dissenting.
Petitioner Bradford has come forward with powerful evidence of both his innocence and his trial lawyers’ ineffective assistance in dealing with the critical scientific issue — the duration of the fire in Tamara Lohr’s bedroom. The State and the courts upholding Bradford’s convictions have not yet offered a plausible theory to account for the physical evidence of the fire and the time Bradford supposedly set it.
Instead, to rebut the scientific foundation for Bradford’s petition for a writ of habeas corpus, the majority has struck out on its own with some modest criticisms based on the majority’s own research and analysis. Those criticisms do not undermine the conclusion that Bradford could not have been the arsonist and murderer. Bradford and his key post-conviction expert also have not had an opportunity to respond to even these modest criticisms. Rather than affirm the denial of relief for Bradford based on these untested critiques, we should order a new trial to test all the relevant evidence. I respectfully dissent.
The scientific issue requires close attention to the timing of Bradford’s actions, to the physical details of the fire and the damage it caused, and to objective information about fire behavior. I cannot offer the reader a shortcut, but I can offer a roadmap. Part I lays out the critical facts about the murder and fire: in I-A, the timing of Bradford’s movements and the fire, in I-B the State’s theory for Bradford’s guilt, and then in I-C the key physi*908cal evidence. Part II turns to the law and how Bradford’s otherwise able counsel provided ineffective assistance by failing to present available and credible expert testimony on the critical issue of fire duration. Part II reviews the evidence of fire experts at trial and then the post-conviction case, applies the law governing ineffective assistance claims to those facts, and finally addresses the majority’s critiques of the new evidence.
I. The Murder and the Fire
A. Timing
The case against Bradford was circumstantial. No one witnessed Tamara Lohr’s murder, and no physical evidence linked Bradford to her death. Yet Bradford was an obvious suspect. He and Lohr had been having an affair. Bradford’s wife had learned of it, and he had recently decided to end it. Lohr responded with emails threatening to tell Bradford’s wife the affair was continuing. Making things worse for himself, Bradford tried to delete some of those angry emails a couple of hours after the discovery of Lohr’s body.
But undisputed facts from the prosecution’s own case make it impossible that Bradford could have murdered Lohr and set the fire in her bedroom. Those undisputed facts confine the State to a theory that ultimately collapses from internal contradictions about the timing and intensity of the fire.
Bradford visited Lohr for a few minutes on the evening of August 1, 1992, leaving around 10:20 p.m. to start his night shift as a police officer. Lohr was alive when he left. She talked with her father by telephone between about 10:25 and 11:00 p.m. The next morning, when the fire department arrived at her home just after 6:30 a.m., her bedroom was ablaze and Lohr was dead. She had been murdered before the fire began, stabbed 21 times. Gasoline had been used as an accelerant.
Bradford’s movements that night and morning provide the first of the facts that confine the State’s theory. After leaving Lohr’s house around 10:20 p.m., he reported for patrol duty on the night shift, starting at 10:30, while Lohr was on the phone with her father. From then until 6:34 the next morning, Bradford was on patrol and accounted for, with one notable exception. Between 11:06 p.m. and 12:11 a.m., there was a gap in his activity log for the night. That would have been his only opportunity to commit the murder. The rest of the night, he was busy with patrol duties until his shift ended.1
The timing of the fire and Bradford’s movements in the morning is critical. The State’s case rested on the theory that Bradford had not started the fire until the morning, just minutes before firefighters extinguished it. We know Bradford left work about 6:30 a.m. and headed for Lohr’s house. A bank’s ATM camera a few blocks away from Lohr’s home camera captured an image of Bradford’s car passing the bank at exactly 6:34:04 a.m.2
Just 65 seconds later, at 6:35:09 a.m., Bradford was at Lohr’s house calling for a fire truck and ambulance. Less than a minute after that, at 6:36:07 a.m., two runners two blocks away saw a column of *909smoke from the house rising above the treeline — smoke they described as “real thick,” and “very dark, black looking, and so thick.” Police and fire arrived at 6:38:30 and 6:38:49 a.m., respectively. The fire was extinguished no later than 6:43:19 a.m. An empty gas can was under Lohr’s bed where her body lay.
Later the day of the fire, an investigator used an electronic “sniffer” to check Bradford’s uniform and shoes. They showed no signs of gasoline or other accelerants. Nor was there any blood on his uniform. Police officers who were close to Bradford at the scene noticed he smelled of smoke but not gasoline. That was consistent with Bradford’s report that when he arrived, he went into the house but it was so full of smoke that he was unable to reach Lohr’s bedroom.
