dissenting in part.
Patrick Wright had a troubled childhood and has led a wretched life. Since the age of 15 Wright has spent most of his days in custody and wreaked mayhem when free. In sentencing Wright to death for home invasion, robbery, sexual assault, and murder, the state judge — acting as trier of fact after Wright waived his right to have his fate determined by a jury — admitted all of the evidence Wright offered in mitigation and observed: “I don’t think any reasonable person who has heard all of the evidence in this case can feel anything but sympathy for the pathetic creature, Patrick Wright”. But the judge added that he would not allow sympathy to affect his sentence.
Wright entered the Specht home intending to steal what he could and kill anyone who got in his way. He slit Connie Specht’s throat, in her mother Carol’s presence, after trying but failing to rape both the girl and the woman; he stabbed *948Carol to death. Connie survived and called the police. Wright confessed; that plus physical evidence and Connie’s account leave no doubt of his guilt. The judge found that aggravating circumstances far outweighed any mitigating factors (particularly the judge’s conclusion that Wright acted “under the influence of [a shoe fetish, which amounted to] extreme mental or emotional disturbance”). My colleagues hold that by declining to treat Wright’s upbringing as an additional factor in mitigation, independent of its effect on his mental state, the judge violated the Constitution. I do not agree with this conclusion. There is a big difference between a legal rule forbidding a judge to give mitigating weight to some factor and a legal rule compelling a judge to give mitigating weight to that factor. Eddings v. Oklahoma, 455 U.S. 104, 102 S.Ct. 869, 71 L.Ed.2d 1 (1982), and Lockett v. Ohio, 438 U.S. 586, 98 S.Ct. 2954, 57 L.Ed.2d 973 (1978), condemn the former but do not require the latter. They hold that “the sentencer in capital cases must be permitted to consider any relevant mitigating factor” (Eddings, 455 U.S. at 112, 102 S.Ct. 869); they do not hold that the sen-tencer must deem any particular factor to be mitigating. Wright’s sentence therefore is lawful, and I dissent from the majority’s contrary decision — though I join its conclusion that the judgment of conviction is valid.
My colleagues start with the fact that the judge who sentenced Wright said, after summarizing the evidence (and making the comments I have quoted);
[A]ny matters dealing with sympathy, outrage, who the victim was, all the matters I just mentioned have no bearing on whether the defendant shall receive the death penalty. And again, I note for the record that I have cited them so the record is clear that I have rejected them, and I have disregarded them in making my decision.
Then they compare this with the “strikingly similar” statement held in Eddings to demonstrate a violation of the Constitution (supra, at 943):
Nor can the Court in following the law, in my opinion, consider the fact of this young man’s violent background.
The judge in Wright’s case “disregarded” his childhood; the judge in Eddings’s case declined to “consider” it; these come to the same thing, my colleagues hold. But they are the same thing only if they reflect the same understanding of the judge’s discretion — which they do not. The judge in Eddings thought that the law forbade him to give weight to the defendant’s troubled childhood. This is the import of “[n]or can the Court in following the law” give weight to that factor (455 U.S. at 109, 102 S.Ct. 869, emphasis in original). As the Justices remarked: “From this statement it is clear that the trial judge did not evaluate the evidence in mitigation and find it wanting as a matter of fact; rather he found that as a matter of law he was unable even to consider the evidence.” Id. at 113, 102 S.Ct. 869. Eddings added that the state’s “Court of Criminal Appeals took the same approach. It found that the evidence in mitigation was not relevant because it did not tend to provide a legal excuse from criminal responsibility.” Ibid. That view conflicted with the constitutional principle that a “sentencer [may] not be precluded from considering, as a mitigating factor, any aspect of the defendant’s character or record ... that the defendant proffers as a basis for a sentence less than death.” Lockett, 438 U.S. at 604, 98 S.Ct. 2954 (plurality opinion; emphasis in original), quoted with approval in Eddings, 455 U.S. at 110, 102 S.Ct. 869.
Lockett and Eddings hold that whoever wields the power to impose capital punishment also must have unfettered discretion *949to dispense mercy. To say that there is discretion is to say that the sentencer may be unmoved — even may treat a wretched childhood as a factor in aggravation because it implies that the defendant is inured to violence and thus more likely to be incorrigible. Nothing in Eddings (or any other decision) compels the sentencer to give favorable weight to a circumstance proffered in mitigation. Burger v. Kemp, 483 U.S. 776, 107 S.Ct. 3114, 97 L.Ed.2d 638 (1987), illustrates this point. Burger accused his lawyer of ineffective assistance for failing to present evidence of his unhappy childhood and abusive family. The Court held that this omission did not violate the Constitution in light of the risk that the sentencer would deem the background damning: evidence showing that a person had a rough past and fell in with a bad crowd often implies future dangerousness. Id. at 793-94, 107 S.Ct. 3114.
