MACK, Associate Judge:
The principal question presented by this collateral attack upon a criminal conviction is the degree of retroactivity to be attached to our constitutional interpretation in
Arnold v. United States,
467 A.2d 136 (D.C.1983). We hold that
Arnold
is fully retroactive, and that appellant Kinston Kirk is entitled to benefit from the double jeopardy rule it sets forth. We hold also that our rejection of appellant’s claim on direct appeal in an unpublished memorandum order issued prior to
Arnold
does not preclude his raising the issue again by way of this § 23-110 motion.
On May 11,1979, appellant was convicted of grand larceny while armed, D.C.Code §§ 22-3202, -3812 (1981 & 1985 Supp.), and unauthorized use of a vehicle, D.C.Code § 22-3815 (1985 Supp.).
On direct appeal, appellant asked us to consider “[w]hether, as a matter of law or under the facts of the case, [his] consecutive sentence for unauthorized use of a motor vehicle merge[d] with his conviction for grand larceny of the same motor vehicle which arose from the same transaction.” On May 21, 1981, a division of this court rejected appellant’s contention and affirmed his convictions in an unpublished opinion.
On October 5, 1983, reaching the opposite conclusion on indistinguishable facts, another division held that the Double Jeopardy Clause of the Fifth Amendment prohibits separate convictions for grand larceny and unauthorized use of a vehicle arising out of the same transaction.
Arnold, supra,
467 A.2d at 138-39. Appellant returned to the trial court and unsuccessfully argued that his unauthorized use conviction should be vacated in view of our decision in
Arnold.
Because
Arnold
is fully retroactive, the trial court should have granted appellant’s § 23-110 motion to vacate an unconstitutional conviction and sentence; moreover, this court’s prior rejection of the same argument on direct appeal does not bar the present division from affording appellant his requested relief in light of the intervening caselaw. We reverse and remand.
I
On the afternoon of July 9, 1979, Eric Mann was driving through the District of Columbia on his way from Atlantic City, New Jersey, to his home in North Carolina. He stopped on Kenilworth Avenue, N.E., to examine his tires and a loose stereo speaker. As Mann did so, he noticed two strangers standing by a nearby fence. Suddenly these men, appellant and codefendant Melvin Thomas, approached Mann from behind, placed a sawed-off shotgun in his back, and ordered their captive back into the car. Mann was forced to drive for a while and then, at gunpoint, to hand over his watch and his wallet. At some point, appellant took over the wheel and the victim was moved into the back seat. About 30 to 45 minutes after the ride began, Mann was abandoned in Marlow Heights, Maryland, and appellant and Thomas drove off in his car.
Mann succeeded in stopping a passing motorist for help and went to a local police station to report the incident. The. police took him back to the neighborhood where the kidnapping began. After a ten-minute search, Mann spotted the stolen car in an alley, with Thomas leaning inside and appellant standing next to it. Both men fled. Thomas was apprehended after a brief chase and identified at the scene by Mann. Appellant was later arrested On a warrant and identified by Mann in a line-up and at trial.
Appellant received three concurrent sentences of fifteen to forty-five years imprisonment, including the term for his grand larceny while armed.
It is the consecutive sentence of eighteen to fifty-four months
for unauthorized use of a motor vehicle that appellant now challenges under the double jeopardy rule in
Arnold.
II
In
Arnold,
we held that dual convictions for grand larceny and unauthorized use of a motor vehicle were improper. We reasoned that because the grand larceny contained within it the unauthorized use of the vehicle, and the unauthorized use conviction required proof of no fact which the grand larceny did not, the defendant had been subjected to multiple punishments for the “same” offense.
See Brown v. Ohio,
432 U.S. 161, 166-68, 97 S.Ct. 2221, 2225-26, 53 L.Ed.2d 187 (1977);
Blockburger v. United States,
284 U.S. 299, 304, 52 S.Ct. 180, 182, 76 L.Ed. 306 (1932). Such cumulative sentences are prohibited by the Double Jeopardy Clause of the Fifth Amendment.
The conviction and sentence for unauthorized use of a vehicle were therefore unconstitutional and we remanded to the trial court with instructions that they be vacated.
