*434SEPARATE OPINIONS
Justice Scalia,with whom Justice Alito joins, concurring.
I join the Court’s opinion in full. I write separately only to address several aspects of the dissent.
1. The dissent’s lengthy excavation of the trial record is a puzzling exertion. The question presented for our review is whether a municipality is liable for a single Brady violation by one of its prosecutors, even though no pattern or practice of prior violations put the municipality on notice of a need for specific training that would have prevented it. See Brady v. Maryland, 373 U.S. 83, 83 S. Ct. 1194, 10 L. Ed. 2d 215 (1963). That question is a legal one: whether a Brady violation presents one of those rare circumstances we hypothesized in Canton’s footnote 10, in which the need for training in constitutional requirements is so obvious ex ante that the municipality’s failure to provide that training amounts to deliberate indifference to constitutional violations. See Canton v. Harris, 489 U.S. 378, 390, n. 10, 109 S. Ct. 1197, 103 L. Ed. 2d 412 (1989).
The dissent defers consideration of this question until the twenty-third page of its opinion. It first devotes considerable space to allegations that Connick’s prosecutors misunderstood Brady when asked about it at trial, see post, at 93-95, 179 L. Ed. 2d, at 447-448 (opinion of Ginsburg, J.), and to supposed gaps in the
[563 U.S. 73]
Brady guidance provided by Connick’s office to prosecutors, including deficiencies (unrelated to the specific Brady violation at issue in this case) in a policy manual published by Connick’s office three years after Thompson’s trial, see post, at 96-98, 179 L. Ed. 2d, at 448-449. None of that is relevant. Thompson’s failure-to-train theory at trial was not based on a pervasive culture of indifference to Brady, but rather on the inevitability of mistakes over enough iterations of criminal trials. The District Court instructed the jury it could find Connick deliberately indifferent if:
“First: The District Attorney was certain that prosecutors would confront the situation where they would have to decide which evidence was required by the constitution to be provided to an accused[;]
“Second: The situation involved a difficult choice, or one that prosecutors had a history of mishandling, such that additional training, supervision, or monitoring was clearly needed [; and]
“Third: The wrong choice by a prosecutor in that situation will frequently cause a deprivation of an accused’s constitutional rights.” App. 828.
That theory of deliberate indifference would repeal the law of Monel1 in favor of the Law of Large Numbers. Brady mistakes are inevitable. So are all species of error routinely confronted by prosecutors: authorizing a bad warrant; losing a Batson2 claim; crossing the line in closing argument; or eliciting hearsay that violates the Confrontation Clause. Nevertheless, we do not have “de facto respondeat superior liability,” Canton, 489 U.S., *435at 392, 109 S. Ct. 1197, 103 L. Ed. 2d 412, for each such violation under the rubric of failure to train simply because the municipality does not have a professional educational program covering
[563 U.S. 74]
the specific violation in sufficient depth.3 Were Thompson’s theory the law, there would have been no need for Canton’s footnote to confine its hypothetical to the extreme circumstance of arming police officers with guns without telling them about the constitutional limitations upon shooting fleeing felons; the District Court’s instructions cover every recurring situation in which citizens’ rights can be violated.
That result cannot be squared with our admonition that failure-to-train liability is available only in “limited circumstances,” id., at 387, 109 S. Ct. 1197, 103 L. Ed. 2d 412, and that a pattern of constitutional violations is “ordinarily necessary to establish municipal culpability and causation,” Board of Comm’rs of Bryan Cty. v. Brown, 520 U.S. 397, 409, 117 S. Ct. 1382, 137 L. Ed. 2d 626 (1997). These restrictions are indispensable because without them, “failure to train” would become a talismanic incantation producing municipal liability “ [i]n virtually every instance where a person has had his or her constitutional rights violated by a city employee”—which is what Monell rejects. Canton, 489 U.S., at 392, 109 S. Ct. 1197, 103 L. Ed. 2d 412. Worse, it would “engage the federal courts in an endless exercise of second-guessing municipal employee-training programs,” thereby diminishing the autonomy of state and local governments. Ibid.
