Marcrum v. Luebbers

The Superintendent of the Potosí Correctional Center, A1 Luebbers, appeals from the district court’s grant of a writ of habeas corpus to petitioner Richard Louis Marcrum. The district court granted the writ on the ground that Marcrum’s Sixth Amendment rights were violated by ineffective assistance of counsel at his trial for murder and armed criminal action in connection with the 1994 killing of Kenneth Reeves. The district court held that trial counsel’s failure to introduce witnesses and medical records establishing that Marcrum was psychotic on the day of the killing and to cross-examine the prosecution’s expert fell below the level of legal representation to which Marcrum was entitled under the Sixth Amendment and that there was a reasonable probability that the result of Marcrum’s trial would have been different without counsel’s errors. The Superintendent contends that trial counsel’s actions regarding the witnesses and records and his decision not to cross-examine the expert were not errors but strategic decisions; that there is no reasonable probability these actions affected the trial’s result; and that in any case, the state courts’ resolution of these questions was not so unreasonable as to warrant federal habeas relief. He also contends that Marcrum’s petition was time-barred. We reverse.

I.

There is overwhelming evidence that on June 3, 1994, Marcrum took a fireplace poker and killed a Presbyterian minister named Kenneth Reeves. The state’s theory was that Marcrum had been blackmailing Reeves and killed Reeves when Reeves balked at giving him money. Marcrum’s theory was that he was at Reeves’s house because they were lovers, and he denied killing Reeves. At the same time, Marc-rum also raised the defense that he was insane because of a seizure disorder that rendered him psychotic on that day. The district court held that Marcrum’s trial counsel was ineffective in presenting his insanity and diminished capacity defenses.

Because the legal issues in this case depend on the difference between what facts were known to the jury at Marcrum’s trial, to Marcrum’s trial counsel, and to the state court hearing Marcrum’s motion for postconviction relief, we must tell this story in layers. We begin with the evidence at Marcrum’s trial.

A. The trial

Around 2:00 to 3:00 in the afternoon of June 3, 1994, Gary and Donna Paszkiewicz drove by the home of their neighbors Kenneth and Katie Reeves in Imperial, Missouri. They saw the petitioner, Richard Marcrum, standing in the road near a blue-gray car, and as they passed him, he seemed to want to talk to them. Gary Paszkiewicz rolled down the window of his truck and Marcrum said to him, “Praise the Lord, I just killed one sorry son-of-a-bitch,” and “I’m going to kill another.” Discomfited by this exchange, the Pasz-kiewiczes drove on, but they soon returned to check on their neighbors. They found Kenneth Reeves lying on the floor by his wheelchair in a pool of blood. His skull had been crushed, and a bloody fireplace poker was on the floor. Reeves died of his injuries two days later.

When police contacted Reeves’s wife, Katie, she tipped them to investigate Marcrum. Police Captain Edward Kemp arrived at Marcrum’s house to find that he *492 had been taken to the emergency room of St. Louis Regional Medical Center in an ambulance. Kemp found Marcrum on a gurney in an examining room. As soon as Marcrum caught sight of Kemp, he said, “I know why you’re here. It’s because of George.... I killed George in Kimmswick. He also goes by the name of Kenny Reeves.” Marcrum said that Reeves was “evil.” Kemp saw blood splatters on Marcrum’s clothes. Those clothes were later introduced as exhibits at trial, and DNA samples taken from blood on the clothing were identified as being Kenneth Reeves’s DNA.

Katie Reeves testified that Reeves was a Presbyterian minister who often helped the poor using the Reeveses’ own money. Reeves had been paralyzed after falling from a tree and was a paraplegic. She said Reeves had employed Marcrum in 1987 or 1988 to refinish some furniture; Reeves had paid Marcrum the promised amount, but Marcrum never finished the job. Nevertheless, Marcrum had telephoned the couple demanding more money. The jury heard a tape of telephone conversations between Marcrum and Katie Reeves in which Marcrum demanded money and made threats to expose Reeves as a homosexual if they didn’t pay him. Katie Reeves also testified that after her husband’s death, she found that he had drawn checks on their checking account that were not reflected in the register, though the balance was adjusted surreptitiously to compensate for the amounts drawn. Katie Reeves was a schoolteacher, and she testified that in the year or so before his death, Reeves would call her at school before she left in the afternoon just to find out if she was still at school. She testified that after Reeves’s death, she learned that Reeves had co-signed an auto loan with Marcrum shortly before he was killed.

A teller from the local bank testified that Kenneth Reeves came through the drive-through often during the winter and spring of 1994 cashing checks for hundreds of dollars at a time. She said that Marc-rum brought in checks signed by Reeves “more than several times” and that her supervisor had checked with Reeves and had given her permission to cash the checks for Marcrum.

Marcrum testified in his own defense. He described a ten-year sexual relationship with Kenneth Reeves that began when Marcrum was about twenty years old and Reeves picked him up at Tower Grove Park in St. Louis. Marcrum testified that he had trysts with Reeves as often as two or three times a week and that Reeves had given him an estimated $90,000 over the course of their relationship. He said that Reeves had bought him at least four cars over the years.

Marcrum testified that he had suffered from seizures since the age of 16, when he went through a windshield in a car wreck. He said he had been hospitalized “a time or two” for psychiatric problems.

Marcrum said that he did not remember going to Reeves’s house the day of the killing, and in fact did not remember that day at all. He was roundly impeached with prior inconsistent statements he had made that he remembered having drinks with Reeves at Reeves’s house that day, and that he had a seizure, woke up to find Reeves in a pool of blood, and fled the house. The prosecutor cross-examined Marcrum about the effects of a seizure and Marcrum conceded that when he has a seizure he is “basically helpless” — he loses control of his bladder and bowels and would not be able to drive a car. The prosecutor asked, “Your seizures and after effects are passive, of a nonviolent nature?” and Marcrum answered, “That’s what I’ve been told, yes.”

*493 Beginning with Marcrum’s mother, Marcrum’s lawyer put on a number of witnesses and asked them whether they saw blood on Marcrum on June 3. Marc-rum’s mother, father, and brother all said they did not see blood on him, despite having checked him when he came into the house. Judy and Don Paszkiewicz, the Reeveses’ neighbors, both said they did not notice any blood on him, either.

Marcrum’s counsel introduced testimony of a car dealer who had sold Kenneth Reeves a gray-blue Buick shortly before the killing. Reeves paid $500 down for the car and co-signed an installment contract for the remainder of the purchase price with Marcrum. Marcrum’s father, mother, brother, and Mend Judith Quick all testified that they saw Reeves pick Marc-rum up many times. They said Reeves would routinely pick Marcrum up around 8:30 in the morning and bring him home about 1:00 in the afternoon. Marcrum’s former girlfriend, Marilyn McManus, testified that when Marcrum lived with her, Reeves used to come by and bring Marc-rum money, or else she and Marcrum would go to Reeves’s house or church to pick up money. McManus said that Marc-rum did not have to work because Reeves gave him money. Marcrum’s mother said that every time Marcrum left with Reeves, he would come home with several hundred dollars. Judith Quick said she saw Marc-rum with checks from Reeves for hundreds of dollars.

