United States v. Carlos Caro

UNPUBLISHED

UNITED STATES COURT OF APPEALS FOR THE FOURTH CIRCUIT

No. 16-1

UNITED STATES OF AMERICA,

Plaintiff – Appellee, v.

CARLOS DAVID CARO,

Defendant – Appellant.

Appeal from the United States District Court for the Western District of Virginia, at Abingdon. James P. Jones, District Judge. (1:06-cr-00001-JPJ-1; 1:13-cv-80553-JPJ)

Argued: September 14, 2017 Decided: May 8, 2018

Before GREGORY, Chief Judge, DUNCAN, Circuit Judge, and SHEDD, Senior Circuit Judge.

Affirmed by unpublished opinion. Judge Duncan wrote the opinion, in which Senior Judge Shedd joined. Chief Judge Gregory wrote a separate opinion dissenting in part.

ARGUED: Timothy Michael Gabrielsen, OFFICE OF THE FEDERAL PUBLIC DEFENDER, Tucson, Arizona, for Appellant. Anthony Paul Giorno, OFFICE OF THE UNITED STATES ATTORNEY, Roanoke, Virginia, for Appellee. ON BRIEF: Jon M. Sands, Federal Public Defender, District of Arizona, OFFICE OF THE FEDERAL PUBLIC DEFENDER, Tucson, Arizona; Fay F. Spence, First Assistant Federal Public Defender, Roanoke, Virginia, Brian J. Beck, Assistant Federal Public Defender, OFFICE OF THE FEDERAL PUBLIC DEFENDER, Abingdon, Virginia, for Appellant. Rick A.

Mountcastle, Acting United States Attorney, Roanoke, Virginia, Jean B. Hudson, Assistant United States Attorney, OFFICE OF THE UNITED STATES ATTORNEY, Charlottesville, Virginia, for Appellee.

DUNCAN, Circuit Judge:

A jury convicted Petitioner-Appellant Carlos David Caro of first-degree murder and sentenced him to death. Following a direct appeal, in which this court affirmed his conviction and sentence, Caro filed a 28 U.S.C. § 2255 Motion for Collateral Relief (“§ 2255 motion”) challenging his death sentence on several grounds. The district court denied Caro’s § 2255 motion but granted him permission to appeal whether the government violated his due process rights under Brady v. Maryland, 373 U.S. 83 (1963), by withholding Bureau of Prisons (“BOP”) data on the amount of time that inmates are housed at U.S. Penitentiary, Administrative Maximum Facility (“Florence ADMAX”). 1 The key legal issue in this appeal is whether Caro can relitigate a subsequent, duplicative Brady claim on the basis of data that was available to him at the time the first claim was made. Because there is no legal basis for Caro’s position, we affirm the denial of his § 2255 motion.

In summary, Caro’s Brady claim fails for at least two independent reasons. First, it is procedurally barred because this court previously denied the same claim on direct appeal. Under Brady, the government must disclose evidence that is (1) “favorable to

1 This court also granted Caro a Certificate of Appealability to consider whether his trial counsel’s decision not to proffer mental-health testimony “fell below an objective standard of reasonableness,” Strickland v. Washington, 466 U.S. 668 (1984). After a thorough review of the record, we conclude that it did not. Trial counsel hired a mental health expert, Dr. Keith Caruso, who informed the trial team that Caro’s evaluation revealed damaging information. In light of Caruso’s assessment, it was reasonable for counsel to decide that the potential benefits of mental-health testimony were outweighed by its risks.

[the] accused” and (2) “material either to guilt or to punishment.” Brady, 373 U.S. at 87 (emphases added). This court rejected Caro’s Brady claim on direct appeal because he failed to demonstrate that the requested data was favorable. United States v. Caro, 597 F.3d 608, 619 (4th Cir. 2010). Caro’s § 2255 motion raises the same alleged Brady violation except that it includes previously available statistics, left out of the direct appeal record, from which to argue that the requested BOP data would be favorable. Additional, previously available statistics are insufficient to distinguish the Brady claim raised in Caro’s § 2255 motion from the claim we denied on direct appeal.

As we explain below, the dissent’s argument to the contrary fails as a matter of law. The dissent argues that a Brady claim is only procedurally barred “if it is made with exactly the same evidence and exactly the same arguments raised on direct appeal.” Infra at 48. But it cites no precedent for this proposition and we have found none. In fact, the weight of Supreme Court precedent indicates that previously available evidence is insufficient to revive a claim that was denied on direct appeal, unless that evidence could not reasonably have been included in the direct appeal record. See Sanders v. United States, 373 U.S. 1, 17 (1963); see also Townsend v. Sain, 372 U.S. 293, 317 (1963); Davis v. United States, 417 U.S. 333, 342 (1974). We are therefore unwilling to create out of whole cloth authority so fundamentally at odds with the central purpose of the Antiterrorism and Effective Death Penalty Act (“AEDPA”)--partially codified at § 2255--which is “to reduce delays in the execution of state and federal criminal sentences, particularly in capital cases.” Woodford v. Garceau, 538 U.S. 202, 206 (2003).

Even if Caro’s Brady claim were not procedurally barred, however, it is unavailing. Caro provides no indication that the requested BOP data would have been favorable. Nor does he satisfy Brady’s materiality requirement that there was a “reasonable probability” of a different sentence if the BOP data had been disclosed, see United States v. Bagley, 473 U.S. 667, 682 (1985), because, at best, the requested data would reiterate undisputed information that the jury found unpersuasive.

I.

We begin with a history of Caro’s criminal career, which culminated in the murder of Roberto Sandoval. Next, we discuss the penalty phase of Caro’s murder trial because the evidence adduced during the penalty phase and its effect on the jury’s decision to impose the death penalty are crucial to our Brady analysis. Finally, we recount the procedural history of this case, which is the basis for our conclusion that the Brady claim in Caro’s § 2255 motion is procedurally barred.

A.

Caro was recruited to the drug trade at a young age and has spent most of his adult life incarcerated as a result. When he was twenty-one years old, Caro was convicted of possession of marijuana with intent to distribute and received a twenty-four-month prison sentence. Upon his release, Caro reentered the drug trade. He was promptly arrested and convicted for a second time of possession of marijuana with intent to distribute. The court sentenced Caro to seventy-one months in prison. After completing this sentence, *6 Caro was arrested with five kilograms of cocaine. In 2001, thirty-four-year-old Caro was convicted of his third drug-related offense and sentenced to 360 months imprisonment.

Since then, Caro has become increasingly violent and repeatedly defied the BOP’s efforts to securely house him. In 2002, Caro was incarcerated at the low-security Federal Correctional Institution in Oakdale, Louisiana (“FCI Oakdale”), where he became a leader in one of the most violent prison gangs: the Texas Syndicate. 2 When members of a rival gang were transferred to FCI Oakdale, the prison staff asked Caro to maintain the peace, but he refused to cooperate. Instead, Caro led an attack against the newcomers, beating one of the rival gang members so severely that he was hospitalized. His clothes and boots covered with blood, Caro boasted to the guards: “I don’t give a fuck if they send me to the United States Penitentiary. My brothers follow orders. They know what they’re getting into. It doesn’t even matter if we’re prosecuted. I have [thirty] years to do. I certainly don’t care about myself.” J.A. 321.

2 BOP facilities have various levels of security. From least to most secure, they consist of: federal prison camps, low-security federal correctional institutions, medium- security federal correctional institutions, U.S. penitentiaries and Florence ADMAX. In addition, every BOP facility has a secure housing unit, which serves to temporarily segregate inmates from the facility’s general population for disciplinary reasons or pending transfer to another institution.

As the security level increases, the amount of contact inmates have with each other and with prison staff decreases. In low-security, medium-security and the general population of high-security facilities, inmates perform jobs and engage in recreational activities that bring them into contact with other inmates and prison staff. However, in the secure housing unit of a penitentiary and Florence ADMAX, inmates have restricted access to other people. At Florence ADMAX, for example, inmates spend twenty-three hours of each day in solitary confinement. They spend the remaining hour in an exercise pen where they can communicate with, but cannot touch, other inmates.

Shortly after this attack, the BOP transferred Caro to the high-security U.S. Penitentiary in Lee County, Virginia (“USP Lee”). The additional security, however, did not deter Caro from injuring another inmate. In August 2003, Caro and another member of the Texas Syndicate stabbed a prisoner twenty-nine times with homemade knives. Caro pleaded guilty to conspiracy to commit homicide and was sentenced to twentyseven years in prison.

Caro was subsequently transferred to USP Lee’s secure housing unit. On December 16, 2003, Sandoval was placed in Caro’s cell. The next day, Caro ate Sandoval’s breakfast. When Sandoval objected, Caro wrapped a wet towel around Sandoval’s neck and strangled him to death. After he killed Sandoval, Caro yelled to a passing guard: “[G]et this piece of shit out of here.” United States v. Caro, 102 F. Supp. 3d 813, 824 (W.D. Va. 2015). The guard asked Caro if Sandoval was alive and Caro responded, “No. At this time he’s stinking up the room, get him out.” Id. The BOP transferred Caro to Florence ADMAX pending his trial for Sandoval’s murder.

B.

On February 1, 2007, a jury convicted Caro of first-degree murder for killing Sandoval. The trial advanced to the penalty phase, which proceeded in two stages. First, the jury determined that Caro was eligible to receive the death penalty under 18 U.S.C. § 3591. Second, the jury found that the aggravating factors established at trial sufficiently outweighed the mitigating factors to justify a death sentence.

1.

Section 3591 provides that the death penalty is only available for defendants who have been convicted of a capital offense and for whom the government has proven at least one of the statutory aggravating factors provided in 18 U.S.C. § 3592(c). Here, the jury found that Caro was eligible for the death penalty because first-degree murder is a capital offense and the government proved two statutory aggravating factors: (1) Caro was previously convicted of two offenses involving distribution of illegal drugs committed on different occasions and punishable by imprisonment for over one year, see 18 U.S.C. § 3592(c)(10); and (2) Caro was previously convicted of a federal drug offense punishable by five or more years, see 18 U.S.C. § 3592(c)(12).

2.

In the second stage of the penalty phase, the jury was asked to determine whether the aggravating factors of Caro’s case--including ones not provided by statute--sufficiently outweighed the mitigating factors to justify a death sentence. The government alleged three non-statutory aggravating factors. At issue here is the government’s allegation that Caro would pose a danger to inmates and BOP staff if he was sentenced to life in prison. To counter the government’s future-dangerousness factor, Caro alleged that he would spend the rest of his life in a secure institution and would grow less violent with age.

The second stage of the penalty phase progressed in four parts that are significant to this appeal: (a) a discovery dispute over BOP statistics regarding the average length of *9 time inmates spend at Florence ADMAX; (b) testimony from Caro’s expert witness that the BOP could prevent Caro from assaulting other inmates and prison staff; (c) testimony from the government’s witness that the BOP could not guarantee that inmates and guards would be safe from Caro; and (d) the jury’s determination that the balance of aggravating factors to mitigating factors justified imposition of the death penalty.

a.

The defense hired Dr. Mark Cunningham to testify that the BOP could prevent Caro from hurting other inmates and prison staff by housing him at Florence ADMAX until he aged out of violence. To prepare Cunningham’s testimony, Caro requested data on the “median length of stay, [] range of length of stay, and [] standard deviation of the distribution of length of stay at Florence ADMAX for all inmates since it was opened in 1994 to the present time.” J.A. 19. After the government failed to voluntarily disclose the requested information, Caro moved to compel disclosure.

At first, a magistrate judge determined that Brady required the government to disclose the requested information. But the government successfully appealed this ruling to the district court. It argued that Brady did not compel disclosure because there was no indication that the requested data existed and, even if it did exist, there was no indication that the data would be favorable to Caro. In a supporting affidavit, Tomas J. Gomez, the Unit Manager at Florence ADMAX, stated that BOP “does not maintain rosters that would allow the defendants to identify every single inmate who was housed at a particular institution during the relevant time period, nor does the computer system allow *10 such rosters to be retrieved after 30 days.” J.A. 113. In other words, the BOP does not maintain a database of all the inmates ever housed at a particular institution. Instead, it keeps an up-to-date list of the inmates currently housed at each institution.

