Leroy Thomas, Jr., was convicted in an Indiana state court in 1979 of murder, and was sentenced to prison for thirty years. After exhausting his state remedies in Thomas v. State, 420 N.E.2d 1216 (Ind.1981), he brought this action for federal habeas corpus, and prevailed in the district court, which ordered the state to either release or retry him.

That Thomas committed the homicide has never been questioned. His only defense was insanity, which he had to prove by a preponderance of the evidence, Ind.Code Sec. 35-41-4-1, and of which there was, indeed, evidence aplenty. He had been diagnosed as a paranoid schizophrenic in 1971 and had been hospitalized for this condition in 1974, 1976, and 1977; the homicide was in 1978. His family testified to many episodes of bizarre behavior on his part. Once, after complaining to his mother that he had been unable to hire a lawyer to sue his family, he shot her three times. Neither party put on a psychiatric witness but the court appointed two psychiatrists who between them had conducted more than 800 court-directed sanity examinations. They testified that Thomas had indeed been insane at the time of the homicide. One testified that Thomas would have behaved the way he did even if a police officer had been present. The other testified that Thomas had been "legally insane" at the time. Cross-examination of the psychiatrists brought out two points. First, one of the psychiatrists had not actually interviewed Thomas (because Thomas had refused to talk to him), but had based his diagnosis on observation of Thomas, discussion with his family, and study of his medical records. Second, the other psychiatrist had not inquired into Thomas's behavior the day before and the day after the homicide, even though the psychiatrist conceded the relevance of Thomas's behavior at those times to a diagnosis of his mental condition when he committed the homicide.

Besides cross-examining the psychiatrists, the prosecutor put in evidence about Thomas's manner and statements at the time of his arrest, which had taken place outside his apartment shortly after the killing. Thomas had refused to come out of the apartment until the press arrived, had asked for a lawyer, and had not seemed crazy. All this was before he was given his Miranda warnings, and raises no problems. But the police were also allowed to testify that Thomas said nothing after being given the warnings; this evidence was allowed in "for the purpose of showing his mental attitude to demonstrate that he was reserved and quiet and not in a manic state." Thomas v. State, supra, 420 N.E.2d at 1220. The admissibility of this testimony is the issue presented by the petition for habeas corpus.

Doyle v. Ohio, 426 U.S. 610, 96 S. Ct. 2240, 49 L. Ed. 2d 91 (1976), holds that it is a denial of due process to use a defendant's refusal to talk or request for an attorney, after being given the Miranda warnings, in evidence against him, even if just to impeach his testimony. Wainwright v. Greenfield, 474 U.S. 284, 106 S. Ct. 634, 88 L. Ed. 2d 623 (1986), adds--even just to rebut his defense of insanity. How mere silence--all we have in this case (there was more in Greenfield, as we shall see)--can be evidence of sanity puzzles us. Insane people frequently are withdrawn and uncommunicative; raving is not the only style of maniacs. A manic depressive has manic phases; a paranoid schizophrenic need not, and need not be the less dangerous for not having them. So far as we know, Thomas was quiet and withdrawn after he shot his mother, yet there is no suggestion that that was the deed of a sane man. But Thomas's silence after being given the Miranda warnings, argued to show rationality, was, along with his coherent requests at the time of his arrest and his outwardly calm appearance, the only evidence that the government introduced to prove his sanity. All the psychiatric evidence indicated insanity, though how strongly was a question for the jury. So we cannot dismiss as harmless the error in admitting the evidence of his silence.

At oral argument we raised the question whether Greenfield, which was decided after Thomas was tried, announced a new rule within the meaning of Teague v. Lane, 489 U.S. 288, 109 S. Ct. 1060, 103 L. Ed. 2d 334 (1989), and of the cases decided by the Supreme Court this past term following and refining that decision. Butler v. McKellar, --- U.S. ----, 110 S. Ct. 1212, 108 L. Ed. 2d 347 (1990); Saffle v. Parks, --- U.S. ----, 110 S. Ct. 1257, 108 L. Ed. 2d 415 (1990); Sawyer v. Smith, --- U.S. ----, 110 S. Ct. 2822, 111 L. Ed. 2d 193 (1990). These decisions hold that, with exceptions unnecessary to discuss in this case, a new rule is not to be applied retroactively in a habeas corpus proceeding--in other words a rule may not be invoked by a defendant whose conviction became final before the rule was declared. By so holding, these decisions may have overruled our decision in Sulie v. Duckworth, 864 F.2d 1348 (7th Cir.1988), which held that Greenfield was to be applied retroactively.