Marcrum’s mother, father, brother, and Marilyn McManus and Judith Quick, all testified about Marcrum’s seizures and about Marcrum’s bizarre behavior, which they associated with the seizures. Each of these lay witnesses had his or her own interpretation of Marcrum’s behavior. Marilyn McManus had her own elaborate typology, classifying Marcrum’s seizures into types, ranging from mild (in which he was disoriented) to moderate (in which he would jerk, wet himself, vomit, and foam at the mouth) to severe (in which he would make sounds like a rabid dog and sleep for a week afterward, not even waking up when he relieved himself in the bed). Mc-Manus described an “aftermath” of the seizures when “Rickie would say he was God,, he was Jesus, he’s taking the children to his kingdom.” McManus said the aftermath could last as long as three or four days, and when Marcrum woke up, he would not remember anything. She later broke the “aftermath” into two stages: first Marcrum was “God and Jehovah and Jesus,” but then he would be “very nice” and “like an angel” and would clean the house obsessively. Later, however, she described an occasion when he twisted her arm when he was in an “aftermath.” The witnesses said that the bizarre behavior could happen before a seizure, but they also said that after seizures, Marcrum would often think he was God and would rant about being God and taking people with him to heaven. Judith Quick described Marcrum pounding on people’s cars in the grocery store parking lot, saying he was God. Sometimes he would think people were trying to kill him.

Various family and friends described taking Marcrum to the hospital during these episodes, where he would be medicated and released, according to his mother, because he did not have health insurance. McManus said his condition got worse in 1993, so that by the time of their last four months together (in 1993), Marc-rum could have six or seven seizures in a day. Their relationship ended on December 27, 1993, when he twisted her arm during one of his “aftermaths” and he was arrested.

Marcrum also presented testimony from a sheriffs deputy who arrested Marcrum for assaulting McManus on that day. *494 When he responded to the call, Marcrum was standing in the driveway saying he was God and that he had molested and abducted children. Marcrum said, “I am the promise and the children come unto me to the palace and they do construction work.” Marcrum resisted arrest, and it took three troopers to subdue him.

Marcrum’s family testified that Marc-rum had been suffering from seizures the last few days before Reeves was killed. His brother was in the room with him the night before and heard him making weird sounds and saw him shaking in his sleep. The morning of June 3, Marcrum accused his mother of offering him poisoned coffee. He left the house that morning and returned about 5 p.m. When he returned, he was calling himself God, “screaming at the top of his lungs about his kingdom and how he’s God and quoting from the Bible, which was nothing but one of those free real estate magazines you pick up at the store.” His brother took away his car keys and called the ambulance.

The ambulance driver testified that he thought Marcrum was intoxicated because the family told him Marcrum had been drinking and because he had slurred speech and a staggering gait. The driver said Marcrum did not seem to have had a seizure because he was “up and walking around.” Marcrum’s counsel introduced the results of a blood test from that night that showed Marcrum’s blood alcohol content on June 3 was from 0 to 10 milligrams per deciliter, but there was no evidence about the significance of those findings. In closing argument, counsel said of those test results, “I don’t know what it means,” but then argued it meant Marcrum was not drunk. By way of comparison, the legal blood alcohol limit for driving in Missouri is eight times higher than 10 milligrams per deciliter, see Mo.Rev.Stat. § 577.012 (0.08% blood alcohol by weight excessive) and Jarrett v. Woodward Bros., 751 A.2d 972, 976 n. 4 (D.C.Ct.App.2000) (100 mg/dL equivalent to 0.1% alcohol in blood by weight). Of course, the test did not distinguish between amounts of 0 to 10 milligrams, so Marcrum may have had no alcohol in his blood at all.

Throughout the trial there had been some evidence that Marcrum had a history of alcohol and other substance abuse. Probably the most damaging testimony of this sort came from Marilyn McManus, who testified on questioning by Marcrum’s lawyer:

Q: Sometimes when he was in this fourth phase or type [of seizure], would he beat you?

A: Not during the seizure, no.

Q: I mean after the aftermath or before or somehow, would he beat you till you’re black and blue? ...

A: Not during the aftermath and the seizure, no....

A: When he was drinking, like I said, he did drink. The times that he did drink he has beaten me while he has been drunk, he has....

A: There was times that he has drank and he has hit me and he has not had a seizure.

McManus also said that the alcohol abuse often precipitated seizures.

The last evidence at trial was the expert testimony on the subject of Marcrum’s mental state.

Marcrum called Dr. Allan G. Barclay, a Ph.D. in clinical psychology. Dr. Barclay opined that Marcrum suffered from a mental disease or defect, specifically that “he suffers from an organic personality disorder subsequent to the seizure disorder and the history of substance abuse in the past and the trauma [from the car wreck].” He testified that a personality disorder is an enduring characteristic of a person, where *495 as delusions and psychosis can come and go. When Marcrum’s counsel asked whether the mental disease or defect would have compelled Marcrum to commit the killing, Barclay answered, “Yes.” Similarly, Barclay said that the seizures could cause impulsive behavior and that this could preclude Marcrum from using a “logical, rational approach, the planning of any kind of activity.”

Barclay did not explain on direct examination what relationship might exist between Marcrum’s seizures and his delusions. The prosecutor touched on this connection during cross-examination, but only to make the point that the symptoms of the mental disease or defect came and went with the seizures, leaving Marcrum lucid and able to think rationally at other times. The prosecutor established that during the actual seizures and the post-ictal (post-seizure) period, Marcrum would be physically and mentally debilitated, so that he would not be able to drive a ear or coordinate his movements, or do any purposeful act of violence, such as picking up a poker and “precisely, accurately and deliberately [landing] four or five direct blows to someone’s skull.” In a telling moment, the prosecutor got Barclay to admit that he could not infer whether Marcrum was in possession of his faculties at the time of the killing:

Q: You don’t know whether on June 3, 1994 he was in the reality phase or in the loss of the reality phase, do you?

A: I can’t testify to that because I wasn’t present.

Further, the prosecutor’s questioning elicited testimony from Barclay tending to show a one-on-one relationship between seizures and “loss of reality”:

Q: The seizure is the trigger, isn’t it?

A: Yes....

Q: If the mental disease or defect non-reality stage is not triggered and he’s in the reality stage, ... it must be your opinion that the mental disease or defect would not cause impulse [sic] behavior; isn’t that true?

A: If I follow your line of reasoning, yes.

Barclay later agreed there was “no history of psychosis without seizure.” The prosecutor asked if Marcrum had ever been violent in the “post-ictal” stage, and Barclay answered that there was no record of that. Furthermore, the prosecutor established through Barclay that if Marcrum were in a “nonreality” state, he would not be able to remember what he had done on June 3, but his statements to the state’s psychiatrist, Dr. Parwatikar, indicated that he did remember many events that took place that day at the Reeveses’.

On redirect, Barclay stated that Marc-rum had “intermittent psychotic episode[s] from time to time associated with the organic brain damage and seizure disorder.” When reminded about Marcrum’s history of seizures the night before June 3, his accusation of poisoning that morning, his claim to be God that evening, Barclay agreed that it would be “consistent” with that history to conclude that Marcrum was psychotic at the time of the killing. Barclay also opined that a person in a psychotic state could drive a car and do other purposeful activities, such as hitting someone over the head.

The testimony of the state’s psychiatrist, Dr. Sam Parwatikar, was largely consistent with Barclay’s. In particular, Parwa-tikar testified that seizures make a person unconscious and physically debilitated, so that a person in a seizure could not attack someone and it was not probable that a person in a post-ictal stage could do so. Parwatikar expressly conflated the post-ictal state with psychosis, referring to Marcrum’s “post-ictal psychotic period,” *496 during which he said it would be difficult or impossible for Marcrum to drive a car. Parwatikar testified that there was no record of Marcrum ever being in a “non-reality” state without having suffered a seizure and that there was no record of Marcrum engaging in violent behavior while psychotic. Marcrum’s counsel did not cross-examine Parwatikar, except to establish that he was paid by the state.