Our conclusion in
Arnold
was arrived at by looking at the facts of the case rather than by limiting our analysis to “abstract consideration of the statutes involved.” 467 A.2d at 139 (quoting
Hall v. United States,
343 A.2d 35, 39 (D.C.1975)). This factual appraisal was necessary because we recognized that in some circumstances “it is possible that a conviction for grand larceny might be had without proof of this particular element of unauthorized use.”
Arnold, supra,
467 A.2d at 139. An examination of the factual record before us here is accordingly required in order for us to determine whether appellant’s conviction for unauthorized use of a vehicle merged with the more serious. offense of grand larceny while armed.
In this regard, the record is crystal clear. Both offenses arose out of the same fact pattern; there was nothing in appellant’s unlawful use of a vehicle conviction that was not also used as proof of the grand larceny while armed. Under our holding in
Arnold,
we have no choice but to conclude that appellant’s unlawful use of a vehicle conviction violated his Fifth Amendment guarantee against double jeopardy, and that the resulting eighteen to fifty-four months sentence of consecutive imprisonment meted out by the trial court was unconstitutional.
Ill
The government concedes the unconstitutionality of the consecutive sentence of imprisonment, acknowledging, with admirable candor, “that the facts of this case cannot meaningfully be distinguished from
Arnold.
But even so,” it continues less generously, “appellant’s claim must fail.” The sole justification urged for keeping appellant incarcerated pursuant to an unconstitutional sentence is that this precise issue has been decided adversely to appellant on his direct appeal. The decision in
Arnold,
the government admits, would otherwise control.
In support of this position, the government cites
M.A.P. v. Ryan,
285 A.2d 310 (D.C.1971), in which we announced that “[a]s a matter of internal policy, we have adopted the rule that no division of this court will overrule a prior decision of this court ... and that such result can only be accomplished by this court en banc.”
Id.
at 312 (footnote omitted). Our Internal Operating Procedures (IOPs) echo this policy: “To avoid conflicts in division decisions and
to preserve stability of the court’s decisions, no subsequent division may overrule a published opinion of a previous division. En banc consideration is required to overrule a previous decision of the court....” D.C.App. IOP § VIII H (1985). Because our prior rejection of appellant’s merger contention was by way of an unpublished opinion, which does not normally constitute binding authority, the government also refers us to another provision in our IOPs:
Decisions designated as not for publication shall not be cited to the court in any material prepared for the court
except
when the opinion is relevant under the doctrines of the law of the case, res judicata or collateral estoppel, or in a criminal action or proceeding involving the same defendant....
Id.
§ IX C (emphasis added). The combined effect of
M.A.P. v. Ryan
and IOP §§ VIII H and IX C, argues the government, precludes this division — though, strangely, not the Superior Court — from considering the acknowledged unconstitutionality of appellant’s sentence of imprisonment. Having unsuccessfully raised the merger issue on direct appeal, the argument goes, appellant must live with his defeat unless he can point to intervening authority from the Supreme Court or else persuade us to revisit the matter en banc.
The government, in making this argument, ignores the purpose behind our rule that one division cannot overrule another. The rule is designed to preserve uniformity in the decisions of this court.
Arnold
did not undermine that policy because its double jeopardy interpretation ran counter to no published opinion on the same point. Yet the government contends that a division such as this should disregard our otherwise binding decision in that case and thus rule differently from every other division that has confronted the issue since publication of
Arnold
Instead, the government contends, we are bound to follow a prior unpublished decision which cannot be cited in any case other than the present. In our view the ideal of uniformity in the decisions of this court, as expressed in
M.A.P. v. Ryan
and our IOPs, is defied by the government’s position.
The government also fails to comprehend the true nature of a collateral attack. One primary purpose of § 23-110 is to enable convicted prisoners to escape the shackles of res judicata when constitutional rights have been violated or other illegalities have occurred in their sentencing. By its very definition, a collateral attack on a tainted sentence involves a challenge to the decision of the court that has previously adjudicated the issue. Despite this fact, § 23-110 requires the courts of the District of
Columbia to be eternally vigilant in ensuring that prisoners are not subject to unlawful incarceration. For this reason we have necessarily held that strict principles of res judicata — and the government invokes them here in their harshest form — do not apply in these proceedings.
