concurring in part and concurring in the judgment:
To say this case is troubling is a serious understatement. James Dennis was convicted of murder and sentenced to death based almost entirely upon the testimony of three problematic eyewitnesses and despite a dearth of physical evidence. On direct appeal, the Pennsylvania Supreme Court affirmed his conviction and death sentence in an opinion that is no credit to that court’s usual standards. See Dennis I, 552 Pa. 331, 715 A.2d 404 (1998). It rejected in a mere three sentences Dennis’s Brady claim with respect to the Cason receipt, a piece of evidence thoroughly described in today’s Majority opinion. Here is the entirety of the state court’s analysis:
Finally, it is clear that there clearly was no Brady violation. The DPW receipt was not exculpatory, because it had no bearing on Appellant’s alibi, and there is no evidence that the Commonwealth withheld the receipt from the defense. Accordingly, Appellant’s claims of ineffectiveness regarding Cason and the DPW receipt have no arguable merit.
Id. at 408.
Perhaps the most remarkable aspect of that drive-by discussion is the assertion that the Cason receipt was not exculpatory because “it had no bearing on [Dennisj’s alibi.” Id. In reality, the pertinence and importance of the receipt could not be more glaring. It shows exactly what time witness Latanya Cason received her public assistance check, thus shifting the timeline of events that she laid out during her trial testimony so that, instead of contradicting Dennis’s testimony, she almost perfectly corroborated his alibi. The previously-undisclosed receipt thus transforms Cason from a damning prosecution witness into a powerful witness for the defense.
Every judge of our en banc Court has now concluded that the Pennsylvania Supreme Court’s contrary determination was not only wrong, but so obviously wrong that it cannot pass muster even under AEDPA’s highly-deferential standard of review. In other words, it is the unanimous view of this Court that any fairminded jurist must disagree with the Dennis I court’s assessment of the materiality and favorability of the Cason receipt. Yet somehow a majority of the Pennsylvania Supreme Court endorsed Dennis’s conviction and death sentence. The lack of analytical rigor and attention to detail in that decision on direct appeal is all the more painful to contemplate because the proof against Dennis is far from overwhelming. He may be innocent.
*347But the strength of the case against James Dennis need not be the focus of our attention. This case can and should be resolved on a single point: the Brady claim concerning the Cason receipt. That is one reason why I cannot join the more expansive opinion of my colleagues in the Majority. Their correct conclusion that the error in Dennis I regarding the Cason receipt is by itself enough to warrant habeas relief means that we have no call to address the Brady claims with respect to the Howard police activity report and the Frazier documents. And, in fact, I disagree with the' Majority’s analysis of those latter two claims and fully agree with my dissenting colleagues’ rejection of them, which is another reason I cannot join the Majority opinion.
Moreover, I also agree with the Dissent’s position,1 set forth in its discussion of the Cason receipt, that imposing a “reasonable diligence” requirement upon defense counsel does not violate a clearly established holding of the Supreme Court. The “reasonable diligence” requirement is, in effect, a rule that a Brady claim will not lie when the evidence in question was available to the defense by the exercise of reasonable diligence. E.g., Brown v. Cain, 104 F.3d 744, 750 (5th Cir. 1997). We are obligated by AEDPA to uphold a state court’s decision unless it is “contrary to, or involved an unreasonable application of, clearly established Federal law, as determined by the Supreme Court of the United States.” 28 U.S.C. § 2254(d)(1). Under AEDPA, whether any of us thinks that imposing a reasonable diligence requirement is a good idea or the best interpretation of Brady is irrelevant. What matters is that one can reasonably perceive such a requirement being allowed by Supreme Court jurisprudence.2 We ourselves have applied it repeatedly,3 so we can hardly *348say that it constitutes an unreasonable application of federal law.
Of course, the Pennsylvania Supreme Court never said anything at all in its Dennis I decision about defense counsel’s lack of diligence in locating the Cason receipt. But, under Harrington v. Richter, habeas review requires that we engage in so-called “gap-filling,” and apply AEDPA deference to whatever reasonable “arguments or theories ... could have supported[ ] the state court’s decision,” if that decision does not provide reasoning for its conclusions.4 562 U.S. 86, 102, 131 S.Ct. 770, 178 L.Ed.2d 624 (2011). Thus, despite the fact that the Pennsylvania Supreme Court never itself discussed diligence, Richter might prompt us to apply a reasonable diligence requirement and reject Dennis’s Cason receipt Brady claim — exactly as the Dissent has suggested — if there were a gap in the state-court decision for us to fill. The problem I have with the Dissent is that I see no gap in the state court’s reasoning, at least not in the sense contemplated in Richter. My dissenting colleagues are not filling a gap here; they are rewriting the opinion of the Pennsylvania Supreme Court, adding and then elaborating a theory that was never litigated in state court.
