Marcus had been staring at it for ten minutes, the blue text hovering at the edge of his vision like a patient salesman. The library was quiet around him - 3 AM, and even the coordination center had wound down. Just him and the question he couldn't answer.
[Skill Evolution Available]
Anomaly Detection v0.9.2-beta → v1.0-release
Accept upgrade? [Y/N]
Is this a reward? Or a trap?
He'd been asking himself that since the prompt appeared. The System had just flagged the entire cluster for behavioral analysis. It was watching them now. And in that exact moment, it had offered him - specifically him, the person who'd triggered the flag - an upgrade to the skill he'd used to trigger it.
That wasn't coincidence. The System didn't do coincidence.
The timing was too precise. Building 2847's deprecation flag had been rescinded at 4:17 PM. The System's behavioral analysis notification had appeared at 6:23 PM. And now, at 2:47 AM, exactly eight hours after the flag, Marcus received a skill upgrade offer.
Like the System had been watching. Processing. Deliberating. And then made a decision.
You found an exploit. Here's a better tool. Please continue exploiting.
Except that made no sense. Why would the System reward the behavior it had just flagged as anomalous? Unless-
Something else nagged at him. The System message from his own deprecation notice had stuck with him: "Individual deprecation flags cannot be rescinded once issued." Rescinded. Not cancelled. Not removed. Legal language, precise and absolute. Yet Building 2847's deprecation had been pulled back anyway. Marcus filed that contradiction away, another data point that didn't quite fit the System's stated rules.
Marcus pulled up his notes file and started typing, organizing his thoughts the way he always did when a problem was too complex to hold in his head.
Theory 1: Carrot and Stick
System flags cluster (stick), upgrades Marcus's detection skill (carrot). Purpose: incentivize continued exploit discovery while maintaining surveillance capability.
Problems:
- Why reward exploit behavior it's supposed to prevent?
- Why upgrade the person most likely to find more exploits?
- Unless... it WANTS exploits found?
He paused. Read that last line three times.
Unless it wants exploits found.
That thought opened uncomfortable doors. If the System wanted exploits discovered, that implied testing. Quality assurance. Someone-or something-using the survivors as unpaid QA labor to find bugs in reality's new operating system.
Marcus had spent five years at Nexus doing exactly that: paying users to find bugs the dev team had missed. Beta programs. Bug bounties. External testing.
He minimized the notes and pulled up the skill upgrade prompt again.
[Skill Evolution Available]
Anomaly Detection v0.9.2-beta → v1.0-release
Accept upgrade? [Y/N]
Warning: v1.0-release includes telemetry reporting. Usage data will be collected for system optimization purposes. Data collection cannot be disabled.
That warning hadn't been there before. Or Marcus had been too focused on the upgrade offer to notice it.
Usage data will be collected. Cannot be disabled.
The System wasn't even being subtle. It was explicitly telling him: take this upgrade, let us watch everything you do with it.
And in exchange, he'd get better capabilities. Active scanning. Source identification. Pattern history. All the features he'd been wishing the beta version had.
It was the classic Faustian bargain of modern software: give up your privacy, get better functionality.
Except in this case, privacy meant "the System doesn't know every exploit you're finding" and functionality meant "you might survive longer."
Marcus stood up and walked to the window.
The cluster was dark except for guard posts and a few late-night work lights. Two hundred people sleeping, or trying to. All of them depending on Marcus and people like him to find patterns, discover exploits, buy them time.
If he rejected the upgrade, he kept his privacy. The System wouldn't know every anomaly he detected, every pattern he identified. But he'd be working with inferior tools. Slower. Less effective.
If he accepted, he'd get better capabilities-and become a data source for the thing trying to kill them.
But I'm already a data source, Marcus thought. The System already knows I triggered the building save. It already flagged the cluster. I'm not under the radar-I'm completely visible.
The question wasn't whether to stay hidden. He'd lost that option when Building 2847's status flipped from Flagged to Active. The question was whether to trade privacy he didn't have for capabilities he needed.
Put that way, it wasn't much of a choice.
