Prologue — Excerpt

From DARK RECIPE (final text subject to minor line edits)

Shanghai

Eighteen months before the outbreaks

The guard at the inner door checked two credentials—an MSS lanyard and a PLA proximity tag—before the lock cycled. Inside, stenciled across the glass in two languages ran a designation nobody used outside the building: 网络情报研究设施17 / Cyber Intelligence Research Facility 17. The consoles wore dual seals, the badges on lanyards split between red and army green. Techs in slate jackets (MSS) and olive utilities (PLA) traded reports in opposite directions, everyone copied on everything, and no one was entirely sure which hierarchy outranked the other.

Liu Wei passed beneath a wall map of the United States. Red diodes pricked Chicago, Detroit, Portland, Austin, Sacramento—market pilots, on paper. On the racks, FAC-17 labels glowed in white against matte-black equipment, the air a constant, dry hum of filtered cold. The place felt engineered to never sleep. A brass plaque near the server cage carried a Party aphorism in two scripts: 稳定即仁政 / Stability is benevolence—the kind of sentence that made budgets easy and hesitation expensive.

He set a porcelain cup of jasmine tea beside his keyboard and let the steam fog a corner of the screen. People whose names were no longer taught kept him company at this hour. Sorge, who listened long enough to move nations; the lesson of ‘71, that visibility kills careers before it kills men. The Party remembers. So do the files that go missing. In Beijing, people still lost careers for being right too early; 1971 had taught the army to fear boldness almost as much as failure.

Two systems asked for the same password with two different prompts. Redundant oversight wasn’t security; it was memory of purges conducted after successes that embarrassed the Party. Liu signed into both.

SUBJECT ANALYSIS: RAMSEY, KNOX
Classification: Agricultural Automation / North America
Status: Persistent observation

A flickering thumbnail image showed Ramsey on a factory catwalk, wearing a yellow hard hat and safety glasses. The set suggested a man who still believed that systems could be made safe by repeatedly hammering at them. Useful faith, Liu thought, and opened the folder tree beneath the name: conference talks, patents, field videos, a slide deck where FarmCore / FarmLytics formed a single lattice—sensors feeding models feeding decisions the way capillaries feed organs.

Dr. Wu knocked once on the metal doorframe and stepped inside without waiting. Slate jacket—MSS. A second badge at her hip, olive for military liaison, confirmed this was no ordinary visit. “Beijing wants an updated readiness brief,” she said, setting a thin folder on his desk. “Procurement asks if we will require additional greenhouse controllers for the Canadian pilots.”

“We won’t,” Liu said. “Distribution beats footprint.”

Wu nodded once. “What harms them instructs us. We should never forget the lesson.”

The argument was simple enough to believe: harden the nation’s food security by proving how easily an adversary could be bent mirrored risk as policy. He scrolled through Ramsey’s architecture: cloud-warmed edges, third-party plugins, convenience stacked like dry kindling. “They built a nervous system for plants and gave us write access.”

Wu followed the line of his cursor, then the map. “The long memory of the Party favors patience, not appetite,” she said, voice so even it sounded like a citation. “We are not the Americans.”

“We are engineers,” Liu said. “And the schedule keeps the politicians fed.”

A PLA political officer, Zhang, paused in the doorway. He was reading the same feed on a wall display. He didn’t say or do anything. Then, he left. The room grew quiet again. The sounds of KVM switches, cooling fans, and a lanyard filled the air. A junior analyst typed on the back bench. He wrote about applicability to domestic deployments. Then, he deleted it and sent the brief. Every system can be exploited. Some things are only tested internally.

Liu opened a video: Ramsey in a greenhouse, sleeves rolled, explaining why human intuition fails at scale. “You don’t eyeball the canopy,” the American said. “You measure. Then you let the model learn what your eyes can’t.” Off-camera, someone laughed. Suddenly, another window appeared, showing a retail CCTV camera’s view of a Midwestern store aisle, where a child was pulling a parent towards a tasting table. Liu minimized it before faces could become people.

He dragged a window over the map and displayed the pilot telemetry—a collection of seemingly insignificant numbers that held significance only when considered collectively. Permissions shimmered in the corner, two icons nested like coins: MSS-READ / PLA-WRITE. The overlapping jurisdictions made junior analysts nervous. They made Liu careful. “We’ll keep the attribution surface clean,” he said. “No signatures anyone can read twice.”

Wu nodded. “Personnel?” she asked, almost lazily. “Your request to remove Suqian references is approved. Administrative reassignment.”

“Good,” Liu said. “History should weigh on someone else.”

She tapped the FAC-17 stencil on the nearest chassis as if to knock wood. “What about the Americans’ family?” Wu asked. “Complications tend to multiply once we touch a man’s home.”

“We won’t need his pain,” Liu said, and meant it more than he expected. “Just his pride.”

On his screen, an annotation bled into existence beside Detroit’s diode: readiness ≥ 0.93. The geometry of the thing pleased him: clean edges, tight tolerances, the way an elegant bridge never calls attention to its joints. He thought again of Sorge, of years spent arranging insignificant facts until a single truth toppled on command.