2. Perhaps for that reason, the dissent does not seriously contend that Thompson’s theory of recovery was proper. Rather, it accuses Connick of acquiescing in that theory at trial. See post, at 102, 179 L. Ed. 2d, at 452. The accusation is false. Connick’s
[563 U.S. 75]
central claim was and is that failure-to-train liability for a Brady violation cannot be premised on a single incident, but requires a pattern or practice of previous violations. He pressed that argument at the summary judgment stage but was rebuffed. At trial, when Connick offered a jury instruction to the same effect, the trial judge effectively told him to stop bringing up the subject:
“[Connick’s counsel]: Also, as part of that definition in that same location, Your Honor, we would like to include language that says that deliberate indifference to training requires a pattern of similar violations and proof of deliberate indifference requires more than a single isolated act.
“[Thompson’s counsel]: That’s not the law, Your Honor.
“THE COURT: No, I’m not giving that. That was in your motion for summary judgment that I denied.” Tr. 1013.
*436Nothing more is required to preserve a claim of error. See Fed. Rule Civ. Proc. 51(d)(1)(B).4
3. But in any event, to recover from a municipality under 42 U.S.C. § 1983, a plaintiff must satisfy a “rigorous” standard of causation, Bryan Cty., 520 U.S., at 405, 117 S. Ct. 1382, 137 L. Ed. 2d 626; he must “demonstrate a direct causal link between the municipal action and the deprivation of federal rights,” id., at 404, 117 S. Ct. 1382, 137 L. Ed. 2d 626. Thompson
[563 U.S. 76]
cannot meet that standard. The withholding of evidence in his case was almost certainly caused not by a failure to give prosecutors specific training, but by miscreant prosecutor Gerry Deegan’s willful suppression of evidence he believed to be exculpatory, in an effort to railroad Thompson. According to Deegan’s colleague Michael Riehlmann, in 1994 Deegan confessed to him—in the same conversation in which Deegan revealed he had only a few months to live—that he had “suppressed blood evidence in the armed robbery trial of John Thompson that in some way exculpated the defendant.” App. 367; see also id., at 362 (“[Deegan] told me . . . that he had failed to inform the defense of exculpatory information”). I have no reason to disbelieve that account, particularly since Riehl-mann’s testimony hardly paints a flattering picture of himself: Riehl-mann kept silent about Deegan’s misconduct for another five years, as a result of which he incurred professional sanctions. See In re Riehlmann, 2004-0680 (La. 1/19/05), 891 So. 2d 1239. And if Riehlmann’s story is true, then the “moving force,” Bryan Cty., supra, at 404, 117 S. Ct. 1382, 137 L. Ed. 2d 626 (internal quotation marks omitted), behind the suppression of evidence was Deegan, not a failure of continuing legal education.
4. The dissent suspends disbelief about this, insisting that with proper Brady training, “surely at least one” of the prosecutors in Thompson’s trial would have turned over the lab report and blood swatch. Post, at 89, 179 L. Ed. 2d, at 449. But training must consist of more than mere broad encomiums of Brady: We have made clear that “the identified deficiency in a city’s training program [must be] closely related to the ultimate injury.” Canton, supra, at 391, 109 S. Ct. 1197, 103 L. Ed. 2d 412. So even indulging the dissent’s assumption that Thompson’s prosecutors failed to disclose the lab report in good-faith—in a way that could be prevented by training—what sort of training would have prevented the good-faith nondisclosure of a blood report not known to be exculpatory?
[563 U.S. 77]
Perhaps a better question to ask is what legally accurate training would have prevented it. The dissent’s suggestion is to instruct prosecutors to ignore the portion of Brady limiting prosecutors’ disclosure obligations to evidence that is “favorable to an accused,” 373 U.S., at 87, 83 S. Ct. 1194, 10 L. Ed. 2d 215. Instead, the dissent *437proposes that “Connick could have communicated to Orleans Parish prosecutors, in no uncertain terms, that, ‘[i]f you have physical evidence that, if tested, can establish the innocence of the person who is charged, you have to turn it over.’ ” Post, at 97, n. 13, 179 L. Ed. 2d, at 449 (quoting Tr. of Oral Arg. 34). Though labeled a training suggestion, the dissent’s proposal is better described as a sub silentio expansion of the substantive law of Brady. If any of our cases establishes such an obligation, I have never read it, and the dissent does not cite it.5
Since Thompson’s trial, however, we have decided a case that appears to say just the opposite of the training the dissent would require: In Arizona v. Youngblood, 488 U.S. 51, 58, 109 S. Ct. 333, 102 L. Ed. 2d 281 (1988), we held that “unless a criminal defendant can show bad faith on the part of the police, failure to preserve potentially useful evidence does not constitute a denial of due process of law.” We acknowledged that “Brady . . . makes the good or bad faith of the State irrelevant when the State fails to disclose to the defendant material exculpatory evidence,” but concluded that “the Due Process Clause requires a different
[563 U.S. 78]
result when we deal with the failure of the State to preserve evidentiary material of which no more can be said than that it could have been subjected to tests, the results of which might have exonerated the defendant.” Id., at 57, 109 S. Ct. 333, 102 L. Ed. 2d 281. Perhaps one day we will recognize a distinction between good-faith failures to preserve from destruction evidence whose inculpa-tory or exculpatory character is unknown, and good-faith failures to turn such evidence over to the defense. But until we do so, a failure to train prosecutors to observe that distinction cannot constitute deliberate indifference.
