Denavi Stays

Greel posted the rotation orders at 0700, printed on the thin paper-stock the unit used for anything that needed a physical record. He handed out copies without ceremony, which was his way with everything administrative. Denavi read hers standing up, her back against the interior wall, and when she finished she folded it in thirds and held it without putting it away.

Three names on the rotation list. Denavi, plus Bael and Sorrith, the two soldiers who'd been on line since before Derek arrived. All three cleared to extract to the staging station on the evening transport, which would leave the unit with the two newer soldiers plus whichever of the originals chose to stay.

Greel was at his workstation when Denavi knocked on the frame of his partition. I was across the room pulling charge packs from the supply crate and I wasn't trying to listen, but the forward position's interior didn't have enough square footage for privacy.

"I'd like to request a rotation delay," Denavi said.

Greel looked up from the workstation.

"Forty-eight hours minimum. Through the next engagement." She set the folded orders on his workstation without dramatic effect. "The unit has two soldiers who've been in this terrain for three weeks. Hetch has the east approach familiarized and Rovan knows the interior node layout, but nobody on that list knows the north quadrant the way I do. Not since the engagement last month."

Greel picked up the orders, looked at them, set them back down. "The north quadrant matters because?"

"Because if I'm wrong about the north, Derek's right about the north, and if Derek's right about the north, we're going to need someone who can move through it at speed in the dark without checking the map." She said it plainly, the way she said most things. "That's me."

Greel considered this for about four seconds. "Approved. I'll amend the orders." He pulled the orders off the workstation and wrote something on the margin. "Bael and Sorrith on the 1800 transport."

Denavi picked up her copy, nodded once, and walked back toward the supply area. She glanced at me as she passed, not asking for a reaction, just registering that I'd heard.

I pulled the next charge pack and checked its level. Eighty-three percent. Good enough.

The terrain argument was accurate. Everything she'd said to Greel was true and operationally sound, and I knew it was also not the complete reason she'd asked to stay. She knew what was coming as well as I did, she'd been in this sector for four months, and she understood that the engagement she was staying for was not going to be a probing contact. She'd made her decision the way she made most decisions: by framing it in terms she could defend in a command conversation and then acting on it.

I didn't say anything about the other reason. It wasn't my place, and she hadn't asked.

Bael and Sorrith shipped out on the 1800 transport. The unit that remained was five people: Greel, Denavi, Hetch, Rovan, and me. Durrik was technically present as an observer, which in practice meant he occupied a position near the interior's secondary entrance and watched everything and said very little. His presence in the forward position instead of the staging station was his own choice, not a Command order, and I'd registered what that meant when he'd shown up three days ago with his gear and the look of a man who had decided where he needed to be.

The twenty-four hours after Denavi's rotation request were preparation in the specific sense: not dramatic, not rushed, just the systematic work of making a position as defensible as a five-person unit could make it before an engagement they knew was coming.

I started on the forward position's northern approach at 0900, after three hours of sleep that the predictive extension had run through at standby without waking me. The north quadrant had two natural chokepoints that Denavi had described to me over the past two weeks: a gap between a collapsed structural section and a stabilized wall remnant, maybe four meters wide, and a drainage channel that ran east-west and created a natural depression that provided cover but also channeled movement. The gap was the obvious problem. The channel was what I wanted to use.

Denavi was already at the north quadrant when I arrived. She had a measurement tool and was doing something to the wall remnant's base, checking the structural integrity, which was the kind of preparation that came from months of making do with what the terrain gave you.

"Gap's four-two," she said without looking up. "I measured it last week but I wanted to confirm after the north probe. They ran two gifted soldiers through the adjacent section and I wanted to know if the approach changed the surface stability."

"Did it?"

"No. Surface is the same." She stood up and looked at the gap. "Standard coverage won't hold this if she pushes gifted soldiers through first. They'll have the crossing done before a standard rifleman can cycle."

"I know."

