In this appeal we review the district court's, 645 F. Supp. 664, grant of a petition for writ of habeas corpus under 28 U.S.C. Sec. 2254.

Otto Graham was convicted in 1981 in the district court of El Paso County, Colorado, of nine counts of aggravated robbery, three counts of first degree sexual assault, one count of third degree sexual assault, one count of aggravated motor vehicle theft, and three counts of a crime of violence, all in connection with rapes and robberies at three fast food restaurants (two Taco Bells and a Der Wienerschnitzel) in Colorado Springs, Colorado. In a motion for a new trial, Graham alleged that newly discovered evidence established that the prosecution had knowingly introduced the perjured testimony of a key government witness, Robert Reddick. Specifically, Graham claimed that Reddick perjured himself during the trial when he denied that his testimony against Graham was part of an agreement with the police to dismiss charges against him; Graham further claims that the prosecution was aware that this testimony was perjured.

In denying the new trial motion the state trial court found, inter alia, that "[t]he evidence against Mr. Graham was overwhelming even without the testimony of Reddick." III R. 49. The Colorado Court of Appeals affirmed, holding that the evidence of perjury either was not newly discovered or could have been discovered with diligence at the time of trial. I R., Doc. 2 at 3. The Colorado Supreme Court granted certiorari and likewise affirmed, finding that the "prosecution's allowing testimony about the scope of the agreement with Reddick, if a misrepresentation, was harmless beyond a reasonable doubt." Graham v. People, 705 P.2d 505, 509 (Colo.1985).

While serving his sentence at a Colorado state correctional facility, Graham filed a petition for a writ of habeas corpus in the United States District Court for the District of Colorado. Pursuant to District of Colorado Rule 605, the petition was referred to a magistrate for consideration. The magistrate recommended dismissal, agreeing with the Colorado Supreme Court that allowing Reddick's testimony was harmless beyond a reasonable doubt. I R., Doc. 5 at 7. The district court rejected this recommendation and granted Graham's petition. After a lengthy discussion of the evidence, the district court found that Reddick's testimony was (1) false; (2) knowingly used by the government; and (3) material in the sense that without it a "reasonable doubt would exist as to whether or not Graham is guilty." Id., Doc. 7 at 14-15.

On appeal the State of Colorado asserts that the federal district court failed to give proper weight to state court findings of fact and improperly applied the harmless error standard.I

We first review whether the district court failed to give proper weight to state court findings of fact. In considering a petition for habeas corpus, a federal court must accord a presumption of correctness to findings of facts made by a state court unless the findings are covered by one of the eight exceptions codified in 28 U.S.C. Sec. 2254(d)(1)-(8). In this case, none of the exceptions apply; we must therefore presume that the Colorado state courts' factual findings were correct.

The government contends that the federal district court improperly failed to apply this presumption of correctness. It supports this contention by alleging a conflict between the state trial court's finding that Reddick may have been confused about whether his testimony against Graham was part of a deal with the prosecution and the federal district court's finding that Reddick's testimony against Graham was part of a deal. After careful review, we find no such conflict.

In analyzing the "presumption of correctness" question, we discern three factual issues: (1) whether Reddick and the government made a deal before trial; (2) whether Reddick believed he was testifying pursuant to such a deal; and (3) whether the prosecution was aware of this deal when Reddick testified that there was none.

The state trial court discussed the first two of these issues together. It found that a deal had been struck, but that Reddick was confused at trial whether his testimony was a part of this deal:

"Given the testimony at trial of Detective Eldridge, it is clear that he told Reddick that part of his deal was to testify concerning Mr. Graham's involvement in the crimes of which he stands convicted. Reddick clearly testified that his testimony was not the result of any such deal. Given the obvious confusion within the District Attorney's Office, the Court is not at all sure that Reddick understood that he would actually have to testify against Mr. Graham to have his charges dismissed."

III R. 48-49. The state trial court's finding that a deal in fact had been struck is clarified later in the paragraph by the court's conclusion that the prosecution "should have known about it [the deal] and corrected the statements made by Mr. Reddick." See also id. (the prosecuting attorney "should have known that [testimony was part of the deal] and, therefore, the other members of the office should have known the same"). We interpret the court's statement in the same paragraph, that "it is not clear that Reddick was testifying under penalty of having his charges which had been dismissed nine months earlier reinstated," id. at 49, to mean that it was not clear to the court that Reddick believed he was testifying pursuant to a deal.

The Supreme Court of Colorado similarly interpreted the trial court opinion. That court stated that the trial court "found that Reddick's agreement required Reddick to testify against Graham, but concluded that confusion in the district attorney's office may have led Reddick himself to be uncertain whether he had to testify." Graham, 705 P.2d at 507. These findings, held the supreme court, were "supported by the record." Id. at 509-10; see also id. at 506. It is true that the Colorado Supreme Court, in basing its holding on harmless error, referred to the deal conditionally, see, e.g., id. at 509 ("We conclude that the prosecution's allowing testimony about the scope of the agreement with Reddick, if a misrepresentation, was harmless beyond a reasonable doubt under the facts of this case") (emphasis added). The supreme court, however, did not explicitly reject the trial court finding of a deal, and generally upheld it. We thus conclude that there was no essential conflict between the state and federal court findings concerning the first factual issue--that Reddick had entered into a deal to testify against Graham.

