The prep bag is by the door.
Magda Prusak packed it last Sunday — gown, socks, the subscriber card with its pale blue gradient and her patient-portal ID, the contraindication list (two items, unchanged since 2019), her pharmacist's name in her own handwriting because she stopped trusting the Intake agent to autofill verification fields after the August sync lag that sent her to the wrong building once already. A book she will not read. The zipper has been checked twice since Sunday: once at 8 AM, once in the early afternoon when she walked past it and her hand went to the tab before she had made any decision to reach.
The bag will stay there until 9:15 AM Tuesday.
Her household agent has been running for three months. Her sister set it up during the July visit that also prompted the imaging referral — the timing is not ironic, her sister is thorough in the way people in insurance verification are thorough. The agent handles grocery restocking, utility renewals, the quarterly scheduling of preventive appointments. It negotiates against her stated preferences and a monthly budget envelope she resets at the start of each month.
What she has noticed: the agent gap runs the wrong direction. She is in the better half. Her agent price-routes her prescriptions across six pharmacies on a cost-and-wait composite score. Her pharmacist of eleven years — three blocks, family-owned, the woman knows which brand of ibuprofen she prefers — receives her scripts now only when the composite ties. This happens less often than she expected. The efficiency is not in question. Something adjacent to it is, something she does not have the right word for yet, a friction she keeps locating in the space between accurate and known.
Her Intake agent belongs to UPMC. It is not the same agent.
The first question: What has changed?
She has been carrying it since July, when CAS-monitor flagged two consecutive weeks of minor variation and then returned without explanation to THRESHOLD-STABLE just before her doctor was ready to escalate. She asked at the six-week check: what does the variation mean? Her doctor said: probably nothing, but we will image it to be sure.
Probably nothing is not nothing. Nothing is not something. Something is not the accumulation she has been sitting with for eleven weeks, which is not fear — fear has an object, and what she has does not yet have an object. It has a shape: peripheral, patient, waiting. She recognizes it as the thing that forms before objects arrive.
CAS-monitor has shown THRESHOLD-STABLE every morning for eleven weeks. She knows what the status means: the system has not flagged an alert condition. She knows what that does and does not mean, because she read the protocol documentation when she was first enrolled, which most monitoring patients do not do. The status is a statement about the monitoring system's thresholds, not a statement about her health. She holds these as two distinct claims.
The second question: What do you need from me?
She wrote this on the train two Fridays ago. She had been reading a trial summary on her phone — a patient in a chronic monitoring protocol who received a clear imaging result and was then recommended three behavior modifications she had already made on her own initiative three months earlier. The clinical record had not caught up. The agent-generated recommendation was running behind her life. Magda does not want that. She wants to know what her data, specifically, requires of her, specifically. Partial is fine. Generic is not.
Two questions. Notebook, her handwriting, the date.
At 4:00 PM, CAS-monitor sent its afternoon calibration: THRESHOLD-STABLE.
She checks the log the way she checks it every day — one look, closed. The monitor is doing its job until 10:00 AM Tuesday. She has confirmed the appointment twice through the UPMC Scheduling agent. Both times it responded with an identical confirmation block and a link to the prep checklist she had already completed through a different UPMC portal that the Scheduling agent does not appear to be able to read. She flagged this to the feedback form in August. The form acknowledged receipt. She has heard nothing. She is not surprised. Two separate agentic systems sharing patient state is still, in October 2026, harder than it looks.
At 3:07 PM, UPMC Intake sent a pre-imaging form. She did not see it until 3:40.
Twelve questions. Standard pre-MRI: six. Monitoring-protocol extension: six more, generated from her CAS-monitor data through the A2A pipeline that standardized in Q3 2026. The extended protocol applies to any monitoring patient within ninety days of imaging.
Questions one through six: standard. Subscriber ID, contraindications, emergency contact (her sister, in Braddock, same number since 2017). The Intake agent generates from her current monitoring record, not her form history. She has answered these in July, August, twice in September. The agent does not know this.
Questions seven through ten: CAS-monitor protocol. Sleep hours. Medication changes (none). Activity variation since July (she has been walking forty minutes evenings, started in August — her doctor mentioned it and she also wanted to, and the two reasons produced the same behavior but are not the same reason). She writes increased daily walking, self-initiated August, approximately forty minutes per evening in the free-text field. The self-initiated is deliberate. She is documenting that the clinical record is behind her life on this one.
