Chapter 22: Arson Investigation: The Science of Fire and the Myths That Convicted the Innocent

"Each and every one of the indicators that I have relied upon to establish that this was an incendiary fire is no longer valid." — the substance of the conclusion reached by fire scientists who re-examined the Cameron Todd Willingham case after his execution [paraphrase of the public findings of the Hurst and Beyler reports, 2004–2009; see §22.6]. A man was already dead when the science that convicted him was withdrawn.

Overview

A house fire is the rare crime scene that consumes itself. By the time anyone reads it, the fire has destroyed much of the evidence of how it began, rearranged the rest, and left behind a chaos of char that has to be interpreted — and for most of the twentieth century, the people interpreting it were working from a body of "knowledge" that almost no one had ever tested. They had learned, from mentors who had learned it from their mentors, that certain marks on a burned building were the unmistakable fingerprints of arson: glass that had cracked into a spiderweb of fine lines, wood charred into shiny blisters like alligator skin, scorch marks on the floor that traced where a liquid had supposedly been poured and lit. An investigator who found these "indicators" would testify, with complete sincerity, that the fire had been set deliberately. Juries believed him, because why wouldn't they? He was the expert.

He was also, it turned out, wrong about nearly all of it. When fire scientists finally did the experiments — building rooms, setting them alight under controlled conditions, and watching what accidental fires actually do — they discovered that the celebrated "arson indicators" are produced routinely by ordinary fires that no one set. Crazed glass comes from water hitting hot glass, not from accelerant. The low burn patterns and "pour" marks are the signature of flashover, the moment a room becomes so hot that everything in it ignites at once — a phenomenon the old folklore had never accounted for. The science that had sent people to prison was not science at all. It was tradition wearing a lab coat, and in at least one case we will study in painful detail, it sent a very likely innocent man named Cameron Todd Willingham to the execution chamber in Texas in 2004.

This chapter is therefore two things at once, and you must hold both. It is a genuine science chapter — fire is a physical process governed by chemistry and thermodynamics, and the modern discipline of fire investigation, anchored by a national standard called NFPA 921, is real and improving. And it is the book's hardest case study in what happens when an unvalidated method is allowed to testify to certainty. The two are not in tension; they are the same lesson. The way you tell the real fire science from the folklore is the way you tell any valid method from a junk one — you ask whether its claims have been tested against ground truth, and you flinch when the answer is no.

In this chapter, you will learn to:

• Explain how fire actually behaves — heat transfer, fuel and oxygen, and flashover — and why fire dynamics is the foundation of everything else.
• Describe origin and cause determination, the NFPA 921 standard that governs it, and why "undetermined" is often the honest answer.
• Name the debunked fire indicators — crazed glass, alligatoring, "pour" patterns — and say exactly what each does and does not prove.
• Explain the negative corpus fallacy and how flashover produces the marks once mistaken for arson.
• Describe how ignitable-liquid residue is collected, and why only the laboratory (Chapter 23), not the scene investigator, can confirm an accelerant.
• Use the Willingham and Han Tak Lee cases to see fire science's human cost, and place the discipline honestly on the validity spectrum.

Learning Paths

🔎 Investigator/CSI: This is your scene more than almost anyone's, and the discipline of not concluding too early is the whole job. Weight §22.2 (how a cause is honestly determined) and §22.5 (collecting residue so the lab can do what your eyes cannot). A burn pattern you over-read at the scene is a wrongful conviction you started. 🧪 Lab analyst: You confirm — or refute — what the scene investigator suspects. §22.5 hands you the debris; the instrumental confirmation is Chapter 23. Note that your "gasoline confirmed" is the only step in this whole chain that rests on validated analytical chemistry, and that the absence of residue does not mean the absence of an accelerant. ⚖️ Law/courtroom: This chapter is a litigator's case study in junk science admitted and a man executed. Weight §22.3, §22.4, and §22.6. The Daubert gate (Chapter 5) should have stopped the indicators; learn why it didn't, and how a post-conviction fire-science challenge is built. 👥 General reader/juror: If you remember one thing, remember that "the fire investigator said it was arson" is exactly the sentence that killed Willingham. §22.1 and §22.6 teach you why a confident expert and a real science are not the same thing.

22.1 How fire behaves: fire dynamics and flashover

You cannot read a fire scene until you understand what a fire is, because every mark the investigator interprets is the frozen record of a physical process. The single greatest failure of the old fire investigation was that it tried to read the marks without understanding the process — it catalogued patterns without a working physics of how fires create them. The modern discipline begins the other way around, with fire dynamics: the study of fire as a phenomenon governed by chemistry and the laws of heat and mass transfer. Get the dynamics right, and most of the folklore collapses on its own.

Start with the chemistry, which is simpler than it sounds. Fire is combustion — a rapid, self-sustaining chemical reaction, usually between a fuel and the oxygen in the air, that releases heat and light. It needs three things present at once, the trio traditionally drawn as the fire triangle: a fuel (something that will burn), an oxidizer (almost always atmospheric oxygen), and heat (enough energy to start and sustain the reaction). Remove any one leg and the fire goes out — which is exactly how extinguishers work, by smothering the oxygen, cooling the heat, or cutting off the fuel. (Modern texts often add a fourth leg, the self-sustaining chemical chain reaction, making a "fire tetrahedron," but the triangle is enough for our purposes.) The crucial point for an investigator is that fire is not a thing; it is a reaction, and reactions obey rules. Where it spreads, how hot it gets, what it leaves behind — none of it is arbitrary.

