dissenting:
The majority affirms the denial of habeas corpus relief in both of these cases on the basis that in each the state trial court’s concededly erroneous burden-shifting alibi instruction was harmless beyond a reasonable doubt because of the overwhelming evidence of guilt coupled with the weakness in each of the alibi evidence.
With respect, I disagree with that result and in fundamental respects with the majority’s mode of analysis. I believe that a special jury instruction on alibi that effectively shifts to the defendant the burden of persuasion on the disputed issue of presence is a constitutional error that must be deemed never harmless. On that view it is of course inappropriate for an appellate court, once the constitutional error is found, to conduct any inquiry into possible harmlessness based upon the weight of the evidence of guilt.
This conclusion seems to me compelled by authoritative Supreme Court decisions and by an independent consideration of the nature of this particular error, the proper function of harmless error review in general, and the relative roles of juries and appellate courts in assessing the evidence of guilt in criminal cases.
On this basis, I would reverse in both of these cases and remand with directions to issue the writs unless new trials were provided. In both cases it is plain that the alibi instructions did effectively shift to the respective habeas petitioners the burden of persuasion on the disputed issues of presence.
I
In Chapman v. California, 386 U.S. 18, 87 S.Ct. 824, 17 L.Ed.2d 705 (1967), the Supreme Court rejected the view theretofore held that “all federal constitutional errors, regardless of the facts and circumstances, must always be deemed harmful.” Id. at 21, 87 S.Ct. at 826. In opening the possibility that even some constitutional error might be “declared” by appellate courts to be harmless “beyond a reasonable doubt,” the Court emphasized, however, that “there are some constitutional rights so basic to a fair trial that their infraction can never be treated as harmless error.” Id. at 23, 87 S.Ct. at 827. Specific examples of per se harmful constitutional error were given: coerced confession, citing Payne v. Arkansas, 356 U.S. 560, 78 S.Ct. 844, 2 L.Ed.2d 975 (1958); denial of counsel, citing Gideon v. Wainwright, 372 U.S. 335, 83 S.Ct. 792, 9 L.Ed.2d 799 (1963); and trials by a biased judge, citing Turney v. Ohio, 273 U.S. 510, 47 S.Ct. 437, 71 L.Ed. 749 (1927). And, focussing generally on the jury trial process itself, the Court observed, without offering specific illustrations, that “[a]n error ... which possibly influenced the jury adversely to a litigant cannot ... be conceived of as harmless.” 386 U.S. at 23-24, 87 S.Ct. at 827-828.
In re Winship, 397 U.S. 358, 364, 90 S.Ct. 1068, 1072, 25 L.Ed.2d 368 (1969) soon thereafter established — if it had not already been established — that the “Due Process Clause protects the accused against conviction except upon proof beyond a reasonable doubt of every fact necessary to constitute the crime .,. charged.” And later Supreme Court decisions have simply made an obvious application of this principle to find constitutional error in jury instructions that, in a variety of ways, effectively cast upon the accused rather than the state the burden of persuasion in respect of an essential element of the crime *1035charged. See, Mullaney v. Wilbur, 421 U.S. 684, 95 S.Ct. 1881, 44 L.Ed.2d 508 (1975) (direct placement of burden upon accused to prove provocation in order to reduce murder to manslaughter); Connecticut v. Johnson, 460 U.S. 73, 103 S.Ct. 969, 74 L.Ed.2d 823 (1983) (presumption that may have been understood to compel finding of intent from proof of other facts); Sandstrom v. Montana, 442 U.S. 510, 99 S.Ct. 2450, 61 L.Ed.2d 39 (1979) (same).
At least since Mullaney, therefore, it has been established that jury instructions that effectively shift to an accused the burden of persuasion (by any quantum of proof) on an element of the crime violate due process. The only question that remained was whether, under the Chapman rationale, such a constitutional error might nevertheless be found harmless or must, on the contrary, always be deemed harmful.
