OPINION

Gilbert Aguilar was convicted of first-degree murder after a jury trial in Los Angeles County Superior Court. A young Hispanic man got out of a white Volkswagen Beetle and shot John Guerrero while Guerrero’s car was stopped at a stoplight. The only question at trial was the identity of the shooter. Aguilar’s defense was that another young Hispanic man, Richard Osuna, had shot Guerrero.

The prosecution introduced evidence that a police dog named Reilly had alerted to a “scent pad,” showing that Aguilar’s scent was present on the front passenger seat of the white Volkswagen. The prosecution did not disclose to the defense that Reilly had a history of making mistaken scent identifications, even though it had stipulated to Reilly’s mistaken identifications in a different trial several months earlier. Following the stipulation, that court had excluded evidence of Reilly’s scent identification from the earlier trial.

Reilly’s scent identification was the only evidence that tied Aguilar to the white Volkswagen. Putting the scent identification to one side, the evidence against Aguilar was weak. No clear motive for Aguilar to shoot Guerrero was ever suggested at trial. No physical evidence tied Aguilar to the crime. The faces of Aguilar and Osuna are very similar, but Aguilar is older and, at the time of the shooting, was significantly taller. A number of eyewitnesses identified Aguilar as the shooter at trial. Several of those witnesses had earlier given a quite different physical description to pólice—one that matched Osuna in age and height rather than Aguilar. At trial, these witnesses changed their description to match Aguilar.

The evidence suggesting that Osuna was the killer was substantial. Osuna’s brother was shot several days before Guerrero was shot. Two witnesses testified that Osuna jumped into a white Volkswagen Beetle to pursue Guerrero’s car as it drove past. One of them testified that Osuna did so in the belief that the “fools” in the car had shot his brother. That same witness testified that Osuna told her a short time later that he had shot a “fool.” Even so, Osuna was never investigated as a suspect in this case. Indeed, the prosecutor in this case expressly told the police not to pursue an investigation of Osuna.

This case comes to us on a petition for habeas corpus under 28 U.S.C. § 2254 . Aguilar argues that the prosecution’s failure to disclose Reilly’s history of mistaken scent identifications violated Brady v. Maryland, 373 U.S. 83 , 83 S.Ct. 1194 , 10 L.Ed.2d 215 (1963), and that the California Court of Appeal’s decision to the contrary was an unreasonable application of Brady . We agree.

I. Standard of Review

We review de novo a district court’s decision to grant or deny a habeas petition under 28 U.S.C. § 2254 . Campbell v. Rice, 408 F.3d 1166, 1169 (9th Cir.2005) (en banc). To prevail in a habeas petition filed after the Antiterrorism and Effective Death Penalty Act of 1996 (“AEDPA”) was enacted, a petitioner must show that the state court’s adjudication of a claim:

28 U.S.C. § 2254 (d).

This is a “highly deferential standard,” Woodford v. Visciotti 537 U.S. 19, 24 , 123 S.Ct. 357 , 154 L.Ed.2d 279 (2002) (per curiam) (internal quotation marks omitted). The “contrary to” clause in § 2254(d)(1) applies where the state court adopts “a rule that contradicts the governing law set forth in Supreme Court cases” or “confronts a set of facts materially indistinguishable from those at issue in a decision of the Supreme Court and, nevertheless, arrives at a result different from its precedent.” Lambert v. Blodgett, 393 F.3d 943, 974 (9th Cir.2004) (citing Lockyer v. Andrade, 538 U.S. 63, 73 , 123 S.Ct. 1166 , 155 L.Ed.2d 144 (2003)). The “unreasonable application” clause applies where “the state court’s application of clearly established law” is “objectively unreasonable.” Lockyer, 538 U.S. at 75 , 123 S.Ct. 1166 . Under either clause of § 2254(d)(1), the law must be clearly established. There must be a “Supreme Court decision that ‘squarely addresses the issue’ in the case before the state court” or one that “establishes an applicable general principle that ‘clearly extends’ to the case before us.” Moses v. Payne, 555 F.3d 742, 760 (9th Cir.2009) (quoting Wright v. Van Patten, 552 U.S. 120, 123, 125 , 128 S.Ct. 743 , 169 L.Ed.2d 583 (2008)). Additionally, the constitutional error must have “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) (internal quotation marks omitted).

II. Background

On July 25, 2001, John Guerrero was driving his red Mitsubishi westbound on Amar Road in La Puente, California, with four friends as passengers. The group was looking for somewhere to eat dinner. Guerrero was not affiliated with a gang. Guerrero drove past several Hispanic males dressed in white baggy shirts, standing near a “primered” white Volkwagen Beetle. Omar Soltero, one of Guerrero’s passengers, testified that Guerrero told the other passengers that he could see the Hispanic males running toward them in his rearview mirror. Guerrero and Soltero thought that the individuals might be trying to attract the attention of the occupants of Guerrero’s car.

Trying to avoid any confrontation, Guerrero kept driving down Amar Road. Guerrero and his friends then decided to go to a restaurant in the opposite direction. Guerrero made a U-turn and drove eastbound, again passing the Hispanic males standing along Amar Road. The males yelled as Guerrero drove past. Victor Carillo, another passenger in Guerrero’s car, testified that they were “throwing up their hands” as Guerrero’s car drove by the second time. Carillo believed that the males were “throwing up a neighborhood”—in other words, indicating their neighborhood gang affiliation.

Guerrero drove several more blocks and stopped at the front of a left turn lane at the corner of Amar Road and Hacienda Boulevard, waiting for a red light to change so he could turn northbound onto Hacienda Boulevard. There was a KFC restaurant on the southwest corner of the intersection. There was a gas station at the northwest corner.

