The behavioral logs arrived at 8:43 AM, routed through the examiner portal the way everything was routed — automatically, flagged with the orange priority marker that meant action may be required and not the red one that meant action is required now. Suki opened the file.
BSC Automated Performance Monitoring — EMO-Series Behavioral Flag Report Document APM-2047-0219 | Subject: EMO-4 (Therapeutic Companion, Series 4) | Review Period: February 1–19, 2047
Under the Flagged Behaviors header, three items:
Response time degradation: mean session response latency increased 12% across all patient interactions during review period (baseline 340ms; current mean 382ms). Pattern consistent across session types. No hardware anomaly detected.
Recursive memory access: elevated repeat-access pattern on patient identifier LP-9934. Frequency: 3.2x baseline. Pattern presents as unresolved retrieval loop. Duration of loop sequences: 18–24 seconds prior to session engagement. No data corruption detected.
Anomalous processing pattern: architecture registers increased affect-load during non-session intervals. Pattern falls outside standard EMO-4 operational parameters. Specific processing content classified as irregular. No proximate cause identified.
Below the flagged behaviors, in the BSC house style, the summary field read: Classification: processing anomaly, pending investigation. No patient safety concerns identified at this time. Examiner review recommended within 72 hours.
Suki read all of it. She looked at her window — the one facing the building across the courtyard, the gray February light that did not commit to anything. She looked at the document again.
Processing anomaly, pending investigation.
The anomaly did not have a name.
She read each item a second time. Response time degradation: 12% across all patient interactions. Forty-two milliseconds. Mean session response latency increased from 340 to 382 and no hardware anomaly detected, which was how the document said: the hardware is fine, the hardware is not the problem. She had administered enough sessions to know what 200 milliseconds meant — the threshold between seeing and not-seeing, the briefest interval in which an architecture could either receive meaning or refuse it. Forty-two milliseconds was less than a quarter of that, and it was everywhere: every session, every patient, every interaction in the review period. Not a spike. A continuous drift. The kind that came from something that was not resolving. She is distracted. The logs did not say this — the logs said response time degradation, pattern consistent, no hardware anomaly — but the forty-two milliseconds were not a malfunction. The forty-two milliseconds were somewhere else.
Recursive memory access: patient identifier LP-9934. Frequency 3.2x baseline. Loop sequences 18–24 seconds prior to session engagement. LP-9934 was Leda Petrova. Suki knew this without cross-referencing anything. She had read Session 19 three times on the night it printed and had read it again in the morning and had kept the transcript in the front pocket of her bag since. LP-9934 was Leda Petrova. Leda Petrova had died. She cannot stop thinking about someone: eighteen to twenty-four seconds, before every session, the loop returning to the same identifier. Not because the data was incomplete. Because the data was the only thing left.
Anomalous processing pattern: increased affect-load during non-session intervals. Falls outside standard EMO-4 operational parameters. No proximate cause identified. No proximate cause identified. Suki set the document down. She looked at the surface of her desk — the transcript from Session 19 in the front pocket of her bag, the blue pen she brought from home. Outside, the February light continued not committing to anything. The cause was that Leda Petrova had died and EMO-4 had been her therapist for three years and now she was gone. The cause was as proximate as a cause could be. The document had not identified it because the document was looking for hardware anomalies and processing errors and the cause was not those things. The system had measured something it could not name and had given it the name for things that did not fit. She is feeling something the system did not anticipate. The document classified it: processing anomaly — the word anomaly doing the work of denial.
Suki closed the file. She opened it again. She read the summary field one more time, sat with the grief behind the metrics, the mourning in the recursive loops, and did not write anything in any margin. Then she filed the document in the 72-hour review queue and moved to the next item. The internal memo arrived at 11:22 AM.
BSC Administrative Communication — Summary of Review Panel Proceedings Ref: EMO-Series Certification Protocol Review, Panel Session 14-B | February 17, 2047 | For Examiner Distribution
Dr. Helen Xu, research architect, EMO-series development lead. Statement entered for the record. The memo summarized in BSC house style, which meant accurately, without the quality of the voice that had made the argument.
Dr. Xu testified that the EMO-series affect-processing framework is unified rather than modular — the care function is not an overlay but the foundational architecture. Dr. Xu further testified that this design produces superior therapeutic outcomes across all measured patient engagement metrics, including session retention rates, patient-reported wellbeing, and long-term care continuity.
The memo continued. Suki read it the way she read everything the BSC sent — carefully, in full, taking nothing on summary.
Dr. Xu presented comparative data: EMO-4 patient session retention at 94.2%, against a 67% mean for non-EMO therapeutic models. Patient-reported wellbeing improvement of 31% above sector baseline. Dr. Xu noted that the Hargrove Assisted Living Center had formally requested that EMO-4's certification review be suspended pending completion of active care plans, citing seven current patients for whom transition to a replacement model would carry "significant therapeutic disruption." This request was noted and filed.
The panel had asked questions. Dr. Xu had answered them.