Question ten: self-reported symptom changes, forty-three options. She reads them. The nearest to accurate is intermittent awareness of monitored area, which covers the right territory without reaching her specific instance of it. She selects it. A form is not a conversation. She does not try to make it one.
Question eleven: logistics. No changes.
Question twelve:
Based on your CAS-monitor readings over the past eleven weeks, how would you characterize your current baseline compared to your July referral?
A. Stable improvement — I believe my condition has improved since July. B. No significant change — My readings appear consistent with my July baseline. C. Possible decline — I have noticed changes that may indicate progression. D. Uncertain — I do not have enough information to assess my own baseline. E. Defer to clinical review — I prefer not to characterize my baseline before imaging.
She reads it once. Then again.
The Intake agent has her eleven weeks of monitoring data. It built this question from that data. Question twelve is asking her to confirm, contradict, or defer to what the agent already suspects.
She knows the shape of what she has been not-forming all day. At 11:30 AM in the break room it almost surfaced — what does it mean if the scan shows — and she let it dissolve before it finished. At 1:45 PM, eating lunch, it tried again. Dissolved again. She does not have the number yet. A question built on a predicted number is a prepared reaction, and she does not want to bring a prepared reaction into the room where the scan gives her the actual information.
The form is asking her to finish the sentence.
She sits with question twelve for three minutes. The timestamp in the corner advances from 3:44 to 3:47 while she holds the cursor.
Option D is honest in the way that honesty can be a category error. She is not uncertain. She has eleven weeks of readings. She is waiting. These are not the same posture.
Options A, B, and C require her to interpret her own monitoring data. She is a trained professional in an unrelated domain. Her interpretation of eleven weeks of a health monitoring protocol she has been enrolled in since July is a worse data source than a radiologist reading an MRI. This is not self-deprecation. It is an accurate description of the instruments available and which one is the better instrument for this specific question.
She moves the cursor to Option E.
Defer to clinical review — I prefer not to characterize my baseline before imaging.
She reads it twice. It is accurate. She is not declining because she fears what she might find. She is declining because she is not the right instrument. The scan is. She has been waiting eleven weeks for it.
Option E is not original to the form. It was added in the second revision, after the first three months of deployment showed a pattern of patients abandoning at question twelve without submitting. The design team reviewed the exit records. A subset were abandoning not from confusion but from a specific reluctance: the question was asking them to pre-characterize data they were about to have professionally characterized. Option E was added to give those patients a valid response. Magda does not know this history. She is making the same choice those patients made.
She selects E and submits.
The Intake agent responds in four seconds: Thank you. Your form has been received. A clinical review of your CAS-monitor data will be completed before your appointment. Please arrive at Building C, Radiology Intake, at 9:45 AM. Bring your subscriber card and photo ID.
CAS-monitor: THRESHOLD-STABLE at 3:47 PM. It does not know she just submitted a form. It only knows whether her readings are within the band it was configured to hold. It will show THRESHOLD-STABLE at midnight. It will show it again tomorrow morning.
She goes to the kitchen to make dinner.
The third question — fully formed — would be: If the reading is worse than July, does that mean the behavior changes I made were not enough, or does it mean I should have made them three years earlier?
She has known the shape of it since August. Once fully formed, it inserts itself between the reading is and whatever comes after. She does not want it in the room tomorrow when the technician is reading the scan. She does not want it to be the lens through which she hears the result.
Option E was on the form. It was there because a hundred patients before her made the same choice she is about to make, and the system eventually learned to accommodate that choice rather than log it as abandonment. She selected it. The agent processed it without comment.
The prep bag is by the door. The zipper is closed.
Tomorrow she will pick it up. She will walk to Building C. There will be a person at the Radiology Intake desk — the hospital decided, sometime in the last eighteen months, that scans warrant a human in the first room. She will put on the gown, sit in the room, and hold still.
She has two questions for tomorrow.
In the Errands world, CAS-monitor is a UPMC-affiliated monitoring application that reached mass adoption after the March 2026 agentic infrastructure buildout, when A2A and UCP protocol standardization made patient-wearable-to-hospital-scheduler pipelines cheap enough to include in standard health plans. The UPMC Intake agent was deployed Q3 2026. The extended pre-imaging form was approved October 2026. Option E (Defer to clinical review) was added in the second revision after exit data showed patients abandoning at question twelve — not from confusion, but from reluctance to pre-characterize data they were about to have professionally characterized. The clinical review queue receives a patient's full monitoring history two hours before the appointment. This has measurably reduced the number of clinicians in imaging who are surprised by what they find. Whether it has reduced the number of surprised patients is not something the form tracks.