The components were arranged on her side of the bench in the order of installation: spring cluster first, then the micro-calibration disc, then the pressure membrane last.

She had looked at this arrangement when she arrived that morning and understood, with the specific understanding that came from years of knowing when a thing was about to be finished, that today was the day. Not because the calendar said so — the calendar had said so for three days and the calendar had been wrong three times because the work required what the work required and not what the calendar demanded. But today the bench had the feeling of a bench on which the final things would happen, and she had learned to trust that feeling the way she trusted her measurements: not because it was infallible, but because ignoring it had never produced better results.

Raymond was at his station when she came in, already working, the compound housing open and the final calibration sequence running. He looked up when she entered through the side door — the look he always gave her, the reading look, the one that moved across before it settled — and said nothing, which was correct. There was nothing to say that the morning didn't already contain.

She put down her coat and tools and stood at her bench and looked at the three components.

The spring cluster was the most complex — forty-seven individual springs of varying tension specifications, pre-assembled into a single unit that had to seat into the machine's secondary housing at a precise angle and click into registration with a tolerance of less than half a millimeter. She had made it three times. The first two had been within specification but not within her own standard, a distinction that the machine would not have noticed and that she could not ignore. The third was correct.

The micro-calibration disc was simpler in concept — a single precision-machined piece of brass with seventeen registration marks cut to within a tenth of a millimeter — and had nearly defeated her twice through problems that had nothing to do with the machining and everything to do with the material's behavior at the specific temperature range the machine operated in. She had solved this by changing the alloy composition rather than the machining approach, which had required sourcing a new material from Lyon and waiting eleven days, which had been eleven of the more difficult days of the project.

The pressure membrane was the one she was proudest of. It was also, technically, the simplest — a flexible seal between the compound housing and the detection array, designed to transmit pressure variations without contaminating the compound's olfactory integrity. The first iteration had worked. The second had worked better. The third had required her to invent a fabrication technique she was fairly certain no one had used before because the technique required thinking about the problem as a acoustic problem rather than a mechanical one, and acoustic thinking was not standard in precision mechanical engineering, and she had only arrived at it by spending three days in the laboratory's silence listening to the machine breathe.

Raymond had not asked why she was sitting still in the laboratory for three days. He had brought her tea.

"The installation sequence," she said. Not a question — a statement that was the opening of a process they had discussed before but needed to speak through one more time before beginning, the way you confirmed a route you knew well before making a journey that mattered.

"Spring cluster first," he said, closing the compound housing and coming to his side of the bench. "While I hold the primary housing at operating pressure. Installation window is approximately four minutes before the pressure differential exceeds the cluster's pre-load tolerance."

"Four minutes is tight."

"Four minutes is the number." He looked at the spring cluster on the bench. "The disc goes in while the cluster is still settling — there's a forty-second window after the cluster seats before the housing geometry shifts. The membrane goes last, at ambient pressure, once the disc has registered."

"So the first six minutes are the critical window."

"The first six minutes are everything."

She looked at the components again. Then she looked at the machine — the machine they had been building since she had pushed her way through his front door with a proposal and an argument and no intention of leaving, the machine that had grown from a schematic and a theory and two people who had, somewhere in the process of building it, built something else as well.

"Ready," she said.

He picked up the primary housing and brought it to operating pressure with the manual pump — she could hear the characteristic sound of it, the specific tone that meant the pressure was correct — and nodded once.

She picked up the spring cluster.

The installation required her to work inside the machine's central cavity, which was approximately twenty centimeters deep and designed for a specific approach angle that meant she had to lean over the bench with her right arm fully extended. This put her at a proximity to Raymond — who was holding the housing steady with both hands on the far side of the machine — that the bench had not previously arranged. Not uncomfortable. Not remarkable. Simply the geometry of the task.

Except that it was also the geometry of two people who had, for three months, maintained an invisible eighteen-inch border and were now on the same side of it for the first time because the work required it.

She did not acknowledge this. She focused on the spring cluster.

The seat for the cluster was at the bottom of the cavity, accessible only by feel — the approach angle didn't allow direct visual confirmation. She had practiced this installation blind, in a mock-up, until her fingers knew the registration marks the way her hands knew the calibration of her personal tools. She found the primary registration mark, oriented the cluster, began the seating sequence.

Five seconds. The first spring engaged. She felt the characteristic click through her fingertips.

