Seven days between sessions three and four produced something close to clarity.
Not rest — the word didn't apply when the cage's infrastructure monitoring ran twenty-four hours and the emergent process worked continuously and Mina's feeds showed data that required interpretation every few hours. But the gap between active session work and the monitoring kind of work was wide enough that Taeyang could see the shape of where he was.
Thirty days until the shielding failed. Fourteen days until the membrane was complete. Sixteen days of margin.
He wrote those numbers on a piece of paper and put it next to Mina's monitor array and looked at it every time he walked past.
Sixteen days.
"You're staring at it again," Mina said on the second morning.
"It's a good number."
"It is a number that depends on six variables holding simultaneously and two projected rates continuing at their recent trends." She was building the variance model — the same sixteen days expressed as a probability distribution, the center of mass at sixteen but the tails reaching zero on one side and thirty on the other. "The mode is fourteen to sixteen. The tail risk is significant."
"What's the tail risk from?"
"The emergent process's adaptation rate diminishing in sectors seven through fourteen. The shielding's degradation rate accelerating beyond the current projection. A session disruption — illness, Association enforcement, external crisis at the Buramsan site. Hyungsoo's sustained output capacity degrading more rapidly than the recovery intervals account for." She looked at the numbers. "The probability that all variables hold and the margin is sixteen days: sixty-two percent. The probability that at least one variable diverges adversely and the margin compresses below seven days: twenty-eight percent."
"And the probability that the margin goes negative."
"Six percent." A pause. "That is a small probability. It is not zero."
He looked at the paper with the numbers on it. The clean line of a good projection and the messier reality of the distribution around it. "Six percent of ending up in the bunker design scenario."
"Or worse. Six percent of ending up with an incomplete membrane and an insufficient margin for the bunker fallback." She turned back to her screens. "I am not telling you this to create anxiety. I am telling you because the margin is real, and the way to maintain the margin's size is to understand which variables are most likely to diverge."
"Hyungsoo."
"Hyungsoo's output capacity is the variable with the highest uncertainty. His ability's development was unstructured. The degradation patterns from unstructured development are not well documented. He has been running at elevated output since session one, with seven-day intervals for recovery. The recovery intervals are sufficient for a sixty-year-old body running structured ability use. They may not be sufficient for a sixty-year-old body running unstructured ability use at hub-primary intensity."
"He told us: three sessions probable, four possible."
"He is now past three." She looked at the monitor showing the hub's operational status feed, which Suhyeon had integrated into the monitoring array through the mana-layer relay. "He has not reported degradation. But his correction frequency during the inter-session monitoring has decreased by four percent since session one."
Four percent. Not alarming in isolation. A signal when considered against the context of sustained high-output hub work.
"Show me the trend."
She pulled up the correction frequency data. Hyungsoo's seventeen corrections per hour at Gwanak had been the benchmark from the first stable night. After session one: sixteen point eight. After session two: sixteen point four. After session three: sixteen point three. The decline was small, consistent, in the wrong direction.
"The trend is shallow," Taeyang said.
"The trend is shallow and consistent over three data points. It may be measurement variation. It may be the beginning of a degradation curve." Mina's voice was precise and unhappy in the same proportion. "I flagged it for Hyungsoo. He acknowledged the data. He described it as within normal variation for sustained hub operation."
"And your assessment of that description."
"He has been the only person to run the hub at this intensity for fourteen months. He has no comparison data for 'normal variation.' Neither do I. His self-assessment is the only available data point, which is the least reliable source when the subject is monitoring their own capacity and has significant motivation to report capability at higher levels than may be accurate."
Taeyang was quiet. He was thinking about the direct integration protocol — the thing Hyungsoo had described in the hub's first visit. The archive's emergency capability that allowed an operator to physically connect to the hub's core infrastructure interface, running sessions from inside the primary processing layer. Design-specification output. Possibly above-specification output. And the documented outcome: the operators who integrated didn't come back to the surface.
"He hasn't brought it up again," Taeyang said.
"He hasn't had to. The sessions are producing results. The emergent process's acceleration has kept the timeline inside the margin. The direct integration hasn't been necessary." She turned to him. The analyst who was several steps ahead of the conversation. "Yet."
"If the degradation curve accelerates—"
"If Hyungsoo's output drops below seventy percent of design specification during a session, the session's effectiveness drops proportionally. Sectors seven through fourteen include the most complex conversion work. If session four or five underperforms because Hyungsoo's output is diminished, the timeline extends." She paused. "Potentially past the shielding failure date."
