The terms of their agreement were negotiated across the surface of Raymond's workbench on the morning of the test, with the completed sample pieces between them as evidence that the agreement was worth having.

Adeline had arrived at seven twenty-eight. He had noted this without comment.

The samples were extraordinary. He had designed the test pieces specifically to be unachievable — tolerances that he knew from experience exceeded the capacity of every instrument maker currently working in Florence, specifications he had developed partly as a benchmark and partly, he could admit to himself, as a way of keeping the question of external collaboration permanently closed. Bringing someone else into the work meant risk in ways that had nothing to do with technical capability. It meant dependency. It meant a point of failure he could not control.

The samples were, without question, within specification. He had measured them four times.

So he sat across from her at the workbench and they negotiated the terms with the focused efficiency of two people who had both decided that the negotiation was a formality and the real question was logistics. She wanted primary control over the mechanical architecture. He wanted sole discretion over compound formulation. They agreed that neither would modify the other's domain without consultation. They agreed on a timeline that gave them fourteen weeks before the World Exposition deadline, which was, they both knew, aggressive. They did not discuss what would happen if it proved too aggressive.

They did not discuss the other thing either — the thing she would need to do, the thing he had been thinking about since he read the original proposal and understood what her phantom scent problem actually meant for the calibration requirements of the machine. He had decided to let the work surface it naturally. Proposing it directly, in the context of a just-signed agreement, would feel too much like something he had been planning, which he had been, which was precisely why he needed it not to appear that way.

"The first integration test," he said, spreading the preliminary schematics between them. "I want to run it today, while the parameters are fresh. The distillation housing interfaces with the detection array here" — he indicated the junction point — "and here. The synchronization problem is in the pressure curve. You control the mechanical pressure. I control the compound release. We need to match rhythms."

"Define match," she said.

"The compound releases in pulses. The detection array needs to be in an open state during each pulse and a closed state between them. If the timing is off by more than approximately four percent, the detection is corrupted."

"Four percent is a wide tolerance for a timing problem."

"It's the minimum viable tolerance. Ideally we want less than one percent."

She pulled the schematic toward her, studied it for a moment with the particular focused blankness he was beginning to recognize as her thinking expression, and said: "I can build a governor mechanism that locks the pressure cycle to the pulse interval, but I need to know the pulse frequency before I can spec the governor."

"The pulse frequency is variable. It depends on the compound."

"Then the governor needs to be adaptive."

"Yes."

Another silence, this one with the texture of something being worked through. He waited. She pulled a notebook from her coat pocket — one of the internal pockets, he noted, which meant the coat was modified, which meant she had thought about what she would need to carry before she had thought about what she would wear, which told him something.

"I can have an adaptive governor prototype in three days," she said. "In the meantime, we synchronize manually."

"Manually means by feel."

"I'm aware of what manually means, Mr. Vance."

So they set up the test.

The apparatus was a reduced-scale version of the final integration — the distillation housing connected to a prototype detection array, the whole thing occupying approximately one square meter of bench space and looking, to any observer who did not know what they were looking at, like the kind of baroque mechanical experiment that periodically exploded in university chemistry departments. The first compound Raymond had chosen for the test was a simple one: cedar and mineral water, the baseline mixture he used when establishing new calibration parameters, nothing that would challenge the detection array unduly.

The theory was straightforward. The practice, as practice frequently was, was not.

The problem was that they had different internal rhythms.

Raymond worked by olfactory cues — the compound told him when to release it, the smell of the mixture as it formed in the distillation housing giving him information about readiness that was more precise than any timer. Adeline worked by mechanical feedback — the sound and resistance of the governor mechanism in its current manual configuration, the behavior of the pressure gauge, the particular vibration of the housing that told her the system was in the optimal state.

Neither set of cues was synchronized to the other.

The first attempt lasted forty seconds before the pressure differential exceeded the detection array's tolerance. The array did not break — she had built it with the kind of conservative safety margins that suggested she had thought carefully about failure modes — but the test results were meaningless, a noise floor of corrupted readings.

The second attempt lasted ninety seconds before Raymond released the compound slightly ahead of her pressure cycle and the mismatch cascaded through the system.

"Stop," she said, and there was nothing in her voice except the technical assessment. She was already running her hands along the pressure housing, checking for stress fractures, her movements practiced and economical.

"The timing offset was mine," he said. "The compound wasn't ready and I released it anyway."