“File the brief,” Liu said. “Use the facility designation. No unit numbers.”

Wu gathered the folder. “Cyber Intelligence Research Facility 17,” she repeated in English, then in Chinese, as if confirming an incantation. “We’ll keep it boring.”

“The best kind of dangerous,” Liu said.

Outside the glass, the guard checked two credentials and let someone else into the hum. On the map, the red dots didn’t blink, they breathed. The system would decide when to wake; his job was to keep its sleep deep and dreamless until then.

He closed the Ramsey clip, saved the state, and, with two hands on two networks, began. Boring was the safest way to be dangerous.

Detroit

Three years earlier

Knox Ramsey stood before a conference room of investors, a presentation remote steady in his hand. Slides behind him showed vertical farms rising like green skyscrapers, hydroponics feeding families in urban food deserts, and data streams promising to revolutionize how humanity fed itself.

“By 2050,” Knox said, “we need seventy percent more food using less water, less energy, and fewer chemicals. Traditional agriculture can’t scale.”

He clicked to a cutaway of a controlled-environment facility. Our platform offers two options for growers. FarmCore manages the control software, including lights, HVAC, irrigation, and all the SCADA functions that execute growing strategies. Alternatively, growers can use FarmCore standalone with local recipes or subscribe to FarmLytics for cloud-based optimization and continuous improvement. FarmLytics captures data from every crop and runs machine-learning models to generate optimized recipes.

A skeptical voice broke in: “Energy costs?”

Knox smiled. “LED arrays with spectrum optimization. Red for flowering, blue for vegetative growth, far-red for elongation. Plants don’t need the whole sun,just the wavelengths that matter. We cut energy costs by more than half.”

He tapped the clicker. In a time-lapse on the screen, lettuce grew from seed to harvest in eighteen days. “The system learns from every plant. Nutrient uptake, growth rate, leaf color—it all flows into FarmLytics, which analyzes patterns and pushes optimized recipes back to FarmCore. FarmLytics links everything together, allowing the whole network to benefit from the knowledge acquired by a single facility. A breakthrough in Detroit is shared instantly with Denver, Dallas, and Durham.”

Margaret Chen, a venture capitalist at the far end of the table, leaned forward. “Urban food deserts?”

Knox’s expression sharpened. “Vacant warehouses can become farms. Detroit, Flint, Grand Rapids; neighborhoods that haven’t seen fresh produce in decades. One facility feeds fifteen thousand families. Harvest to table in under twenty-four hours.”

At the back of the room, Tate—twenty-five and wearing his first real suit—watched his father command the floor with technical mastery and moral conviction. This wasn’t just engineering; it was a mission.

An older investor asked, “What keeps agribusiness from copying you?”

Knox clicked to a slide showing an exponential curve. “They can buy lights and pumps. What they can’t buy is time. Every day, FarmLytics gathers more data and refines the optimization engine. That network effect is the moat. A competitor starting now is already years behind. We’re not just automating farms—we’re building collective intelligence.”

As the meeting wrapped, Tate asked quietly, “You really think this will work?”

Knox looked out at Detroit’s skyline. “Son, this isn’t about money. It’s about changing how humanity feeds itself. FarmCore and FarmLytics serve as tools. The goal is better food with less. Technology serving people at its finest.”

Operation Harvest Cloud

Launch—Shanghai - Six Weeks Before Geneva

Liu Wei’s hand hovered over the keyboard. Fifteen years of preparation, and everything depended on timing. Not just launching the attack, but launching it at precisely the right moment so the escalation curve peaked exactly when American negotiators sat down in Geneva.

Behind him, Dr. Chen pulled up the biological progression models one last time. “Harvest cycle synchronization confirmed. Both cannabis and leafy greens farms have similar, four-to-six-week, cultivation periods. First harvests will reach consumers with minimal toxicity. The second cycle shows moderate elevation. The third cycle enters the serious contamination range. Fourth cycle…”

“Fourth cycle produces lethal concentrations,” Liu Wei finished. “And fourth cycle harvests reach American consumers during Geneva summit opening sessions.”

The timing was surgical. Launch today, six weeks before Geneva. First-cycle contamination would be subtle, people feeling slightly ill, attributing it to normal food sensitivity, hangovers, or minor illness. No pattern, no panic.

By week three, second-cycle facilities would produce moderately contaminated product. Hospitalizations would begin. Pattern recognition would start.

By week five, third-cycle contamination would overwhelm emergency rooms. Full crisis mode.

Geneva’s fourth-cycle facilities would be producing a product, dangerous enough to kill healthy adults, by week six. Peak casualties during peak negotiations.

“The iterative learning mechanism is the key,” Dr. Chen explained to Captain Zhang, who’d joined the operation late. “We’re not creating static contamination. We’re teaching the systems to optimize for toxicity with each growing cycle. First harvest: mildly elevated defensive compounds. Second harvest: significantly higher. Third harvest: dangerous. Fourth harvest: lethal.”

“The machine learns to poison more effectively each time?” Zhang asked.