B. The State’s Theory
The undisputed timing evidence left the State with a theory that seems unlikely on first reading. According to the State, a police officer planned and carried out a scheme — while on patrol duty and subject to calls at any moment during a busy Saturday night shift in high summer — in which he returned to Lohr’s house bé-tween 11:00 p.m. and midnight, murdered her by stabbing her 21 times, and then carefully arranged the premises to fake a break-in and made preparations to start a fire in a return visit. The State’s theory is that Bradford then returned to work (in a clean uniform, no less), worked the rest of his shift, behaving professionally and exhibiting no unusual behavior toward anyone he encountered, and then returned to the scene of the murder just after 6:30 a.m. both to start the fire and to call it in just seconds later.
On further scrutiny, this theory becomes even less likely. It requires that Bradford, in no more than about 45 seconds, encountered a neighbor who he knew would recognize him, walked calmly into Lohr’s house, poured gasoline over her body, the bed, and the floor, lit a fire, and then left the house and immediately called the fire department.3
Why would a police officer use a plan that required him to make a second visit to the scene of the murder and ensured he would be there when Lohr’s body was discovered? Why make that second visit late enough in the morning that a number of neighbors would be awake and out on streets and sidewalks and able to place him there, supposedly before there was any sign of a fire? If the fire was meant to conceal the murder, why not light it at night, during the first visit, when he had at least an arguable alibi, and when the fire would have had a much better chance of going undiscovered long enough to destroy significant evidence?
Despite these problems, this had to be the State’s theory to convict Bradford. Yet murder cases can involve strange behavior, even if it seems improbable. Those problems with the theory make it implausible but do not quite disprove it.
C. Physical Evidence of the Fire
The physical evidence of the. fire, however, finishes the job of refuting the State’s theory. Under that theory, the fire could have burned no more than eight and a half minutes after Bradford supposedly started it. It also would have had to generate a heavy column of smoke visible from two blocks away no more than 90 seconds after ignition.
*910Investigators found many clues to the duration of the fire. The fire had been confined to Lohr’s bedroom, as shown by the only minor damage to areas outside the bedroom. Critically, the bedroom door had been open just about one inch during the fire, prevented from closing entirely by a clothes hanger. Paint at the top of the door frame had been protected by the door, showing that it had remained nearly closed during the fire.
Within the bedroom, the fire caused extensive damage. Lohr’s foam mattress had been entirely consumed. Yet the room had not gone into “flashover” — when the upper layer of gas in a room becomes hot enough to ignite all available combustible surfaces. The room had two windows: the front bedroom window, made up of twelve glass panes, was intact. (A140) The side window had been shattered by firefighters’ efforts to access the bedroom, but one firefighter testified that only one of its panes had had a few long cracks in it before that occurred. (Tr. 976-68) Even the wooden dividers separating the panes were still intact. (Tr. 1955)
The upper panel of the bedroom door had charred through from ambient heat, as had a similar closet door. That hole in the panel of the bedroom door allowed smoke from the bedroom to escape through that gap, and then up through an open hatch to the attic. Until the door charred through, no significant amount of smoke would have escaped the bedroom. Until that happened, there was no way for enough smoke to accumulate in the attic to cause a thick smoke plume and to seep from the house’s eaves.
No witness, lawyer, or judge has offered a plausible theory as to how the door could have charred through and the fire produced a heavy column of smoke within 90 seconds after Bradford entered the house. The fire simply could not have done so much damage to the bed and to the bedroom door in the available time. Even to approach those short times, the fire would have left unmistakable signs of much higher temperatures and a flashover fire in the bedroom, which did not happen. In short, no one has offered a plausible account of the timing and physical evidence that supports Bradford’s convictions.
II. The Claim for Ineffective Assistance of Counsel
The fire’s duration has always been the pivotal issue in the circumstantial case against Bradford. If the fire burned longer than eight and a half minutes, then it does not matter whether he might have had a motive to kill Lohr or that he tried to delete emails about the affair or that the able prosecutor scored some effective points in cross-examining him. If the fire began before 6:34 a.m., Bradford could not have been the arsonist — full stop.
The failure of Bradford’s lawyers to identify and call an appropriate expert on fire duration amounted in this case to ineffective assistance and led to an unreliable verdict. To show the ineffective assistance, I review in Part II-A the expert evidence about the fire at trial, in II-B the expert evidence Bradford offered in the post-conviction case, in II-C the applicable law, and in II-D the majority’s criticisms of the post-conviction expert evidence.
