She came back to the lab at four in the morning with paint on her elbow and a frequency she was afraid to lose.
The overnight crew had gone. The monitoring systems hummed in their racks, the centrifuges running their slow patient cycles. None of it required her. She crossed to her workstation without turning on the overhead lights and pulled up the nanite simulation. The frequency was still inside her chest from the orb, the one Zavo hadn't moved from the low table even when they left. As if it understood where it belonged.
She could not replicate what the orb did. Not without understanding it. But she did not need to replicate it yet. She needed to characterize it.
The sensor rig data she had pulled at his flat was raw: forty minutes of readings with a calibration offset she had not had time to correct on site. She ran the correction now and the waveform came up clean. Smooth. Regular. Not random at all.
She pulled the nanite simulation alongside it and set the vibration settings to match. First pass: nothing. Same jagged peaks. Same red lines. She adjusted the frequency by four decimal points and ran it again. The peaks softened marginally. She noted the value and adjusted again.
This was how her work had always gone. Not a single moment of revelation but iterations accumulating until the evidence became undeniable. She had learned early not to romanticize it. The work was the work.
By six she had narrowed the oscillation range to a twelve-point window. By eight she had it to four. She was still running passes when she hit the edge of the range and the stabilization stopped improving, and she understood that she had the shape of the frequency but not its source. She could describe it perfectly. She could not reproduce it without the orb itself in the room.
She spent the next three days trying to prove herself wrong about this. She tried every synthesis approach in the lab's capacity: signal generators, resonance amplifiers, a pressure-sensitive array she borrowed from the geophysics shelf on the theory that it might approximate the orb's output at the right matching frequency. None of it worked. The organisms stabilized briefly and then drifted, the coherence lasting seconds instead of the hours she needed.
She needed the orb.
She messaged Zavo on the sixth evening and said so.
He was at her lab door by the following morning with the orb in a cloth bag and no questions asked. He set it on the bench beside the sensor rig and watched her recalibrate without speaking. She appreciated this about him: he understood when the work needed to happen and that his commentary was not part of the work.
The first test with the orb present ran for four hours. The organisms held coherence the entire time. The vibration settings she had characterized over the previous week matched the real-time output exactly. She had the signature. Now she needed to broadcast it.
"I need to build an emitter," she said, not looking up from the readout. "Something that can broadcast this frequency at a range of at least ten meters. Potentially much more if we want atmospheric delivery."
"What kind of emitter?"
She told him. He listened. He asked three precise technical questions, the kind that told her he had a working understanding of signal architecture rather than a polite interest in it. She answered all three. He nodded, already thinking.
"There's a component fabricator in the lower ring," he said. "Level 4 industrial. They build custom signal hardware for the mining consortiums. I know someone there."
"I don't want to explain this to someone I don't know."
"You won't have to." He picked up the orb and put it back in its bag. "I'll get you the specs. You build it."
He was back in four days with component diagrams and a bag of parts. She did not ask how he had explained the request or what he had paid for it.
She started the build the same evening. The component architecture was straightforward: signal emitter housing, frequency tuner, broadcast antenna array. She had built more complicated equipment from worse components in her first year at Flora. The emitter took shape quickly. Too quickly, she found out on day six.
The frequency was right. She had characterized it well enough that the signal generator matched the orb's output to within point-four percent across the primary band. The problem was an interference pattern in the secondary signal, a narrow artifact of the frequency tuner's architecture that the orb's output did not have. In isolation it looked like nothing. Against a live suspension of Florabyte organisms, it interfered with the stabilization effect. Not immediately: the organisms would hold for forty minutes, sometimes an hour, and then the coherence would drift exactly as it had in the synthesis attempts before the orb arrived.
She ran the problem for three days. She mapped every variable she could control. She rebuilt the frequency tuner from a different component architecture on day eight and got the same artifact at a different frequency, which told her the problem was not the specific parts but something structural in the emitter design. A fundamental interference that she could move but not eliminate.
She was sitting with this at eleven at night when Zavo arrived.
He came in quietly, the way he moved through spaces he was uncertain about. He had the orb in its cloth bag. He set it on the bench beside the emitter, which was where it always went, and looked at what she was working on.
"The interference pattern is in the frequency tuner," she said, not looking up. "I can't get rid of it."
He was quiet for a moment. Then: "What does it do."
"It mimics a stress frequency. Not exactly, but close enough that the organisms read it as a threat signal and shift their oscillation pattern to compensate. After long enough, the stabilization fails."
He sat down on the second stool. The orb pulsed once on the bench, settling into its low steady rhythm. "The organisms in my paint handled stress signals differently," he said. "I noticed it when I was modifying the strain. They didn't compensate. They absorbed."
