The Discredited, Part III: Fire, and the Indicators That Were Never Indicators

For most of a century, arson investigation ran on a set of burn-pattern signs that fire research eventually explained as ordinary features of accidental fires. Texas received the report four days before it executed Cameron Todd Willingham. Michigan released David Gavitt after twenty-six years. The underlying error was identical.

What I Notice First About Fire Cases

I want to name the thing that separates arson from every other discipline in this series before getting to the cases.

Hair comparison produced wrongful convictions. So did bite mark analysis, and so will several disciplines this series has not reached yet. In each of those, the remedy is late and inadequate and requires someone to find the file. It is still, in principle, available.

Fire cases are different in one respect. Arson is a predicate for capital murder in many states, because a fire that kills a family is exactly the sort of case that draws the maximum charge. That means the validation lag in fire science runs against a clock that other disciplines do not have. When the correction arrives after the execution date, there is no correction. There is only a finding.

That is the structural fact this installment is about. Not that the fire science was wrong, which is now uncontroversial, but that the interval between knowing and acting was allowed to run against a scheduled date, and nobody in the chain had an obligation to stop the clock.

The Indicators That Were Never Indicators

Fire investigation developed as a trade rather than a science. Practitioners learned to read a burned structure from more experienced practitioners, and the reading conventions were transmitted through training courses and apprenticeship rather than through research. By the middle of the twentieth century a stable body of doctrine existed, and it was taught with confidence.

The doctrine held that certain features of a fire scene revealed intent. Crazed glass, meaning a fine web of cracks in window panes, indicated rapid heating from an accelerant. Alligatoring, meaning large shiny blisters in charred wood, indicated an unusually hot or fast fire. Spalled concrete indicated a chemical burning on the surface. Low burn patterns and burning beneath furniture indicated that fuel had been poured, since fire normally moves upward. Irregular patterns on flooring were read as pour patterns tracing where an accelerant had been distributed.

Each of those readings has a surface logic. None of them was ever tested against a control set of accidental fires, which is the step that would have revealed the problem, because every one of those features occurs routinely in fires that no person set.

The explanation is flashover. As a fire burns in an enclosed space, hot gases collect at the ceiling and radiate downward until every combustible surface in the room reaches ignition temperature and the entire compartment goes up at once. Post-flashover fires burn fast, burn hot, burn at floor level, and leave irregular patterns where ventilation and furnishings happened to concentrate the heat. A room that has flashed over presents, in other words, the complete checklist.

Corsicana, 1991

On December 23, 1991, a fire in Cameron Todd Willingham’s home in Corsicana, Texas killed his three young daughters. Investigators from the local fire department and the state fire marshal’s office concluded that the fire had been intentionally set, relying on burn damage they read as arson indicators. Willingham was convicted of murder on August 21, 1992 and sentenced to death.

The case also rested on a jailhouse informant, Johnny Webb, who testified that Willingham had confessed. Webb later recanted. Documents obtained years afterward indicated that the Navarro County district attorney had assisted Webb in ways that were not disclosed to the defense.

A month before the scheduled execution, Willingham’s file reached Gerald Hurst, an Austin fire scientist. Hurst reviewed the investigation and concluded that the features the investigators had treated as proof of arson were simply the ordinary residue of a fire. On February 13, 2004, Willingham’s attorneys released the Hurst report and sent it to Governor Rick Perry and the Texas Board of Pardons and Paroles along with a clemency request. The Navarro County district attorney’s office responded that the report was beside the point given the informant testimony.

No reprieve was granted. Willingham was executed on February 17, 2004, and pronounced dead at 6:20 that evening. He maintained his innocence in his final statement.

Ionia, 1985

Michigan has its own version of this case, and it is closer to home than most readers of the Willingham coverage realize.

