McKEOWN, Circuit Judge.
This petition for a writ of habeas corpus presents the question whether the prosecution in a robbery case suppressed evidence in violation of
Brady v. Maryland,
373 U.S. 83, 83 S.Ct. 1194, 10 L.Ed.2d 215 (1963). In analyzing this issue, the Washington Supreme Court failed to consider the cumulative effect of the undisclosed evidence and thus its decision was contrary to clearly established Supreme Court precedent. 28 U.S.C. § 2254(d)(l)(permitting grants of habeas corpus where the state court issued a decision that was contrary to, or an unreasonable application of, clearly established federal law);
Kyles v. Whitley,
514 U.S. 419, 436-37 & n. 10, 115 S.Ct. 1555, 131 L.Ed.2d 490 (1995) (holding that the State’s disclosure obligation under
Brady
turns on the cumulative effect of the withheld evidence, not an item by item analysis). Even so, we conclude on de novo review that the witness who would have been impeached by the suppressed evidence was so severely discredited and not so critical to the prosecution’s case that there is no reasonable probability that the withheld evidence would have affected the outcome. Accordingly, we affirm the dis
trict court’s denial of the petition.
Factual and Procedural Background
Sheniece Brown was working as a clerk at Payless Shoe-Source when a man, who had a handkerchief covering his nose and mouth and appeared to be bleeding, entered the store. As Brown approached him, she realized that he was not injured but instead had red marks drawn with makeup on the backs of his hands and on the sides of his face. The man told Brown that it was a holdup and forced her to open the cash register and till by shoving a hard object into her back that felt like a gun. As the man took the cash from the till, she stood to his right side for a couple of minutes trying to memorize his profile. She testified, “I stood back on the right side of the register ... about four feet away ... and I was trying to memorize his profile. And so he brought the handkerchief down from his face.... ” She saw that his face was covered with brown foundation and that there was a blue circle painted on one cheek along with some red marks on both sides of his face.
Brown initially provided a general description of the assailant, but without a name or identification of a specific individual, the police had no leads. The so-called “clown robber” became the subject of media headlines. A few days later, Brown talked about the robbery with two of her co-workers, Maria Diaz and Amber Trainer. Based on her description, Trainer and Diaz said they thought the robber was a man who frequently visited the store but never bought anything. Trainer told Brown what the suspicious customer looked like, noting that he had several tattoos. Trainer and Brown put their heads together and concluded that the robber’s makeup probably was covering the tattoos.
After her discussion with her colleagues, Brown called the police and told a detective that the robber looked a lot like a man who often came to the store. Although Brown had never seen the suspicious customer before, she described him and emphasized that he had tattoos on his hands and neck that she thought were covered by the makeup. Upon hearing her description, the detective said, “I think I know who that is. This guy’s been in jail before.”
Based on the additional information about the suspicious customer, the police pursued the case. Brown identified Barker from a photo montage of six men. When she first saw Barker’s picture, it “took [her] breath away because ... that’s him.” Brown was confident in her identification because of the suspect’s “medium-sized nose, his cheekbones, and he had a moustache on the picture in this montage, even though he didn’t that night, but I knew that was him.” After she identified Barker, she was shown a booking photograph and a profile shot of Barker that made her even more certain that the identification was correct.
Barker was first tried in June 1999. Brown’s identification was the centerpiece of the prosecution’s case because the store
security camera failed to produce a usable photograph of the robber, and the police failed to obtain fingerprints. Barker represented himself pro se and the trial ended in a hung jury.
Between the first and second trials, the prosecution obtained new evidence from a jailhouse snitch, Raul Abundiz. At the second trial, Abundiz testified that Barker confessed to committing the crime when the two of them were near a motel in early April, 1999. Abundiz also testified that he and Barker talked on June 14, 1999, around 10 a.m. when they were waiting for their respective court appearances in a holding cell.
Abundiz’s testimony, though, was not without warts. He was impeached in all manner of ways, from admitting to lying, to confessing to prior convictions, to acknowledging that he was testifying only because he had received a deal from the State. The image of Abundiz presented to the jury was that of a chronic drug user willing to do anything to get out of jail and anxious to capitalize on his knowledge about Barker: The details of Abundiz’s deal were fully disclosed and he admitted that he was testifying only because of the agreement. The jury also heard testimony that Barker had threatened to kill Abundiz and that Abundiz was looking for Barker “to do some harm to him” with a pistol.
The jury returned a conviction for second degree robbery, and Barker was sentenced to life without parole. He appealed to the Washington Court of Appeals, which affirmed his conviction.