A. Fire Experts at Trial
The prosecution called Jesse Storey, an investigator for the Evansville Fire Department. He opined that the fire as a whole lasted a total of seven to ten minutes (Tr. 1993), nicely bracketing the maximum, eight and a half minutes allowed by the State’s theory. Storey also said the door would have charred through in two to three minutes. (Tr. 2018) He did not explain the basis for his char-through estimate. He also claimed that smoke would *911have been visible from outside the house within a minute of ignition. (Tr. 1990) If that were true, it might help explain away the runners’ testimony. The problem is that neither Storey nor anyone else has offered an explanation for that claim, let alone any quantitative basis for it, keeping in mind that the bedroom door was nearly closed and remained intact during that time, according to Storey himself.
The defense called Barker Davie, a well-regarded forensic chemist and fire investigator. He was an expert on the causes and origins of fires, including cases of arson. In this case, though, there was no doubt about the cause or origin. Timing was the issue. Despite his other abilities, Davie was not able to provide the needed expertise on that issue.
Davie described the behavior of the fire in qualitative terms, testifying that the fire would have burned intensely at the beginning due to the gasoline, would have died down as the oxygen in the nearly closed room was consumed with little ventilation to supply new oxygen, and then would have flared up again once there was a new source of air — i.e., after the door panel charred through. He also testified about the quality of the smoke the fire would have produced and the path the smoke would have taken. Davie testified that the smoke would have escaped the bedroom, first through the crack in the slightly open bedroom door and then, in greater amounts, through the hole in the door after it charred through. It would have risen through the open scuttle hole into the attic and exited through the roof vent. Eventually, he testified, as the fire continued to burn and more smoke built up in the attic, the plume would have become visible above the roof line; smoke might also have begun to seep from beneath the eaves if it had built up to such a degree that the roof vents could no longer provide sufficient egress.
On the critical matter of timing, however, Davie was not able to offer any reliable quantitative analysis — not for the door to char through or for smoke to become visible. When asked about the time to char through, he offered only an estimate based on the National Fire Protection Association “international rule of thumb” that “wood will char and burn at the rate of roughly one inch in forty-five minutes time frame.” (Tr. 3085) Davie used that rule of thumb to estimate a char-through time of between seven and ten minutes for this door panel about 5/16” thick. (Tr. 3086) As for seeing smoke, Davie said he could not estimate “how long and what range of time it would be before smoke would be clearly visible from the outside of the house.” (Tr. 3088) Instead, he relied on the fact that the two runners had seen a heavy column of smoke as early as 6:36:07 a.m., just two minutes after Bradford passed the bank camera on his way to Lohr’s home. Because generally smoke takes time to build up, he testified, it would be “highly improbable” that the smoke would be visible “in the first minute or two of that fire.” (Tr. 3091; 3099-3101)
When asked for an opinion as to the fire’s earliest possible starting time and the range the fire could have burned, Da-vie was unable to offer an estimate based on the physical evidence at the scene. Instead, he based his estimate on the testimony of Gerald Johnson, a newspaper carrier who testified he smelled smoke in front of Lohr’s home during his paper route. Johnson described hearing fire alarms later that morning and estimated he had smelled smoke seven and a half to ten minutes earlier than that. Bradford’s call was fixed at 6:35:09 a.m., so with an estimated alarm time of 6:36, Johnson would have smelled smoke between 6:26 and 6:29 a.m., five to eight minutes before Bradford drove past the bank. Davie also testified that the smoke would not have *912been perceptible for about two to five minutes after the fire started, because time was required for the smoke to build in intensity, pushing the ignition time even earlier. (Tr. 3122)
On cross-examination, Davie was asked to ignore the witness testimony and to provide an estimate, “based upon the physical evidence,” as to the minimum amount of time the fire would take. He responded:
Based just solely on the physical evidence and nothing else that door frame could burn at the rate of one inch in forty-five minutes time frame with constant direct flame impingement and that would take on the order of about eleven minutes to do that. (Tr. 3227)
That opinion was “[sjolely based on just looking at that door excluding everything else and that’s the only piece of physical evidence that I know of right now that’s present here that would help give you a gauge.” (Tr. 3228)
Davie had made a critical mistake that undermined his credibility. In rebuttal the State called Donald Johnson, a Special Agent with what was then the federal Bureau of Alcohol, Tobacco and Firearms. He testified that Davie’s “rule of thumb” was “no longer considered to be a rule or a standard.” He cited NFPA Publication 921, which was released in 1992 and rejected Davie’s “rule of thumb”:
4-5.2* Rate of Charring. The depth of char measurements should not be relied on to determine the duration of the burning. The rule of 1 in. in 45 min for the rate of charring of pine is based on one set of laboratory conditions in a test furnace. Fires may burn with more or less intensity during the course of an uncontrolled fire than under a controlled laboratory fire. Actual laboratory char rates from exposure to heat from one side vary from 0.4 in. per hr at 750°F (390Ó) to 10 in. per hr at temperatures approaching 2000°F (1090°C) in intense fires. Even these figures will vary with the species of the wood, orientation of the grain, moisture content, and other variables. Charring rate is also a function of the velocity of hot gases and the ventilation conditions. Fast moving gases or ventilation can lead to rapid charring.