Fire spreads by moving heat, and it does so in three ways you should be able to name because they explain burn patterns. Conduction is heat moving through a solid material — a metal pipe carrying fire's heat from one room to the next. Convection is heat carried by a moving fluid, and in a fire the fluid is hot gas: the buoyant column of hot combustion products rises, which is why fire climbs and why the worst damage is so often high and above the burning fuel. Radiation is heat traveling as electromagnetic energy across open space, the way you feel a bonfire's warmth on your face without touching it; radiant heat is what preheats nearby objects until they, too, are ready to ignite. Of the three, convection and radiation do most of the work of spreading a room fire, and both push heat upward and outward from the burning fuel. Hold that geometry; it will matter enormously when we get to "pour patterns."

🔬 At the Bench A concept that quietly demolishes a lot of old arson folklore is the heat release rate — how much energy a burning item puts out per unit time, measured in kilowatts or megawatts. Folklore assumed that a fast, hot, intense fire must mean an accelerant, because "ordinary" household fires were imagined to be slow and cool. But the heat release rate of perfectly ordinary modern furnishings is enormous. A single upholstered chair or a polyurethane-foam couch can release heat at a rate of one to two megawatts once fully involved — comparable to a small gasoline fire — because modern furniture is essentially solid petroleum. The "unnaturally fast and hot" fire that investigators once read as proof of a poured accelerant is, in a room full of synthetic furnishings, simply what a normal fire does. The science did not get more permissive; the fuel load of the average living room got far more flammable over the twentieth century, and the folklore never noticed.

Now the single most important idea in this entire chapter, the one whose absence from the old folklore did the most damage. A fire in an enclosed room does not simply burn the object that ignited and then spread item by item in a leisurely way. As the first fuel burns, the hot gas layer at the ceiling grows hotter and deeper, and it radiates heat back down onto everything in the room. At a critical point — often a ceiling-gas temperature in the rough neighborhood of $500\text{–}600^{\circ}\text{C}$ — that downward radiation becomes intense enough to ignite, nearly simultaneously, the exposed surfaces of every combustible thing in the room at once. The carpet, the curtains, the magazines on the table, the far side of the sofa: all of it bursts into flame together. This transition is called flashover — the near-simultaneous ignition of all exposed combustible surfaces in an enclosed space, driven by radiant heat from the accumulated hot gas layer, marking the change from a growing fire to a fully developed one.

THE PHASES OF A ROOM FIRE  (illustrative — temperatures and times are schematic, not to scale)

  TEMP
   ↑                                   ┌──────────── FULLY DEVELOPED ────────────
   │                                  /  (whole room burning; floor-level damage,
   │                                 /    "low burns," can appear NOW — no accelerant
   │            ── FLASHOVER ───────/     required)
   │           (≈500–600°C ceiling /
   │            gas → everything   /
   │            ignites at once)  /
   │                            /
   │   GROWTH                 /
   │  (one item burns,      /
   │   hot gas layer      /
   │   builds at ceiling)
   │ ____________________/
   │/ IGNITION
   └────────────────────────────────────────────────────────────────→ TIME
   Legend: the marks the old folklore read as "arson signatures" appear AFTER flashover
           in an ORDINARY accidental fire. Flashover is the event the folklore ignored.

Walk the diagram, because it reorganizes the whole chapter. In the ignition and growth phases, a single item burns and the hot gas layer builds at the ceiling — this is the fire the old folklore implicitly imagined, spreading from one thing to the next. But once the room reaches flashover, it enters the fully developed phase, in which everything burns at once, top to bottom, including the floor, the undersides of furniture, and the low corners. Here is the catastrophe hiding in plain sight: many of the marks the old investigators treated as the exclusive signatures of a poured accelerant — burning at floor level, damage under furniture, irregular patterns on the floor — are produced routinely by an ordinary accidental fire that simply reached flashover. The folklore had catalogued the appearance of a post-flashover room and labeled it "arson," never realizing it was looking at the normal end-state of a fully developed fire. An entire discipline mistook a phase of combustion for a crime.

🔍 Check Your Understanding 1. Name the three legs of the fire triangle, and state what happens to "fire is unusually hot, so someone used accelerant" once you know the heat release rate of a modern foam sofa. 2. In your own words, what is flashover, and why does its existence undermine the reading of low, floor-level burn damage as proof of a poured liquid?

Two more dynamics round out the foundation. Ventilation — the supply of air — controls a fire as powerfully as fuel does; a fire starved of oxygen smolders, and one fed by an open door or a broken window roars, which means burn patterns often track where the air came from rather than where a fire was "set." And fires, left alone, tend to burn upward and outward from their origin in a rough V-pattern on a wall, because of that buoyant convective plume — a genuinely useful indicator of origin, when it has not been overwritten by flashover. The honest modern investigator reads all of this together: not a checklist of "arson signs," but a physical reconstruction of how this fire, in this room, with this fuel and this ventilation, actually behaved. That reconstruction is the subject of the next section.

22.2 Origin and cause determination

The fire investigator's central task has a name, and it is a pair of linked questions. Origin and cause determination is the systematic effort to establish where a fire began (the origin) and what started it (the cause). The two are sequential and dependent: you cannot reliably determine the cause until you have correctly located the origin, because the cause is whatever competent, accidental, or deliberate mechanism was present at the origin to bring fuel, oxygen, and heat together. Find the wrong origin and every later inference is built on sand — and as we will see, finding the wrong origin is exactly how the worst fire-investigation failures begin.

Fire investigation — the discipline of determining a fire's origin, cause, and development through systematic examination of the scene and the physical evidence — is supposed to proceed by the scientific method, the same one Chapter 5 described. The investigator observes the scene, gathers data, forms hypotheses about origin and cause, and then tests each hypothesis against all the data, discarding those the evidence refutes. The conclusion is supposed to be the hypothesis that survives. This is not how the old folklore worked; the old folklore pattern-matched the scene against a memorized list of "indicators" and announced a verdict. The difference between testing hypotheses and matching patterns is the difference between the modern discipline and the one that killed Willingham.