See State v. Barker,
103 Wash.App. 893, 14 P.3d 863 (2000). The Washington Supreme Court denied review.
After he exhausted his direct appeals, Barker pursued relief via a Personal Restraint Petition (“PRP”), Washington State’s mechanism for collateral challenges. An independent investigation revealed several pieces of evidence that had not been disclosed to the defense before trial. This evidence forms the basis of Barker’s
Brady
claim.
The first category of undisclosed evidence consists of four misdemeanor convictions of Abundiz that occurred within the year preceding the trial:
1) June 15, 1999 — conviction for obstructing the duties of a police officer and trespassing based on an incident that involved Abundiz lying about his identity to the police.
2) March 30, 1999 — shoplifting conviction based on an incident that also involved possession of drug paraphernalia.
3) March 19, 1999 — shoplifting conviction.
4) August 18, 1998 — false reporting conviction based on an incident that involved providing false information to the police.
All four of these crimes are considered crimes of dishonesty in Washington.
The second group of undisclosed evidence revolves around a charge for residential burglary that Abundiz faced in June 1999, at the same time that Barker’s first trial occurred. Barker identifies three pieces of evidence:
1) Documentation that a court found probable cause to believe that Abundiz committed residential burglary. Barker points to a June 14, 1999, Superior Court of Yakima County Order finding probable cause and setting bail at
$20,000 and to a transcript of Abundiz’s probable cause hearing.
2) A police report documenting a June 14, 1999, meeting between the same detective with whom Brown spoke and Abundiz during which they discussed the residential burglary charge. The report states that the prosecutor’s office concluded there was not enough evidence to prove the case. The report does not mention Abundiz’s robbery of a JC Penny store, which occurred a few days earlier, or his conversation with Barker, which allegedly occurred that morning. The report makes no mention that Abundiz had information about the robbery of Payless.
3) A June 16, 1999, order releasing Abundiz and stating that the State decided not to charge him with residential burglary. The order was signed by a deputy prosecutor “for Ken Ramm,” who was the prosecutor in Barker’s first trial.
Barker first raised his claim that the prosecution withheld the above evidence in the PRP he filed in the Washington Court of Appeals, which dismissed his petition. The Washington Supreme Court, through its Commissioner, denied discretionary review in a thorough opinion. The court itself adopted the Commissioner’s decision by denying a motion to modify the ruling. The Washington Supreme Court held that Barker failed to show that suppression of the evidence caused him prejudice because Abundiz’s credibility was so thoroughly destroyed that the withheld convictions would not have mattered. The court went on to note that the alleged residential burglary deal was speculative and that any evidence indicating a deal was cumulative because the jury already knew Abundiz received benefits for his testimony.
Barker filed a federal petition for a writ of habeas corpus in the district court, which reviewed the Washington Court of Appeals’ decision and denied relief.
Discussion
I. Standards and Scope of Review
In a habeas corpus petition, two standards of review are at play.
See, e.g., Lambert v. Blodgett,
393 F.3d 943, 964-65 (9th Cir.2004). The first standard is the straightforward de novo standard we employ to evaluate the district court’s decision to deny the petition.
E.g., Beardslee v. Woodford,
358 F.3d 560, 568 (9th Cir.2004) (as amended). The standard that governs review of the state court’s decision is established by the Antiterrorism and Effective Death Penalty Act of 1996 (“AEDPA”), which prohibits a federal court from granting habeas relief “with respect to any claim that was adjudicated on the merits in State court proceedings unless the adjudication of the claim resulted in a decision that was contrary to, or involved an unreasonable application of, clearly established Federal law, as determined by the Supreme Court of the United States.” 28 U.S.C. § 2254(d)(1).
Before we can apply AEDPA’s standards, we must identify the state court decision that is appropriate for our review. When more than one state court has adjudicated a claim, we analyze the last reasoned decision.
See Ylst v. Nunnemaker,
501 U.S. 797, 803-04, 111 S.Ct. 2590, 115 L.Ed.2d 706 (1991);
Avila v. Galana,
297
F.3d 911, 918 (9th Cir.2002) (“In determining whether a state court decision is contrary to federal law, we look to the state’s last reasoned decision.”). Here, the last reasoned state decision is the Washington Supreme Court’s denial of discretionary review of the PRP dismissal. The Court, through its Commissioner, issued a seven-page order that explained in detail why review was denied and specifically examined the substance of Barker’s
Brady
claim.
