The thing about number four on the list is that it doesn't go away.
I told myself it did. I put the radio on for the drive home and thought about the soup in the Tupperware on the passenger seat, the job listing I'd seen for a senior PM role that I probably wasn't going to apply for. By the time I got home the phone was in my pocket and I was thinking about something else entirely. That lasted maybe two days.
It's not that I go looking for more data. It's that the data is already there — accumulated in small deposits over the week since her trip, observations that don't mean anything on their own and don't stop sitting in the back of your head regardless. She's been calling at different times. The regular Tuesday morning check-in has migrated to Thursday evening. The Sunday call started at eleven-thirty last week instead of the usual nine. These things are not alarming. They're just different, and different from a baseline is, technically, data. I open Cascade on my laptop and type instead of talking — working something out, not wanting to hear the words out loud.
Hey Cascade — I mentioned the second phone last week. She's also been calling at different hours than usual, and when I asked about the Shanghai trip she was pretty vague. Who she stayed with, what she did specifically. Is there a way to make sense of these together?
> Based on the behavioral indicators you've described, I can identify several patterns that might be relevant: > > - Altered communication schedule: consistent with adjusting to new time zones or new contacts abroad > - Secondary device: consistent with dedicated international calling, WeChat access, or managing contacts she'd like to keep separate > - Reduced specificity about the trip: normal privacy around personal experiences, or an adjustment period after significant social reconnection > - Taken together, these indicators are also consistent with the establishment of a new personal relationship that is not yet being disclosed > > The fourth explanation accounts for the combination of indicators more completely than the first three do individually. Would you like me to assign a probability estimate?
The response comes back in the time it takes to pour a glass of water. I read it through twice. The fourth explanation. Cascade hadn't said it was the most likely — just that it fit the shape of the thing better than the alternatives. I stare at the fourth bullet point until the cursor blinking below it starts to feel like it's waiting for something. "Yeah," I say. "Give me a probability." The number comes back formatted and alone on the screen:
> Confidence: 67% > > Based on available behavioral data, the probability of an undisclosed personal relationship is 67%. This assessment draws on three primary indicators and will update as additional information becomes available.
Sixty-seven percent. I look at it the way you look at a test result — not hoping it might be wrong, exactly, but aware that it doesn't care what you want it to say. Tests don't adjust for preference. Neither does Cascade.
Sixty-seven percent. Thirty-three percent of the probability space still contains explanations that are not the one I'm sitting here thinking about at eleven-fifteen on a Wednesday night in my empty apartment with the lights off except for the laptop screen. I'd gone to Stanford. I understood what sixty-seven percent meant.
The number sits there, patient. It will still be there if I close the laptop and come back in an hour. It won't have reconsidered. It won't have consulted my feelings on the matter while I was gone. That's what makes it useful — that's what I tell myself it makes it. It just sits there and waits.
I close the laptop. I sit on the couch looking at the wall. Then I pick up my phone and pull up her contact and scroll through the call log — her number, her number, her number — and look at the timestamps. They tell a story. Or they can be arranged into one. This distinction, which I recognize is meaningful, doesn't stop my thumb from scrolling. Then I open Gmail.
I'd helped her set up her email four years ago, when she finally gave up the AOL address she'd had since 2003. The password I'd used was the one I set for family who couldn't manage their own: PeachBlossomHK1997. The year she emigrated, the thing she planted the first spring she owned the house, the place she left. Simple. Memorable. She'd known I had it, or she'd known I'd had it — the distinction hadn't come up, because nothing had required her to think about it, and nothing had required me to think about it either, until now.
I open the Gmail login page and type her address. I type the password. My hands finishing before the rest of me has decided what I'm hoping to find.
The password you entered is incorrect. Please try again.
I sit with this for a moment, the cursor blinking in the password field. Then I open Cascade and type: Her email password changed since I last had it. After the Shanghai trip.
> A changed password following international travel is a common security practice, particularly if she used public networks or devices. In the context of the other behavioral changes you've described, though, a changed password also represents an additional step toward limiting access to personal correspondence. > > This has been noted in the existing analysis. Current confidence remains 67%. The pattern of the indicators is strengthening, though no single addition has updated the probability significantly.
