dissenting.
Both the majority opinion and Judge Malo-ney in dissent agree that applicant in this cause has established that he has suffered a deprivation of due process. Thus the entire Court concludes that the State knowingly used false testimony in this cause. As I *378understand it, the majority further concludes that applicant has established the materiality of the false testimony; that is to say, the record admits of a “reasonable likelihood that the false' testimony could have affected the judgment of the jury.” United States v. Bagley, 473 U.S. 667, 678-79, 105 S.Ct. 3375, 3381-82, 87 L.Ed.2d 481, 492 (1985), quoting United States v. Agurs, 427 U.S. 97, 103, 96 S.Ct. 2392, 2397, 49 L.Ed.2d 342, 349-50 (1976). Nevertheless, in the majority’s scheme of things, applicant still cannot prevail. The reason is that applicant has failed to meet the standard of harmfulness announced by the United States Supreme Court in Brecht v. Abrahamson, 507 U.S. 619, 113 S.Ct. 1710, 123 L.Ed.2d 353 (1993). But we have never adopted such a standard for harm in post-conviction habeas corpus in Texas,1 and, indeed, the majority only implicitly adopts it today. At least in the context of the particular error presented here, I believe it is a mistake to apply the Brecht test. But even if I believed that we should adopt that harm test in Texas, I could not agree with the majority today that applicant has failed to satisfy it.
I.
I should preface my remarks with a statement about post-conviction habeas cognizability. I presume the majority considers applicant’s claim cognizable simply by virtue of the fact that it is brought under the auspices of due process. And it is indeed the law in Texas that practically any federal constitutional claim is cognizable in an Article 11.07 application for post-conviction writ of habeas corpus. Ex parte Banks, 769 S.W.2d 539, 540 (Tex.Cr.App.1989). I would not have it that way. Over the last half dozen years I have advocated reducing habeas eognizability of federal constitutional claims to those so “exceptional,”
“so fundamental to the fair operation of the system as to be 1) immune from procedural default, 2) not subject to a harm analysis, and 3) fully retroactive in application. [Ex parte Goodman, 816 S.W.2d 383, 388 (Tex.Cr.App.1991) (Clinton, J., concurring)] I would also adopt a doctrine of excuses, entertaining any federal constitutional claim recognized as of the time of trial but for which a record could not have been made, despite due diligence of the accused, in time to preserve the error for direct appeal. [Ex parte Dutchover, 779 S.W.2d 76, 79 (Tex.Cr.App.1989) ] (Clinton, J., concurring; Ex parte Goodman, supra, at 387-88.”
Ex parte Sadberry, 864 S.W.2d 541, 545 (Tex. Cr.App.1993) (Clinton, J., dissenting).
The Supreme Court has not clearly indicated whether the knowing use of perjured testimony constitutes a violation of due process absent a showing of “materiality”/harm. See n. 2, post. Thus, I am not sure that such a claim would fit my definition of “exceptional” for purposes of my scheme of cognizability of federal constitutional claims. I am confident, however, that applicant’s claim would meet my doctrine of excuses, for it appears he was in no position to raise the due process claim here in time to preserve it for appeal. Thus I conclude that it is cognizable not only under the Court’s criteria for eognizability in Ex parte Banks, supra, but under my own view of habeas eognizability as well.
The majority apparently agrees the claim is cognizable, but turns applicant away for a different reason. The majority concedes that applicant can show “materiality”/harm, at least under the standard for harm ordinarily applicable to federal constitutional errors on direct appeal, viz: Chapman v. California, 386 U.S. 18, 87 S.Ct. 824, 17 L.Ed.2d 705 (1967). However, because he cannot meet the test for harm in the context of federal habeas review of state convictions, as announced in Brecht v. Abrahamson, supra, applicant cannot prevail. It is to this proposition that I turn first.
II.
In order to establish a violation of due process on direct appeal, an appellant must show that the prosecution knowingly used *379false testimony at his trial. If that is so, and if the testimony was “material,” which is to say, “if there is any reasonable likelihood that the false testimony could have affected the judgment of the jury[,]” United States v. Agurs, supra, then an appellant would be entitled to a new trial. From the caselaw in the Supreme Court it is not entirely certain whether this “materiality” requirement is part of the due process claim itself, or is just another way of saying that knowing use of perjured testimony is subject to the constitutional harmless error analysis of Chapman v. California, supra. See United States v. Bagley, 473 U.S. at 679-80, 105 S.Ct. at 3382, 87 L.Ed.2d at 492.2 In any event, the Supreme Court has commented that the “reasonable likelihood that the false testimony could have affected the judgment of the jury” .test is functionally equivalent to the Chapman test for harmless error. Id., n. 9. I take this to mean that if from its review of the record the appellate court cannot say that the false testimony did not contribute to the conviction or punishment, it is obliged to reverse the judgment.