The district court reversed the magistrate judge’s ruling because Caro failed to demonstrate that the requested BOP data would be favorable. The court explained, “While [Caro] obviously hopes . . . the information requested here will support [Cunningham’s] opinion, there is no indication . . . that it will do so . . . .” J.A. 149.

b.

Caro nevertheless called Cunningham as an expert witness in prison violence and prison security measures. Cunningham testified that Caro would be unable to assault another BOP inmate or guard if sentenced to life in prison because the BOP would incarcerate him at Florence ADMAX, where strict security measures would virtually eliminate Caro’s contact with other people. Cunningham stated that at Florence ADMAX inmates spend twenty-three hours per day in solitary confinement and the remaining hour in outdoor pens that allow communication between the inmates but prevent physical contact. He also explained that Caro would be restrained during any interaction with BOP staff. Specifically, Cunningham testified that inmates at Florence ADMAX never leave their cells without a two-guard escort. One officer holds the inmate’s handcuffs while the other carries a baton in case the inmate turns violent.

Cunningham explained that his opinion on Caro’s future dangerousness was based on his belief that the BOP could prevent Caro from assaulting other people through *11 restrictive security measures, not on an assessment that Caro would voluntarily refrain from violence. In fact, Cunningham stated that “in a U.S. penitentiary [Caro posed a] grave risk of serious violence” and would continue to pose that risk for “five to ten years . . . and perhaps much further out.” J.A. 764 (emphasis added).

Cunningham predicted that the BOP would keep Caro at Florence ADMAX until Caro ceased to exhibit violent tendencies, no matter how long this took. He based his prediction on anecdotal examples of particularly dangerous inmates, such as Al Qaeda terrorists and the “Unabomber” Theodore Kaczynski, whom the BOP assigned to Florence ADMAX without the expectation that they would be transferred back to a less secure institution in the foreseeable future. Cunningham nevertheless acknowledged that, according to policy, the BOP did not permanently assign inmates to Florence ADMAX and aimed to transfer inmates to less secure facilities through a “step-down” program, which took an average of five years to complete.

Finally, Cunningham testified that security breaches allowing an inmate to assault another prisoner or guard occur at every BOP facility, including Florence ADMAX. He acknowledged that in 2005 two Florence ADMAX inmates beat another prisoner to death. One month later, a second inmate was murdered. He also acknowledged that security failures at USP Lee had permitted Caro to communicate with members of the Texas Syndicate in code. Caro might exploit this failure to order fellow gang members to carry out assaults on his behalf, even though the restrictive measures at Florence ADMAX prevented him from committing the acts himself.

c.

On rebuttal, the government called Gregory Hershberger, who formerly served as the warden of Florence ADMAX. Hershberger testified that Florence ADMAX “is designed to house those individuals who can’t function in open [U.S.] penitentiary settings . . . . [But] they still go to [Florence ADMAX with] the expectation [] that they will return to an open population after a period of time.” J.A. 834–35.

He then explained the process for reintegrating inmates into a U.S. penitentiary. Inmates that are assigned to Florence ADMAX are typically placed in the facility’s general population unit. If an inmate does not have any disciplinary problems for twelve months, he is moved to the immediate unit and then to the transitional unit. Once he completes a year in each unit without any disciplinary issues, the inmate is transferred back to a U.S. penitentiary. According to Hershberger, the step-down program takes at least three years to complete. 3

3 In closing argument, the government stated, “You heard the testimony of Mr. Hershberger. Three years, three years for him to be stepped down out of [Florence ADMAX] and into a [U.S. penitentiary].” J.A. 924. This statement misrepresented Hershberger’s testimony that the step-down program takes a minimum of three years to complete. While we disapprove of the government’s misrepresentation, Caro does not challenge the statement in this appeal. In fact, Caro does not even suggest that the government misrepresented Hershberger’s testimony. He merely invokes the government’s closing argument to support his position that the requested BOP data is material because it would likely disprove Hershberger’s testimony, which the government emphasized during its closing argument.

Hershberger also testified that especially dangerous inmates are not placed directly into the step-down program. Instead, they are assigned to Florence ADMAX’s control unit. These inmates are evaluated monthly until the prison staff determines that they can be safely transferred to the general population. Hershberger emphasized, however, that the control unit and the general population unit share the same goal: “to return the inmate to an open population [in a U.S. penitentiary].” J.A. 843–44.

Hershberger also stated that, even if Caro were placed in the control unit, he would have regular contact with prison staff at Florence ADMAX and access to materials from which to fashion homemade weapons. Finally, Hershberger told the jury that potential security lapses might allow Caro to send coded messages instructing his associates in the Texas Syndicate to carry out murders on his behalf.

d.

After considering all of the evidence, including the future-dangerousness testimony recounted above, the jury sentenced Caro to death. It unanimously found that Caro was “likely to commit acts of violence against other inmates or staff within the federal prison system if imprisoned for life without possibility of release.” J.A. 881. Moreover, no juror found that Caro was “less likely, as he age[d], to engage in violent behavior.” J.A. 885.

C.

On direct appeal, Caro challenged his conviction and sentence on several grounds. In relevant part, Caro challenged the district court’s denial of his motions to compel disclosure of the BOP data arguing that the district court’s ruling was “a violation of Brady’s constitutional commands.” Appellant’s Opening Br. at 66 n.45, United States v. Caro, 597 F.3d 608 (4th Cir. 2010), (No. 7-5). On March 17, 2010, this court denied the Brady claim because Caro could only speculate as to what the requested information might reveal and, thus, could not show that the undisclosed data was favorable to his case. Caro, 597 F.3d at 619. After disposing of his other grounds for appeal, the court affirmed Caro’s conviction and death sentence.

The dissent objected to Caro’s death sentence, arguing that the statutory aggravating factors provided by § 3592(c)(10) and § 3592(c)(12) were unconstitutional because they target nonviolent drug offenders. But the dissent “concur[red] with the rest of the Court’s analysis,” id. at 636 n.1 (Gregory, J., dissenting), including our rejection of Caro’s Brady claim.

D.

Caro then filed the § 2255 Motion for Collateral Review that is the subject of this appeal. Once again, Caro argued that the government violated his right to due process under Brady by withholding BOP data on the length of time that inmates spend at Florence ADMAX before they are assigned to a less secure facility. However, the § 2255 motion included statistics--absent from the direct appeal record--that identified 155 *15 inmates who spent more than three years at Florence ADMAX, sixty-three inmates who spent more than five years there and twenty-five inmates who spent over ten years there.

These statistics, or at least similar ones, were available to Caro during his trial and direct appeal, because they were compiled from publicly available sources, such as an informal survey sent to Florence ADMAX inmates, the BOP’s inmate locator website, PACER, the Federal Death Penalty Resource Counsel website, documents received from a Freedom of Information Act request, and internet searches of newspaper articles containing names of inmates known to be at Florence ADMAX. In his § 2255 motion, Caro argued that these figures were evidence of favorability because they demonstrated that the requested BOP data would have supported Cunningham’s testimony that Caro would remain at Florence ADMAX until he aged out of violence, regardless of how long that took.

Without holding an evidentiary hearing, the district court dismissed Caro’s § 2255 motion on two alternative grounds. First, it determined that Caro’s claim was procedurally barred because a petitioner cannot relitigate issues on collateral review that were previously decided on direct appeal. Additional evidence supporting the same claim does not make the claim new.

Alternatively, the district court dismissed Caro’s Brady claim on the merits, holding that the requested BOP data did not create a “reasonable probability” of a different sentence because that data was cumulative of testimony proffered by both sides that inmates routinely spend more than the average five years at Florence ADMAX. The district court also found that the requested data would not have affected the jury’s future *16 dangerousness determination because the jury found that Caro would remain dangerous for the rest of his life and there was no indication that the requested BOP data would show that, contrary to BOP policy, Caro would be permanently assigned to Florence ADMAX. This appeal followed.

II.

We review de novo the district court’s legal conclusion that the Brady claim alleged in Caro’s § 2255 motion was procedurally barred. See United States v. Linder, 552 F.3d 391, 395 (4th Cir. 2009). The district court’s determination that the undisclosed BOP data was not material to Caro’s punishment raises a mixed question of law and fact that we also review de novo. See Summers v. Dretke, 431 F.3d 861, 878 (5th Cir. 2005). Because the district court denied the § 2255 motion without an evidentiary hearing, we review the facts in the light most favorable to Caro. United States v. Poindexter, 492 F.3d 263, 267 (4th Cir. 2007) (citing United States v. Nicholson, 475 F.3d 241, 248 (4th Cir. 2007)).

As explained below, we affirm the district court on alternative grounds. First, we hold that the Brady claim alleged in Caro’s § 2255 motion is procedurally barred because Caro raised an identical claim on direct appeal. Alternatively, we hold that Caro’s Brady claim lacks merit because Caro did not show that the requested BOP data would be favorable or material.

A.

To begin, the Brady claim raised in Caro’s § 2255 motion is procedurally barred. It is well-settled that a petitioner cannot “circumvent a proper ruling . . . on direct appeal by re-raising the same challenge in a § 2255 motion.” United States v. Dyess, 730 F.3d 354, 360 (4th Cir. 2013) (internal quotation marks omitted) (quoting Linder, 552 F.3d at 396). Because Caro’s § 2255 motion raised the same Brady claim we previously rejected on direct appeal, we are compelled to hold that Caro is barred from relitigating that claim.

On direct appeal, Caro argued that the district court’s denial of his motion to compel disclosure of BOP data regarding the length of time inmates are housed at Florence ADMAX was “a violation of Brady’s constitutional commands.” Appellant’s Opening Br. at 66 n.45, United States v. Caro, 597 F.3d 608 (4th Cir. 2010), (No. 7-5). We rejected this argument. 4 Caro, 597 F.3d at 619. In his § 2255 motion, Caro raised the same claim arguing that “the Government violated [his] constitutional rights under Brady . . . by withholding material exculpatory and impeachment evidence that the BOP

4 Caro argues that, on direct appeal, he did not intend for us to decide the merits of his Brady claim. Instead, he raised a Brady challenge intending for us to remand the case so that the district court could determine whether the government withheld Brady evidence. Therefore, this court should not have addressed the merits of his Brady claim on direct appeal. This argument borders on the bizarre. This court has the authority to decide whether a claim should be resolved on the merits or remanded for further proceedings. See 28 U.S.C. § 2106 (“The Supreme Court or any other court of appellate jurisdiction may affirm, modify, vacate, set aside or reverse any judgment, decree, or order of a court lawfully brought before it for review, and may remand the cause and direct the entry of such appropriate judgment, decree, or order, or require such further proceedings to be had as may be just under the circumstances.”). This authority is cabined only by the law, not the litigants’ desires.

has housed many inmates at [Florence ADMAX] and its predecessor prison . . . for more than three years.” J.A. 1168.

Caro’s § 2255 motion includes statistics that were absent from the direct appeal record, but this additional information does not suffice to make the Brady claim raised in his § 2255 motion different from the claim we rejected on direct appeal. The presentation of additional, previously available evidence to support the same claim is insufficient to make an old claim new. See Small v. Hunt, 98 F.3d 789, 798 (4th Cir. 1996) (holding that a court may not amend a judgment to account for additional evidence if the movant fails to provide a legitimate justification for not presenting the evidence during the earlier proceeding); see also Restatement (Second) of Judgments § 25, cmt. b (“A mere shift in the evidence . . . will not suffice to make a new claim avoiding the preclusive effect of the judgment”).

