Marcus knocked on the desk edge. "I found something."
Chen looked up, expression shifting from exhaustion to sharp focus in half a second. "Already?"
"The deprecation system isn't random. It's usage-based. The System monitors activity metrics and flags structures that fall below a maintenance threshold. Flagged structures get added to a queue, which processes once a day at 11:47 UTC."
Chen leaned back, studying Marcus. "Prove it."
Marcus spread his notes on the desk. Walked Chen through the timestamp analysis. The structure status correlation. The pattern of abandoned buildings receiving notices. The high-activity cluster buildings remaining safe.
"Every deprecated structure in the dataset was empty for days before the notice hit," Marcus concluded. "Low user presence, no power consumption, no maintenance activity. The System treats buildings like cloud computing resources - if you're not using it, it assumes you don't need it."
"And you think we can prevent deprecation by increasing activity?"
"I think we can test that hypothesis. Take one of the cluster's low-activity buildings. Move people in. Increase power usage. Monitor whether it stays off the deprecation list."
Chen was quiet for a moment, fingers steepled, thinking.
"The three storage buildings on the southeast perimeter," he finally said. "Low traffic. Nobody lives there. Just supply storage. If your theory's right, they're at risk."
"How long have they been low-activity?"
"Maybe a week? We moved most of the supplies to central distribution to reduce runner trips." Chen pulled up a map on his laptop. "If we're going to test this, those are our candidates."
"Do we have enough people to increase occupancy?"
"Depends on what counts as 'occupancy' to the System. Could be actual residence, could just be regular traffic." Chen made notes on a pad. "We can run operations through there. Daily inventory checks. Move some guard rotations to include that sector. Turn on lights, run equipment."
"How quickly can you implement?"
"If the council approves it? Tonight." Chen looked at Marcus. "This is good work, Webb. Real analysis, not just wild speculation. If you're right about the queue structure-"
"We could save buildings. Maybe even reverse deprecations if we catch them early enough."
"Maybe." Chen's expression was cautious. "Don't get ahead of yourself. We need verification first. One positive test result doesn't prove the system works at scale."
Spoken like someone who's managed QA teams, Marcus thought.
"I know. But it's more than we had yesterday."
"It is." Chen stood up. "I need to take this to the council. If they approve the test, I want you running it. Document everything. Activity levels, timing, System response. Full QA methodology."
Marcus hesitated. "There's one problem with the methodology."
Chen's expression sharpened. "Go on."
"Standard experimental protocol would be to leave one building untouched. A control." Marcus pulled up his notes, the old instincts from a decade of QA work surfacing automatically. "We increase activity on two buildings, leave the third alone. If the control gets deprecated and the test buildings don't, we've isolated the variable. If all three survive, we haven't proven anything - could be coincidence, could be the System changed its algorithm, could be they were never actually at risk."
Chen was quiet for a moment. "You're saying we deliberately let a building fail."
"I'm saying that's what proper methodology requires."
"And if the control building has supplies we need? Equipment? What if someone's sheltering there that we don't know about?"
Marcus had already thought about this. It was the same problem he'd faced a hundred times in software testing - the gap between laboratory conditions and production environments. "In a lab, you'd never ship without a control test. But we're not in a lab. We're running experiments in production, on live infrastructure, with real stakes."
"So what do you recommend?"
Marcus made a note in his documentation, the words feeling like a confession: Test methodology compromised - no control group. Cannot definitively isolate variable. Real-world constraints prevent proper experimental protocol.
"We test all three buildings," he said. "We document the limitation. And we accept that our results will be circumstantial, not conclusive." He looked up at Chen. "It's not clean science. But clean science requires conditions we don't have - the luxury of letting something fail just to prove a point."
Chen nodded slowly. "QA in production."
"QA in production," Marcus agreed. "You do your best, you document your assumptions, and you ship knowing the test coverage isn't complete. Because the alternative is shipping nothing."
"The council won't care about methodology," Chen said. "They'll care about results."