Since 1992, the discipline has had a written standard, and you need to know its name. NFPA 921, Guide for Fire and Explosion Investigations, published by the National Fire Protection Association and revised regularly, is the consensus document that establishes the methodology of modern fire investigation. It did something the field had never had: it wrote down, in one authoritative place, that fire investigation must follow the scientific method, it catalogued which traditional "indicators" had been invalidated by research, and it set out how origin and cause should actually be determined. NFPA 921 is not a statute and does not bind investigators by law, but it has become the de facto standard of care, the benchmark against which competent fire investigation is now measured — and the document a good attorney will hold up in court to ask whether the investigator followed it.

⚖️ In the Courtroom Why does a non-binding "guide" carry so much courtroom weight? Because after Daubert (Chapter 5), the question a court must ask of expert testimony is whether it rests on a reliable methodology — and NFPA 921 is the fire-investigation field's own written statement of what a reliable methodology is. An investigator whose conclusions follow NFPA 921 can point to a peer-reviewed, consensus standard. An investigator who reaches a finding of arson using methods NFPA 921 specifically labels invalid — the old indicators — is, in effect, conceding that his own field's standard does not support him. Post-Daubert, "the National Fire Protection Association's guide says the indicator I relied on is not valid" has become one of the most powerful sentences a defense attorney can put to a fire investigator on cross. The standard did not just improve the science; it gave the courtroom a yardstick.

The honest determination of cause sorts a fire into one of a few classifications, and the discipline of the classification matters as much as the science. A fire's cause may be ruled accidental (an unintended mechanism — faulty wiring, an unattended candle, a discarded cigarette), natural (lightning, an earthquake-ruptured gas line), incendiary (deliberately set under circumstances in which the person knew it should not be — the fire-science term for what the law calls arson), or — and this is the category the old folklore could never tolerate — undetermined. A fire is properly classified as undetermined when the evidence does not allow the investigator to establish the cause to a reasonable degree of scientific certainty. "Undetermined" is not a failure or an embarrassment; very often it is the only honest finding a destroyed scene permits. A great deal of the harm in this chapter flows from investigators who refused to say "undetermined" and reached for "incendiary" instead, on the strength of indicators that proved nothing.

⚠️ Junk-Science Alert The most dangerous habit in fire investigation is reasoning to arson by elimination without ever having ruled out flashover or an undetermined cause — the negative corpus error we will name fully in §22.4. In its classic form it goes: "I examined the scene, I could not find an accidental cause, therefore the fire was deliberately set." That is a logical fallacy dressed as a determination. The absence of an identified accidental cause is not evidence of arson; it may simply mean the accidental cause was destroyed by the fire, or that the scene was too damaged to read, in which case the honest classification is undetermined. A finding of incendiary must rest on affirmative evidence that the fire was deliberately set — not on the investigator's failure to find something else. When you hear "we ruled out all accidental causes, so it must be arson," you are hearing the exact reasoning that the modern standard, and this chapter, exist to stop.

So what does support an honest finding of an incendiary fire? Affirmative, physical evidence of deliberate ignition that the science can actually stand behind. The two strongest are the ones the cold case will turn on. Multiple, separate, unconnected points of origin — two or more independent places where the fire clearly started, with no physical pathway by which a single accidental fire could have produced both — are difficult to explain innocently, because accidental fires begin in one place. And a confirmed ignitable liquid in a pattern consistent with deliberate distribution — gasoline poured in a trail or a pool and then ignited — is strong evidence of incendiary cause, provided the liquid is confirmed by laboratory instrumentation (§22.5, Chapter 23) and not merely guessed from a scorch mark. These are the valid grounds. The contrast between them and the debunked indicators is the spine of the chapter, and the next section dismantles the indicators one at a time.

22.3 The debunked indicators: crazed glass, alligatoring, pour patterns

For decades, fire investigators carried in their heads a checklist of visual "indicators" that were taught as the reliable signatures of an accelerated, deliberately set fire. They were passed from instructor to trainee as settled knowledge, repeated in manuals, and testified to in courtrooms as fact. Almost none of them survived contact with controlled experimentation. The debunked fire indicators are the set of once-trusted visual signs of arson — crazed glass, alligator charring, low and irregular burn patterns read as "pour patterns," and others — that systematic fire-science research has shown to be unreliable, because each is produced routinely by ordinary accidental fires. Learn them not because you should ever use them, but because you must be able to recognize them when an investigator reaches for them, and know why each one fails.

Crazed glass. The folklore: window glass that has cracked into a dense network of fine lines ("crazing") shows that the fire burned with unnatural speed and intensity — the kind of rapid heating an accelerant produces. The reality, established by experiment: crazing is caused by rapid cooling, not rapid heating — specifically, by cold water striking hot glass, which is to say, by the fire hose. The very act of extinguishing the fire produces the "indicator" that was said to prove arson. A pane that crazes is telling you that a firefighter sprayed water on it while it was hot, which happens at essentially every structure fire. As a sign of arson, crazed glass is worse than useless; it is an artifact of putting the fire out.

Alligatoring. The folklore: when charred wood forms a pattern of shiny, glossy blisters in small, rounded segments — said to resemble alligator skin — it indicates a fast, hot fire and therefore an accelerant; large, dull, flat char blisters supposedly indicated a slow, "normal" fire. The reality: the size and shininess of char blisters do not reliably indicate the fire's speed, temperature, or the presence of an accelerant. Char patterns are influenced by the type of wood, the ventilation, the duration of burning, and many other variables; the neat "small-and-shiny-means-accelerant" rule was never validated and does not hold. Alligatoring tells you wood burned. It does not tell you how fast, how hot, or whether anyone poured anything.