A different rule would contravene Supreme Court precedent and AEDPA’s purpose. In Sanders v. United States, the Supreme Court held that a petitioner is entitled to an evidentiary hearing on a second or successive § 2255 motion if he demonstrates that “the evidentiary hearing on the prior [motion] was not full and fair.” Sanders, 373 U.S. at 17. The Court explained that the criteria for what constitutes a full and fair hearing was set out in Townsend v. Sain, which stated that “newly discovered evidence” could provide the basis for a new hearing if the evidence “could not reasonably have been presented to the [previous] trier of facts.” See id. at 13 (citing Townsend, 372 U.S. at 317). The same rule applies to cases like Caro’s, where “the prior determination was made on direct appeal from the applicant’s conviction, instead of in an earlier § 2255 *19 proceeding.” See Davis, 417 U.S. at 342; see also Withrow v. Williams, 507 U.S. 680, 721 (1993) (Scalia, J., concurring in part and concurring in judgment) (“[A] prior opportunity for full and fair litigation is normally dispositive of a federal prisoner’s habeas claim. If the claim was raised and rejected on direct review, the habeas court will not readjudicate it absent countervailing equitable considerations.”). Together, these cases establish that evidence proffered for the first time on collateral review is insufficient to overcome the procedural bar against relitigating claims that were denied on direct appeal, unless that evidence could not reasonably have been included in the direct appeal record. See United States v. Palumbo, 608 F.2d 529, 533 (3d Cir. 1979) (“[I]n the absence of newly discovered evidence that could not reasonably have been presented at the original trial . . . a § 2255 petitioner may not relitigate issues that were adjudicated at his original trial and on direct appeal.”); see also Morgan v. United States, 438 F.2d 291, 293 (5th Cir. 1971) (“Where newly-discovered evidence is alleged [in support of a § 2255 motion], it must be such as could not reasonably have been presented to the trier of facts.”). In addition, allowing a petitioner to endlessly revive old claims based on evidence that he could have previously proffered but chose not to, would obstruct the central purpose of AEDPA “to reduce delays in the execution of state and federal criminal sentences, particularly in capital cases.” See Woodford, 538 U.S. at 206.

In this case, Caro could have reasonably proffered the new statistics to support his Brady claim at trial or on direct appeal because those figures were compiled from public sources that he could have accessed at any time. The statistics are consequently insufficient to overcome the procedural bar at issue.

The dissent disagrees with our conclusion for two main reasons. First, it posits that a § 2255 Brady claim is not procedurally barred unless it “is made with exactly the same evidence and exactly the same arguments raised on direct appeal.” Infra at 48. According to the dissent, it should not matter whether the newly proffered evidence was previously available to the petitioner. The dissent, however, cites no precedent for its proposed rule and we have found none. Nor can we discern a rationale under AEDPA for a rule that would impose no limit on serial, marginally reformulated Brady claims based on evidence petitioner could have, but chose not to, proffer on direct appeal.

Second, the dissent takes issue with our conclusion that the newly proffered evidence supporting Caro’s § 2255 motion was previously available because some of that evidence was collected after Caro’s direct appeal. In particular, the dissent cites a survey that Jeanne Dvorak conducted by mailing questionnaires to the inmates at Florence ADMAX several months after Caro’s direct appeal was decided. The dissent’s argument that Dvorak’s survey was previously unavailable is beside the point. The underlying data was available to Caro during his direct appeal. Nothing in the record suggests that his attorneys were prevented from mailing similar questionnaires. An absence of diligence does not render the data previously unavailable.

B.

Even if Caro’s Brady claim were not procedurally barred, it would fail on the merits. Under Brady, the prosecution’s failure to disclose evidence upon request violates due process if the requested evidence is (1) “favorable to [the] accused” and (2) “material *21 either to guilt or to punishment.” Brady, 373 U.S. at 87. Caro’s Brady claim clears neither hurdle.

1.

First, there is no indication that the requested BOP evidence would be favorable to Caro. At trial, Caro sought to prove that he would not assault another inmate or member of the BOP staff if he were sentenced to life in prison because the BOP would house him at Florence ADMAX until he aged out of violence. The government countered by offering evidence that Caro would remain dangerous for the rest of his life but, pursuant to BOP policy, Caro could not be permanently assigned to Florence ADMAX. To disprove the government’s argument, Caro requested BOP data on the length of time inmates spend at that institution.

In this appeal, Caro identifies 155 inmates who have spent more than three years at Florence ADMAX, sixty-three inmates who have spent more than five years there and twenty-five inmates who have spent over ten years there. According to Caro, these figures show that the requested BOP data would have been favorable to the proposed mitigating factor that the BOP would house him at Florence ADMAX until he aged out of violence. We find that these statistics do not support such a conclusion.

The statistics are not relevant, let alone favorable, to the mitigating factor at issue. The jury rejected Caro’s allegation that he would become less violent with age. Accordingly, the requested data would only be relevant to the jury’s future dangerousness finding if the data showed that the BOP would likely house Caro at Florence ADMAX *22 for the rest of his life. The statistics Caro provides in his § 2255 motion reflect that some inmates spend a long time at Florence ADMAX but they do not identify any inmate that has served a full life sentence there. This is consistent with Cunningham and Hershberger’s trial testimony that the BOP does not permanently assign inmates to Florence ADMAX.

For these reasons, Caro has failed to demonstrate that the requested BOP data would be favorable to his sentence.

2.

Caro’s Brady claim also fails to satisfy the materiality element. Evidence is “material” if “there exists a ‘reasonable probability’ that had the evidence been disclosed the result at trial would have been different.” Wood v. Bartholomew, 516 U.S. 1, 5 (1995) (per curiam). A “reasonable probability” exists when “the likelihood of a different result is great enough to undermine[ ] confidence in the outcome of the trial.” Smith v. Cain, 565 U.S. 73, 75 (2012) (internal quotation marks omitted) (quoting Kyles v. Whitley, 514 U.S. 419, 434 (1995)). On the other hand, “[t]he mere possibility that an item of undisclosed information . . . might have affected the outcome of the trial, does not establish ‘materiality’ . . . .” United States v. Agurs, 427 U.S. 97, 109–10 (1976).

In Caro’s case, the BOP records are material if there is a reasonable probability that their disclosure would have persuaded at least one juror to vote for a life sentence. See 18 U.S.C. § 3593(e) (“[T]he jury by unanimous vote . . . shall recommend whether *23 the defendant should be sentenced to death . . . .”). Caro argues that the requested BOP data would have undermined the jury’s finding that he would commit future acts of violence if sentenced to life in prison because that data would have shown that he would be housed at Florence ADMAX until he aged out of violence. During the sentencing phase of his trial, however, none of the jurors found that Caro would grow less violent with age. 5 Accordingly, even if we assume that the jury was convinced that Florence ADMAX could safely house Caro, the requested BOP data would only have affected the jury’s future-dangerousness determination if it showed that Caro would remain at Florence ADMAX for the rest of his life. Caro has not demonstrated that the data would support such a conclusion.

At trial, the parties did not dispute that some inmates take longer than the average five years to complete the step-down program. However, Cunningham and Hershberger both testified that the BOP does not permanently assign inmates to Florence ADMAX as a matter of policy, because the objective of the institution is to rehabilitate prisoners so they can be safely transferred to less secure facilities. Moreover, Caro’s attorney stated during closing arguments, “[E]ven when you’re talking about the super maximum facility in the Federal Bureau of Prisons, where they send the worst of the worst offenders, . . .

5 According to the dissent, our materiality analysis cannot rely on the jury’s refusal to find that Caro would become less violent as he aged because the requested BOP data could have undermined that conclusion. We are not persuaded. Caro’s Brady motion requested information on the BOP’s ability to incapacitate and control him. It did not seek any data on the likelihood of Caro’s rehabilitation. Accordingly, we conclude that the requested BOP data would have had no bearing on the jury’s refusal to find that Caro would age out of violence.

they still believe in the power of redemption, that Step Down Unit program is proof of that.” J.A. 962. At best, then, the requested BOP data--which Caro posits would show that some inmates remain at Florence ADMAX longer than the average five years--would merely reiterate undisputed information that the jurors found was outweighed by the BOP’s policy against permanently assigning inmates to Florence ADMAX and its goal of transferring inmates to less secure institutions. Therefore, Caro has failed to demonstrate a “reasonable probability” that the requested data would have affected his sentence. 6

In addition, Caro failed to demonstrate beyond a “mere possibility” that the statistical evidence he requested even existed. Indeed, there is unrebutted evidence in the record that the BOP does not maintain a database of all the inmates ever housed at a particular institution. See J.A. 113. The argument that data, which the government did not possess in any accessible format, would have changed the result at trial is highly speculative, see United States v. Wolf, 860 F.3d 175, 193 (4th Cir. 2017) (“The government did not have this evidence until after [the defendant’s] trial ended. Therefore there was no Brady violation.”), and suggests that Caro was attempting to engage in the

6 The dissent asserts that our analysis is too narrow because it focuses exclusively on the effect of data showing the amount of time that inmates have served at Florence ADMAX. According to the dissent, we should consider the cumulative effect of all the information Caro requested in his pretrial Brady motion, which included statistics about the frequency of violence at Florence ADMAX and the disciplinary records of inmates at the facility. We disagree. Caro has not challenged the government’s failure to turn over all of the information requested in his pretrial Brady motion. He merely argues that he “was denied his right to due process of law under the Fifth Amendment where the Government withheld Bureau of Prisons’ data on the maximum length of time inmates can be housed at ADX Florence.” Appellant’s Opening Br. at 22 (emphasis added).

type of fishing expedition Brady’s materiality requirement seeks to foreclose, see Caro, 597 F.3d at 619 (“Brady requests cannot be used as discovery devices.”).

For these reasons, we are compelled to hold that Caro failed to satisfy Brady’s requirement that the requested evidence create a “reasonable probability” of a different result at trial.

III.

In summary, the Brady claim alleged in Caro’s § 2255 motion was procedurally barred because it was previously denied on direct appeal. Even if the claim was not barred, it lacked merit because the requested evidence was not favorable or material to Caro’s sentence. For these reasons, the judgment of the district court is

AFFIRMED.

GREGORY, Chief Judge, dissenting in part:

At the heart of this collateral challenge to a capital sentence is a single question: should the jury have been allowed to hear the truth about how Carlos David Caro could be incarcerated before deciding if he was too dangerous to remain alive? The Bureau of Prisons (BOP) certainly does not lack the means to securely house highly dangerous inmates; indeed, the BOP’s highest security prison, Administrative Maximum United States Penitentiary in Florence, Colorado (Florence ADMAX or ADX), currently holds Unabomber Ted Kaczynski, Atlanta Olympics bomber Eric Rudolph, 9/11 conspirator Zacarias Moussaoui, Oklahoma City bomber Terry Nichols, underwear bomber Umar Farouk Abdulmutallab, and Thomas Silverstein, who killed two inmates and a BOP guard over three decades ago. 1 At trial, Caro argued that the BOP can securely house him as well, negating the need to put him to death. The Government disagreed, claiming that the BOP had no facility that could hold Caro securely and therefore his future dangerousness justified the death penalty.

1 BOP, Inmate Locator, https://www.bop.gov/inmateloc/ (last visited Apr. 18, 2018) (saved as ECF opinion attachment 1); J.A. 700. See generally Mark Binelli, Inside America’s Toughest Federal Prison, N.Y. Times Mag. (Mar. 25, 2015), https://www.nytimes.com/2015/03/29/magazine/inside-americas-toughest-federal-prison .html (saved as ECF opinion attachment 2). Known as the “Alcatraz of the Rockies,” Florence ADMAX is “a place to incarcerate the worst, most unredeemable class of criminal—‘a very small subset of the inmate population who show,’ in the words of Norman Carlson, the former director of the Federal Bureau of Prisons, ‘absolutely no concern for human life.’” Id. Another former warden has described Florence ADMAX as “a clean version of hell.” Laura Rovner & Jeanne Theoharis, Preferring Order to Justice, 61 Am. U. L. Rev. 1331, 1404 (2012) (citation omitted).

