Mira had not slept.
He could tell from her station — the arrangement of coffee cups (three, graduated leftward, each one further from the keyboard than the last as caffeine lost its contest with exhaustion), the density of notation on her secondary screen (a full night's worth of calculations, the earliest time-stamped at 0134), and the particular quality of her focus when she turned to face him. Mira without sleep was sharper, not duller. The filter between data and speech thinned. She said what she was thinking at the speed she was thinking it.
"The cost discrepancy isn't an error in Pell's formula," she said. "It's an environmental variable he didn't account for because it didn't exist in his environment."
Voss pulled a chair to her station. 0630. The intelligence center was running its morning shift change, the overnight analysts handing off to the day team. He had slept four hours, which was two more than usual during an active operation and two fewer than Ryn had prescribed. His right hand was warm. Both fingers. The weave's restoration held through the night.
"Show me."
Mira pulled her calculation log to the center screen. "Pell operated in the pre-sealing period. The doorway network was fully active. The dimensional substrate was at natural density — the Loom's organizing radiation saturated the physical dimension's structure the way it had for eons. Pell's restoration costs were calculated in that environment. In a healthy substrate."
She drew the analogy he was already building. "The tissue around the repair site matters."
"Exactly. When a surgeon stitches a wound in healthy tissue, the surrounding cellular matrix supports the repair. Blood supply is intact. The immune response is functional. The stitches hold because the tissue around them is strong enough to bear the load." She tapped the calculation log. "When a surgeon stitches a wound in damaged tissue — burned, irradiated, compromised — the surrounding matrix can't support the repair the same way. The stitches have to do more work. The surgeon has to use more material. The cost per centimeter of closure is higher because the environment around the repair site is weaker."
"The substrate is the damaged tissue."
"Six hundred years of degradation. The sealing of the original network cut off the Loom's radiation to large portions of the dimensional fabric. The thread-density at sealing sites dropped forty percent. The network we reactivated is healing that damage — Nira Sol's thirteen-point-seven percent increase per active node. But the healing isn't complete. The ambient substrate across the dimension is still at roughly sixty to seventy percent of its historical density. We're stitching in damaged tissue."
He processed the numbers. "So the cost multiplier is proportional to the substrate deficit."
"Inversely proportional to the substrate density, more precisely." She had the formula on screen — her own derivation, built overnight from the weave test data and the network's substrate measurements. "At sixty percent substrate density, the restoration cost per unit volume is approximately five-point-three times the cost at one hundred percent density. At seventy percent, it drops to about three-point-eight times. At ninety percent, one-point-five times. At full density, it converges on Pell's original figures."
The implications assembled. "The cost will decrease as the network heals the substrate."
"Yes. Every node that stays active, every month the network operates, the substrate density increases. The cost of structural reinforcement drops with it. In three years, if the network reaches full activation and the substrate recovers to ninety percent or better, the weave would cost roughly one-point-five times Pell's numbers. Manageable. Sustainable."
"Three years."
"Three years." Mira's voice carried the specific flatness of someone stating a timeline they knew was too long. "At current Gradient incursion rates, the network will not survive three years of sacrifice-and-isolate management. Each Gradient takes eight to twelve percent of the network. Two Gradients in a month. If the rate holds or increases, the network is gone within a year."
---
The math on the center screen was clean. Devastating, but clean.
A four-person weave restoring a single drained node at current substrate density: five percent Thread Sight reduction per Carver. Four Carvers spending five percent each to bring back one node.
Thirty-two drained nodes in the first Gradient's isolation zone. To restore all of them: each Carver would need to spend one hundred sixty percent of their Thread Sight capacity. Impossible. You could not spend more than you had.
With a rotation of Carvers — multiple weave teams, each taking a portion of the restoration — the math improved but did not solve. Eight functional Thread Sight Carvers in the Corps. Four in a weave at a time. Two teams, alternating. Each team restores sixteen nodes. Each Carver spends eighty percent of their Thread Sight.
Eighty percent. They would be functionally blind afterward. Months of recovery, if recovery was even possible. And the second Gradient hadn't been addressed yet.
"We can't restore the full isolation zone," Voss said. "Not at current costs."