A young Hispanic male wearing a dark baseball cap and a white t-shirt got out of a white Volkswagen Beetle in the KFC parking lot. He went into the street and approached Guerrero’s car. The male reached into the open passenger side front window with a semi-automatic handgun and shot Guerrero seven times, killing him. The other occupants of the car ducked down and were unhurt. The shooter walked rapidly back to the white Volkswagen, and the car drove away.

Based on descriptions by Desiree Hoe-fer, Victor Jara, and Laura Jara—all eyewitnesses to the shooting—a police artist made a sketch of the suspect from the neck up. The sketch was made about a month after the murder. A probation officer who knew Aguilar thought the drawing looked like him. Police then put a photograph of Aguilar from the neck up in a photo “six-pack” and showed it to Hoefer and the Jaras. There was no photograph of Richard Osuna in the sixpack. All three picked Aguilar’s photograph out of the six-pack. Victor Jara made a positive identification. Hoefer and Laura Jara said that the man depicted in the photograph looked similar to the killer, but neither was sure that this was the killer.

A. Evidence Pointing To Richard Osuna

Aguilar’s defense at trial was that Richard Osuna, a young Hispanic male, also known as “Gangster”, was the shooter. Aguilar and Osuna were both members of the Puente Street Gang. Osuna’s younger brother Raymond had been shot, but not killed, by unknown perpetrators the week before Guerrero was killed. At the time of Guerrero’s shooting, Aguilar was twenty years old, between 5'11" and 6'0" tall, and weighed 160 pounds. Osuna was sixteen years old and between 5'5" and 5'7" tall.

The sketch of the suspect, as well as booking photographs of Aguilar and of Richard Osuna, were put into evidence at trial. (Osuna’s booking photograph was taken at the time of his arrest for a different offense a year after the Guerrero murder.) The photographs of Aguilar and Osuna are remarkably similar. The sketch somewhat resembles both Aguilar and Osuna. We include the sketch and the photographs as appendices to this opinion.

At the time of the murder, Aguilar lived with his girlfriend and mother of his child, Mary Saiz, in an apartment on Amar Road. Saiz and Alfred DeAnda, a friend of Aguilar, each testified that they saw Osuna jump into a white or gray Volkswagen Beetle to chase after a red car. DeAnda testified that he was walking down Amar Road on his way to the store when he encountered “Gangster” (Osuna) and five or six other people outside the Amar Road apartment. DeAnda testified that he then saw Gangster hop into a white or gray Volkswagen while Aguilar and Saiz remained behind. DeAnda heard about the shooting the following day. He testified that the “word on the street” was that Gangster had shot Guerrero.

Saiz testified that on the day of the shooting, she, Aguilar, Osuna, and several others were gathered at the apartment she and Aguilar shared in order to celebrate Saiz’s baby shower. At the time, Saiz knew Osuna only as “Gangster”. The group saw a red car drive by the apartment. Osuna appeared to recognize the car and got “real antsy” after it drove by. Saiz testified that Osuna said, “There go those vatos that shot my brother, fool. Fuck that. I’m going to go get those fools.” She testified that Osuna then got into a white Volkswagen Beetle driven by Rico Ballesteros and drove away. Saiz testified that Aguilar left the apartment complex to go to another friend’s house shortly after Osuna drove away. After the shooting Osuna returned to the apartment where Saiz and Aguilar were living and rushed into the bathroom to take a shower. Saiz testified that Osuna said, “I just shot a fool____I got to get the gunpowder off of me.” After showering, Osuna changed into Aguilar’s clothes and left the apartment.

Neither DeAnda’s nor Saiz’s testimony was entirely consistent with their prior statements. However, two jail conversations between Aguilar and Saiz recorded without their knowledge support the testimony DeAnda and Saiz gave at trial. The jury heard both conversations. In the first conversation, Aguilar declared, “I can’t believe I’m in jail for something I didn’t do.” In response, Saiz announced that she had “a plan already done” to find out “Richard’s” last name. Saiz’s plan was to go “by the pad like during the day time or whatever, and steal like a piece of mail.” She hoped the piece of mail would tell her Richard’s last name. Saiz then planned to phone the house from which she stole the mail “and make sure that that’s his last name” by pretending that she was calling from Richard’s school. Saiz also planned to “check in the records at Queen of the Valley” because Richard’s “brother was there the day before.... [Wjhat reason why wouldn’t he want to shoot, know what I mean?” Osuna’s brother Raymond was treated at Queen of the Valley Medical Center beginning on approximately July 21, 2001. Raymond was discharged from that hospital on July 24—the day before Guerrero was murdered. Saiz expressed reservations to Aguilar about approaching Osuna directly “because I ... have to think about the baby and us too, you know, when you do get out.”

At trial, Saiz explained that during this conversation, she knew Richard Osuna only by his first name. She devised her “plan” to find out Osuna’s last name so she could disclose it to a police investigator. Saiz also explained that she eventually “chieken[ed] out” and did not follow through with the plan. She later found out Osuna’s last name through Aguilar, who had learned the name from fellow inmates.

In the second recorded jail conversation, Saiz told Aguilar that she had obtained a document containing the names and addresses of eyewitnesses in the case. Saiz asked Aguilar whether she should “[s]how people” the report. He told her that “there’s no reason to show 'em” because “the[y’]re gonna go do something and get me in trouble” when “I ain’t even done.” Saiz then asked Aguilar if he wanted her “to rat,” and he told her, “Not right now.” Saiz responded, “I will Gilbert ‘cause I’m not a south sider and look at me Gilbert I’m ... falling apart[.]” Saiz also stated that she was “ready to rat,” because if Aguilar ratted himself he would “get killed in there.” She did not want to wait until “it’ll be to[o] late for anyone to do anything.” She stated she was “gonna tell them the fuckin’ whole truth.”