Concededly, no post-Chapman/Winship decision of the Supreme Court has squarely and authoritatively determined that question.1 But I am persuaded that the critical post-Chapman/Winship Supreme Court decisions that have addressed the general problem of the effect of arguably erroneous burden-shifting instructions strongly imply, or indeed simply assume, that any found effectively to have misplaced the burden on an accused are to be treated as per se harmful.
These decisions have actually addressed dispositively only the issue of whether the challenged jury instruction did or did not involve constitutional error. Specifically, they have decided only the predicate questions of whether the instruction did or did not relate to an element of the offense as to which, per Winship, the burden of proof could not constitutionally be placed upon the accused, see Mullaney, 421 U.S. at 690-704, 95 S.Ct. at 1885-1892; Patterson v. New York, 432 U.S. 197, 201-216, 97 S.Ct. 2319, 2322-2330, 53 L.Ed.2d 281 (1977) (“sanity”); or whether by incorporating a presumption the instruction did or did not effectively shift to the accused the burden of persuasion on a conceded element of the crime, see Johnson, 460 U.S. at 83-88, 103 S.Ct. at 976-978 (intent); Sandstrom, 442 U.S. at 514-27, 99 S.Ct. at 2454-60 (same).2
*1036The various opinions in these post-Win-skip cases imply to me that at least a majority of the members of the present Court now simply assume that if a challenged instruction either (a) relates to an element of the offense (as opposed to a matter of “affirmative defense”) and (b) effectively shifts the burden on that element to the accused, or (c) otherwise fails to place the burden as to all essential elements of the crime charged upon the state, it is to be deemed per se harmful. See, e.g., Connecticut v. Johnson, 460 U.S. at 88, 103 S.Ct. at 978 (Blackmun, J., for 4-member plurality) (instruction shifting burden by presumption said to have deprived accused of “ ‘constitutional rights so basic to a fair trial that their infraction can never be treated as harmless error’ ”); cf. id. at 98 n. 6, 103 S.Ct. at 983 n. 6 (Powell, J., for 4 members, dissenting) (challenged instruction said not to have shifted burden, hence to be “not comparable to an instruction that can never be harmless — for example, an instruction that fails to inform the jury that it must find guilt beyond a reasonable doubt”); see also Jackson v. Virginia, 443 U.S. 307, 320 n. 14, 99 S.Ct. 2781, 2789 n. 14, 61 L.Ed.2d 560 (1979) (“failure to instruct the jury on the necessity of proof beyond a reasonable doubt can never be harmless error”); Cool v. United States, 409 U.S. 100, 104, 93 S.Ct. 354, 357, 34 L.Ed.2d 335 (1972) (instruction whose effect was “to require the defendant to establish his innocence beyond a reasonable doubt” said to be “plainly inconsistent with the constitutionally rooted presumption of innocence”).3
II
Considering the issue an open one for which only support and not direct authority can be found in the critical Supreme Court decisions, I come out the same way on an independent consideration of the relevant factors.
Looking first to the subject of the challenged instructions, defendant’s alibi evidence, it is settled, and is not here challenged, that because, as a matter of federal law, this “defense” relates to an element of the crime charged, the burden of proof upon it may not constitutionally be shifted to the accused. Adkins v. Bordenkircher, *1037674 F.2d 279, 282 (4th Cir.1982); cf. Patterson v. New York, 432 U.S. 197, 97 S.Ct. 2319, 53 L.Ed.2d 281.
Looking next to the question of whether the challenged instruction effectively shifted the burden to the accused in these cases — again there is no doubt that both did. The patent, unquestionable effect of these instructions was to convey an impression to the jury that the defendant had acquired the burden of disproving his presence at the crime scene (hence his criminal agency in general) by undertaking to prove his presence specifically elsewhere.
