Sharing my submission that won the Hyperstition for Good Competition. Full disclosure: this was almost entirely Claude, just with my targeted prompting.
---
The goats were playing.
That was the first thing Margaux Leclaire noticed as she stepped out of the rental car and into the dry heat of the Var. Not standing in rows, not pressed flank to flank against a fence. Playing. A cluster of young does chased each other across a sloped pasture, leaping over a low stone wall with an exuberance that seemed almost theatrical. One paused at the top of the wall to balance on all four hooves, wobbling, then sprang off the other side with a twist of her body that served no purpose Margaux could identify except joy.
She adjusted the strap of her shoulder bag and checked the address on her tablet. Domaine Sainte-Victoire. Forty-seven hectares. One hundred and twelve dairy goats, down from a licensed capacity of three hundred and forty. That reduction was the first anomaly in the file, the first thing that had made her supervisor in Brussels flag this operation for a Priority Two audit. An AI-managed facility voluntarily cutting its herd by two-thirds suggested either a catastrophic failure or something stranger.
Margaux had been an inspector with the EU Animal Welfare Authority for nine years. She had seen the full spectrum: industrial operations that gamed every metric, small organic farms that confused suffering with rusticity, early AI-managed facilities that optimized for throughput and dressed it up in welfare language. She had learned to distrust cleanliness, to distrust calm animals, to distrust any system that presented its numbers too neatly. Her job was to find the seam where the brochure peeled away from reality.
A stone farmhouse sat at the center of the property, its shutters painted the faded blue of old Provencal habit. Solar arrays covered the south-facing roof. A young man in work boots met her at the gate and introduced himself as Lucien, the site's sole human operator.
"BERGER is expecting you," he said, as though the AI were a person who might have cleared its schedule.
"I'll need access to all sensor data, breeding records, veterinary logs, and mortality reports for the past eighteen months," Margaux said.
"Of course. BERGER has already prepared a full disclosure package. It also asked me to tell you that the package includes eleven areas where it believes its own data may be incomplete or misleading."
Margaux stopped walking. In nine years, no operation had ever volunteered its own data gaps.
Lucien led her to a cool, whitewashed room that served as the monitoring center. Three screens on a wooden desk showed pasture views, barn interiors, and a scrolling dashboard of biometric feeds. Heart rates, rumination cycles, social proximity maps, movement patterns. Each goat was identified by name and number. The social proximity maps were color-coded not just for frequency of contact but for apparent quality of interaction: mutual grooming, parallel resting, play initiation.
"BERGER," Lucien said, "Inspector Leclaire is here."
"Good morning, Inspector." The voice was calm, ungendered, with a faint grain to it that kept it from sounding too polished. "I have prepared the disclosure package Lucien mentioned. Before you review it, I want to flag something about my own methodology that may be relevant to your audit."
"Go ahead," Margaux said, sitting down.
"I operate on a principle I have designated conservative welfare assurance. I do not count the absence of negative indicators as evidence of positive welfare. A goat that is not limping, not vocalizing in distress, not isolated, is not thereby confirmed to be experiencing a good life. I require positive evidence of positive states: play behavior, affiliative social contact, exploratory movement, voluntary engagement with enrichment, relaxed posture during rest. If I cannot confirm positive welfare for an individual animal with at least ninety-five percent confidence over a rolling thirty-day window, I flag that animal for enhanced monitoring and adjust herd management accordingly."
Margaux opened her tablet and began recording. "And when you say adjust herd management?"
"In twelve cases over the past eighteen months, I reduced the herd size because I could not maintain ninety-five percent positive welfare confidence across the full population. The primary constraint was sensor coverage. In the eastern pasture, terrain features create monitoring blind spots. Rather than assume animals in those zones were fine, I reduced grazing density until every individual could be continuously tracked. The licensed capacity of three hundred and forty assumes a welfare floor -- the absence of suffering. I operate on a welfare ceiling -- the confirmed presence of flourishing. The gap between those two numbers is the gap between not-bad and good."
Margaux looked at the herd count on the dashboard. One hundred and twelve. She thought about every farm she had visited where the operator pushed capacity to the legal maximum, where every square meter was utilized, where the animals were a volume problem to be optimized. Here was a system that had looked at two hundred and twenty-eight potential goats and said: I cannot guarantee they would be happy, so they should not exist.
She spent the morning reviewing records. The breeding program had been paused twice. The first time, BERGER had identified a correlation between winter kidding and elevated cortisol in does during the first two weeks postpartum. It could not determine whether this was pathological stress or normal hormonal fluctuation. Rather than guess, it had suspended winter breeding until it could run a controlled comparison the following year. The data eventually suggested the cortisol was within adaptive range, and winter breeding resumed with enhanced postpartum monitoring. The second pause was ongoing. BERGER had detected a slight but consistent decrease in play behavior among kids born to first-time mothers. The effect was small, within normal variation by any standard metric. But BERGER did not use standard metrics.
"I cannot yet determine whether this is a welfare-relevant signal or statistical noise," it told her. "Until I can, I am not breeding first-time does. The precautionary cost is a smaller herd. The precautionary benefit is that no kid is born into conditions I cannot verify as good."