Ten seconds. The cluster was tracking correctly. She could feel the pre-load building under her fingers, the resistance increasing in the specific pattern that meant the geometry was right.

Raymond's right hand shifted slightly — she felt the change in the machine's resistance, the minor adjustment he was making to the housing pressure in response to the changing load distribution as the springs engaged. He was reading the machine through his hands the way she was reading it through hers. They had been doing this long enough that the machine had become a communication medium between them — a shared instrument that both of them played.

Fifty seconds. The cluster was half-seated. Her fingers were in the machine and his hands were on the housing and the distance between their working positions was a matter of centimeters and she was thinking about none of this because she was thinking about the spring cluster, which was exactly as it should be.

At ninety seconds, the cluster seated.

The click was distinct and certain and she felt it travel up through her arm and she said, without looking up: "Cluster seated. Disc window."

"Confirmed," he said, and she heard in the word the specific quality of his technical confirmation, the voice he used when the machine was doing what the machine was supposed to do.

The micro-calibration disc was a two-handed installation — she needed both hands in the cavity simultaneously, at angles that required her to be even further over the bench than the cluster installation had required. This brought her shoulder into the space over the machine, which brought her cheek close enough to the apparatus that she could feel the warmth of the compound housing's operating temperature.

She found the disc registration marks by feel. Aligned the primary. Aligned the secondary. Began the seating.

Raymond moved his left hand from the housing to support her right elbow — not gripping, a flat support, steady and impersonal as a ledge. The installation angle required her elbow to maintain a position that the forearm's natural weight would have compromised over the fifteen seconds the disc seating required. He had read this from the geometry of her position without being asked.

She did not acknowledge it. She seated the disc.

The click of registration was softer than the cluster — a delicate, high-pitched sound that meant the seventeen registration marks had engaged simultaneously. She withdrew her hands from the cavity and straightened.

"Disc registered," she said.

"Membrane," he said.

The membrane installation was the simplest of the three — at ambient pressure, no time window, conventional approach angle. She took the membrane from its protective sleeve, aligned it with the junction between the compound housing and the detection array, and seated it with the specialized tool she had made for the purpose. The seal engaged. She ran the verification pass with the pressure gauge. Nominal. She ran it again. Nominal.

She set down the tool.

The machine was complete.

She straightened and looked at it. It looked the way it always looked — complex and purposeful and exactly what it was supposed to be, the physical expression of everything they had understood and solved and argued about and built over the past months. The same machine. Also something new, as of this moment. A thing that had been in the process of becoming and was now, definitively, what it had been becoming.

Raymond was on the other side of the machine. He had set down the housing tools and was looking at the machine with the expression he wore when he was reading something that required full attention. She had seen this expression on his face when he was reading compound formulas, when he was reading her phantom smoke responses in the calibration sessions, when he was reading the room at the Valois masquerade. It was his most complete expression — the one in which the analytical and the other thing that lived alongside it were not distinguishable from each other.

The laboratory was very quiet.

She had completed machines before. She had completed the first escapement mechanism she built at twelve, the one Auguste had supervised, standing in the cold workshop in the morning and holding the finished thing and feeling the specific quality of a thing made coming into contact with the specific quality of a made thing. She had completed the first precision governor she built herself, at fifteen, standing in the same workshop without Auguste because Auguste's hands were gone by then, and feeling the completion differently — heavier, more solitary.

This was not like either of those.

She placed her hand on the machine. The metal was warm from the installation process, the ambient heat of a thing that had just been assembled from components that had been held and worked and calibrated by two pairs of hands over a long season of mornings and evenings and nights.

"It's done," she said.

Raymond placed his hand on the machine beside hers. Not on her hand — beside it. One centimeter of bench between his fingers and hers, the same one centimeter that had been the distance between them all morning, the distance that was not accidental and not arbitrary but was the specific distance that two people maintained when they were in the process of deciding something.

"Not yet," he said.

She looked up. He was looking at her, not the machine. His hand was beside hers on the completed machine and his expression was the complete one and the distance between their hands was one centimeter and outside the window Florence was doing what Florence always did which was continue to exist with complete indifference to the specific quality of what was happening in any given laboratory in any given morning.

She looked at him. The machine was warm under her hand. The spring cluster had clicked into place and the disc had registered and the membrane had sealed and everything was exactly what it was supposed to be.

"Not yet," she said, and understood what she was agreeing to.