"So we need overlapping coverage with a short trigger window." She turned to look at me. "Which means positioning one of us on the east side of the gap before the engagement opens, not during."

She had already worked out the same geometry I had. I'd thought about the north approach twice during the night, running the predictive extension's spatial modeling against the physical layout, and I'd come to the same conclusion about early positioning. Two seconds of lead time from the predictive extension was enough to cycle a trigger if the position was already established. If someone was moving to position when the crossing started, two seconds wasn't enough.

"Rovan can hold the east side if we brief him on the trigger window," I said.

"I thought Hetch."

"Rovan's faster under load. I watched him in the second probe, when we were tracking the northwest approach." I thought through the movement sequence. "Hetch is steady but he takes a half-second to read the situation before he moves. Rovan moves first and reads on the way. For a short trigger window, Rovan's the better fit."

Denavi considered this. Her eyes moved to the gap and then back. "Rovan on the east side. Hetch on the drainage channel coverage."

"Hetch in the channel gives him a crossfire angle on anything that clears the gap." I walked the approach geometry in my head. "If Rovan can't hold the east side crossing, Hetch in the channel catches them while they're still in the gap's exit zone."

"And one of us needs to be mobile." She said this as a statement, not a question. She already knew the answer.

"One of us reads the engagement as it develops and redirects coverage where it breaks." I looked at the gap and then at the channel's eastern end. "That's you or me."

"You," she said. "Your read is faster. I'll hold the secondary position."

This was the same logic she'd applied in the previous engagements, and it was correct. Her terrain knowledge made her more effective in a fixed position than I was, because a fixed position in this terrain required knowing which surface to trust and which would shift underfoot at the wrong moment. I was faster than her on open-ground movement and I could redirect on three seconds' notice without stopping to read the terrain. We each knew what the other was worth.

We spent the next two hours briefing Hetch and Rovan on the north approach positioning, walking them through the physical layout at walking pace and then at move pace, and I ran a terrain familiarization that Denavi conducted and I translated into the pattern recognition's coverage geometry. The predictive extension registered both of them moving through the north quadrant and built positional models. By the end of the second hour, I had their movement signatures indexed.

Hetch asked good questions and moved with the kind of deliberate care that came from wanting to get it right more than wanting to look competent. Rovan asked fewer questions and moved faster, his body reading the terrain in the way bodies do when they're naturally suited to close physical spaces. Both of them were paying attention. That was what mattered.

After the north quadrant, I spent an hour cross-training Rovan on the integration communication protocol Denavi and I had developed, the shorthand signals she'd built from four months of this terrain and I'd adapted to work with the predictive extension's output. The protocol was simple on the surface and specific underneath: twelve hand signals covering the variables that came up most often in close-terrain engagements, with a layered system for range and urgency.

Rovan learned the twelve signals in forty minutes. He had the kind of procedural memory that absorbed operational information and held it without needing repetition, which was a specific type of competence worth knowing about before a fight.

At 1400 I found Sonn's link in the equipment bay and ran the diagnostic.

Sonn answered inside three seconds, which meant they'd been close to the terminal.

"I scheduled this for 1600," Sonn said.

"I moved it up. Greel wants the unit available for perimeter checks at 1600."

There was a brief pause. "Give me sixty seconds to pull the monitoring feed."

The diagnostic was the most comprehensive assessment since the original calibration, and running it required Sonn to be active on the link throughout rather than pulling a passive data file. The procedure was: I held the integration interface junction points and ran through a series of sensory and motor calibration exercises while Sonn read the output from the monitoring feed and correlated it with the baseline data from the previous diagnostics.

It took forty-seven minutes.

The predictive extension ran at reduced load during the diagnostic, which was the only time I'd found that the system willingly throttled its own processing. The theory Sonn had offered was that the diagnostic itself was a form of assessment the device recognized, and it reduced ambient processing to make the readout cleaner. I'd stopped trying to verify this. It happened consistently, which was enough.

