Read Chapter 4: Resonance

Arun found the old model in a cold-storage cluster under a project name that sounded like an indie band.

BLUE ORCHARD.

“Subtle,” Eleanor said.

“Data scientists should not name covert programs.”

They were in a secure testing room after midnight. The walls blocked wireless signals. The coffee machine produced something officially classified as espresso.

Arun loaded the archive.

“Resonance was built from longitudinal relationship data,” he said. “Not just whether couples stayed together. Whether their trajectories changed after meeting.”

“Changed how?”

“Everything measurable. Vocabulary. Sleep cycles. neighbourhood. career. financial risk. political attitudes. friendship networks. health routines. creative output.”

“That sounds terrifying.”

“It is also what humans call influence.”

“How did it score?”

“Mutual prediction error.”

Eleanor turned to him.

“Explain.”

“Vela likes predictability. If I know your preferences and his preferences, I can estimate your future. Resonance looked for pairs where meeting each other made the future harder to predict but not random.”

“Structured divergence.”

Arun smiled faintly. “You remember the papers.”

“I read compulsively when angry.”

“The model identified people who produced new stable states in each other. Not chaos. Expansion.”

Eleanor looked at Theo’s score.

99.84.

“Why the hard conflicts?”

“Because the same qualities that create divergence often create friction.”

“Then why was it ever used for relationship guidance?”

Arun’s expression closed.

“Founders are occasionally overconfident.”

“Nathan.”

“And others.”

“How long did the trial run?”

“Fourteen months.”

“How many users?”

“Eight thousand.”

“How many adverse outcomes?”

“I don’t know.”

She stared.

“You do.”

“I know the dataset was sealed.”

“That is not the same answer.”

Arun rubbed his face.

“Eleanor, I was twenty-eight. I was on a modelling team. I didn’t design the interventions.”

“But you worked on Resonance.”

“Yes.”

“Did you know about Anna March?”

His silence was answer enough.

Eleanor stepped away from the console.

The room felt too small.

“You knew his name.”

“I recognised it when you showed me.”

“And you said nothing.”

“I told you not to dig alone.”

“That is not an apology.”

“No.”

“Try again.”

Arun met her eyes.

“I was afraid.”

That stopped her.

“Of what?”

“Of finding out we had built Lattice on top of something we never properly buried.”

He opened another archive.

After the Continuity Study, the original Resonance system had been decommissioned.

But parts of its embedding architecture had been reused.

Not in matching.

In retention.

Eleanor read the module name twice.

RELATIONAL FRICTION PREDICTION.

“What is this?”

“Premium save intervention.”

She knew the feature. Of course she knew it. When couples using Lattice’s relationship mode showed patterns associated with separation, the app suggested conversations, counselling resources, or temporary communication boundaries. It was one of the company’s proudest public-safety products.

Arun pointed to the code lineage.

“The system predicts which relationships are likely to reshape both people. Those relationships have high emotional investment and high churn risk during conflict.”

“So Growth uses Resonance to identify who not to lose.”

“Indirectly.”

“And we suppress Resonance in matching because it creates unstable couples.”

“Yes.”

“But once unstable transformative couples exist, we monetize keeping them together.”

Arun did not answer.

Eleanor felt anger become very cold.

“Who knows?”

“Nathan.”

“Anyone else?”

“Celia, probably.”

Celia Ward, chief commercial officer.

Of course.

Eleanor returned to Theo’s score.

“Why is my profile being compared?”

Arun ran a trace.

The answer came from an old internal experiment.

Eleanor’s profile had been used four years earlier to calibrate Vela Two because her behavioural history was unusually complete. Every product test, commute experiment, sleep optimisation pilot and communications study had made her, without anyone planning it, one of Lattice’s richest longitudinal subjects.

Resonance had kept her as an anchor.

Theo’s sparse profile should have made him unscorable.

Instead, the model had inferred him from negative space.

Permissions refused. Routes not predicted. Messages unassisted. Choices made without system prompts.

Against Eleanor’s extreme legibility, Theo was nearly illegible.

Yet the model kept producing the same result.

Arun whispered, “That’s beautiful.”

Eleanor looked at him.

“Professionally beautiful.”

“It’s horrifying.”

“That too.”

On the screen, a hidden note appeared beneath the score.

COUNTERFACTUAL STABILITY: HIGH RECIPROCAL NOVELTY: EXTREME IDENTITY EXPANSION: EXTREME COERCION RISK: LOW DEPENDENCY ASYMMETRY: LOW

Eleanor read the final line.

AUTONOMY PRESERVATION: 0.97

Anna March’s relationship, she realised, must have failed that metric.

“Was autonomy preservation added after the study?”

Arun nodded slowly.

“When?”

“After St Pancras.”