That is on direct appeal. What we have before us today is the same kind of claim raised in a post-conviction application for writ of habeas corpus. The same rules do not necessarily apply.' For example, we have said that in the context of state post-conviction habeas corpus, for any federal constitutional claim that is subject to a Chapman harm analysis, the applicant has the burden to establish harm, just as he has the burden to establish every other fact which if true would entitle him to' relief. See Ex parte Dutchover, supra at 77-78; Ex parte Barber, supra at 891-92.3 Thus we place the burden of proof as to harm on the applicant. The federal courts reviewing state convictions in federal applications for habeas corpus place the burden as to harm differently. In O’Neal v. McAninch, 513 U.S. 432, 115 S.Ct. 992, 130 L.Ed.2d 947 (1995), the Supreme Court declared that when a federal habeas court has a grave doubt whether constitutional error in a state prosecution was harmful, it should err on the side of the applicant and grant habeas relief. Although the Supreme Court expressly eschewed the terminology of burden-of-proof, the rule of default that it announced in McAninch nevertheless operates in essentially the same way that a burden of proof does: it tells the court which party wins should it find itself in doubt as to the outcome. Thus, in McAninch the Supreme Court effectively placed the burden on the State to prove harmlessness in federal habeas corpus.
But while the burden of proof on harm remains the same for federal habeas review of state convictions as it is on direct appeal, the standard for harm does not. As Judge Meyers noted in Ex parte Barber, supra at 893, the standard for harm in federal habeas review of federal constitutional errors in state prosecutions has been relaxed, in Brecht v. Abrahamson. After Brecht it is no longer appropriate for federal courts to grant habeas relief from state convictions simply because they cannot say that federal constitutional error did not contribute to the conviction or punishment, as under the Chapman test for harmless error applicable on direct appeal. Instead, a state conviction need only be overturned on federal habeas corpus if the federal court finds (or, after McAninch, has a grave doubt whether) the federal constitutional error “had substantial and injurious effect or influence in determining the jury’s verdict.” Brecht, supra, 507 U.S. at 637, 113 S.Ct. at 1722, 123 L.Ed.2d at 373.
The majority now incorporates the Brecht standard for harm in federal habeas corpus review into our own state post-conviction ha-beas corpus jurisprudence. The majority does not justify this incorporation, or even acknowledge in a straightforward way that that is what it has done. In my view the *380majority errs in three significant respects. First, the majority errs to adopt the Brecht standard at all, at least for pleasuring harmfulness of the kind of constitutional error involved in this cause. Second, it appears to me that the majority has, in any event, misconstrued the Brecht standard. Finally, even if the majority does properly construe Brecht, it misapplies its own construction to deny applicant relief in this cause. I next develop these arguments seriatim.
III.
A.
There are no compelling reasons for this Court to mimic the Brecht standard for harm, at least not for the kind of error we all agree occurred in this cause. The Supreme Court’s justifications for adopting a less onerous standard of harmlessness in the federal habeas context are more than passing familiar by now. As the Court observed in Brecht:
“The reason most frequently advanced in our cases for distinguishing between direct and collateral review [of state prosecutions] is the State’s interest in the finality of convictions that have survived direct review within the state court system. * * * We have also spoken of comity and federalism.”
507 U.S. at 634, 113 S.Ct. at 1720, 123 L.Ed.2d at 371 (citations omitted). But of course, principles of “comity and federalism” cannot by themselves possibly justify this Court preferring the Brecht test for harmlessness over that of Chapman for purposes of state habeas corpus. This Court has none to defer to but itself. Therefore, if any part of the Supreme Court’s rationale could serve to justify our adoption of Brecht, it would have to be the State’s interest in the finality of a conviction, once it has been affirmed on direct appeal. And indeed, that consideration is weighty enough to convince me that as a general rale this Court should not even recognize as cognizable federal constitutional claims that are not “exceptional.” See Ex parte Dutchover, supra at 79 (Clinton, J., concurring); Ex parte Sadberry, supra at 545 (Clinton, J., dissenting).