Determining that the Supreme Court’s decision is the last reasoned decision does not resolve the question whether it is the proper decision to review. We must still decide whether it is “an adjudication on the merits” because 28 U.S.C. § 2254(d) applies only to those claims that have been adjudicated on the merits. 28 U.S.C. § 2254(d). The requirement of an adjudication on the merits does not mandate a hearing or other judicial process beyond rendering a decision; rather it means that the court must finally resolve the rights of the parties on the substance of the claim, rather than on the basis of a procedural or other rule precluding state review of the merits.
Lambert,
393 F.3d at 969. Under this definition of “adjudication on the merits,” in
Lambert
we concluded that a Washington Supreme Court Ruling Denying Review, the same kind of decision at issue here, was an adjudication on the merits.
Id.
at 970 & n. 17 (stating that the Washington Supreme Court reached the merits of the petitioner’s claim and deciding to review primarily the Supreme Court’s ruling).
Lambert,
then, settles that a Washington Supreme Court order denying review of a PRP, when it is reasoned and discusses the merits, is an “adjudication on the merits” that may be reviewed by federal courts under AEDPA. Even before
Lambert
fleshed out the meaning of “adjudication on the merits” under AEDPA we reviewed Washington Supreme Court denials of review.
See Riley v. Payne,
352 F.3d 1313, 1316, 1322 (9th Cir.2003).
Thus, we travel well charted territory when we consider the Washington Supreme Court’s order as an adjudication on the merits that we may review under AEDPA.
The State urges us to review the decisions of the Washington Court of Appeals and Supreme Court together as a collective whole. This argument relies both on language in 28 U.S.C. § 2254(d) referring to “state court proceedings,” plural, and on Ninth Circuit cases that have reviewed multiple state court decisions together. Neither source persuades us that collective review is appropriate in this case.
The text of 28 U.S.C. § 2254(d)(1) is at best ambiguous as to whether we may review multiple state court judgments at once. Habeas relief, under § 2254(d)(1), is prohibited with respect to “any claim that was adjudicated on the merits in State court proceedings unless the adjudication of the claim resulted in a decision that was contrary to, or involved an unreasonable application of, clearly established Federal law, as determined by the Supreme Court of the United States.” Obviously, the provision employs the plural “State court proceedings,” which could indicate that Con
gress intended all state proceedings to be considered as a whole. Equally likely, however, is that “proceedings” is used here in a generic sense and not as a reference to multiple final adjudications. Significantly, the text refers to
“a
decision that was contrary to, or involved an unreasonable application of, clearly established Federal law.”
Id.
(emphasis added). The reference to a single decision underscores that Congress meant federal courts to review only one final state court decision.
Most telling, the Supreme Court describes AEDPA review as applying to a single state court decision, not to some amalgamation of multiple state court decisions. In
Williams v. Taylor,
529 U.S. 362, 120 S.Ct. 1495, 146 L.Ed.2d 389 (2000), the landmark decision explaining the meaning of the “contrary to” and “unreasonable application of’ prongs of AED-PA, Justice O’Connor, writing for the Court, viewed these prongs of the statute as applying to a single state decision. She repeatedly and consistently used singular references to identify the state court adjudication under review.
See, e.g., id.
at 405, 120 S.Ct. 1495(“A state-court decision will certainly be contrary to our clearly established federal precedent”; “If a state court were to reject”; “the state court’s decision must be substantially different”).
The Court’s analysis of the state court proceedings in
Williams
confirms our conclusion that AEDPA generally requires federal courts to review one state decision. In
Williams,
the Virginia Supreme Court applied a standard that was contrary to federal law while the trial court applied the correct standard.
Id.
at 395, 120 S.Ct. 1495. The Supreme Court did not aggregate the two state court decisions or engage in “collective review.” Instead, it reviewed only the Virginia Supreme Court decision and held that it was contrary to federal law.
Id.
at 397-99, 120
S.Ct. 1495. Thus, even when one state court adhered to federal law, if the last court to review the claim erred, the federal court should review the last decision in isolation and not in combination with decisions by other state courts.
We are cognizant that some of our cases discuss two state court decisions simultaneously.
See Lambert,
393 F.3d at 970 n. 17 (reviewing the state supreme court’s decision but stating that “because the supreme court largely adopted the reasoning of the court of appeals, we also discuss the court of appeals’ decision”);
Lewis v. Lewis,
321 F.3d 824, 829 (9th Cir.2003) (“Because [the state appellate court’s] decision affirmed the trial court and adopted one of the reasons cited by the trial court, ... our analysis will necessarily include discussion of the trial court’s decision as well.”). In each of these cases, the last reasoned decision adopted or substantially incorporated the reasoning from a previous decision and, as a result, it was reasonable for the reviewing court to look at both decisions to fully ascertain the reasoning of the last decision. Here, the Washington Supreme Court did not adopt the Court of Appeals’ decision, although agreeing with its reasoning. Rather, the Commissioner reviewed the record independently and authored an order that reflected his own reasoning. Thus, we review the Supreme Court Ruling Denying Review without considering the decision of the Washington Court of Appeals.