The court instructed the jury on first-degree murder (murder with deliberation), second-degree murder, and armed criminal action. It instructed them that they could find that Marcrum was not guilty by reason of mental disease or defect if the greater weight of the credible evidence showed that at the time of the conduct he had a mental disease or defect that made him incapable of knowing and appreciating the nature, quality, or wrongfulness of his conduct. It also instructed them that they could consider evidence that Marcrum did or did not have a mental disease or defect in deciding whether Marcrum had the state of mind required to be guilty of first degree murder.

The prosecutor argued in closing that Marcrum killed Reeves out of “pure, simple, desperate greed,” as part of a blackmail scheme. He concentrated on the gruesomeness of the killing and Reeves’s helplessness and asked the jury to imagine being in Reeves’s position.

Marcrum’s defense was based on both the idea that he did not commit the killing and the idea that he was insane. Marc-rum’s lawyer argued that the Paszkiew-iczes and Marcrum’s family did not see blood on him after the killing, which east doubt on whether Marcrum was the killer. He never offered a theory as to how or when Reeves’s blood got on Marcrum’s clothes that were exhibited at trial. Counsel also argued that Marcrum was psychotic on June 3, 1994. He argued at length that Marcrum’s seizures could set off a psychotic episode that could go on for days.

On rebuttal, the prosecutor argued that if Marcrum had had a seizure at Reeves’s house, he would have been incapable of hitting Reeves and driving home, and that if there was no seizure at Reeves’s house, there could be no psychosis: “The seizure is the trigger, his own doctor said. So if he didn’t have a trigger-if he didn’t have a seizure then the psychosis would not manifest.” He argued, “The evidence in this case is that when he’s psychotic, he’s not violent.” Conversely, he said, “When he’s in his violent stage, he’s rational, he’s not psychotic.”

The jury found Marcrum guilty of first-degree murder and armed criminal action in connection with first-degree murder. He was sentenced to life in prison without the possibility of parole.

Marcrum appealed to the Missouri Court of Appeals, which affirmed in a summary decision. State v. Marcrum, 958 S.W.2d 96 (Mo.App.1997).

B. State Post-conviction Proceedings

Marcrum moved for post-conviction relief in the state courts under Missouri Supreme Court Rule 29.15, contending among other things, that his trial counsel was ineffective for failure to call medical witnesses, failure to introduce into evidence the records of his previous hospitalizations for psychiatric crises, and failure to cross-examine Dr. Parwatikar. He produced a new expert, Dr. William Logan, a forensic psychiatrist. Logan reviewed the same records that had been made available to Barclay, and he agreed that Marcrum has a seizure disorder which had developed into an organic personality disorder. However, Logan added one crucial idea: that when Marcrum was having uncon *497 trolled seizures-that is, when he was not being medicated with anti-epilepsy drugs like Dilantin-he would develop organic psychosis. This was different from the personality disorder diagnosed by Barclay and Parwatikar: “[A] person with an organic personality doesn’t have hallucinations and doesn’t develop bizarre delusional beliefs. A person with organic psychosis does.” Logan found Marcrum’s medical records showed he had a record of failing to take the anti-epilepsy medicine necessary to control his seizures.

Logan opined that, contrary to the trial testimony, there was not a simple one-on-one relation between seizures and psychotic episodes:

The impression left from the testimony was that he would have a seizure and then that immediately would trigger a psychotic episode. It doesn’t happen that way.

What happens is that a person goes through a period where there are a number of seizures and their epilepsy is uncontrolled. When that happens, psychotic features begin to emerge, but there’s no direct triggering effect....

Q: So is it possible for a seizure to trigger an episode of organic psychosis?

A: Not usually one, it would usually have to be a number of them in fairly close succession.... [T]he more usual history is that a person goes through a period where they get off their medication, they have a number of seizures, then they begin to have some odd delusional beliefs.

Later, he explained:

That’s really one of the key problems I saw in the testimony [of the experts at trial] is this was not behavior triggered by a seizure [or] occurring in the immediate aftermath of a seizure. This was in fact, behavior that occurred as a result of a number of seizures which produced a psychotic state which is something quite different.

Logan used the medical records which had been made available to Barclay to substantiate his theses that (1) Marcrum had in the past become psychotic after uncontrolled seizures, which in turn resulted from failure to take his anti-epileptic medicines; (2) the psychosis lasted far beyond the post-ictal period for particular seizures, and in fact did not resolve until Marcrum was back on the anti-epilepsy medicines; and (3) when psychotic, Marc-rum became violent and subject to religious delusions. Logan furthermore used testimony and medical records from the day of the killing to show Marcrum had not been taking his medicine for several days before the killing and was demonstrably psychotic on the morning of June 3 when he accused his mother of poisoning him, at mid-afternoon when he told the Pakiewiczes, “Praise the Lord, I just killed one sorry son-of-a bitch,” and on the evening of June 3 when he claimed to be God. Logan pointed to other incidents documented in the medical records where Marcrum was violent and psychotic, but was not “in the middle of a seizure.”

Logan opined that Marcrum was psychotic at the time of the killing and that he could not appreciate the nature, consequences, and wrongfulness of his actions.

The post-conviction proceeding moved on to explore why Marcrum’s trial lawyer, Alfred Speer, had not called medical witnesses from Marcrum’s past hospitalizations, introduced the records from those crises, or cross-examined Dr. Parwatikar. By way of background, the murder trial transcript showed that on June 6,1996, the third day of trial, Speer first endorsed Denise Hacker and Dr. Kim Young, from the Southeast Missouri Mental Health Center, where Marcrum had been treated *498 for a psychotic episode in December 1993. Also on June 6, Speer endorsed the two ambulance workers who took Marcrum to the hospital the night of the killing. The state objected to the late endorsement of these witnesses, and at the hearing on the state’s objection, Speer further endorsed Drs. Viamontes and Gedden, who treated Marcrum in the emergency room on the night of the killing. The trial court observed that the case had been filed over two years ago and that the defense had had ample opportunity for discovery, but had failed to comply with the court’s scheduling order. The court mediated the situation by allowing Speer to call the ambulance workers, but excluding the doctors, ruling, however, that the defense could use the medical records those doctors would have introduced, since the defense could bring out those prior hospitalizations through either its own or the state’s expert. However, as it happened Speer did not introduce the records in examining either expert.

Speer testified at the post-conviction hearing about why he had not introduced the medical records from the past psychiatric crises or from June 3 and why he had not cross-examined the state’s expert Par-watikar. He said that “those [records] that were necessary to arrive at the medical opinion were introduced. Beyond that, I saw no need for it.” He qualified that to say that “there were some medical records I thought at one point we wanted to get in and because they were late arriving, the Court declined to let them in. I’m not so sure they would have been at all helpful.” The state’s lawyer then asked whether Speer had “some concerns about the length of trial,” and Speer said he did generally. Then he said he did not like to introduce cumulative evidence and that in this case there was plenty of “paper work” without the medical records. He also stated that he sought to avoid introducing evidence of Marcrum’s history of pedophilia and of violence, which would have been revealed by the medical records. As to his failure to cross-examine Parwatikar, Speer said:

Again, it gets down to what I consider, and this is a tactical judgment, Dr. Sam projects a very professional image. He’s a tall, good-looking man. He has a very professional demeanor. He’s commanding in his appearance. He was followed in his presentation and he is eminently skilled at courtroom presentations for the State. And to attack him I think would be fruitless and would most likely backfire.

The medical records that were introduced at the post-conviction hearing showed a pattern of emergency room visits by Marcrum, sometimes followed by longer hospitalizations, increasing in frequency from 1989 to 1993, always showing that Marcrum had not taken his Dilantin and had suffered seizures. The records show three major episodes in 1993. Many of these records showed bizarre, sometimes violent, behavior in conjunction with seizures and low medication levels. For instance:

(1) On January 20, 1991, Marcrum was running naked in the streets, was combative and agitated, had seizures and reported that he had stopped taking his Dilantin a year ago.