Pettaway v. United States,
390 A.2d 981, 986 (D.C.1978);
See Sanders v. United States,
373 U.S. 1, 8, 83 S.Ct. 1068, 1073, 10 L.Ed.2d 148 (1963). In
Sanders,
the Supreme Court considered the analogous federal vehicle for collateral attack, 28 U.S.C. § 2255 (1982), and held that “[t]he inapplicability of
res judicata
to habeas ... is inherent in the very role and function of the writ.” The Supreme Court has accordingly held that a claim already rejected by an appellate court may be reconsidered on collateral attack in light of new law which would have exonerated the defendant had it been in force at the time of the direct appeal.
Davis v. United States,
417 U.S. 333, 342, 94 S.Ct. 2298, 2303, 41 L.Ed.2d 109 (1974). The federal approach has equal validity with regard to § 23-110 collateral attacks. In short, our internal policies do not require a division of this court to stand idly by when an unconstitutional sentence is about to be executed.
Aside from its incompatibility with both our ideal of uniform decisions and the humane purpose of § 23-110, the government’s argument leads to an absurd result: it would penalize appellant for having correctly argued his appeal in the first place. A defendant who had failed to raise the merger contention on direct appeal, in contrast to appellant, would now be at liberty to reap the benefits of
Arnold
by way of a § 23-110 motion,
c.f. Garris v. United States,
491 A.2d 511, 513 (D.C.1985)
(Garris II);
appellant, on the other hand, is to be prevented from doing so only because he has previously pointed out the very issue in which we later identified a constitutional violation. Such a capricious rule is not to be tolerated. We conclude that our rejection of appellant’s merger argument on direct appeal does not preclude him from raising the same claim, citing intervening authority, by way of a subsequent § 23-110 motion.
IV
The question remaining, one not addressed by the government, is whether
Arnold
applies retroactively so as to benefit a defendant challenging a final conviction by way of collateral attack. “As a rule, judicial decisions apply ‘retroactively.’”
Solem v. Stumes,
465 U.S. 638, 642, 104 S.Ct. 1338, 1341, 79 L.Ed.2d 579 (1984) (citation omitted). Until 1965, the Supreme Court had given full retroactive effect to every one of its new interpretations protecting the constitutional rights of criminal defendants. Up to that time, the accepted Blaek-stonian notion was that a new rule, although only recently “discovered” by the court, had nonetheless always existed as one of the eternal truths of the law. As a matter of logical necessity, therefore, every litigant had an equal right to benefit from its application, irrespective of whether the claim was asserted before or after the rule became exposed to public view. In
Linkletter v. Walker,
381 U.S. 618, 85 S.Ct. 1731, 14 L.Ed.2d 601 (1965), the Supreme Court abandoned this declaratory theory. In its place, the Court adopted the Austinian approach that a court makes, rather than discovers, law and should accordingly render retroactive, partially retroactive, or prospective rulings as the circumstances of the case require.
See generally Gary v. United States,
499 A.2d 815, 830-32 (D.C.1985) (en banc);
Mendes v. Johnson,
389 A.2d 781, 787-89 (D.C.1978) (en banc).
After
Linkletter,
the Supreme Court was frequently called upon to determine the extent to which its new interpretations
affecting the constitutional rights of criminal defendants should be retroactively applied. Relying upon a three-part test, the Court normally inquired into: “(a) the purpose to be served by the new standards, (b) the extent of the reliance of law enforcement authorities on the old standards, and (c) the effect on the administration of justice of a retroactive application of the new standards.”
Stovall v. Denno,
388 U.S. 293, 297, 87 S.Ct. 1967, 1970, 18 L.Ed.2d 1199 (1967).
The outcome of the balancing test varied considerably. On one side of the scales some new rules received full retroactive effect in all subsequent cases.
E.g., Brown v. Louisiana,
447 U.S. 323, 330-37, 100 S.Ct. 2214, 2220-24, 65 L.Ed.2d 159 (1980) (plurality opinion);
Berger v. California,
393 U.S. 314, 315, 89 S.Ct. 540, 541, 21 L.Ed.2d 508 (1969) (per curiam);
Arsenault v. Massachusetts,
393 U.S. 5, 6, 89 S.Ct. 35, 36, 21 L.Ed.2d 5 (1968) (per cu-riam). On the other, at least one new rule was applied with complete prospectivity, not even the parties before the court bene-fitting from the change they had crafted in the law.