Donald Johnson’s testimony wholly undermined Davie’s and the defense’s only analysis of the physical evidence to find the fire duration, which had been based on an obsolete standard that had been disavowed more than a year before trial.
Johnson also disagreed with Davie’s seven- to eleven-minute estimate of the char-through time for the door. Johnson testified that the door would have charred through in three to four minutes, though he did not offer any empirical basis for that estimate. (Tr. 3818-19)
Davie’s reliance on the discredited “rule of thumb” left Bradford with no credible expert testimony on the timing of the fire. Instead, all he had was the imprecise and unreliable testimony of the newspaper carrier, Gerald Johnson.4 The State made this point several times in closing. Here was the most telling attack:
[Y]ou could boil down [Davie’s] hours of testimony to this: that if Gerald Johnson smelled smoke coming from Tammy Lohr’s bedroom and if that smoke was part of the fire that burned down her bedroom and burned her body and if that smell of smoke happened before Patrick [Bradford] got there then Pat*913rick Bradford didn’t set the fire. Well, you don’t need to pay someone a hundred and twenty-five dollars an hour to tell you that. (Tr. 3950-51)
See also Tr. 3938, 3943. The.attacks were right on target. On the critical issue of fire duration, Davie simply did not provide expertise that helped the defense.
B. Expert Evidence in the Post-Conviction Case
Bradford was convicted of the murder and arson, and his convictions were affirmed. Bradford v. State, 675 N.E.2d 296 (Ind. 1996). He sought post-conviction relief. After years of delay, the state court held an evidentiary hearing in 2011. Bradford presented a number of claims, but I focus on the scientific evidence he presented to show that the fire could not have developed as the State had claimed to show his guilt.
Bradford offered the testimony of Douglas Carpenter, a fire protection engineer. Carpenter did what Davie and the State’s experts had failed to do for the original trial: he provided a quantitative analysis, rooted' in science and the physical evidence, of the duration of the fire.
Carpenter’s report described flashover (when hot gases reach a high enough temperature to ignite all available combustible surfaces in a room). Generally, the upper layer temperature must be about 500°to 600°Celsius (about 930°to llOOTahr-enheit) for this to occur, which corresponds to a, heat flux of 20 kilowatts per square meter at floor level. Physical evidence shows that flashover did not occur. (A120-21)
Carpenter set out to estimate how long it took for the door panel to char through A17 in non-flashover conditions, and how long it would have taken, based on the known conditions of the fire. First, he analyzed the significance of the smoke plume rising from the attic and the smoke seeping from beneath the eaves. Carpenter, like both the State’s experts and Davie, recognized that the fire was “ventilation limited” — that is, limited by lack of oxygen, not lack of fuel — before the door charred through. The nearly closed bedroom door meant that the oxygen in the bedroom was used quickly and could not be replaced with fresh air.
According to Carpenter, the flow of smoke out of the bedroom also would have been minimal before the door charred through because the smoke had so little space to escape. There would have been no appreciable smoke accumulation in the house or attic before the door charred through. Without accumulation outside the bedroom, there would have been no visible smoke column or smoke leaking from beneath the eaves. The little smoke that escaped the room would have risen to the attic and exited through the small roof vents without significant accumulation in the attic. (A126-27)
After char-through, however, the dramatic increase in area of the outflow openings (from 0.06 square meters with a one-inch bedroom door gap, to 0.80 square meters after char-through) meant that the smoke could no longer be vented in full by the roof vents. Char-through would have led to smoke accumulation in the attic, which would in turn have led to the heavy smoke plume that the runners saw at 6:36 a.m.
Carpenter then turned to the time for the door to char through. Recall that the State’s experts at trial had estimated just two to four minutes, though without offering any quantitative basis. Carpenter, however, obtained exemplar doors from Lohr’s house. He conducted experiments by exposing samples of the door panels to a range .of thermal heat fluxes (between 12 and 20 kW/m2) and recording the char-through times. (A137-38) At 20 kW/m2, which corresponds to an upper layer temp*914erature of about 350° Celsius (660° Fahrenheit), the panel took an average of 11 minutes and 33 seconds to char through — • three minutes longer than the maximum total duration of the fire, according to the State’s theory.5
For these figures to matter, Carpenter needed to determine the heat flux actually produced by the fire in Lohr’s bedroom. To do this, he considered the front window in the bedroom, which contained twelve individual glass panes. None broke, though one pane in the side window might have cracked during the fire. Relying on data from a study that estimated the probability of three-millimeter glass breakage as a function of hot temperatures, he estimated that the upper layer temperature would have been about 280° Celsius (540° Fahrenheit), corresponding to a heat flux- of 16 kW/m2. At 280° Celsius, the probability of glass breakage is about one in twelve. Because none of the twelve panes broke, Carpenter estimated that the temperature had to be 280° or below, corresponding to a char-through time of about 30 minutes.
