The Model

Joon-Ho arrives a day early.

She knows it's him at the door before she opens it because she has been watching the lane from the kitchen window and she saw him coming slight, moving faster than his usual pace, the rumpled coat from the Edinburgh meetings, the quality of someone whose body is moving faster than their composure can keep up with.

She opens the door. He says: "I need to show you something. And I need to apologise first, because once you've seen it the apology is going to be harder to believe."

She steps back from the door. "Show me first. Then apologise."

He comes in. He sits at the kitchen table. He opens his laptop and turns it to face her.

The screen shows a probability model. She has used probability models in fraud investigation for three years the specific architecture of input variables, weighted outcomes, confidence intervals. This one is more complex than anything she's worked with professionally. The variable names in the left column run for two pages. The outcomes in the right column are numbered to forty-seven.

Her name is in the header. Ashcroft, V. Vein Integration Timeline Model. Version 23.

She sits down.

She reads. She reads with the quality of attention she brings to documents that are going to tell her something she doesn't want to know: carefully, without rushing the conclusion, making sure she has the complete picture before she decides what to do with it. It takes twenty minutes. Joon-Ho sits across from her with his hands flat on the table and says nothing, which is the correct call.

Version 23. She counts the variables in the left column. Two hundred and forty-one.

She looks up. "How long."

"Seven months."

"Twenty-three versions means you revised it twenty-two times."

"Yes."

"Each revision incorporated new data."

"Yes."

She looks at the screen again. She finds what she's looking for on the third page — a variable listed as Subject disclosure tolerance: information received vs. processed, set to a value between zero and one, with conditional branches running six rows, collectively representing a map of what she would do at various points if given various amounts of information at various rates.

Variable 89: If Subject receives Isadora document without prior Vein framework, probability of withdrawal 0.71.

Variable 90: If Subject receives Vein framework without prior bloodline context, probability of overload 0.83, followed by self-isolation 0.64.

Variable 91: If Subject receives full context in correct sequence at correct pace, probability of integration 0.61.

She closes the laptop.

"You decided the sequence," she says.

"Yes."

"And the pace."

"Yes."

"You had two hundred and forty-one variables in a model with my name in it and the model was designed to determine the optimal way to give me information about my own life."

"Yes." He says it without deflection. Without the protective language of intention or care. Just yes. "I want to be accurate about what I did. I didn't lie to you. I didn't withhold the truth maliciously. I managed the sequence and pace of your access to your own history with the specific intention of producing the outcome the model identified as most probable for successful integration." He pauses. "I told myself this was care. I understood, in Florence, when you weren't there and the meeting was happening without you, that it was something else. I understood it more clearly when you left. I have been understanding it with increasing precision for the past three weeks."

She looks at him. "What's the difference. Between what you did and caring."

"Caring asks. What I did assumed." He looks at his hands. "Caring would have been showing you the model and asking you whether you wanted to be in it. What I did was build the model and implement it and call the implementation care."

Variable, who has come in from somewhere, is sitting on Joon-Ho's closed laptop. This is Variable's specific form of provisional acceptance.

She says: "Version 23. What changed from version one."

"You did. Every time I met you or received information about you, the model updated. You were more you consistently exceeded the predicted values. Your processing speed, your tolerance for complexity, your ability to hold contradictory information simultaneously without" He stops. He says, more quietly: "You were always better than my model expected. Which means the model was always wrong, which means"

"Which means you were constraining something that didn't need constraining," she says.

"Yes."

She opens the laptop. She looks at variable 34: Subject discovery of optimisation framework. Probability of withdrawal: 0.89.

"It predicted eighty-nine percent probability that I'd leave if I found the model," she says.

"Yes."

"I'm still here."

"Yes."

"Why do you think that is."

He is quiet for a moment. Then: "Because the model assumed your primary response to being managed would be to leave. It didn't account for the possibility that you might decide the managing was something you could work with rather than something you had to escape." He looks at her. "It didn't account for you choosing to stay in a situation despite what someone had done rather than because of it."

She closes the laptop. "I want the model deleted. All twenty-three versions."

He opens it. He navigates to the file. He deletes it, empties the trash, verifies the cloud sync. He shows her the empty folder.

"Thank you," she says.

"I'm sorry. For what it's worth at this stage."

"It's worth something. It's not worth what you think it's worth yet. Those aren't the same thing and the difference matters."

"I understand."

She makes coffee and sets a mug in front of him and sits across from him. She says: "Tell me about ARC. What you actually do. Not the professional summary. The actual thing."

He looks at her. Something under the carefulness in his face that is closer to relief than anything else.

"It started with weather patterns," he says. "I was twelve."

She says: "Tell me."

He does.