Gary Stewart was convicted by a jury in Massachusetts Superior Court of second degree murder. After the Supreme Judicial Court upheld the conviction, Stewart filed a habeas corpus petition and the district court ultimately granted the writ, holding that the evidence at the state trial was inadequate to permit a reasonable jury to convict. That ten very able judges before us have disagreed so sharply over the evidence is a measure of the difficulty of this case.
i.
Stewart was indicted by a Middlesex grand jury on August 1,1986, and charged with the first degree murder of Robert Perry. He was tried by a jury in a trial lasting several days beginning on March 8, 1988. From the outset, the Commonwealth’s theory was that the actual murder of Perry had been committed by John Good who was tried separately and convicted of first degree murder. See Commonwealth v. Good, 409 Mass. 612, 568 N.E.2d 1127 (1991). The evidence in Stewart’s trial, taken most favorably to the Commonwealth, showed the following.
At about 1 a.m., on July 27, 1986 — the day of Perry’s murder and about 11 hours before that event — Stewart, Good and a third man were seen together entering a bar on Cambridge Street in Cambridge, Massachusetts. The bar was located about four blocks east of the Cambridge City Hospital and about five blocks west of the Harrington School, both of which are also on Cambridge Street.
At about 7:50 a.m. that same morning, Stewart was seen driving west on Cambridge Street near the Harrington School. The car was a yellow or off-white Pontiac bearing Massachusetts license plate 104-MND. Good was in the front passenger seat and a third man was in the rear seat. As the ear passed a cat sleeping on a ear hood on the opposite side of the street, Stewart made a U-tum and drove back east on Cambridge Street. Good then pointed a black handgun out of the passenger side window and shot the cat twice, killing it. The car then drove away with the passengers laughing.
At about 12 noon on the same day, Stewart was sitting in the same car, which was parked on Maple Avenue in Cambridge. Maple Avenue is a one-way street that runs south from Cambridge Street commencing just east of the Cambridge City Hospital. The car was parked several car lengths south of the intersection with Cambridge Street. Fifteen minutes earlier, at about 11:45 a.m., Perry had told his former wife that he would be walking west along Cambridge Street from Inman Square, which is on Cambridge Street several blocks east of the hospital.
*612At about noon — this is inference but amply supported — Good shot Perry three times with a .38 caliber handgun on Cambridge Street just west of the intersection with Maple; one of the shots pierced Perry’s heart, another his head, and he was killed. Immediately afterward, Good, carrying the handgun, ran diagonally down Maple Avenue to the Pontiac parked on the east side of the street. Stewart then drove the car south down Maple Avenue, accelerating to approximately 45 miles per hour.1 A few blocks later, at the intersection of Harvard and Dana Streets, Stewart ran a stop sign and flashing red light and crashed into another car.
Good exited from the Pontiac, told Stewart, “I’m getting out of here,” and quickly walked away. Stewart got out of the car and leaned against the door. When the police came, Stewart said to one of the policemen, “It’s an ordinary accident. I was trying to cross Harvard Street and continue on Dana when this car to the left of me came along and struck me.” To another officer Stewart said: “What’s the big deal? This is only an accident. I’m the only one. I’m the only one in the car.” In the Pontiac the police found a brown paper bag on the floor in front of the passenger seat containing several live rounds of .38 ammunition of types similar to that used to kill Perry.
This is the heart of the evidence offered at Stewart’s trial. There was one eye witness to the event at the bar; another eye witness who saw the incident involving the cat; a third eye witness who saw Good running to the car and Stewart driving away after the shots; and a fourth who saw Good looking out the car window as the car drove rapidly down Maple Avenue. Several police officers and several passengers from the other car in the crash testified to events at the crash scene. There was no evidence of motive. Stewart offered relatively little evidence at trial and did not testify.
The trial judge charged the jury that under Massachusetts law Stewart could be convicted of murder on a “joint venture” theory if he aided in the crime and shared an intent to murder. For first degree murder, said the trial judge, the intent required included both premeditation and an intent to kill or do serious injury; for second degree murder, premeditation was unnecessary. The jury deliberated for three hours and returned a verdict of second degree murder. After-wards, the trial judge indicated that a first degree murder conviction might have been expected.