(2) On April 12, 1992, Marcrum was kicking cars and chasing an elderly woman and child down the street. He stated that he was talking to the radio about God. His Dilantin level was too low, and he reported a two-month history of non-compliance with taking his medicine. He was diagnosed as psychotic, but the psychosis cleared up when he was treated with a therapeutic level of Dilantin.

*499 (8) On December 9, 1992, Marcrum was admitted after a seizure, with a sub-therapeutic Dilantin level. He was diagnosed with “schizophrenia, paranoid.”

(4) On April 4, 1993, Marcrum was “acutely psychotic,” having had seizures the past two days, with a “subtherapeutic anticonvulsant level.”

(5) On November 16, 1993, he was admitted to the Southeast Missouri Mental Health Center after he “began striking and beating another individual” “without provocation,” breaking the windshield out of a car. He was described as “violent, combative, and scream[ing].” His Dilantin level was subtherapeutic. The examining psychiatrist described his “grandiose ideations” in which he claimed to have “a tremendous amount of power that he gets from Jesus with whom he maintains a good communication.”

(6) On December 27, 1993, Marcrum was taken to the emergency room after being picked up for domestic violence. He was “combative and delusional.” He was quoted as saying, “I killed the kids, I’m hitching to the lord and I’m taking a bunch with me, I don’t kill unless the lord tells me he needs them.” He was reported to have not been taking his medication for the past week.

(7) On March 12, 1994, Marcrum was taken to the emergency room for a seizure; he was combative and his Dilantin level was sub-therapeutic.

The medical records from the night after the killing, also introduced at the post-conviction hearing, showed that Marcrum was brought to the hospital for bizarre behavior. When he arrived, he was whispering, “I am the chosen one.” A blood test for phenobarbital, an anti-convulsant, showed the level was far below therapeutic. The doctor wrote that Marcrum was in “florid psychosis” and ordered that he be given 500 mg. Dilantin “NOW.”

The state trial court considering Marc-rum’s post-conviction proceeding rejected Marcrum’s argument that his lawyer had ineffectively presented his mental disease defense. Marcrum v. State of Missouri, No. CV198-1317-CC-JI (Mo.Cir.Ct. May 18, 2000). The court held that counsel had acted reasonably in employing Dr. Barclay, based on Barclay’s credentials and performance as a witness in past cases. Since Barclay’s testimony “went to establishing the defense of not guilty by mental disease or defect and that of diminished capacity,” counsel was not obliged to shop for another doctor, even if the second doctor would reach that conclusion by “different diagnosis and different nomenclature.” The court further accepted Speer’s testimony that the medical records would have been cumulative and repetitive and would have lengthened the trial to Marcrum’s detriment. The court rejected the argument that Speer should have cross-examined Parwatikar, concluding that such cross-examination “would [not] have presented a viable defense,” that Marcrum failed to demonstrate what Parwatikar would have said on cross-examination, and that Speer had valid reasons not to prolong the trial by cross-examining the state’s expert.

Marcrum appealed the denial of post-conviction relief to the Missouri Court of Appeals, which also rejected the claim that counsel failed to present the mental disease or defect defenses adequately. Marcrum v. State of Missouri, 50 S.W.3d 824, 825 (Mo.App.2001). The Court of Appeals listed three grounds for rejecting Marc-rum’s claim that Speer failed to introduce information from his medical records through his expert. First, the Court of Appeals said that Dr. Barclay’s testimony at trial had actually covered the facts that *500 Marcrum did not take his medicine, that he had frequent seizures, 2 that he was admitted to the hospital the night of the killing, and that he accused his mother of poisoning him. The Court of Appeals concluded that the omitted records would not add to what was already before the jury. Second, according to the court, there were legitimate strategic reasons for declining to put in certain evidence relevant to Marcrum’s psychosis, such as his violence, his alcohol abuse and his belief that Reeves was evil, since these facts could have prejudiced the jury against Marcrum. Id. at 825. Third, some of the facts that Marcrum contended should have been brought out from the medical records at trial were not simply the facts from the records, but involved an interpretation of those facts that was different from Barclay’s interpretation. The Missouri Court of Appeals held that Marcrum could not impeach his own expert “under the guise of a claim of ineffective assistance of counsel.” Id.

Even assuming that any of the actions Marcrum complained about constituted deficient performance, the Court of Appeals held that there was substantial evidence that Marcrum was sane at the time of the killing and that therefore any deficiency did not prejudice Marcrum. Id. at 825.

The Court of Appeals also held that it was not unreasonable for Speer not to cross-examine Parwatikar because the assertions Parwatikar made on direct examination reflected Parwatikar’s professional opinion. Id. at 825. Implicitly, the court doubted that Parwatikar would have conceded that there was anything doubtful or incorrect about his testimony on direct. Moreover, the Court held that cross-examination of Parwatikar could have “backfired” because it might have made Parwa-tikar’s testimony even more convincing to the jury. Id.

C. Habeas proceedings in the district court.

Marcrum filed this habeas proceeding, contending that Speer’s assistance was ineffective for failure to introduce Marc-rum’s “voluminous records of mental illness,” and in particular, the records from the night of June 3 showing his anticonvul-sant level was far below therapeutic range and he was diagnosed as suffering from “florid psychosis.”

The district court held it was not a reasonable strategy for counsel to fail to introduce these records on the ground that they were cumulative and would bore the jury. Marcrum v. Luebbers, No. 4:02CV01167 AGF, slip op. at 35 (E.D.Mo. Sept.30, 2005). Moreover, trial counsel appeared to think he had introduced at least some of the records, when in fact he had not introduced them. Id. at 36. Specifically, the district court held that the evidence in the records about the connection between Marcrum’s past psychotic episodes and violent behavior was crucial to his case, especially because both experts at trial and Marcrum himself had said that Marcrum was not violent during his psychotic episodes. Further, the evidence showing that Marcrum became psychotic when his anti-convulsant level was too low was also crucial. The district court rejected the idea that testimony about Marc-rum’s past psychiatric episodes from his friends and family was an acceptable substitute for evidence taken from hospital records. The district court also held that it was unreasonable not to cross-examine *501 Dr. Parwatikar when he said that Marc-rum’s records showed no history of violent conduct in connection with Marcrum’s psychiatric crises, whereas the medical records showed that there was such a connection. Id. at 37.

The district court also held unreasonable the state courts’ conclusions that there was no prejudice to Marcrum from his counsel’s failure to introduce the medical records and to impeach Parwatikar with the inconsistency between the facts disclosed in the records and his opinion that there was no connection between Marcrum’s psychosis and his violent behavior. The district court reasoned that the state courts relied on record evidence tending to show that Marcrum was sane during the killing, but that this evidence “came solely” from Parwatikar and would have been significantly undercut by introducing the records and demonstrating that Parwatikar’s testimony was inconsistent with them. Id. The district court conducted its own prejudice analysis and found that, while evidence that Marcrum committed the killing was overwhelming, evidence that he was sane at the time and able to deliberate was not. The evidence that he was sane and able to deliberate was mostly from the tape demanding money, which was made seven years before the killing and was therefore not very probative of Marcrum’s sanity or ability to deliberate on the day at the time of the killing. Id.

Having concluded that Marcrum received ineffective assistance of counsel in connection with presenting his insanity and diminished capacity defenses and that the state courts’ conclusions to the contrary were objectively unreasonable, the district court granted Marcrum’s habeas petition. Id. at 40. The court stayed its order pending this appeal.