Before getting deeper into this thicket, we remark the state's waiver of the issue of retroactivity. Not only did the state fail to raise the issue in the only brief it filed with us (its opening brief, as appellant), or to cite Teague, although Teague had been decided before the brief was filed; the state expressly conceded, on the authority of Sulie v. Duckworth, the retroactive applicability of Greenfield. It has made no effort to withdraw its waiver by a subsequent filing, or to draw our attention to the Supreme Court's decisions repeating and amplifying the holding in Teague, although two of those decisions were decided before the argument in this case. Teague and the cases following it were first mentioned at the argument--and by the judges, not by the state's counsel. True, when we raised Teague, the state's lawyer embraced it. But to inject a new issue into a case at the oral argument of the appeal ordinarily is much too late, and this regardless of whether the delinquent is a prisoner or is the state, as we held in another habeas corpus case recently, Wilson v. O'Leary, 895 F.2d 378, 384 (7th Cir.1990); see also Egert v. Connecticut General Life Ins. Co., 900 F.2d 1032, 1035 (7th Cir.1990); Gomez v. Greer, 896 F.2d 252, 254 (7th Cir.1990); 7th Cir.R. 28(f) (1990); cf. Alerte v. McGinnis, 898 F.2d 69 (7th Cir.1990). And unlike those cases, here the state expressly waived the point in its opening brief rather than merely failed to raise it, and did not bother to file a reply brief. The question of the retroactive application of Greenfield in light of Teague and the cases following Teague is a difficult one, and by not raising it in a proper and timely manner the state denied its adversary an opportunity to address the question and to argue that we should apply Greenfield after all.

Granted, the rule that a waived issue will be disregarded, like most legal rules, is a generalization rather than a universal truth; it has exceptions. A question of subject-matter jurisdiction cannot be waived until the case is completely over (that is, until all appellate remedies have been exhausted). United States v. Stillwell, 900 F.2d 1104, 1110 n. 2 (7th Cir.1990). And delicate questions of comity can be raised on the court's own initiative, as in Younger v. Harris, 401 U.S. 37, 40, 91 S. Ct. 746, 748, 27 L. Ed. 2d 669 (1971), and Schlesinger v. Councilman, 420 U.S. 738, 743, 95 S. Ct. 1300, 1305, 43 L. Ed. 2d 591 (1975). Of particular pertinence here, failure to exhaust state remedies as a preliminary to asking for federal habeas corpus may be raised by the court on its own initiative even though such exhaustion is not a jurisdictional prerequisite. Granberry v. Greer, 481 U.S. 129, 134, 107 S. Ct. 1671, 1677, 95 L. Ed. 2d 119 (1987); Burgin v. Broglin, 900 F.2d 990, 997 (7th Cir.1990); Del Raine v. Carlson, 826 F.2d 698, 703 (7th Cir.1987). The issue of retroactive application of new constitutional rules in federal habeas corpus cases brought by state prisoners is an issue of the same character. In Teague itself, the state had not argued against retroactive application of the rule for which Teague was contending, leading us to remark that "a court may consider retroactivity questions that the parties did not present." Hanrahan v. Greer, 896 F.2d 241, 245 (7th Cir.1990). In like vein the Supreme Court remarked recently that "although the Teague rule is grounded in important considerations of federal-state relations, we think it is not 'jurisdictional' in the sense that this Court, despite a limited grant of certiorari, must raise and decide the issue sua sponte." Collins v. Youngblood, --- U.S. ----, 110 S. Ct. 2715, 2718, 111 L. Ed. 2d 30 (1990) (emphasis in original). This would seem to scotch earlier suggestions that the rule of Teague is jurisdictional, Hopkinson v. Shillinger, 888 F.2d 1286, 1288 (10th Cir.1989) (en banc); Moore v. Zant, 885 F.2d 1497, 1519 n. 5 (11th Cir.1989) (en banc) (dissenting opinion), but to leave open the possibility that a court can invoke the rule even if the parties do not.

All these examples of nonjurisdictional principles that a court is nevertheless entitled to raise on its own initiative pertain, however, to cases in which a party simply failed to raise an issue--not, as here, to a case in which the party expressly declines to raise it until the judges at argument raise it for him. Whether or not we can refuse to accept an express waiver is unclear. Ohio Bureau of Employment Services v. Hodory, 431 U.S. 471, 480, 97 S. Ct. 1898, 1904, 52 L. Ed. 2d 513 (1977); Sequoia Books, Inc. v. McDonald, 725 F.2d 1091, 1095 (7th Cir.1984); Hickey v. Duffy, 827 F.2d 234, 238 (7th Cir.1987). It is a separate question whether we should even if we can. But we do not have to decide either question here. Teague is inapplicable in any event, even if it is not irrevocably waived. The reason is that, so far as pure silence is concerned (an important qualification, as we shall see), Greenfield did not announce a new rule but merely made explicit what was inescapably implicit in Doyle v. Ohio, decided years before Thomas's trial.

