dissenting:
Randy Moore and others beat Kenneth Rogers until he bled, stripped him, bound him in duct tape, placed him in the trunk of a car, drove him to a remote location, and forced him to march up a hill at gunpoint. While marching Rogers through the woods, Moore shot Rogers— accidentally, he said — through the temple. Moore confessed to his older brother, Raymond, and his step-brother’s girlfriend, Debbie Ziegler, what he had done. He then talked to police, corroborating the evidence the police had already obtained. Before Moore could be indicted, he negotiated a plea bargain under which Moore obtained the lowest sentence available under Oregon law for felony murder. Moore’s counsel, an experienced defense *1172attorney, recommended that he accept the offer. Counsel explained that he did not think he had grounds to suppress Moore’s formal confession and, even if he could have suppressed it, Raymond and Ziegler could “repeat[Moore’s informal] confession in full detail.” Under the circumstances, counsel thought he had secured the best deal he could get for Moore. The Oregon courts and the district court agreed.
Not so, says the majority: Moore’s attorney offered constitutionally deficient advice because he advised Moore to accept the plea offer before he moved to suppress Moore’s confession to the police. The majority reasons that Moore would have prevailed on a motion to suppress, and, knowing the state was without his confession, Moore would not have pled guilty to felony murder or would have held out for a better deal. The majority dismisses counsel’s explanation that there was no reason to go to trial in any event because the state had a second confession — Moore’s confession to Raymond and Ziegler — by ignoring the state and the district court’s findings and entering its own findings: The majority finds that “it is far from clear what those witnesses would have said,” and wonders “to what extent their testimony would have been persuasive to a jury.” Maj. Op. at 1148. Furthermore, the majority concludes, Raymond was “a hostile witness, [so] it is unlikely that [the state] would have been able to elicit much of the information it desired from him.” Id. at 1148-49 n. 24.
The majority not only entered its own findings of fact, it found its own law as well. In this AEDPA-governed case, the majority holds that the Oregon state court’s decision was “contrary” to the statement in Arizona v. Fulminante that “a confession is like no other evidence” and is “the most probative and damaging evidence that can be admitted against [a defendant].” 499 U.S. 279, 295, 111 S.Ct. 1246, 113 L.Ed.2d 302 (1991) (quotation marks and citation omitted). See Maj. Op. at 1130, 1131-32, 1136-37, 1140, 1145-47, 1146-48, 1149-50, 1152-54. The majority’s reliance on Fulminante is twice remarkable: First, Moore does not even cite Ful-minante, nor was it cited by the district court, the state court, or any other party. Second, the majority’s repeated insistence that “a confession is like no other evidence” is its own undoing. Counsel negotiated a plea because he knew what the majority cannot bring itself to admit: Moore had confessed to two other people before he confessed to the police, and their confessions were plainly admissible and independently damaging.
In the process of second-guessing counsel, the Oregon courts, and the district court, the majority clearly establishes a dramatic proposition: After Kimmelman v. Morrison, 477 U.S. 365, 106 S.Ct. 2574, 91 L.Ed.2d 305 (1986), Strickland v. Washington, 466 U.S. 668, 104 S.Ct. 2052, 80 L.Ed.2d 674 (1984), and Arizona v. Fulminante, when a motion to suppress a confession is potentially “meritorious,” counsel’s failure to file the motion constitutes deficient and prejudicial conduct because of the possibility that filing the motion “would have placed [Moore] in a far better position to negotiate a reasonable plea.” Maj. Op. at 1140. The majority’s principle applies regardless of how many witnesses a defendant has confessed to, how many co-defendants are available to testify against the defendant, or any other evidence available in the record. See Maj. Op. at 1144-52. Indeed, the majority concludes that even though “Moore had confessed to Raymond and Ziegler, who could have been called as witnesses,” the state court’s conclusion that Moore suffered no prejudice “is wholly lacking in merit” and was objectively unreasonable under Ful-minante. Maj. Op. at 1150-51. According to the majority, if counsel has any grounds for moving to suppress a confession, the *1173failure to move for suppression satisfies both prongs of Strickland. It is, absolutely, error and, absolutely, prejudicial.
I have referred deliberately to “the majority clearly establish[ing]” this proposition, because no court — not ours, and certainly not the U.S. Supreme Court — has previously done so. In fact, we might have thought the contrary proposition was clearly established in McMann v. Richardson: “In our view a defendant’s plea of guilty based on reasonably competent advice is an intelligent plea not open to attack on the ground that counsel may have misjudged the admissibility of the defendant’s confession.” 397 U.S. 759, 770, 90 S.Ct. 1441, 25 L.Ed.2d 763 (1970). It is not only evident that the majority has clearly established its proposition, but that its proposition has been clearly established by the majority. The majority’s efforts to do so contravene our statutory function under AEDPA. See 28 U.S.C. § 2254(d)(1).
For the reasons I explain below, I would affirm the judgment of the district court denying the writ. I respectfully dissent.
I. WHETHER COUNSEL’S CONDUCT WAS DEFICIENT
On AEDPA review, we may only issue a writ of habeas corpus when the state court unreasonably applies “clearly established Federal law, as determined by the Supreme Court of the United States.” 28 U.S.C. § 2254(d)(1); see also Carey v. Musladin, 549 U.S. 70, 127 S.Ct. 649, 654, 166 L.Ed.2d 482 (2006); Clark v. Murphy, 331 F.3d 1062, 1069 (9th Cir.2003). For an ineffective assistance of counsel claim, the clearly established Federal law that governs is Strickland v. Washington, 466 U.S. 668, 104 S.Ct. 2052, 80 L.Ed.2d 674 (1984), which sets forth a two-step inquiry. “In order to establish ineffective representation, the defendant must” overcome a “highly demanding” standard and “prove both incompetence and prejudice.” Kim-melman, 477 U.S. at 381-82, 106 S.Ct. 2574.1
The majority finds that counsel was deficient on two grounds: Moore’s statement to the police should have been suppressed, first, because it was involuntary and, second, because it was obtained in violation of Miranda. Oddly, the district court’s finding that Moore’s confession was involuntary is not challenged by the state on appeal, and I agree with the majority that the question of voluntariness therefore is not properly before us. See Maj. Op. at 1138-39. I note that, were the issue preserved, a persuasive argument could be made that the confession was in fact given voluntarily. However, since the state-inexplicably2 — has not pressed this issue on *1174appeal, I proceed on the assumption that Moore has demonstrated that his confession was involuntarily given.3
Even conceding that the state has failed to challenge the involuntariness finding, I cannot concede that counsel’s failure to move to suppress necessarily constitutes deficient conduct. It cannot be, as the majority today holds, that because counsel could have filed such a motion, he must have filed the motion. This proposition runs directly counter to clearly established Supreme Court precedent, namely McMann v. Richardson, 397 U.S. 759, 90 S.Ct. 1441, 25 L.Ed.2d 763 (1970). The Court in McMann considered
those situations involving the counseled defendant who allegedly would put the State to its proof ... except for a prior confession that might be offered against him.... At least the probability of the State’s being permitted to use the confession as evidence is sufficient to convince him that the State’s case is too strong to contest and that a plea of guilty is the most advantageous course.
Id. at 767-69, 90 S.Ct. 1441. Of these situations, the Court had this to say:
[A defendant’s] later petition for collateral relief asserting that a coerced confession induced his plea is at most a claim that the admissibility of his confession was mistakenly assessed and that since he was erroneously advised ... his plea was an unintelligent and voidable act. The Constitution, however, does not render pleas of guilty so vulnerable. ... In our view a defendant’s plea of guilty based on reasonably competent advice is an intelligent plea not open to attack on the ground that counsel may have misjudged the admissibility of the defendant’s confession. Whether a plea of guilty is unintelligent and therefore vulnerable when motivated by a confession erroneously thought admissible in evidence depends as an initial matter, not on whether a court would retrospectively consider counsel’s advice to be right or wrong, but on whether that advice was within the range of competence demanded of attorneys in criminal cases.
Id. 769, 770-71, 90 S.Ct. 1441 (second and third emphasis added).
McMann is on all fours with Moore’s claim.4 Moore asserts that his trial coun*1175sel was ineffective in failing to file a motion to suppress Moore’s confession and that he would not have pled no contest to felony murder if he had received competent assistance from counsel. Although filing the suppression motion might seem like a good tactical move now, accepting the state’s plea bargain was the strategic thing to do, knowing what counsel knew at the time. Indeed, counsel’s obligation to explain the risks of trial in light of a plea offer is far more nuanced than the majority’s new mandatory scorched-earth-litigation strategy. Moore’s trial counsel has carefully explained by affidavit his reasons for not filing the motion to suppress:
I did not fail to review [Moore’s] statement to the police. I read it many times and discussed it at length and in detail with Mr. Moore. He affirmed to me that it was true, and that it was accurate.
I did not file a Motion to Suppress. My reasons for doing this were two-fold. First of all, [Moore’s] interview with the police, which was taped and transcribed ... makes it abundantly clear that Mr. Moore was not in custody. He never believed that he was in custody and admitted to me that he realized he was not in custody when he and his brothers and another friend voluntarily came to *1176the police department to give the recorded statement....
[I]n the second place, he had previously made a full confession to his brother [Raymond] and to Ms. [Debbie] Ziegler, either one of whom could have been called as a witness at any time to repeat his confession in full detail....
Counsel explained to Moore, as any competent counsel would, that there was a possibility that the state might charge Moore with aggravated murder because
[t]he victim in this case had been assaulted, bloodied, bound with duct tape] placed in the trunk of a car, taken to an isolated rural location, marched into the woods while still bound, and shot. Furthermore, the victim had an extremely large protruding abdominal hernia for which he always wore a truss. I discussed with Mr. Moore the possibility that if he were ever charged with aggravated murder that the jury might, after taking into account all of the facts of the case, conceivably find that he had engaged in “torture” of a helpless and somewhat disabled victim.
For this and other reasons, counsel feared severe consequences if his client went to trial:
Mr. Moore always claimed his actual shooting of the victim was an accident, but there was never the smallest doubt that it occurred during a kidnap[sic] which began with an assault. We discussed at length the felony murder rule. We also discussed at length the fact that he had not yet been indicted for any conduct and that it was possible that when an indictment came down from the grand jury, it could be for any charge up to ... aggravated murder....
