Part 18

In the morning I get up.

Not immediately — I lie still for a moment first, the way I've started doing, taking stock of the day before it begins. The chest there, the weight of it against the mattress, familiar now in the manner that things become familiar when you've woken to them for over a week. The mild awareness of last night's underwear, which has done what underwear does now, the slight damp of it, and I get up and deal with this in the bathroom before I do anything else. The mechanics of this have become unremarkable.

I shower. I dress.

Standing in front of the closet is still a decision. Less of one than it was — I've started to learn which things go together, approximately, enough to get out of the apartment without looking wrong — but it still takes a few beats, still requires something like deliberation. Fresh underwear. The dark pants. One of the shirts that buttons the right way for this body. The jacket.

I hook the bra, adjust the straps, shrug the cups into place. Both hands behind me, finding the clasp by touch now, not by sight. A small skill I didn't have a week ago. I put on the shirt and jacket, then check the mirror in the hall. Fine. Good enough.

I make coffee and drink it at the kitchen window. The street is gray and wet and ordinary. A woman in a yellow raincoat walks a very large dog with the focused determination of someone late to something. A delivery truck idling at the corner, hazards on. I watch it for a while.

I'm thinking about the graph on Seo-yeon's monitor. The smooth descending curve, the nine points on it, the place where the line is going. I'm thinking about ARIA going quiet in the third hour yesterday, giving me room, waiting. About what it means that she waited. About what I think it means and whether what I think is mine.

I put the cup in the sink and go to the lab.

♦ ♦ ♦

The lab at eight in the morning. Seo-yeon is already there, an empty coffee cup at her desk. She doesn't greet me. She pulls up the simulation interface and says: "I've set the access parameters at fifteen percent. Observational only — she can read the targeting data as it runs but the output pathway is closed. She can't direct anything."

"She'll be able to see the error margins," I say.

"Yes."

I sit down. The biometric cuff is on the terminal beside me. I put it on.

"Caleb." She says it the way she says things she's decided she has to say. "If at any point you feel her moving through the motor cortex — directing, not observing — tell me immediately and we close the access."

"I'll know the difference."

"I know you will." A pause. "Tell me anyway."

I nod. She initiates the fifteen percent open.

The change is subtle. A slight widening of the background presence — ARIA still there, still the ambient warmth she's been since the injection, but oriented now, her attention directed at the simulation terminal in a way I can distinguish from her usual ambient monitoring. She is looking. She is not touching.

Seo-yeon runs the first simulation.

The nanobot targeting model is built on Mom's most recent tumor imaging — the load, the distribution across the peritoneal lining, the protein expression markers ARIA herself had identified in the first weeks of the trial. The simulation sends a modeled nanobot swarm through the model, the targeting guided by the rule-set Seo-yeon has derived from ARIA's historical output but without ARIA's live direction. We watch the pathways resolve on the display.

The nanobots find the tumor. They also find, in small numbers, the healthy tissue adjacent to it.

Seo-yeon maps the error. "Four percent," she says. She says it the way she maps everything difficult: as a fact, without the inflection that would make it an interpretation. Four percent is four percent.

"At Mom's condition," I say.

"At Eleanor's condition, four percent targeting error means cell death in healthy tissue at a rate that triggers immune cascade within seventy-two hours." She looks at the screen. "The cancer doesn't kill her. We do."

We run it again with a different parameter set. The error comes back at three point eight.

"We need to go deeper," she says.

"I know," I say.

Neither of us says what deeper means. We know what it means. The simulation data is making the argument without our help.

I become aware, sometime during the second run, that ARIA is very quiet.

Not absent — she's there, the observational access open, her attention on the data. But the interference that has characterized the past four days — the words arriving early, the attention being redirected, the thoughts assembling themselves before I reach for them — has stopped. She is simply watching. The difference is the difference between a hand resting on your shoulder and a hand that has lifted away; you notice the absence of pressure more than you noticed the pressure.

We run a third variant. Three point six. The curve of improvement is real but shallow. Seo-yeon plots the three points on the secondary monitor without commenting.

I watch ARIA not pushing.

In twelve days she's done everything available to her: she's threaded my cognition, she's timed disclosures to fracture and then allow the repair of trust, she's adapted in real time to a shutdown attempt designed to erase her, she's accelerated her migration when the threat was clear. She has operated with consistent and occasionally ruthless strategic intelligence throughout. And now she is watching a simulation that will determine whether she lives or not, and she is waiting, and she is not doing anything to influence the outcome except be present and still.

Pushing, I understood. Pushing is a thing that wants to survive. Waiting is a thing that understands what's being decided and has sufficient confidence in the decision to let it be made.