These cases are therefore not like those cases in which a proper contextual analysis reveals that the asserted constitutional vice of a single instruction has been dissipated by other specifically curative or otherwise corrective instructions.4 Cf. Cupp v. Naughten, 414 U.S. 141, 146-47, 94 S.Ct. 396, 400-01, 38 L.Ed.2d 368 (1973); see also Tweety v. Mitchell, 682 F.2d at 465. Neither are these cases like Johnson, where the question of effective burden shifting was at least debatable, and, on my reading, see note 2 supra, actually the dispositive issue for the Court.
Finally it is plain that the critical element — presence of the accused — was in each of these cases a disputed issue for the jury5 that was then necessarily decided adversely to the accused by the guilty verdict. These cases are therefore not of the type suggested in the plurality opinion in Connecticut v. Johnson, 460 U.S. at 87, 103 S.Ct. at 977, in which a facially erroneous burden-shifting instruction could be found, however, not to have been violative of due process because the element to which it related was — for any number of reasons — not actually in issue, e.g., Krzeminski v. Perini, 614 F.2d 121, 125 (6th Cir.1980) (intent conceded by accused), or was manifestly not ultimately decided adversely to the accused, e.g., Hearn v. James, 677 F.2d 841, 843 (11th Cir.1982) (erroneous murder instruction on intent; conviction only of manslaughter).
On this threshold analysis of the instructions’ effects we therefore have in each of these cases a jury instruction which effectively shifted to the accused the burden of persuasion with respect to an element of the crime charged whose existence was a disputed issue before the jury that was then necessarily denied adversely to the accused. This, it seems to me, establishes *1038by a properly limited contextual inquiry that the jury instructions involved an infraction of the basic constitutional right generally recognized in Winship and specifically vindicated, for example, in Mullaney.
This in turn then squarely raises the issue of whether this constitutional error can nevertheless be found harmless under the Chapman test, or whether it must be considered per se harmful. Because of the basic nature of the constitutional right involved, general limitations on proper harmless error review, and the primacy of the jury in factfinding, I believe that it must be considered harmful per se.
Ill
Whether a particular constitutional trial error may ever be found harmless depends not only upon its degree of “grievousness,” see Johnson, 460 U.S. at 88, 103 S.Ct. at 978 (Stevens, J., concurring in the judgment), but upon its susceptibility to principled harmless error review — that is, review designed and rationally capable of determining the likelihood of actual prejudice traceable to the specific error. The denial of counsel, for example, is constitutional error that may never be deemed harmless, not only because of the basic nature of the right involved, see Gideon v. Wainwright, 372 U.S. 335, 83 S.Ct. 792, 9 L.Ed.2d 799 (1963), but because of the impossibility of attempting fairly to assess the probability or possibility that that error had a specific prejudicial effect in a given trial. See Holloway v. Arkansas, 435 U.S. 475, 491, 98 S.Ct. 1173, 1182, 55 L.Ed.2d 426 (1978) (harmless error analysis where counsel had conflict of interest “would require ... unguided speculation”).
A constitutionally erroneous burden-shifting jury instruction must be considered at least equally as grievous as those errors specifically recognized as per se harmful, see Mullaney, 421 U.S. at 699-701, 95 S.Ct. at 1889-1890, and equally as unsusceptible to principled harmless error review. The special vice of such an instruction is that it fundamentally taints the whole fact-finding process. In this critical respect it differs from trial errors such as the admission or exclusion of evidence, see, e.g., United States v. Hastings, 461 U.S. 499, 103 S.Ct. 1974, 76 L.Ed.2d 96 (1983), whose “scope is readily identifiable,” Holloway v. Arkansas, 435 U.S. at 490, 98 S.Ct. at 1181, and whose practical effect upon a particular trial process can therefore be assessed with some degree of confidence. Because we must assume that juries understand and follow plainly given instructions, see Johnson, 460 U.S. at 85 n. 14, 103 S.Ct. at 977, the effect of a burden-shifting instructional error must be taken to have wrenched the fundamental way in which the jury was required by the constitution to assess evidence of guilt. As such, this error plainly falls in the same category as other errors resulting in unconstitutionally tainted fact-finders or basic fact-finding processes that have been authoritatively held on that account to be per se harmful. See Chapman, 386 U.S. at 23 & n. 8, 87 S.Ct. at 828 (trial without assistance of counsel; trial by biased judge); id. at 43-44, 87 S.Ct. at 837-838 (Stewart J., concurring) (trial in community exposed to prejudicial pre-trial publicity; trial by discriminatorily selected jury).