I discussed at great length with Mr. Moore the definitions of “aggravated murder,” “murder,” and “felony murder.” I did tell Mr. Moore that if he were charged with aggravated murder and if the jury decided that murder had been committed under [OR. Rev. Stat. § ]163.095(e), in the course of or as a result of intentional maiming or torture, that it was not impossible that he might be convicted of aggravated murder....
I frankly believed if we went to trial he would be found guilty of assault, kidnapping, and murder (as was his code-fendant, Roy Salyer, who chose trial as an option),[5] but I did not presume to tell Mr. Moore what he should do. I only told him what I thought the result would be of the various choices he had before him. I explained to Mr. Moore that if he chose not to accept the offer which the State had made to him, I expected that the district attorney would charge him with assault in the first degree, kidnapping, and murder,[6] and would go to trial on those charges. At no time during our association did I ever tell Mr. Moore what he should do. I only explained to Mr. Moore as carefully as I could what I thought the result would be of his actions if he chose one option or another.
A conviction of aggravated murder, of course, would have subjected Moore to the possibility of the death penalty or life imprisonment without the possibility of parole. See OR. Rev. Stat. § 163.105(l)(a). *1177Given the strength of the evidence facing Moore, it is not surprising to learn that counsel and Moore decided to “attempt to secure the best possible resolution of the case” or that counsel, who had nearly three decades of criminal defense experience, thought the plea “was the best we could do under the circumstances.”
These strong and obvious strategic reasons to take the plea and forego the suppression motion are protected under Strickland, see 466 U.S. at 681, 104 S.Ct. 2052 (“[Strategic choices must be respected in these circumstances if they are based on professional judgment.”), especially because Moore was so obviously seeking “to save himself the expense and agony of a trial and perhaps also to minimize the penalty that might be imposed.” McMann, 397 U.S. at 767-68, 90 S.Ct. 1441 (emphasis added). The majority’s opinion sweeps all of these factors away. Had the majority been advising Moore at the time, they might have come to a different conclusion. But even accepting the majority’s morning-after conclusion that counsel “misjudged the admissibility of the defendant’s confession,” McMann, 397 U.S. at 770, 90 S.Ct. 1441, Moore is not entitled to habeas relief. See id. at 770-71, 90 S.Ct. 1441.
The majority limits its consideration of counsel’s explanation to a mere two paragraphs of the trial counsel’s affidavit and refuses to consider many of the reasons that trial counsel gave for pursuing the plea bargain instead of going forward with a trial preparation strategy. The majority erroneously believes that trial counsel offered only two reasons to justify his advice to Moore: (1) because he concluded Moore was not in custody at the time of the confession and (2) because Moore had given a full confession to two other people. See Maj. Op. at 1130, 1137, 1138, 1140-41 & n. 12. The majority summarily concludes that there is no evidence that Moore wanted to press the case to early resolution and, therefore, trial counsel could not have made a reasoned strategic choice to not file the suppression motion so as “not ... to upend plea negotiations.” Weaver v. Palmateer, 455 F.3d 958, 972 (9th Cir.2006); see Maj. Op. at 1140-41.
The extensive portions of the affidavit already quoted make clear that counsel’s advice to forego the motion and take the plea was based on numerous considerations other than these two factors. But see Maj. Op. at 1141-43. And other portions of the affidavit demonstrate that plea negotiations were front and center in both Moore’s and trial counsel’s mind. Trial counsel reported that Moore indicated his willingness to testify against a co-defendant, which is surely the type of consideration defense counsel weighs during plea negotiations. Two entire paragraphs of the affidavit, paragraphs 13 and 14, discuss how Moore was more worried about the plea agreement that was offered to his brother, Lonnie Woolheiser, than he was about his own plea agreement.7 Another entire paragraph establishes that trial counsel “discussed at great length whether it was in [Moore’s] best interest to try to press the case to early resolution.” (emphasis added). The majority’s assertion that “there is no suggestion, let alone any evidence, that Moore expressed a desire to plead guilty and avoid trial, or to forego the filing of his meritorious suppression motion, prior to counsel’s decision not to file [a suppression] motion,” Maj. Op. at 1143 n. 16, ignores the reality of the record evidence.8
*1178This kind of meticulous, informed representation, provided by an attorney who had decades of criminal defense experience, does not “show that counsel’s representation fell below an objective standard of reasonableness.” Strickland, 466 U.S. at 688, 104 S.Ct. 2052; McMann, 397 U.S. at 767-71, 90 S.Ct. 1441. More importantly for purposes of this appeal, the state court’s decision that it did not constitute deficient representation was not an unreasonable application of clearly established federal law as determined by the holdings of the Supreme Court.
The majority’s attempt to parse counsel’s advice on whether to take the plea into two distinct decisions — first, whether to file a motion to suppress the confession, and second, whether to accept the offered plea bargain — reflects an almost willful ignorance of the record evidence and the realities of criminal defense representation. See Maj. Op. at 1140-44. As counsel’s declaration makes clear, see Maj. Op. App. B, the decision not to file the suppression motion and the decision to take the plea necessarily informed each other. In fact, trial counsel’s affidavit demonstrates that the two decisions — whether to file a motion to suppress and how to advise Moore on the plea — were made contemporaneously. Moore was never indicted, but he pled no contest to an information negotiated as part of the plea. Counsel simply could not have moved to suppress a confession at any time before the plea, unless the majority means to find counsel ineffective for not threatening to file such a motion in the plea negotiations.
The net effect of the majority’s approach is pernicious: Instead of deciding whether counsel’s conduct fell below an objective standard of reasonableness, the majority asks whether the motion had merit and collapses the entire first step of Strickland into the question of prejudice. See Maj. Op. at 1138 (“[0]ur inquiry with respect to deficient performance substantially overlaps with our inquiry regarding prejudice.”). In doing so, it largely ignores the obvious strategic reasons detailed in counsel’s affidavit that counsel had to advise Moore to take the plea, and the dispositive question becomes whether the motion to suppress had merit. Paired with the majority’s unprecedented reading of Arizona v. Fulminante, 499 U.S. 279, 111 S.Ct. 1246, 113 L.Ed.2d 302 (1991), see Maj. Op. at 1146 (“Fulminante stands for the proposition that the admission of an additional confession ordinarily ... is therefore prejudicial.”), which I address in detail infra, the implication is that if the motion had merit, then counsel was obligated to bring it, irrespective of any other considerations or strategy. It forces defense counsel to file any motions to suppress a confession that a panel of federal judges later might determine to be meritorious, lest the court of appeals find that counsel “failed to recognize the clear merit of that motion” or “to assess properly the damaging nature of the tape-recorded formal confession.” Maj. Op. at 1140.
The majority’s application of the Strickland standard does not accord with the realities of defending a criminal defendant. Defense counsel must balance competing factors when selecting a defense strategy: for example, the likelihood of success on the motion to suppress, the likelihood of prevailing at trial given the other available evidence, the deal that the state is offering, the potential penalties that a defendant can avoid by taking an offered deal, and, of course, the defendant’s own wishes. Strickland gave “wide latitude” to counsel to avoid unhelpful judicial nosiness in plea negotiations:
*1179No particular set of detailed rules for counsel’s conduct can satisfactorily take account of the variety of circumstances faced by defense counsel or the range of legitimate decisions regarding how best to represent a criminal defendant. Any such set of rules would interfere with the constitutionally protected independence of counsel and restrict the wide latitude counsel must have in making tactical decisions.
Strickland, 466 U.S. at 688-89, 104 S.Ct. 2052. A requirement that defense counsel file any potentially meritorious pre-trial motions or risk being found incompetent on collateral review will skew plea negotiations where the considerations promoting negotiation include whether the defendant will file a motion to suppress. If, in response to the majority’s new rule, counsel must file all motions, defense counsel loses a bargaining chip and will almost certainly face a much less cooperative prosecutor. And if defense counsel loses the motion to suppress, counsel will be in a much weaker bargaining position when he returns to the negotiation table. In those cases, the post-motion deal will nearly always be worse than the pre-motion deal.9
The majority would leapfrog over all of those considerations — if the motion to suppress had merit, then counsel must bring it (even if counsel does not think it will serve the client’s best interests). Strickland and its progeny simply do not allow this new presumption of deficient conduct whenever a potentially “meritorious” suppression motion might have been filed— least of all in habeas proceedings governed by AEDPA.
•Í* 4» »J*
In short, I cannot agree with the majority that counsel’s thorough representation constituted deficient performance. Even assuming the involuntariness of Moore’s confession, counsel gave a detailed explanation why pursuing the plea was in Moore’s strategic interest. We can second-guess counsel’s decisions, but we have no basis for concluding that those decisions were constitutionally deficient, let alone that clearly established Federal law, as announced by the Supreme Court, compels such a finding.
II. WHETHER COUNSEL’S CONDUCT PREJUDICED MOORE
Even if Moore’s counsel was ineffective, Moore is only entitled to habeas relief if he can demonstrate that he has suffered prejudice as a result. Unlike the majority, however, I do not believe that Moore has demonstrated prejudice. To demonstrate prejudice under Strickland, 466 U.S. at 694,104 S.Ct. 2052, a petitioner must show that “there is a reasonable probability that, but for counsel’s unprofessional errors, the result of the proceeding would have been different.” Id. at 694, 104 S.Ct. 2052. In Hill v. Lockhart, 474 U.S. 52, 57-59, 106 S.Ct. 366, 88 L.Ed.2d 203 (1985), the Supreme Court applied Strickland’s prejudice requirement “[i]n the context of guilty pleas” and held that the relevant question is “whether counsel’s constitutionally ineffective performance affected the outcome of the plea process. In other words, in order to satisfy the ‘prejudice’ *1180requirement, the defendant must show that there is a reasonable probability that, but for counsel’s errors, he would not have pleaded guilty and would have insisted on going to trial.” Id. at 58-59, 106 S.Ct. 366.