We have elaborated on this in a series of decisions that, like Burger, the majority disregards. Stewart v. Gramley, 74 F.3d 132 (7th Cir.1996), is the most thorough and worth an extended quotation:
We are mindful of [the argument] that anything which serves to amplify the personal history of the defendant and by doing so furnishes clues to the causality of the crime for which he has been sentenced to death makes such a sentence less likely to be imposed. Causality is mitigation, the lawyer argued. Tout comprendre c’est tout pardonner. It is not an absurd argument. It exploits the tension between belief in determinism and belief in free will. If the defendant’s crime can be seen as the effect of a chain of causes for which the defendant cannot be thought responsible — his genes, his upbringing, his character as shaped by both, accidents of circumstance, and so forth — then a judge or jury is less likely to think it appropriate that he should receive a punishment designed to express society’s condemnation of an evil person. We consider a rattlesnake dangerous but not evil. Maybe if we learned enough about Walter Stewart we would consider him a person who had no more control over his actions than a rattlesnake has over its actions.
But not everyone who believes that human actions are as much the consequences of causal factors as anything else in nature denies that man is “free” to refrain from “voluntary” acts such as murder and is therefore blameworthy if he does not.- “Compatibilists” since Hume have argued that human action is both caused and, for purposes of ascribing moral and legal responsibility, free. We cannot resolve a philosophical debate. And need not. It is enough for our purposes that capital punishment is not premised on — indeed is inconsistent with — the view that the only reason we think that people act “voluntarily” is that we have not studied the antecedents of their actions carefully, and that the purpose of a death-penalty hearing is to investigate the defendant’s history in sufficient depth to dispel the illusion that he was free not to commit the crimes for which he is being condemned. For then the sentence of death would be the proof that the lawyers had not done their job. And since it obviously is not the theory of capital punishment that murderers are compelled to murder by their past and therefore should not be punished, it cannot be right that anything brought out at a death-penalty hearing is certain or even likely to help the defendant to save his life. What is brought out that will help him is what goes to show that he is not as “bad” a person as one might have thought from the evidence in the guilt phase of the proceeding. What is brought out that will hurt him is what goes to show that he is, indeed, as bad a person, or worse, *950than one might have thought from just the evidence concerning the crime.
At the sentencing hearing members of Stewart’s family and his former supervisor at work testified that he had been abandoned by his mother when he was an infant (he does not know who his father was) and had been brought up by his grandparents in Michigan. His grandmother died when he was 15. He had moved to Chicago with his grandfather a few years earlier. He was friendly, worked some and got along with his coworkers, and attended church regularly. Ten years after moving to Chicago he held up the jewelry store. Shortly before that his grandfather had died and Stewart had been distraught. In the closing argument at the sentencing hearing his lawyer did not mention Stewart’s personal history although he had brought it out at the hearing through the testimony that we have just summarized.
The fruits of the investigation by Stewart’s current lawyers are summarized in a report by a “mitigation specialist” backed up by affidavits. According to the report Stewart was born at home without the assistance of a doctor and fell on the floor on his head when he emerged from his mother’s body. His mother was 19 years old and gave Stewart to her parents when he was six months old. She does not know the name of his father. She eventually moved in with her parents, so that she was living in the same house with Stewart, but she did not treat him as her son and displayed no affection for him. When Stewart was 12 his grandfather, whom he doted on, had a mental breakdown. Stewart was a hyperactive child, did badly in school, and dropped out after the ninth grade. Coming from rural Michigan Stewart was not used to the guntoting ways of his Chicago age-mates and imitated their behavior, which included carrying and using guns. At age 13 he fell under the influence of his cousin Geraldine, “a hustler who got her money through swindling people.” Fa-gan-like, Geraldine used Stewart “as her apprentice and bodyguard.” The affidavit of one of Stewart’s sisters states that “Geraldine is the rotten egg of the family.”
Stewart “was enthralled by Geraldine’s lifestyle,” and “turned increasingly to crime, primarily robberies.” He began using drugs, and for a time was a heroin addict. Once, while under the influence of heroin (which, the “mitigation specialist” pointed out, can injure the brain), Stewart “beat Wanzafanother one of his sisters] very badly and the police had to be called.”
