concurring.
Although I agree with the majority opinion that Lara v. Ryan, 455 F.3d 1080 (9th Cir.2006), requires us to treat the instructional error at issue here as a structural error, I write separately to emphasize that the result the majority reaches would be the right result even under a harmless error standard. I also write to express my respectful disagreement with the conclusion of my concurring colleague that Lara should be overruled.
I
Under the harmless error standard, the government bears the burden of demonstrating that the alleged error could not have affected the outcome. Chapman v. California, 386 U.S. 18, 23-24, 87 S.Ct. 824, 17 L.Ed.2d 705 (1967). Specifically, the government must show that it is “clear beyond a reasonable doubt that a rational jury would have found the defendant guilty absent the error.” Neder v. United States, 527 U.S. 1, 18, 119 S.Ct. 1827, 144 L.Ed.2d 35 (1999).
When applying the harmless error standard in a case of instructional error, the court must evaluate the record as a whole. Id. at 19, 119 S.Ct. 1827 (“Of course, safeguarding the jury guarantee will often require that a reviewing court conduct a thorough examination of the record.”). Although it might be possible to conclude that an individual verdict form was sufficiently clear to cure an instructional error, the relevant question must not be so limited. Rather, the court must determine whether the entire record, including not just curative instructions but all aspects of the jury’s deliberative process, undoubtedly demonstrates that a rational jury would have convicted the defendant even if the error had not been made. In other words, as the Supreme Court has specifically held, *679“a single instruction to the jury may not be judged in artificial isolation but must be viewed in the context of the overall charge.” Middleton v. McNeil, 541 U.S. 433, 437, 124 S.Ct. 1830, 158 L.Ed.2d 701 (2004) (quoting Boyde v. California, 494 U.S. 370, 378, 110 S.Ct. 1190, 108 L.Ed.2d 316 (1990)).
II
The government offers two theories for finding that the instructional error was harmless — or at least for finding that the California Supreme Court did not unreasonably apply Chapman v. California, 386 U.S. 18, 87 S.Ct. 824, 17 L.Ed.2d 705 (1967), when it concluded that the error was harmless.
First, the government points out that one of the three special circumstance robbery-murder instructions, CALJIC 8.80.1, required the jury to find that “the murder was committed while the defendant was engaged in or was an accomplice in ... robbery.” The state Supreme Court concluded and the government now contends that that instruction sufficiently cured any confusion arising from the flawed felony-murder instruction and from the typographical error in CALJIC 8.81.17 because it clearly required contemporaneity between the murder and the defendant’s involvement in the robbery.
Second, the government points out that the jury’s verdict form for special circumstance robbery-murder specified that Puli-do “engaged in or was an accomplice in the commission of or attempted commission of robbery during the commission of the crime charged in Count 1 [murder].” According to both the California Supreme Court and the government, that language on the verdict form demonstrates that the jury found “explicitly, unanimously and necessarily! ] that defendant’s involvement in the robbery ... commenced before or during the killing of Flores.” People v. Pulido, 15 Cal.4th 713, 63 Cal.Rptr.2d 625, 936 P.2d 1235, 1243-44 (1997).
As should be apparent, both of the government’s theories rely on isolated documents; in fact, both theories rely on isolated quotations found within isolated documents. The government argues that those isolated quotations sufficiently demonstrate the jury’s understanding of and reliance on contemporaneity. But the government’s and the California Supreme Court’s analysis of isolated documents, as noted in the first part of this opinion, does not suffice under Chapman and Neder to prove harmlessness beyond a reasonable doubt. Instead, the California Supreme Court and this court are required to review the entire record to determine whether the error caused harm. Our conclusion cannot rest on a verdict form and an allegedly curative instruction taken in isolation; it must include a careful review of the entire record to determine whether the government’s interpretation of isolated documents is consistent with other evidence of the jury’s probable and actual thinking.