Carpenter also considered the mattress in Lohr’s bedroom, which was the principal fuel for the fire after the gasoline burned off. The mattress was entirely consumed by the fire. “If the amount of combustible mass (in units of kg) and the mass loss rate (in units of kg/sec) are known, then an estimate of the duration of the fire (in units of seconds) can be obtained.” As noted, before the bedroom door charred through, the fire was ventilation-limited, not fuel-limited, thanks to the nearly closed bedroom door and the lack of another oxygen source. Carpenter estimated that the mattress included about 20 kilograms (about 44 pounds) of combustible polyurethane foam. (A143) According to a model for fire behavior widely accepted by fire experts — Consolidated Fire and Smoke Transport (CFAST) — with enough ventilation, burning such a mattress would produce so much heat that a fire in an average-sized bedroom would proceed to flashover, which did not occur in this case. Thus, Carpenter concluded, the mattress had been consumed entirely before the door charred through and allowed more oxygen into the room. (A145-46)
Available data showed a mass loss rate on the order of one gram per second for flexible polyurethane foam in a closed-door fire. (A147-48) Since the door was slightly ajar here, Carpenter used the CFAST model to predict a faster mass loss rate of about three grams per second for a fire in a bedroom with a door open approximately one inch. That produces a rough estimate of a two-hour fire to consume the entire mattress with the door an inch ajar. Carpenter used this figure to estimate an outer bound for the fire’s ignition time. Combining the independent data from the glass that did not break, the char-through experiments, and the mattress, he estimated that the fire had to have been set between 4:30 a.m. and 6:00 a.m., well outside the time when Bradford could have been responsible. The undisputed testimony of the runners who saw a smoke column from two blocks away at 6:36, which the majority does not mention, provides independent corroboration of Carpenter’s analysis.6
*915C. Ineffective Assistance of Counsel
To show ineffective assistance of counsel, a defendant must show (1) that counsel’s performance was constitutionally deficient, and (2) that the deficient performance prejudiced the defendant. Strickland v. Washington, 466 U.S. 668, 687, 104 S.Ct. 2052, 80 L.Ed.2d 674 (1984). Bradford has made both showings here.
1. Deficient Performance
Deficient performance means that counsel’s representation “fell below an objective standard of reasonableness.” Id. at 688, 104 S.Ct. 2052. A court’s after-the-fact scrutiny of counsel’s performance must be deferential to avoid hindsight and to recognize that a decision may be a matter of strategy. Id. at 689, 104 S.Ct. 2052. If counsel’s performance overall reflected “active and capable advocacy,” establishing a claim for ineffective assistance of counsel is and should be difficult. Harrington v. Richter, 562 U.S. 86, 111, 131 S.Ct. 770, 178 L.Ed.2d 624 (2011). It is not impossible. In some instances, “ ‘even an isolated error’ can support an ineffective-assistance claim if it is ‘sufficiently egregious and prejudicial.’ ” Id., quoting Murray v. Carrier, 477 U.S. 478, 496, 106 S.Ct. 2639, 91 L.Ed.2d 397 (1986); see also Williams v. Lemmon, 557 F.3d 534, 538 (7th Cir. 2009).
The selection of an expert witness is ordinarily a matter of strategy and difficult to challenge under the deferential standards of Strickland and 28 U.S.C. § 2254(d). The Indiana courts concluded that counsel’s selection of Davie was a reasonable strategic choice. Davie had a good reputation in the fire investigation community, and he had never before testified for the defense in a criminal case. Bradford v. State, 988 N.E.2d 1192, 1203-04 (Ind. App. 2013). We owe deference to the state court’s resolution of the claim as well as to counsel’s strategic decisions.
Crucially, though, in a failing the state courts did not recognize, Davie was unable to offer a credible expert opinion on the one question that mattered the most: whether it was physically possible for Bradford to have set the fire. And such expertise was available at the time of trial, using the information and techniques later used by Carpenter.
By Davie’s own admission, he recognized only a single piece of physical evidence (the door) that could help him estimate the fire’s duration. He apparently did not appreciate the importance of the intact windows and the fully consumed mattress. And rather than carry out an experiment on char-through time, he just relied on the obsolete “rule of thumb” that had been discredited more than a year before the trial.