To support his contention, Caro invoked Brady v. Maryland, 373 U.S. 83 (1963), before his trial, diligently seeking data from the BOP about other inmate assaults and murders in the prison system, instances of violence in Florence ADMAX, and the length of time inmates are actually held at Florence ADMAX. But the Government successfully fought to keep this information hidden and then told a jury that Caro would only be held at Florence ADMAX temporarily because of its three-year step down program. That jury then sentenced Caro to death. Eight years ago, we affirmed the denial of Caro’s Brady claim based only on the record developed at trial, concluding that he had failed to show that the requested data would have been favorable to him. United States v. Caro, 597 F.3d 608, 619 (4th Cir. 2010).

Caro now returns to this Court with data vindicating his prior suppositions: the BOP routinely houses dangerous inmates—including specific inmates who have committed particularly violent homicides while in the BOP—at Florence ADMAX well beyond the aspirational three years suggested by the step-down program. The majority and I do not differ on the law: a defendant cannot use her collateral attack to relitigate issues that were “fully considered” on direct appeal, Boeckenhaupt v. United States, 537 F.2d 1182, 1183 (4th Cir. 1976) (per curiam), and a Brady claim has been “fully considered” if the defendant presents the exact same arguments and evidence on collateral review. But we do differ on the facts: Caro has presented new evidence proving that the data he requested pretrial is materially favorable to him.

Viewing Caro’s § 2255 petition in light of the full record, his Brady challenge is both procedurally sound and meritorious. Because we cannot have “fully considered” a *28 Brady claim when the defendant presents new evidence on collateral review, Caro is not barred by the Boeckenhaupt doctrine and is free to bring his claims now. And because he has demonstrated that the suppressed data is favorable and material, he has made out a Brady violation. Because the majority holds otherwise, I respectfully dissent in part. 2

I.

A.

On December 17, 2003, Caro killed Roberto Sandoval, his temporary cell mate at United States Penitentiary (USP) Lee in Jonesville, Virginia. After the murder, Caro was single-celled 3 in the Secure Housing Unit (SHU) at USP Lee for almost two years before being moved in November 2005 to Florence ADMAX. Caro remained in Florence ADMAX until March 2006, when he was moved between USP Lee and a local jail in preparation for trial, again single-celled. He committed no further acts of violence.

In January 2006, a grand jury indicted Caro for Sandoval’s murder and the Government filed a notice of intent to seek the death penalty under the Federal Death Penalty Act (FDPA), 18 U.S.C. §§ 3591–99. Under the FDPA, a defendant can be sentenced to death only if a unanimous jury finds that he is eligible for the penalty (the

2 I join the majority in concluding that Caro has not presented a colorable ineffective assistance of counsel claim. Ante 3 n.1. 3 To “single cell” an inmate is to place him alone in a cell and give him only limited contact with other inmates or prison officials. Single-celled inmates are handcuffed anytime they are moved and take only an hour or two of exercise a day, typically in isolation.

“eligibility phase”) and selects the death penalty as the justified punishment (the “selection phase”). § 3591. After a four-day trial in February 2007, during which the defense conceded that Caro had killed Sandoval, the jury unanimously found Caro guilty of first degree murder and eligible for the death penalty.

Caro’s future hinged on the selection phase: After a hearing in which both sides presented testimony, the jury had to decide whether the death penalty was justified by weighing statutory and non-statutory aggravating factors proved by the Government against mitigating factors proved by the defense. § 3593(c)–(e). The Government alleged three non-statutory aggravating factors but focused almost exclusively on one: Caro’s future dangerousness to other people, including other inmates. 4 In response, Caro presented twenty-two mitigating factors, but focused primarily on undercutting the Government’s allegations of future dangerousness. 5 The crux of the selection phase was the competing testimonies of clinical and forensic psychologist Dr. Mark D. Cunningham, who testified that Florence ADMAX could securely house Caro for as long as necessary, and retired Florence ADMAX warden Gregory L. Hershberger, who

4 The Government’s other non-statutory aggravating factors were the impact of the murder on Sandoval’s friends and family and Caro’s lack of remorse. Sandoval’s daughter testified about the impact his murder had on her and her family. No witness testified about Caro’s remorse or lack thereof. 5 Caro also presented the testimony of five family members and one teacher, who testified about his difficult childhood and his overall character, and a second expert, who provided general information about Florence ADMAX, and opined that the BOP has the ability to control Caro in the long-term.

testified in rebuttal that Florence ADMAX aimed to send inmates back to lower security prisons.

To understand their testimony, I must take a step back and examine the information Caro had attempted to acquire a year earlier under Brady.

B.

A year before trial, the Government told Caro and the court that it intended to prove future dangerousness during the selection phase. To rebut the Government’s anticipated argument, Caro requested data about inmates housed at Florence ADMAX and inmate killings within the BOP, intending to show that the BOP could securely house him just as it had other dangerous inmates. Caro filed four different motions for this data, including one under Brady. In his Brady motion, Caro requested: movement sheets, investigative reports, and histories for all inmates who have killed another inmate within the BOP in the last 20 years; records on all inmates in Florence ADMAX’s control unit, including records of assaultive conduct; disciplinary records on all inmates at Florence ADMAX; records on frequency and level of violence at each security level of Florence ADMAX; records showing how long inmates are kept at Florence ADMAX; and records showing what caused inmates to be transferred to Florence ADMAX and which inmates are still there (the “BOP data”). Specifically, he requested:

A. Data from Florence ADMAX Colorado showing 1. median length of stay, 2. range of length of stay, and 3. standard deviation of the distribution of length of stay at Florence ADMAX for all inmates since it was opened in 1994 to the present time.

B. Data from Florence ADMAX Colorado showing how many inmates who were admitted to Florence ADMAX from the date of its

opening to the present time continue to be confined there, broken down by name, register number, offense conduct that caused them to be transferred to Florence ADMAX, and Security Threat Group classification.

C. Movement sheets from the Bureau of Prisons on every inmate currently at Florence ADMAX who has killed another inmate within the Bureau of Prisons within the last twenty years.

D. Investigative reports on all inmate homicides at Florence ADMAX since it was opened including any “after action reports” indicating any operational or institutional changes in response to each killing and final memorandum from SIS to the Warden of the institution regarding each killing.

E. Regarding each inmate of the above (subparagraph D.) involved in an inmate killing within Florence ADMAX since it opened, the respective inmate’s “Chronological Disciplinary Record” and “Inmate History ADM-REL” and/or movement sheets within the Bureau of Prisons.

F. Records of any assaultive conduct by an inmate in the Control Unit at Florence ADMAX from the time it opened to the present date, showing the inmate involved, register number, Security Threat Group classification, date of occurrence, description of conduct, staff member victim or inmate victim of each assault. Assaultive conduct can be identified and grouped by using the Bureau of Prison’s misconduct codes, including 100 Level Prohibited Acts (Killing, 100; Assault, 101; Escape, 102; Weapon, 104; Riot/Encourage Riot, 105/106) and 200 Level Prohibited Acts (Escape, 200; Fighting, 201; Assault, 224).

G. Names, register numbers, assignment rationale, Security Threat Group classification, and tenures of all inmates in the Control Unit at Florence ADMAX since it opened to the present time showing the date assigned, the reason assigned, and the date exiting the Control Unit to lesser security or release from the BOP, and reason leaving the Control Unit.

H. Names of all correctional officers working on the Control Unit at Florence ADMAX showing date assigned and date left.

I. Disciplinary Incident Reports on all inmates in the Control Unit at Florence ADMAX from its opening to the present time showing inmate name, register number, date of offense, details of the disciplinary incident, and Security Threat Group classification.

J. Correctional Services Significant Incidents Data on level and frequency of violence at each security level at Florence ADMAX by year from and including 2001 to and through 2006.

K. Movement sheets from the Bureau of Prisons on every inmate who has killed another inmate within the Bureau of Prisons within the last twenty years.

L. Investigative reports on all inmate homicides within the Bureau of Prisons within the last twenty years including any “after action reports” indicating any operational or institutional changes within the institution or within the Bureau of Prisons in response to each killing and any final memorandums from SIS to the Warden of each institution regarding each killing.

M. Regarding each inmate in the above (Subparagraph L.) involved in an inmate killing with the Bureau of Prisons within the last twenty years, the respective inmate’s “Chronological Disciplinary Record” and “Inmate History ADM-REL” and/or movement sheets within the Bureau of Prisons.

J.A. 19–20.

In support of his Brady motion, Caro attached a declaration from Cunningham, who had (at the time) testified in over one hundred state and federal capital cases about sentencing determination issues, including “mitigation and capital violence risk assessment.” J.A. 22–48. Cunningham explained that he needed the BOP data in order to conduct a “reliable individualized assessment” of the “likelihood that Mr. Caro will commit acts of serious violence from this point forward while confined for life in the Federal Bureau of Prison.” J.A. 28. To prepare a reliable assessment, he needed to review the behavior of other inmates who had committed a similar crime and had been housed in similarly restrictive conditions. Using group data to predict Caro’s individual behavior—common in any risk-based assessment, from medicine to insurance—was necessary to rebut the Government’s argument of future dangerousness. Indeed, the *33 Government’s own argument was “necessarily relying on a group-based assumption” that killing another inmate in the BOP “is related to future misconduct.” J.A. 35.

Cunningham also sought the BOP data to rebut “the corollary that the federal Bureau of Prisons is unable to safely contain this defendant, and thus a penalty of death is a reasonable preventative measure.” J.A. 33. “Informing the jury of the capabilities of BOP to bring higher levels of security to bear would appear to be the only evidence that might respond to this implicit corollary assertion regarding a particular inmate.” Id. He also noted that the Government “has routinely represented at federal capital sentencing that placement in ADX is temporary,” an assertion he claimed was “suspect at best for a large proportion of the inmates at ADX, given historic refusals of BOP/DOJ to detail length of stay information regarding inmates at ADX and broad data reflecting only 7– 9% of inmates at ADX being transferred to lower custody in any given year.” J.A. 39.

Finally, Cunningham rebutted the Government’s assertions of burden and stated that he would be happy to receive the raw facility census information. But he argued that it was “patently inconceivable that BOP has not calculated detailed length of stay information regarding this unique facility housing the ‘worst of the worst’ when an in-house BOP research unit is available to examine such vitally important performance and outcome data.” Id.

The magistrate judge granted almost all of Caro’s Brady motion, finding that the requested data was both favorable and material, and thus exculpatory. But the Government objected to the magistrate judge’s order, asserting that the information was not favorable under Brady and that it would be burdensome to disclose. The district court *34 held a hearing, after which the Government filed several declarations discussing burden. Cunningham then filed a second declaration specifically rebutting the Government’s purported difficulty or inability to produce the records. He included specific examples of the exact BOP records he needed—documents he had received from the BOP in prior cases, evidently without controversy.

The district court sustained the Government’s Brady objection on the merits, without addressing the Government’s asserted burden. The court concluded that the BOP records were not favorable: “While the defense obviously hopes that the information requested here will support its expert’s opinion, there is no indication before me that it will do so[.]” J.A. 149. Caro proceeded to trial and sentencing without the BOP data.

C.

At the sentencing hearing, Cunningham testified as “an expert in prison violence and security measures in prisons.” J.A. 677. He testified that Caro is likely to pose a high risk of harming someone else if placed in the general population of a USP during the next five or ten years. But he emphasized that Caro’s violent tendencies differ from Caro’s future dangerousness because the latter hinges on the BOP’s capability to incapacitate and control him. Cunningham testified that Florence ADMAX is not intended to be a permanent placement for most inmates, but stressed that there are some individuals “for whom there is no foreseeable plan for their return to a lower level of security.” J.A. 699–702. He testified that a Florence ADMAX official had told him that inmates stay there for an average of five years, but pointed out that he had only “limited information on average length of stay at ADX.” J.A. 699–702. He testified that there *35 had been two murders in Florence ADMAX in 2005, but explained that the prison had taken steps to prevent future violence by isolating inmates in the general population even during recreation and by moving the pre-transfer unit to a different facility.

