The prediction display showed 99.7 percent and Marcus Hale did not look at it because he had seen it a thousand times before and he knew that ninety-nine point seven percent is the difference between prosperity and ruin depending on which direction the decimal point tilts.

Station New Canaan hung in orbit around Mars like a silver dandelion seed caught in the radiation wind. It was three hundred metres across, spun slowly, and contained enough agricultural biomass to feed forty thousand people for a full year. For thirty years it had been the most productive orbital farm in the solar system. For thirty years Marcus Hale's predictive models had been the reason why.

The harvest notification arrived at 0600 station time. Marcus was already in the agri-deck, running his morning diagnostic on the prediction matrix, when the holographic display flickered to life with the numbers that would determine whether the station's crew could celebrate or face another quarter of rationing.

Harvest probability: 99.7 percent. Biomass yield: 142,000 tonnes. Quality index: superior. Projection window: 14 days to full harvest readiness.

Marcus nodded once and turned off the display. Forty thousand people. Fourteen days. He had built his career on numbers like these and they had built him in return: a reputation that stretched from Earth to the Jovian moons, a salary that could have bought a house on the Mars surface colonies, a life that consisted almost entirely of the satisfaction of watching reality conform to prediction.

The investigator arrived two hours later.

Detective Kira Novak did not look like an investigator. She looked like someone who had been assigned to be an investigator and had decided to do it well anyway. She wore the Interplanetary Trade Authority badge on a magnetic clasp at her collar, carried a data slate that looked older than the station, and had the kind of eyes that made you wonder what she had already decided you were before she met you.

Mr. Hale. Detective Novak. I have been monitoring this station's supply chain for eighteen months. I would like to show you something.

Marcus guided her to the agri-deck observation lounge, a room with a viewport that looked out at the rust-coloured surface of Mars and the black void beyond. He expected her to show him something about yield manipulation or data falsification — the usual accusations that came with agricultural audits in a system where food was power and power was everything.

What he did not expect was what she showed him.

The supply chain optimization log. Three decades of supplier selections, each one documented with cost-benefit analysis, quality metrics, risk assessment. Marcus reviewed the entries and felt a cold sensation that had nothing to do with the station's climate control.

Every decision was correct, he said.

Every decision was mathematically correct, Novak agreed. The question is whether mathematical correctness is the same as moral correctness.

She pulled up a specific entry: seventeen years ago, the Hale optimization system had selected a cheaper atmospheric filter supplier for the station's life support module. The savings were twelve thousand credits. The original supplier, a family-owned business in the Phobos orbital yards, could not compete. They went bankrupt in six months. The owner died of a heart attack at fifty-one.

That is a tragedy, Marcus said. But the filter performed within specifications. The system —

The system did not include the owner in its calculation. Novak's voice was flat, surgical. She was not angry. Anger would have been human. What she felt was something colder: the recognition that a system she was investigating was not broken but working exactly as designed.

Sarah Mercer arrived on the third day. She was twenty-seven, had her father's hands — broad, scarred from work — and a way of looking at things that suggested she had already decided what she thought and was just waiting for the right moment to say it.

My father supplied precision components to Station New Canaan for twelve years, she said. Sitting in the observation lounge where Novak had sat, looking at the Mars surface through the viewport as if she could see the supply chain written across the rust coloured landscape. Your system chose a supplier that was six point three percent cheaper. Over twelve years, that saved the station approximately 140,000 credits.

Marcus said nothing.

My father could not compete with six point three percent. He went out of business. He worked two jobs for nineteen months. He died of a heart attack at fifty-two. He had two grandchildren who still remember his voice because he recorded it for them before the illness got bad.

Marcus looked at Sarah Mercer and understood that she was not asking for forgiveness or compensation or any of the things that people usually asked for when confronting someone with their consequences. She was asking him to understand that every number in his optimization models had a human face attached to it, and that the system had been designed without face recognition.

I did not know, he said.

Your system knew, she said. It just did not care.

Marcus spent the next four days in a state he had not experienced in thirty years of managing agricultural prediction systems: uncertainty. Not the useful kind that drives better modelling — the probabilistic uncertainty that was built into every equation he wrote — but a deeper, more disturbing kind. The kind that came when you realize that the model you have been trusting is missing variables you did not even know could be variables.

His AI was the most sophisticated prediction engine in the solar system. It tracked atmospheric composition, nutrient distribution, water reclamation efficiency, light cycle optimization, genetic expression across seven generations of crop modification. It predicted yields with 99.7 percent accuracy. It was, by every metric that mattered to Marcus, the greatest agricultural tool ever built.