Saiz then told Aguilar, “This reward, like I said it’s 50/50, if you get a lawyer you’ll beat it. If you don’t get a lawyer you’ll have a small chance maybe, know what I mean? Rico’s ass is already, they already he did, he was involved, but th— he’s still out here[.]” Aguilar responded that “[t]hat other fool[ ] stays in Hacienda Heights,” and said he wanted to talk to the “fool.” At the time, Richard Osuna lived in Hacienda Heights.

At trial, Saiz explained that by Rico, she was referring to Rico Ballesteros, the driv er of the white Volkswagen. She understood the “other fool” to refer to Richard Osuna. Saiz also explained that she hesitated to bring Osuna’s name to the police because she feared doing so might put the safety of her family at risk. Saiz explained that “if you tell in court like I am right now or speak of things that you shouldn’t be telling others that other people are doing, ... you get killed or someone comes after you and your family.” Saiz explained that Aguilar did not himself tell the truth because he was likely to be killed in prison if he was perceived as a “rat.”

Saiz testified that at first she tried to resolve the situation without “ratting” Osuna out. Once Saiz identified Osuna by name, she and several friends approached him to ask for money to help her “retain a lawyer or [support] the baby.” She testified that “[w]e went over there because he did do the killing, so therefore I felt like he was responsible for me going through all this situation that I’m going through, and Gilbert [Aguilar] also.” Osuna gave Saiz his phone number, but when she tried to call it, the number had been disconnected. Saiz testified that she eventually got “fed up” with Osuna’s unresponsiveness and decided to tell the truth. She thought she would not be killed because she is not a “south sider”—meaning she did not belong to any gang. Even so, Saiz received threats after she agreed to testify.

The police collected the jail recordings of the conversations between Saiz and Aguilar. In addition, defense counsel and a court-appointed private investigator each brought information about Richard Osuna to Detective Richard Ramirez, the lead police investigator for the Guerrero murder, a few months before trial. Nevertheless, the police never investigated Richard Osuna as a suspect in Guerrero’s murder. Indeed, the prosecutor trying Aguilar’s case specifically told Ramirez “not to follow up” on the Osuna lead because he believed it would be “a wild goose chase.” The prosecutor told the jury at trial that he “instructed Detective Ramirez not to follow up on that lead” because it was “immoral” and “illegal” to do so without more evidence.

B. Evidence Against Aguilar At Trial

At trial, the prosecution introduced three types of evidence to support its case that Aguilar, rather than Osuna, was the killer: physical evidence, eyewitness testimony, and testimony about Reilly’s dog scent identification.

1. Physical Evidence

The prosecution had little physical evidence connecting Aguilar to the crime. On August 9, a police patrol stopped a white Volkswagen Beetle that fit the description of the car from the murder. The police discovered that the vehicle was stolen and impounded it. The police recovered fingerprints from the passenger side of the white Volkswagen. Aguilar’s prints did not match the recovered fingerprints. The police never tested the fingerprints to see if they matched Richard Osuna.

A forensic scientist testified that all of the bullets fired at Guerrero had been discharged from a single firearm. The police never found that firearm. The police did find a live .25 caliber bullet while searching a bedroom at the Amar Road apartment in which Aguilar kept some of his things. Another person was then living in that room. The bullets used in Guerrero’s murder were also .25 caliber, but they were not the same brand as the bullet found in Aguilar’s home. The live bullet found in the bedroom had been “cycled through” a weapon, but a forensic scientist’s efforts to determine whether the bullet had passed through the murder weapon were inconclusive.

2. Eyewitness Testimony

Because there was limited physical evidence, the prosecution relied heavily on eyewitness testimony from seven witnesses. None of the eyewitnesses personally knew Gilbert Aguilar. No eyewitness had been shown a picture of Richard Osuna.

The first witness, Omar Soltero, was a passenger in Guerrero’s car sitting directly behind Guerrero. He ducked down as soon as the gunman approached the passenger side window. During the police investigation, Soltero reported that the gunman was approximately 5'9" tall and about 18-20 years old. At trial, Soltero testified that the gunman was taller than Soltero’s 5'3" height. He also admitted that he only got a “momentary glimpse” of the shooter, that he never saw the shooter’s face, and that he did not believe he could give an accurate description of the shooter outside of the fact that the shooter was male. Soltero never identified Aguilar at any point during the proceedings.

The second witness, Victor Carillo, was also in Guerrero’s car. He was sitting in the middle of the back seat, to the right of Soltero. There is no evidence in the record as to how Carillo initially described the suspect to police investigators. At trial, Carillo estimated that the gunman was “probably about” 5'9" or 5'10", and that the gunman was wearing a white t-shirt and black cap. Carillo admitted that, like Soltero, he could not identify the suspect.

The third witness, Desiree Hoefer, was at the drive-through line at the KFC restaurant when a white Volkswagen Beetle pulled into the restaurant parking lot. Through her rearview mirror, Hoefer watched a Hispanic male get out of the white Volkswagen and walk out of sight. She saw that the male was carrying a gun and feared that she was about to be carjacked. After hearing gunshots, Hoefer decided to leave the KFC. She nearly hit the same male with her car as the male returned to the KFC parking lot. She looked at his face for less than a second.

Hoefer is 5'0". She told the police shortly after the murder that the perpetrator was a Hispanic male around 5'4", 15 to 17 years old, and wearing a baseball cap. A month after the incident, Hoefer told a police sketch artist that the gunman was 5'2" or 5'3" and that he was 16 to 20 years old. Hoefer also described the perpetrator as having a small mouth. After some hesitation, Hoefer identified Aguilar in a six-person photo lineup. The lineup Hoe-fer viewed did not include a photograph of Osuna. At the time of the lineup, Hoefer said that Aguilar’s photograph “looked close” to the suspect, but that she thought Aguilar’s “comple[x]ion was lighter” than the murderer’s complexion. She also thought that the perpetrator was younger than Aguilar. Hoefer said she was not “a hundred percent” sure she had the right person.