Ill
Once the entire record is considered in this case, it becomes apparent that the supposedly curative instruction actually aggravated, rather than curing, the instructional error. Furthermore, the entire record reveals that the multiple instructional errors at issue in this case had great potential to — and did in fact — cause jury confusion. As a result, one cannot conclude beyond a reasonable doubt that the flawed instructions were irrelevant to the outcome.
A
First, the portion of CALJIC 8.80.1 that supposedly cured the errors is an extreme*680ly small portion of that instruction; the instruction as a whole, particularly when read together with other instructions given, actually aggravates the errors in the felony-murder instructions and the special circumstance instruction, CALJIC 8.81.17.
The allegedly curative language is contained solely in the legal formulation of the special circumstance, which states that the special circumstance is true if “the murder was committed while the defendant was engaged in or was an accomplice to ... robbery.” The same instruction, however, later clarifies the standards for finding the special circumstance under various factual scenarios. Most importantly for purposes of this case, the instruction provides special standards for cases in which the “defendant was not the actual killer of a human being” or in which the jury is “unable to decide whether the defendant was the actual killer[.]” CALJIC 8.80.1, at 2. That is, the instruction lays out special guidelines for convicting a non-triggerman of the special circumstance.
Because the jury hung on the question of whether Pulido used a gun, we must assume that the jury’s special circumstance finding relied on those specific standards for non-triggermen. In other words, because we know that the jury could not decide whether Pulido was the triggerman and because we must assume that the jury followed all instructions, we must also assume that the jury followed the special circumstance guidelines that apply when a jury is “unable to decide whether the defendant was the actual killer.”
In turn, those standards advise the jury that non-triggermen are ineligible for the special circumstance “unless you [the jury] are satisfied beyond a reasonable doubt that the defendant with the intent to kill aided, abetted, counseled, commanded, induced, solicited, requested, or assisted any actor in the commission of the murder in the first degree, or with reckless indifference to human life and as a major participant, aided, abetted, counseled, commanded, induced, solicited, requested, or assisted in the commission of the crime of robbery ... which resulted in the death of a human being.” Id. The instruction, thus, specifically instructed the jury to find the special circumstance in this case if it concluded either that Pulido, with intent to kill, aided and abetted the murder or that Pulido, with reckless indifference to life, aided and abetted the robbery. As should be clear, those standards allow the jury to find the special circumstance based solely on the theory that Pulido aided and abetted the robbery with reckless indifference to human life.
Once we consider that instruction in the context of the entire record, it becomes apparent that the jury might have found the special circumstance — might have found aiding and abetting with reckless indifference — based solely on Pulido’s proffered factual theory that he knowingly participated in the crime only after the theft and the murder were completed. There are two aspects of the cumulative jury instructions that make this possibility viable. First, the court specifically instructed the jury that aider and abettor liability for robbery “continues so long as the stolen property is being carried away to a place of temporary safety.” CALJIC 9.40.1 (1991 New). The court did not instruct the jury that this definition of aider and abettor liability attached only to the robbery count — that the relevant assistance for felony-murder purposes must begin before or during the robbery and murder. The jury therefore might reasonably but erroneously have believed that Pulido aided and abetted the robbery within the meaning of the robbery-murder instruc*681tion, CALJIC 8.80.1, even if, as Pulido maintained, his participation began after the robbery and the shooting. Because aiding and abetting the robbery (with reckless indifference to life) sufficed to find the special circumstance under the non-triggerman guidelines and because after-the-fact participation sufficed to find aider and abettor liability under the robbery instructions, the jury might erroneously have found the special circumstance to be true even under Pulido’s proffered facts.
The second troubling aspect of the cumulative instructions is that they nowhere define “reckless indifference to human life.” Although there was no evidence introduced at trial that would have supported a legal conclusion of reckless indifference, it seems extremely unlikely that— without instruction — the jury took a limited, legalistic view of the “reckless indifference” language. Under its ordinary meaning, “reckless indifference to human life” might very well encompass the act of assisting someone to flee the scene of a murder. In other words, a lay jury might have concluded that Pulido was recklessly indifferent to the victim’s life even if it concluded that the victim was already dead before Pulido participated in the robbery.