On the fire’s duration, Davie was equivalent to no expert at all, as the prosecutor showed during closing. We and other circuits have found deficient performance before where counsel fails to present expert evidence on a question so critical to the prosecution’s case and his client’s defense as the fire duration was here. E.g., Thomas v. Clements, 789 F.3d 760, 769-70 (7th Cir. 2015) (effect of counsel’s failure to reach out to expert “was to accept [prosecution expert’s] finding of intentional death without challenge and basically doom defense’s theory of the case”); Woolley v. Rednour, 702 F.3d 411, 423 (7th Cir. 2012) (failure to present rebuttal expert where State expert’s testimony “effectively hollowed out the core” of defense and “went to the heart of whether [defendant’s] version of the shootings was physically possible”); Richey v. Bradshaw, 498 F.3d 344, 362-64 (6th Cir. 2007) (deficient perform-*916anee where scientific testimony was critical to State’s specific theory and defense counsel hired an expert but failed to consult with him, while other experts would have “severely undermined the State’s case”); Dugas v. Coplan, 428 F.3d 317, 328-31 (1st Cir. 2005) (importance of challenging State’s arson case, crucial role of arson evidence, and attorney’s awareness of problems with the State’s case all demonstrated “inescapable need for expert consultation in this case”); see also Rogers v. Israel, 746 F.2d 1288, 1294 (7th Cir. 1984) (“under certain circumstances, ‘it may be vital in affording effective representation to a defendant in a criminal case for counsel to elicit expert testimony rebutting the state’s expert testimony’ ”), quoting Knott v. Mabry, 671 F.2d 1208, 1212-13 (8th Cir. 1982).
For a single error to qualify as ineffective assistance, it must be sufficiently egregious and prejudicial — “an omission of something obviously better (in light of what was known at the time) than the line of defense that counsel pursued.” Williams, 557 F.3d at 538. Scientific testimony that exonerates a defendant surely fits that bill, so long as it was “known at the time.”
On that critical point, the state trial court concluded after the post-conviction hearing that “Carpenter provides only a new opinion on the same evidence that was available at the time of trial.” The court also found that “the theories and factors upon which [Carpenter] based his calculations were known or with some work could' have been calculated — and were therefore discoverable by due diligence — at the time of trial.” The Indiana Court of Appeals agreed. Bradford v. State, 988 N.E.2d at 1199-1200.
That reasoning convinced the state courts that Carpenter’s evidence did not qualify as “new” evidence to justify a new trial. But that reasoning is also critical on the issue of ineffective assistance. The State cannot have it both ways. Bradford’s bid for a new trial was rejected because the information and analysis Carpenter has offered was available at the time of trial. His otherwise capable counsel should have recognized such quantitative expertise as obviously better than the weak and equivocal duration testimony that Davie provided. As the prosecutor correctly pointed out more than once, that testimony was not based on expertise at all.
2. Prejudice
As for prejudice, to be entitled to habeas corpus relief, Bradford must show it is reasonably likely that but for counsel’s error, the result of the proceeding would have been different. Harrington, 562 U.S. at 111, 131 S.Ct. 770. The likelihood of a different result has to be “substantial, not just conceivable,” id. at 112, 131 S.Ct. 770, citing Strickland, 466 U.S. at 693, 104 S.Ct. 2052, though Bradford need not show that the error more likely than not altered the outcome in the case. Id. at 111-12, 131 S.Ct. 770; Stanley v. Bartley, 465 F.3d 810, 814 (7th Cir. 2006) (petitioner’s chance of being acquitted need not be 50 percent or greater); see also, e.g., Thomas v. Clements, 789 F.3d at 772 (finding prejudice where it was “substantially likely that [petitioner] could have raised at least a reasonable doubt and had a different outcome at trial”). Essentially, the errors must be serious enough to undermine confidence in the outcome. Cullen v. Pinholster, 563 U.S. 170, 189, 131 S.Ct. 1388, 179 L.Ed.2d 557 (2011), quoting Strickland, 466 U.S. at 694, 104 S.Ct. 2052. The state courts did not reach the issue of prejudice.