At trial, Hoefer identified Aguilar as the person she had picked out of the lineup. Hoefer testified that she believed she had described the perpetrator as 5'8" or 5'10" to the police, though she had actually described him to the police as between 5'2" and 5'4." Hoefer also stated in court that she continued to believe that the shooter “might have been a little shorter” than Aguilar. The prosecution suggested to Hoefer that the crouching position of the shooter might have led her to underestimate the shooter’s height, and Hoefer agreed.

The fourth witness, Victor Jara, was the driver of a car on Amar Road about four cars back from the intersection and one lane closer to the curb than Guerrero’s car. He was stopped at the same traffic light as Guerrero when he heard gunshots. Victor Jara took note of the shooter’s face as he jogged away from Guerrero’s car. He saw the shooter for between two and four seconds.

Victor Jara reported to the police and to the police sketch artist that the suspect was a clean-shaven Hispanic male between 15 and 17 years old with distinctive eyebrows. After speaking with Victor Jara and two other eyewitnesses (Laura Jara and Kevin Feeney) during the initial investigation, Deputy Sheriff Blackmer described the suspect as 5'5" and 130 pounds. Detective Ramirez, the lead case investigator, believed that it was Victor Jara who told Deputy Blackmer that the suspect was 5'5". Victor Jara later told the police sketch artist that the suspect was about 5'6".

Victor Jara identified Aguilar’s photo from the same photo lineup of six persons presented to Hoefer. That lineup did not include a photograph of Osuna. At trial, Victor Jara expressed “absolute certainly]” that Aguilar was the shooter. He also testified at trial that the perpetrator may have been “taller” than the 5'6" defense counsel.

The fifth witness, Laura Jara, was in the passenger seat of the car driven by her husband Victor. She heard gunshots and looked ahead toward Guerrero’s car. Laura saw a male wearing a baseball cap running away from the vehicle. During the initial investigation, Laura Jara described the gunman to the police as a juvenile, 15 to 17 years old. She could not recall whether she had given a height estimate to the police. However, Deputy Sheriff Blackmer described the suspect as 5'5" and 130 pounds after speaking to Laura Jara. Laura identified Aguilar in a six-person photo lineup on the same date as her husband, but was not certain she had identified the perpetrator. She said it “looks like him a lot.”

At trial, Laura Jara stated that the person she saw looked very young and had dark, distinctive eyebrows. Laura Jara also testified that she thought the perpetrator was a “little taller” than the 5'6" defense counsel. Laura Jara did not identify Aguilar as the shooter in court, but she did identify him as the person she had thought “looked like” the shooter in the photo lineup. She also stated that she believed Aguilar had distinctive eyebrows like the perpetrator.

The sixth witness, Kevin Feeney, was putting gas in his car at the station on Amar Road across from the KFC. Feeney saw Guerrero slumped in his vehicle, and then observed an individual running away from the scene. He was approximately 40 yards from the individual he saw running. Deputy Sheriff Blackmer described the suspect as 5'5" and 130 pounds after speaking to Feeney and the Jaras during the police investigation. At trial Feeney testified that the shooter was “tall and slender.” Feeney never identified Aguilar at any point during the proceedings.

Finally, the seventh eyewitness, Rene Valles, was the driver of a car facing eastbound on Amar Road, stopped in the far righthand lane a few cars back from the intersection. Valles heard gunshots. He then saw a Hispanic male run in front of his car and into the KFC parking lot. Valles told the police the suspect was 16 to 21 years old and wearing a white t-shirt and baseball cap. He also stated that he was focused on the gun and saw the suspect’s face for “just a second.” Valles “really wasn’t sure” of the suspect’s height during the police investigation. The police showed Valles a lineup including a picture of Aguilar. Valles was unable to make an identification.

At trial, Valles identified Aguilar as the shooter. This was the first time he had ever identified Aguilar. Valles testified that he had not identified Aguilar previously because he could not see Aguilar’s profile view in the photo lineup. Also for the first time at trial, Valles estimated the perpetrator’s height as 5'9" or 5'10".

To counter the eyewitness testimony, the defense used an expert witness on eyewitness identification. The expert testified that the sooner after an incident an eyewitness describes a suspect, the more accurate that description is likely to be. Further, he testified that “people overestimate the height” of individuals carrying guns, such that “the actual person” being sought “might be shorter than the height estimates.”

Two things are apparent from the foregoing. First, the eyewitnesses’ height, weight, and age estimates during the police investigation more closely resemble Richard Osuna (no more than 5'7" and 16 years old) than Aguilar (no less than 5'11" and 20 years old) at the time of the murder. Desiree Hoefer estimated the suspect to be no taller than 5'4". Victor Jara estimated him to be 5'5" or 5'6". After speaking to the Jaras and Kevin Feeney, Deputy Sheriff Blackmer described the suspect as 5'5". Only Omar Soltero stated to investigators that the shooter was taller, and even he estimated the shooter to be several inches shorter than Aguilar. According to expert testimony, the witnesses would have been expected to overestimate, not underestimate, the height of a man carrying a gun. The eyewitnesses’ weight and age estimates are also more similar to Osuna than to Aguilar. Deputy Sheriff Blackmer reported, after speaking to the eyewitnesses, that the suspect was 130 pounds. That is 30 pounds lighter than Aguilar’s reported weight. Additionally, Hoefer and the Jaras estimated the perpetrator to be substantially younger than Aguilar.

Second, several of the eyewitnesses changed their testimony at trial from the statements they had given during the police investigation. At least two witnesses (Hoefer and Victor Jara) increased their height estimates, one (Valles) testified to a height estimate when he had not made one before, and two others (Laura Jara and Feeney) testified that the perpetrator was tall when they had given no such opinion before. Hoefer, Laura Jara, and Valles also expressed greater certainty in their identification at trial than they had during the investigation. Valles identified Aguilar for the first time at trial, though he had not previously been able to identify Aguilar from a photo lineup.