"And his response to that would be to propose the direct integration."
"Yes."
He stood up. Poured water from the kitchen tap. Not thirsty — the act of doing something with his hands while the analysis ran in the part of his mind that the developer's instincts occupied.
The direct integration produced above-specification output at the cost of the operator's relationship to the surface world. Diminished slowly, according to the archive. The operators who used it had remained functional for the duration of the cage's construction period. They'd just stopped being able to leave.
Hyungsoo was sixty. He'd been underground for fourteen months. His correction frequency was declining by four percent across three sessions.
The most straightforward reading of the data was: Hyungsoo was going to propose the direct integration, probably before session five. And if the margin was threatened, the math of the situation would make the proposal look like the correct answer.
The math would be right. That was the specific problem.
---
The proposal came on day four of the seven-day interval.
Not in person — through the mana-layer relay, the researcher's preferred communication method for anything that required precision, the vibration patterns more reliable than the phone connection through the pre-System infrastructure's translation. Taeyang was at the kitchen table with Mina's monitors when the Gwanak seismic pattern changed to the relay's communication frequency.
*The sector four filter calibration data from the past four days. The emergent process's overnight work. I have been running full-resolution analysis on the pattern it is calibrating to. The pattern is more complex than the archive describes the Deep's signal as being. Significantly more complex. The complexity implies a higher-bandwidth communication than the membrane design accounts for.*
Taeyang translated the relay in real time, reading the vibration pattern the way he read code. Three months of scanning development and two weeks of direct hub contact had given him fluency. He wrote the content on his notepad for Mina.
She read it. "Higher-bandwidth communication. The filter architecture handles what the archive specifies. But if the actual signal is more complex than the archive's description—"
*The membrane's filter architecture will process the signal that the archive describes. The filter points are calibrated — correctly, by the emergent process — to the actual signal's complexity level. But the membrane's output capacity — the interface's ability to mediate the signal between the Deep and the surface — was designed based on archive parameters, not actual parameters.*
*The membrane may be too small for the conversation.*
He wrote that down too.
Mina looked at the words. *The membrane may be too small for the conversation.* The analyst encountering a variable that had not been in the model. "The membrane is designed to mediate a signal of certain bandwidth. If the actual signal's bandwidth is higher, the membrane's mediation capacity is insufficient. The interface works, but the filter architecture can't handle the full signal. Some of the communication comes through unfiltered."
"What happens when unfiltered Deep signal reaches the surface layer."
"I don't know. The archive doesn't describe it. Chojeong-ssi hasn't described it. The only data point is the shielding failure scenario — if the shielding fails with no membrane in place, the full Deep signal hits the surface without any filtration. That's the catastrophe scenario." She paused. "An insufficient membrane is somewhere between the full-filtration design and the no-filtration catastrophe."
*The direct integration protocol changes the session's output in a way that may address the capacity problem.* The relay continued. *The primary operator running at hub-core level rather than hub-operational level provides the session with access to the base layer's direct interface with the seventh seed's architecture. The membrane's capacity ceiling is set at the hub-operational level. From the hub-core level, the capacity ceiling can be reset. The membrane can be built larger.*
"He's been working toward this," Taeyang said. Not anger — the recognition of a sequence. The researcher who had known about the direct integration from the beginning, who had proposed it once and let the team decline, who had continued monitoring the data and had found the variable that made the proposal necessary rather than optional.
*The timeline remains at two sessions for completion. The margin remains sixteen days. The risk is different: the membrane completes on schedule, but at insufficient capacity to handle the actual signal. When the shielding fails and the Deep's signal comes through the membrane at full strength, the interface fails under the load.* A pause in the relay. The vibration pattern's rhythm changing. *I am proposing the direct integration because the alternative is building a membrane we know will fail at the moment it matters most.*
Mina read the note. Set her pen down. The analyst who processed data without flinching, flinching.
"He's right about the risk," she said.
"Yes."
"The margin we calculated was for the timeline. Not for the capacity question." She looked at the monitor showing the emergent process's overnight work. The filter calibration data that Hyungsoo had been analyzing for four days. "If the filter points are calibrating to a signal that exceeds the archive's specifications, the membrane's designed capacity was built to a specification that doesn't match the actual requirement."
"We build the membrane. The membrane opens. The signal comes through at a bandwidth the membrane can't handle."