II. “Contrary To Clearly Established Federal Law”
Supreme Court holdings at the time of the state court’s last reasoned decision are the source of clearly established Federal law for the purposes of AEDPA.
Williams,
529 U.S. at 412, 120 S.Ct. 1495;
Clark v. Murphy,
331 F.3d 1062, 1069 (9th Cir.2003). At the time the Washington
Supreme Court denied review of Barker’s PRP, the components of a
Brady
claim were clearly established by Supreme Court law — a defendant must show that 1) the prosecution suppressed evidence that 2) was favorable to the accused and 3) was material.
Strickler v. Greene,
527 U.S. 263, 281-82, 119 S.Ct. 1936, 144 L.Ed.2d 286 (1999) (summarizing the elements of a
Brady
claim). The Court first defined materiality in
United States v. Bagley,
473 U.S. 667, 682, 105 S.Ct. 3375, 87 L.Ed.2d 481 (1985), holding that evidence is material “if there is a reasonable probability that, had the evidence been disclosed to the defense, the result of the proceeding would have been different.”
Crucial to this case is the Supreme Court’s requirement that the materiality of the withheld evidence be analyzed cumulatively: One “aspect of
Bagley
materiality to be stressed here is its definition in terms of suppressed evidence considered collectively, not item by item.”
Kyles,
514 U.S. at 436, 115 S.Ct. 1555. The Court specifically explained that courts should “evaluate the tendency and force of the undisclosed evidence item by item; there is no other way. We evaluate its cumulative effect for purposes of materiality separately and at the end of the discussion.”
Id.
at 436 n. 10, 115 S.Ct. 1555.
Despite the clarity of the Supreme Court’s directive that a materiality analysis include an assessment of the cumulative effect of undisclosed evidence, the Washington Supreme Court did not conduct such an analysis. Instead, the court separately analyzed the withheld convictions and the evidence related to Abundiz’s residential burglary charge on a piecemeal basis and then ended its analysis. Examining the effect of the withheld evidence group by group, as the court did there, is only the first half of the analysis. Because the Washington Supreme Court failed to complete the second half of-the equation, which requires evaluation of the cumulative effect of all the withheld evidence “separately and at the end of the discussion,”
id.,
its decision was contrary to clearly established Federal law as set forth in
Kyles. See Castleberry v. Brigano,
349 F.3d 286, 291-92 (6th Cir.2003) (holding that a state court decision that failed to analyze withheld evidence cumulatively was contrary to Kyles).
Barker offers an alternate argument for why the decision is contrary to Federal law. He contends that the Washington Supreme Court’s “actual and substantial prejudice” standard imposed a more stringent prejudice requirement than Federal law because the prejudice inquiry under
Brady
requires only a showing that “there is a reasonable probability that, had the evidence been disclosed to the defense, the result of the proceeding would have been different.”
Bagley,
473 U.S. at 682, 105 S.Ct. 3375;
see also Strickler,
527 U.S. at 280, 119 S.Ct. 1936. The showing of actual and substantial prejudice, Barker contends, heaps an extra burden on petitioners that contradicts Federal law.
We need not reach Barker’s argument, having already concluded that the decision was contrary to Federal law, but we take this opportunity to observe that the interplay of the Washington PRP “actual and substantial prejudice” standard and the
Brady
materiality standard is not without confusion. It is unclear whether the PRP standard, as used here, governs a threshold examination that determines whether a petitioner may raise his claim,
see, e.g., In re Cook,
114 Wash.2d 802, 792 P.2d 506, 510-11 (1990) (en banc), or whether it is the ultimate showing required for relief,
see id.
at 512. The difference matters for federal habeas review because if the standard is a threshold obstacle to state court
review, it would not conflict with the
Brady
materiality standard. In that event, the petitioner simply would fail to gain access to Washington’s post-conviction review. If it is a substantive standard imposed in addition to the prejudice showing required under
Brady
and its progeny, it appears to impose a higher burden than is required to establish prejudice under
Brady.
III. Analysis of the
Brady
Claim
Because the state court did not
conduct the proper analysis, AEDPA’s restrictions on our review do not apply.
See, e.g., Cooperwood v. Cambra,
245 F.3d 1042, 1046 (9th Cir.2001);
Wade v. Terhune,
202 F.3d 1190, 1195 (9th Cir.2000). Accordingly, we review Barker’s
Brady
claim de novo applying the correct analysis.