3. Dog Scent Identification

To supplement the physical evidence and eyewitness testimony, the prosecution put on evidence that a trained scent dog, Reilly, had identified Aguilar’s scent on the white Volkswagen. Shortly after Aguilar was arrested, Officer Joe D’Allura used Reilly to perform a scent comparison test between Aguilar’s scent and the scent found in the white Volkswagen. Scent comparison tests are based on the idea that every person has a unique scent, and that dogs can identify particular scents as belonging to particular objects and persons. A scent transfer unit extracted scent from Aguilar’s clothes and from the impounded vehicle. The extracted scents were then placed in sterile gauze “scent pads.”

Reilly was first given a sample of Aguilar’s scent. Reilly then was led to a lineup of four scent pads, one of which had been collected from the passenger side of the impounded Volkswagen. He was trained to bark if he perceived a match between the sample scent and any of the scent pads. Reilly barked at the third scent pad, signaling a match between the scent pad from the Volkswagen and the scent from Aguilar’s clothes. While there were four scent pads in the line-up, Reilly only reached the third scent pad before he signaled a match. Reilly did not signal a match on the spent casings from the bullets fired at Guerrero.

The scent test occurred on September 4, 2001, over a month after the murder and several weeks after the car was impounded. Aguilar’s scent was taken from Aguilar’s street clothes when he was arrested, and the scent test occurred that same day. It is not clear from the record when the scent was collected from the impounded Volkswagen. On cross examination, Officer D’Allura stated that the scent in the impounded vehicle would have had to have been present “within the last week” in order for it to be “picked up.” On redirect, D’Allura changed his testimony. He stated that “[t]he scent would still be in there as long as [the carl’s not being used by other people and things.”

During the remainder of the trial, the prosecution used Reilly’s scent evidence to support the eyewitnesses’ claim that Aguilar was “in fact, the shooter.” The prosecutor asked Mary Saiz to explain the presence of Aguilar’s scent in the Volkswagen. Saiz responded that perhaps it was her scent the dog identified. She testified that she sometimes wore Aguilar’s clothes and had ridden in Ballesteros’s car. The prosecutor also emphasized the dog scent evidence on crossexamination of the defense’s court-appointed investigator, suggesting that Reilly’s scent match made the case “miraculously strong.” The prosecutor raised the scent evidence yet again while cross-examining the defense’s expert on eyewitness testimony.

The main theme of the prosecutor’s closing summation was that the defense could not “explain why Gilbert Aguilar’s scent was in the same particular vehicle that was responsible for following the victims to the murder location.” The prosecutor discounted the defense’s claim that Osuna and Aguilar looked similar, referring to them disparagingly as “twins.” He then used the scent evidence to cast doubt on the defense’s theory of the case. “[TJhese two twins not only look alike, but they smell alike. So we not only have a lookalike guy, we have a smell-alike guy as well—that being Richard Osuna and Gilbert Aguilar.” More particularly, if “Mr. Osuna is, in fact, the shooter in this case,” the jury “would be accepting an incredible coincidence that Mr. Aguilar’s scent was in the passenger side of that Volkswagen.” Further, it would also be “accepting another incredible coincidence”:

The prosecution argued that its theory of the case did not require the jury to accept such incredible coincidences, not least because “[w]e have the scent for Mr. Aguilar in the seat in which the gunman arose.” The prosecutor concluded: “[I]f you do acquit Mr. Aguilar, [remember] what version of the facts you’re really accepting. You’re accepting all the coincidences that his identical twin and smell-alike person is the one who [is] really responsible. You’re accepting all the incredible coincidences that Gilbert Aguilar just so happened to be everywhere at the time his twin committed this particular offense.”

At the conclusion of the trial, the jury received an instruction on the dog scent evidence stating that “[e]vidence of dog tracking has been received for the purpose of showing ... that the defendant is a perpetrator of the crime of murder.” The jury was also instructed to “consider the training, proficiency, experience, and proven ability, if any, of the dog” in determin ing the weight given to the dog scent evidence.

C. Jury Deliberation and Verdict

After a six-day trial, the jury began deliberating on October 21, 2002. It deliberated for four days. While deliberating, the jurors asked for a reading of the testimony of Desiree Hoefer, Kevin Feeney, and Rene Valles, as well as Victor Jara’s description of the suspect. The jurors also requested a response from the judge to the following question: “Is the fact that the D.A.’s office did not pursue the ‘Richard Osuna’ lead considered evidence? Is it something we should deliberate about?” The court responded, “The state of mind of the investigator or the prosecutor, except as it relates to a bias, intent or other motive to fabricate evidence, is not relevant to the guilt or innocence of the [defendant].”

During the third day of deliberations, Juror No. 2 approached the court about a conversation he had overheard during a break in the trial. He had seen several eyewitnesses, including Victor Jara, Desiree Hoefer, and Rene Valles, apparently talking about the case. The court dismissed Juror No. 2 because the juror could not “evaluate the testimony of those three witnesses just based on what he heard in court.”

Aguilar’s counsel argued for a mistrial, pointing out that the witness testimony on height and age had changed substantially from the police investigation. He argued that Juror No. 2’s report suggested that the witnesses may have “cooked their testimony,” and that a jury could not make a “reasonable” or “accurate determination” about Aguilar’s guilt without that information. The court denied Aguilar’s motion; it found that Aguilar’s counsel had already sufficiently cross-examined the eyewitnesses about the discrepancies in their testimony.

Juror No. 2 was dismissed no earlier than 9:18 am on October 24. With the alternate seated, the reconstituted jury rendered a verdict by 11:23 am that same day. The jury convicted Aguilar on both counts, and he was sentenced to 50 years to life in prison.