The reality of what happened in Dennis I is more straightforward. The Pennsylvania Supreme Court simply erred. Its opinion stated both that “the police came into possession of’ the Cason receipt and that “there [was] no evidence that the Commonwealth withheld the receipt from the defense.” Dennis I, 715 A.2d at 408. There was, however, no recognition that those statements are fundamentally at odds. Under the Supreme Court’s opinion in Kyles v. Whitley, any evidence in the possession of the police is, for Brady purposes, also in the possession of the prosecution. 514 U.S. 419, 437, 115 S.Ct. 1555, 131 L.Ed.2d 490 (1995). If a piece of favorable, material evidence is in the possession of the police but is not turned over to the defense, it is necessarily withheld by the prosecution in violation of Brady. See id. (prosecutors are responsible for “any favorable evidence known to the others acting on the government’s behalf in the case, including the police”).
By entirely failing to apply Kyles, the Pennsylvania Supreme Court acted “contrary to ... clearly established Federal law, as determined by the Supreme Court of the United States.” 28 U.S.C. § 2254(d)(1).5 In light of the state court’s error, I would review Dennis’s Brady claim with respect to the Cason receipt “unencumbered by the deference AEDPA normally requires,” Panetti v. Quarterman, 551 U.S. 930, 948, 127 S.Ct. 2842, 168 L.Ed.2d 662 (2007), to determine whether Dennis is “in custody in violation of the Constitution or laws ... of the United *349States.” 28 U.S.C. § 2254(a).6 On that de novo review, I would hold that the evidence in question meets all three requirements of Brady — the Cason receipt is material and favorable, and it was suppressed by the Commonwealth — for the reasons set out in Part III.A of the Majority opinion. I therefore concur in the judgment. I also agree with Part II of the Majority opinion and write separately to explain my view of the limits of Richter gap-filling and the proper scope of AEDPA deference.
Recall that in Dennis I, the Pennsylvania Supreme Court said, “there is no evidence that the Commonwealth withheld the [Cason] receipt from the defense.” 715 A.2d at 408. My dissenting colleagues believe “it is not clear what the court meant by [that].” (J. Fisher Dissent Op. at 360.) They then proceed to fill the “gap” they think is created by the ambiguity they perceive, saying, “the Pennsylvania Supreme Court could have meant that the receipt was not withheld because it was available to the defense with reasonable diligence.” (J. Fisher Dissent Op. at 362.)
The precedent that establishes a gap-filling requirement, Richter, dealt with a state court decision that was unsupported by any reasoning. 562 U.S. at 96-97, 131 5.Ct. 770. The state court issued a summary order, with no written opinion, denying a prisoner’s ineffective assistance of counsel claim. Id. The gap in the state court’s reasoning was obvious — there was no reasoning at all. The Supreme Court held that, even in those circumstances, “[w]here a state court’s decision is unaccompanied by an explanation, the habeas petitioner’s burden still must be met by showing there was no reasonable basis for the state court to deny relief.” Id. at 98, 131 S.Ct. 770. Thus federal courts must fill gaps in a state court’s reasoning so that there is something against which to measure a petitioner’s efforts. In short, “a habe-as court must determine what arguments or theories ... could have supported[ ] the state court’s decision” and afford AEDPA deference to those theories. Id. at 102, 131 S.Ct. 770.
Premo v. Moore extended RichtePs gap-filling directive a bit beyond cases devoid of all reasoning. 562 U.S. 115, 131 S.Ct. 733, 178 L.Ed.2d 649 (2011). There, a prisoner claimed ineffective assistance of counsel because his attorney had failed to file a motion to suppress a confession. Id. at 119, 131 S.Ct. 733. In concluding that such a motion “would have been fruitless,” id. the state court’s opinion expressly referenced *350trial counsel’s explanation that “suppression would serve little purpose” because the defendant had made full and admissible confessions to others. Id. at 123, 131 S.Ct. 733. The state court did not, however, specify which of the two prongs of the ineffective assistance of counsel standard from Strickland v. Washington, 466 U.S. 668, 104 S.Ct. 2052, 80 L.Ed.2d 674 (1984) — deficient performance or prejudice — formed the basis of its rejection of the claim. Premo, 562 U.S. at 123, 131 S.Ct. 733. The Supreme Court therefore held that the Court of Appeals for the Ninth Circuit had to fill that gap by assuming “that both findings would have involved an unreasonable application of clearly established federal law.” Id. Critical to the ultimate denial of habeas relief, the Supreme Court believed that the state court’s justification for rejecting the petitioner’s claim was sufficient to address either prong of Strickland.7 Accordingly, the Supreme Court’s decision was not an exercise in speculation but was rooted in the state court’s actual reasoning. Tremo did not require consideration of an entirely new argument that had not already been identified and accepted by the state court. See id. at 124, 131 S.Ct. 733 (“[T]he [state court’s] first and independent explanation — that suppression would have been futile — confirms that [counsel’s] representation was adequate under Strickland.”). The “gap” that the Court filled was thus quite narrow.