A. The Evidence Against Moore
A straightforward application of the Hill standard demonstrates that Moore cannot establish that he has suffered any prejudice.10 This conclusion flows from two simple facts: First, the state could have called Raymond and Ziegler to testify that Moore confessed his role in the kidnapping and slaying of Kenneth Rogers. We do not need to fret much about what Raymond Moore would have testified had he been called at trial, as the majority does, see Maj. Op. at 1134-35; id. at 1146-47 (suggesting Raymond and Ziegler “would undoubtedly be reluctant to do unnecessary harm to [Moore’s] case”); id. at 1148 (“[I]t is far from clear what[Raymond and Ziegler] would have said or to what extent their testimony would have been persuasive to a jury ....”); id. at 1148 (“Critically, the state court made no findings as to the contents of what Moore had told Raymond or Ziegler or what details they might have been able to recount at trial.”), because the record contains Raymond’s testimony at the state court post-conviction evidentiary hearing. In it, Raymond confirms that, before he took his brothers to the police station where their confessions were recorded, Moore confessed to him the details of what happened. Here is how Raymond described it, under oath, to the state court:
[W]hen [Roy Salyer] got back from Texas and discovered [that Rogers had burglarized Salyer’s home, Salyer] showed up over at Lonnie [Woolhiser’s] house where Randy [Moore] was that morning and ... there was a couple cases of beer and started drinking and basically, to my understanding ... instigated them into going up and, as they put it, spanking their other friend because friends do not rob friends. And in the process of this, I guess, to make an example and put some scare into Mr. Rogers so he did not do this thing again, they had blindfolded him to duct taped him and put him in the trunk of the car and took him out to a place that’s a little remote, not a lot. The gentleman was a large size guy and didn’t walk much and stuff. And their intent was to leave him there and make him walk home, you know, after he freed himself. This here, of course, is the kidnapping that’s involved in this case because they took him somewhere against his will.... [D]uring this period of time, Lonnie who is a little rowdier — he’s a good boy but he’s just a little rowdy. He’ll fight at the drop of a hat. He had in his possession a .22 magnum pistol in which Mr. Rogers had given to him previously — given or traded, I’m not sure which. And Randy, while they were pushing Mr. Rogers up the hill, kind of muddy, it’s during the winter and we have a lot of red mud down there, Randy had taken the revolver from Lonnie and at the time he had taken it, Mr. Rogers had slipped backwards on the mud and the gun discharged. I’m not sure of all the exact details because this is basically hearsay. It’s what was stated in court and it’s what they had basically told me after the incident, too, before I took them in, or on the way in.
(emphasis added). The majority thus is disingenuous in stating that “[t]he state *1181court found only that Moore had ‘confessed’ to” Raymond and Ziegler but did not make specific findings concerning the content of that confession, Maj. Op. at 1134, particularly because trial counsel’s affidavit — which the state court credited — stated that Moore had made a “full confession to his brother and to Ms. Ziegler.” The record clearly showed that Raymond, at least, knew what happened because Moore told him about the details. With this testimony, the state court reasonably concluded that any failure on the part of trial counsel would not have resulted in prejudice to Moore. Trial counsel, the state court, and the district court all recognized that Raymond could have testified about what Moore told him concerning the murder. In its selective treatment of the record, the majority irresponsibly ignores this testimony.
Second, even without his confession to Raymond and Ziegler, the state’s case against Moore would likely have been airtight. I emphasize that the case “would likely have been airtight” because counsel had to judge the strength of the state’s case before the state put it on. Nevertheless, as I show, the case for felony murder was not a difficult one at all.
Under section 163.115 of the Oregon Revised Statutes, Rogers’s killing was a felony murder if “it [was] committed by a person, ... who committed] or attempted] to commit [kidnapping or assault] and [the death occurred] in the course of and in furtherance of the crime the person [was] committing or attempting to commit.” It is undisputed that Moore, Salyer, and Woolhiser went to Rogers’s home, that Rogers was beaten, that he was bound with duct tape, and that he was thrown into the trunk of the car they had borrowed, driven to a remote location, and shot in the temple. It is indisputable that Moore, Salyer, and Woolhiser, by virtue of their involvement in the felonies of kidnapping and assault, were guilty of felony murder under Oregon law. It is equally indisputable that Moore had no affirmative defense. Thus, to convict Moore of felony murder, all that the state needed to do was prove that he took part in Rogers’s kidnapping and that the murder furthered the kidnapping.
This would not have been hard. The state had both Roy Salyer and Lonnie Woolhiser in custody, and both could have been called to testify that Moore took part in the attack. The state court found that Moore and Woolhiser had also confessed to their older brother Raymond Moore as well as to Woolhiser’s girlfriend, Debbie Ziegler.11 Since all of these witnesses *1182were Moore’s close relatives or good friends, their testimony would likely have been very credible.
And yet this only scratches the surface of the damning testimony available to the prosecution. Before they left for Rogers’s residence, Salyer, Moore, and Woolhiser had been drinking with others at Ziegler’s residence. Salyer was “ranting and raving” about how Rogers had broken into his cabin and slashed his tires. This was what prompted the trio to head to Rogers’s residence — to confront him about the robbery and to scare him out of ever committing another one. Four witnesses, including Ziegler, observed this entire interaction and then observed the trio drive off to confront Rogers — in a car that Salyer had borrowed from another one of the guests.
When the trio arrived at Rogers’s residence, other people were there. These people witnessed Moore, Woolhiser, and Salyer arrive, and at least one of those people spoke with Woolhiser about what they were doing there. Another was able to identify all three defendants from a photo lineup. These people would also have been able to testify that at this point the car’s license plates had been covered over with duct tape. Shortly after the defendants arrived, these people all drove away, leaving Rogers alone with Moore, Woolhiser, and Salyer.12
There was also highly inculpatory physical evidence in this case. The day before Moore confessed, Salyer led police to the location of the revolver they had used. The police were unable to find it in the dark, but Woolhiser led them back to the same area the next day, at which point they recovered the weapon. When police found the car that the trio had used, they found blood in the trunk, as well as hair. Although the record does not state that either the blood or the hair was ever scientifically tested, a visual inspection of the hair suggested that it may have been the victim’s, who had long hair that he wore in a ponytail.
So, to summarize, even without Moore’s confession to the police, or his confession to Raymond Moore or Debbie Ziegler, or the testimony of his co-defendants, the prosecution had testimony from multiple witnesses, as well as unchallenged statements from Moore, that: (1) An intoxicated Salyer had been ranting and raving about how Rogers had stolen from his cabin and slashed his tires; (2) Moore had left with Salyer and Woolhiser to confront Rogers; (3) the trio arrived at Rogers’s residence, and that soon thereafter they were alone with Rogers; and (4) the trio returned from Rogers’s residence togeth*1183er. Rogers was found murdered the next day. Woolhiser and Salyer were clearly involved, as they knew where the gun was, and blood and hair were found in the trunk of the car that the trio had borrowed. The police could prove that Moore had lied about what had transpired when he went to Rogers’s residence. Add the testimony of Raymond and Ziegler, and the case is airtight. The state’s felony murder case against Moore could hardly have been any stronger unless the murderers had brought along a camera crew.
All of this evidence — the witnesses, the duct tape, the gun, Moore’s confession to others — were known to Moore and his counsel. They knew what the majority cannot fathom — that the state had a rock-solid case against Moore and his best shot was to strike a plea deal. However, the majority’s formal opinion completely fails to consider almost any of this evidence or its implications for the deal Moore struck. See, e.g., Maj. Op. at 1157 (“Without the fruits of Moore[’s] ... eonfession[ ], the prosecution would have had tremendous difficulty meeting the high burden it faced. In view of the weaknesses in the state’s case, it is highly unlikely that, in the absence of his own recorded confession, Moore would have pled to felony murder. We thus cannot have any confidence that the outcome would have been the same had counsel filed a motion to suppress.”).
In parts of the majority’s formal opinion as well as its appendix, the majority also criticizes my reading of the record; these criticisms lack force. For example, the majority questions my placing any reliance on the testimony of Salyer and Woolhiser, “without any evidence in the record as to the substance or availability of such testimony, or, even more important, its admissibility,” Maj. Op. at 1150-51 (emphasis in original),13 in contrast to its own reliance on “the arguments made by the state on appeal,” id. at 1150-51. First, it is Moore, not the state, who bears the burden of establishing that he has suffered prejudice; his failure to address evidence in the record does not entitle us to ignore it in making our decision. It is utterly absurd for us to put on blinders and pretend that obviously damning evidence — such as the testimony of co-defendants who have pled guilty — simply does not exist.14 Indeed, Moore himself does not expect us to do so; he specifically addresses the issue of Sal-yer’s testimony in his brief. Second, the record leaves little doubt that Salyer and Woolhiser’s testimony would have been admissible, available, and adequate: Both Salyer and Woolhiser were imprisoned by the State of Oregon, so they could easily *1184be produced if necessary.15 As to the substance of their testimony, even if it were not as complete as their formal confessions to police, Moore was on trial for felony murder, so the state needed to prove precious few details of the day’s events to secure a conviction.16
But the most important point is this: Where we are reviewing the state court’s denial of an ineffective assistance of counsel claim, it would be irresponsible for us not to review the record to apprise ourselves of what counsel likely knew. We can only speculate as to what testimony the witnesses actually would have offered had this case actually gone to trial. But it is naive for us to ignore the other evidence in the record. Moore’s counsel did not have the benefit of the majority’s ipse dixit power; when advising Moore on the plea offer, he had to judge his case on the basis of the evidence he thought the state had and the likelihood of various witnesses testifying. This much is clear: There was more than enough admissible evidence, easily obtainable by the state, to convict Moore of felony murder — especially and most importantly, as the state court recognized, the likely testimony of Raymond Moore and Debbie Ziegler. When we consider that evidence, it is plain that not only was counsel’s advice not deficient or prejudicial, it was very good advice. Moore may not have received a lesser sentence than his codefendant who went to trial, but he did avoid a potential death sentence. There is good reason why Strickland requires us to defer to counsel’s on-the-ground judgments over the majority’s own “ ‘post hoc rationalization.’ ” Maj. Op. at 1143 (quoting Wiggins v. Smith, 539 U.S. 510, 526-27, 123 S.Ct. 2527, 156 L.Ed.2d 471 (2003)).
B. The Majority’s Reliance on Fulmi-nante
Ultimately, however, the majority opinion rests on the premise that the state court’s decision resulted in an unreasonable application of Arizona v. Fulminante, 499 U.S. 279, 111 S.Ct. 1246, 113 L.Ed.2d 302 (1991), a case, I note once again, that Moore does not even cite. According to the majority, “Fulminante stands for the proposition that the admission of an additional confession ordinarily reinforces and corroborates the others and is therefore prejudicial.” Maj. Op. at 1146. Fulmi-nante does no such thing. It did not, as the majority’s characterization suggests, adopt a per se rule that the improper admission of a confession is prejudicial;17 in fact, *1185Fulminante explicitly held that the improper admission of a confession is subject to harmless error analysis. See 499 U.S. at 310, 111 S.Ct. 1246. Nor does the fact that the Supreme Court found that the admission of the confession at issue in Fulminante was prejudicial compel the same result here.
