed that his many prior engagements as a prose-cutorial expert would carry even more weight. Hindsight implies as much. The special prosecutor filed misdemeanor charges, to which Roberts pleaded guilty in January 1989. He was sentenced to probation without time in prison, and he agreed to surrender his medical license. The special prosecutor informed the sentencing judge of Roberts’ many appearances as an expert witness over the course of his long career; Lindh’s case was not singled out.
Nothing in the record shows that Roberts altered his views between his initial interview with Lindh (before any allegations of misconduct had come to light) and his testimony at trial. The possibility that Roberts altered his views — telling the prosecutors one thing orally at the outset of the case but offering another view, less favorable to Lindh, in the written report and on the stand — was open to unfettered cross-examination. Lindh’s counsel did not need to mention the sexual misconduct allegations to explore this question in depth. Under these circumstances, a reasonable and responsible judge might conclude that cross-examination based on the allegations that Roberts had sexual relations with three female patients would divert attention from the principal issue at hand— whether Lindh had a mental disease in January 1988 — without producing concrete evidence of bias. When a district judge comes to such a conclusion, a federal court of appeals will exercise deferential review, see United States v. Hernandez, 84 F.3d 931, 933-34 (7th Cir.1996), and will accord greater weight to thoughtfully reasoned decisions. Cf. United States v. Beasley, 809 F.2d 1273, 1278-79 (7th Cir.1987). Hernandez draws, for federal practice, a line similar to that of § 2254(d)(1): De novo appellate review for core legal issues, such as whether the Confrontation Clause applies, and deferential review for operational decisions within areas in which the constitutional norm allows discretion. Section 2254(d)(1) requires a federal court hearing a collateral attack to accord at least that much respect to the state courts.
The opinion of the Supreme Court of Wisconsin in Lindh’s case is careful; it correctly states the holdings of Davis and Van Ars-dall; it does not transgress any clearly established principles; instead it addresses a matter of degree about which thoughtful people can, and do, differ. Several portions of the opinion suggest that as a matter of state law the court was holding the trial judge to an even higher standard, which it thought satisfied. Lindh believes that the state court put too much weight on objective factors (such as whether the Dane County prosecutor could reward favorable testimony) and not enough on subjective ones (such as the possibility that Roberts credited the Dane County prosecutor with more influence than he possessed, or was willing to exercise). By restricting in the new § 2254(d)(1) the scope of the extraordinary relief provided by the writ of habeas corpus, Congress has instructed the inferior federal courts to refrain from this sort of fine tuning. We hold accordingly that Lindh is not entitled to a writ of habeas corpus.
AFFIRMED.