Morrissey v. Brewer,
408 U.S. 471, 490, 92 S.Ct. 2593, 2604, 33 L.Ed.2d 484 (1972). Between these extremes, other new rules received partial retroactivity, measured from a number of points: some new rules applied retroactively only to the litigant in the rule-changing case and to no other,
e.g., United States v. Peltier,
422 U.S. 531, 534-35, 95 S.Ct. 2313, 2315-16, 45 L.Ed.2d 374 (1975);
Michigan v. Payne,
412 U.S. 47, 57, 93 S.Ct. 1966, 1971, 36 L.Ed.2d 736 (1973);
Adams v. Illinois,
405 U.S. 278, 284-85, 92 S.Ct. 916, 920, 31 L.Ed.2d 202 (1972) (plurality opinion);
Stovall, supra,
388 U.S. at 296-301, 87 S.Ct. at 1969-72; some only to cases where the evidence introduced by the government at trial had become illegal by virtue of the
new
interpretation,
e.g., Desist v. United States,
394 U.S. 244, 89 S.Ct. 1030, 22 L.Ed.2d 248 (1969);
Fuller v. Alaska,
393 U.S. 80, 89 S.Ct. 61, 21 L.Ed.2d 212 (1968) (per curiam); some only to those whose trials had not yet begun when the new rule was established,
e.g., De Stefano v. Woods,
392 U.S. 631, 635, 88 S.Ct. 2093, 2096, 20 L.Ed.2d 1308 (1968) (per curiam);
Johnson v. New Jersey,
384 U.S. 719, 721, 86 S.Ct. 1772, 1774, 16 L.Ed.2d 882 (1966); and some decisions applied retroactively to all convictions which were not yet final when the new rule was announced,
e.g., Linkletter, supra,
381 U.S. 618, 85 S.Ct. at 1732.
This extensive array of chosen options drew charges, both from within
and without
the Court, that its application of the
Linkletter
doctrine rested on no principled basis. Perhaps the most influential of the critics was Justice Harlan. In two opinions,
Mackey v. United States,
401 U.S. 667, 675, 91 S.Ct. 1160, 1164, 28 L.Ed.2d 404 (1971) (concurring in part and dissenting in part) and
Desist, supra,
394 U.S. at 256, 89 S.Ct. at 1037 (dissenting), Justice Harlan expressed his view “that all ‘new’ rules of constitutional law must, at a minimum, be applied to all those cases which are still subject to direct review by this Court at the time the ‘new’ decision is handed down.”
Id.,
394 U.S. at 258, 89 S.Ct. at 1039;
see Mackey, supra,
401 U.S. at 679-81, 91 S.Ct. at 1173-74. The imperatives of principled decisionmaking did not, however, prevent Justice Harlan from proposing a different approach for cases coming before the Court by way of collateral attack. Justice Harlan understood the purpose of a collateral attack as being only to ensure proper application of the law prevailing at the time the conviction became final. Consequently, retroactive application of subsequent changes in constitution
al interpretation was unnecessary and, moreover, unwise because it ignored the need for finality.
Mackey, supra,
401 U.S. at 682-94, 91 S.Ct. at 1174-81;
Desist, supra,
394 U.S. at 268, 89 S.Ct. at 1044. Justice Harlan would permit retroactivity on collateral attack in only two situations: “new ‘substantive due process’ rules, that is, those that place, as a matter of constitutional interpretation, certain kinds of primary, private individual conduct beyond the power of the criminal law-making authority to prescribe....”,
Mackey, supra,
401 U.S. at 692, 91 S.Ct. at 1180, and nonobservance of those procedures that are “implicit in the concept of ordered liberty.”
id.
at 693, 91 S.Ct. at 1180 (quoting
Palko v. Connecticut,
302 U.S. 319, 325, 58 S.Ct. 149, 152, 82 L.Ed. 288 (1937)).
In recent years, the Supreme Court has agreed with Justice Harlan’s belief that “[r]etroaetivity must be rethought.”
