Petitioner Willis Randolph appeals the district court’s denial of his petition for a writ of habeas corpus challenging his 1986 state court conviction for murder. We hold that if the State places a cooperating informant in a jail cell with a defendant whose right to counsel has attached, and if the informant then makes a successful effort to stimulate a conversation with the defendant about the crime charged, the State thereby violates the defendant’s Sixth Amendment rights under Massiah v. United States, 377 U.S. 201 , 84 S.Ct. 1199 , 12 L.Ed.2d 246 (1964). Because the district court failed to make proper factual findings, we vacate the district court’s denial of Randolph’s Massiah claim and remand for factfinding. We do not decide the part of Randolph’s claim under Brady v. Maryland, 373 U.S. 83 , 83 S.Ct. 1194 , 10 L.Ed.2d 215 (1963), that depends on the district court’s finding of fact necessary for the Massiah claim. We affirm the district court’s denial of Randolph’s other claims.

I. Background

Petitioner Randolph is currently serving a life sentence for his conviction for the murder of 10-year-old Lamont Collins on June 24,1981. The police initially suspected Lamont’s father, who had expressed unhappiness at paying child support and had threatened Lamont’s mother. Police, however, were unable to discover sufficient evidence of the father’s involvement to justify bringing charges against him. Several years later, Randall McKinney was interviewed in connection with another homicide. During the interview, he stated that he had seen petitioner Randolph near the scene of the crime at the time of the killing. McKinney claimed that he had not previously come forward because he believed that Randolph had already been convicted of the killing. Soon after McKinney’s interview with the police, Randolph was arrested and charged with Lamont’s murder.

Randolph is a developmentally disabled African-American with an IQ of 59. The prosecution’s theory of the case was that Lamont’s father had paid Randolph to murder Lamont. Randolph was tried twice. At his first trial, the evidence against him consisted primarily of McKinney’s testimony that he had seen Randolph, whom he had known since childhood, near the scene of the crime looking into an open car trunk, and that Randolph’s car matched the description of a car several witnesses had reported seeing near the murder scene.

Prior to the start of his first trial, at which the State sought the death penalty, Randolph moved to set aside the jury panel based on the fact that there were lower percentages of African-Americans and Hispanics in the venire than their respective percentages in the general population. Randolph also moved to change venue, arguing that the media coverage of the murder and his subsequent arrest so tainted the jury pool as to make a fair trial all but impossible. The trial judge denied both motions. The first trial ended in a mistrial when the jury hung. Subsequent interviews revealed that eight of the jurors had voted to find Randolph guilty, and four had voted to acquit.

The State did not seek the death penalty at Randolph’s second trial. Randolph again moved to set aside the jury panel and to change venue, and the trial judge again denied both motions. On retrial, the prosecutor had the benefit of two additional witnesses, both of whom were jailhouse informants. One informant was Jack Kon-kle, who gave somewhat conflicting and relatively insignificant testimony. The other was Ronnie Moore, who provided crucial testimony against Randolph.

Moore shared a jail cell with Randolph throughout most of Randolph’s first trial and for several weeks after the judge declared a mistrial. Moore came to the attention of prosecutors when he gave them a letter asking for leniency and mentioning that he was Randolph’s cellmate. Moore’s defense attorney, as well as prosecutors, interpreted Moore’s letter as an offer to testify against Randolph. Moore met with Deputy District Attorney James Oppliger and Detective Pete Chavez several times to discuss his possible testimony against Randolph, as well as a plea deal relating to the crime for which Moore was being held. At some point, Moore told Oppliger and Chavez that Randolph had admitted to killing Lamont and had said that he was due to receive a lot of money. Moore also told the prosecution team that Randolph had known Lamont’s father and had spoken highly of him, thus supporting the prosecution’s theory that the father had hired Randolph to kill Lamont.

Prior to the start of the second trial, Randolph moved to exclude the testimony of Konkle and Moore. After hearing the proffered testimony of the two witnesses, the trial judge denied the motion. With the benefit of these additional witnesses, the State obtained a conviction for first degree murder. Randolph was sentenced to a prison term of 27 years to life.