Low burns and "pour patterns." This is the most consequential of the indicators, the one that put accelerant into the most courtrooms, so spend the most time on it. The folklore: irregular burn patterns on the floor — especially low burning, damage beneath furniture, and pooled or trailing shapes — show that a liquid accelerant was poured on the floor and ignited, since "fire burns up, so it shouldn't burn the floor unless something on the floor was made to burn." Investigators called these "pour patterns" or "puddle-shaped" burns and read them as a map of where the arsonist splashed the gasoline. The reality is the single most important correction in this chapter, and it is the one that exonerates Willingham: post-flashover fires routinely produce low, irregular, pool-shaped burn patterns on the floor, with no accelerant present at all. Once a room flashes over (§22.1), the burning hot-gas layer fills the entire space and burns everything, including the floor, the undersides of furniture, and low corners — producing exactly the irregular floor patterns the folklore attributed to poured liquid. Radiant heat, falling burning debris, melting and burning plastics that pool and then ignite, and ventilation effects all create floor-level "pour-shaped" damage in ordinary accidental fires. The pattern that was read as a fingerprint of arson is, in a post-flashover room, just what a fully developed fire looks like.

🔬 Read the Evidence

text FIGURE 22.1 — "The 'pour pattern' that wasn't" [after the Willingham-era fire research, public record] THE ITEM An irregular, roughly pool-shaped area of deep charring on a living-room floor, with low burning along the base of the walls and damage to the undersides of furniture. To a 1980s investigator: a textbook "pour pattern" proving a poured accelerant. THE CONTEXT A single-room structure fire, fully extinguished. The room is known (from later controlled burns of similar rooms) to have reached FLASHOVER before suppression. No debris has yet been sent to a laboratory. WHAT IT SHOWS Severe, low, irregular floor-level fire damage. That is all it shows on inspection: that this part of the floor burned intensely. The shape is real; its CAUSE is not visible to the eye. WHAT IT DOESN'T It does NOT show that a liquid was poured. Controlled post-flashover burns produce indistinguishable patterns with no accelerant. The eye cannot tell a gasoline pour from a flashover burn from a melted-and-burning-plastic pool. Only laboratory analysis of the debris (Ch. 23) can confirm or exclude an ignitable liquid. THE INFERENCE The HONEST reading is: "low-level floor damage consistent with a fully developed, post-flashover fire; the presence of an ignitable liquid can be neither confirmed nor excluded without laboratory testing." NOT "a pour pattern proving arson." THE LESSON A burn pattern is a question, not an answer. The shape on the floor is data; "someone poured gasoline here" is a hypothesis that only the lab can test. Reading the pattern AS the conclusion is the precise error that executed a man.

The list goes on, and they fail for the same reason. "Spalling" of concrete (chips flaking off the surface) was said to mark where accelerant burned; in fact it has many causes and does not indicate accelerant. Burn marks under a door's threshold or a "trailer" of fire connecting two areas were over-read. The melting of steel or the "depth of char" were converted into temperature claims the materials cannot support. In every case the structure of the error is identical: a visual feature that can occur in many ways was treated as if it could occur in only one way — a poured accelerant — and the investigator's confidence supplied a certainty the evidence never contained. This is the class-versus-individual error from Chapter 1 (§1.3) in a new costume: a class characteristic (this floor burned severely), present in countless ordinary fires, was paraded as an individual signature of a specific cause.

🧠 Cognitive-Bias Watch The fire indicators are a near-perfect engine for confirmation bias (Chapter 31), and you should see the mechanism because it recurs across this book. An investigator arrives already suspecting arson — perhaps because someone died, perhaps because the homeowner had money troubles, perhaps just because the fire "felt wrong." He then looks for indicators, and because the indicators (crazed glass, low burns) are present at almost every fire, he finds them. Their ubiquity, which is exactly why they are worthless, makes them perfect confirmation fodder: whatever you suspect, the scene will seem to confirm. The expectation selects the evidence, the "evidence" confirms the expectation, and the investigator leaves more certain than he arrived, having learned nothing the fire did not let him project onto it. The safeguard is the same one the whole book preaches: form the origin-and-cause hypothesis from the physics, test it against all the data including the data that would refute it, and keep the case context — the dead child, the insurance policy, the "feeling" — out of the technical determination until it is fixed.

22.4 Flashover and the "negative corpus" problem

We have met flashover as a physical event (§22.1) and seen it dissolve the "pour pattern" (§22.3). Now we make it the center of the chapter's logic, because flashover is not merely one more complication — it is the phenomenon whose existence breaks the entire reasoning structure of old arson investigation. And it pairs with a named logical fallacy that the modern standard has explicitly repudiated. Together, flashover and the "negative corpus" problem are how an honest-seeming investigation arrives at a false finding of arson.

Recall what flashover does to a room (§22.1): past a critical hot-gas temperature, the radiant heat ignites every exposed surface at once, and the room enters the fully developed phase in which everything burns, everywhere, including low and on the floor. Now consider what this means for the investigator who arrives afterward. The marks that the old folklore treated as diagnostic of arson — low burning, floor-level damage, multiple areas of severe burning, "pour-shaped" patterns — are precisely the marks a post-flashover room displays regardless of how the fire started. An accidental fire from an overloaded outlet, given enough fuel and air to reach flashover, will leave a scene that looks, to the indicator-trained eye, exactly like an "obvious" arson. The indicators cannot distinguish set fires from accidental ones because flashover erases the distinction they were thought to capture. This is why the discovery of flashover's role was so devastating to the old discipline: it did not just retire a few indicators; it revealed that the indicators had never been measuring what they claimed to measure at all.

