The dungeon was called Gangnam-7. B-rank, standard spawn configuration, cleared twice weekly by rotating Association teams. Three months ago, Wells' forensic unit had run a full energy audit on it as part of a routine monitoring cycle. The Association's instruments had measured the dungeon's total energy output at 847 kilojoules equivalent during a four-hour clear.
The System's actual extraction from the same clear was 12,340 kilojoules.
Alex needed to prove the gap existed. He needed to pull it from the System's own records, format it in a way that Wells' team could cross-reference against their monitoring data, and deliver it within seventy-two hours. He had fifty-three hours left.
The Archivist ran the query parameters while Alex worked the overlay from the fieldwork station's back room, hands flat on the table, eyes closed, the code layer filling his perception with the dense scrolling architecture of three months of Gangnam-7's operational data.
"The extraction records for that date are in the regional harvest archive," the Archivist said. The words appeared in his overlay as precise text, the AI fragment's measured voice rendered in clean sans-serif. "Accessing them requires sustained Level 3 administrative query. At your current bridge integrity, the sustained query will take approximately ninety minutes."
"Ninety minutes at full overlay."
"Correct. With periodic rest intervals to prevent bridge degradation, the total time extends to approximately two and a half hours."
Two and a half hours of running his admin access at full output. At 58% bridge integrity, that was the equivalent of sprinting with a pulled hamstring. Possible. Painful. The kind of thing that left marks.
Bug had set up in the main room with three laptops and what he described as a "correlation engine" that was actually a set of custom scripts running against scraped Association public records. The Association published aggregate dungeon energy data quarterly. Bug had the last six quarters of reports, broken down by dungeon, date, and team composition.
"The gap has to be obvious," Bug said through the doorway. "Not ambiguous, not interpretable, not 'well maybe the instruments were calibrated differently.' I need to show her a number from the System and a number from her own records, and the difference needs to be large enough that no reasonable explanation covers it except the one you're selling."
"The difference is fourteen to one."
Bug stopped typing. "Fourteen to one?"
"The System extracts roughly fourteen times the energy that the Association's instruments detect. Their sensors read the dungeon's physical energy output. The harvest channels operate on a frequency their equipment doesn't measure."
"Why not?"
"Because the frequency isn't in any physics textbook. It's not electromagnetic, not thermal, not kinetic. It's emotional yield converted to containment energy, and the conversion process happens in a layer of reality that human instruments weren't built to detect."
Bug sat with this for a moment. Then he went back to typing. "Fourteen to one will work."
---
Alex started the extraction at 2:00 PM. The overlay blazed to full output, the bridge straining at its 58% capacity, and the regional harvest archive opened like a filing cabinet the size of a building. Three months of Gangnam-7 data, organized by the System's own taxonomy: spawn parameters, difficulty calibration, participant biometrics, emotional yield metrics, and the harvest channel output that represented the real purpose of every dungeon run.
He found the specific date. July 19th. The Association's quarterly report listed a routine B-rank clear by Team Sigma-4, six hunters, four-hour duration, 847 kJ measured output. Standard. Unremarkable.
The System's records for the same clear told a different story.
Six hunters. Average emotional yield: 2,057 units per hunter. Peak yield during the boss encounter: 4,300 units for the tank who'd taken a hit that dropped him to 8% health before the healer pulled him back. Fear response, survival instinct, relief, triumph. All of it measured, quantified, and channeled through the harvest system into the containment infrastructure that kept the Prisoner contained.
Total extraction: 12,340 kilojoules equivalent. Not kJ in the physics sense. The conversion was approximate, the Archivist's translation of containment energy into a unit Wells' team would understand. But the ratio held. Fourteen to one. A gap that couldn't be explained by instrument error or calibration drift or atmospheric interference.
Alex copied the data into a format Bug could work with. The bridge ached. Not pain exactly, more like the deep fatigue that came from holding a position too long, muscles firing past their comfort range. The overlay flickered once at the forty-minute mark. He backed off, let the bridge settle, ran two of Kwon's calibration exercises, and went back in.
At the fifty-minute mark, the Archivist flagged something.
**[PRIORITY ALERT: WATCHER DEPLOYMENT DETECTED — CLASSIFICATION: SCOUT-CLASS — DEPLOYMENT ZONE: SEOUL METROPOLITAN AREA — STATUS: OBSERVATION MODE — THREAT LEVEL: MODERATE]**
Alex pulled back from the harvest archive. The alert sat in his overlay, blinking with the steady cadence the Archivist used for information that required immediate attention but not panic.