Much of this narrative is irrelevant, such as the circumstances of Stewart’s birth, the fact that his grandfather had a mental breakdown, and the heroin addiction. The purpose of these details is to insinuate that Stewart may have brain damage, but of this there is absolutely no evidence.... As for his falling among evil companions, it is obvious that one does not become a teenage robber without being initiated into criminal activity by older hands, not uncommonly a relative. As for drug abuse, the report does not suggest and there is no evidence that Stewart committed the robbery and murders while under the influence of any drug, or, as we have already noted, that he has any organic brain damage. That he comes from a broken home and was brought up in a poor, rough neighborhood of Chicago was known at the sentencing hearing.... The report of the mitigation specialist adds colorful details possibly relevant under a tout comprendre defense if one existed, but not calculated to make a judge or jury think Stewart less deserving of the death penalty.
*951The mitigation specialist’s reference to Stewart’s having embarked on a criminal career consisting “primarily of robberies” after he met cousin Geraldine when he was 13 is highly relevant to appraising the likely impact of current counsel’s thorough investigation on a sentencing judge. The criminal record that was put before the judge understated, we now know as a result of the investigation, Stewart’s actual criminal history. Evidently the nine crimes of which he had been convicted (or sentenced to supervision) were only the tip of the iceberg. He had been a career criminal from the age of 13. This was, it appears, when he began to “use guns.” The implication is that when he started on the robbery trail he used guns in his robberies, and recall that one of his convictions was for armed robbery — committed when he was only 14. The murders in the jewelry store were the predictable culmination of his criminal career. He was by then a hardened, dangerous, and seemingly inveterate criminal.
74 F.3d at 136-37. Recently we have made the same point succinctly:
All that Britz’s meticulously researched personal history shows, as in the other cases we have cited, is that he led a disordered life culminating in the murder for which he has been sentenced to death. People with his background of antisocial behavior are more likely to commit murders than other people, but this does not make them attractive candidates for lenity; rather, it underscores their dangerousness.
Britz v. Cowan, 192 F.3d 1101, 1104 (7th Cir.1999). A judge who shared this perspective would have admitted the evidence of Wright’s background, expressed compassion for him as a victim of others’ wrongs, given mitigating weight to any mental limitations caused by whatever abuse he had endured, and finally declared the offender’s background otherwise not a reason to reduce his punishment. That is exactly what this sentencing judge did do! So where is the constitutional flaw? Is it that a state judge is forbidden to go straight to the bottom line, without articulating the intermediate steps laid out in Stewart? Or perhaps there is a need for magic words, as the majority implies in stating that a judge must declare something like “I have taken X into account but given it no weight”? Why there is a constitutional difference between “I have rejected X” (which this judge said) and “I have given X no weight,” my colleagues do not explain. Anyway, an obligation to state affirmatively that some circumstance has been “taken into account” cannot be found in Eddings (the majority cites no source at all for this directive, which appears at supra, at 943 n. 9), and as a new rule it may not be applied in a collateral attack. See Teague v. Lane, 489 U.S. 288, 109 S.Ct. 1060, 103 L.Ed.2d 334 (1989). My colleagues must think their approach a logical extension of Eddings, but Teague puts extensions (logical and otherwise) off limits.
Let us return, then, to the rule stated by the Supreme Court in Eddings: that no state may forbid a capital sentencer to consider any ground advanced in mitigation. Because the judge’s explanation was curt, it is impossible to exclude the possibility that he sought to convey the same message as the judge in Eddings. Yet if this is what he meant, his expression was inept; the key phrase from Eddings (“[n]or can the Court in following the law ... consider” certain evidence) is missing and not replaced by an equivalent. One would expect a judge to be able to announce a conclusion that the law (as he understood it) disabled him from taking some factor into account. Yet Illinois law was (and is) clear that a troubled youth may be considered in mitigation, see Peo*952ple v. Lewis, 88 Ill.2d 129, 144, 58 Ill.Dec. 895, 430 N.E.2d 1346, 1352 (1981), so it is not surprising that the judge did not mention any legal prohibition. To read this judge’s brief oral statements as foreclosing on legal grounds all possibility that a person’s childhood could mitigate his punishment is to accuse him of failing to know and follow state law — which not only would be inappropriate (a presumption of irregularity?) but also is not the reading given to the judge’s remarks by the Supreme Court of Illinois, which found no violation of the state’s consider-everything norm. People v. Wright, 111 Ill.2d 128, 166-68, 95 Ill.Dec. 787, 490 N.E.2d 640, 656 (1985). (The state’s highest court considered and sustained the conviction and sentence twice more on collateral review. See People v. Wright, 149 Ill.2d 36, 171 Ill.Dec. 424, 594 N.E.2d 276 (1992); People v. Wright, 189 Ill.2d 1, 243 Ill.Dec. 198, 723 N.E.2d 230 (1999). Wright did not renew in either of those proceedings any contest to the interpretation given in 1985 to the sentencing judge’s remarks.)