"No. But we don't need to restore all of it." Mira switched to the network topology. "The isolation zone includes thirty-two nodes, but not all of them are equally important. The critical ones are the junction nodes — the ones with multiple connections to the broader network. There are nine of those. The rest are endpoint nodes with single connections. If we restore the nine junction nodes, the network can rebuild the endpoint nodes naturally over time through standard energy distribution."
Nine nodes instead of thirty-two. Each Carver in a four-person weave spending five percent per node. Nine nodes at five percent each: forty-five percent total per Carver.
Still devastating. But survivable. Barely.
"Forty-five percent," Voss said.
"With two teams alternating, twenty-two-point-five percent per Carver." Mira paused. "That's the loss equivalent of two years of natural Thread Sight aging. Significant but not incapacitating. The Carvers would retain functional Thread Sight. Their range and resolution would be reduced. Recovery timeline: unknown. Pell's notes don't address whether Thread Sight capacity regenerates after weave expenditure."
He filed the unknown. Another variable they would have to learn by doing.
"And the second Gradient?"
"Same calculation. Different network topology. I'd need to model the eastern network to identify the critical junction nodes, but the principle holds. Restore the junctions, let the network rebuild the endpoints." She paused. "Assuming we have Carvers to send. The eastern region has no Carver Corps presence. Our eight Thread Sight carriers are all here."
The resource constraint. The weave required trained, frequency-capable Carvers. They had eight. The dimension was three thousand kilometers wide. Even with transport, deploying a weave team to the eastern network would take time they might not have.
---
Dex came through the door at 0714, carrying two cups of coffee and a file folder.
He set one cup on Mira's desk without asking — he had learned her coffee preferences during the months of working alongside the intelligence center, the way a good soldier learned the terrain. The other cup he held.
"I've been running numbers," he said.
Voss looked at him. Dex running numbers was still a novelty. The man who had once solved problems exclusively through applied violence had spent months in the archives developing a relationship with data that was functional if inelegant.
"The weave test used four Carvers," Dex said. "Cost was zero-point-four percent each. Pell's formula projected zero-point-zero-seven-five. You're explaining the gap through substrate density." He looked at Mira. "Did you check whether the gap narrows with more participants?"
Mira's hands stopped on the keyboard. She turned. "Explain."
"Pell's notes describe weaves of four to eight. He says eight is optimal. He doesn't say why eight is optimal — just that it is. What if the resonance link gets more efficient with more participants? What if eight Carvers in a weave don't each pay half the cost of four Carvers — what if they each pay a third? Or a quarter?"
Mira stared at him. Then at her screen. Then at the calculation log she had built overnight.
"I didn't test for nonlinear scaling," she said. The words came out clipped, annoyed, directed at herself. "I assumed linear cost distribution — double the participants, halve the per-person cost. Pell's notes don't specify the scaling function."
"Because Pell didn't have the instruments to measure it," Dex said. "He said eight was better. He didn't say how much better."
"If the resonance link has a cooperative efficiency component — if the shared vibrational pattern reinforces itself as more participants join, reducing the energy loss per connection point..." Mira was already building the model. "It's like a load-bearing arch. Two stones support weight. Four stones support more than twice the weight of two because the arch distributes the load through mutual compression. Eight stones in a full arch can support loads that no individual stone could bear."
"The weave is an arch," Dex said. "More Carvers, stronger arch, less load per person."
Mira's fingers moved across the keyboard at a speed that made the keys blur. "I need to test this. We need an eight-person weave. If the scaling is nonlinear — if eight Carvers pay, say, one-point-five percent each instead of two-point-five — the restoration math changes completely."
"We have eight Thread Sight Carvers," Voss said. "But only four have achieved the target frequency."
"Then train the others." Dex set down his coffee. "Holst and Kira are close. Push them. And the three who can't reach the frequency through standard training — find another way. Modified protocols. Different breathing sequences. Whatever it takes."
He was right. The eight-person weave was the variable that could change everything. If the cooperative efficiency held — if the arch analogy was even approximately correct — the restoration costs might drop to levels that made the system workable.
"Start today," Voss said to Mira. "Design the test protocol for eight participants. I'll accelerate the frequency training for Holst and Kira."
