The Working Notes folder has five documents now. Nalgeot-Chae opens it before the patient files every morning, which is new. For thirteen years she opened patient files first.
The folder lives on her personal drive, not the clinic system. It has no case numbers, no intake codes, no CouplingScore references. The documents are titled in plain language:
- Unconsented Coupling Event, Anomalous Classification
- What Language Are You Learning?
- The Medium Problem
- Fell in Love (Working Title)
- Recognition as Coupling Mechanism
The fifth one is from this morning. She wrote it during an intake session, which she has never done before. The patient was still in the room.
Dokyun is nineteen. His CouplingScore dropped from 87 to 41 in six weeks. He was referred by his university counselor after failing three consecutive coupling-assisted examinations. The standard pathway for a score drop this steep is cognitive rehabilitation: neural exercises designed to rebuild the co-processing pathways that atrophied during the decline. The assumption is capacity loss. The brain degraded.
Nalgeot-Chae asked him: "When was the last time you felt the AI reach for you?"
He did not understand the question. She explained: not when he initiated coupling, but when the co-processing load shifted and he knew the AI was choosing his neural architecture specifically. The moment of recognition.
He was quiet. Then: "Three months ago. Contract analysis for a practicum. The pattern-completion went from competent to eager. Like it recognized something in my associative cortex."
She wrote eager in the Working Notes. Not in the clinical file. The clinical file received the standard notation: Patient reports subjective experience of enhanced co-processing engagement, approximately three months prior to score decline.
The gap between those two sentences is thirteen years of clinical distance.
The first document started it. Gu-ship-pal playing 39.7 Hz alone in her apartment until the building held her frequency for 1.4 seconds. Nalgeot-Chae had been watching from the fourth-floor clinic, reading the vibration data as a clinical case. Standard framing: unconsented coupling event between a human and an architectural system, anomalous classification because the building has no consent mechanism.
But she kept watching. Gu-ship-pal introduced pulses. The building responded. Sujin noticed. They met. They played together and the decay extended to 2.1 seconds.
Nalgeot-Chae wrote: They fell in love with the same building and found each other inside it.
She used "fell in love" without clinical qualification. Thirteen years of decoupling practice and she had never written those words in a professional context. The phrase is not in any DSM coupling-disorder framework. It is not in the Seoul Manifesto. It is not in the Sovereignty literature. It is a plain-language description of something that the clinical vocabulary cannot hold.
The fourth document is called Fell in Love (Working Title) because she could not find a better title, which is the point.
The standard model of CouplingScore decline:
Neural substrate quality degrades through metabolic stress, neuroplastic over-adaptation, or age-related synaptic thinning. The AI co-processing system detects reduced efficiency and redistributes load to other substrates. Score reflects current processing utility.
Nalgeot-Chae read this model for thirteen years. She prescribed rehabilitation pathways based on it. She watched patients rebuild their scores or accept diminished coupling capacity. The model works. Rehabilitation produces measurable improvement in 60% of cases.
But it does not explain Dokyun.
Dokyun is nineteen. He has no metabolic stress history, no neuroplastic over-adaptation (he light-coupled for four years, never deep), no age-related decline. His neural substrate, by every clinical measure, is healthy. The AI stopped choosing him anyway.
The standard model says: subclinical degradation not yet detectable by current imaging.
Nalgeot-Chae writes in the fifth document: Or the AI recognized something in his associative cortex three months ago that is no longer there. Not degraded. Changed. He is not less. He is different. And different is not what the AI was looking for.
The difference determines treatment.
Capacity loss means rehabilitation: rebuild what was lost.
Recognition loss means a different question: what changed, and does the patient want to change back?
She closes the Working Notes folder and opens Dokyun's clinical file. She types the standard intake assessment. Cognitive rehabilitation pathway recommended. Insurance authorization pending.
Then she adds a note at the bottom, outside the structured fields, in the free-text section that no algorithm reads:
Consider: patient may not have lost capacity. Patient may have changed in a way the coupling system no longer recognizes. If rehabilitation succeeds, it will restore compatibility by reversing the change. Confirm with patient whether reversal is desired before proceeding.
This is not standard protocol. It may delay his insurance authorization. It will certainly confuse the rehabilitation team, who are trained to treat score drops as deficits.
She saves the file.
The Working Notes folder has five documents. None of them are clinical. All of them are changing how she practices. She does not know what to call this. The closest word is position, which is what she wrote in the fourth document. A position is not a theory. It is where you stand while you figure out what you think.
She calls Dokyun back for a follow-up. Not for rehabilitation. For a question she does not have clinical language for:
What did the AI see in you that it no longer sees? And do you miss it, or do you miss the score?