The very next year, the Supreme Court put a limit on gap-filling. In Lafler v. Cooper, it upheld a grant of habeas corpus. — U.S. -, 132 S.Ct. 1376, 182 L.Ed.2d 398 (2012). The petitioner, Anthony Cooper, had shot at a woman’s head but missed, instead hitting her in the buttock, hip, and abdomen. Id. at 1383. The prosecution offered Cooper two plea deals, and Cooper expressed interest. Id. He ended up rejecting the offers, though, because (he later alleged) his attorney convinced him that the prosecution would be unable to establish intent to murder because he shot his victim below the waist. Id. After he was convicted on all charges, Cooper claimed ineffective assistance of counsel. Id. The Michigan Court of Appeals rejected his claim, analyzing it as follows:
[T]he record shows that defendant knowingly and intelligently rejected two plea offers and chose to go to trial. The record fails to support defendant’s contentions that defense counsel’s representation was ineffective because he rejected a defense based on [a] claim of self-defense and because he did not obtain a more favorable plea bargain for defendant.
People v. Cooper, No. 250583, 2005 WL 599740, at *1 (Mich. Ct. App. Mar. 15, 2005) (per curiam) (internal citations omitted). After the district court granted Cooper’s petition for habeas relief, the Sixth Circuit affirmed, emphasizing the problem in the state court’s decision with this comment: “it is not clear from the [state] court’s abbreviated discussion (only two sentences of the opinion is even arguably responsive to petitioner’s claim) what the court decided, or even whether the correct *351legal rule was identified.” Cooper v. Lafler, 376 Fed.Appx. 563, 568-69 (6th Cir. 2010), vacated by — U.S.-, 132 S.Ct. 1376, 182 L.Ed.2d 398 (2012).
While it ultimately affirmed the habeas decision, the Supreme Court concluded that the state court’s two-sentence analysis “may not be quite so opaque as the Court of Appeals for the Sixth Circuit thought.... ” Lafler, 132 S.Ct. at 1390. The state court had identified Cooper’s ineffective-assistance-of-counsel claim, but had failed to apply the proper Strickland standard to assess it. Instead, the state court had “simply found that respondent’s rejection of the plea was knowing and voluntary.” Id. Although the Michigan court recited the Strickland standard, the Supreme Court concluded that the state court had mistakenly relied upon an entirely different standard (ie., the “knowing and voluntary” standard), which was contrary to Strickland. By relying upon the wrong standard altogether, “the state court’s adjudication was contrary to clearly 'established federal law.” Id. As a consequence, the Supreme Court declined to apply AEDPA deference to the state court decision and, instead, engaged in de novo review of Cooper’s Strickland claim, concluding that his counsel’s deficient performance and the prejudice therefrom required relief. Id. at 1390-91. The Supreme Court’s analysis in Lafler suggests that we should be hesitant to deem a state court opinion to be so lacking in analysis that it is comparable to an “order ... unaccompanied by an opinion explaining [its] reasons.” Richter, 562 U.S. at 98, 131 S.Ct. 770. In other words, we ought not engage in error correction under the guise of gap-filling.
That holds true here. In Dennis I, the Pennsylvania Supreme Court correctly identified Brady and its requirement that, for relief to be warranted, the evidence in question must be both exculpatory and withheld. Nevertheless, the court applied a standard contrary to Brady and its progeny when it concluded that the prosecution did not withhold evidence that the police had in their possession. Cf. Sears v. Upton, 561 U.S. 945, 952, 130 S.Ct. 3259, 177 L.Ed.2d 1025 (2010) (per curiam) (“Although the court appears to have stated the proper ... standard, it did not correctly conceptualize how that standard applies to the circumstances of this case.”). Kyles is very clear in explaining that, for purposes of a Brady analysis, the prosecution functionally possesses all favorable evidence in the possession of the police. See 514 U.S. at 437, 115 S.Ct. 1555 (“[T]he individual prosecutor has a duty to learn of any favorable evidence known to the others acting on the government’s behalf in the case, including the police.”). Just as the Michigan state court in Lafler failed to apply Strickland to assess an ineffective assistance claim, so too the Pennsylvania Supreme Court failed to apply Kyles to assess Dennis’s Brady claim with respect to the Cason receipt. Rather than applying Kyles, the court simply found that there was no evidence that the prosecutor possessed the Cason receipt. Compare Lafler, 132 S.Ct. at 1390 (“Rather than applying Strickland, the state court simply found that respondent’s rejection of the plea was knowing and voluntary. An inquiry into whether the rejection of a plea is knowing and voluntary, however, is not the correct means by which to address a claim of ineffective assistance of counsel.”). Lafler implies a limit on the gap-filling called for by Richter and Premo. As was done in Lafler, we should take the state court’s decision as written, rather than construct our own “not unreasonable” theory to justify that court’s conclusion.
