(dissenting). I agree with the court that the jury instructions at Blackwell’s trial contained three distinct errors. The judge (1) defined malice in terms of “frame of mind” contrary to Commonwealth v. Torres, 420 Mass. 479, 486-487 (1995); (2) gave a vague definition of “extreme atrocity or cruelty” held defective in Commonwealth v. Hunter, 416 Mass. 831, 837 (1994); and (3) allowed the jury to consider culpability on a theory of extreme atrocity or cruelty, which the evidence did not support. Yet even in the face of those three serious errors, see Commonwealth v. Glass, 401 *302Mass. 799, 804 (1988) (error not prejudicial only if jury’s consideration of premeditation theory was “substantially unsullied” by inappropriate submission of extreme atrocity or cruelty theory to jury), the court arrives at the conclusion that no remedy is warranted — the errors did not create a substantial risk of a miscarriage of justice.
In my view these errors were prejudicial, regardless of how “straightforward” the evidence may seem. See ante at 301. The right to trial by jury “embodies ‘a profound judgment about the way in which law should be enforced and justice administered.’ ” Carella v. California, 491 U.S. 263, 268 (1989) (Scalia, J., concurring in judgment), quoting Duncan v. Louisiana, 391 U.S. 145, 155 (1968). Commonwealth v. Skinner, 408 Mass. 88, 97 (1990). Of course, it is possible that the jury may have based the guilty verdict on premeditation, see Commonwealth v. Wallace, 417 Mass. 126, 134-135 (1994) (error in malice instruction was nonprejudicial in light of premeditation theory), or felony-murder. On the other hand, this court has no sure way of knowing whether the inappropriate and erroneous extreme atrocity instruction carried the day in the jury room. “[Wjhere the verdict is supportable on one ground, but not on another, and it is impossible to tell which ground the jury selected,” the verdict should be set aside. Yates v. United States, 354 U.S. 298, 312 (1957) (Harlan, J.), overruled on other grounds, Burks v. United States, 437 U.S. 1, 18 (1978). See Commonwealth v. Matchett, 386 Mass. 492, 511 (1982).1 It is not appropriate in *303this case for appellate judges who examine a cold record to do away with the jury of citizens, who share a courtroom with the defendant and the witnesses, to conduct, in effect, a retrial based on their speculation as to what the jury may have thought. See Carella, supra at 267.
It is true that in a narrow set of cases even instructional error can be found not prejudicial, because other verdicts can show what the jury actually concluded. In such an instance, an appellate court may be able to analyze carefully the jury charge and the evidence and find a precise path, free of error, that the jury took to reach their conclusion of guilty beyond a reasonable doubt. See, e.g., Wallace, supra.
Here, the court purports to carve such a path through a complex case of joint venture liability and an underlying felony with a host of lesser included offenses. I begin, as does the court, with the definition of felony-murder — (1) a homicide (2) committed during (3) the commission of a prerequisite felony. Commonwealth v. Pope, 406 Mass. 581, 584 (1990). While the court can logically discern unanimous jury findings as to (1) the killing and (3) the prerequisite felony, armed robbery while masked, still missing is the “committed during” portion of the definition of felony-murder.
The court concludes that the jury in fact found this causal link because only two of Blackwell’s actions could be considered assaults, and both allegedly have a sufficient causal link to the deaths. Ante at 300-301 n.2. This analysis ignores the fact that the armed robbery conviction necessarily rests on a joint venture theory (Blackwell was neither armed nor masked), not a direct assault by Blackwell. The jury could have rested their separate verdict on any action of the armed accomplice that amounted to assault. This opens a real likelihood that the precise events that the jury found sufficiently proved to demonstrate assault (as a component of the separate robbery conviction) were not the cause of the victim’s death. The court relies on Commonwealth v. Berry, 420 Mass. 95, 111-112 (1995), where a similar analysis identified a single, error-free, unanimous basis for a verdict. That case, however, did not involve the complexities of joint venture liability and the facts contained only one set of events that could support *304the separate assault conviction.2 Here the court has supplied a factual finding that we cannot be certain the jury themselves made — that the killing was “committed during” the precise events comprising the underlying felony.3
It matters not whether the evidence is “straightforward.” The simplest of fact patterns can encapsule the thickest morass of legal theories. It is the legal theory of the case as tried to which the jury must apply the facts. That path must be straightforward if we judges are legitimately to be able to discern a particular path that the jury took to the verdict. No such path properly can be discovered in this case. The defendant is entitled to a new trial. I dissent.
The court correctly notes that the Supreme Court has limited this rule to cases where the tainted conviction is legally infirm, as opposed to a “mere” insufficient evidence problem. See Griffin v. United States, 502 U.S. 46, 56 (1991) (in absence of legal error, jury presumed to have found insufficient evidence did not support guilty verdict, so that jury must have convicted only on factually sufficient grounds). It is true that one of the errors here was “factual” — that there was insufficient evidence to support extreme atrocity or cruelty. However, one other relevant error — the Hunter error — is legal and therefore the rule of Yates v. United States, 354 U.S. 298, 312 (1957), overruled on other grounds, Burks v. United States, 437 U.S. 1, 18 (1978), unquestionably applies. Furthermore, our case law suggests that we may apply this rule even to insufficiency problems. See Commonwealth v. Flynn, 420 Mass. 810, 818 (1995).
I did not participate in Commonwealth v. Berry, 420 Mass. 95 (1995). My comments on the court’s use of that case áre intended only to show it distinguishable. This trial occurred before the Berry decision, and the defense here did not request a unanimity instruction as to the theory of liability. Without such a request the lack of a unanimity instruction was not error on the basis of the common law theory set forth in Berry, supra at 112.
There was one other unanimous jury “finding” that the court claims to discern from the verdicts — that the defendant “was responsible for the victim’s death under a theory of joint venture.” Ante at 300. This cannot be used as a functional equivalent for the causation requirement of felony-murder. A joint venture theory can support a premeditation verdict, with the order to shoot demonstrating knowledge of cool reflection and assistance. E.g., Commonwealth v. Cohen, 412 Mass. 375, 380-381 (1992). The finding of joint venture responsibility demonstrates how the jury might have found a shared mens rea to kill the victim with premeditation. It does not show that Blackwell shared the mens rea for an armed robbery during which the killing occurred.