Before moving to the merits of this case, we observe that the Superintendent contends that this petition is barred by the one-year statute of limitations for habeas petitions, 28 U.S.C. § 2244(d)(1). The Superintendent contends that the statute of limitations began to run fifteen days after Marcrum’s direct appeal was decided by the Missouri Court of Appeals, whereas the district court counted the time beginning 90 days after the Missouri Court of Appeals’ decision. Marcrum v. Luebbers, No. 4:02CV01167 AGF, slip op. at 3 (E.D.Mo. Sept.11, 2003). The case of Riddle v. Kemna, No. 06-2542, which involves the same question concerning the running of the statute of limitations, is now pending before our court en banc. In view of our disposition of the merits of this case, we need not resolve the question of whether the petition was timely filed.

II.

In habeas corpus proceedings, we review the district court’s findings of fact for clear error and its conclusions of law de novo. Garcia v. Bertsch, 470 F.3d 748, 752 (8th Cir.2006), cert. denied, — U.S. —, 127 S.Ct. 2937, 168 L.Ed.2d 267 (2007). Ineffectiveness of counsel claims present mixed questions of law and fact, which we review de novo. Id. at 754. That said, in a habeas case alleging ineffective assistance of counsel, we are bound to view what happened at trial through two filters, the first requiring us to defer to judgments of trial counsel, see Strickland v. Washington, 466 U.S. 668, 689, 104 S.Ct. 2052, 80 L.Ed.2d 674 (1984), and the second requiring us to defer to the state courts’ application of federal law to the facts of the case, see Bell v. Cone, 535 U.S. 685, 698-99, 122 S.Ct. 1843, 152 L.Ed.2d 914 (2002). Taking into account the leeway given to counsel under the Strickland standard and that given to the state courts under 28 U.S.C. § 2254(d), we conclude *502 that Marcrum did not show he was entitled to the writ of habeas corpus.

A.

The Sixth Amendment guarantees the right of the accused in criminal prosecutions to “the Assistance of Counsel for his defence.” U.S. Const. amend. VI. “[T]he right to counsel is the right to effective assistance of counsel.” Kimmelman v. Morrison, 477 U.S. 365, 377, 106 S.Ct. 2574, 91 L.Ed.2d 305 (1986). Effective assistance is representation that “play[s] the role necessary to ensure that the trial is fair.” Strickland, 466 U.S. at 685, 104 S.Ct. 2052. To show a constitutional violation of the right to counsel a convicted defendant must show first, that counsel’s performance was deficient, id. at 687, 104 S.Ct. 2052, and second, that counsel’s errors prejudiced the defense, id.

The test we apply for deficiency of performance is an objective standard of reasonableness. Id. at 688, 104 S.Ct. 2052. In Strickland, when the Supreme Court pronounced this standard, it expressly declined to dictate detailed rules for deciding reasonableness: “More specific guidelines are not appropriate.” Id. However, Strickland gave us several guides to decision: we must assess reasonableness on all the facts of the particular case, we must view the facts as they existed at the time of counsel’s conduct, and we must evaluate counsel’s performance with a view to whether counsel functioned to assure adversarial testing of the state’s case. Id. at 690, 104 S.Ct. 2052. Moreover, the reasonableness of counsel’s actions may depend on his client’s wishes and statements. Id. at 691, 104 S.Ct. 2052; see Schriro v. Landrigan, — U.S. —, 127 S.Ct. 1933, 1941, 167 L.Ed.2d 836 (2007) (client’s statement to court that he did not wish to present mitigating evidence supported state court finding of no prejudice from counsel’s failure to investigate such evidence). A court considering a defendant’s attack on his conviction must be “highly deferential” in assessing whether counsel’s course of conduct could be considered a sound trial strategy rather than an error, Strickland, 466 U.S. at 689, 104 S.Ct. 2052, and must “indulge a strong presumption that counsel’s conduct falls within the wide range of reasonable professional assistance,” id. In other words, the burden of proof is on the petitioner to show that “his attorney’s representation was unreasonable under prevailing professional norms and that the challenged action was not sound strategy.” Kimmelman, 477 U.S. at 384, 106 S.Ct. 2574.

The Supreme Court has held in several cases that the habeas court’s commission is not to invent strategic reasons or accept any strategy counsel could have followed, without regard to what actually happened; when a petitioner shows that counsel’s actions actually resulted from inattention or neglect, rather than reasoned judgment, the petitioner has rebutted the presumption of strategy, even if the government offers a possible strategic reason that could have, but did not, prompt counsel’s course of action. 3 Rompilla v. Beard, *503 545 U.S. 374, 395-96, 125 S.Ct. 2456, 162 L.Ed.2d 360 (2005) (O’Connor, J., concurring); Wiggins v. Smith, 539 U.S. 510, 526-27, 123 S.Ct. 2527, 156 L.Ed.2d 471 (2003); Kimmelman, 477 U.S. at 385, 106 S.Ct. 2574. For instance, in Kimmelman, the Supreme Court held that counsel’s performance was deficient when his failure to file a timely motion to suppress a bed sheet seized in a rape case was due not to strategic considerations, but to counsel’s ignorance that the state had the bed sheet and intended to introduce it. Counsel had failed to learn of the sheet because counsel mistakenly believed that the state would turn all incriminating evidence over without the need for counsel to conduct discovery. 477 U.S. at 384-85, 106 S.Ct. 2574. The warden’s arguments that the bed sheet did not turn out to be as important as other aspects of the case did not justify counsel’s failure to learn of the bed sheet or respond to it. The Supreme Court held that counsel’s decisions had to be evaluated as of the time they were made, and counsel who conducted no discovery could not have known what the relative importance of the different kinds of evidence would be. Id. at 386-87, 106 S.Ct. 2574. Similarly, in Wiggins, the Court held counsel’s actions were deficient, despite proffered strategic justifications: “[T]he ‘strategic decision’ the state courts and respondents all invoke to justify counsel’s limited pursuit of mitigating evidence resembles more a post hoc rationalization of counsel’s conduct than an accurate description of their deliberations prior to sentencing.” 539 U.S. at 526-27, 123 S.Ct. 2527.

If, viewed with appropriate deference, counsel’s performance was in fact deficient, the convicted defendant will only be entitled to relief if he shows there is a reasonable probability that, but for counsel’s errors, the result of the proceeding would have been different. Strickland, 466 U.S. at 694, 104 S.Ct. 2052. “A reasonable probability is a probability sufficient to undermine confidence in the outcome.” Id. The reviewing court must not consider the attorney error in isolation, but instead must assess how the error fits into the big picture of what happened at trial. Id. at 696, 104 S.Ct. 2052. “[A] verdict or conclusion only weakly supported by the record is more likely to have been affected by errors than one with overwhelming record support.” Id.

B.

The federal statute for habeas review of state convictions was overhauled in 1996, in the Antiterrorism and Effective Death Penalty Act, Pub.L. No. 104-132, Title I, § 104, 110 Stat. 1218, now universally known as AEDPA. An important effect of AEDPA was to modify the standard by which we review state courts’ earlier decisions in a case. Section 2254(d) provides:

An application for a writ of habeas corpus on behalf of a person in custody pursuant to the judgment of a State court shall not be granted with respect to any claim that was adjudicated on the merits in State court proceedings unless the adjudication of the claim—

(1) resulted in a decision that was contrary to, or involved an unreasonable application of, clearly established Federal law, as determined by the Supreme Court of the United States; or

(2) resulted in a decision that was based on an unreasonable determination of the facts in light of the evi *504 dence presented in the State court proceeding.

(emphasis added).