But I do not believe the instant claim is cognizable because it is “exceptional” in the sense that I spoke of in Dutchover and Sad-berry. Instead, I believe it is cognizable under the doctrine of excuses; because it is a federal constitutional claim that applicant could not have raised, even by due diligence, on direct appeal. It seems to me that principles of comity and federalism dictate that if a federal constitutional claim cannot, for reasons beyond the applicant’s control, be raised any earlier than in a collateral attack in the state forum, the state should address it at that time, not leave it for resolution by a federal habeas court. For the same reasons that I believe such a claim ought to be cognizable on state post-conviction habeas corpus, I also believe that principles of finality should not lead us to apply a less onerous test for harmlessness in the state habeas forum. An applicant who, in spite of all, has been unable to raise a federal constitutional claim on direct appeal, has had no opportunity to reap the benefit of the more favorable standard for constitutional harm in Chapman.
If we are willing, as I am, to reach such a claim of federal constitutional error for the first time on collateral attack, we ought also to be willing to afford the applicant the full panoply of constitutional protections. This ought to include the right to a new trial as long as he can show (because the burden is on him in state post-conviction habeas corpus) that the federal constitutional error contributed to his conviction. The State’s otherwise compelling interest in finality of convictions should give way in this context to the fact that the applicant has.been, through no fault of his own, unable to litigate the claim at any earlier opportunity. As far as I can see, there is no justification for requiring him to satisfy the tougher Brecht standard to show that the error “had substantial and injurious effect or influence in determining the jury’s verdict.” 507 U.S. at 637, 113 S.Ct. at 1722, 123 L.Ed.2d at 373. The majority errs to require it anyway.
B.
On direct appeal, the burden is on the State to prove constitutional error is harmless. Because in state post-conviction habeas *381the burden is on the applicant to show harm, we must articulate the Chapman standard for harm in a slightly different way. On direct appeal the State satisfies its burden if it can show beyond a reasonable doubt that the constitutional error did not contribute to conviction. This means, essentially, that if there is any reasonable possibility the error did contribute to the conviction, the judgment must be reversed. See United States v. Bagley, supra, 473 U.S. at 679, n. 9, 105 S.Ct. at 3382, n. 9, 87 L.Ed.2d at 492, n. 9. Thus it makes sense to say, applying the Chapman standard for harm in state post-conviction habeas corpus, where the burden is on the applicant, that he is entitled to a new trial if the record admits of a reasonable possibility that the constitutional error contributed to his conviction.4 Even the majority concedes that applicant can satisfy this standard of harm in this cause. Op. at 375-376.
Having uncritically accepted Brecht into our habeas jurisprudence, however, the majority denies relief. But when it comes to apply the Brecht test in the state post-conviction habeas context, the majority does a strange thing. The majority does not transpose the Brecht standard into the key of state post-conviction habeas corpus in the way that one would expect. It seems to me that what the majority should be asking itself, consistent with the standard in Brecht, but accounting for the fact that applicant has the burden in state post-conviction habeas, is this: Has applicant shown a reasonable possibility that the constitutional error in this cause “had substantial and injurious effect or influence in determining the jury’s verdict”? 507 U.S. at 637, 113 S.Ct. at 1722, 123 L.Ed.2d at 373. One would expect this transposition of the Brecht standard because it raises the standard by which an applicant must show harm for state post-conviction habeas purposes in exactly the same proportion that Brecht reduces the standard by which the State must show an absence of harm in a federal habeas application.
But the majority asks itself a different question altogether. The majority asks whether applicant can show by a preponderance — that is to say, whether it is more likely than not — that the constitutional error contributed to the verdict. It seems to me that framing the question this way distorts the Brecht test in two significant ways. First, and most obviously, the majority does not really utilize the Brecht standard (“substantial and injurious effect or influence on the jury’s verdict”) at all, but reverts to the supposedly-abandoned standard of Chapman (“contribute to the conviction”). Because I believe we should be applying the Chapman standard in any event, my objection to this first distortion is, of course, purely academic. It is the second distortion that makes all the difference to the majority in this cause.
