The groceries come to three bags. Magda carries them from the bus stop three blocks to her apartment, and by the time she reaches the landing she has redistributed the weight twice to avoid bruising the tomatoes, which is the kind of thing she has been doing for years — accounting for fragility before she knows what is fragile.
The CAS-monitor ran its 4 PM cycle. Output: THRESHOLD-STABLE, confidence 0.87, variance delta 0.003 below the thirty-day rolling average. She read the full line before she got on the bus. She reads every cycle output when it comes in, regardless of what it will say, because not reading it does not make the data different, and the brief interval between the alert arriving and her reading it is worse than just reading it immediately. This is something she learned in the first month of the monitoring program and has not unlearned.
Eleven consecutive THRESHOLD-STABLE cycles. She has kept a tally in a notepad since the third one, when she decided she needed an external record to be sure the algorithm was not simply returning a cached output. It was not. Each cycle carries a timestamp, a confidence interval, a variance delta that the algorithm processes against a dormant-baseline profile calibrated at enrollment. The readings are different each time. The assessment is the same. Eleven times, THRESHOLD-STABLE.
What THRESHOLD-STABLE means precisely is a question for Adamski. She has learned not to try to interpret it herself. Her self-interpretations in the first six months had been largely wrong — she had read stable as safe, and then, later, as possibly-worse-hiding, and then, during the second series, as probably-fine-but-still-uncertain, which is closer to what THRESHOLD-STABLE actually indicates. The algorithm is measuring a relationship, not a condition. What it is measuring her against is a baseline it constructed from her own earlier data, which means the question of what THRESHOLD-STABLE means is always partly a question about who she was when the baseline was set.
She puts the groceries away.
Tomatoes on the counter rather than the refrigerator, because she read somewhere that refrigeration damages the cell structure, which is a preference she has held for years without having tested it scientifically. Yogurt in the back of the fridge, where the temperature is more consistent. Bread in the drawer. The rice she bought — a bag of the long-grain white she always buys from the same shelf at the same market, which has the same brand label it had three years ago but now carries a small secondary patch indicating the carbon-provenance tag for the specific growing region and harvest quarter — she puts this in the cabinet.
The tote bag with the notepad goes on the kitchen chair. She does not take the notepad out.
She has been in the UPMC Integrated Monitoring Program for fourteen months. She enrolled in the second cohort, eight months after the CAS-monitor deployment at Hillman, when Adamski's team had finished calibrating the system for her diagnostic profile. The enrollment process took three visits over two weeks. On the third visit, Adamski explained what the baseline calibration meant: not a target to reach, but a signature of Magda-as-of-enrollment that the system would use as its comparative reference. Everything the CAS-monitor would subsequently measure would be measured against that signature. The monitoring itself has required nothing from her except the weekly passive check-ins and the six-week readouts. The CAS-monitor runs on her home connection, collects readings from the four passive environmental sensors she agreed to install at enrollment — air-chemistry, acoustic pattern, ambient-light rhythm, thermal signature — and routes them to the algorithm, which produces the THRESHOLD-STABLE or THRESHOLD-ACTIVE assessment. THRESHOLD-ACTIVE would trigger a follow-up protocol.
She has not had a THRESHOLD-ACTIVE.
What the CAS-monitor cannot tell her, and what Adamski can, is whether eleven consecutive THRESHOLD-STABLE outputs represent a stable trend or a plateau in an active process. The algorithm is measuring against a baseline. Adamski will read against the full history: the biomarker profile, the MRI data from six weeks ago, the two comparative cohort datasets from the research archive that Magda signed releases for. The readout is not the algorithm's assessment. It is a conversation between Adamski and everything the algorithm has collected, and Magda will be present for the conversation but will not be the one having it.
She has come to understand this as a form of translation. She presents. Adamski reads. The meaning emerges from the interaction, not from either source alone. In the first few months she tried to do Adamski's reading herself, in advance, and arrived at each readout having already decided what the results meant. She stopped doing this at month four, after she was wrong twice in opposite directions.
The soup is simpler than last week's. Onion, broth, rice, black pepper. She found the pot three weeks ago at the back of the cabinet and has used it four times since, more than in the preceding year. She does not think this means anything. She sets it on the burner. Onion in with a little oil, medium heat, and she stands at the window while it softens.
September light at this hour comes in sideways, finding dust she did not know was on the counter, making the kitchen look temporarily inhabited in a way that is more comforting than the alternative. She has been in this apartment twelve years. She knows the light by month now, roughly — what it does in March, what it does in September, the specific quality of November that she does not have a word for but would recognize anywhere. The light is one of the things she measures without meaning to.