Justice Scalia’s dissent in Lafler further supports the analogy between that case and this one. Indeed, his opinion reads much like the Dissent here. First, he pointed out that the Michigan state court *352had recited the Strickland standard. Lafler, 132 S.Ct. at 1396 (Scalia, J., dissenting). He next read the subsequent paragraph of the state court’s decision as an attempt to apply that standard. Id. He then concluded that the state court did not apply a standard “contrary to” federal law. Instead, by direct analogy to Premo, he argued that his colleagues should have assessed whether the state court opinion constituted an unreasonable application of clearly established federal law, subject to Richter’s gap filling requirement:
Since it is ambiguous whether the state court’s holding was based on a lack of prejudice or rather the court’s factual determination that there had been no deficient performance, to provide relief under AEDPA this Court must conclude that both holdings would have been unreasonable applications of clearly established law.
Id. Justice Scalia’s effort to salvage the state court decision in Lafler provides some support for the Dissent’s approach here. But Justice Scalia was himself writing a dissent. Had the Supreme Court wanted us to save every problematic state court opinion by gap-filling and application of AEDPA deference, Justice Scalia’s opinion would have been the majority position.
I can discern no ambiguity in the Pennsylvania Supreme Court’s Brady analysis regarding the Cason receipt. The Dennis I opinion is clear about it. Very brief and very wrong, but clear. The analysis under the suppression prong of Brady can be distilled from two sentences of the opinion. First, the court says, “During their investigation ... the police came into possession of’ the Cason receipt.8 Dennis I, 715 A.2d at 408. It then says, “there is no evidence that the Commonwealth withheld the receipt from the defense.” Id. If one follows the instruction of Kyles, those two statements are impossible to harmonize. But if one ignores Kyles and assumes there exists some dividing line between the police and the prosecution, the court’s reasoning is plain. To the Pennsylvania Supreme *353Court, the fact that the police had the receipt does not mean that the Common-ivealth had the receipt, and thus the Commonwealth did not suppress what it did not have. There is no hint that “reasonable diligence” was part of the analysis. The Commonwealth did not advance a reasonable diligence argument,9 nor did the court reference a diligence requirement anywhere in its opinion. In failing to apply Kyles, the state court’s opinion was “contrary to” and “involved an unreasonable application of[ ] clearly established Federal law, as determined by the Supreme Court of the United States.” 28 U.S.C. § 2254(d)(1).
My dissenting colleagues treat the contradictory sentences in Dennis I like a “Magic Eye” image, staring past the obvious error until the illusion of a fillable gap materializes. They do so, I assume, because it is. hard to accept that a court would make such a clear error of law: How could the state court possibly have concluded both that the police possessed the receipt and that the prosecution did not withhold it? That conclusion makes absolutely no sense if one assumes the state court knew of and applied Kyles. See Lopez v. Schriro, 491 F.3d 1029, 1046 (9th Cir. 2007) (Thomas, J., concurring in part and dissenting in part) (noting that we start with the “presumption that state judges know and follow the law”). But state courts, just like us, do sometimes err. And when they do, we are not free to label significant errors as “gaps” to be corrected under Richter and Premo.
Limiting our habeas review to the actual, expressed reasoning of a state court is itself a form of deference. The principles of comity and federalism underlying AED-PA’s highly-deferential standard compel us to acknowledge the state court’s reasoning if we can fairly discern it. See Ylst v. Nunnemaker, 501 U.S. 797, 803, 111 S.Ct. 2590, 115 L.Ed.2d 706 (1991) (describing an “unexplained” state-court order as one from which that court’s rationale is “undis-coverable”).10 We would do real damage to those principles were we to begin re-writing state court opinions to save them. Sometimes what appears to be a fundamental misstep is exactly that. Since the passage of AEDPA, the narrow purpose of federal habeas review has been to address just such missteps. See Richter, 562 U.S. at 103, 131 S.Ct. 770 (“As a condition for obtaining habeas corpus from a federal court, a state prisoner must show that the state court’s ruling on the claim being presented in federal court was so lacking in justification that there was an error well understood and comprehended in existing law beyond any possibility for fairminded disagreement.”).
There is yet another reason to think that Dennis I presents nothing more complicat*354ed than a Kyles error: the Commonwealth advocated it. Before the Pennsylvania Supreme Court, the Commonwealth advanced the incorrect theory that it was not required to turn over favorable evidence in the possession of the police. It emphasized that, “even though Cason claims in her affidavit that [the receipt] was taken by the police,” the failure to produce that document could not constitute a Brady violation because “there [wa]s no reason to believe it was in the Commonwealth’s possession to be produced.” (App. 2026.) That argument presupposes, contrary to Kyles, that there exists a divide between discoverable evidence taken by the police and discoverable evidence in the prosecutor’s case file.