"I know. But I care about methodology. And when someone asks why we can't replicate this in another settlement, or why the data doesn't hold up under scrutiny, I want to be able to say we knew the limitations going in. That we made a choice, not a mistake."
Chen studied him for a moment. "You're more thorough than I expected."
"I'm a QA engineer. Thoroughness is the only thing we're good at."
"What about my deprecation timer?"
"Still ticking. But if you're right about activity metrics..." Chen paused. "Your building's been empty since yesterday morning, right? Since you evacuated?"
Marcus nodded.
"So when your building's 72 hours are up, you'll find out whether being outside the structure actually saves you or if the deprecation applies to users regardless of location."
Marcus had been trying very hard not to think about that. The possibility that the System's garbage collection was thorough enough to track users even after they'd left deprecated structures.
"Either way, we get data," Marcus said.
"Either way," Chen agreed. "Now go get some rest. You look like shit, and exhaustion penalties are going to hurt your PER stat. I need you sharp for tomorrow."
Marcus didn't rest.
He went back to his carrel and kept working.
The deprecation data had opened up new questions, and his brain wouldn't shut down until he'd explored them. That was the problem with pattern recognition - once you started seeing connections, you couldn't unsee them.
He pulled out the entity behavior reports and started cross-referencing them with deprecation data.
Entity spawn density was higher near deprecated structures. Not dramatically, but consistently. Building 12 had a 30% higher spawn rate in its immediate vicinity compared to the cluster average. His building - Structure ATX-E-4471 - showed a similar pattern.
Why would entity spawns correlate with deprecation?
Hypothesis 1: Deprecated structures become spawn points. System repurposes flagged infrastructure.
This narrative has been purloined without the author's approval. Report any appearances on Amazon.
That was dark but logical. If buildings were being removed from the "valid location" database, maybe they became entity territory instead.
Hypothesis 2: Low activity attracts entities. Same factors that trigger deprecation also draw spawns.
That made more sense from a systems perspective. Entity pathfinding and structure maintenance were probably different systems, but they might reference the same activity metrics. Low human presence = high spawn desirability.
Marcus made notes. Another test to run. Monitor entity behavior around the southeast storage buildings before and after increasing activity.
He checked the time. 8:47 PM. His timer read 63 hours, 29 minutes.
His body was starting to register complaints. The fire escape descent and the underpass sprint had left him with muscle aches that had graduated from "annoying" to "concerning." His eyes felt like they'd been sandblasted, and the desk lamp's harsh fluorescent glow wasn't helping. His status screen reflected it:
STATUS EFFECTS:
- Exhaustion (moderate): -20% STR, -15% AGI
- Sleep Deficit (major): -10% PER, -5% INT
- Deprecated (63h 29m remaining)
The Sleep Deficit had upgraded. Not good. Chen was right - he needed rest, or his cognitive stats would tank and make the analysis work harder.
But he had one more thing to check.
Marcus pulled up the quest notification that had appeared yesterday, the one he'd dismissed without accepting.
[Quest Available: Integration Assessment]
Objective: Demonstrate adaptation to System integration
Requirements:
- Identify 3 System mechanics
- Document 1 exploitable pattern
- Survive 72 hours post-arrival
Reward: +500 XP, Skill upgrade (Anomaly Detection → v0.9.2-beta), Development Credits: +250
Time limit: 72 hours from quest offer
Accept? [Y/N]
Marcus read it three times.
The quest was asking him to do exactly what he'd been doing anyway. Document mechanics. Find exploits. Survive. And the reward was significant - 500 XP was probably enough to hit Level 3, and a skill upgrade might improve his detection range or confidence accuracy.
But the wording was interesting. Integration Assessment. Not "survival" or "combat" or "resource gathering." Integration. Like the System was evaluating how well he was adapting to the new reality.
And the requirements were cognitive, not physical. Identification. Documentation. Analysis.
This quest was designed for Logicians, Marcus realized. Specifically for people who analyze rather than fight.