Section 2254(d)(1), which governs legal determinations, has two clauses, the “contrary to” clause and the “unreasonable application” clause. Because the state courts correctly identified the governing legal rules in Marcrum’s case, 4 only the “unreasonable application” clause (highlighted in the quotation above) concerns us here. “A run-of-the mill state-court decision applying the correct legal rule from [Supreme Court] cases to the facts of a prisoner’s case would not fit comfortably within § 2254(d)(l)’s ‘contrary to’ clause.” Williams v. Taylor, 529 U.S. 362, 406, 120 S.Ct. 1495, 146 L.Ed.2d 389 (2000). Such a decision could, however, involve an “unreasonable application” if the court identified the correct legal rule but unreasonably applied it to the facts of the case before it. Id. at 407-09, 120 S.Ct. 1495 (reserving question of whether the “unreasonable application” clause could also apply where state court unreasonably extends or fails to extend established legal principle to new context).

Although the Supreme Court has not found it necessary to refine the meaning of “unreasonable application,” see Yarborough v. Alvarado, 541 U.S. 652, 663-64, 124 S.Ct. 2140, 158 L.Ed.2d 938 (2004) (“The term unreasonable is a common term in the legal world, and accordingly, federal judges are familiar with its meaning.”) (internal quotation marks omitted), the Court has established that the standard does not require that all reasonable jurists would agree that the application was unreasonable, Williams, 529 U.S. at 409-10, 120 S.Ct. 1495, and that it requires something more than clear error, Lockyer v. Andrade, 538 U.S. 63, 75, 123 S.Ct. 1166, 155 L.Ed.2d 144 (2003). Unreasonableness is judged by an objective standard. Yarborough, 541 U.S. at 665, 124 S.Ct. 2140. The more general the applicable legal rule or principle, “the more leeway courts have in reaching outcomes in case-by-case determinations,” Yarborough, 541 U.S. at 664, 124 S.Ct. 2140, and consequently, the more difficult it is to say that the state court’s application of such a rule or principle is objectively unreasonable. As we discussed above, in Strickland, the Supreme Court expressly determined that the ineffective assistance standard should be left as a general principle, not a set of specific rules, 466 U.S. at 688, 104 S.Ct. 2052, which suggests that it would be rare for the Supreme Court to hold a state court’s application of Strickland to be unreasonable. See Rompilla, 545 U.S. at 381, 125 S.Ct. 2456 (“A standard of reasonableness ... spawns few hard-edged rules.... ”). However, the Supreme Court has recently made clear that “even a general standard may be applied in an unreasonable manner.” Panetti v. Quarterman, — U.S. —, 127 S.Ct. 2842, 2858, 168 L.Ed.2d 662 (2007).

Nevertheless, the Supreme Court has held on three occasions that state courts’ applications of Strickland were unreasonable under the AEDPA and that prisoners were entitled to the writ of habeas corpus because they received ineffective assistance of counsel. Rompilla, 545 U.S. at 380, 393, 125 S.Ct. 2456; Wiggins, 539 U.S. at 520, 538, 123 S.Ct. 2527; Williams, 529 U.S. at 399, 120 S.Ct. 1495; see John H. Blume, “AEDPA: The ‘Hype’ and the ‘Bite,’ ” 91 Cornell L.Rev. 259, 279-80 nn. 105-107 (2006) (noting that before the *505 AEDPA, Supreme Court never held counsel ineffective under Strickland, whereas Court has done so three times under the AEDPA). In each of these three cases, counsel had failed to present important aspects of the personal history of the defendant, and the failure had resulted from inadequate trial preparation. Rompilla, 545 U.S. at 389-92 (failure to look at file for prior conviction which showed parental abuse and alcoholism and would have pointed to defendant’s organic brain damage); Wiggins, 539 U.S. at 524-25, 123 S.Ct. 2527 (counsel failed to investigate or present evidence of childhood abuse); Williams, 529 U.S. at 395-97, 120 S.Ct. 1495 (counsel failed to uncover records describing defendant’s nightmarish childhood and borderline retardation). Though these cases were all death-sentence cases and Marerum’s is not, there is a good deal of similarity between the kind of omissions that led to relief in Williams, Wiggins, and Rompilla and those alleged by Marcrum. This string of cases shows that the AED-PA does not relieve us of the duty to scrutinize counsel’s omissions and the state courts’ assessments of counsel’s omissions.

HI.

A.

The evidence at the trial of this case revealed a serious possibility that on the day of the killing Marcrum was in the throes of a psychosis. The jury nevertheless rejected Marcrum’s defenses of insanity and diminished capacity. However, that jury never saw or heard about medical evidence that, with scrutiny and analysis, could have shown a well-established pattern in which Marcrum failed to take his anti-convulsants, suffered from serial seizures, and fell into a psychosis in which his behavior was paranoid and violent. Nor did the jury see or hear about medical records from the night of the killing diagnosing Marcrum as psychotic and showing a sub-therapeutic level of anticonvulsant in his blood. The record shows that counsel’s decisions about what medical records to introduce at trial resulted from neglect, not reasonable trial strategy. After scrutinizing the record from beginning to end, we conclude that introduction of the medical records themselves would not have been reasonably likely to have changed the jury’s mind, see Strickland, 466 U.S. at 696, 104 S.Ct. 2052 (asking whether decision would “reasonably likely” have been different absent attorney error). The records themselves could not have made the necessary causal connection between lack of anti-convulsants, serial seizures, organic psychosis, and violence. Only an expert who interpreted the records in a different way than the two experts who testified at trial could have made those records tell the story that we now have before us. Because the law is well-established that a lawyer who hires a qualified mental health expert is not ineffective in relying on that expert’s opinion unless the lawyer has reason to doubt the expert’s competence, 5 we cannot say that Speer’s decision to hire Barclay or to proceed to trial on the basis of Barclay’s interpretation of the medical record fell below the standard of performance dictated by the Sixth Amendment. In sum, as we explain below, we conclude that Speer’s decisions that may have fallen below the level of acceptable competence did not result in prejudice to Marcrum, and Speer’s decisions about what expert to present did not fall below the level of acceptable performance.

B.

Marcrum contends that Speer’s decision not to introduce the medical records *506 into evidence resulted from negligence, not from defensible strategy. Supreme Court precedent shows that counsel’s decisions resulting from failure to investigate are not entitled to the presumption of competent performance, whereas decisions about what to do with the results of investigation are strategic decisions that are virtually immune to second-guessing by habeas courts. E.g., Wiggins, 539 U.S. at 522-23, 527-28, 123 S.Ct. 2527 (decision avoided focusing on counsel’s decision of what evidence to present, which was strategic, but found deficient performance because of inadequate investigation, which shaped strategic decision).

The record here shows that well in advance of trial, Speer was in possession of the medical records that Marcrum now relies on. Speer said at trial that he had received a “large bundle of information” when he took over the case from the public defender. Speer supplied those records to Barclay, and Barclay reviewed the records in examining Marcrum and formulating his opinion about Marcrum’s state of mind at the time of the killing. Indeed, the record before us presents the medical records in sets, most of which are labeled: “Records of Alan [sic] Barclay, PhD from Missouri Baptist Hospital-Sullivan,” “Records of Alan [sic] Barclay, PhD from Social Security Administration, Rolla,” etc. Barclay testified at trial that he reviewed the medical records and Marcrum does not argue that the records Barclay had were incomplete. Therefore, whatever else Speer may have done wrong, this is not a case like the ones where the Supreme Court has found deficient performance because the lawyer failed to turn up the evidence in time to formulate a trial strategy. See Rompilla, 545 U.S. at 389-90, 125 S.Ct. 2456; Wiggins, 539 U.S. at 524-25, 123 S.Ct. 2527; Kimmelman, 477 U.S. at 385, 106 S.Ct. 2574.