The most one can say is that the statements were ambiguous — as oral expressions so often are — because they did not explicitly draw a distinction between legal rules and case-specific application. Did the judge mean that Wright’s childhood may not be considered (a legal error) or only that Wright’s childhood will not be considered (a factual conclusion that the evidence did not justify mercy)? The meaning of an ambiguous pronouncement is a question of fact, not of law. Parker v. Dugger, 498 U.S. 308, 320, 111 S.Ct. 731, 112 L.Ed.2d 812 (1991); Wainwright v. Goode, 464 U.S. 78, 83-85, 104 S.Ct. 378, 78 L.Ed.2d 187 (1983); Rivera v. Sheriff of Cook County, 162 F.3d 486 (7th Cir.1998). Because meaning is a fact, the state appellate court’s resolution controls unless not fairly supported by the record. This principle — a holding of Parker, Goode, and Rivera alike' — governs pre-AEDPA cases under the old 28 U.S.C. (1994 ed.) § 2254(d)(8) and newer cases under the current § 2254(e)(1). The majority’s opinion seems to rest on the view that, because the state judge’s words are in black and white on paper, we can decide their meaning as readily as the state’s highest court. The view that findings based on documents may be freely reviewed by appellate tribunals was ringingly rejected in Anderson v. Bessemer City, 470 U.S. 564, 573-75, 105 S.Ct. 1504, 84 L.Ed.2d 518 (1985), with respect to the relations between trial and appellate judges in a unitary judicial system; it is even less persuasive when applied to relations between the state judicial system and the federal.
Reading the sentencing judge’s statement as a finding that he was not swayed by Wright’s background, as opposed to a declaration that he was legally forbidden to consider it, is fairly supported by the record; it is, indeed, the most natural reading. The judge did not imply that he thought himself blocked by a rule of law from dispensing mercy that he deemed warranted. Unless the sentencing judge thought the law an obstacle to acting on mitigating evidence, there is no problem under Eddings. Because the judge did not declare that the law stopped him from reducing Wright’s sentence, the Supreme Court of Illinois’ resolution of the ambiguity is supported by the record, and thus conclusive in this collateral attack. My colleagues say in response that the court’s “holding [does] not withstand scrutiny” (supra, at 942). To call the Supreme Court of Illinois’ conclusion a “holding” is to call it a proposition of law, which in pre-aedpa cases we review without deference. It is not a “holding,” however; it is a finding of fact. The state judge gave thought to (and believed himself legally entitled to reduce a sentence because of) Wright’s childhood, or he did not; the Supreme Court of Illinois concluded that *953he had done so (this is the point of its statement that the judge “considered all potential mitigating facts in the record”, 111 Ill.2d at 168, 95 Ill.Dec. 787, 490 N.E.2d at 656).
Characterizations of what happened are facts, even when they are entangled with legal issues and determine the outcome of the litigation. See, e.g., Pullman-Standard v. Swint, 456 U.S. 273, 288, 102 S.Ct. 1781, 72 L.Ed.2d 66 (1982) (whether particular conduct is “discrimination” is a question of fact); Icicle Seafoods, Inc. v. Worthington, 475 U.S. 709, 106 S.Ct. 1527, 89 L.Ed.2d 739 (1986) (whether a worker is a “seaman” is a question of fact). See also Cooter & Gell v. Hartmarx Corp., 496 U.S. 384, 399-405, 110 S.Ct. 2447, 110 L.Ed.2d 359 (1990) (characterizations under fact-dependent legal standards, such as whether a suit is frivolous, are treated as facts and thus subject to deferential appellate review). What the state judge thought about the extent of his discretion is a proposition about the state of the world, not about the state of the law. That’s the point of Baron Bramwell’s famous observation that “[t]he state of a man’s mind is just as much a fact as the state of his digestion.” Edgington v. Fitzmaurice, 29 Ch. Div. 459, 483 (1885). The Supreme Court of Illinois told us what we need to know about the state of the sentencing judge’s mind. See also Banks v. Hanks, 41 F.3d 1187 (7th Cir.1994).
