The Supreme Court remanded this case to us for resolution under the law that preceded the Antiterrorism and Effective Death Penalty Act of 1996. See Lindh v. Murphy, — U.S.-, 117 S.Ct. 2059, 138 L.Ed.2d 481 (1997). Our full court returned this case to the panel for disposition. Because the case has been briefed and argued twice in this court, the first time before the aedpa’s adoption, further briefs and argument are unnecessary. To abbreviate exposition, we assume that the reader is familiar with the Supreme Court’s decision and our prior opinion, 96 F.3d 856 (7th Cir.1996) (en bane). One element of the en banc decision — the conclusion that Lindh is not entitled to supplement the record on a particular issue, id. at 867 — was not disturbed by the Supreme Court and is conclusive on the panel. Other questions we must consider on our own.
At the mental-condition phase of Lindh’s trial, psychiatrist Leigh Roberts testified that Lindh did not have a mental disease when he killed two people and tried to kill a third. The jury reached the same conclusion. The prosecutor elicited background information that would have led a reasonable jury to infer that Roberts was a pillar of the medical community; other evidence, which the jury did not hear, would have supported a conclusion that Roberts sexually abused some of his patients, was about to lose his medical license and his prestigious faculty positions, and stood a chance of going to prison. Lindh could have used the excluded evidence in two ways: first, to show that Roberts had a reason to be biased in the prosecutor’s favor, hoping that helpful testimony would mitigate his criminal punishment even though the Dane County prosecutor (who charged Lindh) could not directly influence the Milwaukee County prosecutor (who was conducting the investigation of Roberts); second, to show that the background information used to add luster to Roberts (and hence weight to his testimony) was misleading. Lindh believes that the restrictions on his cross-examination of Roberts violated his rights under the confrontation clause of the sixth amendment, applied to the states by the fourteenth.
One potential response is that the confrontation right does not apply after the jury concludes that the accused performed the acts that constitute the crime. The en banc court held that Lindh’s argument called for “a nontrivial extension of current law” (96 F.3d at 876) impermissible under 28 U.S.C. § 2254(d)(1), as amended by the aedpa. Wisconsin does not contend that this is the sort of extension that under pre-AEDPA law was barred by Teague v. Lane, 489 U.S. 288, 109 S.Ct. 1060, 103 L.Ed.2d 334 (1989). We must decide the question without regard to the requirement in the aedpa that the petitioner establish that 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”. Proceedings to determine Lindh’s mental state when he pulled the trigger were not strictly “sentencing”; Wisconsin commits this question to the same jury that hears the evidence on whether the defendant committed the acts alleged. Although that jury renders two verdicts, Wisconsin refers to the process as a “continuous trial” of two separate components of the defendant’s plea:
There shall be a separation of the issues with a sequential order of proof in a continuous trial. The plea of not guilty shall be determined first and the plea of not guilty by reason of mental disease or defect shall be determined second.
Wis. Stat. § 971.165(l)(a). Only after the jury has resolved both questions in the prosecutor’s favor does the case move to a traditional sentencing proceeding before a judge. Separation of issues and evidence over time, without assigning the mental-state issues to a judge as part of a detached sentencing hearing, makes Wisconsin’s hybrid procedure tough to classify. Having considered the question independently, the panel is persuaded by the reasoning of Judge Wood’s separate opinion in the en banc decision, 96 F.3d at 881-83, that the mental-state element of the trial is so closely associated with the issue of guilt or innocence that Roberts was a *901“witness against” Lindh, who therefore had a constitutional right to confront and cross-examine Roberts.