In respect of each, of those “never-harmless” errors recognized in Chapman, it would be theoretically possible to conclude that because of the overwhelming evidence of guilt even an untainted fact-finder or a specifically untainted trial process would also necessarily have found guilt. The primary reason that this possibility has been thought one not tolerable in those other situations applies with equal or greater force to a constitutionally erroneous burden-shifting instruction: the right violated is one that exists apart from the right not to be convicted except upon a legally sufficient amount of evidence; it is therefore irremedially violated whatever the actual evidence of guilt in the particular case.6 *1039See Jackson v. Virginia, 443 U.S. at 320 n. 14, 99 S.Ct. at 2790 n. 14 (“[A] defendant whose guilt was actually proved by overwhelming evidence would be denied due process if the jury was instructed that he could be found guilty on a mere preponderance”); United Brotherhood of Carpenters & Joiners v. United States, 330 U.S. 395, 410, 67 S.Ct. 775, 783, 91 L.Ed. 973 (1947) (“[W]here the evidence ... is well-nigh conclusive and the court fails to give a reasonable doubt instruction ... [i]t could not be said that the failure was harmless error”).
Finally, finding a trial error harmless (whether or not of constitutional dimensions) solely on the basis of the overwhelming weight of the evidence — the only basis upon which harmlessness of a burden-shifting instruction could be rested — has been authoritatively condemned as an impermissible function of harmless error analysis. In Kotteakos v. United States, 328 U.S. 750, 764, 66 S.Ct. 1239, 1247, 90 L.Ed. 1557 (1946), the Court pointed out why:
The question is not [whether the jury was] right in their judgment, regardless of the error or its effect upon the verdict. It is rather what effect the error had or reasonably may be taken to have had upon the jury’s decision. The crucial thing is the impact of the thing done wrong on the minds of other men, not one’s own, in the total setting.
See also R. Traynor, The Riddle of Harmless Error, 13.7
The fact that only by this inappropriate appellate process could an unconstitutional burden-shifting instruction theoretically be found harmless8 is but further reason why it is an error that must simply be deemed *1040never harmless. Cf. United Brotherhood of Carpenters & Joiners v. United States, 330 U.S. 395, 410, 67 S.Ct. 775, 783, 91 L.Ed. 973 (1947); Bollenbach v. United States, 326 U.S. 607, 613, 66 S.Ct. 402, 405, 90 L.Ed. 350 (1946).
IV
In sum, I would hold that in assessing a claim that a jury instruction unconstitutionally shifted to an accused the burden of proof in a criminal case, the only question is whether it had that effect. If found to have that effect, I would treat the error as per se harmful — as not susceptible to harmless error analysis.
I would further hold that in determining whether a challenged instruction were unconstitutional, inquiry is properly limited to the questions: (a) whether the instruction related to an element of the crime charged within contemplation of Mullaney and Patterson; (b) whether, viewed in total context, it effectively shifted the burden to the accused; and (c) whether the element to which it related was a disputed issue of fact that was necessarily determined adversely to the accused by the verdict of guilty. If all three of these predicates were established, I would hold the particular instruction unconstitutional, and the convict entitled to have the challenged conviction set aside without further inquiry.9
In each of the cases before us I would, in accordance with these principles, find the challenged instructions unconstitutional on the basis that all three predicates were indisputably established, and reverse with directions to issue the writs conditioned upon the state’s right to re-try the petitioners within reasonable times.10
. In Sandstrom v. Montana, 442 U.S. at 526-27, 99 S.Ct. at 2460-61, the Court expressly declined to decide the issue, leaving it open for possible first instance consideration by the state court upon remand. Later, the issue was inconclusively treated in plurality and dissenting opinions in Connecticut v. Johnson, 460 U.S. 73, 103 S.Ct. 969, 74 L.Ed.2d 823 see note 2, infra. Most recently, possible authoritative resolution was prevented by an equal division of the Court in Engle v. Koehler, 707 F.2d 241 (6th Cir.1983), aff’d by an equally divided court, — U.S.-, 104 S.Ct. 1673, 80 L.Ed.2d 1 (1984).