"Details," he said.
**[The scout-class Watcher was deployed approximately four hours ago. It is not hunting. It is mapping. Specifically, it is cataloging administrator-class energy signatures within the Seoul metropolitan area. This is consistent with Administrator Prime's standard protocol preceding direct intervention. The scout maps, Prime acts on the map.]**
"Can it detect me from here?"
**[At your current overlay output level, your bridge signature is detectable within a radius of approximately 800 meters. The fieldwork station is located 14 kilometers from the nearest mapped Watcher patrol route. However, a scout-class unit operates on wider parameters than standard patrols. If it passes within 2 kilometers during a high-output administrative query, it will register your signature.]**
**[Current estimated distance of scout unit from your position: 9.3 kilometers. Moving southeast.]**
Nine kilometers. Moving toward them or moving past them on a sweep pattern. The difference mattered enormously.
"Trajectory?"
**[Insufficient data for trajectory prediction. The scout has been active for four hours. Its movement pattern is consistent with a grid search of the greater Seoul area. If the pattern holds, it will pass within your detection radius in approximately six to eight hours.]**
Six to eight hours. He needed two and a half hours to finish the extraction. If he reduced his overlay output, the detection radius shrank, but the extraction time doubled. If he pushed through at full output, he'd finish before the Watcher reached detection range. Probably.
"Bug."
Bug appeared in the doorway.
"How long to correlate the data I've pulled so far with the Association records?"
"Depends on completeness. If you've got the full harvest log for July 19th, ninety minutes. If you've got partial data, longer, because I have to interpolate and that introduces uncertainty that Wells will notice."
"I've got sixty percent of the log. I need another ninety minutes at full output to complete it."
"Then you need ninety minutes." Bug looked at him. "What's the complication?"
"Watcher. Scout-class. Mapping administrator signatures in Seoul. If I keep running at full output, it might pick up my bridge signal."
Bug processed this the way he processed most information: fast, without visible reaction, the calculations happening somewhere behind the flat expression.
"Might."
"Six to eight hours before it reaches detection range. I need ninety minutes."
"So you finish four and a half hours before it could possibly see you. That's not a might. That's a margin." Bug shrugged. "Finish the extraction."
Alex went back to the overlay. Bug was right about the math. Six hours of margin was enough. But the math assumed the Watcher maintained its current grid pattern, and Watchers weren't obligated to maintain anything. They were subroutines with operational parameters, not GPS trackers with predictable routes.
He pushed through.
The second ninety minutes was worse than the first. The bridge at 58% was designed for observation, not sustained deep-archive queries, and the difference was measurable in the way the overlay started to stutter at the edges, the data coming in with occasional gaps that the Archivist filled through interpolation. Like reading through a window that kept fogging.
He pulled the full July 19th harvest log. Then he pulled two more dates for comparison: August 3rd and September 14th. Different teams, different dungeon configurations, same pattern. The Association measured a fraction of what the System extracted. Fourteen to one. Thirteen to one. Fifteen to one. The ratio was consistent, which was the point. Not a single anomaly. A systemic pattern.
He copied everything to Bug's format and shut the overlay down to passive monitoring. The bridge settled back to its resting state with the grudging acceptance of something that had been overworked and was going to remember it.
"Done," he told Bug.
Bug took the data and disappeared into his correlation engine. Alex sat in the back room with his eyes closed and the secondary channel running its steady 0.3% and listened to the fieldwork station's sounds. Mira writing in the main room. Echo turning pages somewhere. Maya's footsteps on the exterior step, where she'd been standing watch since 1:00 PM.
The Watcher was out there. Nine kilometers and closing on whatever grid pattern it ran. Prime's clock at three and a half days. Wells' seventy-two hours at fifty-one and counting.
---
Bug came back at 5:30 PM.
"The correlation package is done," he said. He had a USB drive in one hand and a printed summary in the other. "Three dates, three dungeons, full comparison. Association measured output versus System actual extraction. The gap is clean. No noise, no ambiguity. Fourteen to one average, with dungeon-specific variation that actually strengthens the case because it shows the ratio scales with difficulty rating."
"Meaning?"