She looked up.
"The modification I did two years ago," he said. "To extend survival duration. The key variable was stress signal response. Instead of pattern-shifting, they incorporated the stress signal into their oscillation. Made it part of the frequency rather than fighting it."
She looked at the emitter. At the orb. At the sensor readout from the last failed run.
"If the organisms treated the interference pattern as a supplemental input rather than interference," she said, "the stabilization wouldn't drift."
"That's what mine do."
She pulled up the modification protocol from her original Florabyte documentation and looked at what she had changed and what she hadn't. The stress response subroutine was six months of work she had filed away as a secondary problem. In the context of the emitter's interference pattern, it was the whole problem.
She rebuilt the test suspension with Zavo's modified strain that night. It took four hours. When she ran the emitter against it in the morning, the organisms held coherence for seven consecutive hours before she stopped the test.
She built the full sealed chamber test while he watched from the second stool, neither of them talking much. The orb sat between them. She was aware of it the way she had been aware of it in his loft: present in the room in a way that was not loud but was specific. She had not asked him again where it was from. The question felt less urgent now than the work, which was always the shape of things she would have to come back to later.
She activated the emitter and they both watched the exhaust monitor.
One point two parts per billion. Down from baseline. Measurable, repeatable, real.
She ran it six more times over the following week and got the same result within two percent every time.
On the seventh day, she went to Villanova directly.
She had thought about waiting for the board meeting to schedule through proper channels. Then she had thought about the Project Asphyxia resource draw and the way Zavo had said whatever they say, and she decided that if she was going to be outmaneuvered, she would rather know the shape of it first.
His assistant let her wait twelve minutes. She counted them.
Villanova's office was on the top floor of Flora's northwest tower, its windows looking out over the contaminated skyline. He was at his desk when she came in, the HUD in his left cheekbone pulsing faintly as it synced to something she couldn't see. He gestured to the chair across from him without looking up.
She set the carry case on the table and opened it.
He looked up then.
She ran the demonstration without preamble. The emitter. The sealed chamber. The exhaust monitor showing one point two parts per billion, down from baseline. She walked him through eighty-seven controlled trials in four minutes. She did not oversell it. The data did that.
When she finished, Villanova was quiet. His HUD pulsed. He looked at the emitter the way he looked at things he was deciding how to categorize.
"How long has this been running?" he asked.
"Four months of stable trials. The emitter architecture took two months before that."
"Real-world atmospheric tests?"
"Not yet. That's the next phase."
He nodded slowly, as if this confirmed something rather than established something new. "The board needs to see this," he said. "Formally. Slides, full trial data, failure protocol." He set his palms flat on the desk. "Tuesday."
She read the room. "You want me to present to the board."
"I want the board to make an informed decision." His tone was smooth, nothing in it she could point to. "Prepare the full presentation. I'll arrange the time."
She picked up the carry case. At the door she paused. "Professor Villanova. Is there anything I should know before Tuesday?"
He looked at her. The HUD pulsed once. "Bring the best version of the data," he said. "That's my only advice."
She thanked him and left.
The corridor felt colder than usual on the way back to her lab.
Three days before the board presentation, she found the first reference to Project Asphyxia.
She had not been looking for it. She was cross-referencing her atmospheric monitoring data against Flora's internal quality records, standard practice before a board presentation, and the access log for a secondary partition showed a resource draw flagged with a project code: PX-7713. Project Asphyxia.
The code was not in the public directory. She ran a search. Nothing.
She checked the resource draw instead. The atmospheric monitoring division had transferred data to PX-7713: fourteen months of records flagged as corrupted in the public system, two years of chlorine concentration readings that did not match the official figures Flora published citywide.
She pulled the corrupted records. They were not corrupted. They were clean, intact, consistent. They showed chlorine concentration in the lower ring three to five percent above what Flora's public reports listed. Not measurement error. Not equipment drift. A gap too precise and too sustained for any explanation except a deliberate one.
Someone had been suppressing the real numbers.
She ran the comparison further back, accessing every atmospheric record she had clearance to touch. The gap appeared first four years ago, widened gradually, stayed consistent. The suppressed figures were always stored intact in the internal system while the public system received the altered ones. The method was careful. It left no fingerprint in any individual record. You had to see the full four-year trend before the shape of it became unmistakable.
She printed nothing. She copied nothing she did not already have access to on the record. She closed the comparison and sat looking at her blank screen for a while.
Then she went home and added Project Asphyxia and the date and the resource draw number to the notes document she kept on a personal drive, the one she did not store on any Flora server.
Zavo came by the lab the following afternoon.
He did not knock. He never knocked. He came in and crossed to the bench and looked at the emitter without touching it, which was new. He usually reached for things immediately.