The fact that Moore’s confession to the police was probative does not make it prejudicial. Indeed, it is fair to say that the formal, tape-recorded confession that Moore gave to police in their interrogation room was probably more probative than Fulminante’s confession, which he made to a friend while they were both incarcerated. All of the quotations that the majority pulls from Fulminante to establish the damaging, probative value of a confession — that “[a] confession is like no other evidence,” Maj. Op. at 1148 (alteration in original), that a “defendant’s own confession is probably the most probative and damaging evidence that can be admitted against him,” id., that “[t]he admissions of a defendant come from the actor himself, the most knowledgeable and unimpeachable source of information about his past conduct,” id. — are statements that were originally made by the Fulminante Court about confessions like those Moore made to Raymond and Ziegler — ie., those that are made to third parties who then testify about them in court. Thus, Fulminante does not suggest — let alone hold — that Moore was prejudiced because his confession to police was not suppressed. In fact, it supports the exact opposite conclusion— that Moore’s two confessions to Raymond and Ziegler were already so damaging to his case that the admission of his confession to police would do him no further harm. By the majority’s own logic, Moore’s confessions to Raymond and Ziegler were “like no other evidence” and were “probably the most probative and damaging evidence ... against [Moore].” Maj. Op. at 1148.
The majority tries to avoid the implications of its own argument by comparing the relative value of Moore’s confession to the police with his confession to Raymond and Ziegler. The majority states that “Moore’s lawyer ... thought that the taped confession was not prejudicial because Moore had told his brother and his half-brother’s girlfriend about the crime.” Maj. Op. at 1130. According to the majority) “[s]uch a formal confession would, without question, be far more persuasive to a jury than Moore’s statements to two lay witnesses.” Id. at 1147. The majority concludes that the state courts’ “determination that the taped confession was harmless was contrary to clearly established Supreme Court law as set forth in Fulmi-nante.” Id. at 1154; see also id. at 1145 (“[The state court’s] determination that counsel’s failure to suppress the formal taped confession was not prejudicial because Moore had previously told his relative and a relative’s girlfriend about his participation in the killing of the victim was contrary to clearly established Supreme Court law.”). In the majority’s mind, the question under Fulminante comes down to “determining whether the difference between the weight of Moore’s statements to his brother and his half-brother’s girlfriend and his formal taped confession to the police is such that the exclusion of the latter undermines our confidence that Moore would have entered into so harsh a plea agreement.” Maj. Op. at 1153. This approach misstates the law in at least two ways.
First, Fulminante says nothing about determining the relative weight of the two confessions. Rather, it says that harmless error analysis applies to determine whether an erroneously admitted confession is harmless. Fulminante, 499 U.S. at 308, 111 S.Ct. 1246. The Fulminante Court found the admission of the first confession not to be harmless, not because it was *1186more weighty than the second one, but because of a unique relationship between the two confessions — the veracity of the second was bolstered by the existence of the other. Id. at 298-300, 111 S.Ct. 1246. Here, no such relationship exists.18
Second, Fulminante concerned the application of harmless error after a trial had taken place, not after a guilty plea had been entered, as here. The question under clearly established Supreme Court precedent is thus not whether we are confident that Moore “would have entered into so harsh a plea agreement,” Maj. Op. at 1153, an approach that would permit Moore to show that he suffered prejudice if he could have negotiated a better bargain. Instead, the question is whether Moore can show that, but for the failure to file the suppression motion, Moore would not have pled but would have insisted on going to trial. No other standard of prejudice is clearly established Supreme Court law in the guilty plea context. See Hill, 474 U.S. at 58-59, 106 S.Ct. 366. But see Maj. Op. at 1148 (“[T]he record falls far short of establishing that the potential testimony of Raymond and Ziegler would have been sufficient to cause Moore to accept so harsh a plea agreement .... ”); id. at 1153 (“[W]e are left only with determining whether the difference between the weight of Moore’s statements to his brother and his half-brother’s girlfriend and his formal taped confession to the police is such that the exclusion of the latter undermines our confidence that Moore would have entered into so harsh a plea agreement.”). If we were dealing with a jury verdict, the majority’s objection that a “confession is probably the most probative and damaging evidence that can be admitted against [Moore]” would carry more force because “it is impossible to know what credit and weight the jury gave to the confession.” Fulminante, 499 U.S. at 292, 111 S.Ct. 1246 (White, J., dissenting) (internal quotations and citations omitted). But we are dealing with a plea bargain here. The question is whether a reasonably competent attorney would have advised Moore differently when faced with this evidence, and, if so, whether that would have led Moore to go to trial instead of pleading guilty. No Supreme Court case permits us to ask whether Moore would have entered into “so harsh a plea agreement.”
Yet even if we assume Fulminante applies, its facts are so different from those presented here that it is absurd to suggest that it supports the proposition that the state court’s determination violated clearly established federal law. Fulminante was convicted of murdering his eleven-year-old stepdaughter. He made two confessions: one to Anthony Sarivola, a fellow inmate who befriended him, and another to Donna Sarivola, Anthony’s wife. Fulminante had no accomplices, and “the physical evidence from the scene and other circumstantial evidence would have been insufficient to convict.” Fulminante, 499 U.S. at 297, 111 S.Ct. 1246. Thus, unlike here, where the case against Moore was exceptionally strong even without his confessions, “both the trial court and the State recognized that a successful prosecution depended on the jury believing the two confessions.” Id. (emphasis added). The question before the Supreme Court was whether Fulmi-nante was prejudiced by the improper admission of the first confession at trial in light of the properly admitted second confession. The Supreme Court found that Fulminante suffered prejudice because, under the unique circumstances of that case, the jury likely would not believe that the second confession (to Donna Sarivola) *1187had been made if it had not already heard about the first confession (to Anthony Sa-rivola). See id. at 298-300, 111 S.Ct. 1246.
The facts of Fulminante are simply not analogous to those presented here. Moore’s confessions do not need the same kind of background explanation that Ful-minante’s confession to Donna Sarivola required because Raymond and Ziegler were not strangers to Moore in the way that Donna Sarivola was to Fulminante. Under the circumstances, it only made sense that Fulminante would have confessed to her if he had already confessed to her husband. Moore’s confessions to Raymond and Ziegler preceded his confession to the police, and the confessions were not linked in the same way that Fulminante’s confessions to the two Sarivolas were.19 Moore confessed to his older brother Raymond because Raymond himself had once been in a similar situation and Moore wanted his advice. Ziegler and Moore were good friends; he was at her house with her before he left to confront Rogers, he returned to her house afterwards, and he was at her house when Rogers’s murder was reported in the news. Moreover, Raymond and Ziegler did not have the same incentives to lie that Donna Sarivola did. Also unlike Fulminante, here there was other evidence — for example, the duct tape used to bind Rogers, and the blood and hair found in the trunk — to corroborate the details of the testimony that Raymond and Ziegler would have given.20
Finally, regardless of how Fulminante is read, it certainly does not hold that an attorney who declines to file a motion to suppress the challenged confession in favor of advising his client to take a plea on the basis of the presence of the second confession — which the Fulminante Court acknowledged was admissible, see Fulmi-nante, 499 U.S. at 298, 300, 302, 111 S.Ct. 1246 — has provided ineffective assistance.
I close by returning to a common theme of my dissent. I do not believe that the majority’s reading of Fulminante is correct. However, even if I am wrong and *1188the majority is correct, under AEDPA this is not enough to permit us to grant Moore habeas relief. We may only do so if the state court’s decision was contrary to any law that was “clearly established” by the Court in Fulminante. The majority’s reading simply does not meet this standard.
III. PLEA AGREEMENTS AND PREJUDICE
Judge Reinhardt’s majority opinion and Judge Berzon’s concurring opinion both argue, though in slightly different ways, that Moore can demonstrate Strickland prejudice by showing that but for counsel’s unprofessional errors, Moore would have received a more favorable plea agreement. See Maj. Op. at 1131, 1140, 1150 n. 26, 1152-53; Concurring Op. at 1164-65. Neither of their proposed theories has any support in clearly established Federal law.
A. The Majority’s Interpretation of Hill v. Lockhart
The body of the majority’s opinion correctly quotes the applicable standard of prejudice for ineffective assistance of counsel claims arising from guilty pleas, but in a footnote, the majority strangely hypothesizes “that if, in response to the filing of a suppression motion, the state would have offered Moore a more favorable plea bargain (and he would have accepted it in lieu of going to trial), Moore could still have established Strickland prejudice.” Maj. Op. at 1150 n. 26. In other words, the majority believes that even if it is wrong on the merits of Moore’s motion to suppress, Moore would be entitled to relief on the grounds that had counsel filed a suppression motion he might have been in a position to argue for a better plea.
The majority’s innovative and far reaching assertion cannot go unanswered. As support for its novel legal theory that Moore can claim he might have negotiated a better plea bargain — a theory that Moore, incidentally, has never argued — the majority cites to no Supreme Court precedent but only to one of our own cases, United States v. Howard, 381 F.3d 873 (9th Cir.2004). See Maj. Op. at 1150 n. 26. The majority’s failure to cite anything else is hardly surprising since, as pointed out above, Hill unequivocally held that, “in order to satisfy the ‘prejudice’ requirement, the defendant must show that there is a reasonable probability that, but for counsel’s errors, he would not have pleaded guilty and would have insisted on going to trial.” Hill, 474 U.S. at 59, 106 S.Ct. 366 (emphasis added). Moreover, the Supreme Court reaffirmed Hill — and thereby rejected the majority’s reading— in Roe v. Flores-Ortega, 528 U.S. 470, 120 S.Ct. 1029, 145 L.Ed.2d 985 (2000). In that case, the Supreme Court “h[e]ld that when counsel’s constitutionally deficient performance deprives a defendant of an appeal that he otherwise would have taken, the defendant has made out a successful ineffective assistance of counsel claim entitling him to an appeal.” Id. at 484, 120 S.Ct. 1029. In so holding, the Court stated
[i]n Hill, we considered an ineffective assistance of counsel claim based on counsel’s allegedly deficient advice regarding the consequences of entering a guilty plea. Like the decision whether to appeal [presented here], the decision whether to plead guilty (i.e., waive trial) rested with the defendant and, like this case, counsel’s advice in Hill might have caused the defendant to forfeit a judicial proceeding to which he was otherwise entitled.