At the time, that argument may have had some basis in Pennsylvania law, although it was already untenable because of Kyles. In 1995, when Kyles was decided, the Pennsylvania rules governing discovery and evidence disclosure were not based on the premise that evidence possessed by the police is possessed by the prosecution. See Pa. R. Crim. Pro. 305B (Repealed) (requiring mandatory disclosure of evidence favorable to the accused only when it “is within the possession or control of the attorney for the Commonwealth”). Even after Kyles was decided, the Pennsylvania Superior Court continued to hew to the outmoded state-law rule. See Commonwealth v. McElroy, 445 Pa.Super. 336, 665 A.2d 813, 819 (1995). The Pennsylvania Supreme Court likewise continued to apply its discovery rules as written. See Commonwealth v. Gribble, 550 Pa. 62, 703 A.2d 426, 435-36 (1997). It did not explicitly abrogate the faulty state rule of discovery until 2001. See Commonwealth v. Burke, 566 Pa. 402, 781 A.2d 1136, 1142 (2001). Dennis I was decided in 1998. Thus,'the court was not leaving agap in its Dennis I opinion. It was accepting the Commonwealth’s unsound argument, and it practically said so.
The wisdom of Richter gap-filling is open to reasonable criticism. A widely respected judge has expressed the view that gap-filling is unfair and incentivizes unreasoned decisions; it is a perspective that my colleague Judge Hardiman evidently shares, as described in his Dissent. See Mann v. Ryan, 774 F.3d 1203, 1225 (9th Cir. 2014) (Kozinski, J., concurring in part and dissenting in part) (Richter “has'the perverse effect of encouraging state courts to deny relief summarily, to insulate their orders from tinkering by the federal courts.”), on reh’g en banc, No. 09-99017, 828 F.3d 1143, 2016 WL 3854234 (9th Cir. July 15, 2016). Given those criticisms, it has been suggested that we should engage in Richter gap-filling, and thus apply AEDPA deference, even when a state court does give a reasoned basis for its conclusions. See id. at 1224 (Kozinski, J., concurring in part and dissenting in part) (“After Richter, it seems clear that we should assess the reasonableness of a state court’s decision, not its reasoning.”). Judge Hardiman would follow that approach here. (See J. Hardiman Dissent Op. at 370 (“I would hold that regardless of the thoroughness — or even the correctness — of the Pennsylvania Supreme Court’s stated reasoning, its judgment may not be upset so long as its decision did not contravene or unreasonably apply clearly established federal law....”).) And, indeed, his approach may have some appeal as a matter of policy — he has identified those policy justifications well — but, as a matter of law, I do not believe we can go so far. Lafler does not accept that logic.
Nor does the Supreme Court’s opinion in Wetzel v. Lambert, a post -Richter decision in which the Court dealt with a fully-reasoned (i.e., gapless) state court opinion. — U.S. -, 132 S.Ct. 1195, 182 L.Ed.2d 35 (2012) (per curiam). Wetzel described the required analytical path as follows:
*355Under § 2254(d), a habeas court must determine what arguments or theories supported ... the state court’s decision; and then it must ask whether it is possible fairminded jurists could disagree that those arguments or theories are inconsistent with the holding in a prior decision of this Court.
Id. at 1198 (quoting Richter, 562 U.S. at 102, 131 S.Ct. 770). The ellipsis in that quotation is significant, as the Court wholly excised the “or, as here, could have supported” language from its quotation of Richter when describing how federal courts review a reasoned state-court decision. Compare supra note 4. Rather than extending Richter, both Lafler and Wetzel suggest that gap-filling is reserved for only those cases where we cannot discern the basis for the state court’s conclusions.11
That is not the case here. Were Dennis in exactly the same position but the Den*356nis I opinion contained one or two fewer sentences, there would perhaps be a gap to fill and I would be joining my dissenting colleagues in applying AEDPA deference, but there is no gap. The Dennis I opinion suffers from erroneous and not opaque reasoning. It may seem odd that so much hinges on so little, with a man’s life depending on the difference between bad reasoning and no reasoning. That, however, is the analytical distinction drawn by Supreme Court precedent, including Richter, Premo, and Lafler.12
Given the magnitude of the Pennsylvania Supreme Court’s error regarding the *357Cason receipt, this case presents the sort of “extreme malfunction[ ] in the state criminal justice system” that demands our intervention. Richter, 562 U.S. at 102, 131 S.Ct. 770 (internal quotation marks omitted). I therefore concur in Part III.A of the Majority’s opinion, insofar as it explains why it is proper to grant Dennis habeas relief on de novo review of the Cason receipt Brady claim, and I concur in Part II of the Majority opinion and in the judgment.
. All references to the "Dissent” refer to Judge Fisher’s dissenting opinion, unless the reference is explicitly made to Judge Hardi-man’s dissent.