Into this confusion steps the fallacy. Negative corpus — short for the Latin negative corpus delicti, "the body [of the crime] by negation" — is the discredited reasoning by which an investigator concludes a fire was incendiary not from affirmative evidence of a deliberately set fire, but purely from the elimination of all accidental and natural causes the investigator could identify. The argument runs: "I inspected the wiring, the appliances, the heating system; I found no accidental cause; therefore the fire must have been intentionally set." NFPA 921 and the modern field have explicitly rejected this reasoning, and the reason is rigorous. Eliminating identified accidental causes does not establish a deliberate cause, for two reasons that compound. First, the accidental cause may have been destroyed by the fire itself — a failed appliance can be consumed beyond recognition, so "I found no accidental cause" may mean only "the fire ate the evidence of its own accidental origin." Second, the proper classification when the cause cannot be affirmatively established is "undetermined," not "incendiary." The negative-corpus investigator skips that honest box entirely, treating his own inability to find an accidental cause as positive proof of arson.

⚠️ Junk-Science Alert Watch how flashover and negative corpus combine into a self-confirming machine that can convict an innocent person without a single dishonest actor. (1) An accidental fire reaches flashover, producing low burns and "pour-shaped" floor patterns. (2) The investigator, trained on the old indicators, reads those flashover artifacts as affirmative evidence of a poured accelerant. (3) Reinforced by that "evidence," he inspects for accidental causes, finds none he can identify (the fire may have destroyed it), and invokes negative corpus: no accidental cause found, therefore arson. (4) He testifies, sincerely, that the physical evidence proves the fire was deliberately set. Every step feels like diligence. Not one step is valid. The flashover artifacts are not accelerant evidence; the failure to find an accidental cause is not proof of a set fire; and "undetermined" — the honest classification — never gets considered. This is not a hypothetical chain. In its essentials it is the Willingham investigation (§22.6).

The corrective is the affirmative-evidence standard, and it is worth stating as a positive rule because it is the heart of competent modern practice. A finding of incendiary cause must rest on affirmative physical evidence that the fire was deliberately set — for example, a confirmed ignitable liquid in a deliberate distribution pattern, or genuinely multiple and unconnected points of origin, or an incendiary device. It may not rest on the elimination of accidental causes alone. If the affirmative evidence is absent and accidental causes cannot be confidently established either, the scientifically honest classification is undetermined. This is the same evidentiary discipline the entire book teaches, applied to fire: the burden is to support a conclusion with evidence, not to back into it by failing to disprove it. "I couldn't find another explanation, so it must be the one I suspected" is the structure of nearly every junk-science conviction in these pages, and fire investigation is where it has been deadliest.

🔍 Check Your Understanding 1. Explain in one or two sentences why the existence of flashover makes "low, floor-level burn patterns" useless as proof of a poured accelerant. 2. State the negative corpus fallacy in your own words, and give the two reasons "I found no accidental cause" does not establish arson. What is the honest classification instead?

There is a deeper point here about the validity spectrum (Chapter 6, §6.6), and it is the one to carry out of this section. Modern fire dynamics — flashover, heat release rate, ventilation, the physics of compartment fires — is genuine, tested, quantitative science; you can build a room, instrument it, set it alight, and measure what happens, and researchers have. The old indicator-based investigation was the opposite: a set of untested visual rules with no measured error rate, exactly the kind of method the 2009 NAS report and the 2016 PCAST framework condemn. So fire investigation does not sit at one place on the validity spectrum; it spans it, and which fire investigation you are looking at determines everything. Origin-and-cause work grounded in fire dynamics and confirmed ignitable-liquid chemistry can be solid. Origin-and-cause work resting on the debunked indicators and negative corpus is folklore that has killed. The investigator's name and confidence tell you nothing about which one you have; only the methodology does.

22.5 Ignitable-liquid residues: collecting and confirming (→ Chapter 23)

If the visual indicators cannot prove an accelerant, what can? The answer is chemistry, and it moves the decisive question off the scene and into the laboratory — which is the most important structural fact in modern arson investigation. The scene investigator can suspect an accelerant and can collect the debris that might contain it, but only an instrument in a lab can confirm it. Drawing that line clearly is what separates the modern discipline from the folklore, so we define the terms with care.

An accelerant is any substance used to initiate or speed the spread of a fire — most commonly an ignitable liquid such as gasoline, kerosene, or charcoal lighter fluid, though the term can include other materials. When such a liquid is used and the fire is then extinguished, traces of the unburned or partially burned liquid can soak into porous materials at the scene — carpet, padding, flooring, soil, upholstery, charred wood — and survive the fire. The chemical traces that remain are called ignitable-liquid residue (ILR): the detectable chemical remnants of an ignitable liquid recovered from fire debris, identified by laboratory analysis. The whole valid path to an accelerant runs through ILR, and ILR is a laboratory finding, never a scene-investigator's eyeball judgment. "I can smell gasoline" or "that looks like a pour pattern" is a reason to collect a sample; it is not a confirmation of anything.

🔬 At the Bench Collection technique decides whether the lab ever gets a chance, and the rules are unforgiving because the target evidence is volatile — it evaporates. The investigator identifies areas most likely to retain residue (often the most protected, least-burned porous material near a suspected origin, plus the lowest points where liquid would pool and soak in), and collects a generous sample of that material into an airtight, vapor-tight container — most commonly a clean, unused metal paint can with a tight lid, or a specialized nylon evidence bag, never an ordinary plastic bag (whose plastic can both leak the vapors and contribute petroleum-based contaminants that mimic an accelerant). The container is sealed immediately, because every minute of exposure lets the volatile residue escape. Crucially, the investigator also collects a comparison sample of the same kind of material from an area away from the suspected accelerant — an unburned piece of the same carpet, for instance — so the lab can distinguish a genuine foreign ignitable liquid from the ordinary petroleum-derived background that modern materials normally emit when heated. (Carpet, adhesives, and synthetic padding are made from petroleum; heated, they release hydrocarbons that an untrained nose or a careless analysis could mistake for "accelerant." The comparison sample is the control that guards against that error.)