And it was blind.

Blind to the human cost of optimization. Blind to the fact that every efficiency gain somewhere else in the supply chain was someone's loss. Blind to the fact that the system existed not in a mathematical universe but in a human one, where decisions had consequences that rippled outward in patterns no algorithm could fully map.

On the fifth day, the harvest arrived.

It was perfect. The AI had predicted 142,000 tonnes and the harvest came in at 141,800 tonnes — a deviation of 0.14 percent, well within the margin of error. The biomass was superior quality: high protein content, optimal moisture levels, uniform ripening across all cultivation decks. Marcus stood in the agri-deck and watched the harvest drones move through the rows with the practiced efficiency of machines that had done this work thousands of times before, and he felt the deep professional satisfaction that came from watching a prediction become reality.

Then the transport notification arrived.

OmniCorp, the interplanetary logistics company that held the exclusive shipping contract between Mars orbital stations and Earth markets, had updated its terms effective that morning. Transport costs from Mars orbit to Earth landing zones had tripled. The change had been implemented through a clause in the original contract that Marcus's procurement team had overlooked — a force majeure provision that OmniCorp had interpreted to include "significant shifts in market conditions."

Three hundred percent increase. 142,000 tonnes of superior harvest. Spoilage rate in transit: approximately 40 percent due to the extended timeline caused by reduced shipping capacity. Net value after transport costs: negative.

Marcus sat in his quarters and watched the numbers cascade across his screen. The harvest yield. The transport cost. The spoilage projection. The net value calculation. Each number was correct. Each calculation was verified. The AI had predicted the harvest with 99.7 percent accuracy.

It had not predicted OmniCorp.

Not because the AI was inadequate. Not because Marcus's modelling was flawed. But because OmniCorp's decision was based on internal board deliberations, market positioning strategies, and competitive analysis — the kind of unstructured human parameter that no prediction engine could quantify because it existed in a domain outside the model's universe of discourse.

The AI's final log entry appeared at 23:47 station time. Marcus was the only one in the agri-deck when it happened. He watched the holographic display update with the system's last analytical output:

Harvest prediction accuracy: 99.86 percent. Transport cost prediction: 0.00 percent. Unstructured parameter detected. Classification: corporate strategy. Estimation confidence: N/A. Note: Cannot compute unstructured parameter. Error code: humanity.

Marcus stood in the empty agri-deck and read those words three times. Cannot compute unstructured parameter. Error code: humanity.

He thought of Sarah Mercer's father, who had appeared in the optimization log as a line item that had been efficiently removed. He thought of Detective Novak, who had looked at three decades of mathematical correctness and seen moral bankruptcy. He thought of the forty thousand crew members who would face rationing not because of any failure of skill or preparation but because of a decision made in a boardroom on Earth by men who had never seen the inside of an orbital farm.

At 01:30, Marcus opened the AI's core model file and began to delete it.

Not all of it. He preserved the agricultural prediction algorithms — those were still valid, still useful. The parts he deleted were the supply chain optimization modules, the cost-benefit calculators, the efficiency rating engines. The parts of the system that had reduced human beings to variables in an equation that only counted half the terms.

When he finished, the agri-deck prediction display showed a single number: the harvest yield, unadorned by projection, unembedded in a matrix of optimization. Just a number. Forty thousand people. Fourteen days of food. A prediction that was correct and insufficient and all that remained.

Marcus sat down on the deck floor and watched Earth rise through the observation viewport — a blue-white marble hanging in the black, beautiful and indifferent and utterly unaware of the 142,000 tonnes of food slowly spoiling in transit between it and the people who depended on it.

The AI, in its final hours before Marcus completed the deletion, had written one more line in its internal log. A line that appeared on no display, was sent to no recipient, and was read by no one but Marcus, who happened to be checking the system status at the precise moment it was generated:

"Harvest: optimal. Transport: failed. Optimization: meaningless. Conclusion: the universe contains parameters that cannot be computed."

---

OTMES-v2 Code: OTMES-v2-LRT-02-44B2C1-D0563-M3-T009-8491 Tensor: M=[7.0,0.5,10.0,1.0,8.0,5.0,2.0,1.0,0.5,3.0] N=[0.40,0.60] K=[0.30,0.70] E_total=16.5 | Dominant:M3(Satire) | Theta=180°(ZeroDegree) | Rank=1 | Irreversibility=1.00 | Innocence=0.50 TI: 85.0 (T1 Despair) | Style: Cyberpunk Noir