On appeal, Stewart argued that the evidence was inadequate for a reasonable jury to find that he had known in advance of Good’s intent to commit murder. The Massachusetts Appeals Court, by a two-to-one vote, agreed and ordered the entry of judgment in Stewart’s favor. Commonwealth v. Stewart, 30 Mass.App.Ct. 569, 571 N.E.2d 43 (1991). On further review, the Supreme Judicial Court reinstated the guilty verdict, all five justices agreeing that the evidence was adequate to allow the jury to conclude “that Good planned in advance to kill Perry and that [Stewart] kn[ew] of this plan and intended to aid Good in committing the murder.” Commonwealth v. Stewart, 411 Mass. 345, 582 N.E.2d 514, 518 n. 3 (1991).
From all the Commonwealth’s evidence, the jury reasonably could have inferred that [Stewart] and Good had been driving up and down Cambridge Street looking for Perry so Good could shoot him; ... and that [Stewart] deliberately parked on a side street near Cambridge Street, and around the comer from where Perry was soon to pass, in order to facilitate a speedy escape after the murder. [and] that [Stewart’s] intent to assist was further shown by evidence that Good ran directly to [Stewart’s] automobile after the shooting, that without any hesitation or conversation, [Stewart] immediately pulled away and sped up the street at a high rate of speed; and that [Stewart] subsequently covered Good’s escape by lying to the police.
582 N.E.2d at 518.
Stewart then began the present habeas proceeding in federal district court. 28 *613U.S.C. § 2254. There, he argued that a constitutional violation had occurred because upon the record evidence adduced at the state trial no rational trier of fact could have found proof beyond a reasonable doubt. Jackson v. Virginia, 443 U.S. 307, 324, 99 S.Ct. 2781, 2791-92, 61 L.Ed.2d 560 (1979). The district court reviewed the state trial record, took the same “view” of the scene at Cambridge Street that the trial jury had taken, considered the briefs and heard oral argument. On June 6, 1994, the district court granted the writ and ordered Stewart’s release after a brief period to permit the Commonwealth to seek a stay.
The district court said that — contrary to the inference drawn by the Supreme Judicial Court — there was insufficient evidence that Stewart had been parked on Maple Avenue with the knowledge that Perry was soon to pass nearby or to facilitate a speedy escape after Good killed him. The district court also said that the testimony did not show that the car moved instantly upon Good’s return or that Stewart and Good did not converse at least briefly. At most, said the district judge, Stewart might have been convicted as an accessory to murder after the fact, a lesser crime with which he was never charged and could not now be because the statute of limitations had run. The Commonwealth then appealed and we stayed the judgment pending review.
II.
The district court and the parties do not greatly differ as to the applicable legal standard. Under Jackson, the question that the habeas court must answer is “whether, after viewing the evidence in the light most favorable to the prosecution, any rational trier of fact could have found evidence sufficient to prove the essential elements of the crime beyond a reasonable doubt.” 443 U.S. at 319, 99 S.Ct. at 2789. The Commonwealth stresses that the habeas judge can only consider the rationality of the verdict and is not to make his or her own evaluation of guilt or innocence. But the Commonwealth does not dispute Stewart’s claim that the habeas court is to apply the quoted standard independently and without otherwise deferring to the state courts.
We are not so sure that this standard does reflect the current thinking of the Supreme Court. On the issue of the proper constitutional standard, Jackson was a flve-to-three decision; every member of the Jackson majority is gone from the Court; and the concurring trio — Justice Stevens joined by Chief Justice Burger and Justice Rehnquist — argued for a standard that asked whether there was some evidence to support the disputed finding. Further, since both opinions in Jackson held that the evidence was adequate to convict, the choice between the two calibrations of the standard did not matter in that case.
At first blush it may appear startling that federal judges should effectively substitute themselves for state judges in deciding whether a judgment of acquittal is warranted in a state criminal case for lack of adequate evidence. No one suggests that federal judges should review ordinary state court rulings on evidence or procedure under a de novo standard. And if one looks to core due process principles, see Rochin v. California, 342 U.S. 165, 172, 72 S.Ct. 205, 209-10, 96 L.Ed. 183 (1952), a conviction on no evidence could well shock the average conscience; but some might find nothing shocking about a simple disagreement between federal and state judges on how far to stretch an inference.