What Speer did do wrong was to fail to timely disclose as witnesses the doctors and other health-care providers who had treated Marcrum in the past and who would have been able to testify about the contents of the records. As we discussed in the statement of facts, Part I B supra, Speer failed until well into the trial to name Dr. Kim Young and Denise Hacker, who could have testified about Marcrum’s records from severe episodes in November and December 1993, and Drs. Gedden and Viamontes, who treated Marcrum the night of the killing. The trial court excluded those witnesses because of Speer’s neglect to comply with the pretrial order, 6 but the court endeavored to limit the damage to Marcrum’s case by ruling that Speer could introduce the records through Barclay. However, at the post-conviction hearing, Speer said that he did not introduce the records because the court excluded them: “[T]here were some medical records I thought at one point we wanted to get in and because they were late arriving, the Court declined to allow them in.” Speer offered his opinion that the records would not have been “at all helpful,” but he did not say that this was his contemporaneous reason for failing to introduce them. The reason he gave for his failure to introduce the records was that the trial court excluded them, which though it may show that Speer misunderstood the details of the trial court’s ruling, also shows that the failure was the result of a situation caused by Speer’s neglect. Moreover, Speer also said that he had introduced some of the records, which he had not; this shows that the failure may have resulted in part from *507 still another mistake, and, again, it shows that the failure was not the result of strategy. Upon prompting he also said that he was against “cumulative” evidence, but he obviously did not discard as cumulative the records that he wrongly believed he had entered into evidence or those that he tried to enter, but which he believed had been excluded for lateness. Furthermore, the trial transcript shows that Speer regarded the excluded treating doctors as crucial, since he argued to the trial court: “We need these people. They’re essential to his defense of mental disease or defect.” The law is clear that we must judge Speer’s performance on facts as they appeared to him at the time of trial, not on whether he thinks in retrospect the action would have been advantageous or not. Kimmelman, 477 U.S. at 386-87, 106 S.Ct. 2574. Thus, the reasons offered by Speer and the state courts why not introducing the records could have been strategic are irrelevant. 7 We need not reach the question of whether the state courts’ reliance on such reasons was an unreasonable application of federal law, because we conclude that there was no prejudice from the failure to introduce the records.

C.

To obtain relief on the theory that Speer was ineffective for failing to introduce the medical records, Marcrum must show a reasonable probability that the failure to introduce the records affected the outcome of the trial. We have before us some seven hundred pages of medical records, much of it in illegible doctor’s handwriting, much in the form of laboratory reports, couched in medical terminology that would mean almost nothing to a jury of lay people. Even if the notations had been read to the jury by the treating doctors, there is no reason to think the jury could have digested them and discerned a causal connection between lack of medication, serial seizures, psychosis, and paranoia and violence. Clearly, it is not the failure to introduce the records themselves that would have changed (or had a reasonable probability of changing) the outcome of the trial. It would have required a specific use of the records to have an effect on the jury.

Even if Speer had drawn the jury’s attention to such notations in the record as “florid psychosis” on June 3, 1994, and the laboratory report showing a sub-therapeu-tie level of phenobarbital in Marcrum’s blood that night, we conclude that there is no reasonable probability the jury would have found these items of evidence determinative. 8 The testimony of both experts at trial led the jury to believe that the psychosis occurred only in the immediate aftermath of a seizure and that Marcrum did not have a seizure at Reeves’s house that day. Without an expert’s opinion that the psychosis resulted from a series of seizures, not from a particular one, and that the psychosis would not resolve until Marcrum was medicated with anti-convul-sants, the jury was not reasonably likely to piece together the medical records on its own to come to the conclusion that Marc- *508 rum’s psychosis that night proved he was psychotic during the afternoon.

Barclay’s testimony left the jury in doubt as to whether there was proof that Marcrum was psychotic on the afternoon of June 3, even assuming he was psychotic that morning and that night. On cross-examination, when the prosecutor asked whether Marcrum was “in the reality phase or in the loss of reality phase” on the day of the killing, Barclay said, “I can’t testify to that because I was not present.” On redirect, Barclay testified that he believed Marcrum was psychotic when he left home on June 3 and he was psychotic when he returned and that “it’s part of a long-running pattern, those repeated seizures.” Speer asked whether there was “a high degree of probability” that during the interim Marcrum was also psychotic, and Barclay answered, “I believe that would be consistent with a persistent psychotic state, yes.” This was not a definitive rebuttal of his earlier admission.

Moreover, Barclay acceded to the prosecutor’s insistence that there was a one-on-one relationship between seizures and psychotic episodes:

Q: The seizure is the trigger, isn’t it?

A: Yes, the seizure is an event, right. The prosecutor labored that point throughout the trial, and he drove it home on closing: “The seizure is the trigger, his own doctor said.” This would have led the jury to think that unless there was a seizure at Reeves’s house, which would have rendered Marcrum unable to wield a fireplace poker or drive home, there must have been no psychosis. In contrast, according to Dr. Logan, once the serial seizures leading up to June 3 had pushed Marcrum into organic psychosis, he would have stayed psychotic-violent and hyper-religious and able to hit people with pokers and still drive home-whether or not he had another seizure. Indeed, he would have stayed psychotic until he had a therapeutic level of anti-epilepsy medicine in his bloodstream.

The jury had no way to assess the significance of the medical records showing Marcrum had a subtherapeutic level of anti-epilepsy drugs. There was some evidence in the record of the fact that Marcrum was off his medicine: Barclay mentioned at trial that Marcrum’s seizure disorder appeared to be progressing over time, “partly as a result of his noncompliance with his medical regimen,” and one of the ambulance drivers from the night of June 3 testified that Marcrum’s family had told him that night that Marcrum “hadn’t been taking his Dilantin, which is an anti-seizure medicine, for two to three days.” These comments were before the jury, but they were drops in an ocean of evidence, which the jury was not equipped to interpret as Logan later interpreted it. Both Barclay and Parwati-kar, and for that matter, even the lay witnesses at trial, characterized the psychosis as something that was associated with the seizures, but the relation was never made clear. Some witnesses said the bizarre behavior happened before a seizure, some said it happened after. Speer himself said in his opening statement that Marcrum’s seizures “don’t necessarily have anything to do with psychosis. Sometimes, I believe the testimony will show, that sometimes one may precede the other or they may be entirely unrelated.” Parwatikar expressly said the psychosis could clear up within two hours to two days, and he referred to it as “postictal psychotic period,” which suggested that it only happened immediately after a seizure. The experts at trial never told the story Logan told: that lack of medicine led inexorably to multiple seizures to organic psychosis that would not clear until Marcrum had been medicated *509 with anti-epilepsy medicine. Without that clear, causal explanation linking the medical observations into a story line, evidence that Marcrum had not been taking his medicine and that he was paranoid that morning or in “florid psychosis” that night did not prove Marcrum was insane at the time of the killing. With Barclay’s and Parwatikar’s testimony as the jury’s expert guidance, introducing a lab result showing a low level of phenobarbital on June 3 or an emergency room doctor’s note of “florid psychosis” would not have appreciably changed the mix of evidence before the jury.

Nor would it have helped to impeach Parwatikar 9 for saying there was no evidence of any connection between Marc-rum’s psychosis and violent behavior. 10 Speer could have impeached Parwatikar with Parwatikar’s own report that said, “[W]hen he becomes psychotic ... [h]e becomes violent, running around the neighborhood without any clothes on, screaming and thinking people are trying to kill him, etc.” Or Speer could have used medical records such as those from the December 1993 incident in which Marcrum was picked up for domestic assault, and taken to the emergency room, saying, “I killed the kids, I’m hitching to the Lord and I’m taking a bunch with me.”