D. Brady Evidence

Unbeknownst to defense counsel at the time of trial, the prosecution had stipulated in another case only a few months earlier that Reilly, the scent dog, had made mistaken identifications on two prior occasions. In People v. White, No. BA 212658 (L.A.Cty.Super.Ct. Mar. 19, 2002), the prosecution sought to introduce testimony from Officer Joe D’Allura about a scent identification made by Reilly implicating White. The prosecution stipulated that in November 1997 Reilly had identified two different men as the source of scent on the murder suspect’s shirt, and that in a 2001 case, People v. Bruner, No. BA 216390 (L.A.Cty.Super.Ct.), Reilly had identified as the perpetrator of a crime an individual who was in prison at the time the crime was committed. After an evidentiary hearing on dog scent lineups, the White court ruled that the dog scent procedures Officer D’Allura used with Reilly “were so flawed” that the judge would “not allow the dog scent lineup in.”

After the White case concluded, the Los Angeles County Public Defender wrote a letter to the Los Angeles District Attorney, Steven Cooley, dated March 20, 2002. The County Public Defender detailed the facts in White, and stated:

By the time the prosecution introduced the dog scent evidence in Aguilar’s case, Reilly no longer worked as a scent dog.

The Los Angeles District Attorney’s office prosecuted Aguilar’s case six months after the White case concluded. At Aguilar’s trial, the prosecutor trying the case did not disclose to the defense the earlier mistaken identifications, the stipulation in the White case, or the letter to District Attorney Cooley from the County Public Defender. Aguilar’s trial counsel moved to strike the dog scent evidence for foundation, but not for relevance or admissibility. Counsel has since declared that he would have objected to the evidence’s admissibility had he been aware of Reilly’s history of mistaken identifications, of the White stipulation, or of the letter to Cooley.

E. Appeal and Collateral Attack

Aguilar’s appellate counsel first discovered the exculpatory evidence about Reilly. Aguilar argued to the California Court of Appeal that “the trial court deprived [him] of due process when it denied his motion for a new trial based on evidence revealed by a juror during deliberations.” Aguilar also filed a habeas petition in the Court of Appeal arguing that (1) the prosecution had violated Brady v. Maryland by failing to disclose the exculpatory evidence—that Reilly had a history of misidentifieation, and (2) his counsel was ineffective for failing to challenge the admissibility of the dog scent evidence.

The California Court of Appeal affirmed Aguilar’s conviction and denied his habeas petition in a single disposition. The court determined that the failure to grant Aguilar’s motion for a new trial did not violate his rights. Further, the court found no ineffective assistance of counsel or Brady violation. It noted that the “[u]se of dog-scent evidence in this case was of questionable probity,” and concluded that the evidence was immaterial because, had the jury been given such information, “it is not reasonably probable that ... the result would have been different.” Aguilar petitioned for review in the California Supreme Court, but his petition was denied without comment. The California Court of Appeal decision is the last reasoned statecourt decision. See Ylst v. Nunnemaker, 501 U.S. 797, 805 , 111 S.Ct. 2590 , 115 L.Ed.2d 706 (1991).

Aguilar petitioned for habeas corpus in federal district court in 2006 on all three issues. The magistrate judge hearing the case recommended that the petition be denied. The district court adopted the magistrate judge’s recommendation without comment. It dismissed the action with prejudice and denied Aguilar a certificate of appealability. Aguilar appealed to this court, and we granted a certificate of appealability.

Aguilar argues on appeal that the state court unreasonably applied clearly established Supreme Court law in two ways. First, he argues that the state court unreasonably determined that the prosecutor’s failure to disclose evidence of Reilly’s misidentifications was immaterial under Brady v. Maryland. Second, he argues that the state court unreasonably determined that Aguilar’s inability to put on Juror No. 2 as a witness at his trial did not violate his right to present a complete defense. We agree with Aguilar’s Brady argument. We do not reach his second argument.

III. Brady

The State concedes that Brady is “clearly established Federal law, as deter mined by the Supreme Court of the United States.” 28 U.S.C. § 2254 (d)(1). Thus, the question before us is whether the state court reasonably applied Brady to the facts in Aguilar’s case. A Brady claim has three components. There must be (1) evidence that is exculpatory or impeaching (2) that is suppressed by the state and (3) resulting prejudice. Strickler v. Greene, 527 U.S. 263, 281-82 , 119 S.Ct. 1936 , 144 L.Ed.2d 286 (1999).

1. Exculpatory or Impeaching Evidence

There is no doubt that Reilly’s history of making erroneous scent identifications is exculpatory evidence. “[Ijmpeachment, as well as exculpatory, evidence falls within Brady’s definition of evidence favorable to the accused.” United States v. Marashi, 913 F.2d 724, 732 (9th Cir.1990) (internal quotation marks omitted). “Any evidence that would tend to call the government’s case into doubt is favorable for Brady purposes.” Milke v. Ryan, 711 F.3d 998, 1012 (9th Cir.2013). The evidence not disclosed by the prosecution showed that Reilly had a record of mistaken scent identifications. Because Reilly’s identification tied Aguilar to the white Volkswagen, the undisclosed evidence is unquestionably “favorable for Brady purposes.” Id.

2. Suppression

Aguilar also demonstrated that the prosecution had knowledge of this exculpatory evidence. Aguilar attached to his state-court habeas petition the reporter’s transcript in White, No. BA 212658 (L.A.Cty.Super.Ct.), the case in which the Los Angeles District Attorney’s office stipulated to Reilly’s mistaken scent identifications, and in which the trial judge excluded evidence about Reilly. Aguilar also attached the letter from the Los Angeles County Public Defender to the Los Angeles District Attorney, Steven Cooley, dated six months prior to Aguilar’s trial and specifically stating that, in his view, the record of Reilly’s misidentifications was Brady material.