The second way in which the majority distorts the Brecht standard is to require the applicant to prove harm to a level of confidence that Brecht itself does not call for, viz: a preponderance. Rather than reverse upon a showing that there is a reasonable possibility of harm (by whatever standard, “contribution” or “substantial and injurious effect or influence”), the majority requires the applicant to show that the constitutional error more likely than not harmed him. The majority adopts this preponderance level of confidence because it perceives that the difference between the Chapman standard and the Brecht standard is “the difference between a possibility and a probability.” Op. at 376. This notion the majority borrows from an opinion of the First Circuit Court of Appeals. Gilday v. Callahan, 59 F.3d 257, at 269 (C.A.1 1995). For its part, the First Circuit derives the notion from a passage in the Supreme Court’s opinion in O’Neal v. McAninch, — U.S. at -, 115 S.Ct. at 994, 130 L.Ed.2d at 951-52. But the passage in O’Neal does not speak to the question by what level of confidence harm must be proved or disproved in order to justify ha-beas relief. It addresses only the issue of which party bears the loss if the record is insufficient to resolve the question of harm, *382by whatever level of confidence (the Supreme Court simply did not say). Thus, both the First Circuit and the majority today are mistaken. O’Neal does not support the proposition that the difference between the standards in Chapman and Brecht is “the difference between a possibility and a probability.” The only real difference between the two is that Chapman merely asks whether error “contributed,” while under Brecht the federal habeas court must be able to say the error “substantially and injuriously affected or influenced the jury’s verdict” before it can reverse a state conviction.
The issue in this cause — a post-conviction application for writ of habeas corpus in state court, not federal — ought to be whether the record admits of a reasonable possibility the State’s knowing use of false testimony contributed to applicant’s conviction. We are all in agreement that it does. Therefore applicant is entitled to the relief he seeks.
C.
Finally, even if I could agree that the question is whether applicant has shown more likely than not that the error contributed to his conviction (or, for that matter, that it more likely than not “substantially and injuriously affected or influenced the jury’s verdict”), I would grant him relief. But for the false testimony in this cause, applicant’s confession would in all likelihood have been suppressed. Instead it was admitted. I do not understand how the majority could conclude that applicant’s confession did not more likely than not “substantially and injuriously affect and influence the jury’s verdict” in this cause, much less that it did not more likely than not at least “contribute.”
The majority ultimately finds it “more probable than not that the outcome of applicant’s trial would have been the same absent [his] confession.” Op. at 376. This is an erroneous application of both the Chapman and the Brecht standard for harm, at least as I understand them. For neither Chapman nor Brecht adduced a test for measuring the harm vel non of constitutional error that is outcome-determinative. • To say that error “contributed to,” or even that it “substantially and injuriously affected or influenced” a jury’s verdict is not at all the same as saying it was a sine qua non of the conviction. Whether or not the outcome of trial would have been the same is not the test under any standard for measuring harmfulness of federal constitutional error that the Supreme Court has announced. When the Supreme Court does want to announce a standard for harm/materiality that is outcome-determinative, it has no trouble making that intention clear. Accordingly, in Strickland v. Washington, 466 U.S. 668, at 694, 104 S.Ct. 2052, at 2068, 80 L.Ed.2d 674, at 698 (1984), in defining the prejudice prong of the test for Sixth Amendment ineffective assistance of counsel, and in United States v. Bagley, supra, 473 U.S. at 682, 105 S.Ct. at 3383, 87 L.Ed.2d at 494, when defining “materiality” of Brady-type error (other than the knowing use of perjured testimony), the Supreme Court unmistakably constructed a standard that is outcome-determinative, viz: whether there is a “reasonable probability that ... the result of the proceeding would have been different.”5 Neither the Chapman nor the Brecht standard is phrased this way, and surely the difference in language stands for something.
My conclusion is bolstered by an observation the Supreme Court made in Kyles v. Whitley, 514 U.S.-, 115 S.Ct. 1555, 131 L.Ed.2d 490 (1995). There- the Court was considering, inter alia, whether a finding of ordinary Brady error in a federal writ, which includes a finding that there is a “reasonable probability that ... the result of the proceeding would have been different,” calls for a Brecht analysis for harm as a predicate to reversal. The Supreme Court held a Brecht analysis unnecessary because “the harm to *383the defendant [must be] greater” to justify reversal under the standard for harm/materiality in Strickland /Bagley on direct appeal than is required under the Brecht standard for reversal on federal habeas corpus. Kyles, supra, — U.S. at-, 115 S.Ct. at 1567, 131 L.Ed.2d at 507. The Court did not say that the likelihood of harm must be greater, but that the harm itself must be. This suggests to me that the difference is not one of levels of confidence; instead it suggests that the harm recognized in Strickland /Bagley is of a kind that has a bearing on the outcome of the case, whereas the harm under Brecht is qualitatively something less than that. We effectuate this distinction if we construe Chapman and Brecht to require, not a showing that the error was a sine qua non of the jury’s verdict, but only that it facilitated that verdict in some substantial way. In other words, the error “contributed,” “affected,” or “influenced” the verdict, but was not necessary to the verdict.