On the merits, the confrontation question is close. Applicable precedents, particularly Delaware v. Van Arsdall, 475 U.S. 673, 106 S.Ct. 1431, 89 L.Ed.2d 674 (1986), and Davis v. Alaska, 415 U.S. 308, 94 S.Ct. 1105, 39 L.Ed.2d 347 (1974), say that the defendant is entitled to cross-examine a witness about potential sources of bias, but that the trial judge may place limits on the examination to prevent what the rules of evidence call “the danger of unfair prejudice, confusion of the issues, or misleading the jury”. Fed.R.Evid. 403. Lindh wanted to cross-examine Roberts on a hot-button subject whose nature creates a risk that the jury would be diverted from its tasks by salacious material unrelated to the nature or extent of potential bias. Lindh’s counsel did not ask the judge to permit a generic cross-examination referring to unspecified “accusations of impropriety” and “investigations”; counsel wanted to make it crystal clear that the charges had to do with sex. But the judge and prosecutor did not offer the option of redaction; like Lindh, they treated the subject as an all-or-none matter. With all and none the only options on the table, the trial court exceeded the discretion it possesses under the confrontation clause by giving the answer “none.”
Roberts, a veteran expert witness, likely knew that the Dane County prosecutor could not reward favorable testimony in Lindh’s ease even indirectly, for the disqualified Dane County prosecutor could not ethically seek to influence the Milwaukee County prosecutor by making a recommendation. Lindh does not contend that any deal between Roberts and the Milwaukee County prosecutor was in prospect or under negotiation. But Roberts may have believed that testimony helping the prosecution in this case, which achieved notoriety throughout Wisconsin, would aid his cause, if only because it was bound to come to the attention of the judge who presided in the prosecution against him. What makes this case close is that there is no evidence that Roberts changed his analysis of Lindh’s mental state between the first interview (before Roberts was aware that any accusations had been made against him) and the trial. Lindh could have cross-examined Roberts about whether his view changed; if Roberts held the same position throughout, the question of bias does not arise. Or does it? Maybe on reflection Roberts would have come to a conclusion more favorable to Lindh, but refrained from doing so because of his desire to achieve an advantage in the prosecution pending against him. A psychiatrist would recognize the possibility that the testimony was tailored in light of subconscious desires, even if the witness attempted to be objective. Psychiatry has not been reduced to the evaluation of objective signs; Roberts could not point to the results of some test that dictated his opinion. When so much rides on imponderables, an inquiry into potential sources of bias is more important. Still, the difficulty of establishing that Roberts shaded his testimony would make this case a candidate for a conclusion that any error was harmless under Brecht v. Abrahamson, 507 U.S. 619, 113 S.Ct. 1710, 123 L.Ed.2d 353 (1993) — but for the rosy glow in which the prosecutor bathed Roberts.
When going through Roberts’ credentials as an expert, the prosecutor led the jury to understand that it was not hearing from just any psychiatrist. Roberts was a faculty member at the University of Wisconsin and an adjunct faculty member of the San Francisco Theological Seminary. He had been the chairman of conferences about religion and mental health. Three years before trial he had been “honored ... as essentially the mid-west psychiatrist of the year”. Roberts was a devoted family man whose grandchildren were approaching Lindh’s age. The prosecutor offered this evidence knowing that psychiatric testimony cannot be verified or refuted empirically, making the witness’ personal and professional stature important to a jury. Roberts came across as a scientist of impeccable credentials and high moral standards — an unimpeachable man. At the *902time he painted this picture of Roberts, the prosecutor knew that Roberts was unimpeachable only because the trial judge would not allow impeachment. Roberts’ sterling credentials were about to acquire tarnish, but the jury never learned this. Because the jury may have placed great weight on Roberts’ stature — the prosecutor, who invited it to do so, is in no position to say otherwise — it is not appropriate to label the denial of cross-examination harmless error. It may have altered the outcome, for two other psychiatrists opined that Lindh had a mental disability (although one of those two believed that the disability did not prevent Lindh from understanding and controlling his acts). If the jury treated the disagreement as a battle of credentials, which it may have, Lindh’s defense was undercut by restriction of cross-examination.
The judgment of the district court is reversed, and the case is remanded with instructions to issue a writ of habeas corpus requiring Wisconsin to treat Lindh as not criminally responsible for the killings, unless within 120 days it offers him a new trial on the mental-disease-or-defect question.