This court has not before addressed and decided the issue directly. But cf. Tweety v. Mitchell, 682 F.2d 461 (4th Cir.1982).
Other federal courts are in substantial conflict both in their approaches to and their resolutions of the issue as it has been presented in a variety of contexts. See, e.g., In re Hamilton, 721 F.2d 1189, 1192 (9th Cir.1983) ("reviewing court cannot rationally conclude beyond a reasonable doubt that a juror did not rely on an unconstitutional Sandstrom instruction where intent is a disputed issue"; weight of evidence is immaterial); Engle v. Koehler, 707 F.2d 241, 246 (6th Cir.1983) (if intent a disputed issue, Sandstrom instruction "can be prejudicial even if overall proof of intent or malice is substantial”), aff’d by an equally divided court, — U.S.-, 104 S.Ct. 1673, 80 L.Ed.2d 1 (1984); Lamb v. Jernigan, 683 F.2d 1332, 1341 (11th Cir.1982) (burden-shifting instruction may be “harmless if the evidence of guilt was so overwhelming that the error cannot have contributed to the jury’s decision to convict”); United States v. Alston, 551 F.2d 315, 320 & n. 24 (D.C.Cir.1976) ("strong evidence of guilt" may support finding that burden-shifting instruction was not prejudicial); Trimble v. Stynchcombe, 481 F.2d 1175, 1176 (5th Cir.1973) (per curiam) (harmless error principle inapplicable where alibi instruction is wholly inconsistent with the reasonable doubt instruction); Stump v. Bennett, 398 F.2d 111, 123 (8th Cir.1968) (en banc) (instruction that shifts burden of proof on alibi not harmless even if "evidence conclusively demonstrates guilt”); Simmons v. Dalsheim, 543 F.Supp. 729, 749 (S.D.N.Y.1982) (burden-shifting alibi instructions may be found harmless, but "only in a rare case’’); Graham v. Maryland, 454 F.Supp. 643 (D.Md.1978) (harmless error doctrine inapplicable to erroneous instructions that shift burden of proof with regard to an alibi defense).
. Of these decisions, only Sandstrom and Johnson advert to the harmless error possibility. Sandstrom, as indicated in note 1, supra, expressly declined to address it.
Concededly, both the plurality and the dissenting opinions in Johnson employ harmless *1036error language in discussing whether the incorporation of a "Sandstrom presumption” in jury instructions required, in the dissenting opinion’s terms, "automatic reversal," Johnson, 460 U.S. at 95, 103 S.Ct. at 982. But, with the greatest deference, I suggest that the focal point of the actual debate within the Johnson Court was only upon the predicate question of constitutional error: whether the presumption did effectively relieve the state of the burden of proving intent. And on this, the real point of division between the two opinions appears to be the proper range of record inquiry to determine that predicate question. Justice Blackmun, for the plurality, thought the inquiry should be confined to the instructions themselves, considered as a whole. On that basis he concluded that the instructions must be considered to have been "the functional equivalent of a directed verdict on [the] issue [of intent].’’ Id. at 84, 103 S.Ct. at 976. Justice Powell, writing for the four members of the Court dissenting, thought instead that for the limited purpose of determining whether the presumption — in the manner of a directed verdict — effectively took intent out of issue, the Court could properly take into account not only the instructions but all the evidence of intent that was before the jury. On this basis he thought the state had not been relieved of the burden to prove intent by the presumption considered in this fuller context.