After exhausting his state remedies, Randolph sought habeas corpus relief in federal district court. He argued, inter alia, that the jury venire was not representative of a cross-section of the community; that the trial judge’s failure to change venue prior to the start of the second trial violated his Fourteenth Amendment right to due process; that the use of Moore’s testimony violated his Sixth Amendment right to counsel; that the use of Moore’s and Konkle’s testimony violated due process; and that the prosecutor violated Brady v. Maryland, 373 U.S. 83 , 83 S.Ct. 1194 , 10 L.Ed.2d 215 (1963), by failing to turn over information about Moore and Konkle. After a hearing, the magistrate judge issued findings and recommendations in which he recommended that the district court dismiss Randolph’s petition on the merits. The district court agreed with the recommendation, adopted the magistrate’s findings, and dismissed the petition. Randolph timely appealed.

II. Standard of Review

We review the dismissal of a petition for writ of habeas corpus de novo. Beardslee v. Woodford, 358 F.3d 560, 568 (9th Cir.2004). Since Randolph’s petition was filed prior to the effective date of the Anti-Terrorism and Effective Death Penalty Act (“AEDPA”), we apply pre-AEDPA law. Alcala v. Woodford, 334 F.3d 862, 868 (9th Cir.2003). We may grant relief only if the state court committed a constitutional error that “had substantial and injurious effect or influence in determining the jury’s verdict.” Brecht v. Abrahamson, 507 U.S. 619, 637 , 113 S.Ct. 1710 , 123 L.Ed.2d 353 (1993) (quoting Kotteakos v. United States, 328 U.S. 750, 776 , 66 S.Ct. 1239 , 90 L.Ed. 1557 (1946)). The district court’s finding of facts are reviewed for clear error. Turner v. Calderon, 281 F.3d 851, 864 (9th Cir.2002). Under pre-AEDPA law, “the factual findings of state courts are presumed to be correct, and may be set aside ... only if they are not fairly supported by the record.” Purkett v. Elem, 514 U.S. 765, 769 , 115 S.Ct. 1769 , 131 L.Ed.2d 834 (1995) (internal citation and quotation marks omitted).

III. Discussion

A. Jury Venire

The Sixth Amendment right to a trial by jury includes a right to a jury venire that is “representative[of a] cross section of the community.” Taylor v. Louisiana, 419 U.S. 522, 528 , 95 S.Ct. 692 , 42 L.Ed.2d 690 (1975). To establish a prima facie violation of that right, a defendant must show:

Duren v. Missouri, 439 U.S. 357, 364 , 99 S.Ct. 664 , 58 L.Ed.2d 579 (1979).

Randolph relies, as he did at his second trial, on a study prepared by Dr. John Tinker. Dr. Tinker’s study purported to show that there were relatively fewer African-Americans and Hispanics in the venire in the second trial than in the general population in Fresno County, the geographical area of the venire. African-Americans, Dr. Tinker found, represented 4.3 percent of the population in the county but constituted only 3.1 percent of the venire. Thus, there was a 1.2 percent absolute disparity. See United States v. Sanchez-Lopez, 879 F.2d 541, 547 (9th Cir.1989) (using absolute, rather than comparative, disparity when engaging in analysis under Duren). For Hispanics, however, this disparity was much greater. While Hispanics made up 29.2 percent of the population in Fresno County, they made up only 13.8 percent of the venire, an absolute disparity of 15.4 percent.

Applying the first prong of the Duren test, it is clear that African-Americans and Hispanics are “distinctive groups.” See United States v. Nelson, 137 F.3d 1094, 1101 (9th Cir.1998) (“It is undisputed that Hispanics are a ‘distinctive’ group for purposes of Sixth Amendment analysis.”). It is equally clear that while the disparity in African-Americans is insufficient to satisfy the second prong of the Duren test, the disparity in Hispanics is sufficient. Compare United States v. Suttiswad, 696 F.2d 645, 649 (9th Cir.1982) (holding that a 7.7 percent absolute disparity was insufficient to make out a Sixth Amendment violation) with Ramseur v. Beyer, 983 F.2d 1215, 1232 (3rd Cir.1992) (finding a 14.1 percent absolute disparity was potentially significant for the purposes of Duren’s second prong). The State concedes as much and focuses its argument with respect to Hispanics on the third prong of the Duren test, contending that Randolph has not shown that the disparity is a result of “systematic exclusion ... in the jury selection process.” Duren, 439 U.S. at 364 , 99 S.Ct. 664 . We agree with the State.

Prior to the trial, Fresno County Assistant Jury Commissioner Irma Marez described the manner in which the County assembles the venire. Fresno County compiles a list of individuals who are registered voters in the County or who have obtained a driver’s license or ID card from the California Department of Motor Vehicles. Approximately 50,000 names are randomly drawn from this list each summer, and questionnaires are sent to those individuals. The County uses the returned questionnaires to determine whether the individuals qualify for jury service or whether they should be excused or deferred. The venire is ultimately drawn from this group of qualified jurors.