Which meant the System had different quest templates for different classes. Combat classes probably got "kill X entities" quests. Support classes probably got "gather X resources." And Logicians got "find patterns."
The System is optimizing quest design for class archetypes.
That was actually good game design. Personalized content. Role-appropriate challenges.
It was also manipulative as hell. The System was giving him a quest that appealed specifically to his strengths, rewarding him for doing what he was naturally inclined to do anyway.
Classic engagement optimization. Make the user feel like they're choosing their path when really you're just validating their existing behavior.
Marcus should refuse it on principle.
But 500 XP and a skill upgrade would make him more effective at the analysis work. Would increase his chances of surviving the deprecation timer. Would make him more valuable to the cluster.
Pragmatism beats principles in the apocalypse, Marcus decided.
He accepted the quest.
[Quest Accepted: Integration Assessment]
Progress: 1/3 mechanics identified, 1/1 exploit documented, 0/72 hours survived
Time remaining: 69h 14m
The lane marking exploit counted as the documented exploit. The cron job discovery probably counted as one mechanic. He needed two more mechanics and approximately three days of not dying.
Achievable goals.
Marcus closed his laptop, finally giving in to the exhaustion that was dragging at his consciousness. His carrel had become a nest of papers and notes and half-formed theories. He'd organize it tomorrow. Right now, he needed sleep before his INT penalty got worse.
He made his way back to the dorms, navigating through the cluster's nighttime routine. Fewer people moving now. Guard shifts changing. The smell of dinner - something with beans, maybe chili - drifting from the dining hall.
His bunk was exactly as he'd left it. Blanket still folded from this morning's half-hearted attempt at making the bed. His backpack in the corner, laptop inside.
Marcus set an alarm on his phone for 6 AM. 59% battery. He'd charge it tomorrow.
He lay down, and for the first time in three weeks, his brain actually agreed to stop running.
He dreamed about spreadsheets.
Not metaphorical spreadsheets. Actual Excel files, cells filled with timestamps and deprecation notices and entity spawn rates. In the dream, Marcus was trying to find a pattern, but every time he highlighted a row, it changed values. The data was alive, shifting, refusing to be categorized.
Somewhere in the dream, Chen appeared. "You're not looking at the right column," he said, pointing. But when Marcus looked, the column was empty. Just blank cells, stretching down forever.
"The data's missing," Marcus said.
"The data's never missing. You're just asking the wrong question."
Marcus woke up with that sentence echoing in his head.
6:02 AM. His alarm hadn't gone off - he'd woken up two minutes late. Close enough.
His deprecation timer read 54 hours, 14 minutes.
Outside, the cluster was waking up. Voices in the hallway. Someone running water in the bathroom. The ambient noise of two hundred people starting another day of survival.
Marcus checked his quest progress.
[Quest Active: Integration Assessment]
Progress: 1/3 mechanics identified, 1/1 exploit documented, 18/72 hours survived
Eighteen hours down. Fifty-four to go.
He got up, ready for Day Two of being useful instead of deprecated.
The library was already busy when Marcus arrived after breakfast - more questionable oatmeal, this time with raisins that might have been expired even before the apocalypse.
Chen was at his desk, looking like he'd pulled an all-nighter. His status screen showed a new debuff: Exhaustion (severe): -30% STR, -25% AGI, -15% PER.
"You didn't sleep," Marcus observed.
"I took the deprecation queue theory to the council. Took four hours to convince them it was worth testing." Chen pulled up a document. "We're implementing phase one today. Increased activity in the southeast storage buildings. I need you to establish baseline metrics before we start."
"What kind of metrics?"
"Everything. Power consumption logs, foot traffic counts, entity proximity data. Then we repeat measurements post-implementation and compare." Chen slid a clipboard across the desk. "Julie's organizing the field team. You're coordinating data collection."
Marcus took the clipboard. It had a structured observation template already printed - timestamps, location codes, measurement categories. Professional work.
"How many people on the field team?"