"Integration stability," Sonn said when the diagnostic completed. They were reading from the output feed, their voice carrying the quality of focused attention they brought to technical analysis. "The predictive extension's load profile has shifted from the acute pattern I read after the retreat engagement. The system has integrated the new capability into the baseline load structure."

"What does that mean in practice?"

"The nosebleeds were acute stress responses to a sudden capability expansion. The system was running the predictive extension on top of a baseline that hadn't incorporated it yet." A brief pause. "It has incorporated it now. The predictive extension is load-bearing structure, not additional load. Your current operational baseline includes it."

"The nosebleeds stopped two weeks ago."

"Yes. This is consistent with the integration timeline." Sonn moved through something on their end, the soft sound of data being navigated. "There are two flag items in the diagnostic that I want to address directly."

"Go ahead."

"Visual acuity in low-light conditions has increased by a margin that exceeds the previous assessment significantly. The increase appears to be associated with the predictive extension's spatial processing: your visual system is providing higher-resolution input to the pattern recognition because the pattern recognition is using that input to generate more accurate positional data. The two systems are reinforcing each other."

"Is that a problem?"

"It is not a problem in the operational sense. It is significant in the long-term trajectory sense, which we should not pretend is the same thing as no problem." Sonn's voice was careful without being soft. "The second flag item is the sleep data."

"Four hours."

"You are reporting four hours of sleep and characterizing your functional state as adequate. The diagnostic confirms functional adequacy at four hours." Another pause. "The diagnostic also confirms that the pattern recognition's ambient processing does not fully disengage during sleep. What I am reading is a system that has found the minimum rest window required for physical recovery and is treating the remainder of the rest period as available processing time."

"The system runs through my sleep and I get four hours of rest instead of eight."

"The system runs through your sleep, you get four hours of physical recovery, and the remaining time is used for processing that I cannot fully characterize because it is not producing outputs I can read cleanly." Sonn chose the next words with the specific care they used when they wanted to be accurate rather than alarming. "I want to be honest about what I don't know here. I don't know what the processing is doing during the rest period. I know it is happening."

I sat with that for a moment. The tone in my right ear ran its frequency, steady. The diagnostic had the extension at reduced load, thirty percent of its normal operational baseline.

"Can I increase the rest duration by choice? Go to bed earlier, sleep longer?"

"The physical recovery threshold appears to be approximately four hours regardless of the sleep duration offered. At six hours, the diagnostic shows four hours of recovery signature and two hours of active processing. At three hours, the diagnostic shows three hours of recovery and the system compensates by running higher processing load during waking hours."

"So the system will take what it needs and use the rest."

"That is the most accurate summary, yes."

It was the answer I'd expected but not the answer I'd wanted. The thing about the ear tone was that I'd learned not to be bothered by it, which was different from it not being there. The sleep data was in the same category: not painful, not acutely impairing, simply present, running in the background of everything I did at its structural baseline pace.

"Sonn," I said. "Overall assessment. Stable or developing?"

"Stable and developing," Sonn said. "The base structure is stable. The developmental curve is ongoing. These are both true at the same time and they will remain both true at the same time for the foreseeable trajectory." They paused. "You are not approaching cascade. The gap between your current integration load and the threshold I flagged in Chapter 10 has not meaningfully closed in the last two weeks of high operational tempo. The system is managing the load distribution effectively."

That was the answer that mattered. The rest of it was things I would need to account for, adapt to, carry, in the way I'd learned to carry the ear tone and the changed sleep pattern and the visual acuity in the dark.

"One more thing," Sonn said. "Derek."

I waited.

"The diagnostic confirms that the predictive extension is integrating rather than destabilizing. The system is adapting to the new capability. But the adaptation itself creates new structures, and those structures will produce new capabilities in time." They were quiet for a moment. "I have been trying to tell you this in a way that is accurate without being alarming, and I've decided that accuracy matters more than managing how it lands. You will continue to change. I do not know at what rate or in what direction. I am running projections against the historical data and the historical data does not have adequate precedent for your current position on the development curve."