Id. at 485, 120 S.Ct. 1029. The Court proceeded to cite Hill for the proposition that “when, in connection with a guilty plea, counsel gives deficient advice regarding a potentially valid affirmative defense, the prejudice inquiry depends largely on whether that affirmative defense might *1189have succeeded, leading a rational defendant to insist on going to trial.” Id. at 486, 120 S.Ct. 1029.
Yet the majority ignores this language by misconstruing our opinion in Howard, where we stated that “[t]o satisfy Strickland’s, prejudice prong, [a defendant] must allege that but for counsel’s errors, he would either have gone to trial or received a better plea bargain.” 381 F.3d at 882. Considered in context, it is clear that our statement in Howard was merely loose language and not a holding. No other panel has ever read Howard the way that the majority now interprets it. Howard cites Hill only in passing, provides no justification for its deviation from our prior case law, does not discuss its novel reading, and did not turn on whether the defendant would have been able to secure a more favorable plea agreement. Moreover, Howard itself does not consistently hold to the majority’s reading; every time that a question turned on the standard of prejudice, the Howard court hewed to the accepted construction of Hill — that a petitioner must demonstrate that, but for counsel’s ineffective assistance, he would not have pled guilty. It interpreted unclear allegations in Howard’s pro se brief “as a sufficient statement that he would not have entered the plea and would have taken the case to trial if his counsel had not permitted him to plead while incompetent,” and stated that “[t]his ... allegation, if true, would establish prejudice.” Howard, 381 F.3d at 883.21 Finally, the majority asserts that the principle it derives from Howard “makes good sense” because “[t]he vast majority of cases in criminal courts are resolved by plea bargains.” Maj. Op. at 1150 n. 26. There is some force to this argument, but it is only an argument that the Supreme Court should have adopted the majority’s rule, not — as required under AEDPA — that it actually did.22 Indeed, commentators have *1190criticized the holding in Hill precisely because it does not afford relief to many prisoners who received constitutionally deficient representation. See, e.g., Richard Klein, Due Process Denied: Judicial Coercion in the Plea Bargaining Process, 32 HofstRA L. Rev. 1349, 1369 (2004) (“We are left by the [Hill u] Lockhart decision with a most unfortunate result: in the vast majority of instances where an effective, competent counsel could have negotiated a better plea for the defendant than his incompetent counsel did, there will be no remedy.”); Emily Rubin, Note, Ineffective Assistance of Counsel and Guilty Pleas: Toward a Paradigm of Informed Consent, 80 Va. L. Rev. 1699, 1705-06 (1994) (“[I]t is a very rare defendant who insists on taking his case to trial”; in most cases, a defendant with the benefit of adequate representation “would have eventually pleaded guilty pursuant to a more generous [plea offer than the one he accepted]. Such a defendant has no recourse under Hill because he will be unable to demonstrate that he probably ‘would have insisted on going to trial.’ ”). The majority simply installs its own view of what Hill should have said, and it infects the entire majority opinion. See, e.g., Maj. Op. at 1140 (“[Invalidating Moore’s formal, tape-recorded confession would have placed him in a far better position to negotiate a reasonable plea and obtain a lesser sentence ....”); id. at 1148 (“[T]he record falls far short of establishing that the potential testimony of Raymond and Ziegler would have been sufficient to cause Moore to accept so harsh a plea agreement....”); id. at 1153 (“[T]he[failure to] exclu[de] the [formal confession] undermines our confidence that Moore would have entered into so harsh a plea agreement.”).
B. The Concurrence’s Prejudice Standard
Judge Berzon’s concurring opinion argues that Moore could choose to satisfy the Strickland prejudice prong either under Hill or directly under Strickland as applied through the lens of Kimmelman. See Concurring Op. at 1165. However, Judge Berzon views the Kimmel-man/Strickland framework as more appropriate for resolving this case because Hill governs prejudice determinations in plea bargains concerning counsel’s advice on whether to take the plea, after “motions practice and discovery have set the legal landscape.” Id. at 1165. On the other hand, Strickland and Kimmelman, in her view, “deal with counsel’s failure to create a proper legal landscape — by, [for example], failing to file a plainly meritorious suppression motion.” Id. at 1165. Judge Berzon concludes that the Kimmelman standard, applied in the plea context, permits prejudice to be established by showing that had defense counsel properly shaped the legal landscape prior to the plea proceedings, the defendant might have obtained a more favorable plea bargain from the prosecutor. Id. at 1165-66. The distinction she attempts to draw is precluded not only by Hill itself, but also by the vast weight of precedent in both the courts of appeals and the district courts. Her “legal landscape” argument is interesting but it has never been “established,” much less clearly established, by the Supreme Court. It also raises substantial concerns about federalism and separation of powers.
The concurrence’s critical mistake is its failure to consider fundamental principles governing the appeal of guilty pleas. A *1191criminal defendant who has entered a plea generally waives his right to challenge defects in the pre-plea proceedings. Instead, a defendant who has entered a plea “may not thereafter raise independent claims relating to the deprivation of constitutional rights that occurred prior to the entry of the guilty plea. He may only attack the voluntary and intelligent character of the guilty plea by showing that the advice he received from counsel was not within the standards set forth in McMann [v. Richardson, 397 U.S. 759, 770-72, 90 S.Ct. 1441, 25 L.Ed.2d 763 (1970) ].” Tollett v. Henderson, 411 U.S. 258, 267, 93 S.Ct. 1602, 36 L.Ed.2d 235 (1973); see also McMann, 397 U.S. at 770-71, 90 S.Ct. 1441 (“Whether a plea of guilty is unintelligent and therefore vulnerable when motivated by a confession erroneously thought admissible in evidence depends ... not on whether a court would retrospectively consider counsel’s advice to be right or wrong, but on whether that advice was within the range of competence demanded of attorneys in criminal cases.”); Ortberg v. Moody, 961 F.2d 135, 137-38 (9th Cir.1992) (“Petitioner’s nolo contendere plea precludes him from challenging alleged constitutional violations that occurred prior to the entry of that plea.” (citing Tollett, 411 U.S. at 266-67, 93 S.Ct. 1602)).
With that fundamental principle of pleas in mind, it becomes obvious that Hill provides the only appropriate standard for evaluating claims of ineffective assistance in the plea context. The only thing that can be challenged after a plea is the advice to enter a particular plea, for all other defects are waived by the plea. To be sure, Strickland is not irrelevant to the analysis under Hill; the Hill Court explicitly adopted the Strickland standard in the context of guilty pleas. “We hold, therefore, that the two-part Strickland v. Washington test applies to challenges to guilty pleas based on ineffective assistance of counsel.” Hill, 474 U.S. at 58, 106 S.Ct. 366. But the process of applying Strickland to guilty pleas was set forth in Hill, and there is no reason to use a different prejudice analysis than that established in Hill.
In Hill, the Court wrote that “[i]n many guilty plea cases, the ‘prejudice’ inquiry will closely resemble the inquiry engaged in by courts reviewing ineffective-assistance challenges to convictions obtained through a trial.” Id. at 59, 106 S.Ct. 366. “[W]here the alleged error of counsel is a failure to investigate ..., the determination whether the error ‘prejudiced’ the defendant by causing him to plead guilty rather than go to trial will depend on the likelihood that ... [counsel] would have ... ehange[d] his recommendation as to the plea.” Id. But this inquiry only goes to the question of whether the advice to enter the plea caused prejudice — i.e., whether the defendant would not have pled guilty with better advice — not whether the prosecution might have offered a different plea agreement. Put differently, even accepting Judge Berzon’s proposed distinction between the advice to enter a plea and the process of creating the legal landscape in which plea bargaining occurs, if a defendant still would have pled guilty or nolo contendere despite trial counsel’s unprofessional errors in crafting the legal landscape in which the plea was entered, then there is no constitutional prejudice.
Judge Berzon’s response that Hill did not deal with the creation of pre-trial landscapes while Kimmelman did, see Concurring Op. at 1166-67, only highlights why the Hill standard is the correct one. Under Hill, if it was reasonable to advise a defendant to take a plea, there is no deficient conduct. If counsel has been unreasonable in giving that advice, then we proceed to ask whether the defendant would have taken the plea anyway. In other words, Judge Berzon’s concerns about the *1192pre-plea landscape are adequately addressed in the first step of Hill.
There is good reason for ns to follow Hill in these circumstances. Judge Ber-zon would have us consider the “legal landscape” and ask whether “the plea bargain outcome would have been improved upon the filing of the meritorious suppression motion that was not filed because of ineffective assistance of counsel.” Concurring Op. at 1167-68. We have no way of evaluating whether the prosecutor, having been forced to answer the motion to suppress, would even be willing to offer a new plea bargain, much less whether the prosecutor would have offered an “improved” “plea bargain outcome.” Id.
Given the multiplicity of factors that a prosecutor must consider when offering a plea bargain, it is highly doubtful that a federal court, reviewing a state prosecutor’s decision to offer a particular plea bargain, even has the tools necessary to decide what bargaining posture a prosecutor would take in the face of a hypothetical motion to suppress. When deciding what plea bargain to offer a particular criminal defendant, for example, a prosecutor might consider the willingness of the defendant to cooperate, the defendant’s past criminal history, department resources, and pressure from the public in high profile or emotionally charged cases. See Wayte v. United States, 470 U.S. 598, 607, 105 S.Ct. 1524, 84 L.Ed.2d 547 (1985); see also United States v. Estrada-Plata, 57 F.3d 757, 760 (9th Cir.1995) (“[T]here is no constitutional right to a plea bargain, and the decision whether to offer a plea bargain is a matter of prosecutorial discretion.” (citation omitted)). To further complicate Judge Berzon’s proposed counterfactual analysis, all of this second-guessing will be conducted, in most cases, years after the decision to offer the challenged plea bargain. Political winds may have shifted or a new prosecutor may have taken office. Permitting a habeas petitioner to demonstrate prejudice simply by showing that a different plea bargain might have been offered calls for an answer to an impossible question, and will have the effect of unsettling scores of negotiated state convictions, encouraging needless litigation, and creating a mass printing press in the federal courts for writs of habeas corpus.