What happens to the can in the laboratory is the subject of Chapter 23, but the principle belongs here because it completes the logic of valid arson investigation. In the lab, the sealed container is gently warmed so any volatile residue evaporates into the air space above the debris (the "headspace"), that vapor is captured — often onto a small adsorbent strip — and then analyzed by gas chromatography–mass spectrometry (GC-MS), the instrumental method Chapter 23 teaches in full. GC-MS separates the complex mixture of compounds and identifies them, and a trained analyst compares the resulting chemical "fingerprint" against reference patterns for known ignitable liquids. Gasoline, for example, produces a characteristic and recognizable pattern of aromatic compounds. The output is a genuine analytical-chemistry identification, with the kind of validated, reproducible foundation that places it near the top of the validity spectrum — the same tier as DNA and toxicology, and a different universe from the visual indicators of §22.3.

Two honest limits keep even this strong method from being overstated, and a competent analyst states both. First, absence of residue is not absence of an accelerant. A fire hot enough and long enough can consume an ignitable liquid completely, leaving no detectable residue; a negative laboratory result therefore does not prove that no accelerant was used — it proves only that none was detected. This asymmetry matters: ILR is strong evidence when present and weak evidence when absent, the same exclusion-versus-proof asymmetry from Chapter 1 (§1.6). Second, the presence of an ignitable liquid is not, by itself, proof of arson. Gasoline and other ignitable liquids have countless legitimate reasons to be present in a home or a garage — a lawnmower, a can of solvent, stored fuel — so a confirmed ILR must be interpreted in context: its location, its distribution, its quantity, and whether an innocent explanation fits. A trace of gasoline in a garage where a mower is stored proves nothing; a trail of gasoline poured across a bedroom floor and ignited is another matter. The chemistry confirms the liquid; the investigation must still establish that its presence was incendiary.

⚖️ In the Courtroom The division of labor here is also a division of testimony, and good lawyers exploit the seam. The laboratory analyst can testify, on solid analytical-chemistry grounds, that "the chemical pattern recovered from this carpet is consistent with gasoline." That is a strong, defensible statement. The scene investigator's testimony that the gasoline was "poured deliberately in this pattern to set the fire" is a far weaker inference resting on scene interpretation, and it is exactly where the old folklore creeps back in. A careful cross-examination separates the two: the chemist's "this is gasoline" may be unassailable while the investigator's "therefore it was arson" is challenged hard. Conflating the validated laboratory finding with the contestable scene interpretation — letting the strength of "it's gasoline" rub off onto "it was deliberately poured" — is one of the most common overstatements in arson testimony. The chemistry and the conclusion are two different claims with two different strengths.

22.6 Willingham and Han Tak Lee: when bad fire science kills

Everything in this chapter converges on two cases. They are not illustrations chosen for color; they are the reason the chapter exists, and the reason the field reformed. Both are matters of public record, and both must be handled with sobriety, because in one of them a very likely innocent man was executed and in the other a man spent decades in prison, on the strength of fire science that has since collapsed. We will state only documented public facts, and we will not invent quotations, dates, or numbers.

Cameron Todd Willingham. On the morning of December 23, 1991, a fire swept through the Willingham home in Corsicana, Texas, and killed his three young children — two-year-old Amber and one-year-old twins, Karmon and Kameron. Willingham escaped the burning house. Fire investigators examined the scene and concluded the fire had been deliberately set, and their conclusion rested squarely on the debunked indicators this chapter has dismantled: "crazed glass" said to show an unnaturally hot fire; multiple low burn areas and "pour patterns" on the floor said to show a poured liquid accelerant; charring said to indicate the fire's intensity and origin in a way consistent only with arson. On the strength of this fire science — together with the testimony of a jailhouse informant who claimed Willingham had confessed (informant testimony of the kind Chapter 6 flagged as a recurring cause of wrongful conviction), and a portrait of Willingham's character as the kind of man who would commit such an act — he was convicted of capital murder in 1992 and sentenced to death.

Here is the part this chapter exists to make you confront. By the time of Willingham's scheduled execution, the fire science underpinning his conviction was already known to be wrong, and qualified fire scientists said so. A noted fire scientist and chemist, Dr. Gerald Hurst, reviewed the case shortly before the execution and concluded that the original investigation did not support a finding of arson — that the supposed indicators were artifacts of an ordinary fire and post-flashover behavior, exactly as controlled fire research had by then established. That assessment reached the authorities with the power to halt the execution. It did not stop it. Cameron Todd Willingham was executed by the State of Texas on February 17, 2004. After his death, the Texas Forensic Science Commission undertook a review of the fire-science evidence, and commissioned an independent analysis by another respected fire expert, Dr. Craig Beyler, whose report found that the original investigators' conclusions could not be supported by modern fire science and did not comport with the standards of their own field. Across multiple independent expert reviews, the verdict on the fire science was consistent: every indicator used to call the fire incendiary was invalid, and the evidence did not support arson. The fire that killed the Willingham children may well have been an accident; what is established is that the science used to call it arson was folklore.

🧠 Cognitive-Bias Watch It is essential — and far more unsettling — to understand that the Willingham investigators were, in all likelihood, not villains. They were applying the methods their entire field had taught them and believed in, with no idea those methods had never been validated. That is the deepest lesson of the case and of this book: junk science does not usually arrive as deliberate fraud. It arrives as the sincere, confident, good-faith application of an unvalidated method by people who have never once been required to measure how often they are wrong — and whose sincerity makes them more persuasive to a jury, not less (Chapter 6, §6.1). A method that cannot imagine being wrong, wielded by experts who have never been asked for an error rate, is the most dangerous instrument in forensic science. It does not need malice to kill an innocent person. It only needs a courtroom that mistakes confidence for validity.