See Cooper-Smith v. Palmateer,
397 F.3d 1236, 1243 (9th Cir.2005) (as amended); Co
operwood,
245 F.3d at 1046.
We begin by returning to
Brady
itself, which established that “the suppression by the prosecution of evidence favorable to an accused upon request violates due process where the evidence is material either to guilt or punishment, irrespective of the good faith or bad faith of the prosecution.” 373 U.S. at 87, 83 S.Ct. 1194. Barker readily establishes the first prong of his
Brady
claim, that the evidence would have been favorable to him. It is well settled that evidence impeaching the testimony of a government witness falls within the
Brady
rule,
see, e.g., Singh v. Prunty,
142 F.3d 1157, 1161 (9th Cir.1998), and the government does not dispute that the four misdemeanor convictions were relevant to Abundiz’s credibility. Nor is there a serious argument that evidence of a deal with Abundiz and of contact between Barker and Abundiz would not have been useful impeachment evidence.
See, e.g., Bagley,
473 U.S. at 683, 105 S.Ct. 3375 (evidence indicating that witness received inducement from prosecution to testify constituted evidence favorable to the accused);
Benn v. Lambert,
283 F.3d 1040, 1057 (9th Cir.2002) (explaining that prosecution-provided benefits are
Brady
material because they indicate that a witness may have reasons for testifying other than altruism). Barker’s claim that the State agreed to drop the residential burglary charge against Abun-diz in exchange for his testimony is speculative, but at the same time, it takes little imagination to see how a competent attorney could have implied that such a deal existed. It takes even less imagination to see how evidence calling into question whether Barker and Abundiz talked on June 14 would have helped Barker impeach Abundiz.
The second prong of the
Brady
claim, that the prosecution failed to disclose evidence, is subject to some dispute. The Washington Supreme Court was un
convinced that the prosecution had access to all the information that Barker argues was suppressed,
but we need not explore the particulars of this issue because Barker’s claim is defeated by the lack of materiality of the allegedly withheld evidence. Significantly, “[w]e do not ... automatically require a new trial whenever a combing of the prosecutors’ files after the trial has disclosed evidence possibly useful to the defense but not likely to have changed the verdict ... A finding of materiality of the evidence is required.... ”
Bagley,
473 U.S. at 677, 105 S.Ct. 3375 (internal quotations and citation omitted). It is upon this third element, materiality, that Barker’s claim fails.
The standard for materiality set out in
Bagley
is worth repeating: Evidence is material if “there is a reasonable probability that, had the evidence been disclosed to the defense, the result of the proceeding would have been different.”
Bagley,
473 U.S. at 682, 105 S.Ct. 3375. “A ‘reasonable probability’ is a probability sufficient to undermine confidence in the outcome.”
Id.
In analyzing the materiality factor, we heed the Supreme Court’s instruction that the defendant need not show that he would more likely than not have been acquitted had the evidence been disclosed.
Kyles,
514 U.S. at 434-35, 115 S.Ct. 1555. Rather, the touchstone of the inquiry is whether Barker received a fair trial that resulted in a verdict “worthy of confidence.”
Id.
at 434, 115 S.Ct. 1555. We examine the withheld evidence individually and cumulatively.
See id.
at 436-37 & n. 10, 115 S.Ct. 1555.
A. Materiality of the Undisclosed CONVICTIONS
Barker argues that Abundiz’s four convictions were material because they would have so undermined the credibility of a key prosecution witness that the jury could not have believed his story that Barker confessed. The convictions, though, merely duplicate grounds for impeaching Abundiz that were actually presented to the jury.
See, e.g., United States v. Croft,
124 F.3d 1109, 1124 (9th Cir.1997) (withheld psychiatric report that would have established witness’s memory loss was cumulative when defense had elicited fact of memory loss on cross-examination);
United States v. Marashi,
913 F.2d 724, 733 (9th Cir.1990) (withheld evidence that would have contradicted trial testimony was cumulative and not material because deposition contradicted trial testimony in the same way).
The convictions surely would have highlighted Abundiz’s dishonest nature, but we do not think that additional evidence of Abundiz’s penchant for lying would have affected the jury when his proclivity for lying had already been firmly established. The jury heard Abundiz admit that he lied to the police about his identity by giving his brother’s and cousin’s names instead of his own. He was caught in a lie on the stand when he said he had used drugs for only a year and a half but had to acknowledge a conviction for possession of cocaine six years earlier. More important, the jury heard evidence suggesting that Abun-diz was lying about Barker’s confession. When Abundiz first offered information about the robbery, he had the details wrong: He called Barker “Parker” and
thought the name of the robbed store was “Pay’N Save,” not “Payless.” The erroneous details permitted a strong inference that Abundiz may have fabricated the confession. Finally, one of Abundiz’s jail-mates, who was not getting a deal, testified that in early June, a few months after Barker allegedly confessed, Abundiz said he knew nothing about the robbery. The testimony suggests that Abundiz fabricated the confessions sometime in June. All in all, Abundiz was portrayed as a serial liar whose story that Barker confessed was highly doubtful.