Taking the cumulative instructions, the jury might have concluded that Pulido’s after-the-fact assistance met both qualifications for non-triggerman liability under the special circumstance. Pulido was an aider and abettor within the meaning of CALJIC 9.40.1, and he acted with reckless indifference to the victim’s life within the lay meaning of those words.
Given these two characteristics of the cumulative jury instructions — that the jury was specifically instructed to find Pulido guilty of aiding and abetting even if he became involved only after the robbery was completed and that the jury was never given a legal definition of “reckless indifference to human life” — it seems at least possible that the jury erroneously found the special circumstance to apply to post-murder involvement, even if it relied wholly on the supposedly curative CALJIC 8.80.1. The specific standards for non-triggerman liability in CALJIC 8.80.1, which could encompass post-murder involvement, severely undermine the government’s and the California Supreme Court’s single-minded reliance on the instruction’s legal formulation of the special circumstance. Reading the entirety of CALJIC 8.80.1 and taking it together with all other relevant instructions, the jury was essentially instructed as follows: “You may find it to be true that the murder was committed while the defendant was aiding and abetting robbery if you find (1) that the defendant began assisting a robber after the theft was complete and (2) that he did so with a reckless indifference to human life.” When stated succinctly and then applied to this case, the combined instructions are, undoubtedly, internally inconsistent. In a case in which the theft and murder occurred simultaneously, it should not be possible to conclude that the murder was committed while the defendant was assisting a robbery if one concludes that the defendant did not start assisting the robbery until after the theft was committed. Nevertheless, the combined instructions nowhere precluded that illogical conclusion and impliedly allowed it. Without clarifying instructions from the court, it seems entirely possible that the jury simply accepted the internal inconsistency of their finding as a legal fiction, assuming that the legal formulation of the special circumstance was less important to its conclusion than the extensive standards it was given for determining whether the special circumstance applied to a non-triggerman like Pulido. At the very least, the cumulative instructions give rise to reasonable *682doubt as to the jury’s thinking; it is not clear beyond all doubt that the jury relied on the minimal contemporaneity language contained in the legal formulation of the special circumstance.
For the same reasons, the California Supreme Court’s and the government’s reliance on the verdict form does not suffice to remove all reasonable doubts as to the jury’s thinking. While the verdict form specified that the special circumstance applies only if Pulido aided robbery “during the commission of the [murder],” the standards that the jury was given for determining whether it could issue that verdict allowed it to do so based on post-murder involvement. Given the combination of instructions that the jury received, it seems possible that at least some jurors simply accepted that the verdict form did not mean what it said. Again, in light of the various instructions that allowed after-the-fact involvement to count (including not only the allegedly curative CALJIC 8.80.1 but also the admittedly defective CALJIC 8.81.17), the jury may have assumed that the contemporaneity language on the verdict form was simply a legal fiction.
In conclusion, neither the supposedly curative instruction nor the verdict form removes all reasonable doubts as to the jury’s thought process. In fact, when read together with all jury instructions, the supposedly curative instruction compounds rather than alleviating the possibility that the jury found a special circumstance of robbery-murder even while accepting Puli-do’s theory of after-the-fact assistance.
B
This theoretical possibility becomes more realistic in light of the questions that the jury submitted to the judge during its deliberations. The Supreme Court has held that questions from the jury can demonstrate that the trial court has failed adequately to instruct the jury. See, e.g., Shafer v. South Carolina, 532 U.S. 36, 52-53, 121 S.Ct. 1263, 149 L.Ed.2d 178 (2001); Simmons v. South Carolina, 512 U.S. 154, 178, 114 S.Ct. 2187, 129 L.Ed.2d 133 (1994); Bollenbach v. United States, 326 U.S. 607, 612, 66 S.Ct. 402, 90 L.Ed. 350 (1946). In this case, three jury questions reveal that the jurors were struggling with the requirements of aider and abettor liability, felony-murder, and the special circumstance robbery-murder. First, the jury asked for “instructions special circumstances for both count 1 and count 2” (emphasis original). This question demonstrates a certain degree of confusion given that the special circumstance does not attach to either count but, rather, intends to be a special finding of contemporaneity between the two counts. Second, the jury asked whether aider and abettor liability requires “knowledge [of the purpose of the perpetrator] prior to the commission of the crime or during the commission of the crime[.]” This question clearly demonstrates that the jurors were confused as to the timing requirements for aider and abettor liability, unsure whether Pulido could be found guilty on an aider and abettor theory if he became aware of the principal’s purpose only during or after commission of the crime.