Cunningham reiterated throughout his testimony that he could only offer anecdotes and estimates because the Government had denied him access to accurate data about Florence ADMAX and inmate violence in the BOP. J.A. 699–702, 736–40, 792– 98, 799–802. For example, Cunningham testified that the BOP had at one point provided him with “the assaultiveness conduct that took place on the Control Unit from the time ADX opened in December of 1994 through June of 2001.” J.A. 738. During that time period, there were seventeen attempted or actual minor assaults by inmates in the Control Unit, most of which involved throwing liquids and ten of which were committed by the same inmate. Cunningham had “asked specifically for an update on assaultiveness conduct on the Control Unit, as well as length of stay information on the Control Unit because it’s so critical to this question of how long can an inmate be held, what’s typical in terms of holding them.” J.A. 740. But the BOP had refused. Later, Cunningham criticized as “misleading” the Government’s evidence that several inmates initially placed in Florence ADMAX were now in lower-security facilities because “the critical issue is what happened to [the inmate] between the time he was guilty of the killing, and now,” not simply where he ended up. J.A. 793–94. Cunningham stated that he could not “comprehend why that simple scientific data would be something that the U.S. Department of, Department of Justice would resist.” J.A. 797.

Cunningham explained that the BOP data was critical to developing an accurate risk assessment of Caro’s future dangerousness; in its absence he was limited only to discussing the conditions of Caro’s confinement:

If I want to know the best way of gauging the risk that killing another inmate in prison has for future conduct, if I want to know what effect does it have for somebody to kill another inmate in prison, how does that affect the rest of their time in prison, and how much violence they commit, then I need to collect the data on individuals who have done that. If I want to know what the risk is of a 16 year old male unmarried driver, then I need to track 16 year old male unmarried drivers and their driving records so I will know whether being 16 is a risk factor for driving, or not, and how much of a risk factor it is. This is fundamental to accurate risk assessment, is to collect data about individuals that have a similar background. The same thing happens in medicine. If I want to know what the prognosis is for a given disease, I need to track the outcomes of people with that disease.

So, that’s what I asked for here, is – there are computer print outs, it’s relatively easily obtained, there are three or four computer print outs that would show the inmate’s movement history within the Bureau of Prisons, so I could identify whether they were being held at a SHU, or went to ADX, or went to some other facility. I also want the print out of their chronological disciplinary record that would have let me view what offenses they had gotten in prison before the homicide, and what offenses they had in prison after the homicide. Then I would have a body of data about prison homicide offenders in the Bureau of Prisons so that we wouldn’t have to speculate about how long are inmates held, going to be held at ADMAX, and does it make any difference whether they have a gang affiliation, or those kind of things. We would have data about that, and would also have data about what to expect from those offenders over time when they came out from under being locked down on a SHU or ADX. It was fundamental scientific data to inform a risk assessment of Mr. Caro.

Now, in the absence of that data, it’s not possible to do that kind of risk assessment. It’s only possible to talk about what conditions of confinement are available that the Bureau of Prisons can bring to bear, and what the effect of those conditions are on what, on rates of violence on the Control Unit, which is the kind of unit where, essentially, ADX is functioning as at this point. It’s simply critical to informing this, informing

an understanding of the future prison behavior of an inmate homicide offender.

J.A. 799–801 (emphasis added).

In rebuttal, the Government called Hershberger, who emphasized that Florence ADMAX officials expect to return inmates to lower security prisons. J.A. 835, 837–38, 841–44, 863. He stated that the “primary program” at Florence ADMAX “is to get them in, work them through a minimum three year program and out to another open penitentiary,” even if the inmate had been convicted of killing another inmate. J.A. 837– 38. He agreed that once inmates complete “12 months in general population, 12 months in the immediate, and 12 months in transition, then it’s anticipated they would leave ADX to go to this pre-transfer unit at USP Lee.” J.A. 841–44. Hershberger did agree that Thomas Silverstein, who killed two inmates and a BOP officer, has been in solitary confinement since 1983, but called him “a very special case” and his review “a very special review.” J.A. 858–61, 870. Despite the danger that the Government claimed Caro posed, Hershberger testified that Caro would not be treated the same as Silverstein.

In its closing argument, the Government focused almost exclusively on how Caro’s future dangerousness justified a capital sentence. The Government argued that Caro’s past history of violence meant that he will be violent in the future and claimed that the BOP cannot control him. The Government also repeatedly asserted that, if sentenced to life in prison, Caro would be imminently released from Florence ADMAX:

What do we know? We know that if, if Carlos Caro goes to that facility he’s not going to stay there. Whether it’s through the Control Unit, or whether it’s through the general population at the ADX, he eventually, ladies and gentlemen, will, will graduate out, be stepped down out of that

facility back into a United States penitentiary just like the United States Penitentiary in Lee County. If he goes, he can probably still communicate with his gang buddies because we know that despite the best precautions at the ADMAX facility, people send out coded letters. They have certain privileges which would allow the communication, and also increasing contact. He can use the telephone. He can have visitation with his buddies. He has exercise. He can use a library. We know that he can write letters. He has a right to medical services, and as all those contacts increase, particularly as we go to the step down, that his contact, his access to inmates, his access to staff is going to increase, ladies and gentlemen, and we also know that he will eventually end up back in the USP just like USP Lee unless he harms someone else before going there.

How long is it going to take to do that? You saw the regulation. You heard the testimony of Mr. Hershberger. Three years, three years for him to be stepped down out of ADX and into a USP. Can he be controlled with ADMAX? We know ADMAX is the most secure federal prison, but it’s not failsafe, and I think what Mr. Hershberger said, where there’s a will there’s a way. . . . And Hershberger told you, based on his experience as a warden, if Mr. Caro was given a light sentence, he may initially go to ADMAX, but he will be moved out to the USP on a three year program, well within the life of violence of Carlos Caro.

• • * *

Again, you decide what the facts are. You decide, is he going to get out in three years as Warden Hershberger says from ADX, or is he going to get out in five years as Dr. Cunningham says? Does that really matter? Everyone agrees, every witness agrees he’s getting out of ADX, that in some time within three to five years he will be back at a USP, right where he stabbed Rick Benavidez, and right where he strangled Roberto Sandoval. That is the evidence. Those are the facts. You have to decide what significance that is.

J.A. 923–24, 979 (emphasis added).

After deliberating for two hours, the jury unanimously imposed the death penalty. All jurors found that the Government had proved all three non-statutory aggravating factors beyond a reasonable doubt. All jurors also found that Caro had proved twelve *39 mitigating factors by preponderance of the evidence, 6 while some jurors found that Caro had proved an additional four mitigating factors. 7 On March 30, 2007, the district court sentenced Caro to death.

D.

Caro filed a direct appeal of his conviction and sentence. My colleagues and I affirmed the district court’s denial of Caro’s Brady motion because Caro could “only speculate as to what the requested information might reveal” and so had “failed to establish that the information requested would be favorable to him.” Caro, 597 F.3d at 619. The majority otherwise affirmed Caro’s conviction and sentence. Id. at 636.

Caro timely filed a motion to vacate his sentence under 28 U.S.C. § 2255. He claimed that the Government had violated Brady by withholding the BOP data. In

6 The jury unanimously found that Caro (1) was exposed to domestic violence growing up, (2) was not encouraged in school, (3) came from an impoverished community, (4) was well-behaved growing up, (5) failed to reach high school after needing special education, (6) was shy and respectful compared to his brothers, (7) was brought into illegal drug trafficking by his uncles, (8) never abused his wife or daughter, (9) was not violent or aggressive until his thirty-year prison sentence, (10) has never attacked prison staff, (11) has never tried to escape, and (12) has been securely detained “at various high security federal institutions” since December 18, 2003. Caro, 597 F.3d at 613 n.6. 7 One juror voted that Caro’s father had a corrupting influence, five voted that Caro’s execution would grieve his family, eight voted that Caro’s life has value to his family, and nine voted that during a life sentence Caro would be “incarcerated in a secure federal institution.” J.A. 882–85.

No juror found any of the remaining six factors: (1) Caro exhibited symptoms of failure to thrive as an infant, (2) Caro’s mother was not able to nurture her children because of her violent and abusive husband, (3) Caro was sometimes a good father and husband, (4) Caro was not involved in gang-related activity while in the community, (5) Caro was not involved in gang-related activity in prison until he was sentenced in 2001, and (6) Caro is 40 years old and is less likely to engage in violence as he ages.

support, Caro presented newly uncovered evidence that revealed some of the suppressed BOP data. This new evidence showed that a substantial portion of the Florence ADMAX population, including specific inmates who committed homicides within the BOP, has been held there for more than three years.

First, Caro presented a November 2011 affidavit from Jeanne Dvorak, an employee of Rothstein, Donatelli, Hughes, Dahlstrom, Schoenburg, & Bienvenu, LLP in New Mexico. J.A. 1338–46. In the affidavit, Dvorak describes the survey she sent to 129 inmates at Florence ADMAX in November 2010. Between late 2010 and early 2011, 69 inmates responded. Fourteen other surveys were returned unfilled because the inmates were in Special Administrative Measures (SAMS)—an extreme form of isolation that places special restrictions on an inmate’s communications—and unable to receive mail.

Of the 69 respondents, 43 inmates stated that they had been at Florence ADMAX (or Florence ADMAX and USP Marion 8) for eight or more consecutive years. Twenty-four inmates stated that they had been at Florence ADMAX (or Florence ADMAX and USP Marion) for 13 or more consecutive years. Dvorak included a table listing the

8 Florence ADMAX opened in 1994; before that, USP Marion was the BOP’s most secure prison. Justin Peters, How a 1983 Murder Created America’s Terrible Supermax-Prison Culture, Slate (Oct. 23, 2013), http://www.slate.com/blogs/crime/2013/10/23/marion_prison_lockdown_thomas_silverst ein_how_a_1983_murder_created_america.html (saved as ECF opinion attachment 3). After Silverstein and another inmate murdered two prison officials in 1983, USP Marion went into a 23-year lockdown. Id. In 2006, USP Marion came out of lockdown and was downgraded to a medium-security prison. Id. See also J.A. 836–37, 848.

names of the surveyed inmates and the years they entered USP Marion and Florence

ADMAX.

Second, Caro presented two 2013 declarations from Mark A. Bezy, who worked for the BOP for 28 years. J.A. 1220–28, 1689–90. Bezy worked as a captain at USP Marion and oversaw the transfer of high security inmates from USP Marion to Florence ADMAX. Bezy stated that numerous inmates were held well beyond three years. He recalled at least eight inmates by name who had been housed at Florence ADMAX for more than three years and who were still there as of December 2012.

Finally, Caro presented an October 2013 declaration from Susan Richardson, an investigator with the Federal Public Defender’s Office for the Western District of Virginia. J.A. 1750–68. Richardson compiled data from: the Dvorak affidavit; documents produced by the Government in response to a 2010 subpoena issued to the BOP in United States v. Basciano, No. 05-cr-60 (E.D.N.Y.); the BOP Inmate Locator; PACER; the Federal Death Penalty Resource Counsel website; documents received pursuant to a FOIA request; and internet searches for articles. She included tables listing the name of each inmate, when they entered Florence ADMAX, how many years they had been there, and, if they had committed a homicide in the BOP, details about the homicide.

Richardson estimated that as of October 2013, 126 inmates have been held at Florence ADMAX for more than five years, and at least 155 inmates have been held there for more than three years. Of these 155 inmates, 125 were still designated to Florence ADMAX. In other words, almost 30% of Florence ADMAX’s October 2013 population *42 of 434 had been held there for more than three years. At the time of Caro’s trial in January 2007, Richardson found that there were at least 79 inmates held at Florence ADMAX for more than three years, at least 63 who had been held there for more than five years, and at least 25 inmates who had been held there for at least 10 years. 9

Richardson located ten cases nationwide in which the Government sought the death penalty for a defendant who committed homicide within the BOP, but where the jury imposed a life sentence. Nine of these ten defendants had been continuously held at Florence ADMAX since the imposition of their life sentences, while the tenth had been held elsewhere due to significant mental disorder. She also located at least 54 inmates who have been convicted or accused of committing a homicide within a BOP facility and who were sent to Florence ADMAX. All 54 were still at Florence ADMAX, including 22 who were placed there in or before 2007.