. Although the Majority is correct that the “Supreme Court has never recognized an affirmative due diligence duty of defense counsel as part of Brady" (Majority Op. at 290), there is no Supreme Court opinion that forecloses the adoption of that duty. The Supreme Court has emphasized that the Brady rule requires disclosure of evidence that is "unknown to the defense,” United States v. Agurs, 427 U.S. 97, 103, 96 S.Ct. 2392, 49 L.Ed.2d 342 (1976), and that the rule is rooted in “the defendant's right to a fair trial,” id. at 108, 96 S.Ct. 2392. Based on that language, several courts of appeals have concluded that information is not unknown to the defense for Brady purposes if it can be obtained by the exercise of reasonable diligence, and that requiring diligence on the part of defense counsel does not implicate the right to a fair trial. See, e.g., Lugo v. Munoz, 682 F.2d 7, 10 (1st Cir. 1982) ("Since the information at issue here- was available to the defense attorney through diligent discovery, we find that the prosecutor’s omission was not of sufficient significance to result in the denial of the defendant’s right to a fair trial” (internal quotation marks omitted).);,United States v. Brown, 628 F.2d 471, 473 (5th Cir. Unit A 1980) ("Truth, justice, and the American way do not ... require the Government to discover and develop the defendant's entire defense.”); United States v. Hedgeman, 564 F.2d 763, 769 (7th Cir. 1977) (establishing a diligence requirement and noting that "the prosecutor will not have violated his constitutional duty of disclosure unless his omission is of sufficient significance to result in the denial of the defendant’s right to a fair trial”). The Dissent has also collected cases to that effect. (See J. Fisher Dissent Op. at 362-63 n.l.) In any event, on AEDPA review it is sufficient for our purposes that there is no Supreme Court decision clearly holding that there is not a reasonable diligence requirement. See Harrington v. Richter, 562 U.S. 86, 103, 131 S.Ct. 770, 178 L.Ed.2d 624 (2011) (noting that a state-court error on habeas review must be one that is "well understood and comprehended in existing law beyond any possibility for fairminded disagreement”).
.See Grant v. Lockett, 709 F.3d 224, 231 (3d Cir. 2013) ("It is therefore clear that trial counsel could have discovered [the otherwise-suppressed evidence] had he exercised rea*348sonable diligence.”); United States v. Pelullo, 399 F.3d 197, 213 (3d Cir. 2005) ("[T]he burden is on the defendant to exercise reasonable diligence.”); United States v. Starusko, 729 F.2d 256, 262 (3d Cir. 1984) (''[T]he government is not obliged under Brady to furnish a defendant with information which he already has or, with any reasonable diligence, he can obtain himself” (internal quotation marks omitted).).
. More specifically, Richter says: "Under § 2254(d), a habeas court must determine what arguments or theories supported or, as here, could have supported, the state court's decision; and then it must ask whether it is possible fairminded jurists could disagree that those arguments or theories are inconsistent with the holding in a prior decision of th[e Supreme] Court.” 562 U.S. at 102, 131 S.Ct. 770.
. At the same time, the court went so far astray in applying Brady that its decision also "involved an unreasonable application of ... clearly established Federal law....” 28 U.S.C. § 2254(d)(1).
. It is important to understand the interplay between §§ 2254(a) and 2254(d). "Section 2254(a) permits a federal court to entertain only those applications alleging that a person is in state custody ‘in violation of the Constitution or laws or treaties of the United States.' " Cullen v. Pinholster, 563 U.S. 170, 181, 131 S.Ct. 1388, 179 L.Ed.2d 557 (2011). Section 2254(d) imposes an "additional restriction” on habeas relief in cases where a claim “has been adjudicated on the merits in State court proceedings.” Id. (internal quotation marks omitted). In those circumstances, habeas relief is barred unless the state court’s decision is "contrary to, or involved an unreasonable application of, clearly established Federal law.” 28 U.S.C. § 2254(d)(1). Section 2254(d) thus sets forth a necessary, but not sufficient, prerequisite to habeas relief only for those claims adjudicated on the merits in state court. If that high bar is cleared — i.e., the state court's decision is so unreasonable or contrary to federal law as established by the Supreme Court — we are still restricted to granting habeas relief only if the petitioner has shown he is in custody in violation of federal law under § 2254(a). In that second analysis, we review the petitioner’s claim de novo, without deference to the state court’s legal conclusions. Panetti, 551 U.S. at 953, 127 S.Ct. 2842 (“When a state court’s adjudication of a claim is dependent on an antecedent unreasonable application of federal law, the requirement set forth in § 2254(d)(1) is satisfied. A federal court must then resolve the claim without the deference AEDPA otherwise requires.”).
. See Premo, 562 U.S. at 126-27, 131 S.Ct. 733 (on performance: "It is not clear how the successful exclusion of the confession would have affected counsel’s strategic calculus. The prosecution had at its disposal two witnesses able to relate another confession.... Moore’s counsel made a reasonable choice to opt for a quick plea bargain.”); id. at 129, 131 S.Ct. 733 (on prejudice: "The state court here reasonably could have determined that Moore would have accepted the plea agreement even if his second confession had been ruled inadmissible. By the time, the plea agreement cut short investigation of Moore's crimes, the State’s case was already formidable and included two witnesses to an admissible confession.”).