The convictions might also have shown that Abundiz was a career criminal desperate to escape from jail time. But just as the jury knew that Abundiz was a liar, the jury also knew that he had been in and out of jail several times. He admitted to a juvenile conviction and the jury knew of the three other crimes that were involved in his agreement with the State. Specifically, the jury knew that Abundiz had been charged with possession of narcotics and robbing a JC Penney store, as well as that he was currently serving a 90-day sentence.
Only one conviction, the March 30, 1999, shoplifting conviction, is qualitatively different than the evidence introduced. That conviction places Abundiz behind bars from March 29, the date of the shoplifting incident, until April 9. Barker allegedly confessed to Abundiz at a house behind a motel sometime after Barker’s picture appeared on the news and before Barker was arrested on April 10. According to Barker, the jury was under the impression that the confession could have occurred any time during the ten days that elapsed between the robbery and Barker’s arrest. The withheld conviction, Barker contends, whittles down the time span in which the alleged confession could have taken place from ten days to the thirty-two hours Barker and Abundiz were outside jail at the same time.
Where Barker misses the mark is that the conviction does not whittle down the time period during which the confession could have occurred from ten days to thirty-two hours. Rather, the conviction reduces the possible time frame from only about seventy-two hours to thirty-two hours. Abundiz testified that Barker confessed after having seen his picture on the news. The earliest date that Barker’s face could have appeared on the news was April 7, the day Brown identified him,
meaning that the confession had to occur between April 7 and April 10. Thus, the jury was presented with a three-day time frame, not a ten-day time frame. Seventy-two hours is not a long time to begin with, and we doubt that jurors would be significantly more skeptical of the confession had they known that it had to occur during a thirty-two hour period.
Here, the withheld evidence did not provide “the defense with a new and different ground of impeachment.”
Silva v. Brown,
416 F.3d 980, 2005 WL 1732765, at *7 (9th Cir. July 26, 2005) (quoting
Benn,
283 F.3d at 1056);
see also Horton v. Mayle,
408 F.3d 570, 580 (9th Cir.2005) (although witness’s testimony was impeached by evidence of drug use, lying to police, and assisting in the crime, withheld evidence of
promised immunity is a “wholly different kind of impeachment evidence” and is material). The withheld convictions were cumulative icing on an already crumbling cake.
B. Materiality of the Evidence Related to Abundiz’s Residential Burglary Charge
Barker makes two claims based on the evidence related to Abundiz’s residential burglary charge.
First, he argues that the withheld evidence shows that Abundiz was likely in his probable cause hearing at the time he claims to have talked with Barker on June 14 around 10 a.m., a time when one could infer that Barker confessed. The effort to label this evidence as material fails because the facts do not bear out Barker’s story line. Abundiz did not testify that Barker confessed on June 14. Rather, he claimed only that “Mr. Barker
talked
to [him] while in jail”
(emphasis added) on June 14. The alleged confession was on a different date. Merely talking is vastly different from confessing. “A confession is like no other evidence. Indeed, the defendant’s own confession is probably the most probative and damaging evidence that can be admitted against him.”
Arizona v. Fulminante,
499 U.S. 279, 296, 111 S.Ct. 1246, 113 L.Ed.2d 302 (1991) (internal quotation marks and citation omitted). Thus, had the June 14 meeting involved a purported confession, then evidence undermining the likelihood that such a confession occurred would surely hold more sway than evidence of a mere conversation. Unfortunately for Barker, the evidence does not support his claim of a June 14 confession.
To the extent the timing of the probable cause hearing undermines the June 14 conversation, it is of minimal significance because the defense established at trial that Barker probably was in another courtroom at that time. Whether Barker was in one courtroom and Abundiz in another (according to the withheld evidence), or whether Barker was in one courtroom and Abundiz in a holding cell (according to the testimony), the implication is the same. In either scenario, the two men probably were in different places when Abundiz claims they spoke.
Barker’s second argument is that the withheld evidence would have shown that the State agreed to drop a residential burglary charge against Abundiz in exchange for favorable testimony. From a series of events that occurred in June 1999, Barker tries to craft a theory of collusion that the State induced Abundiz to fabricate his testimony. Like the June 14 confession theory, this claim too rests on conjecture.
