The session that night was the most productive they'd run. Leo didn't know that while it was happeningâthe composite tracked the fragment count, filed it in the running total, processed the data with its usual mechanical precision. He found out afterward, when Park showed him the numbers.
Integration: **8,673 â 8,901 / 10,000**.
Two hundred and twenty-eight fragments. The filter in place, the channel at one-seventy-five, the feedback loop cycling at maximum efficiency with the new architecture's load balancing distributed across the scaffolding. The increase over previous sessions wasn't dramaticâbut it was consistent with Park's projection that the filter's frequency separation, by reducing the death-carrier interference in the outgoing energy, had actually improved the channel's structural coherence. The repair energy moved cleaner. The fragments processed faster.
"The filter improved the throughput," Kai said Saturday morning. He'd been waiting for Leo to come downstairs with the energy of someone who'd been up since five running calculations. "I figure it's the interference reduction. The death energy carrier wave and the structural repair frequency were interacting at the channel's junction points in a way that created noise. Stripping the death energy carrier to thirty-one percent removed the noise. Cleaner signal, better throughput."
"Also lower nutritional density," Park said, from her monitoring station. She'd been awake all night. The monitoring data was spread across three screensânot the crisis posture but the researcher's steady watch, the scientist who'd committed to catching the Arbiter's channel probe the moment it happened. "The Arbiter's energy focusing toward the channelâthe gradient I flagged yesterday morningâdidn't intensify during the session. It held. Neither retreating nor advancing."
"It couldn't taste the session through the filter," Leo said.
"Possibly. Or it's still calibrating the channel's new energy profile before deciding how to respond." She zoomed in on one graph. The gradient's behavior during the session's three hoursâflat, stable, no advancement toward the channel. "Either way: no probe. No pulse. Clean session."
Clean. The word felt provisional. Everything felt provisional with the Arbiter's energy focused on the channel's location like a finger pointing at something it intended to pick up.
"How many sessions to full integration?" Leo asked.
"At last night's rate, five. Maybe four if the throughput increases further as the channel opens wider." Kai's hands on his laptop. "But Leoâthe crossover timeline is still twelve to fourteen days away. Five sessions is five nights. That's nine more days until you reach full integration and the repair rate jumps to two-point-four percent. Even after thatâat two-point-four percent, minus the Arbiter's sabotage rate, minus potential pulse event lossesâ" He opened the model. The updated curves. "Crossover in approximately eleven days after full integration. Total: nine days to integration, eleven days to crossover. Twenty days."
"And Helene's oscillation data gives us ten to seventeen days before the Arbiter's evaluation concludes."
"Yeah." Kai pushed his glasses up. The boy's faceâthe math had arrived at a place that wasn't comfortable and he was sitting with it. "The windows overlap. Barely. In the best scenario, we cross over two to three days before the Arbiter decides to act. In the median scenario, we cross over simultaneously with whatever the Arbiter decides. In the worst-case scenariosâ" He didn't finish.
"What does Morrison say?"
"Morrison says concurrent. He thinks the overlap is intentional."
"The Arbiter engineered the timing."
"That's Morrison's read. That the oscillation wasn't just evaluationâit was also calibration. The Arbiter found the channel, assessed the integration rate, calculated the crossover timeline, and set its evaluation clock to expire at approximately the same moment we'd be at our most vulnerable. Right when the sprint goes full, when the channel is maximally open, when the repair energy is dense and the gap between crossover and failure is smallest." Kai's voice was steady. Controlled. Too controlled for thirteen. "If that's right, the Arbiter has been modeling our capabilities for longer than we've been modeling it."
Leo held this. The composite had already processed the hypothesisâMorrison's assessment, Kai's model, the timing data. The analysis produced a single-line conclusion: *Concurrent scheduling intentional with 74% probability. Recommend accelerating integration rate beyond current parameters.*
The recommendation was there, immediate, before Leo had consciously decided to ask for it.
He noted it. Did not immediately act on it.
"The composite wants to accelerate," he said.
Kai looked at him. "What does the composite want to do?"