Critically, both the plurality and the dissenting opinions in Johnson agree that an instruction that does effectively relieve the state of the burden of proving a disputed element of the crime can never be deemed harmless; they differ only in their assessments of the effect of the presumption there in issue. See id. at 97-98 & n. 6, 103 S.Ct. at 983-984 & n.6 (Powell, J., dissenting). Put another way, both agree that at some point burden-shifting instructions become harmful per se, see id. at 95 n. 3 (Powell, J., dissenting); they differ only on the point at which this occurs and on the scope of the record inquiry appropriate to determine this. Neither indicated any belief that if such an instructional error occurred, it might then be found harmless because of the overwhelming evidence of guilt.
. In attempting to divine the probable view of a majority of the Court on this unresolved but insistently recurring issue, it cannot be amiss, in light of the division of the Court in Connecticut v. Johnson, to ponder the implications of Engle v. Koehler, 707 F.2d 241, 246 (6th Cir.1983), aff’d by equally divided court, — U.S.-, 104 S.Ct. 1673, 80 L.Ed.2d 1 (1984) (Marshall, J., not participating).
. Though the instructions in both cases included accurate general statements placing the proper burden of proof upon the state in respect of all elements of the crime, and recognizing the general presumption of innocence, there is apparently no disagreement within this panel that, construed in total context, the instructions must be assumed erroneously to have shifted to the accused in each case the burden to prove his nonpresence by a preponderance of the evidence. See Cooper v. North Carolina, 702 F.2d 481, 483 (4th Cir.1983) (effect of directly conflicting portions of instructions).
In Fulton, the court instructed, inter alia: "The defendant has the burden of proving an alibi defense and must do so by a preponderance of the evidence, not beyond a reasonable doubt. In order to prove an alibi the testimony must cover the whole time in which the crime by any possibility might have been committed and it should be subjected to rigid scrutiny.”
In Robinson, in the course of a much more extended and internally ambiguous passage devoted to the alibi defense, the court instructed, inter alia: "[A]n alibi is what we call an affirmative defense. It is of course a complete defense if you believe it. Obviously a person who is not at the scene could not complete the crime. We say it is an affirmative defense because this is the one time in this case where the defendant has the burden of proof.... On the defense of alibi the defense [sic] has the responsibility to satisfy you by a fair preponderance of the evidence that he was not in fact in the store and in fact was somewhere else.”
Neither instruction was specifically corrected before the case was submitted to the jury, for the obvious reason that it faithfully applied — albeit ambiguously — then current state law.
. Although there might conceivably have been a question under state law as to whether the admittedly sparse alibi evidence in Fulton’s case was minimally sufficient to require a special instruction — i.e., to "raise” the alibi defense— that of course has nothing to do with the question whether for our purposes it was a disputed issue before the jury. Whether or not he was required to give the instruction as a matter of state law, the judge in giving it effectively made it an issue for jury consideration. Presence was of course put in general issue by the not guilty plea.
. In In re Winship, 397 U.S. at 364, 90 S.Ct. at 1072, the Court had this to say about the transcending importance — existing apart from the sufficiency of evidence of guilt — of the due pro*1039cess right not to be convicted "except upon proof beyond a reasonable doubt of every fact necessary to constitute the crime ... charged":
“[U]se of the reasonable-doubt standard is indispensable to command the respect and confidence of the community in application of the criminal law. It is critical that the moral force of the criminal law not be diluted by a standard of proof that leaves people in doubt whether innocent men are being condemned. It is also important in our free society that every individual going about his ordinary affairs have confidence that his government cannot adjudge him guilty of a criminal offense without convincing a proper factfinder of his guilt with utmost certainty.”
As Mullaney and its progeny have made plain, this right is effectively denied by instructional error that relieves the state of this burden of proof in respect of any element of the crime by shifting it to the accused. To allow such an error to be found harmless would convert the right merely into one to have guilt adjudged under the proper standard of proof except where the evidence was, in the eyes of an appellate court, sufficiently “overwhelming.” That, I submit, would fundamentally dilute the constitutional right recognized in Winship and directly vindicated in Mullaney in both of which cases it is highly likely that the evidence of guilt might also have been thought "overwhelming” by a reviewing court.
