Gareth had pulled the Kellerman Assessment Hall evaluation data before Damien reached the integration ring.
He knew the fight outcome from the registry pushâit was automatic, the moment assessment staff logged the evaluation result, the data went to the Association's record system and was publicly accessible to any registered guild. He didn't comment on it when Damien walked in. He looked at Damien for a moment, the way he looked at things that required analysis before comment, and said:
"The Archival Mage."
"Dead. Cardiac event this morning. He collapsed before the Tuesday appointment."
Gareth took his glasses off. He set them on the equipment bench with the specific care he used for things he was doing mechanically while thinking about something else.
"What was his name."
"Aldric Verne. Forty-nine years with the class. He spent them preserving historical records for a private foundation."
Gareth was quiet. He put his glasses back on.
"The Foundation's preservation cycle," he said. "His collection will need stabilization within six months or the mana imprint on the older records begins to degrade." He opened his notebook. "I'll contact the Association registry of Archival Mage class holders through Yuki's network. There are three other confirmed living holders. One of them may be able to assist with the Foundation's transition." He looked up. "That's within our ability to arrange. We should arrange it."
Damien looked at him.
"His work continued for forty-nine years," Gareth said. "That continuation doesn't stop because the fragment acquisition did." A pauseâmeasured, not performed. "Those are separate things. I know that."
He'd been thinking about the fragment. He hadn't been thinking about the collection.
"I'll get Yuki to start the contact."
"Good." Gareth turned to the monitor. The assessment data was on the secondary screenâthe evaluation metrics, the exchange duration, the sequence log that the hall's monitoring equipment auto-generated from its mana sensors. He looked at it without rushing.
"The Bladesmaster. Vael."
"I went in overconfident."
"The overconfidence was a symptom. The mechanism was a misapplication of successful dungeon entity combat tactics against a trained human specialist." He didn't say *you failed.* He said *mechanism.* Which was how Gareth communicated that he was interested in why, not what. He moved to the whiteboard. "Walk me through the engagement at the thirty-second mark."
Damien walked him through it. The phase-step and Vael already positioned at the destination. The Duelist timing sense identifying the opening that Vael had already closed before Damien could act on it.
Gareth listened without interrupting.
"He was operating one layer above your adaptation logic," Gareth said when Damien finished. "Not reading your current stateâreading the process that generates your current state." He wrote on the board: *ADAPTATION LOGIC.* Under it: *READ â ASSESS â SYNTHESIZE â APPLY.* "The four-phase structure of multi-class adaptation is consistent and learnable. A specialist who has encountered multi-class combatâeven theoreticallyâcan recognize the phase gaps between Assess and Synthesize, and Synthesize and Apply. Those gaps are small. For a class at full development, small is sufficient."
"He was in the gap before I was."
"Yes." He circled *SYNTHESIZE.* "This is the phase a class at full development doesn't need. The specialist doesn't synthesize anything. Their class's function is already synthesizedâyears of practice has made the full response library immediately accessible. Zero latency between observation and application." He looked at Damien. "You're running a synthesis step they aren't. That's the cost of generalism at this development level."
"At Fragment Harmonyâ"
"At Fragment Harmony, the synthesis step changes character. The network's simultaneous channel access means synthesis isn't a gap anymoreâit's a background process that's already running." He turned to look at the integration monitor. "The Harmony is four fragments away, after today." He met Damien's eyes. "Before the thresholdâthe combat approaches that work against human specialists in the interim."
He came back to the whiteboard. Drew two columns. *READABLE.* *UNREADABLE.*
"Readable," he said, "is the standard multi-class configuration. Fragments you use regularly, combinations that follow the read-assess-synthesize-apply structure. A trained observer sees the logic. Unreadable is any application that doesn't follow that structureâeither because it's a domain the specialist has no frame for, or because the output doesn't match what the input suggests it should be."
"Fragments I have from unusual domains."
"At six percent development," Gareth said. "The constraint is fragment power. A six percent Chronomancer fragment isn't going to produce a temporal distortion a Bladesmaster at full development can't respond to. But a Chronomancer fragment combined withâ" He stopped. He was looking at the fragment load list on the whiteboard's left edge. "The Chronomancer fragment in combination with the Phantom Blade's phase-sensitivity. What does the phase-step look like when the Chronomancer's temporal sense is running in the channel network simultaneously?"
