Order; Dissent by Judge KLEINFELD.

ORDER

The panel has voted to deny the petitions for rehearing and petitions for rehearing en banc.

The full court was advised of the petitions for rehearing en banc. A judge requested a vote on whether to rehear the matter en banc. The matter failed to receive a majority of the votes of the nonre-cused active judges in favor of en banc reconsideration. Fed. R. App. P. 35.

The petitions for rehearing and rehearing en banc are denied.

with whom Circuit Judge BEA joins, dissenting from denial of rehearing en banc:

I dissent from denial of rehearing en banc.

Our panel decision in Irons v. Carey1 makes two significant legal errors. The legal errors are (1) the application of circuit law instead of Supreme Court law to habeas corpus petitions, and (2) the application of “good time” law to parole. These legal errors are likely to lead to practical harm by requiring parole boards to release prisoners on parole, no matter what their crime, if they behave themselves in prison.

The opinion in Irons is particularly mischievous because of an oddity in the way our court treats the traditional distinction between holdings and dicta. In our court, two judges on a panel make law for 20% of Americans as a “supervisory” matter, even where the case does not require the matter to be decided.2 Our en banc decision in Barapind v. Enomoto says that an issue presented for review and addressed and decided in a panel opinion “became law of the circuit, regardless of whether it was in some technical sense ‘necessary’ to our disposition of the case.”3 We characterized this rejection of the traditional distinction between holdings and dicta as an exercise of our “supervisory function ... by instructing three judge panels and district courts about how to determine what law is binding upon them.”4 The traditional view, which we seem to have rejected in Barapind, is that since we are empowered only to decide cases, not to legislate, only those principles necessary to the decision are binding law of the circuit. Thus, under Barapind, the statements of law made in Irons that were unnecessary to the decision may nevertheless be construed by subsequent three judge panels, district courts and the state courts as binding law of the circuit.

Irons v. Carey was a petition for a writ of habeas corpus, challenging a state denial of parole, for a murderer who had served sixteen years of his seventeen-years-to-life sentence for murder. Irons fired twelve shots into his housemate, stabbed him twice in the back, hid his body in the room for ten days, then dumped it in the ocean. In prison, though, “his conduct has been exemplary.”5

Irons says that deprivation of parole for this murderer would violate due process of *953law under “clearly established” Supreme Court law, unless it was supported by “some evidence in the record.” Irons cites previous circuit law and the Supreme Court decision in Superintendent v. Hill for this proposition.6 Irons nevertheless allows denial of parole to this prisoner, during the sixteenth year of his 17-to-life sentence, noting that the parole board thought he needed therapy, that his crime was “carried out in an especially cruel and callous manner,” and he was himself equivocal about his rehabilitation. When asked whether this especially cruel and callous crime would happen again, Irons said “I don’t think so” instead of a definite “no.”

Irons did not stop there. We went on to “note” that in all cases, including Irons, where we upheld denial of parole on the basis of the crime, the inmate had not yet served even the minimum period of his sentence. After that (one more year for Irons), “indefinite detention based solely on the inmate’s commitment offense, regardless of the extent of his rehabilitation, will at some point violate due process.”7 Irons seems to say that no matter how monstrous the crime, a prisoner against whom there is no evidence adverse to parole and subsequent to the crime, has a constitutional right to be paroled after serving the bottom end of his sentence.

The first thing wrong with Irons is that it arrogates to our court more authority than we have. The Supreme Court in Carey v. Musladin8 reversed us in another habeas case, where we applied circuit law (as Irons does) without a Supreme Court decision to the same effect. The federal habeas statute says that we cannot overturn a state decision as contrary to federal constitutional law, unless the federal law is “as determined by the Supreme Court of the United States.”9 That phrase carries a negative pregnant, an exclusion of federal law as determined by a circuit court but not the Supreme Court. The Court held in Musladin that the statutory phrase “refers to the holdings ... of [the Supreme] Court’s decisions.”10 Where, as here, the Supreme Court “has not broken sufficient legal ground to establish an asked-for constitutional principle,” our court “cannot [itself] establish such a principle with clarity sufficient to satisfy theAEDPAbar.”11

The second thing wrong with Irons is that it inaccurately claims that the Supreme Court has “clearly established that a parole board’s decision” must be supported by “some evidence” in Superintendent v. Hill.12 Irons errs because Hill requires “some evidence” to deprive a prisoner of “good time,” not parole.

“Good time” and parole are different kinds of decisions. Good time is retrospective, looking to whether the prisoner has misbehaved. Parole is prospective, looking to whether the prisoner, if no longer confined in a total institution, is likely to misbehave in the future. The first kind of decision must be based on evidence, the *954second requires discretion, judgment, and experience. Evidence of behavior in prison may be insufficient for predictions of behavior outside of prison.