Neither of the two sequels to Jackson is illuminating. Wright v. West, — U.S. -, -, 112 S.Ct. 2482,120 L.Ed.2d 225 (1992), involved a fractured Court with no majority opinion; and in Herrera v. Collins, — U.S. -, 113 S.Ct. 853, 122 L.Ed.2d 203 (1993), the majority opinion by Chief Justice Rehnquist capsulized Jackson solely in order to distinguish it. On a more general plane, the Court’s narrowing of habeas in the last decade is widely acknowledged. At the same time, the Court has not formally retreated its prior position that in habeas proceedings, federal courts — when they do reach the merits — normally make independent decisions on constitutional issues. Miller v. Fenton, 474 U.S. 104, 112, 106 S.Ct. 445, 450-51, 88 L.Ed.2d 405 (1985).
*614We are not certain how the Supreme Court will resolve the matter. The majority’s statement in Jackson represents the pole most favorable to defendants; at the other end of the spectrum lies the possibility that the Court might adopt the “some evidence” formulation of the Jackson concurrence or— pretty much the same thing under a different label — a notion of limited deference to state tribunals. Possibly, even the Supreme Court does not yet know how it would handle a case like ours that is so close that the precise formulation could dictate the result.
While adopting a “some evidence” standard would greatly simplify this case, we agree with the district court that the more stringent literal language of Jackson should control our inquiry. There is nothing ambiguous about the Jackson standard, however hard it may be to apply it in a close case; it was the stated position of a majority of justices; and the standard has never been overruled. It is a tricky matter for lower federal courts to anticipate the Supreme Court. In this instance — especially because the Jackson rule is directly concerned with innocence, see 443 U.S. at 323, 99 S.Ct. at 2791 — we do not think that we would be justified in diluting Jackson.
III.
Accordingly, we proceed to consider whether on the record made in the trial court “any rational trier of fact” could have found Stewart guilty of murder beyond a reasonable doubt. Jackson, 443 U.S. at 319, 99 S.Ct. at 2789. This is an inquiry we make de novo on a cold record without any special deference to either the state’s highest court, see Jackson, 443 U.S. at 318-25, 99 S.Ct. at 2788-92, or the federal district court, see Scarpa v. Dubois, 38 F.3d 1, 9 n. 5 (1st Cir.), cert. denied, — U.S. -, 115 S.Ct. 940, 130 L.Ed.2d 885 (1994) beyond the persuasive power of their (conflicting) interpretations of the record.
In many criminal cases that are close on the facts, the closeness is concealed because an eye witness testifies to the defendant’s guilt. Eye witnesses, of course, can make mistakes; but the newspaper reader or the appellate court reading the transcript after conviction sidesteps the doubt because the factfinder has credited the witness, and that is the end of the matter in all but the most extreme cases. Only in circumstantial-evidence cases like this one do we face head-on the disturbing truth that guilty verdicts rest on judgments about probabilities and those judgments are usually intuitive rather than scientific.
The essential facts of this case — those that the jury was unquestionably entitled to find — are rather simple: the prior association earlier that day of Good and Stewart; Stewart’s knowledge that Good was armed and vicious; the parking of the car on a side street with Stewart at the wheel; the murder of Perry by Good around the corner on Cambridge Street; Good’s hasty return to the car after the shots; the high-speed getaway and the subsequent crash of the car; and Stewart’s lies to the police tending (unsuccessfully) to shield Good from capture.
There is also no doubt about what inference of fact the jury had to draw in order to convict Stewart of murder under a joint venture theory. Although Massachusetts’ label is uncommon, its joint venture theory is essentially an aiding and abetting concept. Stewart clearly aided Good’s escape after the murder. The difficult factual question is whether Stewart also knew that Good was planning to commit a murder or, as would also be adequate under Massachusetts law, knew that Good was planning to do grievous bodily harm to his victim. Commonwealth v. Moore, 408 Mass. 117, 556 N.E.2d 392 (1990); Commonwealth v. Grey, 399 Mass. 469, 505 N.E.2d 171 (1987).