Damien thought about it. The Phantom Blade's phase-step was a spatial displacementâmoving through the phase-layer to a new position. He'd been using it as a positional reset, which was exactly how Vael had predicted it. But if the Chronomancer's temporal sensitivity was running in the same channel simultaneouslyâ
"The temporal read would affect the destination calculation," he said slowly. "Not just where I can land. Where I can land accounting for what the opponent's next movement is going to be."
"That's not how phase-steps work in the Phantom Blade's isolated domain. But in a networked combination, where the Chronomancer's function is a lateral input to the Phantom Blade's spatial calculationâthe output would be a phase-step whose destination is informed by temporal probability rather than just positional geometry." He was writing quickly. "A Bladesmaster with twenty-two years of experience has a model for phase-step geometry. He does not have a model for phase-steps with temporally-informed destinations."
"It would still be six percent Chronomancer. Low accuracy."
"Six percent temporal probability read is sufficient to make phase-step destinations unpredictable. The Chronomancer's accuracy at six percent development is lowâbut unpredictability doesn't require accuracy. It requires noise in the pattern." Gareth put down the pen. "We'll work on this combination after today's acquisition. The training objective: make the adaptation logic invisible by routing it through domains a specialist has no frame for."
---
The harbor district dungeon was in a bonded warehouse building three blocks from the waterfront, abandoned for twenty years since the port authority had switched to automated logistics. The entity had been developing in the mana-rich environment of a space that had handled, organized, and moved cargo for its entire commercial existenceâdecades of documented flow and management had saturated the structure's walls.
The Harbor Administrator was a class expressed as a dungeon ecology of managed flow. Not the Tide Warden's territorial control through environmental manipulationâthis was a different relationship to a maritime system. The entity organized its dungeon territory the way a harbor administrator would: cargo zones, transit corridors, holding spaces, manifest records written in mana-light on the walls. Even in dungeon form, the entity's class function was management rather than combat.
It was harder to fight than a combat entity in one specific way. It kept creating new routes.
Every time he established a path through the dungeonâstructural read from the Earth Mage fragment, spatial map from the Scoutâthe Harbor Administrator rerouted the territory's access. Not aggression. Administration. It was managing the intruder as a resource to be redirected.
He used the meta-read against it.
The meta-read was reading the dungeon's internal channel architecture the same way it read the fragment network's architecture. The Harbor Administrator's rerouting had a logicâit was managing access based on administrative priorities. Once the meta-read had the priority map, the rerouting became predictable.
He was taking its organizational logic and using it to predict it. The entity, at its fundamental operational level, could not stop doing what its class function required it to do. Management didn't stop. The meta-read just learned the management language.
Twenty-seven minutes. The entity's defeat came when its administrative function tried to redirect him into a dead end that the meta-read had flagged as the entity's core zone, and he walked in willingly.
[Fragment 96: Harbor Administrator (B-Rank)]
[Retained: Bounded System Administration 10%, Resource Flow Management 10%, Maritime Logistics 10%]
He noted the fragment count in his notebook in the bonded warehouse's loading dock, standing in the afternoon light coming through a ceiling grate.
Ninety-six.
Four more.
---
The Chronomancer-Phantom Blade combination felt wrong for the first forty minutes.
Not wrong as in unstableâthe oscilloscope showed clean channel outputsâbut wrong as in it didn't behave the way he'd expected. The Chronomancer's temporal sense ran as a background read in the channel network. When he activated the Phantom Blade's phase-step with the Chronomancer channel open, the phase destination calculation wasn't random. But it wasn't what he'd projected, either.
The temporal read didn't give him a destination. It gave him something closer to a discomfort signal about destinations that were predictable.
"It's not probabilistic," he said to Gareth. "It's not telling me where to go. It's telling me where not to go."
Gareth came to look at the oscilloscope. He looked at it for a long time.
"The Chronomancer fragment is ten percent of the class," he said. "At ten percent, the temporal sensitivity is too low for predictive accuracy. The class at full development would show you probable futures with some reliability. At ten percent, the signal isâ" He stopped. "Interference on predictable outcomes. Not projection forward. Noise on familiar patterns."
"So the phase-step's destination is whatever isn't triggering the noise."