"Higher-rank dungeons have a higher extraction ratio. Because the emotional yield is higher when people are more scared." Bug set the USB drive on the table. "Wells will understand that immediately. She's been analyzing dungeon data for years. The pattern will make sense to her even before she accepts the explanation."
Alex picked up the drive. Small. Plastic. Containing data that could change Wells' entire understanding of the System she'd been policing for a decade.
"One more thing," Bug said.
He said it in the tone Alex had learned to recognize. The tone that meant Bug had found something he hadn't been looking for.
"While I was pulling the Association's public records for correlation, I ran a broader search on Wells' published work. Quarterly reports, internal memos that hit the public database, conference presentations." He paused. "Six years ago, Wells presented a paper to the Association's internal research board. Title: 'Unexplained Energy Variance in High-Rank Dungeon Monitoring Data.' The paper was rejected. Filed as 'instrument calibration recommendation' and buried."
"She mentioned the anomalies. In our meeting."
"She mentioned noticing them. She didn't mention that she wrote a forty-seven-page research paper about them and kept investigating after the board rejected it." Bug pulled his laptop around. "I found the paper in the Association's archived database. It's classified internal, but the classification is administrative, not security. Sloppy filing. The paper documents energy variance in twelve dungeons across three cities, exactly what she told you. But that was six years ago."
"And since then?"
"Since then, she's been adding to it. Quietly. Off-record. I found fragments across six different Association internal servers. Updated data sets, new dungeon comparisons, expanded analysis. She has a file, Alex. A working file. And it's not a forty-seven-page paper anymore."
Bug turned the laptop so Alex could see the screen. A directory listing. File names with dates spanning six years. Data sets, analysis documents, correspondence logs, cross-referenced dungeon monitoring records.
"Three hundred and twelve documents," Bug said. "Six years of work. All filed under her personal research allocation, which is technically allowed but practically invisible to anyone who isn't specifically looking for it."
Alex stared at the screen. Three hundred and twelve documents. Six years. Wells hadn't just noticed anomalies. She'd been building a case. Alone, without institutional support, against the dismissal of her own research board.
She'd been looking at the same gap he was about to show her. For six years. From the other side.
"Can you access the content?" Alex asked.
"Some of it. The personal files are encrypted. But the analysis documents reference specific data points that I can cross-reference with what you just pulled." Bug's fingers moved on the keyboard. "She's been tracking the same ratio. She doesn't call it fourteen to one. She calls it the 'secondary energy signature.' And she has—Alex. She has three hundred and twelve data points. From dungeons all over the country."
Three hundred and twelve data points. Against his three.
Wells didn't need his evidence to understand the gap. She already knew the gap existed. What she needed was the explanation. The why. The framework that turned six years of unexplained variance into a picture that made sense.
He was going to walk into that meeting with a USB drive containing three data points.
She was going to be sitting there with three hundred and twelve.
"Bug," Alex said. "Pull everything you can from her file. Everything that's accessible. I need to know what she knows before I show her what I know."
"Already started." Bug was typing. "I'll have a summary by midnight."
Alex sat back. The USB drive was still in his hand. Three data points. He'd spent two and a half hours at full bridge output, risked Watcher detection, pushed his 58% integrity to its limits, to pull three data points that Wells had been collecting by the hundreds for six years with nothing but standard instruments and a research question nobody wanted her to ask.
Maya's footsteps stopped on the exterior step. The door opened.
"Something happened," she said. Not a question. She'd read it in the room's energy, the way Bug was typing, the way Alex was sitting.
"Wells has been investigating the harvest gap for six years," Alex said. "She has more data than we do."
Maya stood in the doorway. She looked at Bug's screen. At the directory listing. At the six years of dates.
"Good," she said.
Alex looked at her.
"Good," she repeated. "It means when you give her the framework, she'll already have the evidence to fill it." She paused. "It also means she's been keeping secrets from her own organization for six years. Which means she understands what it costs to know something nobody else wants to know."
The fieldwork station was quiet. Bug typed. The Watcher moved through its grid somewhere in Seoul. Prime's clock ran down. The secondary channel held at 0.3%.
And somewhere in the Association's internal servers, three hundred and twelve documents sat in a personal research file, waiting for a context they'd never had.
Wells was going to get her context in forty-nine hours. Alex just hadn't expected her to be this ready for it.