Wright tries to get mileage from the context of the sentencing judge’s declaration: “any matters dealing with sympathy, outrage, who the victim was, all the matters I just mentioned have no bearing on whether the defendant shall receive the death penalty.” To lump “sympathy” with “outrage” is to imply that the same rule applies to both, and as “outrage” should not be considered, the sentencing judge must have thought that “sympathy” should not be considered either. That is another possible reading — one my colleagues embrace, supra, at 944 — but not one that the state courts were compelled to adopt. The statement “X, Y, and Z have no bearing on whether Wright shall be sentenced to death” could mean “X, Y, and Z have no bearing on whether Wright shall be sentenced to death, for one legal reason applicable to all three grounds.” Or it could mean “X, Y, and Z have no bearing on whether Wright shall be sentenced to death, because none of them makes any difference to me as a matter of fact.” The Supreme Court of Illinois could complete the sentence either way — and it chose the latter way, with considerable logical support, for reasons I have covered. My colleagues demand “constitutionally grounded certainty that Wright [received] the individualized sentence ... mandated by Ed-dings” (supra, at 946, emphasis added), but no rule of law calls for “certainty”. Such a demand contradicts Parker and Goode (a state court’s findings of fact matter only when there is im-certainty!) and transgresses Teague as well (the majority does not cite any pre-1985 case, indeed any case at all, for its insistence on “certainty”). The idea that a record’s silence on some important matter, and consequent lack of “certainty,” requires a federal court to annul a state court’s judgment finds no support in the Supreme Court’s cases, and many hold the contrary — most recently, Mickens v. Taylor, — U.S. -, 122 S.Ct. 1237, 152 L.Ed.2d 291 (2002) (silence about the effect of counsel’s conflict of interest, as a result of judge’s failure to inquire, does not lead to relief even in a capital case).
Giving an ambiguous statement a meaning that renders it unconstitutional (and a violation of state law to boot) disrespects the state judge’s legal skills and denigrates the role of the state judiciary in resolving disputed issues of fact. My colleagues treat a state judge imposing sentence as if he were an administrative law judge in a *954Social Security disability case, obliged to use prescribed verbal formulas and required to tie up all loose ends on pain of remand, and they disdain the decision of the state’s highest court. That is not the right relation between the state and federal judiciaries. A state judge is not an alj, and it is a state’s appellate judiciary rather than a federal court on collateral review that untangles ambiguities attending oral expositions. Nor is an insistence that only “certainty” suffices the right relation between a court of appeals and the Supreme Court of the United States.
Finality and expedition are important in every criminal case. The meaning of this statement was resolved in 1985. Words offer only a cloudy window into the sentencing judge’s thoughts, so reasonable people could disagree about their import; the Justices of the Supreme Court of Illinois did disagree in 1985, dividing five to two about this subject. Seventeen years later two federal appellate judges sign on with the dissenters. That delay does more to show the wisdom of the rule treating these matters as factual (which allows the state’s decision to stand) than to justify throwing the state-court decision out the window and treating the issue as res nova. If matters had been resolved differently in 1985, all doubt could have been resolved, likely just by asking the sentencing judge what he had meant by his statements. Now it is too late — and delay is a powerful reason to treat this subject as one of fact, in order to bring disputes such as this to timely closure.
No resentencing 20 years after the fact can be as good as the one now upset: human memory fades, so the accuracy of sentencing degrades. New errors are bound to occur, mistakes that likely will be more substantial than the ambiguity that concerns my colleagues. Yet no substantial federal question compels a new proceeding. Lockett and Eddings settle the major legal question; only implementation remains. Nonetheless, by holding that both the passage of time and the findings of the state’s highest court are irrelevant, my colleagues create legal problems concerning federalism, the vitality of the rule stated by Parker and Goode, the novel demand for “certainty” in adjudication, and the application of Teague to the majority’s innovations.
To treat an ambiguity in two off-the-cuff sentences as spoiling a sentence reviewed and sustained three times by a state’s highest court — while giving the findings of that court no respect — is to abuse the power conferred by 28 U.S.C. § 2254. Wright is now 59 years old; he will die of old age before we permit the State of Illinois to carry out its judgment.