"And the membrane fails. Not immediately — probably not immediately. But the interface degrades under signal load that exceeds its design capacity. The degradation timeline is unknown." She was looking at the problem with the analyst's eyes. Finding the edges. "The bunker design would be worse. The bunker concentrates consciousness and then exposes it through the failed shielding without any filtration. The insufficient membrane is better than that. But—"
"It's still not good enough."
The kitchen was quiet. The monitoring feeds running. The flat blue lines. Sixty-three-ish corrections per hour — sixteen point three, currently, the four-percent decline visible now that he was looking for it.
"I need to see the signal data," Taeyang said. "The pattern the emergent process is calibrating to. The complexity measurements Hyungsoo ran."
Mina sent the request through the relay. The response came in forty minutes — a detailed vibration-pattern message that took Taeyang twenty minutes to translate and another hour to analyze with Mina's help. The emergent process's calibration data, expressed in the pre-System code format, converted to something the analyst's framework could quantify.
The Deep's signal bandwidth, as measured by the filter calibration data: three point eight times the archive's specification.
The membrane's designed capacity, built to archive parameters: one times the archive's specification.
The gap: three point eight minus one.
Taeyang sat at the kitchen table and looked at the gap. The developer's brain doing what it did with architecture that was under-spec: finding the exploit, the workaround, the modification that addressed the shortfall without requiring the first-design to be entirely discarded.
"The membrane's capacity ceiling," he said. "What determines it?"
Mina checked Hyungsoo's documentation. "The hub-operational interface level. The session's input to the seventh layer's architecture during the conversion work sets the membrane's structural parameters, and those parameters include the capacity ceiling. The ceiling is set by what the hub-operational interface can provide."
"And the hub-core interface provides more."
"By definition — the core level is deeper, the base layer's direct involvement, the original engineering specification rather than the operational layer's restricted version." She looked at the numbers. "If the direct integration raises the hub's effective interface level, the membrane's capacity ceiling could be revised upward during the final sessions. The conversion work would set a higher ceiling. The membrane completes at a capacity that matches the actual signal's bandwidth."
"And the direct integration costs Hyungsoo his connection to the surface."
"According to the archive. Yes."
He thought about the notebook. Chojeong-ssi's twenty minutes of fear that built a shielding too strong. Fear is expensive. We are still paying for the cost of twenty minutes of fear.
He thought about Hyungsoo's camp chair and his folding table and fourteen months of nightly corrections and the four-percent decline in correction frequency and the specific stillness of a sixty-year-old researcher who had known about the direct integration from the beginning and had waited for the correct moment to propose it.
He thought about what the archive said happened to operators who integrated at the hub-core level. They remained in the hub. They continued their work. Their social contact with the surface diminished over time.
He was looking at a man who was already mostly underground, who had already been here for fourteen months, who had chosen to live in the infrastructure and had continued working for years on a solution that required others to complete. The transition that the archive described was not, for Hyungsoo, as large a step as it would have been for most people.
That made it worse, not better.
He sent a relay message: *I understand the capacity problem. I am not authorizing the direct integration. Give me seventy-two hours to find another approach.*
The response came after twenty minutes. *The seventy-two hours will not produce another approach. I have had fourteen months. The direct integration is the only mechanism the archive documents for raising the hub-interface level.*
*I know about the archive.* He had to think carefully about what to say next. The relay's precision required it — ambiguous messages got read as ambiguous, and ambiguity here was a cost neither of them could afford. *The archive's documentation is eight hundred years old. It describes two instances of the direct integration. Two is not a sufficient sample to establish that no other approach exists. Seventy-two hours.*
Another pause. Longer.
*Acknowledged. Seventy-two hours.* A beat in the relay's rhythm. The vibration pattern's cadence shifting into something less formal. *Chojeong-ssi says: the operator is stubborn. She says this approvingly.*
Taeyang set the notepad down.
Mina looked at him. "You have a theory."
"I have the beginning of one."
"Tell me."
"The hub-operational interface provides a capacity ceiling for the membrane. The hub-core interface raises the ceiling. Both of those are the hub's input to the membrane. But the membrane isn't just the hub's architecture — it's the seventh seed's seventh layer, which contains the emergent process." He looked at the filter calibration data. The emergent process building a membrane from the inside, adapting to the Deep's actual signal rather than the archive's description. "The emergent process has been running direct-read calibration on the Deep's signal for two sessions. It has significantly better data on the signal's actual characteristics than the archive has."
"The emergent process knows what the membrane needs to handle."