"You've said that before."

"I keep saying it because the gap between what I know and what I need to know to give you a complete picture keeps getting wider rather than narrower." A long pause. "I am not saying this to alarm you. I am saying it because you should make operational decisions with complete information, including the information that my picture is incomplete."

I thought about the engagement Eshet was running on her own timeline, patient in the northeast sector, and the forward position's north approach with Rovan on the east side and Hetch in the channel, and Denavi at the secondary position because she was the best terrain reader I had.

"After the engagement," I said. "We'll do another diagnostic after the engagement and you can update the projections."

Sonn said: "Yes. After."

I closed the link and sat in the equipment bay for two minutes without moving. The predictive extension registered the bay's ambient signatures in the background, forty-three stable elements, no anomalous movement, structural baseline load. The ear tone ran at its frequency.

Something else had changed.

I'd noticed it two days ago and been finding precise language for it since. The visual acuity increase was real and measurable. But the thing I was trying to describe was less about vision and more about the quality of the information the pattern recognition was generating at its new baseline. Before the retreat engagement and the predictive extension's first appearance, the system had felt like a tool I was using. After the nosebleeds stopped and the extension went structural, the tool feeling was different.

It wasn't that the system felt like it was running me. It was that the boundary between where I ended and where the system began had gotten harder to locate. The extension projected lines of intent forward in time and my decision-making incorporated those projections the same way it incorporated memory or learned instinct, without a seam.

I didn't know whether that was integration working the way it was supposed to or the beginning of something Sonn's historical records described in the careful-around-it language of cases they didn't want to document directly.

What I knew was that the diagnostic said stable. What I knew was that the engagement was coming on Eshet's timeline. What I knew was that the north approach needed Rovan on the east side and Hetch in the channel and Denavi at the secondary position.

The rest of it was going to have to wait.

I went to find Greel at the perimeter.

He was at the northern edge, looking at the gap with the expression he wore when he was running calculations against factors he didn't have complete numbers for. The light was moving toward late afternoon, the shadows long across the industrial terrain, and the northeast sector had been static for twenty-two hours now.

"Diagnostic done," I said.

"And?"

"Stable. Visual acuity has increased. Sleep pattern is fixed at four hours. Predictive extension has gone structural." I gave him the relevant summary, the parts that affected operational planning, not the parts about Sonn's incomplete projections and the widening gap between what was documented and what was happening. "I'm fully operational."

Greel looked at the gap. "Rovan on the east side is good. Brief him again tonight. I want him running that position in the dark twice before morning."

"I'll set it up."

"Denavi told me about the signal shorthand you've been running with the newer soldiers. The twelve-signal protocol." He turned from the perimeter. "Cross-train me. Whatever you and Denavi have built, I need to be fluent in it before this engagement."

I spent the next forty minutes walking Greel through the protocol in the fading afternoon light, the same twelve signals, the layered system for range and urgency, and Greel learned it with the focused efficiency of a commander who understood that the gap between briefed and fluent was the gap between surviving the engagement and managing it. He was fluent by the time the light dropped.

After dinner, which was the forward position's standard ration situation, adequate and uninteresting, Durrik found me in the equipment bay where I was running Rovan through his second dark navigation of the north approach on the position layout map.

He waited until Rovan finished his run-through and went to prep his gear before he said anything.

"How's the diagnostic?"

"Stable."

"Sonn give you a complete picture?"

I looked at him. "Sonn gave me the most accurate picture they have. The accurate picture includes significant gaps."

Durrik nodded. He picked up a charge pack from the supply crate, checked its level, set it back. "The extension is running continuously now."

"Yes."

"At what load right now?"

"Forty percent. Structural baseline."

He set down the pack. "And at full operational load during an engagement?"