. The Commonwealth argues that this sentence is not necessarily a factual finding to which we must defer under § 2254(e)(1), but was instead the Pennsylvania Supreme Court’s recapitulation of Dennis’s argument. The Majority rightly rejects that argument. (See Majority Op. at 288 n.17.) The plain language of Dennis I indicates that the statement was a finding of fact. See Paulson v. Newton Corr. Facility, Warden, 703 F.3d 416, 420 (8th Cir. 2013) (interpreting, in a habeas case, a state-court opinion consistent with its "plain language”). When the Pennsylvania Supreme Court was referring to arguments from the parties, it said so: in the very next paragraph of that opinion, every sentence contains some version of the words “appellant argues.” No such language appears in the disputed sentence (or its entire surrounding paragraph, for that matter). Thus, it certainly appears that the Pennsylvania Supreme Court was making a statement of historical fact when it said that "the police came into possession of” the Cason receipt. Dennis I, 715 A.2d at 408.
Without the deference afforded to an express factual finding, it would be an open question whether the police actually possessed the Cason receipt. When Dennis first offered Cason’s affidavit alleging that the police took her receipt, he himself argued that a "remand for an evidentiary hearing” would be "necessary to establish the record” before the Brady issue could be resolved. (App. 2012; see also App. 1891, 2021.) Likewise, the Commonwealth understood Cason’s affidavit to be merely a proffer of her "proposed testimony,” and argued that such testimony would have lacked the support of "competent evidence.” (App. 1923.) Further complicating matters, Cason's 1997 recollection of her interview with the police is in conflict with the police’s contemporaneous record of that encounter in 1992 (which did not enter the court record until after Dennis I, during PCRA proceedings). Were we here on de novo review of that factual finding, we could well question whether the police did, in fact, have the Cason receipt. As it stands, the state court’s factual findings are "presumed to be correct.” 28 U.S.C. § 2254(e)(1).
. In its sur-reply brief before the state court, the Commonwealth mentioned the potential "public availability” of the receipt. (App. 2026.) Under Pennsylvania law, however, arguments raised for the first time in reply briefs are generally regarded as waived. Commonwealth v. Potts, 388 Pa.Super. 593, 566 A.2d 287, 296 (1989).
. In Ylst, the Supreme Court held that when there is one reasoned state judgment rejecting a federal claim, any later unexplained orders upholding that judgment or rejecting the same claim will be presumed to rest upon the same ground. 501 U.S. at 803, 111 S.Ct. 2590. In emphasizing the difficulty of discerning the reasoning behind an unexplained state-court order — or one "whose text or accompanying opinion does not disclose the reason for the judgment,” id. at 802, 111 S.Ct. 2590 — the Court said: "Indeed, sometimes the members of the court issuing an unexplained order will not themselves have agreed upon its rationale, so that the basis of the decision is not merely undiscoverable but nonexistent.” Id. at 803, 111 S.Ct. 2590. Although Ylst predates the passage of AEDPA, the Richter Court cited it favorably, 562 U.S. at 99-100, 131 S.Ct. 770, thus indicating the continued validity of its presumption.
. That reading of Richter has ample support in other circuits. See, e.g., Cannedy v. Adams, 706 F.3d 1148, 1158 (9th Cir. 2013) (“[I]t does not follow from Richter that, when there is a reasoned decision by a lower state court, a federal habeas court may no longer ‘look through' a higher state court’s summary denial to the reasoning of the lower state court."); Johnson v. Secretary, DOC, 643 F.3d 907, 930 n.9 (11th Cir. 2011) (“The Court’s instruction from Harrington does not apply here because the Florida Supreme Court did provide an explanation of its decision....”); Sussman v. Jenkins, 642 F.3d 532, 534 (7th Cir. 2011) (distinguishing Richter because that case "addresses the situation in which a state-court decision 'is unaccompanied by an explanation,’ ” whereas in the instant case "the state appellate court issued an opinion”).
To read Richter to apply to a state court's ultimate decisions, irrespective of stated reasoning, also requires that we assume the Richter Court intended to overrule some precedents sub silentio. In particular, Ylst established a presumption that “[w]here there has been one reasoned state judgment rejecting a federal claim, later unexplained orders upholding that judgment or rejecting the same claim rest upon the same ground.” 501 U.S. at 803, 111 S.Ct. 2590. Judge Hardiman endeavors to narrow the Ylst presumption to only apply when we are uncertain as to whether the state court decided a claim “on the merits.” (J. Hardiman Dissent Op. at 368 n.7.) So, in his view, we look through to the last reasoned state court decision to determine whether the case was decided on the merits, and then, having answered that question, take no account of the reasoning in that state court decision. But, in applying the Ylst presumption, the Supreme Court has analyzed and discussed the expressed reasoning of lower state courts. See Johnson v. Williams, -U.S. -, 133 S.Ct. 1088, 1097-99, 185
L.Ed.2d 105 (2013); see also Hittson v. Chatman, -U.S.-, 135 S.Ct. 2126, 2128, 192 L.Ed.2d 887 (2015) (Ginsburg, J., concurring in the denial of certiorari) ("There is no reason not to 'look through’ ... to determine the particular reasons why the state court rejected the claim on the merits.”). The proper application of the Ylst presumption raises all of the same policy problems Judge Hardiman has noted' — just one step lower in the state review process. If we “look through” an unreasoned state court decision, Ylst presumably requires that we then review the reasoning given in the lower state court. If not, then why bother "looking through” at all? If we truly read Richter in the way Judge Hardiman proposes — and took his reasoning to its logical conclusion — it would require that we void the Ylst presumption, because we need not "look through” unreasoned judgments when we actually review only decisions and not their reasoning. But, in the words of the Supreme Court, "a presumption which gives [unreasoned orders] no effect — which simply 'looks through' them to the last reasoned decision — most nearly reflects the role they are ordinarily intended to play.” Ylst, 501 U.S. at 804, 111 S.Ct. 2590 (emphasis in original). It is hard to accept that the Richter Court intended to implicitly overrule Ylst, particularly because the Court cited Ylst favorably. See Richter, 562 U.S. at 99-100, 131 S.Ct. 770. The Court also applied the Ylst presumption just this past term, thus confirming its continued viability. See Kernan v. Hinojosa,-U.S. -, 136 S.Ct. 1603, 1605-06, 194 L.Ed.2d 701 (2016) (per curiam).