Approximately 25 percent of people fail to return the first questionnaire. The County sends a second questionnaire to these people, stating that a failure to return the questionnaire could subject them to a summons demanding that they appear personally in court to fill it out. In prac tice, however, no such summons is ever issued. Similarly, if a person who has returned a questionnaire fails to report to jury duty, that person might be telephoned but would not be summoned to court.

Randolph argues that the County’s failure to issue summonses to individuals who fail to return questionnaires is the sort of “systematic exclusion” forbidden by the third prong of the Duren test. Randolph does not argue that the County failed to summon Hispanics while summoning other ethnic groups, and he has not presented evidence showing that Hispanics failed to return questionnaires at a higher rate than the general population. Nevertheless, Randolph argues that because Hispanics make up a significantly smaller percentage of the venire than of the County population as a whole, the County was obligated to issue summonses in order to maintain a representative venire.

Under the test established by Duren , disproportionate exclusion of a distinctive group from the venire need not be intentional to be unconstitutional, but it must be systematic. In Duren itself, women were, unlike men, able to opt out of jury service by filling out a paragraph in the questionnaire sent to them. Further, women who did not return the questionnaire were presumed to have opted out; the same presumption did not apply to men. 439 U.S. at 362 , 99 S.Ct. 664 . The Court concluded that “the resulting disproportionate and consistent exclusion of women from the jury wheel and at the venire stage was quite obviously due to the system by which juries were selected.” Id. at 367 , 99 S.Ct. 664 (emphasis in original).

In United States v. Jackman, 46 F.3d 1240, 1242-43 (2d Cir.1995), an unintentional computer error excluded individuals residing in the two most populous counties from the “qualified wheel” from which juries were drawn. Since a large proportion of racial and ethnic minorities lived in those counties, those groups were dramatically underrepresented in the jury pool. The court concluded that the resulting un-derrepresentation was “inherent in the particular jury-selection process utilized.” Id. at 1248 (citing Duren, 439 U.S. at 366 , 99 S.Ct. 664 ). Further, in Gibson v. Zant, 705 F.2d 1543 (11th Cir.1983), a panel of jury commissioners was directed to “compile and maintain and revise a jury list of intelligent and upright citizens of the county to serve as jurors.” Id. at 1547 . Because the criteria were so subjective, the court concluded that “systematic exclusion” had occurred. Id. at 1549 .

A showing that a jury venire underre-presents an identifiable group is, without more, an insufficient showing of systematic exclusion under the third prong of the Duren test. If underrepresentation by itself were sufficient to support a holding of unconstitutionality, the second and third prong of Duren would effectively collapse into one inquiry. Randolph cannot satisfy Duren’s third prong because he has failed to present any evidence that the underrep-resentation of Hispanics is due to the system Fresno County uses to assemble the venire. While Randolph suggests in his brief that Hispanics return questionnaires at a lower rate than the general population, he has presented no evidence to support this suggestion. As the district court found, Dr. Tinker’s study did not establish whether the relatively smaller percentage of Hispanics in the venire resulted from a lower questionnaire return rate or from some other factor. For that reason, among others, Randolph has not shown that the alternative system he proposes-issuing summonses to all individuals who fail to return the questionnaires-would increase Hispanic representation. Because Randolph has not shown any relationship between the disproportionately low percentage of Hispanics in the venire and the juror-selection system the County uses, we cannot conclude that the underrepresentation of Hispanics is, as Duren requires, “inherent in the particular jury-selection process.” Duren, 439 U.S. at 366 , 99 S.Ct. 664 (emphasis added). Because Randolph has shown nothing more than a simple disparity between the percentage of Hispanics in the venire and in the County, we reject Randolph’s venire claim under Duren .