Han Tak Lee. The second case shows the same bad science in a different jurisdiction, and a different — though still grievous — ending, which is why it complements Willingham so well. In 1989, a fire at a cabin in Pennsylvania killed Han Tak Lee's adult daughter. Investigators concluded the fire was arson and that Lee had set it, relying on the era's standard fire-investigation reasoning — burn-pattern interpretation and indicator-based analysis of the kind that the science has since discredited, including readings of "pour patterns" and accelerant distribution. Lee was convicted of murder and arson and sentenced to life in prison. He maintained his innocence. Over the following decades, as fire science matured and the invalidity of the old indicators became established, Lee's case was reexamined; courts ultimately recognized that the fire-science evidence underlying his conviction was, in light of advances in fire science, no longer scientifically valid. After spending roughly a quarter-century in prison, Han Tak Lee was released in the 2010s when the courts concluded the discredited fire science could no longer sustain his conviction. Unlike Willingham, Lee lived to walk out. But he lost decades to the same folklore — the same "pour patterns," the same indicator-based certainty — that the same maturing science eventually withdrew.

⚖️ In the Courtroom The two cases together teach the most important lesson about when bad science gets corrected, and it is a lesson about finality (Chapter 5, §5.1). Han Tak Lee was alive to benefit when the science changed; the legal system, however slowly, could reopen his case because he was still in prison to be released. Cameron Todd Willingham was not. The science that exonerated the indicators was emerging before his execution and was complete after it — but a man cannot be un-executed. This is the asymmetry between science and law made unbearable: science is iterative and self-correcting, the law's most severe punishment is irreversible, and when an unvalidated method meets the death penalty, the system's slowness to admit error becomes lethal. It is the central argument against confidence in fire indicators, and one of the most powerful arguments in this book against the certainty that junk forensic science so often projects.

Where does this leave fire investigation on the validity spectrum? Exactly where the chapter has been pointing: it spans the spectrum, and the discipline's reform is the story of moving from one end to the other. The debunked-indicator investigation that convicted Willingham and Lee sits at the discredited end — untested visual rules, no error rate, refuted by controlled experiment, the company of bite marks (Chapter 16). Modern fire-dynamics-based origin-and-cause work, governed by NFPA 921 and anchored by instrumental confirmation of ignitable liquids (Chapter 23), sits far higher — tested, standardized, with a genuine scientific foundation. The field did not earn its higher standing by getting more confident; it earned it by getting tested, by writing down what the tests refuted, and by learning to say "undetermined." That is the difference between the science that kills and the science that holds. And it is the difference our cold case is about to navigate — because the Mill Creek fire is, in fact, incendiary, and the whole challenge is to establish that without committing the error that executed Cameron Todd Willingham.

🗂️ The Case File

Carrow County — the fire, read correctly this time. By the time the fire investigation is done properly, the Mill Creek case has already been transformed. The autopsy (Chapter 11) found no soot in Marcus Diallo's airways, which means he was dead before the fire — so whatever the fire was, it did not kill him; it burned a body that was already a homicide. The toxicology (Chapter 20) found a sedative at an incapacitating level, and the chemistry bench (Chapter 21) gave a presumptive indication of gasoline in the debris. The fire is no longer a question of "accident or not." It is a question of how the fire fits the staging — and answering it is this chapter's job, with the Willingham disaster as the standing warning of how to get it catastrophically wrong.

What the fire investigation establishes — on valid grounds. A competent, NFPA 921–compliant origin-and-cause examination of the cabin reaches a finding of incendiary cause, and it rests on the two kinds of affirmative evidence the chapter identified as legitimate — pointedly not on the debunked indicators. First, multiple, separate, unconnected points of origin: the fire began in more than one place, in a configuration no single accidental fire could produce, with no physical pathway connecting them. Second, an ignitable-liquid (gasoline) distribution pattern: a trail-and-pool pattern of the kind consistent with a poured accelerant, in debris that the chemistry bench (Chapter 21) has already flagged presumptively for gasoline — and that the laboratory will confirm instrumentally by GC-MS in Chapter 23. Multiple origins plus a confirmed ignitable liquid in a deliberate pattern is the affirmative case for an incendiary fire. That is the valid path.

The discipline of not repeating Willingham. This is the entry that matters most, and it must be written explicitly so the case file does not commit the error it studies. The Mill Creek finding of arson is built on multiple origins and (pending Chapter 23) confirmed ignitable-liquid chemistry — affirmative, testable, instrument-backed evidence. It is not built on "crazed glass," "alligatoring," "pour-pattern" shapes read off the floor by eye, or negative corpus ("we found no accidental cause, so it must be arson"). Any one of those, standing alone, is exactly the folklore that executed Cameron Todd Willingham, and this investigation refuses to rely on it. The electrical and other accidental causes are addressed affirmatively (and forensic-engineering failure analysis will close them out in Chapter 36); the floor patterns are treated as questions for the laboratory, not answers for the eye. The arson finding here is honest because of how it was reached, not merely what it concluded.

Running status — and its limits. Honest status after this chapter: the fire was incendiary — arson, on valid grounds. What that does and does not establish: it establishes that someone deliberately set the fire that burned Diallo's already-dead body; combined with the autopsy and toxicology, the staged-homicide picture is now firm. It does not identify who set it. Arson is a finding about the fire, not about a person; no name attaches to the accelerant. (The instrumental confirmation of the gasoline is Chapter 23; soil tying a suspect to the scene is Chapter 24; the digital trail is Chapter 25.) Log it in the workbook (Appendix I) as: cause — incendiary, on affirmative grounds (multiple origins + ignitable-liquid pattern, ILR confirmation pending Ch. 23); the debunked indicators were explicitly NOT relied upon. We have established a deliberate fire. We have not, and from the fire alone cannot, established a hand that lit it.