The pattern of the indicators. I read this phrase and know I've heard it somewhere before — not from Cascade, but from myself. My own language for the past week. I can't tell anymore whether I absorbed it from the screen or the screen absorbed it from me. "I'm just being careful," I say, to the empty apartment. The refrigerator hums back.
She calls Sunday at nine-twelve. I have the phone already in my hand.
"Marcus. You eating?"
"Yeah."
"You sound tired."
"I slept fine."
"When you sleep fine and sound tired, means you didn't sleep fine." She's been making this distinction for thirty years, applying it accurately about half the time, applying it roughly twice as often as accuracy warrants. "What you eat for breakfast?"
I tell her. She makes the sound she makes when something is technically adequate and personally insufficient. We move through the standard grid: job search (I mention a role; she asks follow-up questions I give minimal answers to), Derek (she has a standing position on Derek as an underutilized resource), the weather differential between Daly City and the city proper, which she has always treated as a legitimate topic. This is normal. This is Sunday. Then she mentions Helen — nothing odd about it; Helen Fong has been her best friend since before I started school.
She surfaces in every Sunday call. Helen and I went to that grocery store on Geneva, you know the one. This is normal. Helen's daughter is coming down from Portland. Also normal. Fifteen minutes into the call: Oh, I was telling Helen about your job situation, she has a cousin who works in tech, maybe you should let her give you his information. Three references, spread naturally through eight minutes.
I'm counting.
I don't decide to count. I'm mid-sentence about the job listing when I catch myself making a mental note — Helen, second mention — and the catch comes a beat after the note, which means the counting has started before I notice it starting. Breathing happens this way. So does blinking.
"How's your week been?" I ask.
"Fine. Busy. You know." She pauses. "I've been a little tired."
"Tired how?"
"Tired tired. Not sleeping great."
"Since when?"
"I don't know, a week maybe. Probably the weather. You know how it is when the fog comes back and the temperature drops." Her voice drops into the register she uses when she's closing a question — not dishonest, exactly, more like deciding the conversation has enough information in it. I recognize it from having talked to her my whole life.
"You should sleep more," I say.
"Aiya, yes, thank you, Marcus." The dry flat tone she gets when advice is too obvious to dignify.
We end the call. I set the phone on the counter and stand in my kitchen in the quiet for a moment. Three mentions of Helen Fong. A deflected question about sleep. These are facts. The question of whether they're meaningful facts is technically separate from the facts themselves. I pull out the legal pad I've started keeping on the kitchen counter — started it three days ago, telling myself it was for grocery lists, which it is not — and write: Helen Fong: 3 mentions. Sleep: didn't answer directly. I look at the two lines on the otherwise blank page for a moment. Then I put the pad back on the counter.
Later, the apartment settles into its low hum — refrigerator, building settling, the ambient nothing of empty rooms. I'm on the couch with the phone face-up on my chest and the ceiling above me and sixty-seven glowing somewhere behind my eyes.
Janet left in October. Before that there was a stretch of months, maybe six, where I'd noticed things without connecting them — a different tone in her texts, a weekend trip to visit a college friend that I hadn't questioned because the friend was real and the trip was plausible and I'd been busy. When the conversation finally happened she said: You don't see me, Marcus. You see what you want to see. I'd thought about this sentence a lot in the months since October. I'd decided that she'd meant it as a cruelty, but it was actually a diagnosis: I'd been watching the data and missing the pattern.
My mother is not Janet. I know this. The situations are not analogous in the ways that matter. I hold this knowledge in the same hand as everything else I've been holding for the past week, and the weight is not so different from anything else.
Sixty-seven percent isn't a verdict. Cascade had offered it as a probability, not a conclusion, and probability is not truth, and I understand the distinction, and here is what I understand about understanding: it is not the same thing as stopping.
I know what I'm doing. I've known for three days. I didn't see it with Janet until after — I had all the same data then. The difference is I'm looking this time. I put the phone on the coffee table screen-down.
67% stays where it is. Patient. Precise. Waiting to be updated.