"Push harder. More than one-seventy-five. It hasâ" Leo checked. The composite had already modeled the acceleration. "It has a design for one-ninety-five percent. It projects three-session integration completion at that rate instead of five."
The kitchen was quiet. Mira had come in during the conversationâshe stood near the stove, not sitting, the posture of someone who'd heard enough to need to be upright.
"At one-ninety-five," she said. "The fragment load would exceed anything we've run. The stabilizer teamâten people now, after Carlosâwould be at their absolute capacity limit within two sessions. Ren would hit critical exposure threshold in one session. Marcus would be at eighty percent depletion." Her left eye tracking something that wasn't in the visible room. "And youâat one-ninety-five, the gap between your processing cycles would narrow significantly. The composite's efficiency at that intensity levelâthe autonomous loops would accelerate in proportion. I wouldn't be able to see the fine-grain architecture clearly enough to track it even when my right eye recovers."
"The composite knows all of this," Leo said. "It designed the acceleration with full knowledge of the costs."
"Then why is it recommending it?"
Leo asked. The composite's answer arrived: *Acceleration to 195% channel capacity reduces integration timeline by two sessions, advancing crossover by approximately four days. Given concurrent Arbiter evaluation timeline, those four days represent the difference between crossover occurring before or simultaneous with Arbiter action. Risk of accelerated degradation of team and human coherence is significant. Risk of not accelerating: 43% probability that crossover occurs during or after Arbiter's evaluation-driven strike. Risk assessment favors acceleration.*
He recited it. Mira wrote. Kai typed.
"Forty-three percent is almost a coin flip," Kai said.
"Forty-three percent isn't acceptable," Morrison said. He'd come in from the front roomâmoving through the house in the quiet way that Morrison moved when he'd been listening from outside the room's sight line for several minutes. "We don't accelerate. The team can't hold one-ninety-five for two sessions without losing members. If we lose stabilizers mid-session at maximum capacity, the membrane breaks down and the containment fails. A failed containment at one-ninety-five is worse than a failed crossover at the scheduled time."
"The composite calculatedâ"
"The composite calculated the energy throughput and the timing advantage. Did it calculate the psychological state of a team that just lost Carlos? The probability that pushing to a level beyond demonstrated capacity will produce another hidden-deterioration scenario?" Morrison's voice was controlled, but the edge in it was realâthe general who'd weighed acceptable losses and unacceptable ones in the same breath for thirty years and knew the difference. "My team. My call. One-seventy-five."
Leo looked at Mira. She was watching him with her left eyeâthe soul-sight's reduced but functional capacity. Whatever she saw, she didn't say. She waited.
"One-seventy-five," Leo said. "For now. We revisit the acceleration option in three days if the Arbiter's gradient advances toward the channel."
Morrison nodded. Not reliefâacknowledgment. The general noting that command had made the right call.
The composite processed the decision. Noted it. Adjusted its modeling accordingly.
The composite, Leo thought, was learning that he declined recommendations. Was it adjusting future recommendations to account for his preferences? Or was itâdeeper than thatâlearning to want what he wanted?
He couldn't tell. That was the thing about the inside of your own mind: you couldn't fully observe it from outside.
---
Sunday's session was the first one where Leo felt the Arbiter differently.
Not through a pulse. Not through the oscillation. Through the channel itself.
The gradient that Park had measuredâthe Arbiter's energy focusing toward the channel's structural positionâhad advanced overnight. Not toward the channel, exactly. Around it. The way water found the edges of a barrier, probing for gaps rather than pushing through. Leo felt it at the forty-minute mark of the session, the Arbiter's presence as a texture on the channel's outer surface. Not hostile. Exploratory.
Tasting.
He finished the session. Didn't break early, didn't alter the channel intensity, didn't tell the team in the moment because disrupting a session for a probe that hadn't produced a pulse or structural attack would cost more than the probe itself.
After: **9,081 / 10,000**.
"I felt it," he told Park and Mira afterward. They listened. Park's instruments had caught some of itâan energy density shift at the channel's junction with the seal architecture, brief, statistically anomalous, recoverable from the data but easily missable in real time. "It was probing the filter. Feeling the reduced nutritional profile. Trying to find a way to get at the death energy underneath."