Twenty-seven hours.
She calculated this on the bus. Adamski's readout is Monday at 10 AM, which is thirty-nine hours from when she sat down on the bus with the grocery bags, or thirty-six from when the groceries are put away, which she has just done. She uses twenty-seven because it is the number she can hold most comfortably. Thirty-nine is unwieldy. Twenty-seven has a shape she can work around.
Her phone, face down on the counter, has not notified her. The UPMC patient portal does not update on Saturdays. Adamski does not push data on Saturdays. She knows this. She checked the portal this morning before the market, and once at the market while looking at bread, and once when she got home. She has put the phone face down because looking at a blank screen is a different category of action from not looking at the phone at all, and she is trying to do the second thing.
Rice goes in when the onion is translucent. She adds the broth from the carton, black pepper from the cabinet behind the paprika. Waits. The window is still open. Outside, the September street is doing its Saturday things: a dog, a car moving slowly through, someone with a stroller. She watches none of it in particular.
She has been thinking about the tally marks differently this week.
She started keeping them because she needed an external record — a way to verify the algorithm was running and not recycling a previous output. She has kept them because she has not decided to stop, which is a different reason. The marks are in a notepad rather than a spreadsheet because a notepad does not feel like data management. She knows this distinction is arbitrary. She maintains it anyway.
Eleven marks. Each one is a cycle date and not a verdict. She understands this now in a way she did not at three marks or seven. The CAS-monitor is measuring her against her own enrollment signature. She is not moving toward it or away from it — she is being measured against a portrait of herself at one specific moment, and the portrait is asking whether she has changed significantly in ways the algorithm can detect.
She has not changed significantly in ways the algorithm can detect. Eleven times.
What Adamski will tell her Monday is whether that means something, and if so what.
The soup is done by 5:45. She eats standing. This is something she does when the table has too many associations with decision-making, which is a state she does not quite feel but cannot entirely exclude on the evening before a number like twenty-seven. The soup is good. The rice absorbed more broth than expected. She adds water, which resolves it without complication.
She washes the pot at 6:10. Hangs it on the hook over the stove. This is where it has lived since she started making soup three weeks ago, a small reorganization she considers permanent.
She sits on the couch with the book she has been reading for three months. A novel about an archivist who has made a mistake. She has been reading it slowly, which is how she reads when she wants to remain inside something rather than finish it. She reads twelve pages. She understands eleven of them clearly. The twelfth she follows well enough. She does not go back.
The archivist does not know yet that she has made a mistake. The reader knows. The gap between what the character holds and what the page holds is the thing Magda has been thinking about all afternoon without the right name for it. The archivist is not wrong that she cannot know what she does not know. She is doing everything correctly given the information available to her. The problem is that the information her methods could not reach is the information that turns out to matter. Magda understands this structure from both sides: the side where you are the archivist, processing the available data correctly, and the side where you are waiting to find out whether what arrives on Monday is the kind of data that changes the picture or the kind that confirms it.
She reads three more pages.
At 8:47 PM her phone notifies her. Her sister, asking about Sunday plans Magda has already declined, checking whether the decline was firm or provisional. She replies: firm. The reply takes forty seconds.
She does not open the patient portal.
She turns off the lamp at 9 PM, which is not a rule but has become a practice in the fourteen months since enrollment. The lamp going off is an action that does not require a decision because the decision was already made earlier in the evening, when she registered what time it was and what she was doing. She is not afraid of the dark. She is occasionally afraid of the 10 PM check, the one she does automatically, the one she has been trying to replace with a deliberate action instead of a reflexive one. Turning off the lamp is a deliberate action. It helps.
In the dark, the apartment sounds different: the refrigerator's cycle, a neighbor's television two floors up, the street still moving outside the open window. She is not thinking about Monday. She is thinking about whether she left the kitchen window open, which she did, and which is fine because September is still warm enough.
The twelfth mark will be written Monday morning before she gets dressed. She decided this at mark seven, when she understood she needed a transition ritual — not to make the readout better or worse, but to mark the day as categorically different from other days in a way she had chosen. She will wake at 6:30. She will open the notepad. She will add the mark. She will close it.
Then she will get dressed.
Then she will go.
The window in the kitchen is open. The September light is gone but the air it let in is still there. Twenty-three hours, which is different from twenty-seven in the way that being in motion is different from being at the bus stop. She is already inside the twenty-three hours. They are already happening. The notepad is on the chair. The soup pot is on the hook. The phone is face down. The lamp is off.
She has decided this is enough.