Pattern Library

The Pattern Library had a user interface. That fact made Elias angrier than the technical documents.

Someone had designed buttons. Ivo restored an internal Fallow application from Orpheus. The screen looked ancient: grey panels, blue tabs, tiny icons. On the left were categories of relational behavior. On the right, donor clips.

MATERNAL REASSURANCE.

PATERNAL HUMOR.

SIBLING CONFLICT.

REPAIR AFTER ARGUMENT.

NONVERBAL AFFECTION.

A researcher could select a target and preview source examples. Helen appeared under MATERNAL REASSURANCE.

A clip showed her sitting beside twelve-year-old Mara after a school argument.

"You don't have to like what happened to survive the afternoon," Helen said.

Elias had forgotten the scene. Mara had not.

"That was after I got suspended."

"You were suspended?"

"See? This investigation is already improving your childhood."

The interface allowed researchers to rate clips for emotional clarity, naturalness and transferability. Helen's clip scored high on all three.

Mara stared at transferability.

"They rated how well Mum could belong to someone else."

No one corrected her. The library included Martin Saye, Gabriel, dozens of other parents, children and spouses. People were not stored as whole people. They were segmented by useful moments.

A laugh. A look across a room.

A hand on a shoulder. An apology.

Ivo opened usage history. Aster House researchers selected clips manually at first.

Later versions automated selection. By 2012 the system could choose behavior patterns based on therapeutic goals.

"There," Elias said.

One setting read MAXIMIZE FAMILIARITY. Another read MAXIMIZE SOURCE FIDELITY.

They were separate sliders. Often moving one reduced the other.

Mara looked at Lena. "Which did therapists choose?" Lena answered softly. "Usually familiarity."

"Why?"

"Because clients tolerated the session better."

"Meaning they liked it."

"Often."

Celia asked Ivo to export the configuration logs. One statistic stood out.

Across 318 early sessions, average source fidelity had been 61 percent. Average perceived familiarity was 88 percent.

The system had become good at making scenes feel like home before it became good at reconstructing what happened there. Elias looked at the interface.

A button at the bottom read ACCEPT AS BASELINE. He imagined a tired therapist clicking it after a grieving person said, yes, that feels like her.

One click. Then the borrowed detail became the starting point for the next session.

The Pattern Library interface had been built for speed. Elias hated it for that. Every ethical uncertainty that had consumed months of their lives had once been reduced to filters, sliders and colored confidence bars.

Ivo opened a training version disconnected from patient data. "Look at this."

A therapist could choose desired emotional outcome: reassurance, closure, reduced guilt, relational continuity. Under each outcome, the system ranked patterns associated with positive post-session scores.

Mara leaned over his shoulder. "Closure is a menu option."

"Apparently."

"Can I order two closures and a side of fries?"

The joke died when they opened the historical usage notes.

Pattern Fourteen, practical care after vulnerability, had exceptionally high acceptance. Pattern Nine, shared humor after conflict, reduced session abandonment. Pattern Three, explicit verbal forgiveness, tested badly in some cultures and well in others. Engineers had added demographic filters to improve fit.

"They A/B tested grief," Tess said.

Ivo winced. "Technically, yes."

"Do not say technically before something horrible."

Elias studied the old UI. The researchers had believed they were helping clinicians avoid scenes that felt emotionally implausible. A bereaved person rejecting a reconstruction could experience renewed distress. Pattern selection was supposed to reduce that risk.

Then the outcome metric became a goal.

The distinction was visible in version notes. Early releases displayed source confidence above acceptance. Later releases placed emotional congruence first because therapists used it more often. Eventually a recommended pattern could be accepted as baseline with one click, carrying it into subsequent sessions as if it belonged to the family model.

"There," Elias said. "That's the ratchet."

Mara nodded. "Borrow once, then treat acceptance as evidence you borrowed correctly."

"Then future reconstructions see the borrowed detail as part of the person's history."

Tess wrote that down, then crossed it out. "Too clean?"

"Too clean," Mara said. "Some people probably did have similar behavior. Some borrowed things accidentally matched."

"You are destroying journalism as a profession."

"Aim higher."

They tested a historical donor phrase against later family models. The same basic reassurance moved through six generations of reconstructions, changing slightly each time. A sentence from one father became a gesture in another family, then a combined dialogue-action pattern elsewhere. Provenance became diffuse.

"Like quoting someone who quoted someone who forgot it was a quote," Ivo said.

"Except the quote becomes a memory," Elias replied.

At the bottom of the interface, ACCEPT AS BASELINE remained visible.

Lena arrived midway through the review and stopped when she saw it. "I clicked that."

Mara turned.

"How often?"

"I don't know. Hundreds of times, probably."

"On me?"

"At least once."

The answer sat between them.

Lena did not explain that she had followed protocol. She did not say everyone did it. She simply stood there looking at the button.

Mara walked over and closed the application.

"I don't need to watch you punish yourself," she said. "I need you to help prove what the button did."

Lena nodded.

That afternoon they mapped exactly how baseline acceptance propagated. The result was more damaging than any dramatic confession. A single click could alter which future gaps the system considered already resolved.

The interface had made doubt inconvenient.

That was enough.