Why is it that nine judges (including the majority on this panel) think that the stated facts permit a clear and compelling inference of Stewart’s guilt and four others (including our dissenting colleague) think it plain that an acquittal should have been ordered? After all, the fact that no one overheard Stewart and Good planning a murder is not dis-positive; agreements are frequently inferred from circumstances. United States v. Moran, 984 F.2d 1299, 1300 (1st Cir.1993). The difference between the two viewpoints expressed by the thirteen judges, we think, lies *615primarily in the different probabilities that each side implicitly assigns to the 'possible alternative versions of what happened.
The Supreme Judicial Court thought it reasonably clear that Good and Stewart must have been looking for Perry so that Good could murder him; saw Perry proceeding east on Cambridge Street; and then parked the car on a side street to permit a quick getaway after Good accosted Perry. The district court, by contrast, thought this to be conjecture and said that there were other plausible explanations that did not involve advance knowledge by Stewart that Good meant to kill or assault Perry. Good lived a few blocks away from Maple Avenue; the district court conjectured that Stewart might have been awaiting Good’s return from his house when Good encountered Perry on the way back.
We agree with the district court that there is very little to show that Good and Stewart were searching for Perry; but that is not a necessary component in a scenario leading to Stewart’s guilt. It would be adequate if it could be inferred beyond a reasonable doubt either that Good and Stewart were searching for Perry or that they happened to see Perry while cruising on Cambridge Street and Good then proposed to murder Perry. Either possibility is consistent with the evidence, and both involve prior knowledge by Stewart of Good’s intent. Both possibilities are also quite plausible.
By contrast, we think that it is implausible to suppose that Good was visiting or purporting to visit his home located near but not on Maple Avenue. If this were so, as the district court thought quite possible, there is no apparent reason why Stewart’s car would be parked on Maple Avenue instead of in front of Good’s home. This brings us to Stewart’s brief which, in its only conjecture about other explanations, asserts:
[T]he evidence provides no clue whether Mr. Good left Mr. Stewart’s car with the innocent intention of buying a newspaper, visiting a sick friend at the hospital, or even, — taking into account Mr. Good’s bad character — robbing the hospital gift shop.
Assuming any of these purposes, it is at least possible that Good might have rushed back to the car after murdering Perry and spurred the unwitting Stewart into driving quickly away.
But why would anyone wanting a newspaper wait at the wheel down the block on a side street where, so far as the evidence goes, no newspaper vending machine is known to be located? If Good were visiting a sick friend, why would there not have been evidence at trial that he had such a friend in the hospital — a fact that could almost certainly be proved without Stewart’s testimony? As for the robbery, one would not normally think that a hospital gift shop would make an inviting target — given the limited proceeds and the likely presence of hospital guards — compared to any neighborhood convenience store.
The point is not that the explanations proffered in Stewart’s brief are impossible but none of them seems at all likely. And the two innocent ones — or any like them — are also not easily squared with the high-speed getaway or with the cover-up lies told by Stewart after the crash. The record is, as the district court said, unclear as to whether Stewart pulled instantly out of the space or had to maneuver; but the evidence does show that Stewart drove away very fast, and the later accident reinforces the point. Similarly, Stewart’s lies to the police — especially the spontaneous denial that there was anyone else in the ear — do not prove that he was privy to the crime in advance but reinforce the impression.
Of course, where there is no eye witness one can imagine innocent explanations for almost anything — here, for example, that Good, having left to buy a newspaper, disclosed the murder as he reentered the car and Stewart instantly decided to protect his friend (by fleeing at high speed, and then lying to the police) although at grave risk to himself. But this could reasonably seem farfetched to a jury; and each new gap-closing assumption — e.g., the supposed conversation in the ear — adds a new strain to the story.
Guilt beyond a reasonable doubt cannot be premised on pure conjecture. But a conjecture consistent with the evidence becomes less and less a conjecture, and moves *616gradually toward proof, as alternative innocent explanations are discarded or made less likely. Here, there is nothing at all unlikely about the hypothesis that Good and Stewart were either stalking Perry or chanced upon him and decided that Good would kill him. No other explanation that is at all likely has been suggested to us. “Beyond a reasonable doubt” does not require the exclusion of every other hypothesis; it is enough that all “reasonable” doubts be excluded. United States v. Oreto, 37 F.3d 739, 753 (1st Cir. 1994); United States v. Whiting, 28 F.3d 1296, 1303-04 (1st Cir.1994).
