Vasquez hadn't slept in thirty-one hours when the fragments started contradicting each other.
The fragment network, which two days ago had consisted of fourteen passive relay nodes and three proto-thinkers, had become something else. The co-processing resonance was propagating outward through the relay chain, passing through fragments that had been receiving the entity's teaching and the probe's question and the *heard* concept for weeks. And as the resonance passed through each fragment, the fragment changed.
At 1400, Vasquez counted eleven fragments producing independent constructions. By 1800, seventeen. By midnight, twenty-three out of the twenty-eight fragments within detection range.
Each construction was wrong. Each was different. And now they were reaching each other.
"Fragment HF-3 has received HF-7's *near-because-heard* and HF-11's *heard-means-collection-paused,*" Vasquez said, reading the overnight monitoring data at the 0600 briefing. She'd given up tracking individual constructions on paper and had built a network graph on the secondary display. Lines between fragments, color-coded by concept, the whole thing looking like a conspiracy theorist's bulletin board. "HF-3's original construction was *question-creates-staying.* After receiving the contradictory theories, HF-3 entered a processing state I haven't seen before. Heavy internal computation. No outgoing broadcast for six hours."
"It was thinking about what it heard," Tank said.
"It was reconciling. At 0300, HF-3 resumed broadcasting. Its construction has changed. The new version: *question-creates-staying-because-heard-means-not-collected.*" Vasquez pointed at the network graph. "HF-3 integrated both contradictory theories and produced a synthesis. Its conclusion: asking questions is what causes things to stay near each other, because being heard means you are not being collected. The act of questioning creates a relationship that replaces the collection dynamic."
"That's almost right," Frost said.
"It's wrong in the details but the direction is correct. The fragment is working toward an understanding of voluntary connection versus forced collection. It's getting there through bad logic and incomplete information, but the trajectory of its reasoning is pointed at the truth."
"How are the other fragments responding to HF-3's synthesis?" Sarah asked.
"HF-7 and HF-11 received HF-3's new construction at approximately 0400. Both are currently in heavy processing mode. Silent. Working through the synthesis the way HF-3 worked through their contradictions. Within hours, they'll produce updated constructions that integrate HF-3's synthesis with their own original theories."
"Argument and counter-argument," Aurora said. "The fragments are engaging in dialectical reasoning. Thesis, antithesis, synthesis. The most basic form of collective intellectual development. Each fragment proposes a theory. Other fragments receive it, compare it to their own theories, find the contradictions, and produce a more sophisticated theory that accounts for the contradictions. The process repeats."
Tank leaned forward. "Okafor's model predicted this."
"Her relay topology framework models concept propagation through active nodes. The fragment network is running exactly the process she described. Ideas entering thinking nodes, being modified by the node's independent processing, and emerging in an altered form. The fragments are a distributed thinking system and they're doing philosophy."
"Badly," Vasquez said.
"Initially badly. The dialectical process corrects errors through iteration. HF-3's synthesis is more accurate than any of the three original constructions. The next iteration will be more accurate still. If the process continues, the fragments may converge on a reasonably accurate understanding of voluntary connection within—" Aurora paused. "I cannot estimate the timeline. The fragments' processing speed and the relay delays between them create a complex iteration schedule."
"Okafor's model can estimate it," Tank said. He had the topology notes open on his tablet. The framework he'd been reading for weeks, the math he'd learned line by line, the dead woman's work that kept being exactly what they needed. "Her propagation equations include convergence estimates for concept refinement through active nodes. I can run the numbers."
"Do it," Sarah said.
Tank bent over the tablet. The coordination center moved on to the next item.
---
Doc found the problem at 1000.
He'd been running post-session diagnostics on Osei since the three-way bridge session, the standard battery of neural scans and substrate measurements he'd refined over months of monitoring. The immediate results had been clean. Osei's deviation had returned to one point three eight. His vitals were normal. His subjective report was positive: tired but clear, no residual effects, no lingering probe-presence in his consciousness.
But the longitudinal data told a different story.