On the night of March 9, 1985, a fire in David Lee Gavitt’s home in Ionia, Michigan killed his wife Angie and their two young daughters. Gavitt was badly burned. He escaped by breaking a window and going out into the snow, and neighbors physically restrained him from going back into the house.

Local investigators suspected arson based on burn patterns inside the home. A Michigan State Police crime laboratory test on carpet samples reported traces of gasoline. Gavitt was convicted on April 18, 1986 of arson and three counts of felony murder and sentenced to life without the possibility of parole. No accelerant container was ever found, and no motive was established.

The University of Michigan Innocence Clinic accepted the case in 2010 and retained fire experts, among them John Lentini, who had also reviewed the Willingham evidence. Lentini’s affidavit concluded there was no credible evidence that the Gavitt fire was intentionally set. The evidence was re-examined independently by the Michigan State Police, the federal Bureau of Alcohol, Tobacco, Firearms and Explosives, and a private laboratory. The Ionia County prosecutor reviewed the case, agreed the trial evidence was unsound, and joined the motion to vacate. Gavitt walked out of the Carson City facility on June 6, 2012, carrying two steamer trunks and a typewriter, and went to see his family’s graves for the first time.

The Second Error Nobody Counted

The Gavitt case contains two distinct failures, and the second one has received almost no attention.

The first is the doctrinal failure this installment has been describing. Investigators read burn patterns through a framework that fire research had already undermined. That failure is systemic, it affected every jurisdiction, and it is the reason a national reexamination of arson convictions became necessary.

The second is narrower and, on the record, more troubling. The Michigan State Police crime laboratory reported gasoline in the carpet samples. That finding was wrong. The Ionia County prosecutor acknowledged as much when he joined the motion, describing the laboratory work on the carpet as botched. The analyst responsible had died before the error came to light.

A doctrinal failure is diffuse. An analytical error by a specific person in a specific laboratory is not. It has a name attached to it, a date range, and a caseload.

This is the point where Part II and Part III meet. The hair installment argued that a witness and testimony register would have made affected cases findable at scale. The Gavitt record shows what its absence costs even when the error is not doctrinal but personal, discrete, and documented. One analyst’s work product was demonstrated to be unreliable in a case that produced a life sentence, and the demonstration produced no inquiry into anything else that analyst had touched.

Three Files, One Error, Three Outcomes

Put the three cases beside each other and the argument makes itself.

The evidentiary error is constant across all three. What varies is whether a person with capacity and standing reached the file while a remedy still existed, and none of those three outcomes was produced by a system designed to produce it. One turned on an open federal proceeding. One turned on a clinic’s intake decision twenty-five years later. One turned on nothing, because the clock ran out first.

What Fire Adds to the Validation Lag

Part II described the lag as the interval between admission and validation, and described the failure that follows validation as the absence of any duty to notify. Fire adds a third element, and it is the one that should govern how this series is read.

Some remedies expire.

That sounds obvious stated plainly, and it is almost entirely absent from how forensic reform is discussed. Reform conversations treat the problem as a knowledge problem, and knowledge problems feel like they resolve on their own timeline. Fire cases demonstrate that the timeline is not neutral. While an institution deliberates about whether a discipline is sound, the population affected by that discipline continues to move through a process with its own dates, and some of those dates are terminal.

Why This Matters Beyond These Cases

Let me close where I opened, because the arithmetic here is unusually clean.

The Ionia fire was in March 1985. The Corsicana fire was in December 1991. The doctrine applied in both was the same doctrine, and by 1991 it was already being dismantled in the fire science literature that neither investigating agency was reading. Willingham was executed in February 2004. Gavitt was released in June 2012.

Between those two dates, eight years and four months, nothing changed about the underlying science. It had been settled before the first of them. What separated the two outcomes was entirely procedural: whether a case was still open, whether a clinic had capacity, whether a prosecutor was willing, and whether a date had already passed.

That is not a scientific dispute resolving itself over time. That is a lottery, and the tickets were distributed by the calendar.