July 1999, College Park, Maryland / Washington, D.C.
The heat that summer felt almost algorithmic, predictable, repetitive, a pattern that refused to decay. By the second week of July the humidity over College Park had the texture of static. The air conditioning in the criminology lab coughed instead of cooled, and the fluorescent lights hummed in an almost perfect sixty hertz rhythm. Stephen caught himself syncing his breathing to it without meaning to.
The field study brief was simple. Audit a decade of rehabilitation outcome data from multiple states.
Simple never stayed simple for long.
Reyes slid the files toward him at the first orientation meeting. "Federal dataset. Anonymous identifiers, but the methodology's older than you are. Find what still works."
He skimmed the headers, age, location, education, prior offenses, sentence length. Then he found an economic class indicator buried three columns deep, paired with a set of adjustment coefficients labeled Regional Behavioral Variance. It looked neutral on paper. Most biases do, at first glance.
He set up the equations on the whiteboard. R equals alpha one times social, plus alpha two times economic, plus alpha three times behavioral. Delta equals actual outcome minus predicted outcome.
If the model was sound, delta should have read as random noise. Instead the scatterplot tilted, bias hiding behind mathematics, a slope disguised as certainty.
That night, after the rest of the cohort left, Stephen stayed behind with two cups of stale coffee and a stack of printouts. Reyes had told the room earlier, numbers don't lie, but they can omit. She was right. The omission was the actual crime.
Vale called from Quantico three days later.
"Tell me what you've found."
"A reflection problem. The dataset thinks poverty predicts relapse better than opportunity does. It's been weighting the wrong side of the equation for years."
"We build mirrors again, don't we."
"This one reflects policy, not people."
He paused. "Then show them the distortion."
Paige's message arrived the next morning from her Jeffersonian account. Subject line, ethics paper, final proofs. The attachment name made Stephen grin. EFL_FP_v7_FINALfinal.doc.
She called an hour later, voice bright through the static of the lab phone line.
"They accepted it. Journal of Applied Computational Ethics. Out next quarter."
"What title did they give you?"
"Ethical Feedback Loops in Forensic Pattern Recognition. They said it sounded less terrifying than my first draft."
"Which was."
"When the Machine Learns Prejudice."
He laughed, then felt the symmetry between their two projects settle in. Same ghost showing up in different data.
"How's the field study," she asked.
"Broken. Beautifully broken."
She went quiet for a second. "Brennan told me my model's empathy is leaning. I told her maybe the equation needs a conscience."
"You're going to quote yourself in the next one, aren't you?"
"Probably. You?"
"Still arguing with mine," Stephen said.
When they hung up, he opened her draft and read the abstract's first line. Neutral design is not the same as moral design. He read it twice.
The next week he presented preliminary findings to Reyes. She listened in silence, pen hovering over a legal pad.
"So your correction model raises rehabilitation scores for low income regions by eleven percent," she said finally. "And lowers them for high income regions by eight."
"Right."
She glanced up. "That's uncomfortable."
"It's accurate."
"Accuracy and comfort rarely correlate." She said it like approval, because for her it was.
He sent Vale a copy of the draft that night. The storm rolled through afterward, heat finally breaking.
By mid month Stephen was living inside spreadsheets, the air conditioning's hum turning into a kind of metronome for thought. Every new regression output felt like peeling paint, familiar, inevitable, flawed. Every coefficient, every adjustment, carried a quiet confession about whoever had written it first. Bias didn't announce itself. It sat in the corners of formulas, in the weights nobody questioned anymore.
One afternoon Greenaway stopped by the lab on her way through campus, fresh off a Quantico symposium on behavioral forecasting. Stephen showed her the scatterplots, the slope of inequity obvious once you actually looked for it.
She tilted her head. "You're profiling systems now, Dr. Cooper."
"I'm just trying to get them to tell the truth."
"You're learning the same lesson we do. The story depends on who gets to tell it."
When she left, Stephen realized that was most of what profiling actually was, finding who'd been silenced inside the data.
Paige invited him to D.C. that weekend. They met near the Reflecting Pool, the air thick but gentler now that the rain had cooled it. The water mirrored the sky, still smeared with leftover heat.
She carried a copy of her printed paper, her name alone at the top, and handed it over like a trophy and a burden at once.
"They gave me ten author copies. Figured you deserved one."
He flipped through it, reading the abstract again in actual ink. Neutral design is not the same as moral design.
"You changed the world a little."
She shook her head. "Just measured it differently."
They sat on the edge of the pool, shoes off, reflections blurring in the ripples.
"You changed your formula too," she said.
"No. I just remembered the people in it."
"That's a change. You just don't want to call it one."
"It's a correction. Changes imply the original direction was arbitrary. This wasn't arbitrary, it was incomplete."
"You're allowed to just say you grew, Stephen."
"I'm allowing myself to say it precisely."
She bumped his shoulder with hers. "Same thing, dressed up."
"Possibly."
They didn't talk much after that. The silence held its own kind of balance, two systems settling around the same constant.
Two weeks later came the presentation. The seminar room smelled like chalk dust and coffee gone bitter, a projector fan whining behind him while the screen glowed white.
"Latent Variables in Behavioral Forecasting. A case study in how assumptions bias outcomes."
Some of the faculty nodded. Others frowned like he'd just criticized their handwriting. When he showed how the dataset's regional corrections penalized poverty directly, a few eyes narrowed.
One professor interrupted. "Dr. Cooper, aren't you moralizing statistics."
Before Stephen could answer, Reyes spoke from the back row. "Morality is what tells us the model is wrong."
The room went still. The questions softened after that.
He closed with the revised equation projected in large type. Outcome equals Behavior plus Environment plus Mercy, beta term. The beta wasn't real math. It was an idea standing in for a variable, mercy given a place in the formula instead of being treated as noise.
Afterward, Reyes shook his hand. "Submit the paper to the review board. Keep the mercy term. They'll hate it, which means it matters."
That night Stephen took the Metro into D.C. again. Paige met him outside her building, hair pulled back, still wearing her Jeffersonian badge lanyard.
"They already cited my paper," she said. "Someone at NIH sent a memo about bias mitigation standards. I think they're borrowing it without asking."
"Good. Means it's useful."
They walked without much destination, the air still humming faintly from the city's generators. Somewhere downtown leftover fireworks from the Fourth popped late, reflecting off the glass of nearby office buildings.
Paige looked up. "You ever wonder if we're just teaching our equations to see us the way we want to be seen?"
"Constantly. Still better than letting them see nothing at all."
On the thirtieth, Stephen finished the practicum summary and sent it to Reyes, then a copy to Vale. He didn't add anything extra to the file before closing it. The monitor light faded to black, and the lab felt different in the dark, less sterile, closer to ordinary.
Outside, past the old oak trees behind the building, the night was thick with cicadas. He thought briefly about Vale's mirror, Paige's algorithm, Greenaway's read on profiling, none of them chasing anything close to perfection, just trying to find what was actually true underneath the clean lines everyone preferred.
Then he headed for the train, the rails humming under the carriage the whole way back toward Takoma Park.
(Thanks for reading, feel free to write a comment, leave a review, and Power Stones are always appreciated. Let me know if you find any mistakes)