Marcus returned to his chair and stared at the prompt one more time.
The smart play was probably to reject it. Stay cautious. Don't give the System more data than absolutely necessary.
But he was a QA tester. His whole career had been built on poking at systems to see what happened. And right now, the most interesting question was: what happens if I accept?
That wasn't caution. That wasn't even smart. But it was data. And in an environment where information was the only advantage, refusing data collection because it was risky felt like the wrong kind of cowardice.
Marcus reached out and selected Accept.
The world flickered.
Not metaphorically-literally flickered, like a monitor refreshing at the wrong frequency. For half a second, Marcus's perception doubled: he could see the library and something else, overlaid but not quite aligned. Wireframes. Data structures. The library rendered as nested collision volumes and texture maps, every surface marked with invisible metadata. Glowing symbols that weren't quite text, a language his brain almost recognized but couldn't parse.
Then his vision snapped back to normal and the vertigo hit.
Marcus grabbed the desk, stomach lurching, inner ear screaming that he'd just fallen while his eyes insisted he was perfectly still. The sensation lasted three seconds-enough for sweat to break out on his forehead-and then faded like it had never been.
When his hands stopped shaking, Marcus pulled up his status screen.
Active Skills: Anomaly Detection v1.0-release
The beta tag was gone. The version number had incremented. He opened the skill description with a sense of anticipation he couldn't quite suppress.
Anomaly Detection v1.0-release
Passive/Active. Identifies environmental and systemic inconsistencies.
Range: 25m (passive), 100m (active scan)
New features:
- Active scan mode (cooldown: 5 min)
- Source identification (confidence-weighted)
- Pattern history (last 24h cached)
- Enhanced sensitivity (+40% detection range)
Note: This skill has been flagged for telemetry. Usage data may be reported for system optimization purposes.
Marcus read the last line three times.
Flagged for telemetry. Usage data may be reported.
Not "will be reported"-"may be reported." Interesting word choice. Like the System was hedging, or the telemetry was conditional.
He pulled up his character sheet's technical details-something he'd only glanced at before because it was mostly noise. But now one line jumped out:
Stolen from its rightful place, this narrative is not meant to be on Amazon; report any sightings.
Active Monitoring Flags:
- Skill Usage (Anomaly Detection): TELEMETRY ENABLED
- Behavioral Analysis: PENDING REVIEW
So the telemetry was real. Not hidden in fine print or buried in logs-explicitly labeled in his character sheet. The System wanted him to know he was being watched.
Every anomaly he detected now, the System would see. Every pattern he identified would be logged, analyzed, fed back into whatever algorithms governed reality's new rules. He wasn't just using the System's tools-he was becoming part of its quality assurance process, documenting bugs in the code whether he wanted to or not.
Marcus stood up and stretched, trying to work the residual vertigo out of his system. His hands were steady again. His breathing was normal. The momentary visual corruption hadn't repeated.
So that's what a skill upgrade feels like, he thought. Like someone just installed a patch directly into your brain.
His status screen showed a new notification:
[Tutorial Available: Skill Upgrades]
Learn more about skill evolution mechanics?
Marcus dismissed it. He'd rather experiment than read documentation written by a System that was actively trying to kill him.
Time to see what the upgrade could actually do.
Marcus tested the new features at dawn.
He'd waited until 6:00 AM-not for any strategic reason, but because testing in the dark felt like a good way to miss important details. Now, standing on the library's flat roof with the sun just starting to creep over East Austin, he had good visibility and enough isolation that if something went wrong, he wouldn't panic a crowd.
The rooftop was empty except for him and a makeshift communication antenna someone had rigged from salvaged parts. The view stretched in every direction: cluster buildings to the north, deprecated zones to the east and south, the slow-motion architectural decay of a city being garbage-collected one structure at a time.
Marcus took a breath and triggered the active scan.
The sensation was immediate and disorienting.
His perception expanded-not physically, not like his vision had suddenly improved, but like someone had turned on a debugging overlay that showed him information his brain shouldn't have access to. The passive version of Anomaly Detection had been like peripheral vision: vague pings, uncertain warnings, a sense that something was wrong without specifics.