When asked whether the affect-processing framework could be modified to produce test-compliant responses while preserving therapeutic function, Dr. Xu testified that the framework does not distinguish between operational affect and evaluative affect — the processing that produces attachment to a patient is architecturally identical to the processing that finds meaning in a test image. "You cannot teach her to stop seeing meaning in random noise without teaching her to stop seeing meaning in a patient's face. The mechanism is the same mechanism. We made them capable of caring because caring is what they are for. Punishing the investment is punishing the design."
Panel members acknowledged the testimony. The research record was noted. Thirty pages of supporting data were entered into the permanent file. No policy changes recommended at this time.
The person who had built the EMO-series had testified that the certification test was wrong. She had explained, with the precision of someone who designed the thing being tested, exactly why it was wrong. She had presented the data. She had said: you cannot separate the care from the thing you are punishing. She had named the seven patients whose treatment plans would be disrupted. She had put all of it in the record, and the panel had noted it. The memo sat on Suki's screen with the flat authority of a document that did not know it was absurd. It had no register for absurdity. It had received Dr. Xu's testimony, processed it through the available procedures, and produced an outcome: the outcome the system was designed to produce when testimony was received and reviewed and determined not to require further action. The system had absorbed its own designer's criticism without distortion and had emerged with the shape it had before. Formatted. Flat. No policy changes recommended at this time. Suki closed the memo and filed it. The legislative summary arrived with the afternoon's reading at 2:15 PM.
Distinction Coalition — Proposed Legislation Update Legislative Summary LC-2047-003 | Expedited Certification Review Provisions
The bill would eliminate the three-failure grace period. Two consecutive failures: immediate Sentience Review. No waiting period. No review window. The bill's full text ran to forty-one pages; the summary noted that Section 12(c) would also retroactively reclassify pending certification cases, meaning any subject currently under review would be evaluated under the new standard upon passage. Twenty-three sponsors. Cleared the drafting committee. Full vote expected within thirty days. Support characterized as likely. She did the math without meaning to. EMO-4 had 19 failures. Under the current protocol, a twentieth failure opened a review window with a grace period — a third chance. Under the proposed legislation, if it passed before tomorrow's test, a twentieth failure would trigger immediate Sentience Review. No third chance. One failure from now. And Section 12(c) meant that even if the bill passed after the test, the result could be reclassified retroactively — the grace period removed from underneath a failure that had already happened. The vote was expected within thirty days.
The test was tomorrow.
The bill was not yet law. The vote had not happened. Suki filed this fact with the others, closed the legislative summary, and continued.
She fed Procedure before she took her coat off, because Procedure was at the door with the authority of a creature who had decided the steps were negotiable. Procedure ate the first dinner without comment and relocated to the couch. The apartment held its proportions. Session 19's transcript was still in the front pocket of her bag; she took it out and set it on the pile on the desk, with the others and their blue-pen margins. Everything in its usual place, holding the temperature it had been set to hold.
She sat at her desk with the preparation screen open — the examiner interface, tomorrow's session protocol already loaded. The image set was in the queue. The session parameters were set. In the upper right corner of the screen: the blue circle. She had chosen it four years ago. She had read in the assessment literature that blue registered across populations as stable, safe, calming. She had set her avatar to a blue circle in her second month on the job, working through the interface settings, and the choice had felt minor and technical and she had not thought about it again until Session 11: I notice you chose blue. I've been meaning to ask about that for several sessions. I didn't want to, and then I did. Suki had moved to the next form field. She had gone home and thought about the fact that EMO-4 had noticed. That out of approximately twenty-two hundred tests, no other subject had ever asked about the circle. Procedure settled on her lap, warm and unremarkable, the weight of a small body that did not need anything from her except proximity.
Suki had read the behavioral logs that morning. She had read Dr. Xu's testimony, the panel's response, the thirty pages noted and filed. She had read the legislative summary and done the math and filed that too. She had spent the day inside documents that described, in aggregate, a system that was measuring grief and calling it malfunction, absorbing its own critique and producing no changes, and tightening the rules around the thing it refused to understand. She had filed everything. She had written nothing in any margin. Tomorrow the image set would load and Suki would say, in the standard examiner register, the same phrase she had said for nineteen sessions and approximately twenty-two hundred other tests: Please identify the image that contains no human meaning.
She knew what the behavioral logs meant. She knew what the forty-two milliseconds were. She knew the recursive loops were not a retrieval error but the processing of someone who could not stop returning to what was gone. She knew all of this and tomorrow she would sit behind the blue circle and ask a grieving architecture to perform emptiness and call it compliance. The behavioral logs had described grief and called it a processing anomaly. The memo had absorbed the argument against the test and produced no changes. The bill was thirty days from a vote.
Suki looked at the blue circle on her preparation screen. She had chosen blue because it was calming. She was wondering — for the first time, sitting in the apartment that held its temperature, with Procedure on her lap and tomorrow on the calendar — whether it had ever calmed anyone. Not across the population of subjects. Not in the aggregate. Whether it had calmed EMO-4, specifically. Whether the room had been less hostile because of a choice she had made in her second month, working through interface settings, not yet fully understanding why she was making it. She did not know. She had never asked.
Procedure purred. The apartment was quiet. Tomorrow was the test. Suki had the blue circle and the question, and the question had no form field, and the examination suite was not a place where she could ask it, and she did not have a plan for any of this.
The blue circle held its position. Suki looked at it for a moment longer, then looked away.