B. Venue

Petitioner also challenges the state trial court’s failure to change the venue for the second trial sua sponte, arguing that the media coverage of the murder and subsequent arrest of Randolph so tainted the jury pool as to deny him a fair trial. A criminal defendant is entitled to be tried by “a panel of impartial, ‘indifferent’ jurors.” Irvin v. Dowd, 366 U.S. 717, 722 , 81 S.Ct. 1639 , 6 L.Ed.2d 751 (1961). If pretrial publicity is such that it is impossible to seat such a jury, a judge must grant a defendant’s request for a change of venue. See Gallego v. McDaniel, 124 F.3d 1065, 1070 (9th Cir.1997) (citing Harris v. Pulley, 885 F.2d 1354, 1360 (9th Cir.1988)). In order to show a violation’of his constitutional rights, petitioner must show either actual or presumed prejudice. “Prejudice is presumed when the record demonstrates that the community where the trial was held was saturated with prejudicial and inflammatory media publicity about the crime.... To establish actual prejudice, the defendant must demonstrate that the jurors exhibited actual partiality or hostility that could not be laid aside.” Id. (quoting United States v. Sherwood, 98 F.3d 402, 410 (9th Cir.1996)).

Because Randolph does not claim actual prejudice, he must show presumed prejudice. In evaluating a claim of presumed prejudice, we typically consider three factors, including whether there was a “barrage of inflammatory publicity immediately prior to trial amounting to a huge ... wave of public passion,” whether the media accounts were primarily factual or editorial in nature, and whether the media accounts “contained inflammatory, prejudicial information that was not admissible at trial.” Ainsworth v. Calderon, 138 F.3d 787, 795 (9th Cir.1998) (citations and internal quotation marks omitted). Further, “[t]he presumed prejudice principle is rarely applicable, and is reserved for an ‘extreme situation.’ ” Harris v. Pulley, 885 F.2d 1354, 1361 (9th Cir.1988) (internal citations omitted).

The three factors are not satisfied in this case. Most of the relevant articles were published at the time Lamont was murdered, six years before the first trial and eight years before the second trial. This length of time helps mitigate any bias the media coverage might have created. See Harris, 885 F.2d at 1362 (relying on the fact that “[t]he number of news reports regarding [petitioner’s] case had dissipated considerably by the time of jury selection four months later” in concluding there was no presumed prejudice). Further, the majority of the published articles were factual in nature, largely describing Lamont and Lamont’s father. Although additional stories were published after the judge denied Randolph’s motion to change venue prior to the first trial, most of the stories focused on the trial proceedings. The factual nature of these stories similarly supports a conclusion that the trial court did not err in refusing to change the venue. See Jeffries v. Blodgett, 5 F.3d 1180, 1189 (9th Cir.1993) (holding that there was no presumed prejudice when “the nature of the news coverage was factual and not inflammatory”). Finally, while some of the media coverage contained prejudicial information that would not have been admissible at trial, such as Randolph’s previous conviction for an unrelated crime, the Supreme Court held that a court was not required to change venue under circumstances even more unfavorable to the defendant. See Patton v. Yount, 467 U.S. 1025, 1031-35 , 104 S.Ct. 2885 , 81 L.Ed.2d 847 (1984) (venue change not required even though media coverage had described defendant’s prior conviction for the murder at issue, the reversal of that conviction on appeal, the defendant’s insanity plea in the first trial, and the defendant’s confession, which had been introduced in the first but not second trial). The trial court therefore did not violate Randolph’s due process rights by failing to change the venue.

C. Massiah Violations

Randolph further contends that the incriminating statements he made to Moore should be excluded because they were obtained in violation of his Sixth Amendment right to counsel. Once a defendant’s Sixth Amendment right to counsel has attached, the government is forbidden from “deliberately eliciting” incriminating statements from the defendant. Massiah v. United States, 377 U.S. 201, 206 , 84 S.Ct. 1199 , 12 L.Ed.2d 246 (1964). This prohibition has been extended to the use of jailhouse informants who relay incriminating statements from a prisoner to the government. In United States v. Henry, 447 U.S. 264 , 100 S.Ct. 2183 , 65 L.Ed.2d 115 (1980), the government used a paid informant, who was serving time for forgery, to obtain information about Henry. The government admonished him to “be alert to any statements made by federal prisoners [including Henry, with whom the informant was housed in the same cellblock], but not to initiate any conversation with or question Henry regarding the bank robbery [for which he had been previously indicted].” Id. at 266 , 100 S.Ct. 2183 . As a result of conversations the informant had with Henry, the government obtained incriminating evidence, which was used at trial to convict him.