Were this the only side effect of Judge Berzon’s method, perhaps it would be tolerable. But there are at least two additional problems. First, it places federal courts in the role of instructing state prosecutors — members of the state executive branch — of how to conduct plea negotiations, or at least how much prison time a prosecutor is permitted to offer if the state decides to proceed with reprosecution after the writ of habeas corpus issues. To put it mildly, this kind of interference with a state executive branch function raises substantial federalism concerns. Cf. Printz v. United States, 521 U.S. 898, 933, 117 S.Ct. 2365, 138 L.Ed.2d 914 (1997).
Second, and related to the federalism problem, Judge Berzon’s approach implicates the separation of powers with potential effects far beyond the current case.23 Conducting this type of inquiry into whether a better plea bargain would have been available would require the kind of judicial review of prosecutorial decisions that courts have almost uniformly shunned. We have previously described the reasons *1193for avoiding judicial review of the plea bargaining decision:
Prosecutorial charging and plea bargaining decisions are particularly ill-suited for broad judicial oversight. In the first place, they involve exercises of judgment and discretion that are often difficult to articulate in a manner suitable for judicial evaluation. Such decisions are normally made as a result of careful professional judgment as to the strength of the evidence, the availability of resources, the visibility of the crime and the likely deterrent effect on the particular defendant and others similarly situated. Even were it able to collect, understand and balance all of these factors, a court would find it nearly impossible to lay down guidelines to be followed by prosecutors in future cases. We would be left with prosecutors not knowing when to prosecute and judges not having time to judge.
Assuming these problems of guidance and understanding could be overcome— and it is unlikely that they could be— there is an added constitutional consideration based on the peculiar relationship between the Office of the United States Attorney and the federal district courts: The United States is necessarily a party to every criminal case presented to a district court. It would raise serious separation of powers questions — as well as a host of virtually insurmountable practical problems — for the district court to inquire into and supervise the inner workings of the United States Attorney’s Office.
The very breadth of the inquiry ... would require that the government divulge minute details about the process by which scores, perhaps hundreds, of charging decisions are made. The court would also have to consider the validity of various rationales advanced for particular charging decisions, which would enmesh it deeply into the policies, practices and procedures of the United States Attorney’s Office. Finally, the court would have to second-guess the prosecutor’s judgment in a variety of cases to determine whether the reasons advanced therefor are a subterfuge.
United States v. Redondo-Lemos, 955 F.2d 1296, 1299-1300 (9th Cir.1992) (footnotes and citations omitted), overruled on other grounds, United States v. Armstrong, 48 F.3d 1508, 1515 n. 5 (9th Cir. 1995) (en banc); see also Wayte, 470 U.S. at 607-08, 105 S.Ct. 1524 (recognizing that the “broad discretion” afforded the executive to evaluate such factors is “not readily susceptible to the kind of analysis the courts are competent to undertake”); United States v. Banuelos-Rodriguez, 215 F.3d 969, 976 (9th Cir.2000) (en banc) (“Courts generally have no place interfering with a prosecutor’s discretion whom to prosecute, what charges to file, and whether to engage in plea negotiations.”).
To be sure, prosecutorial discretion, including the discretion to negotiate plea bargains, does not give the executive branch license to violate a criminal defendant’s due process rights, and courts widely agree that a prosecutor cannot hide discriminatory motives under the guise of prosecutorial discretion. See United States v. Arenas-Ortiz, 339 F.3d 1066, 1068 (9th Cir.2003) (“One important restriction on prosecutorial discretion, however, is that ‘the decision whether to prosecute may not be based on an unjustifiable standard such as race, religion, or other arbitrary classification.’” (quoting United States v. Armstrong, 517 U.S. 456, 464, 116 S.Ct. 1480, 134 L.Ed.2d 687 (1996)) (internal quotations removed)); Redondo-Lemos, 955 F.2d at 1299 (“Given the significance of the prosecutor’s charging and plea bargaining decisions, it would offend common notions of justice to have them made on the basis of a dart throw, a coin toss or some other arbitrary or capricious *1194process.”)- But it is one thing to say that the executive branches of either the federal or state governments cannot prosecute selectively, for there are methods through which the judiciary can evaluate such claims. It is an entirely different, and far more bothersome, thing to instruct district courts to ask whether “the plea bargain outcome would have been improved upon the filing of the meritorious suppression motion.” Concurring Op. at 1167-68. That analysis would require inquiry into the precise questions that this court has said courts should avoid and which the Constitution likely protects from judicial intrusion. See Redondo-Lemos, 955 F.2d at 1299-1300.
Judge Berzon responds by claiming that this is the type of inquiry that courts regularly undertake in ineffective assistance of counsel claims. See Concurring Op. at 1169-70. Rather than considering the prosecution’s actions, Judge Berzon says that her approach would focus on “the defendant’s and defense counsel’s choices, defense counsel’s judgment, and defense counsel’s actions.” Id. at 1170. But if the question posed is whether “the plea bargain outcome would have been improved upon the filing of the meritorious suppression motion,” as she says it ought to be, id. at 1167-68, I fail to see how that could be done without looking at the prosecution’s decisions “in the first instance.” Id. at 1170. In an ineffective assistance of counsel claim following a trial, where the coun-terfactual question posed to the court concerns evaluating what a jury might have done, at least the evidence presented to the jury and the legal instructions it was given are available for review. In contrast, in a plea bargain situation, there is no record at all about what other deals the prosecution might have offered.24
If, in contrast, “the question is whether, but for counsel’s ineffective assistance, a defendant would [be] in a better position to negotiate with the prosecutor,” as Judge Berzon articulates the test later on, id. at 1170, the new standard would entirely swallow Hill. Numerous cases decided under Hill can also be characterized as “dealing] with counsel’s failure to create a proper legal landscape” by failing to take some strategic action. Concurring Op. at 1165. See, e.g., Weaver, 455 F.3d at 970-71 (9th Cir.2006) (applying Hill to counsel’s failure to investigate mental defect defense); Langford, 110 F.3d at 1386-87 (using Hill in guilty plea case alleging ineffective assistance of counsel based on failure to file various suppression motions). And filing a potentially meritorious suppression motion will always strengthen defense counsel’s bargaining position (at least until a potentially adverse ruling is handed down). If the possibility that a more favorable plea bargain might have been offered if a potentially meritorious motion was not filed is sufficient to establish Strickland prejudice after a guilty plea, virtually every plea bargain in the country is now open to habeas relief.
Viewed within the proper standard of review under AEDPA, it was not an unreasonable application of Supreme Court law for the Oregon courts to evaluate Moore’s claim under Hill. The case fits squarely within the rule of Hill: Moore asserts that his trial counsel failed to advise him that a motion to suppress might be successful. On the basis of that advice, Moore pled no *1195contest to the charge of felony murder. Other courts considering such claims have uniformly looked to Hill for the correct standard.25 Given that the federal courts have been applying Hill’s prejudice requirement to situations similar to the one presented by this case, it is difficult to see how a state court should have been able to divine this kind of extension of Strickland and Kimmelman. Judge Berzon’s prejudice standard is not only not clearly established Supreme Court law; it is also contrary to all other federal authority. I therefore puzzle over Judge Berzon’s objection that I do not cite to any case holding that her reading of the prejudice standard is precluded by Hill, see Concurring Op. at 1169, for all the cases she cites applying the Kimmelman standard occurred after a trial and did not involve the plea context. She also fails to respond to the numerous cases I have cited which apply Hill in the guilty plea context, other than to say that “[s]ome other cases ... assume (as Judge Reinhardt does today) that Hill also applies in the motions context when a plea bargain is involved, as well as in the advice context.” Id. From that, despite the absence of any case applying her proposed distinction to a guilty plea case, she draws the inexplicable con-elusion that “these cases do not provide support for using Hill as the exclusive standard in such circumstances.” Id.
The Supreme Court would have had difficulty being more clear than it was in Hill about the proper prejudice standard for guilty pleas. It said, <cWe hold, therefore, that the two-part Strickland v. Washington test applies to challenges to guilty pleas based on ineffective assistance of counsel. In the context of guilty pleas, ... to satisfy the ‘prejudice’ requirement, the defendant must show that there is a reasonable probability that, but for counsel’s errors, he would not have pleaded guilty and would have insisted on going to trial.” The Hill Court did not recognize Judge Berzon’s distinction between pretrial landscape setting ineffective assistance claims and advice to enter guilty plea claims, and she cannot point to a single federal court since Hill that has done so. Until the Supreme Court tells us otherwise, her proposed distinction is foreclosed by Hill. See Carey v. Musladin, 549 U.S. 70, 127 S.Ct. 649, 654, 166 L.Ed.2d 482 (2006) (“Given the lack of holdings from this Court ..., it cannot be said that the state court ‘unreasonably] appli[ed] *1196clearly established Federal law.’ ” (quoting 28 U.S.C. § 2254(d)(1))).
The concurrence’s argument is a disguised attempt implicitly to extend Howard, to permit satisfaction of Strickland’s prejudice prong by demonstrating simply that a more favorable plea might have been obtained, instead of doing it explicitly in the manner advocated by Judge Reinhardt. See Maj. Op. at 1150 n. 26. Although Judge Berzon couches her argument as giving defendants the choice of proving prejudice under what is commonly perceived as the more difficult Strickland standard rather than proving prejudice under Hill, her approach actually achieves the opposite effect: Instead of being required to demonstrate that he would not have pled guilty but would have insisted on going to trial, Moore would only be required to demonstrate that he might have obtained a more favorable plea agreement. This attempt misconceives the nature of plea bargaining, potentially violates principles of federalism and separation of powers, and ignores Supreme Court precedent on how to conduct the prejudice inquiry following plea agreements. Even if I were to agree that an ambiguity about the relationship between Hill, Kimmelman, and Strickland existed, it is not our job on AEDPA review to resolve it.
