She started with three columns.
Active: the guest in her left temporal recognizing something in the client's history, orienting toward it, the way you orient toward a sound in a room. Quiet-absent: nothing, the kind of silence that was absence. Quiet-present: something there, no signal, a texture she could not name.
She had built the taxonomy from four clients. Isabel, who had given her the active column. Aurelio and Marisol, who had divided the quiet ones. Eduardo, who had given her the third.
She did not show the notes to anyone. She wrote at the bottom of the page: do not show anyone until you have enough to be wrong.
The first full week: thirty-one sessions. Active eleven times, quiet-absent thirteen times, quiet-present seven times. She looked at the numbers. No clean pattern — the distribution could be noise, or real variation, or her categories bleeding into each other in ways she had not noticed.
Then she noticed something else.
The guest in her left temporal had been mostly quiet all week. Not absent — the texture was still there. But quieter than usual. She sat with that for a while. She was tracking the guest's responses to clients. The guest was also, apparently, aware of being tracked.
She wrote at the bottom of the page: observer effect possible.
She did not know what to do with this. If the guest went quieter when she paid close attention to it, her data was compromised. She was not measuring the guest's natural response to clients. She was measuring the guest's response to being measured.
Or — and she held this possibility carefully — the guest going quiet while she was tracking was itself a response. Not to the clients. To her. The guest knew she was paying attention in a particular direction and had something to say about that.
She did not know which interpretation was correct. She was not sure yet that she would know in two more weeks either.
She kept taking notes. The uncertainty was data. The data was incomplete. She had learned by now that incomplete data was not the same as no data, and that no data was not the same as nothing there.