Johnson, supra
note 11, 457 U.S. at 548, 102 S.Ct. at 2586 (quoting
Desist, supra,
394 U.S. at 258, 89 S.Ct. at 1038). In
Johnson,
decided in 1982, the Court was asked to give retroactive effect to a Fourth Amendment interpretation prohibiting the police from.making a warrantless and nonconsen-sual entry into a suspect’s home to carry out a routine felony arrest.
Payton v. New York,
445 U.S. 573, 100 S.Ct. 1371, 63 L.Ed.2d 639 (1980) (evidence inadmissible under Fourth Amendment right against unreasonable searches and seizures). Embracing Justice Harlan’s view to the extent necessary to decide the case, the
Johnson
Court held that its Fourth Amendment decisions are to be applied retroactively to convictions not yet final at the time the new rule of interpretation is established.
Id.,
457 U.S. at 562-63, 102 S.Ct. at 2593-94. The Supreme Court expressly avoided any extension of the Harlan approach to cases, such as the present, implicating rights outside the Fourth Amendment or arising on collateral attack.
Id.
The doubtful retroactivity attaching to new constitutional interpretations not falling within the bounds of
Johnson
was partially resolved two years later in
Stumes, supra,
465 U.S. 638, 104 S.Ct. 1338. A prisoner alleged that his conviction rested on a confession obtained contrary to a later decision that, once a suspect has invoked the right to counsel, any subsequent conversation must be initiated by him or her.
Edwards v. Arizona,
451 U.S. 477, 101 S.Ct. 1880, 68 L.Ed.2d 378 (1981) (statement inadmissible under Fifth Amendment right against self-incrimination). The
Stumes
Court assumed what it later referred to as “an obvious
Edwards
violation.”
Shea v. Louisiana,
470 U.S. 51, -, 105 S.Ct. 1065, 1069, 84 L.Ed.2d 38, 45 (1985). Noting that
Stumes,
unlike
Johnson,
involved a collateral attack rather than direct review, and a new Fifth rather than Fourth Amendment interpretation, the Court applied the three-part
Stovall
balancing test to determine the retroactive effect of the new rule. It concluded, all things considered, that
Edwards
should not be retroactively applied.
Stumes, supra,
465 U.S. at 643, 104 S.Ct. at 1341.
More recently, the Supreme Court further refined the rethinking of retroactivity doctrine begun in
Johnson.
The setting of
Shea, supra,
470 U.S. 51, 105 S.Ct. 1065, 84 L.Ed.2d 38, was another attempt to obtain retroactive application of
Edwards.
The only material difference from the previous year’s unsuccessful effort in
Stumes
was that
Shea
came before the Court on direct review rather than collateral attack. The question to be decided, therefore, was whether Justice Harlan’s view — that new rules should be applied retroactively to all cases pending on direct appeal — should be adopted outside the Fourth Amendment context where it had won relative support in
Johnson.
The Court, with some limitations, answered yes: “Justice Harlan’s reasoning — that principled decisionmaking and fairness to similarly situated petitioners requires application of a new rule to all cases pending on direct review — is applicable with equal force to the situation presently before us.” 470 U.S. at -, 105 S.Ct. at 1070, 84 L.Ed.2d at 47.
In the wake of
Johnson, Stumes
and
Shea,
it appears that the Supreme Court’s nonretroactivity doctrine has undergone a comprehensive revision since first launched in
Linkletter,
at least with respect to Fourth and Fifth Amendment violations. A modified version of the Harlan approach has prevailed with respect to cases coming before the Court on direct review, so that most cases not yet final can henceforth be expected to receive the benefit of the new rule.
Shea, supra,
470 U.S. at ---, 105 S.Ct. at 1069-70, 84 L.Ed.2d at 46-47 (Fifth Amendment);
Johnson, supra
note 11, 457 U.S. at 562-63, 102 S.Ct. at 2593-94 (Fourth Amendment). But the Supreme Court has declined to adopt Justice Harlan’s restrictive attitude towards the retroactive application of new rules when the benefit is sought on collateral attack.
Stumes, supra,
465 U.S. at 643, 104 S.Ct. at 1341 (Fifth Amendment);
Johnson, supra
note 11, 457 U.S. at 562-63, 102 S.Ct. at 2593-94 (Fourth Amendment). Instead, collateral attacks remain subject to
Sto-vall’s
three-part balancing test,
Stumes, supra,
465 U.S. at 643, 104 S.Ct. at 1341, the test which has now been abandoned with respect to cases pending on direct appeal at the time the new rule is announced.