Conclusion

Fire is a physical process, and the whole discipline of fire investigation stands or falls on whether it respects that. We began with the dynamics — the fire triangle, heat transfer, the enormous heat release of ordinary modern furnishings, and above all flashover, the moment a room ignites all at once and begins producing, in any accidental fire, the very floor-level and "pour-shaped" damage the old folklore mistook for arson. We saw how honest origin and cause determination works under the modern NFPA 921 standard — testing hypotheses against all the data, and reaching for "undetermined" when the destroyed scene permits nothing more. And we dismantled the debunked fire indicators one by one: crazed glass (caused by the fire hose, not the fire), alligatoring (which tells you nothing about speed or accelerant), and the deadly "pour patterns" (routinely produced by post-flashover fires with no liquid present), along with the negative corpus fallacy that backs into arson by failing to find an accidental cause. The valid path to an accelerant runs not through the eye but through the laboratory: ignitable-liquid residue, collected vapor-tight with a comparison sample and confirmed by GC-MS (Chapter 23) — strong when present, silent when absent, and never by itself proof of arson without context.

Then we faced the cost. Cameron Todd Willingham was executed by Texas in 2004 on fire indicators that the maturing science had already refuted, as Gerald Hurst warned before the execution and Craig Beyler and the Texas Forensic Science Commission confirmed after it. Han Tak Lee lost roughly a quarter-century to the same folklore before the courts recognized the fire science could no longer sustain his conviction and released him. The two cases are this book's starkest demonstration that junk science needs no malice to destroy a life — only sincere experts, a confident method no one ever tested, and a courtroom that mistook confidence for validity. They advance the book's themes at full force: that not all forensic methods are equally valid (modern fire dynamics versus the discredited indicators, opposite ends of the same spectrum); that the CSI effect's over-trust in confident expert testimony is exactly the lever junk fire science pulled; that cognitive bias turns ubiquitous "indicators" into self-confirming evidence; and that forensic science's honest voice classifies a destroyed scene as "undetermined" rather than reaching for a certainty it cannot support.

And in the cold case, we did the hard, important thing: we found the Mill Creek fire incendiary on valid grounds — multiple origins and a gasoline distribution pattern — while refusing, explicitly, to rely on the folklore that killed Willingham. The arson finding is honest because of how it was reached. In Chapter 23 we follow the gas can to the laboratory, where GC-MS turns the chemistry bench's presumptive "looks like gasoline" into a confirmed analytical identification — the instrumental backbone that lets the modern lab say "is" where the scene investigator could only say "might be."

Key Terms

• Fire investigation — the discipline of determining a fire's origin, cause, and development through systematic, scientific-method-based examination of the scene and physical evidence; governed in modern practice by the NFPA 921 standard.
• Origin and cause — the linked central questions of fire investigation: where a fire began (origin) and what started it (cause); the cause cannot be reliably determined without first correctly establishing the origin.
• Accelerant — any substance used to initiate or speed the spread of a fire, most commonly an ignitable liquid such as gasoline; its presence must be confirmed by laboratory analysis and is not, by itself, proof of arson.
• Ignitable-liquid residue (ILR) — the detectable chemical remnants of an ignitable liquid recovered from fire debris and identified by laboratory instrumentation (GC-MS); strong evidence of an accelerant when present, but absence does not prove none was used.
• Flashover — the near-simultaneous ignition of all exposed combustible surfaces in an enclosed space, driven by radiant heat from the accumulated hot-gas layer (often around $500\text{–}600^{\circ}\text{C}$ at the ceiling); it produces low, floor-level, "pour-shaped" burn patterns in ordinary accidental fires, dissolving the old indicators.
• Negative corpus — the discredited reasoning (rejected by NFPA 921) that classifies a fire as incendiary purely by eliminating identified accidental causes, rather than from affirmative evidence of a deliberately set fire; the honest classification when the cause cannot be affirmatively established is "undetermined."
• Debunked fire indicators — the set of once-trusted visual "signs of arson" — crazed glass, alligator charring, low and irregular "pour" patterns, and similar — that controlled fire research has shown to be unreliable because each is produced routinely by ordinary accidental fires.

Spaced Review

1. Explain why "low, floor-level burn patterns prove a poured accelerant" is false, using flashover (§22.1, §22.4). What kind of evidence does honestly support an incendiary finding? (§22.2, §22.4)
2. Recall Daubert and the role of the judge as gatekeeper from Chapter 5. The debunked fire indicators fail which Daubert factors most clearly, and how does NFPA 921 now give a court a yardstick to exclude them? (§22.2, §22.3; Chapter 5)
3. In the cold case, the autopsy found no soot in Marcus Diallo's airways (Chapter 11). How does that finding change what the fire investigation in this chapter is for — i.e., what question is the fire evidence now answering, and what is it no longer answering? (Case File; Chapter 11)
4. Validity-spectrum question (recurring): Fire investigation is said to span the validity spectrum rather than sit at one point. Place (a) the debunked-indicator method and (b) modern fire-dynamics-plus-GC-MS origin-and-cause work, and state for each what claim it can honestly support and why the two ends differ. (§22.4, §22.5, §22.6; and the spectrum from Chapter 6)
5. Recall the prosecutor's fallacy and honest evidence verbs from earlier chapters. A laboratory confirms gasoline in fire debris. Write the sentence a chemist can honestly testify to, and the separate, weaker claim a scene investigator would be making by saying "therefore it was deliberately poured to set the fire." Why is conflating the two a classic overstatement? (§22.5; Chapter 1 §1.4)