5. By now the reader has doubtless guessed the best-kept secret of this case: There was probably no Brady violation at all—except for Deegan’s (which, since it was a bad-faith, knowing violation, could not possibly be attributed to lack of training).6 The dissent surely knows this, which is why it leans heavily on the fact that Connick conceded that Brady was violated. I can honor that concession in my analysis of the case because even if it extends beyond Deegan’s deliberate actions, it remains irrelevant to Connick’s training obligations. For any Brady violation apart from Dee-gan’s was surely on the very frontier of our Brady jurisprudence; Connick could not possibly have been on notice *438decades ago that he was required to instruct his prosecutors to respect a right to untested evidence that we had not (and still have not) recognized. As a consequence, even if I accepted the dissent’s conclusion that failure-to-train liability could be premised on a single Brady error, I could not agree that the lack of an accurate training regimen caused the violation Connick has conceded.
[563 U.S. 79]
. Monell v. New York City Dept. of Social Servs., 436 U.S. 658, 98 S. Ct. 2018, 56 L. Ed. 2d 611 (1978).
. Batson v. Kentucky, 476 U.S. 79, 106 S. Ct. 1712, 90 L. Ed. 2d 69 (1986).
. I do not share the dissent’s confidence that this result will be avoided by the instruction’s requirement that “ ‘more likely than not the Brady material would have been produced if the prosecutors involved in his underlying criminal cases had been properly trained, supervised or monitored regarding the production of Brady evidence.’ ” Post, at 102, n. 17, 179 L. Ed. 2d, at 452 (quoting Tr. 1100). How comforting that assurance is depends entirely on what proper training consists of. If it is not limited to training in aspects of Brady that have been repeatedly violated, but includes—as the dissent would have it include here—training that would avoid any one-time violation, the assurance is no assurance at all.
. The dissent’s contention that “[t]he instruction Connick proposed resembled the charge given by the District Court,” post, at 102, n. 18, 179 L. Ed. 2d, at 452, disregards his requested instruction concerning the necessity of a pattern of prior violations. It is meaningless to say that after “the court rejected [Connick’s] categorical position,” as it did, he did not “assail the District Court’s formulation of the deliberate indifference instruction,” post, at 103, n. 18, 179 L. Ed. 2d, at 452. The prior-pattern requirement was part of Connick’s requested formulation of deliberate indifference: “To prove deliberate indifference, a plaintiff must demonstrate ‘at least a pattern of similar violations arising from training that is so clearly inadequate as to be obviously likely to result in a constitutional violation.’ ” Record, Doc. 94, p. 18 (emphasis added).
. What the dissent does cite in support of its theory comes from an unexpected source: Connick’s testimony about what qualifies as Brady material. See post, at 98, n. 13, 179 L. Ed. 2d, at 449. (“Or Connick could have told prosecutors what he told the jury when he was asked whether a prosecutor must disclose a crime lab report to the defense, even if the prosecutor does not know the defendant’s blood type: ‘Under the law it qualifies as Brady material’ ’’ (quoting Tr. 872)). Given the effort the dissent has expended persuading us that Connick’s understanding of Brady is profoundly misguided, its newfound trust in his expertise on the subject is, to the say the least, surprising.
. The dissent’s only response to this is that the jury must have found otherwise, since it was instructed that “ ‘[f]or liability to attach because of a failure to train, the fault must be in the training program itself, not in any particular prosecutor.’ "Post, at 105, n. 20, 179 L. Ed. 2d, at 454 (quoting Tr. 1098). But this instruction did not require the jury to find that Deegan did not commit a bad-faith, knowing violation; it merely prevented the jury from finding that, if he did so, Connick was liable for a failure to train. I not only agree with that; it is part of my point.