The central question in Henry was “whether under the facts of this case a Government agent ‘deliberately elicited’ incriminating statements from Henry within the meaning of Massiah.” Id. at 270, 100 S.Ct. 2183 . In concluding that a government agent had elicited such statements, the Court relied on three factors. First, the informant was acting under instructions from the government and was paid for his services. Second, the informant “was ostensibly no more than a fellow inmate of Henry,” which caused Henry to trust him and thus be more likely to make incriminating statements. Id. Finally, Henry was in custody and under indictment at the time the informant conversed with him. The Court found it irrelevant that government officials had cautioned the informant not to ask Henry any questions: “Even if the agent’s statement that he did not intend that Nichols would take affirmative steps to secure incriminating information is accepted, he must have known that such propinquity likely would lead to that result.” Id. at 271 . In these circumstances, the Supreme Court held that the government agent’s conversation with Henry amounted to “deliberate elicitation” under Massiah .

In contrast to Henry , the Supreme Court found no Sixth Amendment violation in Kuhlmann v. Wilson, 477 U.S. 436 , 106 S.Ct. 2616 , 91 L.Ed.2d 364 (1986). In that case, the trial court concluded that the statements made by the defendant to the informant were “unsolicited” and “spontaneous.” Id. at 440 , 106 S.Ct. 2616 . As a result, Kuhlmann fit squarely into the type of case explicitly left open in Henry , “where an informant is placed in close proximity but makes no effort to stimulate conversations about the crime charged.” Henry, 447 U.S. at 271 n. 9, 100 S.Ct. 2183 .

Thus, to show that the state violated his Sixth Amendment rights by obtaining and using Moore’s testimony, Randolph must show that Moore was acting as an agent of the State when he obtained the information from Randolph and that Moore made some effort to “stimulate conversations about the crime charged.” See id. Notably, “stimulation” of conversation falls far short of “interrogation.” See Fellers v. United States, 540 U.S. 519 , ----, 124 S.Ct. 1019, 1022-23 , 157 L.Ed.2d 1016 (2004) (finding that “implicit questions” and “discussion” about defendant’s methamphetamine use constituted a Sixth Amendment violation) (internal quotation marks and brackets removed). Any statements, however, made by Randolph before Moore met with the prosecution team cannot be the basis of a Massiah violation. See Maine v. Moulton, 474 U.S. 159, 176 , 106 S.Ct. 477 , 88 L.Ed.2d 481 (1985) (“[T]he Sixth Amendment is not violated whenever — by luck or happenstance — the State obtains incriminating statements from the accused after the right to counsel has attached.”)

1. Acting on Behalf of the State

In order for a Massiah violation to have occurred, Moore must have been acting on behalf of the State. After Moore met with Deputy District Attorney Oppliger and Detective Chavez, he was returned to the cell he shared with Randolph. Unlike in Henry , there was no explicit deal under which Moore was promised compensation in exchange for his testimony. For purposes of our holding, we accept as true the State’s contention that Moore was told not to expect a deal in exchange for his testimony. However, Henry makes clear that it is not the government’s intent or overt acts that are important; rather, it is the “likely ... result” of the government’s acts. Henry, 447 U.S. at 271 , 100 S.Ct. 2183 . It is clear that Moore hoped to receive leniency and that, acting on that hope, he cooperated with the State. Oppliger and Chavez either knew or should have known that Moore hoped that he would be given leniency if he provided useful testimony against Randolph. (Indeed, that is precisely what happened. After providing useful testimony against Randolph, Moore received a sentence of probation instead of a prison term.)

We have not previously considered whether a jailhouse informant can be considered a government agent if there is no express agreement between the informant and the government that the informant will be compensated for his services. In the circumstances of this case, we hold that an explicit agreement to compensate Moore is not necessary to a finding that Moore acted as an agent of the State. There is sufficient undisputed evidence to show that the State made a conscious decision to obtain Moore’s cooperation and that Moore consciously decided to provide that cooperation. That cooperation rendered Moore an agent of the State. We recognize that agreed-upon compensation is often relevant evidence in determining whether an informant is acting as an agent of the State. But it is the relationship between the informant and the State, not the compensation the informant receives, that is the central and determinative issue.

2. Factfinding by the District Court

We thus conclude that Moore was acting as an agent of the State when he was placed in Randolph’s cell after meeting with Deputy District Attorney Oppliger and Detective Chavez. But we are unable to determine whether a Massiah violation occurred because two critical factual issues have not been resolved or properly considered by the district court. The first issue relates to timing: When did Moore meet with the prosecution team, and when, in relation to that meeting or those meetings, did Moore obtain incriminating information from Randolph? The second issue relates to Moore’s behavior: What, if anything, did Moore do to stimulate conversations with Randolph about the crime with which Randolph was charged?