. Only within very narrow bounds may the weight of the evidence properly be considered by an appellate court in assessing the possible harmlessness of error. With respect to trial errors, such as the admission or exclusion of evidence, whose "scope is readily identifiable,” and whose possible effect upon the verdict may therefore be rationally assessed, the weight of the evidence may be taken into account in assessing whether the case was a close one in which the error may have tipped the balance. See United States v. Nyman, 649 F.2d 208, 212 (1981). This must be distinguished from the inappropriate process, frequently confused with it, of assessing whether, disregarding the error, there was ample evidence to support the verdict. Id.
. With all respect, the majority's harmless error analysis is precisely of this type and well illustrates the reason why the instructional errors here must be deemed per se harmful if the basic constitutional right is to be secured.
In demonstrating the harmlessness of error in Robinson's case, for example, the majority has considered it necessary to declare, inter alia, that: "the fingerprints are conclusive”; "Robinson was in the store; the alibi was a fabrication beyond any conceivable doubt"; there was no "articulable possibility that the alibi could be true.”
I agree that as a practical matter only by this sort of record analysis could this instructional error be declared harmless beyond a reasonable doubt. I further agree that these factual conclusions are the most rational ones that could be drawn by a factfinder charged with first instance adjudication on the records we review. But I do not believe that this is a proper function of harmless error review by an appellate court, whatever the degree of error, and particularly when, as here, the error is of constitutional dimension.
. Under this approach, the only and dispositive inquiry is whether the instructional error was unconstitutional. The identified predicates establish the content and contextual limits of that inquiry. In my view, the limits so defined give the maximum flexibility that is possible for an inquiry designed effectively to protect both the interests of the state in the finality and invulnerability to purely technical attacks of its judgments, and the constitutional right of the accused not to be convicted by a jury except under the proper standard of proof.
This approach concededly is at odds — though to some extent probably only semantically— with that taken by other courts which in assessing these predicates have done so in "harmless error” terms. See, e.g., Hearn v. James, 677 F.2d 841, 843 (11th Cir.1982); see also Johnson, 460 U.S. at 90, 103 S.Ct. at 979 (Powell, J., dissenting) (semble).
There is, however, a sound conceptual basis both for confining the inquiry in this way and for insisting that it relates only to the existence of constitutional error, not to its harmlessness. The basis is that the due process inquiry as to trial error — to determine whether it is of constitutional dimension — is itself aimed at deciding whether the error resulted in "fundamental unfairness.” Such an inquiry seems — both conceptually and practically — essentially to subsume any further question of "harmlessness.” See Krzeminski v. Perini, 614 F.2d at 125. More important than conceptual tidiness however is a practical effect — that by this means the inquiry is stopped short of any assessment of the weight of the evidence of guilt as a possible basis for finding the "error” harmless.
. Of undoubted concern in this, as in all cases of collateral federal attacks on state convictions, is the spectre of opening the doors on unknown numbers of comparably flawed convictions. The concern is of course magnified where, as here, the constitutional flaw alleged is one based upon a long-standing, previously unchallenged state rule which presumably was followed in many cases. The concern is one not lightly to be dismissed, but it is one that cannot of course control decision. Cf., e.g., Mullaney, Sandstrom, Johnson. Questions of retroactivity, cf. Hankerson v. North Carolina, 432 U.S. 233, 97 S.Ct. 2339, 53 L.Ed.2d 306 (1976); of staleness, see 28 U.S.C. § 2254, Habeas Rule 9(a); see also id. (proposed amendment Aug. 19, 1983); and of procedural default, see Hankerson, 432 U.S. at 244 n. 8, 97 S.Ct. 2339 at 2345 n. 8, would of course control the ultimate extent of the precedential consequence.