"That'sâyes." He made a notation. "That is not the elegant combination I projected. But it may be operationally more useful." He was thinking. "A phase-step whose destination is not the geometrically optimal position, not the tactically obvious position, but whatever position the Chronomancer fragment's interference pattern isn't reacting toâthat's inherently unpredictable. Because the interference pattern is reacting to your opponent's predictive model, which you don't have access to consciously."
Damien thought about that. The Chronomancer fragment was reading something about predictable trajectories and flagging them. He couldn't see the read directlyâhe just felt the discomfort of a destination being wrong, and moved away from it.
"It's reading their model and routing around it. Without my knowing what their model is."
"Technically," Gareth saidâand then stopped, and looked at Damien, because it was Damien's verbal tic and Gareth had used it without intending to. He made a notation. "Yes. Train on it. The combination needs to be practiced until the discomfort signal becomes automatic response rather than conscious processing. If you're thinking about it, there's latency."
He trained for another two hours. The combination became gradually less effortfulânot reliable yet, but the signal was clarifying. By the end he could activate the phase-step with the Chronomancer channel open and feel clearly which potential destinations were triggering noise and which weren't.
He couldn't always reach the non-triggering destination. His own movement was limited to what his body could execute. But the signal gave him information he hadn't had before.
At the end of the training session, Gareth said: "I reviewed the Association registry for Archival Mage holders. Aldric Verne was not the most senior living holderâthere's a woman in the Fourth District. Seventy-one. Her name is Sefa Olan. She was his professional contact for a decade before his retirement from active archival circuit work."
Damien looked at him.
"Yuki can facilitate the introduction through the professional connection." Gareth was looking at the oscilloscope data. "The Foundation doesn't know Yuki's network. The frame is that Verne had mentioned Olan's name as a potential successor for the collection's maintenance before his death." He met Damien's eyes. "Which he may well have done. He knew his health was declining. We don't know that he hadn't."
"He hadn't said that specifically."
"He'd said the collection needed to be transitioned properly. He'd said a week for the transition." He closed his notebook. "We make the transition happen. We use the correct social pathway to make it possible." A pause. "It's what he wanted."
Not entirely honest. What needed to happen.
"Set it up," Damien said.
Gareth nodded once. No other commentary.
---
He called Tomas on the drive home.
"The Perfect One threat profile," he said.
"Ready. I sent it to your secure channel an hour ago." Tomas's voice had the quality of someone delivering information they found genuinely alarmingâcontrolled, but carrying something underneath. "The interval projection was accurate. There's been a fourth incident. It happened two days ago, northeast of the city. The interval between the third and fourth incidents was seventeen days."
"The projected interval was under four weeks."
"Seventeen days. The acceleration is consistent with the interval-reduction pattern but more aggressive than the baseline projection." A pause. "The incidents are getting closer together and the geographic progression is accelerating. If the pattern holds, the fifth incident will be within ten days. The current trajectory puts the fifth incident within the regional boundary of our operating area."
He was still.
"Within the regional boundary," he said.
"Within two hundred kilometers of the city center. The fourth incident was at two hundred and eighty." Tomas was careful. "The profile document has the full analysis. I want you to read it before we discuss contingency options. The profile's conclusions require the full context to interpret correctly."
"I'll read it tonight."
"Tonight." Another pause. "The Fragment Harmony timeline. If Gareth's resistance hypothesis has any validityâif the Harmony creates a system state that's harder to fully absorbâthe timeline matters significantly more than it did a week ago."
"Four fragments. Not five. The Harbor Administrator was this afternoon."
"Four fragments." A mental arithmetic pause on Tomas's end. "At current acquisition rateâ"
"Five to seven days."
The silence that followed was the kind that meant Tomas was looking at a gap between two numbers and not liking what was in it.
"Read the profile tonight," he said.
---
He read it.
He read it at the desk in the Sixth District apartment, the city outside, Maya at her own screen across the room working the board session analysis while he worked through Tomas's documentation.
Tomas had built the threat profile from four sources: Yuki's incident reports, the Association's public documentation of what they were calling "unrelated awakener safety events," the Eastern Hunter Association's seismic monitoring data for mana signature disturbances, and a cross-reference against historical documentation of class absorption phenomena that Gareth had provided from the previous Class Shift case research files.