These tactics would have done no good because Barclay also agreed that there was no relationship between the “nonreality stage” and violence: 11

Q: Even during the previous seizure patterns, isn’t it true that your report indicated that “these patterns are nonviolent in nature”?

A: Yes.

Q: And if we remember the seizure triggers that mental disease or defect to go into the nonreality stage. So that nonreality stage, based upon your report and your research, is nonviolent in nature, isn’t it?

A: Yes.

Again, Barclay said:

Q: During the post-ictal stage, which means after the seizure, when he’s saying he’s God and he’s Gordon Gundaker, that he’s holding a book, a real estate book and thinks it’s the Bible or whatever, you never found one violent episode in his past, did you?

A: Not by a report, to the best of my knowledge, Counselor.

Neither of these experts said, as Logan said later, that the hallmark signs of Marc-rum’s psychosis were religious delusions and violence, so that the very evidence linking Marcrum to the killing (blood and the statement, “Praise the Lord, I just *510 killed one son-of-a-bitch would support the conclusion that he was insane at the time it happened. The record was not so clear that it could interpret itself. While the medical records themselves show that Marcrum was sometimes violent during his psychotic episodes, the evidence at trial indicated he was also violent at other times, 12 and the lay witnesses said not every episode of delusions and bizarre behavior led to violence. 13

Perhaps the closest case from our Circuit factually is Hill v. Lockhart, 28 F.3d 832 (8th Cir.1994), a pre-AEDPA death-penalty case. Hill’s psychologist testified that Hill had chronic paranoid schizophrenia and that it “would be reasonable to assume” the defendant was psychotic at the time of the killing. Id. at 841. He also said that paranoid schizophrenics often quit taking their medication. Id. at 842. What he did not say was that Hill had in fact missed an appointment to take his drugs about three weeks before the killing and so was unmedicated at the time of the killing. We found counsel’s performance deficient for failing to raise the “obvious” defense that Hill was insane because of failure to take his drugs. Id. at 842. However, in light of other evidence that Hill abused drugs and alcohol before the killing, we held that Hill did not prove prejudice so as to require vacatur of his conviction (although he did prove prejudice in connection with proving mitigating factors for the death penalty). Thus, despite our conclusion that the jury never heard the best reason for finding Hill insane at the time of the killing, we did not vacate the conviction. The enactment of the AEDPA after Hill was decided would make Hill even less entitled to relief under current law. Thus, Hill does not point clearly either way on whether there was prejudice in this ease.

We conclude that, without expert testimony such as Logan’s, linking the lack of anti-epilepsy medicine to an organic psychosis, marked by violent behavior and religious delusions, that would not resolve until Marcrum had received a therapeutic dose of medicine, it would not have helped appreciably for Speer to introduce the medical records and cross-examine Parwa-tikar. The defect at trial was not lack of primary evidence, it was lack of an interpretation. It was not objectively unreasonable for the Missouri Court of Appeals to find that Marcrum suffered no prejudice from Speer’s failure to introduce the medical records or cross-examine Parwatikar.

But this conclusion simply delays getting to the heart of the matter: if the medical records and the cross-examination would not have been reasonably likely to change the result of the trial without expert testimony like Logan’s testimony at the post-conviction hearing, was it ineffective for Speer to fail to come up with a Dr. Logan or someone else who would have testified to the same theory? No one disputes that Dr. Barclay was qualified or that he took sufficient time and trouble with Marcrum. He was a Ph.D. clinical psychologist from a prestigious university, with graduate training in psychopathology and impressive professional credentials, who has prac *511 ticed extensively. See Mo.Rev.Stat. §§ 552.020, 632.005(19) (requiring evaluation of certain defendants by psychiatrist or psychologist or by physician with one year experience in treating retarded or mentally ill patients), and § 337.021 (educational criteria for licensure satisfied by doctoral degree in psychology and one year of practice). Barclay examined Marcrum, tested him, interviewed members of his family, and reviewed statements of the police and extensive medical records given to him by Marcrum’s lawyer. Barclay estimated he had spent between 12 and 20 hours on Marcrum’s case.

Where counsel has obtained the assistance of a qualified expert on the issue of the defendant’s sanity and nothing has happened that should have alerted counsel to any reason why the expert’s advice was inadequate, counsel has no obligation to shop for a better opinion. Sidebottom v. Delo, 46 F.3d 744, 753 (8th Cir.1995); Six v. Delo, 94 F.3d 469, 474 (8th Cir.1996), abrogated on other grounds, Smith v. Bowersox, 311 F.3d 915, 918 n. 2 (8th Cir.2002). The fact that a later expert, usually presented at habeas, renders an opinion that would have been more helpful to the defendant’s case does not show that counsel was ineffective for failing to find and present that expert. See, e.g., King v. Kemna, 266 F.3d 816, 824 (8th Cir.2001) (en banc); Sidebottom, 46 F.3d at 753. Admittedly, we have found ineffective assistance in the sentencing phase of a death case where counsel failed to look for a second expert opinion after a cursory first examination (twenty minutes) produced an expert’s opinion which did not fit with the facts as known to counsel, Antwine v. Delo, 54 F.3d 1357, 1368 (8th Cir.1995). But there was no cursory examination here, where Barclay spent 12-20 hours working on Marcrum’s case. The very fact that Barclay’s interpretation of the record was consistent with Parwati-kar’s would have given Speer every reason to believe that both experts were making a correct analysis of the medical records.

Nor would it necessarily have been obvious to Speer, the layman, that he should contend that there was a link between Marcrum’s psychosis and violence, when both the experts said there was not. The record is not so crystal clear that Speer was on notice that his expert was missing something. As we noted above, there was evidence in the record of violent behavior even when Marcrum was not psychotic and not every episode of bizarre behavior resulted in violence. Even if Logan’s interpretation of the record makes a more coherent story and a story that would have been more likely to produce an acquittal, this does not necessarily mean that Barclay’s interpretation was wrong or that Speer was wrong to rely on it. Far less could we say that the state courts were unreasonable in holding Speer’s representation was constitutionally acceptable. See Ringo v. Roper, 472 F.3d 1001, 1006 (8th Cir.), cert. denied, - U.S. -, 128 S.Ct. 445, 169 L.Ed.2d 311 (2007).

We cannot conclude that Marcrum has proved that his constitutional right to counsel was violated and that the Missouri courts were unreasonable in finding to the contrary.

IV.

Marcrum argues that we should affirm on the alternative ground that Speer was ineffective in failing to object to the prosecutor’s personalization in closing argument. The district court held that there was no reasonable probability that the outcome of the trial would have been different if Speer had objected, and we agree there is no such probability.

Marcrum also contends that Speer was ineffective in failing to file Marcrum’s new trial motion in time. He contends that this constituted deprivation of counsel at a critical stage, thus entitling him to *512 relief even without proving he suffered prejudice. Speer’s mistake in failing to file the motion on time was just that — a mistake by counsel, not a deprivation of counsel. Marcrum must therefore prove prejudice. See Bell v. Cone, 535 U.S. 685, 697-98, 122 S.Ct. 1843, 152 L.Ed.2d 914 (2002). As the district court held, he showed none, since the Missouri Court of Appeals considered his arguments for new trial at the post-conviction stage and held none of the arguments would have prevailed.

# * ❖

While habeas counsel has done a commendable job in presenting this difficult case and the district court’s analysis was thorough and thoughtful, we ultimately must conclude that Marcrum was not denied his right to effective assistance of counsel at trial. It follows that the state courts did not unreasonably apply established federal law in reaching the same conclusion. Accordingly, the judgment of the district court must be and is reversed.