“Good time” is a prison discipline device. Prisons need some way to get prisoners, who are after all felons, to leave the recreation yard or mess hall, go into their cells, quit fighting and keeping contraband in their cells, and so forth. Otherwise the prisoners could taunt the guards, “What are you going to do about it, send me to jail?” Many prison rule violations are not crimes. Prisons use reductions in “good time” credits, among other sanctions, to induce compliance. For example, federal law generally credits prisoners with 54 days per year, starting with the second year of imprisonment, for compliance with prison disciplinary and educational regulations.13 The California scheme allows for reductions of up to one-third of the sentence for good behavior and participation in assigned programs.14

Parole, on the other hand, is a discretionary and prospective decision about whether it is prudent and appropriate to let the prisoner out early.15 The Supreme Court has characterized parole decisions as “necessarily subjective in part and predictive in part.” Release on parole is “essentially an experienced prediction based on a host of variables.”16 Requiring “some evidence” for a subjective prediction is dangerous. The prisoners who are the subjects of parole decisions have already proved to be dangerous. Some people behave better in the total institution of prison than outside. It is striking that the panel felt moved to expound on the likely future entitlement of parole to Irons, an especially vicious murderer.

Wolff v. McDonnell, the Supreme Court’s seminal “good time” revocation ease, holds that due process requires, inter alia, “that there must be a ‘written statement by the factfinders as to the evidence relied and reasons’ for the disciplinary action.” 17 Wolff explains that “the provision for a written record helps to insure that administrators, faced with possible scrutiny by ... perhaps even the courts, where fundamental constitutional rights may have been abridged, will act fairly.”18 Relying on Wolffs procedural requirements for revoking good time, the Supreme Court, in Superintendent v. Hill, conclud*955ed that, to take away a prisoner’s good time, due process requires there to be “some evidence” of a disciplinary infraction.19

No Supreme Court case holds that the “some evidence” rule applies to parole determinations. Two Supreme Court cases suggest the opposite.20 Despite these two Supreme Court decisions, Irons inappropriately applies the Hill “good time” rule to what the Supreme Court has said are the quite different decisions made by parole boards.

The Court in Greenholtz v. Inmates of the Nebraska Penal and Correctional Complex holds that “nothing in the due process clause ... requires the Parole Board to specify the particular ‘evidence’ in the inmate’s file or at his interview on which it rests the discretionary determination that an inmate is not ready for conditional release.”21 Greenholtz explains that the parole decision “is necessarily subjective” and demands “experimentation involving analysis of psychological factors combined with fact evaluation guided by practical experience.”22 Thus, “to require the parole authority to provide a summary of the evidence would [incorrectly] tend to convert the process into an adversary proceeding and to equate the parole-release determination with a guilt determination.” 23

Greenholtz specifically notes that the “procedures designed to elicit specific facts, such as those required in ... Wolff [the good time case] are not necessarily appropriate....”24 The panel ought to have noticed that the Supreme Court expressly distinguished the procedures that are appropriate in parole board decisions from what procedures are appropriate in good time decisions. Under Greenholtz, all that is required for due process is “an opportunity to be heard” and “inform[ing] the inmate in what respects he falls short of qualifying for parole ... as a guide to the inmate for his future behavior.”25

In Wilkinson v. Austin, the Supreme Court confirmed the distinction between the requirements of due process for the revocation of good time credits and the requirements of due process for parole determinations.26 Wilkinson rejected a challenge to the administrative decision to transfer a prisoner to a maximum security facility.27 Wilkinson held that the prisoner transfer decision is more like a parole decision than a revocation of good time credits, and thus requires only the “nonadversary procedures set forth in Greenholtz,” not the “more formal adversary-type procedures” set forth in Wolff.28 Again, the Supreme Court expressly distinguished between procedures for good time revocation and procedures for parole.

Irons says that after a certain time, the crime committed by the prisoner does not *956amount to “some evidence.”29 According to the panel, the savagery of the murder committed by Irons only counts as “some evidence” because he had not yet satisfied the “minimum” of his 17-to-life sentence.30

Implicit in Irons is a psychological theory, untested, not especially likely to be true, and dangerous. If we were to predict Irons’ behavior based on how he behaved when he was last free, we would regard him as dangerous, but if we made our prediction based only on how he behaved in prison, we would regard him as much less dangerous. The theory implicit in Irons is that past conduct in freedom does not, after a while, predict future conduct in freedom, and if long enough ago, is not even “some evidence” of how a prisoner will conduct himself if free. That could be true. It could also be false. It may be that especially savage conduct in freedom better predicts future dangerousness in' freedom, than behavior in prison. The Irons psychological theory is not Supreme Court law and there is no particular reason to assume its truth. We have no justification for forcing this unproved theory upon the states in the guise of constitutional law.

In addition to the predictive, empirical concern with recidivism, states may have a moral concern with parole of prisoners who have committed especially savage crimes. Even if there is not “some evidence” of likely recidivism, and even if parole boards are satisfied that there is no likelihood of recidivism, states may justifiably deny parole. States are free to take the view that vindication of principles of right and wrong, and a decent respect for the victims of crime, require denial of parole to especially vicious criminals. States are entitled to deny parole and require prisoners to serve their full sentences less “good time,” even without “some evidence” beyond the crimes for which the sentences were imposed.