The profile's central finding was this: the Perfect One was not acquiring classes randomly. The four incidents had a pattern in the class types targetedâeach acquisition had been a class with high mana-processing density. Chronomancers, Arcane Architects, Mana Sculptors. Classes whose core function involved the manipulation of mana at a structural level.
Not combat power. Mana architecture.
"The target selection is deliberate," Tomas had written. "The Perfect One is not maximizing combat capability through acquisition. The Perfect One is acquiring a specific functional profile. The mana-architecture class cluster suggests a target: the ability to interact with, modify, or disrupt a complex mana network structure at the architectural level."
Damien read that line twice.
*The ability to interact with, modify, or disrupt a complex mana network structure at the architectural level.*
At ninety-six fragments, his channel network was among the most complex mana architectures in the country. The combination bridge, the meta-read cluster, the lateral connections between rare class channelsâit was a network that Gareth spent hours studying and still found surprising.
The Perfect One wasn't just coming for Class Shift.
The Perfect One was coming prepared to take it apart.
He set the document down. He looked at the window.
Maya glanced at him from across the room. She'd been watching him read for the last twenty minutes and hadn't said anything, which was Maya's way of respecting that the content needed full attention before it could become conversation.
"Tomas's profile," he said.
"I read it an hour ago." She turned to face him. "The target selection analysis."
"Yes."
"The mana-architecture class cluster is not incidental. The Perfect One is building a toolkit specifically suited to interacting with your network's structure." She was precise. "The question is whether that toolkit is for absorptionâtaking your fragmentsâor for something more specific."
"More specific like what."
"Disruption." She looked at the document on her screen. "If the full absorption mechanism works by establishing a forcible connection to the target's class channel architectureâseizing control of the mana flow and redirecting itâthen a mana-architecture class cluster would give you the tools to make that connection faster and more complete than pure force." She met his eyes. "It's the difference between breaking a door down and picking the lock."
He thought about Gareth's hypothesis. *Fragment Harmony at a hundred is the first threshold where your Class Shift becomes something that might resist a full absorption attempt. Before thatâI'm not certain.*
Before that. He was at ninety-six.
Four fragments away from uncertain resistance.
Unknown distance away from the Perfect One reaching the regional boundary.
"Seventeen days between the third and fourth incident," he said. "Ten days to the fifth, if the pattern holds."
"Seven days margin between Fragment Harmony and the fifth incident's projected timeline," Maya said. "That's not enough margin for comfort."
"No."
She was quiet for a moment. "The board session is in five days. I can't accelerate that timeline. The fragment acquisitionâhow fast can you safely push?"
He thought about the runs. About the Harbor Administrator's twenty-seven minutes. About the combination bridge's stability at four activations.
"Two runs a day," he said. "Not three. The channel load at ninety-six fragments with the combination bridge at full capacityâthree runs would push the network toward interference threshold. Gareth's modeled it. Two per day, with recovery time."
"Two per day is two to three days to Fragment Harmony."
"Best case," he said. "If the acquisitions land. If the entities are in range." He looked at the threat profile on his screen. "I'll push the schedule."
She nodded. She didn't say *be careful.* She never said that. She said: "Yuki needs to widen the monitoring perimeter for the Perfect One's incidents. The geographic acceleration suggests the timeline model may already be out of date."
"I'll call her in the morning."
"After the board briefing." She looked at her screen. "The off-calendar Wells meeting with the fourth persuadable board memberâI have a read from Cho's behavioral pattern analysis. He issued a statement this afternoon about the board's responsibility to balance individual awakener rights with systemic safety concerns. That's Wells's language. He's been turned."
Three persuadable members. Now two.
"What does that do to the vote count," he said.
"If the investigation delay member holds his position and the other two are genuinely undecidedâWells needs both remaining persuadables. She has enough to push the monitoring-without-mandatory-registration outcome, barely, if she can secure them both." Maya made a notation. "The investigation delay member received an outreach from the advocacy group Petra connected to us last week. He's meeting with them tomorrow. That's the best current leverage on that side."
Five days. Two persuadable board members. A Wells ally who'd just shifted.
He thought about how Gareth described managing the three timelines simultaneously.
The board session was political. The fragment count was capability. The Perfect One was threat.
All three were real. None would wait for the others.
He closed Tomas's profile.
"Get some sleep," Maya said. Not softlyâpractically. She turned back to her screen.
He went to the window and looked at the city for a while before he did.
[Fragments: 96 / 1000]