A conflict does exist between the state and federal findings about the second factual issue--whether Reddick was confused. The federal district court appears to have gone beyond the state courts to find that Reddick knew there was a deal at the time of his testimony, see I R., Doc. 7 at 11 ("For Reddick to lead the jury to believe there was no plea arrangement involving the exchange of immunity for his testimony against Graham (and others) was an outright lie"); id. at 15 ("Reddick perjured himself").

This conflict, however, is not relevant to a determination of error. In our habeas corpus consideration of the introduction of false or mistaken testimony, the question of error turns not on the witness' knowledge of falsity, but on the government's knowledge. Napue v. Illinois, 360 U.S. 264, 269, 79 S. Ct. 1173, 1177, 3 L. Ed. 2d 1217 (1959) ("a conviction obtained through use of false evidence, known to be such by representatives of the States, must fall under the Fourteenth Amendment"); Graham v. People, 705 P.2d at 507 ("the prosecution's knowing or reckless use of false or mistaken testimony warrants reversal of the defendant's conviction").

On the third factual issue, government knowledge, there is no conflict. Both state and federal courts imputed knowledge of the agreements to the prosecuting attorneys "because they had possession of the same police report that informed the defendant of the agreement and other members of the district attorney's office were aware of the agreement." Graham, 705 P.2d at 508; see I R., Doc. 7 at 12; III R. 49. Because there was no material conflict between the district court's and the state courts' findings as to this issue, we hold that the federal district court afforded proper weight to the state courts' relevant factual findings. We thus affirm the district court's conclusion that the prosecution's use of false testimony was error.

Having found prosecutorial error, we must decide if it was harmless. Determination of harmless error is a mixed question of law and fact. Although the effect of a particular error on the jury verdict has a factual dimension, the inquiry also requires the court to weigh the sufficiency of the evidence--whether the evidence properly admitted is such that no reasonable possibility exists that the error contributed to the conviction. See Chapman v. California, 386 U.S. 18, 23, 87 S. Ct. 824, 827, 17 L. Ed. 2d 705 (1967). In Miller v. Fenton, 474 U.S. 104, 106 S. Ct. 445, 88 L. Ed. 2d 405 (1985), the Supreme Court addressed an analogous fact/law situation, deciding that the voluntariness of a confession was not a question of fact entitled to the Sec. 2254(d) presumption. In reaching this result the Court reasoned:

"[T]he decision to label an issue a 'question of law,' a 'question of fact,' or a 'mixed question of law and fact' is sometimes as much a matter of allocation as it is of analysis. At least in those instances in which Congress has not spoken and in which the issue falls somewhere between a pristine legal standard and a simple historical fact, the fact/law distinction at times has turned on a determination that, as a matter of the sound administration of justice, one judicial actor is better positioned than another to decide the issue in question."

Id., 474 U.S. at 113-114, 106 S. Ct. at 451-452 (citation omitted).

As in Miller, we believe "the state-court judge is not in an appreciably better position than the federal habeas court" to determine whether an error is harmless. Id. at 117, 106 S.Ct. at 453. Because a determination of harmlessness results in an unremedied constitutional violation, see Napue, 360 U.S. at 269, 79 S.Ct. at 1177, independent de novo federal review of this issue plays a necessary and "important parallel role in protecting the rights at stake." Id.; see also 2 S. Childress & M. Davis, Standards of Review Sec. 13.10 at 216 (1986) ("[M]ost courts apply independent review to a harmlessness holding by a court below without discussion...."). Thus this court must review de novo, as did the district court, whether the error was harmless as a matter of law.

In Chapman v. California, 386 U.S. 18, 87 S. Ct. 824, 17 L. Ed. 2d 705 (1967), the Supreme Court rejected the argument that all errors of constitutional dimension require automatic reversal of criminal convictions. Id. at 22, 87 S.Ct. at 827. Although the Supreme Court has recognized that "some errors necessarily render a trial fundamentally unfair," Rose v. Clark, --- U.S. ----, ---- 106 S. Ct. 3101, 3106, 92 L. Ed. 2d 460 (1986), it has emphasized that these are the exceptions and not the rule:

"Accordingly, if the defendant had counsel and was tried by an impartial adjudicator, there is a strong presumption that any other errors that may have occurred are subject to harmless error analysis. The thrust of the many constitutional rules governing the conduct of criminal trials is to ensure that those trials lead to fair and correct judgments. Where a reviewing court can find that the record developed at trial establishes guilt beyond a reasonable doubt, the interest in fairness has been satisfied and the judgment should be affirmed."