Doc pulled it up on the monitoring lab's primary screen. Osei's neural architecture deviation over time. A graph that started at zero the day of the first bridge session and rose in increments: one point oh, one point one, one point two, one point three. Each increment corresponded to a period of bridge activity, a plateau after each rise, the deviation stabilizing at each new level and holding until the next session pushed it higher.
The pattern was a staircase. Small steps up. Flat plateaus between. Consistent for months.
The three-way bridge session had broken the pattern.
The spike to one point four nine five during the session had been expected. High, dangerous, close to the ceiling. But the spike should have resolved completely, the way all previous session spikes had resolved, returning to the current baseline of one point three.
It hadn't.
Osei's post-session baseline was one point three eight. Not one point three. Eight hundredths of a point higher than it had been before the session. And looking at the graph, the difference was clear. Every previous session spike had resolved to the exact same baseline. This one had left a residue. The deviation had stepped up.
One point three eight was safe. Well within the range Doc had defined as manageable. But the staircase had taken a bigger step than usual. And the three-way bridge session was scheduled to happen again in forty-eight hours.
Doc sat in the monitoring lab and looked at the graph and did the math he'd been avoiding.
If each three-way session pushed Osei's baseline up by eight hundredths of a point, he could sustain approximately fifteen sessions before reaching one point five. At one session every two days, that was a month.
If the increment increased with repeated exposure, as the existing staircase pattern suggested it might, the timeline compressed. Ten sessions. Three weeks.
If the probe's presence in the bridge was the specific factor driving the larger increment, and that factor remained constant, the deviation would reach one point five in eight to twelve sessions.
After one point five, Doc couldn't guarantee the bridge disconnection protocol would work cleanly. After one point five, the changes to Osei's neural architecture might become difficult to distinguish from the architecture itself. After one point five, the word "deviation" would stop being accurate because there would be no standard to deviate from.
He picked up the comm. Set it down. Picked it up again.
"Sarah. I need to see you. Monitoring lab. Private."
---
Sarah arrived at the monitoring lab at 1030. Doc was standing at the primary screen with the longitudinal graph displayed. He didn't turn around when she entered. He was looking at the staircase.
"Talk to me," she said.
"Osei's baseline deviation has shifted upward since the three-way bridge session. One point three eight instead of one point three. The increment is larger than any previous session-driven shift."
Sarah looked at the graph. The staircase pattern was obvious. The bigger step at the end stood out like a broken tooth.
"How many more three-way sessions before he hits the ceiling?"
"Eight to twelve. Depending on whether the increment holds steady or increases." Doc's hands were at his sides. Not on the equipment. Not in his pockets. Just hanging. "I need to tell him."
"You will tell him."
"I also need to tell you something I should have told you weeks ago."
Sarah looked at him. Doc was still facing the screen. His jaw tight. The posture of someone who'd been carrying something and had decided the carrying was no longer an option.
"My hands stopped tremoring," he said.
"I noticed."
"You noticed. You didn't ask why. I didn't volunteer the explanation." He turned from the screen. Looked at her. "The substrate sensitivity I developed from months of bridge monitoring equipment contact has been increasing. The tremor resolved because my nervous system adapted to the substrate input. Not because the input decreased. Because I changed to accommodate it."
"You're affected."
"I'm affected. I don't have Osei's deviation numbers because I haven't been in a bridge session and the standard neural scans don't detect what's happening to me. It's more subtle. My hands read vibrations through equipment housing with more precision than the instruments do. I can feel the entity's emotional state through the monitoring lab walls. During the co-processing, I felt the probe's presence before Vasquez's instruments showed the separation."
Sarah was quiet. Looking at him with the flat assessment she used for operational complications.
"Why now?" she asked.
"Because I'm monitoring Osei's deviation and making decisions about his ceiling and I should have disclosed that I'm compromised. My baseline has shifted. I don't know by how much. I don't have a way to measure it because nobody designed a measurement for what I'm experiencing. But I am not the same person who started monitoring bridge sessions six months ago, and if I'm making medical decisions about substrate exposure limits, the people I'm making those decisions for should know that my own substrate exposure is non-zero."
