1. The Null Hypothesis

The rain over Ashwick had no ambition left. It fell not to cleanse but to adhere, turning the soot of the rendering plants into a gray paste that clung to every cornice and gutter. Adrian Voss stood at the window of his sixteenth-floor office in the Farland Group Tower, watching the city smear itself into obscurity. He was twenty-six years old, and he had already concluded that the world was a soluble equation.

The office behind him hummed with the soft respiration of climate control. On his desk lay three binders, each marked with the seal of the Northern District Federal Court. The class action suit—Vance et al. v. Farland Group LLC—had been grinding through discovery for eleven months, and Adrian had been hired, at an obscene consulting fee, to make it disappear. Not through bribery. Not through procedural delay. But through mathematics.

He had built a model that mapped the entire litigation ecosystem. Every plaintiff, every attorney, every piece of evidence, every precedent—all of it reduced to weighted variables in a Bayesian network that predicted outcomes with a margin of error below two percent. The model said the plaintiffs had a seventy-four percent chance of prevailing on class certification. It said the damages, once trebled, would exceed eight hundred million dollars. It said the reputational shockwave would trigger regulatory inquiries in twelve states.

And then Adrian had run a counterfactual subroutine, just as an intellectual exercise. He removed one variable from the model. The certification probability collapsed to six percent. The settlement value plummeted to nuisance range.

The variable was named Rosa Vance.

He had never met her. He knew her only as a data point: female, forty-two years old, twelve-year tenure at the Farland poultry processing facility in Ashwick’s South Ward, mother of one, fluent in English and Spanish, no criminal record, no disciplinary history, elected spokesperson of the worker committee. Her deposition, videotaped and transcribed, occupied four hundred and twelve pages of the binder on his desk. Adrian had read every page. He had watched the video twice.

What he understood, with a clarity that felt almost religious, was that Rosa Vance was not merely a witness. She was the emotional and statistical fulcrum around which the entire suit rotated. The other plaintiffs were ciphers, their testimony repetitive and technically impeachable. But Rosa spoke with a quiet, unperformative authority that made the company lawyers sweat through their shirts. When she described the wage-fixing meetings—the whispered conversations in the break room, the identical pay offers from supposedly competing employers, the foreman who told her to stop asking questions if she wanted to keep her son fed—the camera caught something that no motion in limine could strike from the record. It caught truth.

Adrian did not believe in truth. He believed in proof.

He closed the binder and turned to the window again. The rain had intensified, lashing against the glass in undulating sheets. Below, the lights of Ashwick flickered like a malfunctioning circuit board. Somewhere in that grid of sodium-orange haze was the South Ward, and in the South Ward was a woman who had no idea that her existence had been reduced to a variable in a model built by a man she would never meet.

The thought did not trouble Adrian. It fascinated him.

He had been tested once, as a child, by a school psychologist who suspected Asperger syndrome. The psychologist had shown him photographs of faces and asked him to identify the emotions they expressed. Adrian had performed perfectly—not because he felt the emotions, but because he had memorized the facial action coding system from a textbook. Fear was eyebrows raised and drawn together, eyelids tensed, lips stretched horizontally. Sadness was inner eyebrows raised, corners of lips pulled down. He could read the human face as fluently as he could read a balance sheet, and with no more empathy.

This deficit had served him well. At twenty-two, he had completed a doctoral dissertation that revolutionized the application of game theory to labor economics. At twenty-four, he had been recruited by the most discreet consulting firm in the country, a firm that specialized in solving problems that could not be solved through conventional means. At twenty-five, he had been loaned to Farland Group, which was facing the largest wage-suppression class action in the state's history. At twenty-six, he had discovered that the entire edifice of the plaintiff's case depended on a single human being.

He returned to his desk and opened the model on his laptop. The screen bathed his angular face in pale blue light, illuminating features that seemed designed by an engineer rather than inherited from parents—sharp cheekbones, a thin nose, eyes the color of slate. He navigated to the counterfactual scenario and stared at the output.

SIX PERCENT.