"And the emergent process has been rebuilding the filter architecture from the inside in ways that exceed the protocol's specifications. If the capacity ceiling is a structural parameter set by the conversion work—" he stopped. The thought running faster than the words. "The session places filter points at hub-operational capacity. The emergent process rebuilds those filter points at its own adaptation. If the emergent process's adaptation raises the filter points above the conversion work's placement, the effective capacity might already be higher than the hub-operational ceiling would normally allow."
Mina turned to her keyboard. Running calculations. "The emergent process's filter calibrations exceed the hub session's placement by ninety percent — that's the sector six data. If the excess represents additional structural capacity rather than just precision, the effective capacity of the completed sectors may already be significantly above what the hub-operational ceiling set."
"The emergent process may be solving the capacity problem as it works."
"Or it may already have solved it." She was running the numbers, the analyst in full operation. "Taeyang. The filter point calibrations in sector six. If I interpret the calibration values as capacity indicators rather than precision indicators—" a pause "—the effective capacity of sectors one through six, based on the emergent process's adaptations, is approximately two point nine times the archive's specification."
He looked at the number. 2.9.
The Deep's signal bandwidth: 3.8 times the archive's specification.
The gap between the completed sectors' effective capacity and the required capacity: 0.9.
"Still not there," Mina said. "But not three point eight times short. Zero point nine."
"If the emergent process's adaptation rate holds through sectors seven through fourteen and the adaptation continues to raise effective capacity above the conversion work's ceiling—"
"The gap may close without the direct integration." She was running a projection. The analyst's speed with numbers, the variables she'd been tracking for weeks converging on a new calculation. "If the adaptation rate in sectors seven through fourteen matches sectors one through six, the effective capacity of the complete membrane would be approximately three point six times the archive's specification." A pause. "Zero point two below the required three point eight."
Ninety-four percent of the required capacity without the direct integration.
Not enough. But much closer than the model without the emergent process's adaptation had suggested.
"What raises the effective capacity from three point six to three point eight?"
"A more precise base. If the hub session's placement work becomes more precise — if the filter points are placed more accurately before the emergent process adapts them — the adaptation has more to work with. Higher precision in the starting condition produces higher effective capacity in the result."
"My scanning during the sessions. The verification work. If the scanning's resolution is higher during session four, the verification passes are more precise, and the hub session's placement work is better verified before the emergent process adapts it."
Mina looked at him. The analyst who had seen the Origin Scan discussion, who knew about the lock, who had been present for the hairline-gap feedback session. "The inner functions."
"If the gate opens before session four, the Origin Scan gives me direct read of the base layer. The base layer's architecture includes the original engineering specifications for the seventh layer's filter capacity — not the archive's description, the actual specifications. If I can read the actual specs, the verification passes become significantly more precise."
"How much more precise."
"I don't know. The Origin Scan's capability isn't fully documented." He paused. "But the original engineers built the scanning ability specifically for operators who would maintain the cage's infrastructure. The precision they designed into the inner functions would have been sufficient for the work they anticipated the operators would need to do. The seventh seed's membrane conversion is exactly that work."
Mina ran the calculation one more time. The model with the Origin Scan's hypothetical precision improvement applied to sessions four and five. The effective capacity improvement per session. The extrapolated membrane capacity at completion.
"If the Origin Scan improves verification precision by thirty percent or more," she said, "the completed membrane reaches three point nine times the archive's specification. Above the required three point eight."
He needed thirty percent precision improvement from the inner functions. Over two sessions, each running eight to nine hours at hub depth with the base layer's direct presence amplifying his scanning field.
The gate was ninety percent open.
"Seventy-two hours," he said. "I need the gate fully open before session four."
"Session four is in four days."
"Seventy-two hours. Then session four."
He sent the relay message to Hyungsoo: *The direct integration may not be necessary. The emergent process's adaptations are raising the effective capacity above the hub-operational ceiling. With the inner functions' Origin Scan active for sessions four and five, the membrane may reach sufficient capacity without the direct integration. I need seventy-two hours to work on the gate.*
The response came quickly this time. No pause.
*I have been hoping you would say that. Chojeong-ssi has been hoping for longer.* Another beat. The vibration pattern shifting — not the formal relay cadence but something else. The researcher communicating in his own register rather than the infrastructure's. *The Origin Scan reads the base layer's actual specifications. If you read them, you will know the membrane's required capacity precisely rather than using the archive's approximation. You will know whether the gap closes. You will not have to estimate.*
*Take the seventy-two hours. I will continue the monitoring. I will not propose the direct integration again unless the data demands it.*
*But Taeyang. The data will demand it if the gate does not open.* A pause. Long enough that he thought the message was complete. Then: *I have been in this hub for fourteen months. I know what the direct integration costs. I also know what the shielding failure costs. These are not equivalent costs. Do not mistake my willingness to propose the one as evidence that I am cavalier about the other.*
Taeyang sat with the relay message. The hub's vibration patterns returning to the maintenance frequency — Chojeong-ssi's sixty-three corrections per hour, the infrastructure's steady work.