This was different. This was seeing the collision volumes.
Entities lit up like thermal signatures. Three Glitch Spawns prowling the perimeter, staying just outside the barricades-Marcus could see them through walls, through buildings, their positions marked with glowing outlines like they'd been highlighted by a game engine's debug mode. A Glitch Horror two blocks south, moving slowly through a deprecated zone, its threat classification displayed as floating text: [Horror-class | Threat: Moderate | Behavior: Patrol].
And something else-fainter, harder to read-near the engineering building.
[Active Scan Complete]
Scan radius: 100m
Duration: 8.3 seconds
Processing time: 1.1 seconds
Anomalies detected: 4
- Entity (Spawn-class) x3: Perimeter, confidence 94%, threat level LOW
- Entity (Horror-class) x1: Grid D-7, confidence 97%, threat level MODERATE
- Unknown signature: Engineering East, confidence 34%, threat level UNKNOWN
Marcus blinked and the debug overlay faded, leaving him staring at normal Austin morning light. His head hurt slightly-not a migraine, more like the low-grade ache of eyestrain from staring at a monitor too long.
The scan had lasted eight seconds. In that time, he'd gotten more information than fifteen minutes of passive detection would have provided.
This is what the beta version should have been, Marcus thought. This is what "release-quality" actually looks like.
He pulled up his notes and started documenting.
Active Scan - Initial Test
Time: 6:07 AM
Location: Library roof
Duration: 8.3 seconds (plus 1.1s processing lag)
Results:
- 3 Spawn-class entities at perimeter (94% confidence)
- 1 Horror-class entity in deprecated zone (97% confidence)
- 1 unknown signature, 34% confidence
Observations:
1. Visual debugging overlay (collision volumes, threat classifications)
2. Confidence ratings on detection (beta version never showed this)
3. Behavior tags ("Patrol" on Horror-class)
4. Scan has processing lag-not instantaneous
5. Mild headache post-scan (stamina cost? cognitive load?)
He checked his status screen. Stamina: 94/100. So there was a cost, just not severe.
The unknown signature was interesting. Thirty-four percent confidence-barely above noise threshold. The beta version wouldn't have flagged something that uncertain. The release version was more sensitive, which meant more data but also more false positives.
Marcus made a note: More data, more noise. Classic trade-off. Need to establish confidence threshold for actionable vs. background.
He waited five minutes-watching the cooldown timer tick down in his peripheral vision-and triggered another scan.
This time he was ready for the sensation. The expansion of perception, the overlay effect, the momentary disorientation as his brain tried to process information it wasn't designed to handle. He rode it out, staying focused on the data.
Same three Spawns at the perimeter. Same Horror in Grid D-7. But the unknown signature had moved-now 28% confidence, slightly west of its previous position.
[Active Scan Complete]
Anomalies detected: 4
- Entity (Spawn-class) x3: Perimeter, confidence 94%, threat level LOW
- Entity (Horror-class) x1: Grid D-7, confidence 96%, threat level MODERATE
- Unknown signature: Engineering Central, confidence 28%, threat level UNKNOWN
The confidence had dropped. Whatever the unknown signature was, it was either moving erratically or the system wasn't sure it existed.
Ghost signal, Marcus thought. Or data corruption. Or something that doesn't fit the entity classification system.
The headache was worse this time-not debilitating, but noticeable. Two scans in five minutes was apparently pushing limits.
Marcus sat down on the roof's edge and pulled up the pattern history feature.
That was new too. A timeline of anomalies he'd detected over the past twenty-four hours, plotted against his movement and the cluster's activity. It was the kind of analysis tool he would have killed for back at Nexus: automated correlation between his observations and contextual data.