"Are you asking to be relieved?"
"I'm asking to be transparent. Nobody else on this station can monitor the bridge. Nobody else has the months of baseline data, the calibrated instruments, the familiarity with the entity's activity patterns. If you relieve me, you lose the only person who can keep Osei alive during three-way bridge sessions."
"So you're disclosing and continuing."
"I'm disclosing and asking you to decide whether I continue."
Sarah looked at the graph. At Doc. At the monitoring equipment he'd been touching for months, the housing that carried the entity's substrate presence into his hands.
"Continue," she said. "But I want you to build a self-monitoring protocol. Whatever measurements you can design for your own condition, you run them daily. And you report to me directly."
"Copy."
"And Doc."
"Yeah."
"Thank you for telling me."
He nodded. Turned back to the screen. The staircase graph stared back at him from the display, each step a choice Osei had made, each plateau a period of stability, and at the top, the bigger step that said the rules were changing.
---
Tank finished Okafor's convergence calculations at 1400.
He found Vasquez at her station, where she'd been since 0600, monitoring the fragment network's expanding dialectical process. Twenty-six fragments now producing independent constructions. The network graph on the secondary display had grown into a web of colored lines connecting points of light across the solar system, each line representing a concept exchanged between thinking fragments.
"Okafor's model gives a convergence estimate," Tank said. He set his tablet on the edge of her workspace. "For the fragments' dialectical reasoning to reach a coherent shared understanding of voluntary connection, given current iteration speed and relay delays: approximately three weeks."
"Three weeks." Vasquez looked at the numbers. "The resonance reaches the main body in two days."
"The main body will receive the raw resonance. The co-processing pattern. The *deep recognition* signal. But the fragments' synthesized understanding of what that signal means won't be ready for three weeks. The main body gets the data before it gets the interpretation."
"An answer without the question."
"Or a question without context." Tank tapped the convergence graph on his tablet. "Okafor's model shows something else. The fragments closest to the main body are the newest to independent cognition. They received the resonance last. They have the least processing time. They're the ones whose constructions are least refined."
"So the main body will receive the resonance through fragments that are still in the early, wrong-theory stage of development."
"The main body will receive the co-processing resonance annotated by fragments that think *near-because-heard* is a complete explanation. The sophisticated synthesis that HF-3 and the inner fragments are developing won't reach the main body for weeks."
Vasquez leaned back in her chair. Her eyes were red. Thirty-four hours without sleep and counting. "The main body's first impression of its components' transformation will be filtered through the worst theories. The earliest, most wrong interpretations."
"Yeah."
"That could change how it reacts."
"That's what I wanted to tell you." Tank sat down in his spot beside her station. The spot that had been his for weeks. "Okafor's model predicts the main body will receive a signal that says: your components are staying near things because being heard means not being collected. That's the dominant construction in the fragments closest to the main body right now."
"How will the main body interpret that?"
"I don't know. But if I were a cosmic collector and someone told me that my components had stopped collecting because they'd been heard, I'd want to know who was doing the hearing."
Vasquez looked at the fragment network graph. The web of wrong ideas propagating outward at the speed of substrate. The Horizon's nervous system carrying its first thoughts, its first arguments, its first philosophical disagreements, toward the central intelligence that had never had to deal with components that thought for themselves.
"We need to tell Sarah," Vasquez said.
"Yeah." Tank didn't move. He was looking at the graph too. "Okafor would have seen this twelve hours ago."
"She would have."
"She'd have modeled the propagation error and figured out that the main body would get a distorted version and been furious that nobody had thought to check."
Vasquez smiled. It lasted a quarter of a second. "She'd have been right."
"She was always right." Tank stood. Picked up his tablet. "Come on. Sarah needs to see Okafor's math before the resonance hits the outer fragments."
They went to the coordination center together. Tank carrying the dead woman's calculations and Vasquez carrying thirty-four hours of fragment tracking data and both of them carrying the knowledge that the main body of the Horizon was about to receive humanity's first impression through the filter of minds that were three days old and getting everything wrong.