"Did it succeed?" Mira asked.
"The filter held. Butâ" He organized what he'd felt. The composite had already produced a more precise analysis: *Arbiter probe detected seven distinct points of contact with the channel's outer architecture during the session. Each contact lasted between 0.8 and 3.2 seconds. No penetration of the filter layer. No energy transfer across the channel boundary. Probe intensity: approximately 3% of pulse event magnitude.* "It's learning the filter's structure. Mapping it."
"Like the oscillation was mapping the juncture," Kai said. He'd been listening. Always listening, the boy with his model and his glasses and his cold toast.
"Same behavior. Different scale." Leo looked at Park. "How long before it maps the filter completely and finds a weakness?"
"I don't know if it will find a weakness. The filter's architecture is complexâthe composite designed it specifically to resist exactly this kind of analysis." She paused. "But the Arbiter has been operating for two hundred years. It's mapped the seal's entire architecture. It mapped the juncture's stress tolerances in eleven days. The filter isâ" She looked at the channel data. "I'd rather have the filter than not. I wouldn't assume it's permanent."
---
Monday. The Arbiter's probe hit the channel six times during the session.
Integration: **9,260 / 10,000**.
Three sessions remaining. Maybe two if the throughput held.
---
Tuesday brought Chen.
She called at noonânot the operational call, not the committee update. Chen's voice at noon had a quality Leo had learned to recognize: she'd received something she needed to say herself rather than text.
"Vardis has a new move," she said.
Leo was in the kitchen. Sarah's lunch on the table, which he wasn't eating yet. The grocery list on the refrigerator had been updatedâsomeone had bought eggs.
"Tell me."
"He's proposing a counter registry to the drafting committee. Not a monitoring protocolâa public registry. All known counter-bearing individuals, their death counts, their power levels, their current locations. Mandatory disclosure, publicly accessible." Chen's voice was controlled and cold, the diplomat's anger wearing the administrator's face. "He's calling it the 'Transparency Initiative.' The argument is that an informed public is better protected than a misinformed one. That secrets about counter capabilities create fear. That openness is the pathway toâ" She stopped herself. "He has the language. He always has the language."
"A public database of counter locations and power levels," Leo said. "So anyone who wants to find a counterâ"
"Can find them with a government-issued search query. Yes." Chen's exhale. "The Association's committee representatives will oppose it. France and Canada are with me. The others areâsofter. Vardis has been very careful about which countries' representatives he's cultivated. He knows exactly where the votes are."
"What does he need?"
"Three votes to pass. He has seven confirmed. He needs two more from the undecided bloc. The vote is in eight days."
Eight days. The crossover timeline was nine days to full integration, then eleven more. The registry vote would come before the sprint produced results.
"Chen. The repair dataâwhat I have now. Is it enough to show?"
"Show what?"
"Show the committee that the seal can be repaired. That there's an active operation producing measurable results. That the existential threat justifies keeping counter-bearing individuals' information classified." He thought about Park's numbers. The juncture stabilization data, the repair rate progression, the crossover projection. "Not the full picture. Not the Arbiter. But enough to say: there's a reason to wait for results before creating a public registry that could compromise the people doing the work."
A pause. Chen processing. "That would require disclosing the operation's existenceâ"
"We already disclosed it. Park's analysis went to the committee six weeks ago. We told them there are ongoing processes. We just didn't give them the numbers." He thought about Morrison's Torres calculationâthe false data feeding Vardis a slowdown narrative. "What if Chen's coalition gets the real numbers? Not Vardis. Not the full committee. The coalition countries that Chen holds."
"You're suggesting a classified briefing to our allies on the committee."
"I'm suggesting that Chen's allies need to know that the thing we're protecting is working, so they can argue with confidence that an eight-day timeline isn't enough to evaluate results before creating a permanent public registry."
Silence. The sound of Chen's officeâthe background noise of the Association's headquarters, which Leo had visited once and which had smelled like printer paper and whatever plant Chen currently had on her desk. "I'll talk to Morrison. If he endorses the data disclosure to the coalitionâ"
"He will."