Bradford easily satisfies the prejudice requirement. No reliable scientific evidence on fire duration and char-through time was offered at trial — not by Storey, not by Johnson, and not by Davie. The *917bank camera drew a bright line at 6:34 a.m. If the fire started earlier than that, Bradford did not light it. If counsel had presented Carpenter’s analysis on duration and if that evidence was scientifically valid, a reasonable jury could not have found Bradford guilty. At the very least, Carpenter’s analysis undermines confidence in this verdict.7
D. The Majority’s Critique of Carpenter
Carpenter’s scientific analysis showing that Bradford could not have set the fire was unscathed by cross-examination in the state courts. The prosecutor focused on trying to show that another expert could have done the same sort of analysis at the time of trial. That is also my point, of course, for it is key to showing the ineffectiveness of relying on Davie. The prosecution tried a few substantive attacks, such as trying to show some uncertainty about the thickness of the window glass, how similar the door samples were for the char-through experiments, and the exact size of Lohr’s mattress. Carpenter easily parried those, showing why his estimates were reliable and/or that minor changes in the variables would not change the bottom line conclusion. Not surprisingly, then, the state trial and appellate courts did not try to engage Carpenter’s analysis on the merits. In this federal appeal, we find the first attempt by a court to engage Carpenter on the merits. The majority offers some new criticisms based on its own research. None of those criticisms are serious enough to undermine the key conclusion.
The majority’s most telling point is that Carpenter’s analysis of the probability of glass breakage was wrong, and I agree with the point. A one in twelve probability that a pane of glass will break does not mean that in twelve separate, independent events, one will certainly break. (In simpler terms, if you flip a coin twice, the probability of heads is one-half on each flip but there is no guarantee that one of two flips will turn up heads.) Carpenter also did not explain why all twelve panes might have been exposed to approximately the same heat flux, given that heat rises. He also could not be sure that the glass in Lohr’s bedroom windows was three millimeters thick, though there is no evidence of non-standard glass in this small and modest home. And so on.
The prosecutor did not challenge Carpenter on the probability of the window panes remaining intact. The data in Carpenter’s report show that the probability of breakage rises dramatically after about 275° Celsius (530° Fahrenheit). By the time the temperature reaches about 350° Celsius (660° Fahrenheit) — which is not hot enough to produce a char-through time of eight and a half minutes based on Carpenter’s tests — the probability of glass breakage rises to about 0.6. (A140) If each of the twelve panes had an independent six in ten chance of breaking, the probability that none would break would be (4/10)12, or *9181.677 x 10'5 — less than two in a hundred thousand. Also, the data refer to glass breakage, meaning that pieces of the glass fall out, further ventilating the fire. Ba-brauskas, cited by the majority, also says it is fairly well-established that a windowpane of ordinary float glass “tends to crack when the glass reaches a temperature of about 150-200°C.” Carpenter’s report indicates that the front bedroom window “showed no signs of cracking.” (A138)
In other words, even if we correct the error in Carpenter’s probability analysis of the glass, the State still cannot explain how, if the fire burned hot enough to char through the door fast enough to support its theory, all of the front windowpanes remained unbroken.
More fundamental, most of the majority’s critiques were not even raised during the cross-examination of Carpenter at the post-conviction hearing. Carpenter never had the opportunity to expand upon his reasoning, to defend his work, or, if there was in fact a significant error, to explain why it might or might not affect his ultimate conclusion.
For example, at page 6, above, the majority attacks Carpenter for having confused floor- and ceiling-level heat flux during his testimony. That apparent error seems to have been limited to his testimony at the hearing. His report did not make that mistake, nor were the report’s overall conclusions based on it. His report makes clear that his estimate of the char-through time is based on the intact glass. (A138) It is not unusual even for experts (and lawyers and judges) to misspeak occasionally in such ways. The questioning prosecutor did not identify the point or ask for an explanation.
The majority also criticizes Carpenter for failing to explain why, since heat rises, all the windowpanes would have been exposed to the ceiling level heat flux. One could just as easily say it is not clear why the entire door panel from top to bottom would have been exposed to the ceiling level heat flux, given that the door, like the window, extended down from the ceiling into what were presumably slightly cooler temperatures. More to the point, Carpenter simply was not asked to justify his assumption. We do not know why he made it, whether it was appropriate, or whether changing it would have a significant effect on his conclusion.
In contrast, one such critique was raised on cross-examination — Carpenter’s assumption that the window was made of plain three-millimeter glass. He parried the attack easily, noting that tempered glass “isn’t traditionally used in residential situations,” and that three millimeters is a typical thickness. This seems a reasonable explanation to me, and it suggests Carpenter may well have been able to rebut the majority’s other critiques if he had been asked about them.
Carpenter’s evidence is not perfect. It depends on certain assumptions and estimates, both because of the passage of time and because the behavior of an uncontrolled fire can be difficult to pin down. But his evidence is powerful. It is the only analysis of the fire’s duration based on empirical data, derived from actual tests instead of conjecture, obsolete rules of thumb, or inexact witness guesstimates. His analysis is quantitative and testable. His results appear to be so definitive that minor adjustments for surprisingly thick glass or having the bedroom door open a little farther could not solve the basic problems with the State’s theory.