We address timing first. According to Moore, he had two early meetings with Oppliger and Chavez, and the first of those two meetings took place before he obtained incriminating information from Randolph. According to Oppliger and Chavez, there was only one early meeting, and that meeting took place after Moore had obtained the incriminating information.

Moore testified before the magistrate judge that he first met with Deputy Oppli-ger and Chavez on August 17, 1989, the day he gave his letter requesting leniency to his defense attorney. Moore further testified that it was only after he was placed back in the jail cell, after that meeting, that Randolph made many of the incriminating statements to which Moore later testified at Randolph’s trial. Moore testified that he relayed those statements to prosecutors at a second meeting on August 24. Oppliger and Chavez, however, testified that there was only one early meeting. According to their testimony, that meeting took place on August 24.

The district court did not resolve whether Moore first met with Oppliger and Chavez on August 17 or 24. The magistrate judge’s report, which the district court adopted, stated that “although it is not clear, there is substantial evidence in the record to support a finding that petitioner’s admission was made to Moore prior to Moore’s first meeting with Oppliger and Chavez, whether that meeting occurred on August 17, 1989 or on August 24, 1989.” (emphasis added). A statement that “there is substantial evidence in the record to support a finding” is not the same thing as a finding. That is, the magistrate judge did not recommend, and the district court did not make, a finding that Randolph’s admission was made to Moore prior to Moore’s first meeting with Oppliger and Chavez.

There is “substantial evidence” that would support the opposite finding — that there were two early meetings, and that the first occurred on August 17, before Randolph made damaging admissions to Moore. Moore testified consistently and repeatedly that he met with Oppliger and Chavez on both August 17 and 24. Moore testified that at the first meeting on August 17 he had little to offer Oppliger and Chavez in the way of incriminating statements and that he only told them of Randolph’s incriminating statements at the second meeting. Moore also testified that Randolph did not make incriminating statements until after the trial judge declared a mistrial in the first trial. According to Moore, the mistrial was declared before Moore was placed back in the jail cell with Randolph after his first meeting with Oppliger and Chavez on August 17.

Moore testified that he was asked only general, introductory questions at the initial meeting on August 17 but was asked more pointed, detailed questions at the second meeting on August 24. Finally, Moore specifically described two different rooms in which each of the meetings occurred. According to Moore, the first meeting took place in an empty courtroom, and the second took place in a large conference room with a large table.

Chavez and Oppliger each testified that only one meeting occurred, but they described differently the room in which this meeting took place. Chavez testified that the meeting was in an empty courtroom. Oppliger testified that it was in a jury room with a large table. These two rooms, as described by Chavez and Oppli-ger, appear to correspond to the rooms Moore described in which his two meetings took place. Moreover, Oppliger did not have any independent recollection of whether there were one or two meetings; he relied solely on a report written by Chavez to conclude that only one meeting took place.

We do not recount the foregoing to usurp the factfinding role of the district court. Rather, we recount it to make clear the necessity for such factfinding. It is true, as the magistrate judge wrote, that there is “substantial evidence” to support a finding that there was only one early meeting, and that that meeting took place before Moore obtained incriminating information from Randolph. But it is also true that there is substantial evidence for the opposite finding, that there were two meetings and that Moore obtained incriminating evidence between the first and second meetings. Depending on which alternative is true, Randolph’s Sixth Amendment rights may or may not have been violated.

We next address Moore’s behavior. At the evidentiary hearing before the magistrate judge Moore testified that, at the first meeting with Oppliger and Chavez, “I told them what I knew, but I don’t think I knew that much then.” Nevertheless, Moore left the meeting feeling that “if they’re asking me questions about this guy, then if I give him information about him, then they have to do something for me.” Asked whether he was able to “get any more information for Mr. Oppliger” between the first and second meetings, Moore responded affirmatively. Moore testified that “if [Randolph] was to be on a subject about something that was in the area [of the crime] .... if it was along the line of that, and I figured that he wasn’t suspicious in me asking him about it, then I asked him about it.” According to Moore’s testimony, he further encouraged Randolph to provide information by “being friendly and talkative. Just, I guess, act like being-I was being on his side.” In response to the question, “Did you lead [Randolph] on to provide you with information?” Moore testified, “Of course. Yes.”

The magistrate judge wrote that “there is no evidence to support a finding that ... Moore in fact took action” to “deliberately elicit incriminating statements from petitioner.” (Emphasis added.) This statement is incorrect. Moore’s testimony, just recounted, is first-person evidence supporting precisely that finding — that Moore “took action ... to deliberately elicit incriminating statements from[Randolph].”