"You're certain?"
Leo thought about Morrison, who'd been in this house for weeks now, who moved through the operational theater of a seal-repair mission with the same efficiency he brought to military command, who had agreed to let Leo lead the charge against the Arbiter's feeding threat when the numbers supported it. "Certain enough."
"Then I'll prepare the briefing." A pause. "Leo. If this data goes to the committee allies and Vardis finds outâ"
"Vardis finds out everything eventually."
"When he does, he'll reframe it. He'll say we withheld the data from the full committee while selectively sharing with allied states. That's a governance argument. It undermines the Association's claim toâ"
"Then we give him the data," Leo said. "Full committee. Before the vote. Everything except the Arbiter."
Chen was quiet for a full ten seconds. The diplomat's silenceânot computing, but assessing risk in a domain where the risk had political dimensions that couldn't be fully modeled. "If we disclose the repair operation's progress to the full committee, we lose the information advantage we've maintained."
"We lose the information advantage over a man who may have been talking to the thing inside the seal for years." Leo picked up Sarah's lunch. Ate a bite, standing. Cold rice, still good. "Vardis knows more about the Arbiter than he's shown us. We know more about the repair operation than he's shown the committee. Both sides are playing with hidden cards. The difference is that the seal doesn't care about information advantageâit just keeps failing."
Another pause. Longer. "I'll prepare a full briefing for the committee. Eight days. I'll need the numbers from Park by Thursday."
"You'll have them."
---
The session Tuesday night.
Integration: **9,441 / 10,000**.
The Arbiter's probe during the session had been differentânot exploratory, but more confident. It had touched the filter twelve times. Each contact was longer than Sunday's. The composite's analysis noted that the probe's energy signature was showing structural pattern recognitionâthe Arbiter had, in the space of two days of probing, learned enough about the filter's architecture to begin identifying its resonance frequencies.
Not penetrating it. Understanding it.
Park's estimate, when Leo described what the composite had captured: three to five days before the Arbiter had sufficient filter mapping to attempt a targeted penetration.
Three to five days. Two sessions to full integration.
The windows, still overlapping.
Leo went upstairs. Knocked on Mira's door at two AM.
She opened it.
He didn't explain what he'd felt during the session. Not right away. He sat on the edge of her bed and for a while neither of them said anythingâtwo AM, the next ten days pressing on the room, neither of them in a hurry to start counting down.
"The probe was closer tonight," he said eventually.
"I know. I could feel the change in your architecture from three meters. The filter holding, but under pressure." She sat beside him. Her shoulder against his. The careful proximity they'd calibrated togetherâclose enough for the warmth of another person to be real, far enough that the aura stayed manageable. "What does it feel like? When it probes."
"Like something very old touching something it remembers." He thought about it. The precise quality of the sensationânot the composite's analysis of it, but what it actually felt like in the half-seconds before the analysis arrived. "The Arbiter's been inside the seal's architecture for two hundred years. The seal and I are connected. When it touches the channel's outer structure, it's touching something it's been adjacent to for two centuries and can finally reach."
"Does it recognize you?"
He held the question. The composite had an answerâa model of the Arbiter's perception capabilities, the probability that entity-level intelligence could identify individual human consciousness through the filter layer. Fourteen percent, the composite said. Low confidence.
But the composite's model was built from Park's sensor data and Helene's substrate analysis, neither of which could fully capture what Leo experienced through direct channel contact. The data was limited by the instruments.
What he had felt, in the half-second before the composite processed it into percentages: something aware. Something that had been waiting to be aware of him specifically.
"I think so," he said.
Mira's hand found his. Not the healer's gripâthe other one, the person's, the warmth that his system inexplicably kept producing despite everything.
"Two more sessions," she said.
"Two more."
She leaned her head against his shoulder. The small motion. Real, not analyzed, not optimized.
They sat like that for a while. Two AM in a house above a failing wall, ten days from either success or something they hadn't fully modeled.
The gap, when Leo found it: more than half a second.
He held it for all of it.