At this point, Jackson’s own objective standard turns against the defendant. It makes no difference whether we or the district judge would as jurors have voted to acquit Stewart or whether we ourselves think that there is some reasonable doubt. The question posed by Jackson is whether “any” rational jury could on the evidence presented think Stewart’s knowing participation so likely as to exclude all reasonable doubts. And rational people can have quite different views about the likelihood that a quick getaway implies prior planning or that one otherwise innocent would he to the police to protect a murderous companion.
The problem is that no scientific data exists on these probabilities (and it might not be admissible if it existed). Each judge and juror brings to the courthouse a bundle of unarticulated assumptions about how the world works and about the respective likelihoods of different concatenations of events. That does not mean that we, or the district court, or the state tribunals can escape the task of second-guessing the jury to the limited extent necessary to direct verdicts, apply Jackson, or consider such issues on appeal. But variations in human experience suggest that one should expect a considerable range of reasonable estimates about what is likely or unlikely.
We do not have the same confidence as either the district court or the Massachusetts Supreme Judicial Court in assigning the probabilities in this case. All that we can say, with the advantage of having both sets of views before us, is that the case against Stewart was not overwhelming and involved some uncertainties that cannot be erased, but it was also not so weak as to render the jury verdict irrational. A rational jury might well have acquitted without violating its oath; but, drawing all reasonable inferences in favor of the prosecution, a rational jury could also convict.
We have considered the case thus far in terms of the scenarios suggested by the parties and the four other courts that have had the Stewart case before them. Although other Massachusetts joint venture cases have been cited to us by the parties, each is distinguishable; and none would relieve us of the duty under Jackson to make our own independent assessment as to what a reasonable jury could infer on the facts of this case. But we think that one additional possibility is so patent that it calls out for comment even though neither side has thought it useful to draw our attention in this direction.
The association of Stewart and Good, the prior shooting, the strategically parked car, the reasonably quick departure and high speed escape, and the lies to the police cumulatively suggest that Stewart was the knowing participant with Good in a criminal joint venture. A rational jury could, given the absence of other likely explanations, find the joint venture alternative so likely as to be beyond reasonable doubt. In other words, we think that the jury, although not obliged to do so, was entitled to reject any notion (as posited by the district court) that Stewart was at most shown to be an accessory after the fact. But just what was the joint venture?
The only theory argued at trial by the prosecution, and the only one covered by instructions to the jury, was that the venture was one to murder Perry or to do him grievous bodily harm. We think that another possibility suggests itself as consistent with the evidence: that Good left the car as part of a plan by Good and Stewart to rob someone on Cambridge Street, whether Perry in particular or whoever might come along, and that Perry was the victim of a robbery gone awry. This scenario does not involve a shared intent to commit murder or do bodily harm.
*617It may be enough to say that it is a substantially less likely scenario than a planned murder, if only because of the wounds inflicted on Perry. Two of the three shots- — one to the head and one to the heart — suggest an intent to kill; and this is especially so of the former since the coroner said that the shot entered at the back of the skull. Of course, it is still possible that Perry resisted a robbery attempt and was then killed, but the shot from behind makes this less likely. We also have no reason to think that Perry (en route to meet his former wife for lunch) was armed, and there is no evidence — -which the prosecution would have had good reason to offer if it existed — that Perry’s wallet or property was found in the ear.
Ironically, if the joint venture were one to commit robbery, it appears that Stewart would still have been guilty of murder under the felony murder rule followed in Massachusetts and in many other states. See, e.g., Commonwealth v. Claudio, 418 Mass. 103, 634 N.E.2d 902, 906-07 (1994). Of course, that would not be a basis for sustaining Stewart’s conviction on direct appeal since he was never charged with felony murder. But we doubt whether it would be part of the office of habeas corpus to release a prisoner whose “defense” in seeking the writ was that he had committed murder but only on a theory not properly presented to the jury.
IV.
Judges who have presided over criminal jury trials are wont to say that the juries usually reach a correct, or at least defensible, result on the evidence presented to them. That is no reason to diminish further safeguards, such as the directed verdict, against the tragic risk that an innocent person may be sent to prison. It is a reason to hesitate long and hard before concluding that a jury’s judgment is irrational.
The judgment of the district court is reversed.
Dissent follows.
. See Mass.Gen.L. ch. 90, § 17 (establishing a speed limit of 30 miles per hour for streets like Maple).