By this understanding I can only conclude that admission of applicant’s confession at his trial not only “contributed to,” but indeed, “substantially affected or influenced” the jury’s verdict, even to a level of confidence of more likely than not. The Supreme Court has recognized that “an involuntary confession may have a more dramatic effect on the course of a trial than do other trial errors— in particular cases it may be devastating to a defendant[.]” Arizona v. Fulminante, 499 U.S. 279, at 312, 111 S.Ct. 1246, at 1266, 113 L.Ed.2d 302, at 333 (1991). It is indeed hard to imagine other evidence so overwhelming that a confession from the defendant’s own mouth would not nevertheless “substantially affect or influence” the jury’s verdict.
The majority is only able to find harmlessness (even under its distorted Brecht standard) by envisioning a hypothetical record in which the applicant’s confession was not admitted. Without the confession, the majority reasons, the only eyewitness to the offense is not shown to be an accomplice, and the jury would have no reason to doubt his veracity. But this method of harm analysis does not measure the harmfulness of the constitutional error upon the trial that actually occurred. At applicant’s real trial, the eyewitness was an accomplice, whose veracity was intrinsically suspect. Under those circumstances it is impossible to believe anything but that applicant’s own confession substantially influenced the jury — even if it might have reached the same verdict eventually anyway. The prosecutor must have thought at least as much, for he testified he would not have proceeded with the prosecution had the confession been suppressed unless he could have corroborated the accomplice — even though, according to the majority, such corroboration would have been unnecessary to the sufficiency of the evidence. The majority errs to deny applicant relief in this cause even under its own distorted version of the federal habeas corpus standard for determining harm.
IV.
We all agree applicant was deprived of due process of law at his first trial for capital murder. Nevertheless the majority denies him another trial because, for reasons beyond his control, he was only able to raise his due process contention for the first time in post-conviction habeas corpus. I would apply the Chapman standard for harm and hold that applicant has met his burden to show a reasonable possibility that the error contributed to his conviction. If I believed that Brecht announced an appropriate standard for harm in this cause, I would hold applicant satisfied that test as well. Even if I thought the majority was right to apply its bastardized Brecht standard, I would hold applicant has met that too. Because it is apparent that under no circumstances will the majority grant relief in this cause, I dissent.
. Judge Meyers has suggested that we adopt the Brecht standard for harm in state post-conviction habeas corpus. Ex parte Barber, 879 S.W.2d 889, 893 (Tex.Cr.App.1994) (Meyers, J., concurring). So far the Court has not applied itself to the question.
. It is for this reason that I find it difficult to say whether the knowing use of perjured testimony would fit my definition of an "exceptional” federal constitutional claim. See text, ante.
. Thus, I agree with the majority that in state post-conviction habeas review the applicant has the burden to show harm, by whatever standard. Op. at 371-372. The question remains what the standard for harm in state habeas should be. The majority implicitly and uncritically accepts the Brecht standard.
. If after assaying the record the habeas court cannot determine whether there is a reasonable possibility the error contributed to the conviction, the applicant must lose, since he bears the burden of proof. Contrast O'Neal v. McAninch, supra (in federal habeas corpus review of state convictions, where habeas court has "grave doubt” whether error harmless, State must lose).
. It is true that the Strickland /Bagley tests are phrased in terms of a "reasonable probability ...” It could not be clearer from context, however, that by "reasonable probability” the Supreme Court did not mean it must be "more likely than not” that the outcome of the proceeding would be different. In Strickland the Court expressly eschewed a "preponderance” level of confidence. 466 U.S. at 694, 104 S.Ct. at 2068, 80 L.Ed.2d at 698. It is apparent that the Court meant that reversal should follow if there is a reasonable chance the result would be different, not that it is more likely than not a different outcome would result.