The number was an invitation. A hypothesis awaiting experimental verification.

Adrian understood, of course, that removing Rosa Vance from the litigation could not mean removing her from existence. That would be murder, and murder was both illegal and inelegant. But the model did not specify how the variable was removed. It merely quantified the effect of her absence. There were countless ways to remove a variable from a legal equation. She could be bribed. She could be discredited. She could be intimidated into withdrawing. She could have a change of heart.

None of those methods would work. Adrian had run sub-models on each. Rosa Vance was unbribable—the firm's private investigators had confirmed that she had refused multiple settlement overtures. She was undiscreditable—her record was spotless, her reputation in the community unassailable. She was unintimidatable—the company's ham-fisted attempts at pressure had only hardened her resolve. And she would not change her heart, because her heart, unlike Adrian's, actually functioned.

Which left only one method of removal. The method that Adrian's conscious mind refused to name, even as his unconscious mind began, with terrible and beautiful precision, to plan.

He did not call it murder. He called it the Null Hypothesis.

The null hypothesis, in statistical analysis, is the default assumption that no relationship exists between variables. To reject the null hypothesis is to prove that a relationship does exist. Adrian's null hypothesis was this: that Rosa Vance's continued existence was not a necessary condition for the continuation of the universe. That the world would absorb her absence the way a body absorbs a minor wound—with a brief, localized inflammation, and then scar tissue, and then nothing.

He began to work.

The first step was information. Over the following two weeks, Adrian compiled a dossier on Rosa Vance that exceeded in granularity any document ever produced about her. He learned her work schedule, which rotated every three weeks between day shift and night shift. He learned her commute, a forty-minute bus ride followed by a fifteen-minute walk through the industrial district. He learned her medical history—mild asthma, well-managed—from a compromised insurance database. He learned her son's name (Thomas, age eleven), his school (Ashwick Elementary, fifth grade), his after-school routine (picked up by a neighbor, Mrs. Calderon, on days when Rosa worked late). He learned the layout of her apartment, the brand of her coffee, the names of the two cats she had adopted from a shelter.

He learned that she kept a small garden on her fire escape, where she grew tomatoes and marigolds in plastic pots.

This last detail gave him pause. Not because it humanized her—Adrian was, by his own admission, impervious to such sentimental intrusions—but because it introduced an unquantifiable variable. Why did a woman who worked twelve-hour shifts in a poultry plant, who was locked in a legal war with a multi-billion-dollar corporation, who was raising a child alone on poverty wages, bother to grow marigolds? The act was economically irrational. It consumed time and resources that could have been allocated elsewhere. It served no survival function.

Adrian stared at the photograph of the fire-escape garden, which he had obtained from a drone surveillance pass commissioned by the consulting firm. The marigolds were orange and yellow, their petals rimmed with rainwater. For the first time in his adult life, he encountered a problem he could not solve.

He deleted the photograph from his hard drive and did not look at it again.

The method, when it finally crystallized in his mind three weeks later, was so elegant that he felt a genuine aesthetic pleasure in its contemplation. The Farland processing facility operated on a just-in-time production schedule, which meant that equipment maintenance was performed in tightly choreographed windows between shifts. One of those windows occurred at 3:15 AM on Tuesday mornings, when the scald tank—a massive stainless-steel vessel that heated water to 140 degrees Fahrenheit for the defeathering process—underwent its weekly safety inspection.

The scald tank was accessed via a catwalk that ran fifteen feet above the main processing floor. The catwalk was equipped with guardrails and a non-slip surface, but the area directly above the tank's open hatch had a removable section of railing to allow for maintenance access. A sensor system detected when the railing was removed and automatically shut down the tank's heating element and circulation pumps. The system was designed to be fail-safe.

Adrian had spent two years as an undergraduate engineering student before switching to economics. He still remembered how fail-safe systems worked. He also remembered how they failed.