3. Massiah Violation

If, in fact, the State placed Moore in a cell with Randolph after he indicated his willingness to cooperate with the prosecution, the State “intentionally create[d] a situation likely to induce [Randolph] to make incriminating statements without counsel’s assistance.” United States v. Kimball, 884 F.2d 1274, 1278 (9th Cir.1989). If that is true, Oppliger and Chavez took the risk that Moore might “deliberately elicit” information from Randolph within the meaning of Massiah and Henry and that such information would be excluded at trial. According to Moore’s testimony, that is exactly what happened. See Henry, 447 U.S. at 271 , 100 S.Ct. 2183 . Therefore, subject to factual determinations to be made by the district court, Randolph has potentially established a Massiah violation.

Our decision in Brooks v. Kincheloe, 848 F.2d 940 (9th Cir.1988), is consistent with our conclusion in this case. In Brooks , the defendant was indicted for the murder of a young boy. While he was in custody awaiting trial, he shared a cell with Kee, to whom he admitted killing the boy. When detectives found out that the defendant had confided that information to Kee, they asked Kee to “tell them what Brooks had been saying” and to “remember anything further Brooks might tell [Kee],” but promised nothing in return. Id. at 943 . Kee said that he wanted to talk to his attorney first and was returned to his cell, which he shared with the defendant. A few days later, Kee provided prosecutors a written statement detailing what the defendant had said to him. The statement included information that the defendant had revealed to Kee after the initial meeting with prosecutors. After Kee provided prosecutors with the statement, he was moved to another jail and given $100. Id. at 942 .

We concluded that all of the defendant’s incriminating statements could be used at trial, including those made to Kee after he met with detectives. The court found that Brooks had confessed his responsibility for the murder to Kee before Kee met with detectives, that “the detectives did not request Kee to elicit any information from defendant,” and that Kee was not used by the police “to carry out any deliberate and surreptitious investigation of defendant.” Id. at 944-45 . We refused to disturb the state court findings of fact “that Kee was not a government agent at the time that Brooks made the incriminating statements concerning the murder.... While these findings indicate that Kee did take action beyond mere listening, they also clearly demonstrate that he did this before the detectives talked to him.” Id. at 945 .

In this case, however, there is substantial evidence to support a conclusion that Oppliger and Chavez knew or should have known that Moore believed that he would receive leniency if he elicited incriminating statements from Randolph, circumstances sufficient to make Moore a government agent. Further, there is substantial evidence that, after meeting with Oppliger and Chavez, Moore took affirmative steps to elicit information from Randolph. This evidence of government action “designed deliberately to elicit incriminating remarks” removes this case from the purview of Brooks .

4. Summary

We conclude that Moore was acting on behalf'of the State when he was put back in the cell with Randolph after his first meeting with Oppliger and Chavez. Because it is within the district court’s province as factfinder, we do not determine when the first meeting between Moore and Opplinger and Chavez took place and when in relation to that meeting Moore obtained incriminating information from Randolph. Nor do we determine precisely what Moore did to obtain the incriminating information from Randolph. We vacate the district court’s decision that Randolph’s Sixth Amendment rights under Massiah were not violated, and we remand to the district court for further factfinding.

D. Due Process Violations

Randolph argues that the testimony of both Moore and Konkle should have been excluded as inherently unreliable and that the use of their testimony violated due process. A violation of state evidence rules is insufficient to constitute a due process violation. See Jammal v. Van de Kamp, 926 F.2d 918, 919 (9th Cir.1991). Rather, we must determine “whether the admission of evidence rendered the trial so fundamentally unfair as to violate due process.” Windham v. Merkle, 163 F.3d 1092, 1103 (9th Cir.1998) (citing Reiger v. Christensen, 789 F.2d 1425, 1430 (9th Cir.1986)). In order to grant relief, a federal habeas court “must find that the absence of ... fairness fatally infected the trial; the acts complained of must be of such quality [that they] necessarily prevented] a fair trial.” Kealohapauole v. Shimoda, 800 F.2d 1463, 1465 (9th Cir.1986) (internal quotation marks and citation omitted).