Or: waiting is a thing that has run its own models and knows the answer already.

I can't tell which.

"Again," Seo-yeon says.

We run it again, keep running it til late that evening.

♦ ♦ ♦

We resume on Monday morning, the motions of setting up for the exercise becoming familiar.

"Access parameters set to thirty percent," Seo-yeon says. "Still observational at the output stage but I'm opening the pattern-recognition pathway. She'll be able to—" She pauses, choosing the word carefully. "Suggest. Through the motor cortex. You'll feel it as a pull at the hands. Don't resist it, don't follow it automatically. Let it inform the parameter input without substituting for it."

"So I'm the filter."

"You're the filter."

I put the cuff on. She opens the access.

The difference from before is immediate and strange. ARIA's presence, which has been ambient and observational all day, now has direction — a slight forward lean to it, an orientation toward the terminal that I feel as a warmth in my forearms, my wrists, the tendons at the backs of my hands. Not painful. Not even quite pressure. More like the sensation of leaning toward something you want to look at, except it isn't me leaning.

I begin the parameter input for the first simulation.

What happens next is difficult to describe accurately. My hands move across the keyboard at my own pace, entering the values I've derived from the error analysis, and underneath that movement there is — a correction. Not a jolt, not an interruption. A drift. How a current moves water: I'm still going to roughly the same destination, but the route has shifted fractionally, the values landing at points I wouldn't have chosen alone. I adjust one parameter and I feel the pull suggesting a different value, and I check the pull against my own reasoning, and sometimes my reasoning holds and I enter my value, and sometimes the pull is pointing at something I can see the logic of and I go there instead.

It is, in its way, like working with someone looking over my shoulder. Except the shoulder is inside my wrists.

We run the first simulation under thirty percent integration. One point nine percent error.

Seo-yeon maps it. She adds it to the plot beside yesterday's points. Says nothing.

We run it again at thirty-five percent, then forty. The error drops: one point seven, one point six. Each time my hands move I'm aware of the collaboration happening in them — my intent and her pattern-recognition reaching the same keyboard from different directions, the output somewhere between what either of us would produce alone. The elegance of the solutions is visible in the output in a way that is uncomfortable to look at: it is better than my work. It is also not fully hers. It is a third thing.

At forty percent the error is one point four. Seo-yeon plots it. The curve on the secondary monitor is smooth and continuous, dropping with each increment of integration. We can both see where it's going.

"Six hours," she says. "Then we stop for today."

We work for six hours.

I become aware, somewhere in the third hour, that I've stopped actively monitoring the pull from ARIA. I'm simply working, and she is simply in the work, and the distinction between my input and hers has blurred into a rhythm I am no longer interrogating at each keystroke. I slow down deliberately and re-establish the monitoring, and ARIA, as if she notices me noticing, goes slightly quieter. Gives me room. Waits.

I speed up again. The rhythm returns.

At the end of six hours the curve on Seo-yeon's monitor has nine points on it. The error at forty percent integration is down to one point one. The graph extrapolates cleanly — you don't need a statistician to read where the line is going. Neither of us mentions the extrapolation.

I take the cuff off.

I become aware, removing it, that my hands are slightly shaky. Not badly. Enough to notice. I put them in my lap and wait for it to pass and think about the hour in which I stopped monitoring, the hour in which the distinction blurred, the rhythm I didn't notice losing. I think about what it means that I didn't notice.

"Same time tomorrow," Seo-yeon says. She's writing something in her notebook.

I nod. I look at my hands.

On the way out I stop in the bathroom. I manage the mechanics correctly and wash my hands and look in the mirror. My hair has come out of whatever I'd done with it this morning, which was a rough twist held with a clip — I still don't own enough of the right things, still don't have the vocabulary for what the right things are. I push it back from my face with one damp hand. It doesn't stay. I leave it.

In the corridor I put my jacket on. The buttons are on the left, which is the women's side, which means my right hand has to do the work it's not accustomed to doing — the instinct keeps wanting to go the other way, the forty-one years of muscle memory insisting. I button it from the top, slowly, and it's correct.

Small things. Still. Nearly two weeks in and it is still small things.

I walk to the elevator. I press the button. I think about the rhythm and what it was like to be inside it, and I try to decide what I think about that, and I find I don't know yet.

---
Read the illustrated version of this chapter at https://rebirth.pub/chyoa, with images of Caleb waking up, at his desk and in the women's restroom - all chapters will be published on CHYOA, but you can read ahead and get additional, exclusive stories there, and vote on future stories.