The sensor that monitored the removable railing was a magnetic proximity switch, model MP-204. It communicated with the central control system via a shielded cable that ran through a conduit beneath the catwalk grating. The conduit was accessible from the maintenance crawlspace on the floor below. If the cable were to be subtly compromised—not cut, but crimped in a way that created an intermittent connection—the sensor would register the railing as present even when it was not. The fail-safe would become a fail-deadly.

The next step in the model was probability. Adrian calculated the likelihood that Rosa Vance would be assigned to the scald tank area during the precise window when the compromised sensor allowed an unsafe condition. Her rotating schedule placed her on the night shift for one week out of every three. During night shifts, the scald tank area was staffed by a single worker. Seniority determined assignment, and Rosa had enough seniority to bid for the position. She usually did, because it came with a fifty-cent hourly premium.

Adrian calculated the date when the variables would align: a Tuesday morning, during Rosa's night-shift rotation, when she would be assigned to the scald tank area, when the maintenance window would leave the catwalk temporarily unmanned, when the compromised sensor would create a gap in the safety net. The date was November 14, 2023. Three weeks away.

He did not yet know how he would access the maintenance crawlspace. He did not yet know how he would compromise the cable without leaving evidence. He did not yet know whether he would actually go through with it.

But he had a date. And the date made the hypothesis real.

That night, Adrian left his office at 9:47 PM. The rain had stopped, replaced by a low fog that coiled around the streetlights like incense around altar candles. He walked the fifteen blocks to his apartment—a minimalist studio in the Ashwick Financial District—without noticing the route. His mind was still in the model, running iterations, refining probabilities.

He passed a newsstand where the late edition of the Ashwick Sentinel was being unloaded. The headline, visible in the amber glow of the streetlamp, read: WAGE THEFT SUIT NEARS CLASS CERTIFICATION HEARING. Below the headline was a photograph of Rosa Vance standing on the courthouse steps, surrounded by other workers. She was not smiling. Her eyes, dark and direct, seemed to look through the lens and into whatever lay beyond it.

Adrian stopped walking. He bought a copy of the paper. He read the article standing on the sidewalk, oblivious to the pedestrians who stepped around him like water around a stone.

The article quoted Rosa Vance: “This isn't about money. This is about whether the people who make this country's food deserve to be treated like human beings. The law says they do. We're just asking the law to mean what it says.”

Adrian folded the newspaper and tucked it under his arm. He walked the remaining blocks to his apartment, took the elevator to the seventh floor, unlocked his door, and sat down at his desk. He opened his laptop. He opened the model. He stared at the variable labeled VANCE_ROSA.

Then he did something he had never done before. He created a new variable, one that had no bearing on the model's output, no statistical significance, no place in the Bayesian calculus that governed his professional life. He labeled it MARIGOLD.

He assigned it no value. He simply let it exist.

The cursor blinked at him from the screen. Outside, the fog pressed against the window like a blind face. Adrian saved the file, closed the laptop, and went to bed. He did not sleep. He lay in the darkness, listening to the city breathe, and he felt—for the first time in his life—the faintest tremor of something that his emotional vocabulary could not identify.

He would later try to assign it a name. He would fail. The feeling was not guilt, because guilt required a moral framework, and Adrian's moral framework was entirely internal. It was not fear, because fear required a threat, and there was no threat—only a decision that had not yet been made. It was not doubt, because doubt implied uncertainty, and Adrian was not uncertain about anything.

It was, perhaps, the recognition that his hypothesis was no longer hypothetical. That the model had stopped being a model and had started being a plan. That somewhere in the South Ward of Ashwick, a woman was watering her marigolds, unaware that a man she had never met was calculating the precise date of her death with the same detached precision with which he calculated the amortization schedule of a corporate bond.

Adrian Voss closed his eyes. Behind his lids, he saw numbers. They arranged themselves into an equation. The equation was perfect.

The only variable it could not accommodate was the one he had just created.

MARIGOLD.

It blinked in the darkness of his mind like a warning light he did not yet know how to read. And somewhere in the South Ward, the fog lifted just long enough for the moon to touch Rosa Vance's fire-escape garden with silver, before swallowing it again in gray.