"Four. Including you." Chen highlighted three names on a list. "Maya Reyes - Scout class, good at entity detection. Marcus Webb - you. Owen Park - Defender class, combat support. And Julie Tran - Synthesist, pattern recognition backup."
"When do we start?"
"Thirty minutes. Briefing in the east conference room." Chen looked at Marcus directly. "This test matters, Webb. If you're right about the activity metrics, we can protect the cluster's infrastructure long-term. If you're wrong, we're wasting resources we don't have. No pressure."
"I love impossible pressure. It's my natural habitat."
Chen almost smiled. "Get to the briefing."
The east conference room was a repurposed study room, now equipped with maps and whiteboards and the kind of organized chaos that made Marcus nostalgic for actual office work.
The team was already assembled.
Maya Reyes was mid-twenties, built like a distance runner, with the alert posture of someone whose job involved not getting eaten by glitching reality monsters. Level 4 Scout. Her status screen listed a skill called Threat Detection v1.1.3.
Owen Park was older, maybe forty, stocky and calm in the way that suggested military or law enforcement background. Level 5 Defender. His gear was practical: reinforced vest, baton, knife. No flashy weapons. Just tools that worked.
Julie Tran Marcus already knew. She nodded at him as he entered, her pen already tapping that three-beat rhythm against her notebook.
"Okay," Julie said, taking charge of the briefing. "Webb, you want to explain the theory?"
Marcus walked them through it. The deprecation timing pattern. The activity correlation. The hypothesis that structures were flagged based on usage metrics and added to a processing queue.
"We're testing whether we can prevent deprecation by increasing activity in at-risk buildings," he concluded. "Southeast storage buildings are our candidates. Currently low-traffic. We need baseline data, then we implement activity increases and monitor the System's response."
Maya raised a hand. "What counts as 'activity' to the System?"
"Unknown. We're guessing: user presence, power consumption, movement, resource interaction. Anything that generates data the System tracks."
"And if it doesn't work?" Owen asked. Not challenging - just planning for contingencies.
"Then we've lost some time and learned that activity metrics aren't the right variable. But the buildings were already at risk, so minimal downside."
"Except for the part where we're spending time and resources on your theory instead of other priorities," Owen said.
Marcus met his gaze. "That's why we're testing. To validate before scaling. Standard QA methodology."
Owen nodded, apparently satisfied.
Julie stopped tapping her pen. "Wait," she said slowly. "If deprecation follows activity metrics... do entity spawns follow the same pattern?"
Marcus blinked. He'd noticed the correlation, but hadn't made the explicit connection. "The data shows higher entity density near deprecated buildings."
"Right. So if low activity attracts deprecation and entities..." Julie was already pulling up her notes. "Maya, you've been tracking spawn patterns. Do you see clustering around abandoned structures?"
Maya nodded. "Consistently. I thought it was random, but if there's a shared activity metric-"
"Then entity deployment might be inversely proportional to human presence," Julie finished. "The System spawns more entities where there are fewer humans. Which means if we increase activity in the storage buildings, we should see entity density drop as deprecation risk drops."
The room went quiet.
"That's testable," Marcus said. "We can measure both variables during the same observation period."
Julie looked at Owen. "If Maya's right, this isn't just about saving buildings. We could create entity-free zones by manipulating activity levels. That's strategically huge."
Owen's expression had shifted from skeptical to engaged. "We'd need to verify it. But if we can predict and control entity spawning patterns..."
Marcus was mentally recalculating the test parameters. Julie had just connected two patterns he'd been analyzing separately. That was Synthesist thinking - seeing the relationships between disparate systems.
She's good, Marcus thought. Really good.
"Okay," Julie said. "Maya, you're on entity monitoring. Track spawn density in a hundred-meter radius around the target buildings. Marcus, you're logging structure activity. Owen, you're on security - keep us alive while we collect data. I'll handle pattern synthesis if we spot anything unexpected. Questions?"
Nobody had questions.
"Then let's move. We've got three buildings to baseline before Chen implements phase one."