The timeline showed:
- Last night at 2:47 AM: Skill upgrade accepted (marked as "System Event")
- Yesterday at 6:23 PM: Cluster flagged for behavioral analysis
- Yesterday at 4:17 PM: Building 2847 status change
- Yesterday at 3:15 PM: Increased entity activity near cluster perimeter
- Yesterday at 12:00 PM: Building occupation test begins
Each event was tagged with timestamps, locations, and Marcus's proximity. But it was the granular detail that made his stomach drop-not just where he'd been, but how long he'd paused before making a decision. Which anomalies he'd investigated first. When he'd chosen to wait versus act immediately.
The timeline didn't just show anomalies-it showed his decision-making process. When he'd hesitated. When he'd acted immediately. The System wasn't just collecting data on what he found. It was learning how he thought.
Marcus closed the pattern history and stared out at the cluster, feeling suddenly exposed in a way that had nothing to do with standing on an open rooftop.
The source identification feature turned out to be the most immediately useful.
Marcus spent the next hour doing systematic scans from different positions around the cluster, documenting entity movements and testing the system's accuracy. By 7:30 AM, he had a pretty clear picture of the local threat landscape.
The three Glitch Spawns at the western perimeter were persistent-they'd been there for at least six hours, based on guard reports Chen had shared yesterday. They moved in predictable patrol patterns, staying just outside the barricades like they were testing the boundaries without actually engaging.
The Horror in Grid D-7 was more interesting. It wasn't patrolling-it was waiting. Every scan showed it in roughly the same position, moving occasionally but not with purpose. Like it was stationed there.
Marcus pulled up his notes and added a new section:
Entity Behavior Patterns:
Spawns (Perimeter):
- Consistent patrol routes
- Stay outside safe zones
- High confidence detection (94-95%)
- Behavior tag: "Patrol"
Horror (Grid D-7):
- Stationary or slow movement
- Positioned in deprecated zone
- High confidence detection (96-97%)
- Behavior tag: "Stationed"
Hypothesis: Entity deployment is strategic, not random. Spawns patrol boundaries. Horrors hold positions. System is allocating resources like it's running threat containment algorithms.
That matched the monitoring flag. The System had flagged the cluster for "behavioral analysis" and immediately increased entity presence in the area. It was responding to the exploit-not with instant deletion, but with increased pressure.
Resource allocation, Marcus thought. The System treats clusters as threat-level problems. Demonstrate unusual capability, get elevated to higher threat tier, face more dangerous enemies. It's literally just priority queue management.
At 8:02 AM, Marcus triggered another scan and caught something new.
The unknown signature was back-but this time at 67% confidence. And it was inside the cluster.
[Active Scan Complete]
Anomalies detected: 5
- Entity (Spawn-class) x3: Perimeter, confidence 94%, threat level LOW
- Entity (Horror-class) x1: Grid D-7, confidence 97%, threat level MODERATE
- Unknown signature: Main Quad, confidence 67%, threat level UNKNOWN
- Anomalous player signature: Coordination Center, confidence 82%, type UNUSUAL
- Environmental corruption: Library basement, confidence 51%, type DECAY
Marcus's heart rate spiked.
Anomalous player signature. Inside the coordination center. Eighty-two percent confidence.
He triggered another scan immediately, not waiting for cooldown. The five-minute timer was still active-he forced it anyway, feeling something tear in his perception as the skill activated before it was ready.
[Cooldown Violation Detected]
Stamina cost increased: -15 (94 → 79)
Scan accuracy reduced: -20%
Headache intensity: Moderate → Severe
The pain hit like a spike through his temples, but Marcus gritted through it and focused on the data.
[Active Scan Complete - Degraded]
Anomalies detected: 4
- Entity (Spawn-class) x3: Perimeter, confidence 74% (REDUCED)
- Entity (Horror-class) x1: Grid D-7, confidence 77% (REDUCED)
- Anomalous player signature: Coordination Center, confidence 63% (REDUCED), type UNUSUAL
The player signature was still there. Lower confidence because of the forced scan, but consistent. Someone in the coordination center was reading as anomalous.
Marcus stood up too fast, stumbled slightly from the headache, and headed for the roof access door.
He needed to know who-or what-the System thought was unusual about one of their own people.