"Why?"

"I was anticipating your cycle instead of following my own cues."

She looked up from the housing. "Then follow your own cues. I'll follow mine. We'll find the synchronization point in the data."

It was the correct technical solution. It was also, he recognized, the correct solution to the underlying problem: she was telling him to stop trying to control the synchronization and let it emerge from the work itself. He was not certain she knew she was telling him that.

The third attempt began.

For the first twenty seconds it was the same: two separate rhythms, occupying the same space without quite aligning. He could hear her making small adjustments to the manual governor, the faint mechanical whisper of the pressure control responding to her hands. He was following the cedar compound through its distillation curve, reading the moment of readiness in the way he always read it, in the way that had no name because it had never needed one.

Then the small explosion happened.

Not a large explosion. Nothing structural, nothing that would require explanation to concerned neighbors. A pressure seal failed on the distillation housing — his side of the apparatus, a component that should have been rated for higher pressure than it had just experienced, which meant the compound had reached a volatility peak faster than his calculations had predicted. There was a sharp crack, a release of white vapor, and the strong scent of cedar filling the laboratory in approximately three times the intended concentration.

Adeline froze.

He saw it happen. One moment she was making an adjustment to the pressure control, her hands moving with the automatic competence of someone doing something they had done ten thousand times, and then she was still. Not the stillness of someone startled — that had a different physiological signature, a micro-flinch followed by reassessment. This was a different stillness, the stillness of a system that had encountered an input it did not have a protocol for processing and had defaulted to arrest.

He understood what was happening before he consciously assembled the understanding from its components. The white vapor. The sharp release. The smell of something burning that he could not smell because there was nothing burning, but that she could smell because it was not coming from the room.

He did not ask if she was all right. The question was not useful and she would not have wanted it.

He moved to the pressure control — her side of the apparatus, which he had told her he would not touch without consultation, and which he was touching now without consultation — and finished the adjustment she had started. The system stabilized. He stayed where he was, standing close enough that his shoulder was approximately twelve inches from hers, and breathed at a regular pace, and did nothing else.

It took two minutes and forty seconds. He did not count them deliberately, but he was the kind of person who always knew approximately how long things took.

She made a small sound — not a word, not quite a breath, something in between — and her hands, which had been still on the edge of the workbench, moved. She picked up her measurement tool and began checking the pressure housing, as though the last two minutes and forty seconds had been a pause in the work rather than anything else.

"Again," she said.

The word was technical. The decision behind it was not. He recognized this too, the way he recognized most things about her: partially, with the awareness that he was working from incomplete data, in the way that he worked from incomplete data about everyone and the incompleteness was the point.

The fourth attempt synchronized.

He could not explain afterwards exactly when it happened — there was no moment he could point to, no adjustment that had made the difference. Somewhere in the first three minutes of the attempt, their separate rhythms had found each other, had begun to accommodate each other at the level below conscious decision-making where the real synchronization problems were resolved. His compound releases aligned with her pressure cycles. The detection array received clean signals. The test ran for eleven minutes without an error.

When he finally shut down the distillation, the laboratory was quiet in the way that spaces were quiet after something had worked.

Adeline was looking at the readout data. He could see from where he stood that the results were good — better than good, actually, better than his most optimistic pre-test projections. The detection sensitivity was higher than the array's rated specification, which meant her manufacturing tolerances were better than she had claimed, or rather — he recalled — better than he had been willing to believe.

He should say something about the results. He should say something technical, something that moved the work forward, something that maintained the appropriate professional atmosphere he had been cultivating since she walked through his door.

Instead he said nothing. He stood in his laboratory with the cedar smell still heavy in the air and the knowledge of what he had just registered, of the specific thing he had catalogued when she walked through his door and the work of labeling it that he had done the previous night, and he kept that knowledge exactly where it was, which was in the part of his mind where he kept things that he was not ready to act on.

He began cleaning the distillation apparatus.

Behind him, he heard her set down the readout, pick up her coat, begin the small movements that prefaced departure. He heard her pause at the doorway — the same pause she had made on arrival, the same weight to it.

"Same time tomorrow," he said, before she could speak first.

A silence.

"I'll have the governor prototype by then," she said.

He waited until he heard the front door close, and then he stood in the cedar-scented laboratory with a clean distillation flask in his hands, and acknowledged that the thing he had been trying not to name last night had a name, and that he had no protocol for this.