IV. CONCLUSION
I cannot join anything the majority has written. For the reasons I have stated, I believe the majority to be wrong on the facts and the law, and I believe that it fails to accord the state court’s decision the deference that AEDPA commands. At the end of the day, it is not clear what the majority has accomplished, for Moore or for anyone else. The majority grants Moore a writ of habeas corpus and orders the state either to permit Moore to withdraw his plea or to release him. Oregon will surely allow Moore to withdraw his plea and then prosecute him to the hilt. When it does, Oregon will be under no obligation to offer Moore any kind of a deal, and if it does decide to bargain, it has no obligation to offer Moore a plea bargain as attractive as what he got in this case. It may even decide to seek the death penalty. And even if Oregon were to offer a new plea deal, Moore’s counsel must reject it until he has filed every conceivable pre-trial motion he can. After today’s decision, no conscientious defense attorney should even consider accepting a plea deal — no matter how good the bargain and no matter what other evidence the prosecutor has — if there are potentially “meritorious” motions that can be filed.
Oregon will try Moore and, given his confessions to family and friends, the available eyewitnesses, and other incontrovertible evidence, Moore will likely be found guilty of murder. For that, he is likely to receive a sentence well in excess of the bargain he negotiated. It is quite possible that Moore will be worse off for having prevailed here. Nor is it clear that anyone else after Moore will actually benefit from today’s ruling. In fact, defendants whose counsel cannot negotiate plea agreements until after exhausting their pre-trial motions are likely to be worse off for the majority’s effort.
Today’s decision is not a liberty-enhancing decision. It will actually hamper defense counsel’s ability to avoid trial and negotiate plea agreements. And our decision is so unnecessary. Moore is plainly guilty of felony murder, or worse. He took a fair deal from the prosecutor on the advice of competent counsel. Justice was served. There is no reason for us to upend the orderly administration of justice in Oregon in this way.
I respectfully dissent.
. To meet the first prong, “the defendant must show that counsel’s representation fell below an objective standard of reasonableness,” Strickland., 466 U.S. at 687-88, 104 S.Ct. 2052, and overcome the "strong presumption that counsel’s performance falls within the 'wide range of professional assistance,' ” Kimmelman, 477 U.S. at 381, 106 S.Ct. 2574 (quoting Strickland, 466 U.S. at 688-89, 104 S.Ct. 2052). When the habeas petitioner alleges that his counsel was ineffective because of counsel's failure to file a motion, a necessary, but by no means sufficient, condition for a successful showing of incompetence is that the motion would have been meritorious. See Kimmelman, 477 U.S. at 375, 382, 106 S.Ct. 2574. And, to meet the second, “[t]he defendant must show that there is a reasonable probability that, but for counsel's unprofessional errors, the result of the proceeding would have been different.” Strickland, 466 U.S. at 694, 104 S.Ct. 2052.
. The district court based its finding of involuntariness on an implied promise of leniency that the police allegedly gave to Moore. Yet, to use Moore’s own words, the officers promised him “[njothing other [than] helping the best [they] could.” This offer to recommend leniency to the district attorney is inadequate *1174to establish that "all of the attendant circumstances” indicate that " ‘the defendant's will was overborne at the time he confessed.' ” Haynes v. Washington, 373 U.S. 503, 513, 83 S.Ct. 1336, 10 L.Ed.2d 513 (1963) (quoting Lynumn v. Illinois, 372 U.S. 528, 534, 83 S.Ct. 917, 9 L.Ed.2d 922 (1963)); see also United States v. Guerrero, 847 F.2d 1363, 1366 (9th Cir.1988) ("An interrogating agent's promise to inform the government prosecutor about a suspect's cooperation does not render a subsequent statement involuntary, even when it is accompanied by a promise to recommend leniency or by speculation that cooperation will have a positive effect.”). The record demonstrates that the police could not have promised anything other than a recommendation of leniency — and certainly could not promise a reduction of charges — because Moore had not been formally booked or charged when he made his confession.
. The conclusion that the state has forfeited its argument concerning the voluntariness of Moore's confession relieves us of the obligation to discuss the Miranda issue. It is not relevant to the rest of our inquiry — not even to "buttress[the majority's] conclusion that [counsel's] performance was highly deficient.” Maj. Op. at 1139 & n. 11. The majority's discussion of the Miranda claim in footnote 11 is thus double dicta — resolving the Miranda claim is unnecessary to the majority’s decision, and its discussion is way beyond our charge under AEDPA. I will not prolong my dissent with further discussion of the Miranda claim, except to note that there is substantial merit in the state’s argument that Moore was not in custody when he invoked his right to counsel and, therefore, Miranda did not apply. And, certainly, the state courts' decision is not an unreasonable application of any Supreme Court decisions.
. The majority laments that my "invocation of McMann v. Richardson is misplaced” because *1175“Moore's challenge is not to counsel’s plea advice, as was the case in McMann, but to counsel’s failure to file a motion to suppress.” Maj. Op. at 1142 n. 14. True enough, the claim in McMann was styled as a challenge to the voluntariness of a plea. But the McMann Court treated the claim as one of ineffective assistance of counsel. See McMann, 397 U.S. at 770, 90 S.Ct. 1441 ("[D]efendants facing felony charges are entitled to the effective assistance of competent counsel.... [The] matter ... should be left to the good sense and discretion of the trial courts with the admonition that if the right to counsel guaranteed by the Constitution is to serve its purpose, defendants cannot be left to the mercies of incompetent counsel. ...”). Further, although McMann was decided before Strickland and Hill, those two cases were built upon and reaffirmed McMann. "In the context of guilty pleas, the first half of the Strickland v. Washington test is nothing more than a restatement of the standard of attorney competence already set forth in Tollett v. Henderson and McMann v. Richardson.” Hill v. Lockhart, 474 U.S. 52, 57-58, 106 S.Ct. 366, 88 L.Ed.2d 203 (1985). McMann informed the creation of the first part of the two-part test in Strickland. See 466 U.S. at 687, 104 S.Ct. 2052 (”[T]he proper standard for attorney performance is that of reasonably effective assistance. The Court indirectly recognized as much when it stated in McMann v. Richardson that a guilty plea cannot be attacked as based on inadequate legal advice unless counsel was not a ’reasonably competent attorney’ and the advice was not 'within the range of competence demanded of attorneys in criminal cases.’ ” (quoting McMann, 397 U.S. at 770-71, 90 S.Ct. 1441)); see also Kimmelman,477 U.S. at 377-78, 106 S.Ct. 2574 (citing McMann); Hill, 474 U.S. at 57-58, 106 S.Ct. 366; accord Langford v. Day, 110 F.3d 1380, 1386-87 (9th Cir.1996) (applying McMann, as interpreted by Hill, in case alleging ineffective assistance of counsel based on failure to file various suppression motions). Indeed, the question in McMann and the question here are identical: whether a reasonably competent attorney would have foregone the suppression motion and advised the client to plead guilty. If McMann means anything, it means that counsel's advice on the admissibility of a confession is not unreasonable simply because two federal judges disagree with it years later.
The majority cites two cases that it holds out as establishing that we have recognized “Kimmelman-type Strickland claims ... in cases in which the defendant pled rather than going to trial.” Maj. Op. at 1142 n. 14. These cases provide, at best, weak support for the majority's assertion that McMann should not inform our analysis here. In both cases, we concluded that the defendant’s claim of ineffective assistance failed because the decision not to file the motion was a strategic one or because the defendant insisted on forgoing the motions. See Weaver v. Palmateer, 455 F.3d 958, 972 (9th Cir.2006); Langford, 110 F.3d at 1386-88.
. Unlike Moore, Salyer chose to proceed to trial and was found guilty of murder, kidnapping, burglary, and assault. See Salyer v. Belleque, 2005 WL 555403, at *1 (D.Or. March 4, 2005).
. Very strong circumstantial evidence contradicted Moore's "accidental shooting” explanation and would have justified an intentional murder charge. Moore shot Rogers with a revolver — a gun that would not discharge unless it had been cocked by the shooter. Furthermore, Rogers was shot in the temple, as if he was executed, and not in the back, as might be expected if he had slipped and fallen backward onto Moore.
. The affidavit indicates that trial counsel properly informed Moore that the offer to him was independent of the offer made to his brother.
. Contrary to the majority’s claims, see Maj. Op. at 1141-43, the record does not reflect whether the plea negotiations occurred before or after trial counsel advised Moore that a *1178suppression motion would be unlikely to succeed or would otherwise be useless. The record does reflect that Moore wanted to press for an early resolution.
. Moreover, if, as the majority effectively holds today, defense counsel must always move to suppress a confession or risk a claim of ineffective assistance of counsel, there is no end to the second-guessing game. If counsel moved to suppress and lost on that motion, then, by the majority’s reading, counsel must see the case through trial and take an appeal. And if he loses on appeal, then he must file for habeas — all so that he can be in the best position to negotiate a plea bargain. The bottom line is that counsel, in defense of his own reputation, will not seek plea agreements or will counsel against accepting a plea bargain. This makes no sense.
. Judge Berzon argues in her concurrence that Moore had the option of meeting the prejudice standard under either Strickland/Kimmelman or Hill and, further, that he meets both of those standards. See Concurring Op. at 1164-65. I address her arguments in Part III.B, infra.
. The majority spends a great deal of time discounting the value of these two confessions by speculating on what the witnesses would or would not have done. It finds that "Raymond would likely have been a hostile witness” and that "it is unlikely that [the state] would have been able to elicit much of the information it desired from him.” Maj. Op. at 1148 & n. 24. This is raw speculation, and it is belied by the testimony Raymond actually provided to the state court. It is also patently absurd. Because the broad outlines of what happened were very clear — i.e., that Rogers was beaten in his home, bound, kidnapped, taken to a remote location and shot in the head while still bound — the state needed to establish very little to convict Moore of felony murder. Focusing on statements made by Ziegler during Moore's initial questioning, the majority states that "there is little evidence that Ziegler could have contributed anything.” Maj. Op. at 1148. But the state court did not rest its finding that Ziegler could have repeated Moore’s confession solely on the statements she made at questioning. The state court explicitly found that Moore’s trial counsel's affidavit, including the paragraph explaining that no motion to suppress was filed because Ziegler or Raymond Moore could have been called to testify, was truthful. As already noted, see supra at 1175-76, the affidavit clearly states that Moore admitted to making a full confession to Ziegler. Contrary to the majority's assertions, see Maj. Op. at 1147-48 & n. 24, the state court's ruling that Moore confessed to Ziegler did not conflict with the interrogation transcript: Just be*1182cause Ziegler learned of the murder on the day of the interrogation does not mean that she learned of it at the interrogation. I find it remarkable that the majority, twelve years after the incidents in question, and without knowing any of the parties involved, can divine what Raymond and Ziegler were thinking and what they would or would not have done. I also find it remarkable that the majority brushes aside, with almost no discussion, the state court’s factual findings concerning these confessions.