To cause the physical damage within the State’s eight-and-a-half-minute maximum burn time, the fire would have had to have been much more intense, consuming the entire mattress in just a few minutes and resulting in broken windows and the flashover that the experts agree never *919happened. And there is no plausible explanation how the fire in the nearly closed bedroom could have produced, in no more than 90 seconds, a heavy smoke plume visible to the runners more than two blocks away. To explain that smoke plume at all, the char-through time would need to be measured not in minutes but in seconds. And that would mean in turn that the fire would-have been ventilated almost from the beginning, contrary to all the fire experts’ views, and it would have reached flashover, which did not happen.
If we are going to apply our own analysis to Carpenter’s analysis, we also should not close our eyes to the grisly report from Dr. John D. DeHaan, which focuses on the fire damage to Lohr’s body. His analysis, based on the autopsy report and tests that have been done by exposing human cadavers and the bodies of other animals to flames, corroborates Carpenter. DeHaan found that the heavy fire damage to Lohr’s body was inconsistent with a fire of only five to seven minutes and was “much more consistent with a fire exposure of approximately 20 minutes to direct flames” — or perhaps to a much longer fire, one of reduced intensity due to the ventilation limits on the fire in this case.
To sum up, even if Carpenter’s analysis has a couple of minor errors in it, his analysis is the only analysis of the fire’s duration with a scientific and empirical basis. His conclusion appears sound, and the State did not show otherwise. If it is correct, it exonerates Bradford. In a new trial, the State would have the chance to attack the scientific analysis on its merits. Perhaps the State would then be able to show that Carpenter’s work is not to be trusted. Or perhaps it could present, finally, an expert who could explain in reliable scientific terms how a fire could have done the damage this fire did so quickly without leaving unmistakable signs of a much hotter fire.
Given the state court’s conclusion that the analysis Carpenter provided was available at the time of the original trial, I would treat the failure to seek and find such critical expert evidence as ineffective assistance of counsel. In the alternative, I would treat this case as appropriate for an actual innocence grant of habeas' corpus. We should order issuance of a writ of habeas corpus directing that Bradford be released or retried.
. The majority claims that Bradford "appears to have concealed his whereabouts on the fatal night.” If he had known he would want an alibi, I expect he would have done a better job of manufacturing one. Bradford's activities- in those 65 minutes were hotly disputed at trial, but my focus is on the duration of the fire, even assuming that his alibi evidence for those 65 minutes need not be believed.
. This time and the others presented in the evidence were all synchronized to ensure exact comparisons of times from different systems, such as the ATM camera and the police and ñre dispatch systems.
. The tíme between Bradford's appearance on the bank camera and the call to emergency services was 65 seconds. A detective testified that completing the drive from the bank would have taken 18 to 23 seconds, leaving about 45 seconds to complete all the other tasks.
. As the State pointed out in closing, Johnson testified that the smoke smelled like burning wood, which is not the smell the foam mattress — the primary source of fuel — would have given off. Johnson also said that he had been "convinced” the smoke smell was coming from behind the house, not Lohr’s bedroom; otherwise, he would have called in an alarm.
. Carpenter's estimate of char-through time is consistent with data in the new NFPA 921 standard, which notes a char-through rate of 0.4 inches per hour at a temperature of 750° Fahrenheit, or about 400° Celsius. PCR Tr. 198. The bedroom door was about 0.3 inches thick (5/16 of an inch), and the heat on the door could not have reached anything close to 750° Fahrenheit in light of the lack of glass breakage.
. The parties have also debated Carpenter’s assertion that the concentration of carbon monoxide in Lohr’s dog’s blood proved that the fire had burned longer than eight minutes. There is enough controversy about the relia*915bility of that analysis that I do not rely on it here.
. Bradford recognizes how difficult it is to obtain federal habeas relief from a state conviction on a theory of ineffective assistance of counsel. He has also argued that this would be an appropriate case to decide whether the federal constitutional guarantee of due process of law should support an independent claim for relief based on only actual innocence, without any other constitutional violation. See House v. Bell, 547 U.S. 518, 554-55, 126 S.Ct. 2064, 165 L.Ed.2d 1 (2006) (question remains open); Herrera v. Collins, 506 U.S. 390, 417, 113 S.Ct. 853, 122 L.Ed.2d 203 (1993) (threshold for such theory would be "extraordinarily high”); id. at 442, 113 S.Ct. 853 (Blackmun, J., dissenting) (petitioner would need to show he “probably is innocent”). In my view, Bradford has shown that he is probably, in fact almost certainly, innocent. The only reasons I hedge at all on his innocence are (a) I am not a fire expert and (b) no true expert has tried to rebut Carpenter’s analysis. Perhaps he is mistaken, but no one has shown that yet.