"What about the three who can't reach the target?"
"I'll work on that. The resonance model says the substrate signal determines the threshold. If we train near an active node — where the substrate is densest — the signal might be strong enough to push borderline Carvers past the ceiling."
Mira nodded. She was already building the eight-person test protocol, her coffee untouched, the overnight fatigue forgotten in the momentum of a new model to build.
---
The comm alert came at 0803.
Not a call. A priority flag from the Dragon Bone Island relay. Nira Sol requesting an emergency meeting. The request carried a marker he had not seen before — a thread-modulation identifier that Mira's translation system tagged as "urgent, architectural concern, new information."
He was at the arch within ninety minutes. Transport from the mainland, landing on the island's southern pad, walking to the coastal position where Nira Sol waited. She was not at her usual four-meter distance. She was at two meters. The closest she had ever stood to him.
Her thread-architecture was in a configuration he had seen only once before — the day she had described what happened to unmanaged dimensions. The cognitive threads pulled tight. The communication channels narrowed to single-output pathways. She was compressed around information that required careful handling.
"The second Gradient," he said.
*Yes.* Her thread modulation was dense. Compressed. More information per signal than her usual communication style. *I have consulted with the network architects. They have analyzed the movement patterns of both Gradients — the one approaching your bait node and the one detected in the eastern network. The analysis is complete.*
"What did they find?"
*The eastern Gradient is not a new incursion.*
He stood still.
*A new incursion would enter the network through a boundary point — a location where the inter-dimensional space interfaces with the network's outer edges. The eastern Gradient did not enter through a boundary point. It appeared at a location internal to the network. Specifically, at junction node 6-31 — a high-connectivity node in the network's central corridor, approximately fifteen hundred kilometers from the metropolitan region.*
Junction node 6-31. He knew the network topology well enough to place it. A major hub in the transcontinental connection architecture. Multiple pathways converged there.
"The first Gradient passed through junction 6-31," he said. He had reviewed the trajectory data. The first Gradient's path from the northwestern wilderness to the metropolitan area ran through the central corridor. Through junction 6-31.
*Yes. And when it did, it divided.*
The word sat in the coastal air.
*The first Gradient reached junction 6-31 approximately six days ago. The junction presents multiple high-energy pathways — connections running east, south, and west. The Gradient followed the western path toward your metropolitan network. But at the junction point, a portion of the Gradient separated and followed the eastern path. The portion that went east is what your monitoring system detected as a second incursion.*
"It split."
*It propagated.* Nira Sol's threads carried the frequency he had read when she said "Gone." The low, old vibration from beneath the cognitive layer. *Gradients in low-density substrates do not split. The energy differential is not sufficient to sustain two zones simultaneously. In a high-density network with multiple high-energy pathways, the conditions are different. A Gradient encountering a junction with multiple viable paths can divide and pursue both paths independently. Each fragment is smaller than the original but capable of growing through further feeding.*
He ran the network topology in his head. How many junction nodes were there in the full network? Mira had cataloged them. The answer came immediately. Fifty-three high-connectivity junctions across the continental network.
"How many times can it split?"
*At every junction where the energy conditions support division. The fragments that result are fully independent. Each one feeds, grows, and can split again at the next viable junction.*
Fifty-three junctions. A Gradient that could propagate at each one. Not two feeders. A branching network of feeders, multiplying through the infrastructure that was supposed to protect the dimension.
Not two problems. One problem that had learned to breed.
*This behavior has been observed before*, Nira Sol sent. The low frequency was sustained now, running beneath every signal she transmitted. *In dimensions with dense, high-energy networks. It is the terminal scenario. The one that produces the outcome I described.*
Gone.
He stood on the shore. The arch glowed behind her. The Loom visible through it, vast and coherent and under siege from something that had no mind and no malice and no vulnerability that anyone had found in eons of trying.
"How long?" he asked.
*At current propagation rates, with the network at its present energy density, a single Gradient fragment requires approximately two weeks to reach the next viable junction and divide again. The growth is geometric. Two become four. Four become eight. Within three months, every major corridor in the network will contain an active Gradient fragment.*
Three months. Not three years. Three months.
The race between restoration and decay had just changed its terms.