"I don't have a ceiling number. I've run it at high load for sustained periods and it holds." I thought about the node assault's final hours, the toll on the extension's output quality when Eshet's dense integration signature was generating interference. "The limit I've found is interference from high-density signatures. Specifically Eshet's."

"But you got the cycle read in the gap."

"At close range, yes. Her interference attenuates at close range." I'd worked through the geometry of this twice since Chapter 36 was still ahead of me, but I knew the mathematics of it from the third probe's engagement data. "The interference comes from the density of her integration signature reading like static on my pattern recognition. At close range, the predictive extension can resolve individual signals out of the static because the signal strength increases faster than the static at close range."

Durrik absorbed this in the way he absorbed tactical information: without visible processing, just taking it in and setting it somewhere useful. "So closer is better."

"Against her specifically, yes. At operational range, her signature creates coverage gaps I can't fully read around. At close range, I can work with it."

"That's a hell of an operational recommendation." He wasn't criticizing. He was noting.

"It's a constraint. Not a plan."

He picked up another charge pack, checked it. "You're going to be running the extension at full load for how long, when this engagement starts?"

"Until it ends."

"And you don't have a ceiling number."

"No."

Durrik set the pack down. He looked at the equipment bay's north wall, which was the interior side of the approach he'd spent the last two days familiarizing himself with in the same way he did everything: methodically and without announcement. "The nosebleeds."

"Haven't recurred since the extension went structural."

"But the ear tone is louder under load."

I'd mentioned that once, two weeks ago, in a brief diagnostic follow-up conversation with Sonn that Durrik had been present for. He'd filed it and held it.

"Yes," I said. "Gets louder when the system is running hard. Settles back to baseline afterward."

"That's a physical limit marker," Durrik said. "Even if it doesn't affect function."

"Sonn thinks so too."

He was quiet for a moment. His hand rested on the supply crate with the stillness he brought to everything he was taking seriously. "You're going to monitor it during the engagement and you're not going to say anything unless it gets bad enough to affect your output."

That was accurate. I didn't confirm it.

"Make sure Sonn has access to your integration readings live during the engagement," Durrik said. "If something is happening at your end that you're not tracking clearly, I want someone with eyes on the data."

"Sonn doesn't have real-time access during ground operations. Link latency."

"Sonn can run a passive monitoring feed that buffers locally and uploads at link opportunity." He looked at me. "It's in the protocol documentation for integration-active operators in forward positions. Vasht can authorize it. Send her the request tonight."

He knew the protocol documentation for integration-active operators in forward positions. I'd had no particular reason to know it existed. He'd looked it up.

"I'll send the request," I said.

Durrik picked up the equipment he'd come for, the small kit in the corner that he'd been servicing for an hour before I'd arrived, and moved toward the bay's exit. At the frame he stopped without turning around.

"Get four hours. Whatever the system needs besides that is its business. You need to be sharp when she comes."

I sent the monitoring request to Vasht at 2100. She authorized it at 2113 and had Sonn's passive feed active by 2140. I confirmed the feed was running on the integration display at 2145 and then went to run Rovan through the north approach one more time in the dark.

He ran it clean. His movement through the gap-to-east-side coverage distance was six seconds faster than his afternoon run, and he hit the position naturally, the way bodies do when the muscle memory has gone in properly.

"You're good," I told him.

He nodded once, checking the grip on his rifle in the dark, and we walked back to the interior position without talking.

At 2300 I went to sleep. The predictive extension ran its standby tracking through the northeast sector and the north approach and the equipment bay and the interior position, forty-three stable elements plus two on-watch signatures plus the rotating signature of Greel running his own check of the perimeter. The system ran its structural baseline processing and my brain rested inside it the way it had learned to rest, four hours of recovery while the extension did whatever it did with the remaining time.

The ear tone ran its frequency in my right ear.

The northeast sector was still static. Twenty-nine hours without movement, the signatures sitting in their positions with the quality of something that had stopped waiting and started deciding.

I went to sleep.