THE ERROR BETWEEN US
Read Chapter 1: The Impossible Match
At 6:42 on a wet Tuesday morning, Eleanor Vale discovered that the most expensive relationship algorithm in Europe thought she should date a man it had already forbidden her to meet.
She was standing barefoot in the kitchen of her flat in Clerkenwell, watching rain stitch silver lines down the glass. Her kettle had finished boiling. Her coffee had not. She was reading the overnight anomaly digest because she had spent eleven years training herself to believe that an unread dashboard was a form of negligence.
Lattice’s internal console floated above the counter in pale blue panes.
MATCHING HEALTH Global latency: nominal. Recommendation drift: nominal. Safety overrides: nominal. Outlier events: 3.
Eleanor pinched the first pane open.
Two were ordinary. A user in Bristol had connected three incompatible identity wallets after a legal name change. A couple in Glasgow had caused a recursion loop by mutually blocking and unblocking each other twelve hundred times in twenty minutes.
The third event made her coffee go cold.
SUBJECT A: E. VALE / INTERNAL PROFILE SUBJECT B: T. MARCH / USER 8F2-19A PRIMARY COMPATIBILITY: 99.84% RELATIONSHIP VIABILITY: 04.11% HARD CONFLICT FLAGS: 17 RECOMMENDATION STATUS: SUPPRESSED
Eleanor stared at it.
“Again?”
Her flat, naturally, offered no answer.
She expanded the history.
The same pair had surfaced six times in nineteen days.
The production system should never have compared her to anyone. Employees could volunteer anonymized profiles for quality testing, and Eleanor had done so years ago, but internal profiles were air-gapped from live recommendation queues. More importantly, even if hers had escaped into production, a match carrying seventeen hard conflict flags should have been discarded before primary scoring.
Instead, the system had done something computationally absurd. It had scored them first, fallen almost violently in favour of the match, and only afterward remembered that it was supposed to reject them.
She opened Subject B.
Theo March, thirty-six. London. Occupation: acoustic systems designer. Relationship objective: long-term partnership. Children: uncertain. Political disclosure: withheld. Religion: none declared. Social graph permission: minimal. Behavioral history permission: minimal. Mood prediction permission: refused. Adaptive messaging assistance: refused. Biometric intimacy layer: refused. Predictive date routing: refused.
Eleanor gave a small, involuntary laugh.
Lattice users could refuse those permissions, but hardly anyone refused all of them. It was like buying a self-driving car and insisting on using the steering wheel.
She scrolled.
Lifestyle mismatch: high. Conflict style mismatch: high. Planning horizon mismatch: severe. Novelty preference mismatch: severe. Career centrality mismatch: severe. Algorithmic trust mismatch: absolute.
That last category was new enough to irritate her.
“Absolute,” she repeated.
The kettle clicked as it cooled.
At seven-fifteen, she was in a driverless cab moving south through streets that smelled of rain, diesel heritage engines, and wet brick. London had changed unevenly. Glass towers in the City adjusted their skins to daylight. Delivery drones moved in regulated corridors above the Thames. Yet outside Smithfield Market, two men in fluorescent coats were arguing beside a van that looked old enough to vote.
Lattice occupied six floors of a converted printing warehouse near Blackfriars. The company liked the symbolism. Old machines had once turned ink into public opinion. New machines turned data into private decisions.
Eleanor entered through biometric security and found Arun Patel already in the product war room.
“You look offended,” he said.
“I am offended.”
“By a person, a metric, or the existence of Monday?”
“Tuesday.”
“That bad.”
Arun was Lattice’s director of applied machine intelligence and one of the few people who had known Eleanor before her job title became a public noun. He wore a jumper that said PLEASE DO NOT HUMANISE THE MODEL, which was funny because he humanised models constantly.
Eleanor dropped the anomaly onto the wall.
His expression changed.
“Oh.”
“You’ve seen it.”
“No.”
“You made that face.”
“I have a broad repertoire of professional concern.”
“Arun.”
He approached the display. “Ninety-nine point eight?”
“Four viability.”
“That should not be possible.”
“Exactly.”
He reached toward the metadata, stopped, and glanced at her.
“Why is your profile in production?”
“It isn’t.”
“That was not a philosophical question.”
She folded her arms. “Find out.”
Arun zoomed into the scoring lineage. Lattice’s public matching product was called Vela, after the constellation. Users saw a clean number, a short explanation, and a sequence of suggested actions. Underneath it sat seven interacting models: preference alignment, stress response, conflict recovery, practical compatibility, attraction persistence, network friction, and long-horizon stability.
But this match included an eighth lineage.
A line Eleanor had never seen.
RZ-CORE / RESONANCE weight contribution: 0.781
“What is RZ?” she asked.
Arun did not answer.
That was answer enough.
“You know.”
He lowered his hand.
“I know the name.”
“Then use the name.”
“Resonance.”
“I can read.”
“It was a research model. Years ago.”
“How many years?”
“Before Vela Two.”
Eleanor felt something tighten behind her ribs. Vela Two had launched four years earlier, after a redesign she had led.
“Why isn’t it in the architecture map?”
“Because it was killed.”
“Apparently not.”
Arun looked at the rain-blurred windows.
“Eleanor, don’t go digging alone.”
The sentence was wrong in at least three ways. It sounded like a warning. Arun hated melodrama. And he had not told her to stop.
She turned back to Theo March’s profile.
His photo was not visible. He had declined visual pre-screening.
Of course he had.
“What did Resonance measure?”
Arun’s jaw shifted.
“Change.”
“What kind?”
“The kind people say they want until it happens.”