If silence cannot be used even to impeach the defendant's testimony, not because it has no probative significance but because its use breaks the promise implicit in the Miranda warnings, by what logic could it be used to show the defendant's sanity? A request for counsel need not be treated the same way, for reasons explained in our first Sulie decision, Sulie v. Duckworth, 689 F.2d 128 (7th Cir.1982), and by the concurring Justices in Greenfield. 474 U.S. at 296-301, 106 S.Ct. at 641-44. But that is not an issue in this case, which is a case of pure silence. There was, it is true, testimony about Thomas's request for an attorney, but since the request was made before he was given his Miranda warnings this testimony did not violate Greenfield.

If Greenfield 's holding with respect to silence was "dictated" by Doyle (Teague v. Lane, supra, 109 S.Ct. at 1070)--if it would have been "an illogical or even a grudging application" of Doyle "to decide that it did not extend to the facts of" Greenfield (Butler v. McKellar, supra, 110 S.Ct. at 1218)--then Greenfield did not state a new rule. This is the proper characterization of Greenfield with respect to pure silence. It is true that the decision resolved a conflict among the circuits, 474 U.S. at 289 n. 4, 106 S. Ct. at 637 n. 4, a consideration bearing on whether a rule is new that was emphasized in Butler v. McKellar, supra, 110 S.Ct. at 1217. But neither of the lower-court cases that the Supreme Court cited in Greenfield to show that some courts had thought Miranda and Doyle inapplicable to the use of silence to prove sanity was a case of pure silence. Each was in fact a request-for-counsel case--as was Greenfield itself. Sulie v. Duckworth, 689 F.2d 128 (7th Cir.1982); United States v. Trujillo, 578 F.2d 285 (10th Cir.1978). It was common ground among all the Justices in Greenfield and, so far as we are able to determine, among all other federal judges between Doyle and Greenfield, that the use of mere silence (after Miranda warnings had been given) for any purpose violated the defendant's constitutional rights. The conflict that produced a grant of certiorari in Greenfield, and the disagreement among the Justices in Greenfield, had to do with the situation in which the defendant, after being arrested and given his Miranda warnings, asks for a lawyer (Sulie ) or says he will give no statement until he talks to his lawyer (Greenfield and Trujillo ). The panels in Sulie and Trujillo and the concurring Justices in Greenfield thought that the probative value of such a statement on the issue of sanity outweighs the damage that admitting the statement into evidence would do to the credibility of the Miranda warnings. Insofar as Greenfield resolved this disagreement, it may well have stated a new rule--contrary to our second Sulie decision, which preceded Teague. But that we need not decide. All that is important to the present case is that the inadmissibility of evidence of pure silence was established by Doyle; in this limited respect, at least, Greenfield did not establish a new rule.

It is true that in affirming Thomas's conviction the Supreme Court of Indiana distinguished Doyle. But it distinguished it on the ground that Thomas "never took the stand." 420 N.E.2d at 1220. That is just to say that Doyle was about the use of silence to impeach, and Thomas (and Greenfield ) about the use of silence to prove sanity. The Indiana court suggested no principled distinction between the two situations, and we can think of none either.

Since Greenfield is applicable to this case, the only remaining question is whether the admission of the evidence made inadmissible by Greenfield was harmless. The state argues that we should apply an attenuated standard of harmless error. We need not even consider this argument. Even if we applied the most attenuated standard of all, and placed on Thomas the burden of showing that the error in the admission of the evidence of his silence was "prejudicial," in the sense of likely to have affected the outcome of his trial, United States v. Silverstein, 732 F.2d 1338, 1349 (7th Cir.1984), the state would lose this appeal. The evidence of Thomas's insanity was very strong, and the refusal of the jury to acquit him on this ground is more likely to have reflected the traditional hostility of juries to the insanity defense than a failure by Thomas to carry his burden of persuading the jury by a preponderance of the evidence that he was insane. Thomas's silence was one of the very few pieces of evidence of sanity that the state had. As we said earlier, we have difficulty understanding how silence can be evidence of sanity unless the defendant claims to have been not just a maniac, but a raving maniac. The jury, however, may have been looking for a peg to hang its guilty verdict on, and may have fastened on what the prosecutor claimed, however implausibly, was evidence of sanity. Weak as that evidence was, it may well have been decisive in a case as close as this one on the issue, itself essential to the determination of guilt, of the defendant's insanity at the time of the crime.

It is with no relish that we uphold an order for a new trial of a dangerous man whom the passage of time may have made infeasible to retry. But the inadmissibility of key state's evidence having been settled by a Supreme Court decision three years before Thomas's trial, we have no choice but to affirm the district court's grant of habeas corpus.

At oral argument, Thomas's counsel moved to withdraw from the case because Thomas had threatened his life. The motion to withdraw is GRANTED; the judgment of the district court is AFFIRMED.