Third and most importantly, the jurors asked whether they could find felony murder based solely on actual “facilitating] by aiding” or whether they must also find that Pulido intended for the robbery to occur. Like the previous question, this question demonstrates that at least some jurors believed that Pulido developed knowledge and intent only after the actual theft had occurred; otherwise, on the alternative facts presented, there could have been little doubt that Pulido intended to commit robbery. Critically, that question also demonstrates the jurors’ belief that *683Pulido’s intent to commit robbery would be the determining factor in a felony-murder conviction, not Pulido’s intent to commit murder. That distinction raises a very real possibility that the jury found the special circumstance under the “reckless indifference” prong rather than the “intentional murder” prong of the standards for non-triggerman liability.
Importantly, the judge did not answer any of those questions directly. Instead, he merely referred the jurors back to their flawed and inconsistent instructions.
Based on these jury questions, it seems not only possible but probable that the jurors were confused as to the timing and intent requirements in the felony-murder and special circumstance instructions. Those questions, at the very least, raise a reasonable doubt as to whether the jurors understood and relied on the contemporaneity requirement that was formally included in the verdict form and in the legal formulation of the special circumstance.
C
In conclusion, the government’s theories fail to establish beyond a reasonable doubt that the error in this case was harmless. The cumulative instructions allowed a conviction for felony-murder and a finding of special circumstance robbery-murder even under Pulido’s asserted factual theory that he became a knowing participant only after the robbery and shooting were completed. Furthermore, the jurors’ questions raise a real possibility that at least one juror would have voted to acquit Puli-do of felony-murder and special circumstance charges if the jury had been properly instructed that contemporaneity was a necessary finding. Therefore, I conclude that Pulido would be entitled to relief even under a harmless error standard.
IV
The special concurrence urges us to reexamine Lara. I believe Lara was correctly decided. Lara was founded on Sandstrom v. Montana, 442 U.S. 510, 526, 99 S.Ct. 2450, 61 L.Ed.2d 39 (1979), which held that “when a case is submitted to the jury on alternative theories the unconstitutionality of any of the theories requires that the conviction be set aside.” Id. (citing Stromberg v. California, 283 U.S. 359, 368, 51 S.Ct. 532, 75 L.Ed. 1117 (1931)). Sandstrom was not an anomalous decision. In Zant v. Stephens, 462 U.S. 862, 881, 103 S.Ct. 2733, 77 L.Ed.2d 235 (1983), the Supreme Court held that “a general verdict must be set aside if the jury was instructed that it could rely on any of two or more independent grounds, and one of those grounds is insufficient, because the verdict may have rested exclusively on the insufficient ground.” Similarly, in Yates v. United States, 354 U.S. 298, 312, 77 S.Ct. 1064, 1 L.Ed.2d 1356 (1957), overruled on other grounds by Burks v. United States, 437 U.S. 1, 98 S.Ct. 2141, 57 L.Ed.2d 1 (1978), the Court stated that:
[T]he proper rule to be applied is that which requires a verdict to be set aside in cases where the verdict is supportable on one ground, but not on another, and it is impossible to tell which ground the jury selected.
Id.
In sum, the conclusion in Lara that the Supreme Court meant what it said in Sandstrom is firmly founded in Supreme Court jurisprudence. I would not revisit the holding in Lara.
Y
Because I agree that Lara controls this case, because I believe that Lara was properly decided, and because Pulido would be entitled to relief even under a *684harmless error standard, I concur in the majority opinion.