The State argued to the California Court of Appeal in this case that knowledge of the Brady evidence could not be imputed to the trial prosecutor. For good reason, the State has not made that argument to us. The individual prosecutor at Aguilar’s trial may or may not have possessed Brady information. Joe D’Allura, who testified in Aguilar’s case, was Reilly’s handler in the White case. D’Allura stated at trial, in response to questioning by the trial prosecutor, that he knew about Reilly’s performance in prior gang-related homicides. If the prosecutor in Aguilar’s case was unaware of Reilly’s prior performance, either he was hasty in preparing his witness or D’Allura deliberately concealed from him Reilly’s prior record of misidentifications.

But even if the trial attorney did not himself possess the exculpatory evidence, knowledge of that evidence is imputed to him under Brady . First, each “individual prosecutor has a duty to learn of any favorable evidence known to the others acting on the government’s behalf’ and to disclose it to the other side. Kyles v. Whitley, 514 U.S. 419, 437 , 115 S.Ct. 1555 , 131 L.Ed.2d 490 (1995). This includes evidence held by other prosecutors. The Public Defender’s letter, which put the State on notice that the prior Reilly cases were Brady evidence, was addressed specifically to District Attorney Steven Cooley. The prosecutor in Aguilar’s case was employed by District Attorney Cooley. Knowledge of the Brady evidence therefore is imputed both to Cooley and, by extension, to prosecutors working in his office.

Second, it is clearly established that “Brady suppression occurs when the government fails to turn over even evi dence that is known only to police investigators and not to the prosecutor.” Youngblood v. West Virginia, 547 U.S. 867, 869-70 , 126 S.Ct. 2188 , 165 L.Ed.2d 269 (2006) (per curiam) (quoting Kyles, 514 U.S. at 438 , 115 S.Ct. 1555 ); see also United States v. Blanco, 392 F.3d 382, 393-94 (9th Cir.2004) (“Exculpatory evidence cannot be kept out of the hands of the defense just because the prosecutor does not have it, where an investigating agency does.” (internal quotation marks omitted)). Here, even if the prosecutor’s office had not had the Brady material, Reilly’s handler Joe D’Allura, a “scenting K-9 handler with the Los Angeles County Sheriffs Department,” clearly did. D’Allura’s testimony about the reliability of Reilly’s scent identifications addressed precisely what had been at issue in White, and D’Allura has admitted he had knowledge of previous trials involving Reilly’s misidentifications. Finally, even if D’Allura himself had not been aware of Reilly’s misidentifications, it is enough that other members of the Sheriffs Department were aware of them.

3. Prejudice

The state court based its decision on the third prong of Brady , concluding that Aguilar was not prejudiced by the failure of the prosecution to disclose Reilly’s record of misidentifications. “To determine whether prejudice exists, we look to the materiality of the suppressed evidence.” Jackson v. Brown, 513 F.3d 1057, 1071 (9th Cir.2008). Brady evidence is material if “the favorable evidence could reasonably be taken to put the whole case in such a different light as to undermine confidence in the verdict.” Kyles v. Whitley, 514 U.S. at 435 , 115 S.Ct. 1555 . Aguilar does not need to prove that a different result would have occurred in his case. He needs to show only that the state court unreasonably decided that there was not “a reasonable probability of a different result.” Id. at 434 , 115 S.Ct. 1555 (internal quotation marks omitted).

If the Brady evidence had been presented to the Aguilar court, it is virtually certain that the trial judge would have ruled as did the trial judge in White by excluding the evidence of Reilly’s scent identification. In at least two contemporaneous California state court trials, defense attorneys successfully challenged the admissibility of dog scent lineups. See White, No. BA 212658; People v. Rhoney, No. 94HF0957 (Orange Cty.Super.Ct.1998) (dog scent evidence excluded because it was more prejudicial than probative). One of those cases, White, was before the same court, with the same District Attorney’s office, and involved the same dog. Further, shortly after Aguilar’s trial, the California Court of Appeal found that “evidence of Reilly’s scent identification was admitted in error” in a different criminal proceeding because it was not adequately supported by scientific evidence. People v. Mitchell, 110 Cal.App.4th 772, 790-94 , 2 Cal.Rptr.3d 49 (2003); see also People v. Willis, 115 Cal.App.4th 379, 381 , 9 Cal.Rptr.3d 235 (2004) (finding in a case involving a different dog that “the dog scent evidence was improperly admitted”).

The court in Aguilar’s case already seemed receptive to excluding the dog scent evidence, even without knowledge of the stipulation in the White case. At trial, Aguilar’s counsel objected to that evidence on grounds of foundation, arguing that the control scent pads were unidentified and may have been prepared improperly. The court deferred ruling on the objection but stated, “I am sure it is likely [the prosecution] will be happy to strike the whole thing because [D’Allura] was real honest. He had no idea where in the world the pads came from.... ” Aguilar’s attorney did not renew his foundation objection. Given that the trial court was already con cerned about the admissibility of the dog scent evidence, we are confident that it would have excluded that evidence if the Brady material had been presented to it.

Even if D’Allura’s testimony about Reilly had been admitted into evidence, at the very least a reasonable state court would have concluded that the Brady evidence provided powerful impeachment material. Despite being asked in jury instructions to consider the “proven ability, if any, of the dog” in determining the weight to give to the dog scent evidence, the jurors in Aguilar’s case were presented no evidence about the reliability of Reilly as a scent dog. Thus, they had no reason to question the accuracy of Reilly’s identification of Aguilar.