"Board presentation is Tuesday," he said.
"I know when it is."
He picked up the sensor log. Scrolled through it. Set it back down without commenting.
"Zavo." She waited until he looked at her. "Is there something I should know before Tuesday."
He held her gaze for a beat longer than the question required. The particular quality of it was one she had learned to read over the past months: the moment before the thing he actually thought replaced the thing he had planned to say. She waited.
"It's good work," he said. "Whatever they say."
Whatever they say. Not: I think it will go well. Not: Villanova will recognize what this is. Whatever they say. As if he already knew the shape of the outcome and was making room for her to hear it without breaking.
She thought about the Project Asphyxia resource draw. Fourteen months of suppressed data, intact in a partition she was not supposed to know about. The same atmospheric division Zavo had contacts in through his painting work, through the lower ring supply networks she had never asked him about in detail.
She did not ask now. She filed it.
"I'll be fine," she said.
He nodded. He was already looking at the emitter again with the expression she had come to recognize as his version of admiration: not warm, not demonstrative, the way someone looks at something that worked exactly as intended.
"I know," he said. He left.
She sat at the bench for a moment after the door closed. She picked up the emitter and held it. Then she put it back in the carry case and locked it.
She spent the two days between the Project Asphyxia discovery and the presentation running the board meeting in her head.
Not the optimistic version. The real version.
Villanova would open with interest. He was good at that: genuine-seeming engagement with the early slides, the kind that built her hope before he used it. He would ask about failure protocols. She had failure protocols. He would ask about real-world trial data. She didn't have real-world trial data, and that was the wedge. The gap between controlled chamber results and atmospheric deployment at scale was legitimate. She had always known it was legitimate. In any other context it would have been a reasonable scientific objection.
In this context, she now knew, it was a door he had been planning to open since before she walked in.
Project Asphyxia had been running for four years. Flora had been suppressing real chlorine concentration data for four years. Which meant that someone at the director level had decided, four years ago, that the city's air was not going to get better on its own timetable, and had built an internal project to manage that fact while publicly pretending otherwise. The Florabytes were not a threat to a company trying to solve a problem. They were a threat to a company that had decided the problem was a product.
She had known this abstractly for years. The Project Asphyxia resource draw had made it concrete.
The board meeting was not a scientific review. It was a performance. Villanova would play the responsible executive and she would play the idealistic scientist and at the end he would table the project for further review and she would leave the room and the Florabytes would go into a file that Flora controlled.
Unless there were witnesses who understood what they had seen.
Unless the data was already off site.
Unless whatever happened to her in that room, the work survived.
She checked the data stick in her jacket lining for the third time that week. It was there.
She was calibrating the board demo the night before the presentation when the drones changed pattern.
The lab occupied the sixth floor of Flora's northwest tower, facing the street. She had worked in this room for two years and she knew the drone paths: a standard perimeter sweep every forty minutes, two drones running offset loops over the commercial district below. She knew their tone by the time of night, the higher pitch when they were facing into the wind, the lower harmonic when they were downwind and coasting.
Tonight there were three drones on the northwest route instead of two. Running a tighter overlap. The same section of street covered twice on every pass.
She stood at the window and watched them run the pattern four times before she was sure of what she was seeing.
Then she went back to the carry case she had packed for the presentation. She took out the extra data stick in the inside pocket. She plugged it into her personal drive, copied the full prototype documentation, and wrote a note on the tag: synthesis method, emitter specs, full trial log, Project Asphyxia resource draw. She put the stick inside the lining of her work jacket, the one she wore every morning.
She checked the carry case. Prototype, demonstration chamber, datapad with trial results. Everything locked.
She looked at the emitter one more time. The small device that had started as a frequency in her sternum while paint dried on her skin. Four months. Eighty-seven trials. A number too small to see but provably, replicably real.
Tomorrow she would carry it into a room full of people who had already decided, and she would show them anyway.
She turned off the lab lights and left.
Outside, the three drones ran their overlapping loops. Below them, the city breathed its slow contaminated air and nobody in the lower ring could afford to notice the difference.
She walked to the transit line without looking up at them, keeping her pace even, her hands easy at her sides. The data was copied. The backup was in her jacket. Whatever the board said tomorrow, the work survived.
Her mother's mask had cracked along the seal because there was no replacement.
She knew exactly why she was still here.
Her communicator buzzed as she reached the transit line.
A new message. No caller ID. The same anonymous source that had warned her before.
They moved the timeline. Board decision tomorrow morning, not next week. Someone wanted it done before you had time to prepare.
She looked at the backup drive in her jacket pocket. She had prepared. She just hadn't known she'd need to be this prepared this fast.