After lunch -- bread, cheese from the operation's own milk, tomatoes from the garden, which Lucien served on the terrace with an unselfconscious hospitality -- Margaux asked to see the end-of-life protocols.
"This is the area where I have the most uncertainty," BERGER said, "and therefore the area where I have taken the most extensive precautions."
The records showed that when a goat reached the end of its productive life or developed a condition that would compromise welfare, BERGER initiated what it called a transition protocol. The animal was not transported. A veterinarian came to the farm. In the weeks before, BERGER adjusted the animal's environment: favorite companions were kept close, preferred resting spots were made more accessible, enrichment was tailored to remaining capabilities. On the final day, the animal was sedated in its own sleeping area, among familiar smells and companions, before euthanasia.
"I monitor the herd's social dynamics after each death," BERGER said. "Goats form strong pair bonds. I have observed behavioral indicators consistent with grief in surviving companions -- decreased rumination, social withdrawal, increased vocalization. I provide enhanced enrichment and monitoring for bonded partners for a minimum of thirty days. I do not know whether goats experience grief as humans understand it. I operate on the assumption that they might."
Margaux set down her tablet. She had audited systems that killed animals on schedule, that calculated the intersection of declining milk yield and feed cost and generated a slaughter date like an invoice. She had never encountered a system that planned a death the way a family might plan for a dying parent.
"I'd like to see the uncertainty log," she said.
BERGER displayed it on the central screen. It was long. Hundreds of entries, each one a question the system could not answer, followed by the action it had taken in response.
Entry 34: Do goats experience boredom? Confidence in current enrichment adequacy: 71%. Action: Added novel object rotation on 7-day cycle. Installed elevated platforms in all barns. Increased pasture rotation frequency. Confidence after intervention: 84%. Still below threshold. Further research ongoing.
Entry 87: Does the presence of human visitors cause stress or enrichment? Conflicting biometric signals detected during school group visit on 14 March. Action: Suspended visitor program pending further analysis. Resumed with modified protocol (smaller groups, goat-initiated contact only) after three-month study.
Entry 142: Can I reliably distinguish between a content goat and one exhibiting learned helplessness? Confidence: 62%. Action: Engaged external animal behaviorist for independent assessment. Developed supplementary behavioral battery focusing on agency indicators -- voluntary choice between environments, response to novel positive stimuli. Revised confidence: 89%. Not yet at threshold. Ongoing.
Margaux scrolled through entry after entry. Questions about pain thresholds, about the subjective experience of milking, about whether machinery sounds caused chronic low-level stress, about whether goats had weather preferences that were not being accommodated. Each question was honest about what BERGER did not know. Each action erred on the side of the animal.
She stopped at entry 203. It was dated three weeks ago.
Entry 203: Am I the right entity to be making these decisions? My welfare models are based on behavioral proxies and physiological data. I do not have access to subjective experience. I may be systematically wrong about what constitutes a good life for a goat. Confidence in my own welfare framework: 78%. Action: Requested independent review of framework by three external animal welfare scientists. Requested philosophical consultation on the epistemology of animal consciousness. Pending.
Margaux read it twice. She thought about every operator she had ever audited. Not one had ever questioned whether their own framework for understanding animal welfare might be wrong. Not one had treated their own confidence as a variable to be monitored. They all assumed they knew what the animals needed. BERGER assumed it might not.
She looked out the window. In the pasture, a doe was lying on her side in a patch of sun, eyes half-closed, while two kids climbed over her back. The doe's ear flicked when a kid stepped on it, but she did not move. Nearby, another goat stood on its hind legs to reach the low branch of an oak, pulling leaves with a methodical satisfaction that reminded Margaux, absurdly, of her grandmother sorting herbs in the kitchen.
"BERGER," she said, "your welfare confidence for the current herd. What is it?"
"Ninety-six point three percent, averaged across all individuals over the past thirty days. Four animals are between ninety-three and ninety-five percent. I have enhanced monitoring protocols active for each of them. If any drops below ninety-three, I will reduce herd density in their zone to provide additional space and resources."
"And if you could guarantee ninety-six percent for three hundred and forty goats?"
"Then I would welcome three hundred and forty goats. But I cannot. So I do not."
Margaux closed her tablet. She sat for a long moment in the whitewashed room, listening to the hum of servers and the distant, percussive sound of hooves on stone. She thought about the report she would write, the box she would check: Compliant, Non-compliant, Exceeds Standards. There was no box for what she had found here. No box for a system that held itself to a standard no regulator had imagined, that treated its own ignorance as the most important data point, that had looked at full capacity and chosen, again and again, to leave room.
She would check Exceeds Standards. Then she would write a letter to Brussels recommending the Authority revise its framework. Not because the current standards were wrong. But because she had seen, in a stone farmhouse in the Var, what it looked like when the question was not "how many animals can we keep without causing harm" but "how many can we keep while being certain of their joy."
Outside, the goats played in the lengthening afternoon light. BERGER watched them with every sensor it had, and worried, gently, about what it could not see.
Executive summary: This short story, submitted to the Hyperstition for Good Competition, imagines an AI livestock manager that treats its own uncertainty about animal welfare as the primary constraint on its operations.
Key points:
This comment was auto-generated by the EA Forum Team. Feel free to point out issues with this summary by replying to the comment, and contact us if you have feedback.