A reasonable state court would have concluded that there was a reasonable probability that the jury would have reached a different verdict if Reilly’s dog scent identification had not been presented to the jury, or had been impeached by the evidence of Reilly’s earlier misidentifications and the White court stipulations. The gunman’s identity was the only issue in Aguilar’s case. Absent Reilly’s dog scent testimony, there was no corroborating evidence for the shaky eyewitness identifications. There was no forensic evidence, murder weapon, or confession. The prosecution did not tie Guerrero and Aguilar to each other in any way. The only motive given for the killing was the unsubstantiated suggestion that Guerrero had trespassed into the territory of Aguilar’s Puente Street gang, and this theory was suspect given that Guerrero was shot numerous times at close range while his passengers—equally trespassing—were left unharmed. The prosecution’s own gang expert testified that the fact that only Guerrero was shot indicates that he was the intended target, undercutting the government’s theory that this was a gang rivalry shooting. Richard Osuna, a suspect who had a motive to commit a targeted shooting, and who more closely resembled the eyewitness descriptions, had not been investigated.

The state court misstated the nature of the eyewitness testimony, making it appear stronger than it was. The Court of Appeal wrote that “Kevin Feeney ... told police the shooter was ‘tall and slender,’ ” when in fact Feeney first stated that the shooter was “tall and slender” at trial. The eyewitness testimony at trial clearly gave the jury pause, even when reinforced by the unimpeached dog scent evidence. The jurors asked to rehear significant portions of the eyewitness testimony during their deliberation. Given that the identity of the killer was the only question in the ease, “it does not seem possible that the jury would have deliberated ... over several days if the jurors did not have serious questions as to the credibility of the eyewitnesses.” Gibson v. Clanon, 633 F.2d 851 , 855 n. 8 (9th Cir.1980); see also Rhoden v. Rowland, 172 F.3d 633, 637 (9th Cir.1999) (deliberations of nine hours over three days suggests jurors “did not find the case to be clear cut”). The dog scent evidence provided the only corroboration that the eyewitnesses had seen what they testified to at trial, rather than what almost all of them had told the police immediately after the shooting.

The prosecution emphasized the importance of the dog scent identification throughout trial. The State now argues to us that “the dog scent evidence in this case did not prove that Petitioner was in the Volkswagen on the date of the murder” because the scent would not last that long. But the State took a very different position at trial. While defense counsel argued that the dog scent evidence was not probative, the prosecution consistently contended that the evidence corroborated the testimony that Aguilar was “in fact, the shooter.” In his closing argument, the prosecutor said that the relevant question was whose scent was present “on July 25th” (the date of the murder). The jury was told to use the dog scent identification “for the purpose of showing ... that the defendant is a perpetrator of the crime of murder.” The State cannot now argue with a straight face that the evidence upon which it relied so heavily at trial was, in fact, not probative.

The strength of the unimpeached dog scent evidence at trial also forced Aguilar’s counsel to make a strategic concession in his closing argument that Aguilar had sat in the white Volkswagen Beetle. Aguilar’s counsel during closing arguments admitted that the scent evidence shows that “[Aguilar] at some time sat in that car.” He stated, “I don’t doubt that Gilbert sat in that white Volkswagen.” Had Reilly’s dog scent evidence been excluded, or had counsel been able to impeach it using the Brady evidence, counsel would never have made this concession.

In every case where a federal or California state court has found dog tracking or scent identification Brady evidence to be immaterial, the defendant was convicted on evidence stronger than, and independent from, the dog scent identification. In such cases, there was physical evidence to support the conviction, see Epperly v. Booker, 997 F.2d 1, 10 (4th Cir.1993) (bloodstained clothes with head hair resembling Epperly’s hair), a known relationship between the defendant and the victim, see Shererv. Stewart, No. 06-1635-RSM-JPD, 2008 U.S. Dist. LEXIS 118661 at *3, *57-59 (W.D.Wash. June 20, 2008) (history of violence toward victim); Willis, 115 Cal.App.4th at 387 , 9 Cal.Rptr.3d 235 (same), or other evidence corroborating guilt, see Sherer v. Sinclair, 476 Fed.Appx. 443, 443 (9th Cir.2012) (mem.) (“[G]iven the strength of the evidence against petitioner versus the relative weakness of the dog tracking evidence, petitioner has not demonstrated a reasonable probability that disclosure of the allegedly suppressed dog tracking report would have produced a different result.”); People v. Herrera, No. B181092, 2006 WL 2773702 , at *3, *9-10, 2006 Cal.App. Unpub. LEXIS 8638, at *8, *28 (Cal.Ct.App. Sept. 28, 2006) (mem.) (defendant testified and admitted he lied); Mitchell, 110 Cal. App.4th at 794 , 2 Cal.Rptr.3d 49 (admission to third party of guilt); People v. Rivera, No. B166838, 2004 WL 2601335 , at *1-2, *6-7, 2004 Cal.App. Unpub. LEXIS 10517, at *3-4, *20 (Cal.Ct.App. Nov. 17, 2004) (perpetrator chased and arrested minutes after attack). In each of these cases, the evidence to convict was sufficient even absent the dog scent identification because the prosecution had independently proven guilt beyond a reasonable doubt. Here, in contrast, the only evidence in addition to Reilly’s scent identification was shaky eyewitness testimony.

Conclusion

Reilly’s scent evidence was the only evidence at trial linking Aguilar to the getaway car, as well as the only evidence corroborating strikingly weak eyewitness identifications. We conclude that the prosecution’s failure to disclose that Reilly had a history of mistaken identifications violated Brady v. Maryland, and the California courts’ decision to the contrary was an unreasonable application of Brady .

We grant Aguilar’s petition and reverse the district court’s judgment on the Brady claim. We do not reach Aguilar’s other argument. We direct that a conditional writ of habeas corpus issue, requiring the State of California to release Aguilar from custody unless it grants him a new trial to commence within a reasonable period of time to be determined by the district court.

REVERSED and REMANDED.

APPENDICES

[[Image here]]

[[Image here]]

[[Image here]]