II.

“When the district court denies § 2255 relief without an evidentiary hearing, the nature of the court’s ruling is akin to a ruling on a motion for summary judgment. In such a circumstance, we review the facts in the light most favorable” to Caro, the § 2255

9 In November 2006, Florence ADMAX had a capacity of 490 cells, and held approximately 470 inmates. J.A. 697, 835. As of April 2018, Florence ADMAX holds 405 inmates. Generate Inmate Population Reports, Florence ADMAX, BOP, https://www.bop.gov/about/statistics/population_statistics.jsp (last visited Apr. 18, 2018) (saved as ECF opinion attachment 4).

movant. United States v. Poindexter, 492 F.3d 263, 267 (4th Cir. 2007) (citing United States v. Nicholson, 475 F.3d 241, 248 (4th Cir. 2007)).

The majority concludes that Caro’s Brady claim is both procedurally barred and meritless. I disagree with both conclusions.

A.

In finding that Caro’s Brady claim is procedurally barred, the majority relies on a well-established doctrine: A defendant cannot use her collateral attack to relitigate issues that were “fully considered” on direct appeal. Boeckenhaupt, 537 F.2d at 1183; accord United States v. Dyess, 730 F.3d 354, 360 & n.5 (4th Cir 2013); United States v. Linder, 552 F.3d 391, 396–97 (4th Cir. 2009); United States v. Roane, 378 F.3d 382, 396 n.7 (4th Cir. 2004). But the majority’s invocation of Boeckenhaupt here is misplaced. 10

We have never before applied Boeckenhaupt to an alleged Brady violation—and for good reason: Boeckenhaupt and its progeny concerned exactly the same claims made with exactly the same evidence and exactly the same arguments on both direct and collateral review. E.g., Dyess, 730 F.3d at 360 (rejecting on collateral review the defendant’s argument that “the indictment did not allege a specific drug quantity” and therefore his sentence violated Apprendi v. New Jersey, 530 U.S. 466 (2000), because we rejected that precise argument on direct appeal); Linder, 552 F.3d at 396–97 (rejecting on collateral review the defendant’s challenge to his sentence under United States v. Booker,

10 The majority only cites to Dyess, 730 F.3d at 360 and Linder, 552 F.3d at 396, which are the most recent iterations of the doctrine. Ante 17. But because Boeckenhaupt is one of our earliest articulations of the doctrine, I refer to it by that case name.

543 U.S. 220 (2005), because we rejected the identical argument on direct appeal); Roane, 378 F.3d at 396 n.7 (rejecting on collateral review the defendants’ claims of discrimination, unconstitutional delegation of legislative authority, insufficient evidence, and juror misconduct because we “already addressed and rejected” them on direct appeal); Boeckenhaupt, 537 F.2d at 1183 (rejecting on collateral review the defendant’s arguments that he was arrested without probable cause, unlawfully detained, and unlawfully sentenced because we had “fully considered” those issues on direct appeal). Had Caro brought the exact same Brady claim, supported by the exact same evidence and the exact same arguments, I would agree with the majority that he cannot relitigate it now. Ante 17–18. But he has not. The majority’s conclusion to the contrary, and its holding that Caro’s newly uncovered evidence was “previously” or “publicly” available, ante 15, 18, has no basis in the record.

1.

As the Supreme Court has recognized, there are three types of Brady violations: undisclosed evidence unknown to and unrequested by the defense, undisclosed evidence requested generally by the defense pretrial (e.g., a request for “Brady material”), and undisclosed evidence specifically requested by the defense pretrial (e.g., the BOP data here). United States v. Agurs, 427 U.S. 97, 104–07 (1976); see Kyles v. Whitley, 514 U.S. 419, 433–34 (1995) (stating that the Government has equal obligation to disclose materially favorable evidence in all three circumstances). The commonality between all three is nondisclosure: a Brady claim by definition involves an assertion that the Government has suppressed (willfully or inadvertently) materially favorable evidence at *45 trial. Monroe v. Angelone, 323 F.3d 286, 299–300 (4th Cir. 2003). And because the Government has suppressed the evidence at trial, a Brady claim also necessarily means that the evidence is not part of the trial record—and thus not part of the record to which a court of appeals is limited on appeal. See United States v. King, 628 F.3d 693, 702 (4th Cir. 2011) (“Brady cases . . . typically involve a defendant’s post-trial discovery of evidence that the Government has assertedly suppressed.”); United States v. Dodson, 291 F.3d 268, 275 (4th Cir. 2002) (stating that Brady claims “often arise for the first time in collateral proceedings”).

In this way, Brady claims resemble ineffective assistance of counsel (IAC) claims, which also almost always turn on facts outside the trial record. Massaro v. United States, 538 U.S. 500, 505 (2003). Because of their unique posture, the Supreme Court has held that IAC claims can proceed on collateral challenge without fear of procedural default, a doctrine that ordinarily bars collateral review of claims not raised on direct appeal. Id. at 504. Like the Boeckenhaupt rule and other rules of procedure, the procedural default rule is a judge-created rule intended to “‘induce litigants to present their contentions to the right tribunal at the right time,’” to “conserve judicial resources,” and to “respect the law’s important interest in the finality of judgments.” Id. (quoting Guinan v. United States, 6 F.3d 468, 474 (7th Cir. 1993) (Easterbrook, J., concurring)). Because a trial court record is “often incomplete or inadequate” for litigating IAC claims, barring them from collateral review would risk preemptively eliminating meritorious claims and would waste judicial resources. Id. at 506–08.

The same is true with Brady claims. Unsurprisingly, the Fourth Circuit has never held that a Brady claim raised for the first time in a collateral challenge under § 2255 is procedurally defaulted. To the contrary, we have declined to review Brady claims on direct appeal when the allegedly suppressed evidence was not part of the trial record. E.g., United States v. Russell, 971 F.2d 1098, 1112 (4th Cir. 1992). We do so because we recognize that plaintiffs should be allowed to present Brady claims, like IAC claims, “to the right tribunal at the right time.” Massaro, 538 U.S. at 504 (quoting Guinan, 6 F.3d at 474 (Easterbrook, J., concurring)); see Sanchez-Llamas v. Oregon, 548 U.S. 331, 359 (2006) (“In the case of a Brady claim, it is impossible for the defendant to know as a factual matter that a violation has occurred before the exculpatory evidence is disclosed.”); United States v. Dominguez Benitez, 542 U.S. 74, 83 n.9 (2004) (stating that Brady claims may be raised in § 2255 proceedings because they “permit greater development of the record,” citing Massaro, 538 U.S. at 500).

But unlike IAC claims, Brady motions are often filed by the defendant pretrial, making the motion’s denial inevitably part of the record we review on appeal. Fed. R. App. P. 10(a). Applying the Boeckenhaupt doctrine, collateral consideration of an unsuccessful pretrial Brady motion would be barred—especially because Boeckenhaupt applies even to claims buried in the trial record that we never squarely address on direct appeal. Dyess, 730 F.3d at 360 & n.5. And yet we have never before today used Boeckenhaupt to bar collateral review of any Brady claim, even in an unpublished opinion. To the contrary, in our only opinion addressing both doctrines, we recognized that a matter “considered on direct appeal . . . cannot be revisited collaterally absent a *47 violation of Brady.” United States v. LaRouche, 4 F.3d 987 (Table), at *2 (4th Cir. 1993) (per curiam) (emphasis added) (addressing the merits of a Brady claim on collateral review even though we had previously addressed the Brady claim on direct appeal). This reticence to apply Boeckenhaupt to Brady claims indicates our acknowledgement that a defendant’s inability to locate pretrial what the Government has suppressed—and the appellate court’s subsequent review of that insufficient trial record—should not bar the defendant, upon discovering that evidence post-trial, from raising it in a collateral challenge.

This mirrors how the Supreme Court has instructed us to approach, on collateral review under 28 U.S.C. § 2254, a Brady claim that failed in state court for lack of evidence. Banks v. Dretke, 540 U.S. 668, 690 (2004). In that situation, the state habeas petitioner’s Brady claim is procedurally defaulted and he is barred from an evidentiary hearing in federal court—unless he can “show cause for his failure to develop the facts in state-court proceedings and actual prejudice resulting from that failure.” Id. at 690–91 (quoting Keeney v. Tamayo–Reyes, 504 U.S. 1, 11 (1992)). But the Supreme Court has observed that cause and prejudice “parallel two of the three components of the alleged Brady violation itself”: a petitioner shows “cause” when “the reason for his failure to develop facts in state-court proceedings was the State’s suppression of the relevant evidence,” and a petitioner shows “prejudice” when “the suppressed evidence is ‘material’ for Brady purposes.” Id. at 691 (quoting Strickler v. Greene, 527 U.S. 263, 282 (1999)). In other words, even a Brady claim that was adjudicated on the merits in *48 state court should be considered on the merits in federal court if the defendant presents new favorable evidence.

At bottom, the majority and I agree that Boeckenhaupt can theoretically bar relitigation of a fully considered Brady claim on collateral review—we differ (in the first instance) on whether Caro’s Brady claim was fully considered, given the new evidence he has uncovered. I write here only to emphasize the narrowness of today’s holding: a Brady claim is procedurally barred under Boeckenhaupt and its progeny only if it is made with exactly the same evidence and exactly the same arguments raised on direct appeal. Because the vast majority of Brady claims will not meet this strict requirement, Boeckenhaupt will likely return to dormancy in Brady cases.

2.

Caro’s case is a variation of the typical Brady case: he requested disclosure of specific BOP data that he knew existed but could not prove pretrial would be favorable to him. On direct appeal, we concluded that Caro could “only speculate as to what the requested information might reveal.” Caro, 597 F.3d at 619. Now Caro returns to court with evidence validating his speculations: the BOP data would show that the Government could securely house him at Florence ADMAX well beyond three years, the same way it routinely houses other violent inmates who have committed homicides within the BOP. Had Caro possessed the BOP data at trial, he could have undercut the Government’s future dangerousness allegations, bolstered the testimony of Cunningham, and impeached Hershberger.

Rather than recognizing this evidence for what it is—newly discovered data, vigorously suppressed by the Government and therefore beyond the limited trial record we reviewed eight years ago—the majority concludes that it was “compiled from publicly available sources” and “previously available” but “left out of the direct appeal record.” Ante 4, 15. 11 Thus, says the majority, Caro’s new evidence “does not suffice to make the Brady claim raised in his § 2255 motion different from the claim we rejected on direct appeal.” Ante 18.

The majority provides no case for the proposition that evidence being “previously” or “publicly” available means an issue was “fully considered” under Boeckenhaupt. Instead, the majority cites to Small v. Hunt, which involved a motion to alter or amend a judgment under Federal Rule of Civil Procedure 59(e). 98 F.3d 789, 798 (4th Cir. 1996). There, we held that Rule 59(e) relief could be granted “to account for new evidence not available at trial,” provided that the moving party produced a “legitimate justification for not presenting the evidence during the earlier proceeding.” Id. (internal quotation marks omitted). But Caro has not moved to alter or amend any civil judgment under Rule 59(e); instead, he argues that the new evidence proves that his Brady claim was not “fully considered” on direct appeal. And Small says nothing about “previously” or “publicly” available evidence. To the contrary, the relevant evidence was previously available to the

11 Even the district court here found that Caro’s new evidence was collected from various sources, “some of which were not available at the time of Caro’s trial.” J.A. 1955.

state (it was the state’s own plans)—the state had simply declined to present those plans until ordered by the court, which we considered a “legitimate justification.” Id.