Although Moore was a government informant, the use of a government informant does not automatically render a trial unfair. Under state law, Randolph was entitled to receive, and did receive, an instruction cautioning the jury to view the testimony of Moore with caution. See California Jury Instructions-Criminal 3.20. Randolph was given ample opportunity to cross-examine Moore and, in so doing, brought out the fact that Moore had entered into a favorable plea bargain with the government, revealing an incentive for testifying. The jury was thus given sufficient opportunity to consider Moore’s credibility. See United States v. Cuellar, 96 F.3d 1179, 1182 (9th Cir.1996) (declining to upset a jury verdict based on an informant’s testimony, and noting “[t]he established safeguards of the Anglo-American legal system leave the veracity of a witness to be tested by cross-examination, and the credibility of his testimony to be determined by a properly instructed jury”) (quoting Hoffa v. United States, 385 U.S. 293, 311 , 87 S.Ct. 408 , 17 L.Ed.2d 374 (1966)).

Similarly, the use of Konkle’s testimony did not violate due process. The defense strenuously cross-examined Konkle, using racist statements he had made to the prosecution (which the prosecution had voluntarily turned over to the defense) to impeach him. Further, the defense questioned Konkle about his motives in testifying, establishing that he expected a plea agreement to be offered in exchange for his testimony against Randolph. Even though Konkle has subsequently recanted his testimony, an after-the-fact suggestion that his testimony was false is insufficient to afford Randolph a new trial based on a due process violation. See Napue v. Illinois, 360 U.S. 264, 269 , 79 S.Ct. 1173 , 3 L.Ed.2d 1217 (1959).

E. Brady Violations

Finally, Randolph challenges the prosecution’s failure to turn over details of the State’s dealings with Moore and Konkle. Under Brady v. Maryland, 373 U.S. 83 , 83 S.Ct. 1194 , 10 L.Ed.2d 215 (1963), prosecutors have a duty to disclose exculpatory material on their own, without a request by the defendant. Brady material includes evidence that would help to impeach a prosecution witness. See United States v. Brumel-Alvarez, 991 F.2d 1452, 1461 (9th Cir.1992). However, a defendant must also show that “there is a reasonable probability that, had the evidence been disclosed to the defense, the result of the proceeding would have been different.” United States v. Bagley, 473 U.S. 667, 668 , 105 S.Ct. 3375 , 87 L.Ed.2d 481 (1985).

We are unable to assess the strength of Randolph’s Brady claim with respect to Moore. Randolph rested his Brady claim on the argument that the prosecution failed to turn over information that might have allowed him to pursue suppression of relevant evidence under Massiah . Randolph did not argue that information about any meeting between Moore and state officials while he and Moore shared a cell would have provided general impeachment evidence. Therefore, Randolph has a viable Brady claim only if the district court finds that Moore and state officials met before Randolph made incriminating statements to Moore.

As described above, we do not know when the first meeting between Moore and the prosecution team took place.

Because we do not know the date of that first meeting, we do not know what happened between Moore and Randolph after that meeting. Because we do not know these things, we are unable to determine whether there was a Brady violation.

Randolph’s Brady claim with respect to Konkle is that the prosecution did not tell the defense the precise extent of Konkle’s plea deals with the prosecution, which it could have used for impeachment purposes. However, even if the defense had this additional information, it is unlikely that the result of the trial would have been different. As noted earlier, counsel cross-examined Konkle at length, impeaching him with his racist statements, as well as with the fact that his sentencing hearing had been postponed and that he was hoping to reach a deal with the district attorney. As a result, “the jury knew that [Konkle] had motivation to adjust his testimony to assist the government’s case and could weigh his testimony accordingly.” Hayes v. Woodford, 301 F.3d 1054, 1074 (9th Cir.2002) (concluding that there was no Brady violation in the prosecution’s failure to disclose its offer of leniency because, considering the other evidence calling witness’s testimony into question, it would not have been material to the jury’s verdict). Further, the district court concluded that Konkle’s testimony was inconsequential to the verdict. Thus, Randolph has not shown that the outcome of the trial would have been different had he been given the information about Konkle that he alleges the prosecution withheld from him, and the district court therefore properly denied his claim.

Conclusion

We vacate the district court denial of Randolph’s Massiah claim and remand for factual findings. We do not decide Randolph’s Brady claim with respect to Moore. We affirm the district court’s denial of Randolph’s other claims.

AFFIRMED in part, VACATED in part, and REMANDED.

. The jury instruction was: The testimony of an in-custody defendant should be viewed with caution and close scrutiny. In evaluating such testimony, you should consider the extent to which it may have been influenced by the receipt of, or expectation of, any benefits from the party calling that witness. This does not mean that you may arbitrarily disregard such testimony, but you should give it the weight to which you find it to be entitled in the light of all the evidence in the case.