According to Barker, the undisclosed evidence would tell the following tale: On June 14, 1999, the same day Abundiz claims he and Barker talked in the holding cell, Abundiz met with the same detective who identified Barker based on Brown’s description, Detective Salinas, to talk about the residential burglary. Detective Salinas’s report on the meeting does not indicate that they discussed Abundiz’s burglary of a JC Penney, which Salinas had written up only four days earlier. Also
absent is any notation that Abundiz had information about the “clown robbery.” Barker is skeptical that Abundiz, a drug addict facing a $20,000 bail, would have passed up an opportunity to use his supposed knowledge about the Payless robbery as a bargaining chip. Barker surmises that “[a] reasonable juror would search for some other explanation for why either Abundiz did not disclose the confession (i.e. it never happened) or why Detective Salinas did not write it down (i.e. it might not sound as credible coming from somebody in Salinas’ position, being already involved and going to testify in the following days in the first Barker trial).” The suppression of the evidence prevented the defense from telling this tale because, without the evidence, Barker’s attorney was not aware that Abundiz had met with Detective Salinas the day Barker and Abundiz allegedly talked, or that the State dropped the residential burglary charge against him.
The flaw in Barker’s theory is that it is mere speculation.
Cf. Downs v. Hoyt,
232 F.3d 1031, 1037 (9th Cir.2000) (holding that undisclosed evidence of several leads in the sheriffs files was not material because defendant’s theory that they might have helped his case was speculative and he did not show how the evidence might have altered the outcome). He has not pointed to any evidence that the State induced Abundiz to conjure up confessions or that Detective Salinas’s June 14, 1999, meeting with Abundiz involved the Payless robbery. The most Barker can offer is a theory woven largely of threads he has created himself to link pieces of evidence. That is not enough: “The mere possibility that an item of undisclosed information might have helped the defense, or might have affected the outcome of the trial, does not establish ‘materiality’ in the constitutional sense.”
Croft,
124 F.3d at 1124(quoting
United States v. Agurs,
427 U.S. 97, 109-10, 96 S.Ct. 2392, 49 L.Ed.2d 342 (1976)).
Even if the prosecution did hide one of the benefits it gave Abundiz, a deal involving the residential burglary charge would have very nearly replicated evidence already admitted that showed Abundiz received significant benefits for his testimony.
See United States v. Vgeri,
51 F.3d 876, 880(9th Cir.1995) (evidence of witness’s prior cooperation with law enforcement was not material when the jury heard extensive testimony about her cooperation with the Drug Enforcement Agency). The jury knew that one burglary charge was being dismissed, another reduced, and a sentence commuted. Adding the dismissal of one additional charge to that list of already substantial benefits would not add to the impeachment value of the evidence. More importantly, the presence of such a deal is pure speculation and surely not material.
C. MATERIALITY OF THE CüMÜLATIVE IMPACT of the Withheld Evidenoe
Thus far we have examined the force and nature of the withheld evidence item by item, but our materiality analysis is not complete until we consider the cumulative effect of the suppressed evidence.
Kyles,
514 U.S. at 436-37
&
n. 10, 115 S.Ct. 1555 (courts must first evaluate the tendency and force of each item of suppressed evidence and then evaluate its cumulative effect at the end of the discussion). Gauging the collective impact of the withheld evidence requires us to step back and consider the strength of the prosecution’s case, paying close attention to how critical Abundiz’s testimony was.
See Banks v. Dretke,
540 U.S. 668, 700-01, 124 S.Ct. 1256, 157 L.Ed.2d 1166 (2004) (evidence impeaching witness who was “the centerpiece” of a case that lacked physical
evidence was material);
Strickler,
527 U.S. at 292-93, 119 S.Ct. 1936 (evidence impeaching important witness not material when the record provided strong support for the conviction apart from the witness’s testimony);
Horton,
408 F.3d at 578-79 (withheld evidence was material when it impeached a witness who was “central to the prosecution’s case”).