The majority also invokes Supreme Court precedent and the Antiterrorism and Effective Death Penalty Act of 1996 (AEDPA), ante 18–19, but to no avail. It is true that Sanders v. United States, 373 U.S. 1, 16–17 (1963), abrogated in relevant part by AEDPA (codified at 28 U.S.C. § 2255(h)), incorporated Townsend’s definition of “newly discovered evidence” 12 to second or successive § 2255 petitions raising previously rejected claims—but Caro is on his first petition. Contrary to the majority’s mischaracterization, Davis v. United States did not extend this provision of Sanders to direct appeals; indeed, Davis did not discuss newly discovered evidence at all. 417 U.S. 333, 341–42 (1974). Instead, the “sole issue” resolved in Davis was affirming that a petitioner can base a § 2255 petition on a “change in the law of [a] Circuit” that occurred after the petitioner’s direct appeal. Id. And the majority points to no provision of AEDPA itself that bars first-time § 2255 petitions if newly discovered evidence could “reasonably have been included in the direct appeal record.” Ante 19. Indeed, nothing in the majority’s cited cases suggest that Boeckenhaupt is limited to only a subset of newly discovered evidence.

Finally, the majority appears to indirectly invoke the “other sources” doctrine, which holds that “the Brady rule does not apply if the evidence in question is available to

12 Townsend v. Sain, 372 U.S. 293, 317 (1963) (noting that “newly discovered evidence” is “evidence which could not reasonably have been presented to the state trier of facts”), overruled in nonrelevant part by Keeney v. Tamayo-Reyes, 504 U.S. 1 (1992).

the defendant from other sources,” United States v. Wilson, 901 F.2d 378, 380 (4th Cir. 1990) (internal quotation marks and citation omitted), “including diligent investigation by the defense,” Fullwood v. Lee, 290 F.3d 663, 686 (4th Cir. 2002) (internal quotation marks and citation omitted). But this doctrine determines whether the Government has an obligation to provide the BOP data in the first instance, not whether we “fully addressed” Caro’s Brady claim on direct appeal such that he is barred now under Boeckenhaupt.

Even if these doctrines applied to the Boeckenhaupt framework, Caro has a legitimate justification for not providing the new evidence sooner: it was not available, much less “reasonably” capable of being included in the direct appeal record. The Dvorak affidavit summarizes a survey sent to Florence ADMAX residents by an unrelated New Mexico firm in November 2010, while the Richardson declaration relies in part on the Dvorak affidavit and documents produced by the Government in response to a 2010 subpoena. Neither Dvorak’s survey nor the subpoena existed in 2007; therefore, they were not “previously available” to Caro. In addition, we have applied the “other sources” doctrine only when the evidence was either already known by the defendant or reasonably accessible. 13 But Caro had no knowledge of or access to the

13 E.g., United States v. Catone, 769 F.3d 866, 872 (4th Cir. 2014) (evidence was form submitted by defendant himself to Department of Labor and could also have been obtained by written request); Roane, 378 F.3d at 402 (evidence was witness statements providing defendant an alibi, but defendant knew where he was); Fullwood, 290 F.3d at 686 (evidence was defendant’s own statements to police); United States v. Bros. Const. Co. of Ohio, 219 F.3d 300, 316 (4th Cir. 2000) (defendant obtained the same information via FOIA); Barnes v. Thompson, 58 F.3d 971, 976–77 (4th Cir. 1995) (evidence was (Continued)

underlying BOP data. Nor is evidence reasonably available from other sources when even diligent investigation only exposes fragments. And just because some information is publicly available now (such as the BOP Inmate Locator, PACER, the Federal Death Penalty Resource Counsel website, and miscellaneous internet articles relied on in part by Richardson) does not mean that it was readily available then. These are “legitimate” and “reasonabl[e]” explanations for not presenting this new evidence at trial. See Townsend, 372 U.S. at 317; Small, 98 F.3d at 798.

To the contrary, the majority’s suggestion that Caro’s attorneys should have conducted a piecemeal survey of individual inmates at Florence ADMAX, ante 20, is unreasonable. Townsend, 372 U.S. at 317. It is not reasonable to expect inmates to systematically and accurately self-report sensitive personal information, such as their assault histories. More fundamentally, inmates incarcerated in the BOP’s highest security prison are not “publicly available.” Indeed, only half of the surveys sent by Dvorak were even filled out. An additional fourteen were returned unfilled because the inmates were in SAMS and unable to receive mail. Because the Government tied Caro’s future dangerousness in part to his ability to communicate with the outside world in code,

location of victim’s gun, which defendant either knew or could have obtained from his co-defendant’s earlier trial); Stockton v. Murray, 41 F.3d 920, 927 (4th Cir. 1994) (defendant was aware of evidence and never requested it); Epperly v. Booker, 997 F.2d 1, 9 (4th Cir. 1993) (defendant could have obtained evidence through discovery, independent expert testimony, or cross-examination); Wilson, 901 F.2d at 381 (evidence was statements of witness the defendant was free to question ahead of trial but did not).

e.g., J.A. 923, this means that the very inmates Caro would be most interested in surveying were literally inaccessible.

Moreover, contrary to the majority’s assertion, ante 20, Caro showed diligence before trial: he filed four motions for the BOP data and hired an expert (Cunningham) who filed two declarations in support of Caro’s motions. See United States v. Ellis, 121 F.3d 908, 914 (4th Cir. 1997) (finding that defendant made “substantial efforts” to obtain evidence in dispute by filing a Brady motion). That Caro did not uncover all of the information the Government was working so hard to hide should not keep him from seeking that information now that new evidence vindicates his original claims. As the Supreme Court has said, “[a] rule thus declaring ‘prosecutor may hide, defendant must seek,’ is not tenable in a system constitutionally bound to accord defendants due process.” Banks, 540 U.S. at 696.

The bitter irony is that Caro would have been better off had he never filed the Brady motion to begin with. Free of the Boeckenhaupt doctrine, he could have proceeded to the merits of his Brady violation on collateral review, using the evidence he discovered in the interim. Caro’s pre-trial diligence, frustrated by the Government’s suppression efforts, should not bar his post-trial claims when he has provided the Court with new evidence.

B.

In the alternative, the majority concludes that Caro’s Brady claims fail on the merits. “[A] Brady violation has three essential elements: (1) the evidence must be favorable to the accused; (2) it must have been suppressed by the Government, either *54 willfully or inadvertently; and (3) the suppression must have been material, i.e., it must have prejudiced the defense at trial.” Monroe, 323 F.3d at 299–300 (citing Strickler, 527 U.S. at 281–82). It is undisputed that the BOP data has been suppressed, but the majority errs in concluding that the BOP data is neither favorable nor material.

1.

Evidence is favorable if it is exculpatory or if it can be used to impeach a witness. Wolfe v. Clarke, 691 F.3d 410, 423 (4th Cir. 2012) (citing Banks, 540 U.S. at 691). The majority concludes that the BOP data is not favorable because it does not support one mitigating factor raised by Caro, that the “BOP would house him at Florence ADMAX until he aged out of violence.” Ante 21. But this reads Caro’s claim far too narrowly. Caro did not seek the BOP data to support only a single mitigating factor. Instead, he sought the BOP data because it would have impeached Hershberger’s testimony and exculpated Caro of a capital sentence by undermining the Government’s key factor of future dangerousness. 14

First, the requested BOP data would have allowed Caro to show that he could be held indefinitely at the BOP’s most secure prison. The new evidence proves that a

14 Even the Government acknowledges that Caro’s new evidence is favorable. Appellee’s Resp. Br. 40 (stating that Caro had now “presented some statistical evidence extrapolated from raw data he located independently, that appears favorable to his position on future dangerousness”). The Government then shifts the goalposts, arguing two pages later that “Caro has again failed to show that the requested evidence is favorable” because his new evidence does not establish that the BOP data would show exactly how long the BOP would hold Caro at Florence ADMAX. Id. 42. But Caro never sought to show exactly how long he would be held at Florence ADMAX, only that Florence ADMAX would be able to house him securely.

substantial number of inmates at Florence ADMAX do remain there much longer than the aspirational three years anticipated by the step-down program. Even more importantly, nine out of ten inmates sentenced to life in prison for killing another inmate have been held at Florence ADMAX since convicted, which shows that the BOP can and does securely house inmates with a history of dangerousness. This evidence would have directly undermined the Government’s arguments that it would only take “three years for him to be stepped down out of ADX and into a USP,” that “if Mr. Caro was given a light sentence, he may initially go to ADMAX, but he will be moved out to the USP on a three year program, well within the life of violence of Carlos Caro,” and that “in some time within three to five years he will be back at a USP, right where he stabbed Rick Benavidez, and right where he strangled Roberto Sandoval.” J.A. 923–24, 979.

Second, the BOP data would have allowed Caro to impeach Hershberger. For example, Hershberger testified that Silverstein, who has been housed since 1983 in solitary confinement at USP Marion and then Florence ADMAX, was a “very special case” who receives “a very special review.” J.A. 858–61. Statistics and case studies about other inmates held long-term in solitary confinement at Florence ADMAX would have shown this to be untrue. Indeed, that nine out of ten inmates convicted of killing another inmate have been held at Florence ADMAX since being sentenced to life in prison would certainly have contradicted Hershberger’s claim that only Silverstein was treated in such a “special” way. In addition, Hershberger testified that “the program [at Florence ADMAX] is to get them in, work them through a minimum three year program and out to another open penitentiary.” J.A. 837–38. He said that inmates who killed *56 other inmates and were placed in Florence ADMAX would be “in the three year program.” J.A. 863. He responded “That’s correct” when the Government asked him whether inmates who spend 12 months at each step of the step-down program would leave Florence ADMAX. J.A. 842–43. Hard data about how long inmates actually stay at Florence ADMAX would undermine Hershberger’s testimony that Florence ADMAX operated as advertised.

The majority claims that because no juror found that Caro would age out of violence, the BOP data “would only be relevant to the jury’s future dangerousness finding if the data showed that the BOP would likely house Caro at Florence ADMAX for the rest of his life.” Ante 21–22. Not so. The majority “confuses the weight of the evidence with its favorable tendency.” Kyles, 514 U.S. at 451. That inmates are routinely held at Florence ADMAX well beyond the three year program would have allowed Caro to challenge the Government’s arguments to the contrary, and ultimately undermine the Government’s primary aggravating factor of future dangerousness. This is plainly favorable; there is no sufficiency requirement for favorability.

Caro seeks the BOP data to support his argument that he can be securely housed at Florence ADMAX. The data he has uncovered since his sentencing vindicate this argument. Therefore, the BOP data is favorable.

2.

The majority also errs in concluding that the BOP data is not material. “[E]vidence is material only if there is a reasonable probability that, had the evidence been disclosed to the defense, the result of the proceeding would have been different.” *57 United States v. Bagley, 473 U.S. 667, 682 (1985). A “reasonable probability” does not require the defendant to show that he more likely than not would have received a different sentence. Kyles, 514 U.S. at 434. Nor does it turn on the sufficiency of the evidence. Id. at 434–35 (“A defendant need not demonstrate that after discounting the inculpatory evidence in light of the undisclosed evidence, there would not have been enough left to convict.”). Instead, there is a reasonable probability of a different result “when the government’s evidentiary suppression ‘undermines confidence in the outcome of the trial.’” Id. (quoting Bagley, 473 U.S. at 678). The majority concludes that there is no reasonable probability that the BOP data would have affected any juror’s vote. Ante 23–25. But the majority ignores what Caro actually requested in his Brady motion and consequentially fails to recognize the material impact its absence had on the jury’s decision.

In the penalty context, materiality does not require a showing that the balance of evidence would still justify the death penalty. See Kyles, 514 U.S. at 434–35. In Strickler, for example, the Supreme Court struck down as “incorrect” an appellate court’s holding that even “without considering [witness]’s testimony, the record contained . . . evidence sufficient to support the findings of vileness and future dangerousness that warranted the imposition of the death penalty.” 527 U.S. at 290. Instead, the touchstone of Brady materiality is whether 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, 514 U.S. at 435; accord Strickler, 527 U.S. at 290; Juniper v. Zook, 876 F.3d 551, 567 *58 (4th Cir. 2017). The materiality of the evidence “turns on the cumulative effect of all such evidence suppressed by the government,” not on each item. Kyles, 514 U.S. at 421.