. With the exception of the lineup identification, the prosecution could establish all of the facts in the last two paragraphs by relying solely on statements made by Moore, the admissibility of which has never been questioned. The day before the interrogation at issue in this case, the police approached Moore and Woolhiser while they were eating at a local restaurant, and the pair agreed to come by the station house and answer questions after they had finished their dinner. After arriving at the station house, Moore gave a description of events that was largely accurate up until the point that Woolhiser and Salyer began beating Rogers. The next day, Woolhiser admitted to asking the owner of the car if he had anything that they could use to tie Rogers up, to which the owner responded that he had some duct tape.
. This criticism seems particularly strange in light of the extensive discussion that the majority provides of why, in its opinion, Ziegler and Raymond would not have made good prosecution witnesses. See, e.g., Maj. Op. at 1147-48; see also supra at 1181-82 n. 11.
. The majority’s criticism on this score seems all the more hypocritical because it has no hesitation, apparently, in manufacturing new arguments on behalf of Moore, who did not cite on appeal the two cases most important to the majority’s opinion, Kimmelman and Fulminante, or make any argument related to the theories on which the majority bases its decision. But see Maj. Op. at 1152 ("The forfeiture rule (sometimes erroneously called the waiver rule) applies equally to arguments, factual assertions, and legal theories that were not urged below.”). To borrow from the majority opinion, I "recognize that [my] colleague[s] believe[] that Moore deserves to [get a less harsh penalty], but disregarding the [appellate] arguments as well as the state court record and findings, and substituting one's own, is hardly the manner in which federal appellate courts are supposed to determine appeals,” Maj. Op. at 1131 n. 1, and "[w]e may not as appellate judges manufacture such arguments from scratch, especially where, as here, the facts in the record are directly contrary to the theory we are seeking to create on behalf of one of the parties,” Maj. Op. at 1143.
. The majority's inadmissibility argument— an argument that Moore himself only makes in passing and, even then, only with reference to Salyer — fails for the same reason. The majority’s claim that any of the evidence I cite above flowed from "Woolhiser's involuntary confession[],” Maj. Op. at 1155, and is therefore inadmissible is simply wrong because Woolhiser could have been called to testify directly.
. The majority cites Lee v. Illinois, 476 U.S. 530, 545, 106 S.Ct. 2056, 90 L.Ed.2d 514 (1986), for the proposition that testimony by co-defendants is "inherently] unreliable].” See Maj. Op. at 1135 n. 4. This citation is misleading both because Lee involved application of hearsay law, which is not implicated by Salyer and Woolhiser’s potential testimony, and because it relied on Ohio v. Roberts, 448 U.S. 56, 66, 100 S.Ct. 2531, 65 L.Ed.2d 597 (1980), which is no longer good law. See Crawford v. Washington, 541 U.S. 36, 124 S.Ct. 1354, 158 L.Ed.2d 177 (2004).
.The majority attempts to conceal its mis-characterization of Fulminante, assuring its readers that it does not adopt a per se rule that improperly admitted confessions are prejudicial. Maj. Op. at 1146 n. 20. That is a remarkably contradictory claim to make, as footnote 20 of the majority’s opinion is appended to a sentence that says, “Fulminante stands for the proposition that the admission of an additional confession ordinarily ... is therefore prejudicial." Maj. Op. at 1146. If that sentence does not adopt a per se rule, it comes dangerously close.
. The majority’s accusation that, under my logic, Fulminante "would have come out the opposite way,” Maj. Op. at 1149, is thus flatly incorrect.
. On the contrary, the confessions in this case bear far more resemblance to the confessions in Milton v. Wainwright, 407 U.S. 371, 92 S.Ct. 2174, 33 L.Ed.2d 1 (1972), the case that the Fulminante Court cited as an example of a case where confessions were made independently of each other. The challenged confession in Milton was made to a police officer who was posing as a fellow murder suspect incarcerated with the defendant. The defendant had previously made three separate confessions that were not related to the confession he made in prison, other than in the sense that the confessions all corroborated each other. Id. at 372-73, 92 S.Ct. 2174. To the extent that the majority distinguishes Milton because Moore's confession at issue was his only formal confession, it misses my point. See Maj. Op. at 1147 n. 21. The Fulminante Court held that admission of the confession was prejudicial, even though a second confession was also admitted, because the credibility of each confession was strongly related and each could easily be attacked independently. This was not the case in Milton, and, accordingly, the Court did not find prejudice; similarly, it is not the case here.
. The majority claims that Anderson v. Terhune, 516 F.3d 781 (9th Cir.2008) (en banc), supports its remarkable reading of Fulmi-nante. See Maj. Op. at 1149-50. It cites Anderson for the proposition that a habeas court applying Fulminante need "not ... consider the other evidence the state had presented to tie the defendant to the crime, or whether the confession would have repeated such evidence.” Maj. Op. at 1149-50. That is an inappropriate reading of Anderson, which simply did not discuss any of the additional evidence that was present before the state court. Its entire discussion of prejudice is basically a single sentence: "The confession was central to the conviction.” Anderson, 516 F.3d at 792. Anderson provides no support for the majority’s broad reading of Ful-minante. If the majority's reading of Anderson and Fulminante is correct, it would swallow the Fulminante rule that harmless error analysis applies to erroneously admitted confessions.
. The majority’s implicit reading of Hill through its interpretation of Howard also completely ignores the plethora of cases— both pre-and post -Howard — where we have adopted the more limited construction of Hitt that I propound. See, e.g., Weaver v. Palmateer, 455 F.3d 958, 966-67 (9th Cir.2006) ("In assessing prejudice, we do not ask what a defendant might have done had he benefited from clairvoyant counsel.... Instead, we consider whether there is a reasonable probability that Weaver would have proceeded to trial had he been given proper advice from counsel. See Hill, 474 U.S. at 60, 106 S.Ct. 366.... [T]he prejudice requirement of Strickland and Hill requires Weaver to show that but for [counsel’s] alleged errors Weaver would have rejected the State’s plea offer and insisted on facing trial....”); United States v. Davis, 428 F.3d 802, 806 (9th Cir.2005) (”[D]efendant satisfied the prejudice prong of Hill ... only if he proved that, but for his counsel's erroneous advice, he would have insisted on going to trial.”); Washington v. Lampert, 422 F.3d 864, 873 (9th Cir.2005) ("[TIo prevail [under Hill ], the petitioner must show that ... 'there is a reasonable probability that, but for counsel's errors, he would not have pleaded guilty and would have insisted on going to trial.’ ... Washington points to no evidence indicating that, had his attorneys performed differently, he would have rejected the sentencing stipulation and insisted on proceeding with the sentencing phase of trial.” (quoting Hill, 474 U.S. at 58-59, 106 S.Ct. 366)); Langford, 110 F.3d at 1389 ("The state trial court applied the proper test under the Sixth Amendment. It invoked Hitt and Strickland and ruled that Langford 'would not have entered a not guilty plea even if his trial counsel [had not made the errors the defendant alleges].’
... The trial court properly applied the federal standard.”); Doganiere v. United States, 914 F.2d 165, 168 (9th Cir.1990) ("Even if his attorney’s performance was deficient ..., Do-ganiere's defense was not prejudiced. To show prejudice [under Hitt ], Doganiere must show that 'there is a reasonable probability that, but for counsel's errors, he would not have pleaded guilty and would have insisted on going to trial.’ Doganiere does not allege that had his attorney [not been deficient], he would have pleaded not guilty and insisted on going to trial.” (quoting Hill, 474 U.S. at 59, 106 S.Ct. 366)).
. In fact, it is not so obvious that the Howard rule makes very much sense, because it is essentially a truism: If counsel had filed a *1190potentially’ meritorious motion, the state's case would have been at least marginally weaker, and the defendant might have obtained a plea agreement that was not quite as "harsh.” The majority's reading reopens every plea agreement to a possible ineffective assistance of counsel claim if counsel had a potentially meritorious motion to file.
. Although the current case does not involve a direct extension of the judiciary’s constitutional powers vis-a-vis coequal branches of the federal government because this case is before us on review of a state court’s judgment, Judge Berzon’s novel prejudice analysis, if adopted, also would apply to review of federal convictions, either under 28 U.S.C. § 2255 or on direct review.
. I note that this problem is avoided under the Hill standard of prejudice: If the question is whether the defendant would have insisted on going to trial instead of taking the plea, that question is amenable to a reasoned answer by looking at the evidence presented by trial counsel and the defendant on habeas review about the defendant’s discussions and choices. The Hill prejudice standard depends not at all on the prosecutor's considerations.
. See, e.g., Gilbert v. Merchant, 488 F.3d 780, 791 (7th Cir.2007) (“[G]iven that he was convicted based on his own plea, Gilbert was obliged to complete the demonstration of prejudice by showing that had his confession been suppressed, it is reasonably likely that he would have gone to trial rather than plead guilty.” (citing Hill, 474 U.S. at 59, 106 S.Ct. 366)); United States v. Salazar, 323 F.3d 852, 857 (10th Cir.2003) (applying Hill to determine if defendant who pled guilty "should be allowed to pursue his suppression claim”); Langford, 110 F.3d at 1386-88; Banks v. Hanks, 41 F.3d 1187, 1189 (7th Cir.1994); Hale v. Lockhart, 903 F.2d 545, 548-50 (8th Cir.1990); United States v. Carasis, 863 F.2d 615, 616 (8th Cir.1988) ("[B]ecause Davis’ lawyer did not unreasonably forego filing a suppression motion on behalf of his client, we cannot say Davis has shown 'a reasonable probability that, but for counsel's errors, he would not have pleaded guilty and would have insisted on going to trial.’ ” (quoting Hill, 474 U.S. at 59, 106 S.Ct. 366)); Hibbert v. Poole, 415 F.Supp.2d 225, 232-33 (W.D.N.Y.2006) (applying Hill to prejudice determination when counsel advised defendant to plead guilty before court ruled on suppression motion); United States v. Hawkins, 973 F.Supp. 825, 828 (S.D.Ill.1997) ("Even if counsel’s failure to file a motion ... to suppress the evidence was unreasonable, defendant has not ... alleged that he would have pled differently had the motions been filed, and had they later succeeded.”); Friedman v. Gamble, 919 F.Supp. 1440, 1447 (D.Mont.1995).