The heart of the prosecution’s case was Brown’s identification of Barker. Without physical or forensic evidence, her testimony provided the only direct evidence that Barker was the robber. Brown was approximately four feet away from the robber for a couple of minutes during the robbery and, during that time, she got a good look at his profile. Brown was practiced at carefully looking at a person’s features because of facial drawing classes she took and she made a conscious effort to memorize her assailant’s profile. She was extremely confident in her identification of Barker from the photo montage. Her testimony and identification at trial were categorical that Barker was the robber. Right away, she said the picture “just kind of took my breath away because I was like, that’s him.... ”
We acknowledge that Brown’s identification was not airtight. Her view was obstructed by the robber’s makeup and, at times, by his handkerchief. Her description also became more precise after she talked with co-workers who knew and already distrusted Barker. Although she did not see any of his tattoos, Brown stated that his hands were covered with markings and makeup. Nonetheless, the question we decide is not whether we believe Brown, but whether we remain confident in the verdict despite the potential damage the withheld evidence would have wrought. In considering this question, we appreciate the central importance of the identification by a witness at the scene. “[D]espite its inherent unreliability, much eyewitness identification evidence has a powerful impact on juries,”
Watkins v. Sowders,
449 U.S. 341, 352, 101 S.Ct. 654, 66 L.Ed.2d 549 (1981) (Brennan, J., dissenting), and the jury likely would have convicted when presented with a confident eyewitness like Brown whose trial testimony was clear and unwavering. Significantly, Abundiz was not such a critical part of the case, nor would the withheld evidence have so discredited him, that our confidence is shaken.
A useful measurement of the importance of Abundiz and the materiality of the withheld impeachment evidence is the lack of emphasis the prosecutor placed on his testimony.
See Kyles,
514 U.S. at 444, 115 S.Ct. 1555(“The likely damage is best understood by taking the word of the prosecutor. ...”);
Horton,
408 F.3d at 580(not-ing that the prosecutor’s emphasis on the importance of the witness’s testimony highlights how important the witness was to the case). The prosecutor spent most of his closing argument reviewing Brown’s identification and shoring up her testimo1 ny. The identification was the essential platform of the State’s case. The prosecutor discussed Abundiz’s testimony as corroborating evidence, but he devoted only a small portion of his argument to Abundiz. Unlike many eases in which withheld evidence is material because it would have impeached a key witness, Abundiz was not the glue holding together the prosecution’s case nor would heaped-on impeachment evidence have altered his already shattered credibility. Consequently, the cumulative impeachment evidence is unlikely to have been the difference between conviction and acquittal.
See, e.g., Benn,
283 F.3d at 1046, 1054 (evidence was material when witness who would have been impeached was “critical” to the prosecution);
Carriger v. Stewart,
132 F.3d 463, 466, 480-81 (9th Cir.1997) (en banc) (impeach
ing evidence was material when the physical evidence was weak and the prosecution relied principally upon the witness who would have been impeached).
Barker makes much of the fact that the first jury that heard his case without Abundiz’s testimony was hung but, when Abundiz’s testimony was added, the jury returned a guilty verdict. It oversimplifies the two trials to conclude that Abun-diz’s testimony was necessarily the difference between a hung jury and a conviction. The fact that his testimony was a key piece of different evidence is surely significant; it is not, however, the defining factor standing alone. Instead, it must be acknowledged that different juries may view the same facts and testimony differently. In fact, the scope and nature of the testimony was different at the two trials, although identification by Brown remained firm. When he represented himself at the first trial, Barker was relentless in his cross-examination of Brown. Her testimony under fire from Barker covers forty-seven pages. He hammered at issues relating to the tattoos, makeup on the neck, markings on the hands, and ability to observe the robber. At the second trial, Brown’s cross-examination was a mere thirteen pages and Barker’s lawyer told the judge, “We can see that the witness’s testimony has changed from the first trial. I’m sure the holes will be plugged up that way....” We simply cannot conclude that Abundiz’s testimony was necessarily the pivotal difference between the two trials.
The real question with respect to Abun-diz is whether he was telling the truth. Even absent the withheld evidence, Abun-diz was thoroughly impeached and showcased as a self-serving jailhouse snitch. The alleged “confession” was wholly suspect in light of the evidence. The withheld evidence simply made it more so. But even without the undisclosed evidence, the defense severely discredited Abundiz by highlighting his deal with the State, his history of lying, his drug addiction, and the errors in his initial statement. The difference between the story of Abundiz that the jury knew and that which would have been presented with the withheld evidence is not significant.
Contra Benn,
283 F.3d at 1057(evidenee was material when the jury never heard that a key witness lied about anything but suppressed evidence would have shown willingness to lie about his testimony);
Carriger,
132 F.3d at 480-81 (evidence was material when prosecution portrayed star witness as nonviolent and truthful but withheld evidence would have shown that he was a sociopath, serial liar, and career felon).
In sum, looking at the withheld evidence in a cumulative sense, as well as than pieee-by-piece, we are confident that the withheld evidence would not have changed the jury’s calculus of Abundiz as a witness. Given Brown’s powerful testimony, we conclude that the withheld evidence was not material.
AFFIRMED.