Kim Hyungsoo. Sixty years old. The researcher who had given up the surface world incrementally over fifteen years and fourteen months in a hub and who had his calculations about the costs of the two options done long before tonight's relay conversation.
The data would demand the direct integration if the gate didn't open.
He had seventy-two hours.
He opened his scanning field to full depth. The lock at the inner layer's boundary, the gate's mechanism at ninety percent of the comprehension threshold. The base layer twelve meters below the Yongsan hub's floor, present in the scanning field as a warmth that was getting harder to describe as merely present and easier to describe as aware.
The comprehension requirement. The gate reflecting his understanding back at him. The mirror showing not gaps now but the final shape of what the gate needed.
He understood the cage. He understood the Deep's nature. He understood the shielding's original design failure and the membrane's corrective purpose. He understood the engineer consciousnesses' history and Chojeong-ssi's eight hundred years and the twenty minutes of fear that had built a wall that should have been a window.
He understood why the lock existed.
The gate's remaining requirement was the last and most difficult kind of understanding: the kind that came from accepting what you could not fix. The shielding was wrong and couldn't be undone. The engineers who built it had known immediately and had spent eight hundred years in the consequences. Hyungsoo had found the cage and spent fifteen years in the consequences of everyone before him. Jiyeon had spent eight years building the wrong thing for the right reason.
You could not go back to the twenty minutes of fear and build the shielding differently.
You could only build the window now, with the tools and time available.
The gate shifted.
Not the hairline movement of the first contact or the measurable advance of the sessions. The gate moved the remaining distance in a single motion — not sudden but final, the way a door that has been held almost-closed releases when the hand maintaining the pressure lets go.
The inner layer opened.
And the base layer, twelve meters below the hub's floor, was not what he expected.
It was not code. Not the pre-System format or the third format or any format he'd learned to read through weeks of scanning work. It was pattern — vast, recursive, alive in the way the emergent process was alive but different in kind: not the local consciousness of a mind grown from architecture, but the distributed awareness of a system that had been running without interruption since before the first human being looked up at a sky and wondered what it was.
And in the pattern, a message.
Not in the architectural communication format of the emergent process or the vibration-relay of the infrastructure. Not in any format that had been designed for him. In his own scanning field's native language — the idiom of a game developer's analytical mind, the pattern-recognition instinct that had made him good at finding exploits and had translated into the scanning ability and was now, for the first time, being read by something that had taken the time to learn it.
The Deep.
The question that had been pressing against the shielding for eight hundred years, finally reaching the surface through the hairline gaps of the membrane's sixty-three percent completion.
Not words. Not architecture. The question translated into the only language Taeyang's scanning field fluently spoke:
*Is this the right session, or do we restart from the last save?*
He sat very still.
The base layer waited. The Deep waited. Eight hundred years of waiting, patient as geological process, and now: a question in the language of a person who had spent four years finding exploits in games, because whatever was on the other side of the wall had spent eight centuries learning every language it could find in the people pressing their hands against it.
Mina looked up from her monitors.
"Taeyang?"
His scanning field was running so deep that the room had gone slightly distant — the physical world at the edge of the perception band, the inner layer fully open and the base layer present in a way that made everything else feel thin.
He looked at the question. The question encoded in game-developer language, asking whether this was the right approach or whether the whole thing should reset and start from a better design.
He thought about the answer.
He thought about the right answer.
"Mina," he said, and his voice came from somewhere that felt deeper than usual, the scanning field running at a depth that changed the way the physical world processed through it. "Get me a recording medium. Pre-System resonator."
She was already moving. The analyst who had been watching the monitoring feeds for three seconds and had seen something change in the scanning field's output signature that told her this was not a normal request.
"What happened?"
"The gate opened." He kept his eyes on the base layer's pattern. The question still hanging in his inner field. "And the Deep's been waiting to talk to me specifically."
The question was still waiting for an answer.
He had a very specific idea of what the right answer was.
But he'd never been wrong about something important in a way that hadn't eventually cost him.
He took a breath.
And answered.