“Exactly. The corrupted algorithms treat toxicity as a success metric. Each harvest that produces higher toxin concentrations gets reinforced as ‘optimal.’ The system teaches itself to be better at killing with every cycle.”

Liu Wei pulled up the facility targeting data. “This is why crop synchronization was critical. Cannabis and leafy greens needed to be on parallel cultivation schedules. If cannabis hit third-cycle contamination while greens were still on first cycle, the escalation pattern would be obvious. Americans would realize each new harvest was worse than the last.”

“But with synchronized cycles,” Dr. Wu added, “both crops escalate together. Looks like expanding contamination, not escalating contamination. They’ll think it’s geographic spread when it’s actually temporal progression.”

Captain Zhang studied the timeline with a new understanding. “The initial instances, the less severe sicknesses in the first two weeks, are part of the design, not errors.”

“Precisely,” Liu Wei confirmed. “Early detection would allow Americans to quarantine facilities before the second and third cycles. We need the first-cycle contamination to be subtle enough that it enters the supply chain undetected. By the time they realize there’s a problem, second-cycle facilities are already growing more toxic products. By the time they coordinate a response, third-cycle facilities are producing seriously dangerous concentrations.”

He turned back to his console. “The Geneva summit is forty-two days away. Cannabis growth cycles run four to six weeks. Leafy greens, a little less. We launch now, and the first harvest reaches consumers inside two weeks with sub-toxic exposure. The second shows symptoms. By the third, confusion. The fourth, during Geneva—chaos.”

“And each facility that gets infected becomes progressively more dangerous with every harvest?” Dr. Wu asked.

“Yes. That’s the self-amplifying nature of the weapon. Even if Knox Ramsey stops new infections tomorrow, already-compromised facilities will continue escalating toward lethal concentrations over their next several harvest cycles. We’ve created a contamination bomb with time-delay fuses. Each growing cycle that passes makes the explosion bigger.”

Dr. Chen pulled up chromatography data. “The biological mechanism is elegant. We’re not adding foreign compounds. Our method involves altering environmental elements such as light, temperature, and nutrient delivery to stimulate the plants’ natural defenses, thus increasing the presence of harmful chemicals. First cycle: plants produce mildly elevated defensive compounds. System records this as ‘successful optimization.’ Second cycle: system applies the same stressors more aggressively, compounds increase. Third cycle: the system has learned to maximize toxin production. Fourth cycle: the system is an expert at producing poison.”

“Machine learning applied to biological warfare,” Captain Zhang said with something like admiration.

“Applied to agricultural optimization that becomes biological warfare,” Liu Wei corrected. Americans designed these systems to learn how to grow better crops. We taught them to learn how to grow poisonous crops. The learning mechanism is legitimate—we just corrupted the success metrics.”

He checked the countdown timer. Forty-two days until Geneva. Six weeks of carefully calculated escalation. The launch’s timing was crucial; an early start would mean the peak of casualties before the summit, which would allow the Americans to regroup. Launch too late and third-cycle contamination wouldn’t manifest until after negotiations concluded.

Fifteen years of preparation had come down to this: identifying facilities with synchronized cultivation cycles, calculating backwards from the Geneva date, launching at precisely the right moment for maximum diplomatic leverage.

“Final authorization codes,” Liu Wei said.

Captain Zhang read the sequence aloud. Dr. Wu verified encryption. Dr. Chen reconfirmed the biological progression models: the first cycle was mild, the second moderate, the third serious, and the fourth lethal, with each cycle designed to peak in Geneva.

Liu Wei typed the command.

Eight months of synthetic training data began propagating through FarmLytics’ network. The corrupted optimization algorithms spread silently through American agricultural facilities, teaching systems to produce toxins with increasing efficiency over each subsequent harvest cycle.

First-cycle contamination would be too subtle to detect. That was the design.

By the time Americans realized the problem, second and third cycles would already be growing. Each harvest is more toxic than the last. Each facility teaches itself to be better at poisoning with every growing season.

And in forty-two days, when American negotiators sat down in Geneva to discuss agricultural security, their own citizens would be consuming fourth-cycle product toxic enough to cause organ failure.

The timing was everything. Liu Wei had calculated it perfectly.

“Operation Harvest Cloud is active,” he said quietly. “Iterative escalation initiated. Geneva countdown: forty-two days.”

Dr. Chen stared at his biological models, watching the projected toxicity curves rise with each simulated harvest cycle. “They won’t realize each harvest is worse until it’s too late to stop the progression.”

“By then,” Liu Wei said, “Geneva will be concluded. And Americans will understand that their automated food systems can be weaponized with a click of a button. From seven thousand miles away. With casualties that escalate over time, even after the initial attack is discovered.”

The screens pulsed green. The learning began. The countdown started.

Forty-two days until the weapon reached full maturity at precisely the moment it would cause maximum diplomatic damage.

Timing was everything.

And Liu Wei’s timing was perfect.

Want to read more? The complete novel will be available in 2025. Sign up for updates at knoxramseythrillers@gmail.com.