Id. at ----, 106 S.Ct. at 3106-3107. The harmless-error doctrine reflects "the central purpose of a criminal trial ... to decide the factual question of the defendant's guilt or innocence, and ... focus[es] on the underlying fairness of the trial rather than on the virtually inevitable presence of immaterial error." Delaware v. Van Arsdall, 475 U.S. 673, ----, 106 S. Ct. 1431, 1436-1437, 89 L. Ed. 2d 674 (1986) (citation omitted). Thus, "an otherwise valid conviction should not be set aside if the reviewing court may confidently say, on the whole record, that the constitutional error was harmless beyond a reasonable doubt." Id.

In Van Arsdall, a Delaware trial court refused to allow defense counsel to cross-examine a prosecution witness about his agreement to testify in exchange for dismissal of an unrelated criminal charge. Id. at ----, 106 S.Ct. at 1433-1434. In finding the trial court's error to be harmless, the Court reasoned:

"The correct inquiry is whether, assuming that the damaging potential of the cross-examination were fully realized, a reviewing court might nonetheless say that the error was harmless beyond a reasonable doubt. Whether such an error is harmless in a particular case depends upon a host of factors, all readily accessible to reviewing courts. These factors include the importance of the witness' testimony in the prosecution's case, whether the testimony was cumulative, the presence or absence of evidence corroborating or contradicting the testimony of the witness on material points, the extent of cross-examination otherwise permitted, and, of course, the overall strength of the prosecution's case."

Id. at ----, 106 S.Ct. at 1438. The instant case is similar to Van Arsdall, except that here the prosecution, not the court, hindered disclosure of a witness's deal with the prosecution.

Reviewing the record in light of the factors suggested in Van Arsdall, we hold that the error at issue was harmless. We note initially that the question is not whether the verdict might have been different without any of Reddick's testimony, but whether the verdict might have been different if Reddick's testimony were further impeached by disclosure of the deal. Cf. id. at ----, 106 S.Ct. at 1437-1439. Further impeachment would have had minimal impact, for two reasons. First, a clear disclosure of the deal would have had merely cumulative effect in impeaching Reddick's veracity. The issue of whether a deal existed was before the jury in the form of conflicting testimony by two detectives.1 And Reddick's character had already been severely impeached.2 Thus, we agree with the conclusion of the Colorado courts that the impeachment effect of a clear disclosure would be largely cumulative. Second, the effectiveness of impeachment would have been limited, as suggested in Van Arsdall, id., by the presence of evidence corroborating the testimony on material points.3

The overall strength of the prosecution's case underscores the minimal impact that would have resulted from such further impeachment of Reddick's testimony. Reddick was only one of many witnesses in the trial. Reddick's testimony was devastating, of course, especially his statements that he drove his brother-in-law Graham to the site of the Taco Bells just before the incidents and that he picked Graham up after the Volkswagen Graham stole at the Der Wienerschnitzel broke down, and his identification of the name that appeared on the patch on Graham's jacket. But there were ten eyewitnesses to the various offenses: two at each of the Taco Bells, and six at the Der Wienerschnitzel. Although the perpetrator was masked during part of the Der Wienerschnitzel incident, he took off his mask for an extended period. All of the six eyewitnesses at Der Wienerschnitzel positively identified Graham as the perpetrator at trial; four of those had identified no one else. We consider the case against Graham from that encounter overwhelming and not subject to any reasonable doubt by the impeachment or elimination of Reddick's testimony.

The case against Graham on the Taco Bell incidents is not as strong, partly because the Taco Bell witnesses had previously identified other individuals instead of Graham in pretrial lineups and photo arrays. But the perpetrator of the crimes was masked during the Taco Bell rapes and robberies, except for brief exposures of his face to the rape victims. Further, this identification issue was clearly before the jury and indeed was the principal focus of Graham's defense. The two other suspects most often identified in photographs or line-ups testified at Graham's trial and were required to stand together with Graham before the jury, where they could be compared for physical resemblance. In addition, both eyewitnesses to the first Taco Bell rape and robberies identified Graham's jacket, which had a distinctive patch on the front, as being worn by the perpetrator. The victim of the second Taco Bell rape early and consistently identified the rapist as having a small scar beside his nose similar to Graham's scar. In both Taco Bell incidents the rape victims identified the perpetrator as having unusually long (one-half inch or more) fingernails; a police officer identified Graham as having inch-long, claw-like fingernails when he was arrested less than three months after the first Taco Bell rape. Finally, there were similarities between the Taco Bell incidents and the Der Wienerschnitzel incident, supporting a conclusion that they were all done by the same person. They all occurred at closing, rapes occurred each time, the robber-rapist required all witnesses to remove most or all of their clothing, and he threatened to "blow out the brains" of the victims if they did not comply with his commands.

We have examined the entire record and are satisfied beyond a reasonable doubt that any further impeachment of Reddick's testimony would not have changed the jury's verdicts. We therefore vacate the order of the district court granting Graham's petition for a writ of habeas corpus.

REVERSED.