The broad principles emerging from the Supreme Court’s recent retroactivity decisions must be read subject to certain limitations. In particular, the Court has not given unqualified approval to Justice Harlan’s absolutist view that a new rule should be applied retroactively to
every
case not yet decided on direct appeal; in
Shea,
the Court specifically declined to consider “situations clearly controlled by existing re-troactivity precedents, such as where the new rule of law is so clear a break with the past that it has been considered nonretroac-tive almost automatically.” 470 U.S. at - n. 5, 105 S.Ct. at 1070 n. 5, 84 L.Ed.2d at 46 n. 5;
see also Stumes, supra,
465 U.S. at 647, 104 S.Ct. at 1343;
Johnson, supra
note 11, 457 U.S. at 449-50, 102 S.Ct. at 2586-87. Conversely, there are other constitutional interpretations that are automatically retroactive, whether invoked on direct review or collateral attack. The Supreme Court has observed that when its decision “merely has applied settled precedents to new and different factual situations ... it has been a foregone conclusion that the rule in the later case applies in earlier cases, because the later decision has not in fact altered that rule in any material way.”
Johnson, supra
note 11, 457 at 549, 102 S.Ct. at 2586. Another such case is “the sort of decision that goes to the heart of the truthfinding function, which we have consistently held to be retroactive.”
Stumes, supra,
465 U.S. at 645, 104 S.Ct. at 1343. Finally, and for our purposes
critically, the Supreme Court “has recognized full retroactivity as a necessary adjunct to a ruling that a trial court lacked authority to convict or punish a criminal defendant in the first place.”
Johnson, supra
note 11, 457 U.S. at 550-51, 102 S.Ct. at 2587. In this last category of cases, where jurisdiction was always lacking, the new rule does not so much apply retroactively as reveal the fact that the prior judgment or sentence was void ab initio.
Id.
Appellant claims on this collateral attack that our constitutional interpretation in
Arnold
should be applied retroactively so as to invalidate his sentence of eighteen to fifty-four months imprisonment for unauthorized use of a vehicle.
We need not concern ourselves here with the Supreme Court’s apparent rejection of the three
Sto-vall
factors when a conviction comes before it on direct review. A collateral attack, such as the present, remains generally subject to the
Stovall
balancing test. We are not required to weigh the
Stovall
factors in the present case, however, because
Arnold
is the classic example of “a ruling that a trial court lacked authority to convict or punish a criminal defendant in the first place.”
Johnson, supra
note 11, 457 U.S. at 550, 102 S.Ct. at 2587. The Fifth Amendment guarantee against double jeopardy serves to “prevent a trial from taking place at all, rather than to prescribe procedural rules that govern the conduct of a trial.”
Robinson v. Neil,
409 U.S. 505, 509, 93 S.Ct. 876, 35 L.Ed.2d 29 (1973) (double jeopardy decision in
Waller v. Florida,
397 U.S. 387, 90 S.Ct. 1184, 25 L.Ed.2d 435 (1970), fully retroactive);
see Ashe v. Swenson,
397 U.S. 436, 437 n. 1, 90 S.Ct. 1189, 1191 n. 1, 25 L.Ed.2d 469 (1970) (citing
North Carolina v. Pearce,
395 U.S. 711, 89 S.Ct. 2072, 23 L.Ed.2d 656 (1969)) (double jeopardy decision in
Benton v. Maryland,
395 U.S. 784, 89 S.Ct. 2056, 23 L.Ed.2d 707 (1969), fully retroactive).
Arnold,
as a holding that the trial court was without power to convict or punish, receives full retroactivity in all cases without resort to any balancing test; appellant is therefore automatically entitled to benefit from the wholly retroactive effect of our double jeopardy decision in
Arnold.
Our previous rejection of appellant’s contention does not preclude his raising the same claim, citing intervening authority, by way of a § 23-110 collateral attack. On the merits, we hold that
Arnold
is fully retroactive, and that appellant is entitled to the benefit of it. We reverse and remand for vacation of the conviction and sentence for unauthorized use of a vehicle, both having been imposed in violation of appellant’s Fifth Amendment right not to be twice put in jeopardy for the same offense.
So ordered.