Applying these principles, the BOP data is material because its absence undermines confidence in a juror’s vote for death. Caro’s Brady motion requested not just how long inmates have stayed at Florence ADMAX since it opened in 1994, but also what offense caused them to be transferred there; the disciplinary and assaultive conduct records for inmates in the Control Unit at Florence ADMAX; records about violence at each security level of Florence ADMAX; and the movements sheets, disciplinary records, and histories of inmates (including those at Florence ADMAX) who killed another inmate in the BOP over the last 20 years. See supra Part I.B; see also Kyles, 514 U.S. at 436 (holding that suppressed evidence must be “considered collectively, not item by item”). 15 Had Caro received this information, his expert Cunningham would have been

15 The majority claims that the Court cannot consider everything Caro requested in his Brady motion because Caro states in his brief that the Government withheld the BOP data related to “the maximum length of time inmates can be housed at ADX Florence.” Ante 24 n.6 (quoting Appellant’s Opening Br. 22). But this quote is from a header in Caro’s opening brief that summarizes the many categories of information requested by Caro in his Brady motion. See supra Part I.B. The maximum length of time inmates can be held at Florence ADMAX is not a category of information requested by Caro in his Brady motion, and he does not limit himself to only that information. Instead, Caro’s briefs make repeated references to all the data sought by Caro in his Brady motion, indicating that the full BOP data, not a small subset, are properly before this Court. E.g., Appellant Opening Br. 18 (describing the suppressed BOP data as reflecting “how long BOP would hold Caro at ADX Florence,” which was likewise not a specific category of information requested and something that could only be discovered if the full BOP data were disclosed); id. 26 (summarizing his Brady motion as “records relative to the security of BOP facilities and the length of time the BOP could hold him in the supermax prison in Florence, Colorado, ADX Florence”), id. (stating that his Brady motion is “[s]ignificant to the certified claim brought in this appeal”), id. 27 (stating that the BOP (Continued)

able to prepare an actual risk assessment based on how the BOP has handled inmates with similar criminal histories. Cunningham also would have been able to testify about what the BOP actually does with high risk inmates, rather than what it aspires to do. And Caro could have impeached Hershberger’s testimony about Silverstein and his affirmance that Florence ADMAX’s step-down program applies to everyone. Rather than rely on dueling expert witnesses, the BOP data would have conclusively shown that the Government can—and routinely does—keep dangerous inmates at Florence ADMAX securely and for far longer than the aspirational three-year step down program suggests. By ignoring the full scope of this information, the majority incorrectly assumes that Caro’s penalty phase arguments would have remained the same.

In addition, the majority incorrectly assumes that because all twelve jurors found Caro likely to commit acts of violence against other inmates and not likely to grow less violent with age, they would necessarily do so again. Ante 23. But that is the crux of this case—the Government urged a capital sentence based almost exclusively on Caro’s likelihood of committing future acts of violence. Had Caro received the BOP data, he could have rebutted the Government’s allegations. The majority’s circular reasoning

records requested by Cunningham “are the subject of the Brady motion at issue here”); id. 34 (stating that “the Brady claim in the trial court, in the absence of the production of the BOP data requested by Caro in discovery and initially ordered produced by the magistrate judge, was not ‘fully considered’” and thus cannot preclude review by this Court); Appellant Reply Br. 7–9 (same); id. 15 (arguing that a Government assertion at trial “could have been disproved had the district court affirmed the magistrate judge’s decision to compel the production of the BOP data”); id. 16 (describing the suppressed BOP data in part as “length of stays at ADX Florence during its history, including for those who have killed while in federal custody”).

presumes that the BOP data will have no effect on the outcome of the proceeding, in direct contravention of what a materiality analysis requires.

The majority also sidesteps the Government’s closing arguments, which told the jury that it would only take “three years for [Caro] to be stepped down out of ADX and into a USP,” that “if Mr. Caro was given a light sentence, he may initially go to ADMAX, but he will be moved out to the USP on a three year program, well within the life of violence of Carlos Caro,” and that “in some time within three to five years he will be back at a USP, right where he stabbed Rick Benavidez, and right where he strangled Roberto Sandoval.” J.A. 923–24, 979. The majority rightly chastises the Government for misrepresenting Cunningham’s and Hershberger’s testimonies. Ante 12 n.3. 16 But the majority ignores the fact that materiality can turn on what the Government emphasizes in closing. In Kyles, for example, the Supreme Court found suppressed evidence to be material in part because it would have impeached two witnesses identified by the Government in closing as “the State’s two best witnesses.” 514 U.S. at 444–45. Just so here. Hard data about how long inmates are actually held at Florence ADMAX would have “undercut the prosecution” in closing by providing the jury with an objective baseline for how the BOP handles dangerous inmates like Caro. See id at 445.

Indeed, the BOP data would be material even if it did not “show[] that Caro would remain at Florence ADMAX for the rest of his life.” Ante 23. True, both sides testified

16 The majority errs in concluding that Caro has not challenged these statements— he did, in both his opening and reply briefs. See Appellant’s Opening Br. 31; Appellant’s Reply Br. 3, 14–15.

that although the “BOP does not permanently assign inmates to Florence ADMAX,” “some inmates take longer than the average five years to complete the step-down program.” Ante 23. But Cunningham repeatedly explained that he was hamstrung in his testimony by the BOP’s refusal to provide hard data. J.A. 699–702, 736–40, 792–98, 799–802. The majority notes that Cunningham “based his prediction on anecdotal examples of particularly dangerous inmates,” ante 11, ignoring that this is precisely the point: Because the BOP data was suppressed, Cunningham was deprived of accurate data and case studies. He was not able to conduct a risk assessment of Caro’s future dangerousness or provide evidence to support his contention that the BOP can securely house Caro. Had the BOP data been disclosed, Cunningham likely would have testified about the dozens of Florence ADMAX inmates who had been there for over a decade, including inmates who had likewise committed homicides within the BOP. He also would have testified about how the BOP actually addressed the security concerns of these other dangerous inmates. From these real examples, a juror could have concluded that the Government can house Caro securely and that executing him is unnecessary.

The majority claims that Caro had failed to show that the “statistical evidence he requested even existed” because “there is unrebutted evidence in the record that the BOP does not maintain a database of all the inmates ever housed at a particular institution.” Ante 24. But Caro had not requested a list of all inmates in the BOP system; most of the requested records concern only Florence ADMAX and the remaining records concern inmate homicides within the BOP. See supra Part I.B. Moreover, Cunningham’s two declarations and testimony effectively rebutted many of the Government’s arguments *62 about the BOP data’s existence by noting inconsistencies between the several Government declarations while clarifying exactly what records he needed. J.A. 126–43. Indeed, Cunningham had previously received from the BOP the exact type of records he requested, apparently without controversy. It strains plausibility that the BOP would not update their records about the inmates who commit violent acts behind bars and where they are held. See J.A. 39 (declaration of Cunningham stating that “it is patently inconceivable that BOP has not calculated detailed length of stay information regarding this unique facility housing the ‘worst of the worst’ when an in-house BOP research unit is available to examine such vitally important performance and outcome data.”). Whatever the measure of materiality, the BOP data requested by Caro undoubtedly does exist.

In sum, Caro sought information about how the BOP has managed similarly situated inmates—inmates who have committed assaults and even murders behind bars. He sought this information to prove that the BOP could manage him securely as well. In denying him this information, the Government deprived not only the jury of accurate data but also Caro’s expert of the ability to develop a risk assessment and rebut the Government’s expert. Had the jury known that the BOP securely houses other highly dangerous inmates and routinely keeps them in Florence ADMAX for well beyond three years, I am not confident that every juror would still have concluded that Caro’s future dangerousness justified the death penalty. And because a capital sentence in this context is not “worthy of confidence,” there is a “reasonable probability” that disclosure of the BOP data would have led to a “different result.” Kyles, 514 U.S. at 434 (quoting Bagley, *63 473 U.S. at 678, 682)). Reviewing the facts “in the light most favorable” to Caro, Poindexter, 492 F.3d at 267, I would find that the BOP data is material.

C.

But even if Caro has not met the favorability and materiality prongs of Brady, his claim is at worst one of the “atypical cases” in which “‘it is impossible to say whether’ requested information ‘may be relevant’” to the defendant’s case. King, 628 F.3d at 703 (quoting Pennsylvania v. Ritchie, 480 U.S. 39, 57 (1987)). Under our established precedent, the solution to a Brady problem created and perpetuated by Government suppression is not dismissal—it is remand for in camera review.

In King, the defendant was indicted for felony possession of a firearm that he said belonged instead to a cooperating witness named Bilal. Id. at 698–99. Bilal had also told police that King had kidnapped and assaulted him, but King was never federally indicted or convicted for the purported crime. Id. at 697. Before trial, King repeatedly requested and was repeatedly denied copies of Bilal’s grand jury testimony, which the Government claimed “contained no exculpatory information.” Id. at 698. At trial, King argued that the firearm belonged to Bilal, but without success. Id. at 698–99. The district court then applied an eight-level sentencing enhancement based on Bilal’s unsubstantiated claim that King had kidnapped him. Id. at 699.

On direct appeal, we sustained King’s Brady objection and vacated the firearms conviction. Id. at 704. We recognized that “a defendant cannot demonstrate that suppressed evidence would have changed the trial’s outcome if the Government prevents him from ever seeing that evidence.” Id. at 702. In these “atypical cases,” the defendant *64 is not required to “make a particular showing of the exact information sought and how it is material and favorable.” Id. at 703 (quoting Love v. Johnson, 57 F.3d 1305, 1313 (4th Cir. 1995)). Instead, “a defendant need only ‘make some plausible showing’ that exculpatory material exists.” Id. (quoting Ritchie, 480 U.S. at 58 n.15; Love, 57 F.3d at 1313). A “plausible showing” requires the defendant to “identify the requested confidential material with some degree of specificity.” Id. (quoting United States v. Trevino, 89 F.3d 187, 189 (4th Cir. 1996)). Once a defendant makes a “plausible showing,” he “becomes ‘entitled . . . to have the information’—not immediately disclosed to him—but ‘submitted to the trial court for in camera inspection’ to determine if in fact the information is Brady material subject to disclosure.” Id. (quoting Love, 57 F.3d at 1313); see also Ritchie, 480 U.S. at 58.

We concluded that King had made such a “plausible showing” that the grand jury transcript could be materially favorable to both his culpability and its sentence. King, 628 F.3d at 703. Even though “the jury disbelieved King’s story about Bilal,” we held that “it remains plausible that Bilal’s grand jury testimony contained information that might have affected that disbelief.” Id. at 704 (emphasis added). And because the district court judge credited Bilal’s statements about kidnapping, the grand jury transcript could reveal information that significantly reduced King’s sentence. Id.

King should have guided our decision here. Caro has identified specific records maintained by the BOP that would likely show the BOP’s ability to securely incarcerate him long-term in Florence ADMAX and would have likely allowed his expert to prepare an accurate risk assessment. Given what Caro has now uncovered, it is at least *65 “plausible” that the BOP data “contain[s] information that might have affected” the jury’s belief about Caro’s future dangerousness. See id. at 704. At the very least, Caro is entitled to have the district court review those records and determine whether their absence undermined confidence in the jury’s sentence—a sentence that will otherwise lead to Caro’s imminent execution.

III.

“[D]eath is a different kind of punishment from any other which may be imposed in this country.” Gardner v. Florida, 430 U.S. 349, 357 (1977) (plurality opinion). It is the “ultimate sanction,” Furman v. Georgia, 408 U.S. 238, 286 (1972) (Brennan, J., concurring)—there is no more severe or final punishment, nor any more grave exercise of state power. We must tread cautiously when the Government claims that a defendant is too dangerous to be kept alive—and then fights tooth and nail to prevent that defendant from accessing data that he says will prove otherwise. Justice demanded that Caro receive an opportunity to fully rebut the Government’s claim of dangerousness with information about how the Government handles those with equally dangerous histories. Because Caro was denied that opportunity, I respectfully dissent.