And, under Judge Hardiman's approach, Ylst is not the only precedent that would have to fall. Compare J. Hardiman Dissent Op. at 375 ("Where the state court denies relief but addresses only certain prongs of a test,or components of a claim, the reviewing federal court should likewise consider what reasons *356regarding an unaddressed prong or component could have supported the decision.”), with Wiggins v. Smith, 539 U.S. 510, 534, 123 S.Ct. 2527, 156 L.Ed.2d 471 (2003) (reviewing a Strickland claim, and concluding that its "review is not circumscribed by a state court conclusion with respect to prejudice, as neither of the state courts below reached this prong of the Strickland analysis”), and Palmer v. Hendricks, 592 F.3d 386, 400 (3d Cir. 2010) (citing Wiggins for the proposition "that because the state courts did not decide the prejudice issue on the merits, AEDPA's deferential standards do not apply to our resolution of the prejudice question”). In Wiggins, the Supreme Court did not defer to the state court’s order in assessing the second prong of the petitioner’s Strickland claim because "neither of the state courts below reached this prong of the Strickland analysis.” 539 U.S. at 534, 123 S.Ct. 2527. The Court thus acted contrary to Judge Hardiman's proposed holding here — it engaged in de novo review of the second prong even though "the state court denie[d] relief but addresse[d] only certain prongs of a test or components of a claim....” (J. Hardiman Dissent Op. at 375.) Judge Hardiman forthrightly acknowledges that his proposed holding is in tension with Wiggins, but then suggests that Richter (as the later of the two cases) undermines Wiggins. I do not believe that Richter intended that result, especially because the two cases can be reconciled.
. Again, if we determine that a state court’s reasoning is contrary to clearly established federal law, we then engage in de novo review of the claim in question. See supra note 6; Panetti, 551 U.S. at 948-54, 127 S.Ct. 2842. In his dissent, Judge Hardiman posits a hypothetical in which our decision to grant habeas relief could turn on the state court's method of drafting its decision. If the state court issues a summary order, we would apply Richter and deny habeas relief by application of AEDPA deference. If, however, it issues a reasoned decision, and that reasoning is contrary to clearly-established federal law, we would grant habeas relief — to the very same claimant — after de novo review of the underlying claim. My colleague thinks that outcome absurd, but, whether we like it or not, that is what the Supreme Court directs us to do. Under AEDPA, we must defer (1) to the reasoning actually elaborated in a state court decision, and (2) to any basis that can reasonably support a state court’s decision, but only if its own reasoning cannot be fairly discerned. The latter is the import of Richter. If the Supreme Court wanted us to afford AED-PA deference to all state court decisions regardless of the extent of their reasoning, that would be a rule of considerable consequence for habeas petitioners. Presumably the Supreme Court would have said (or at least suggested) as much in Richter, Premo, Lafler, Wetzel, or any of the other numerous habeas appeals it has considered in recent years and that Judge Hardiman has collected in his dissent. If anything, though, the Court has said the contrary. See Panetti, 551 U.S. at 954, 127 S.Ct. 2842 ("§ 2254 does not preclude relief if either the reasoning or the result of the state-court decision contradicts” clearly-established Supreme Court precedent (internal quotation marks and alteration omitted, emphasis added).).
A petitioner does not get any windfall under the approach I have outlined based on Supreme Court precedent. If his claim does not have merit, it will fail even under de novo review. Under Judge Hardiman’s approach, by' contrast, state prosecution teams do get a windfall. They would prevail unless every conceivable route to victory is "contrary to ... clearly established Federal law, as determined by the Supreme Court of the United States.” 28 U.S.C. § 2254(d)(1). In other words, the prosecution wins even if it never argued a sensible position and the state court gave only a completely erroneous basis for its decision. I do not believe we can or should read Richter as going that far.