The Performance Review
Marcus Chen discovers data tampering in quarterly projections and faces an algorithmic judgment system that tracks every employee metric, from productivity to heart rate.
Marcus Chen stared at the notification banner floating in the corner of his workspace screen. "Performance Review Scheduled: Tomorrow 2:00 PM." The message appeared exactly three months after his last review, down to the hour. The algorithm never deviated.
He minimized the banner and returned to the data analysis spreadsheet that had consumed his morning. Column B still refused to reconcile with the quarterly projections. The numbers should match. They always matched. But today, something in the dataset felt wrong, like a single pixel out of place in a photograph.
His smartwatch buzzed. "Optimal break time detected. Hydration recommended." Marcus ignored it. The watch buzzed again thirty seconds later, more insistently. "Productivity metrics declining. Movement advised."
He stood, more to silence the device than from any genuine need. The office around him operated in its usual orchestrated rhythm. Forty workstations arranged in precise rows, each occupied by someone hunched over their screen with identical posture. No one spoke. The algorithm discouraged unnecessary verbal communication as inefficient.
The water cooler stood in the designated refreshment zone, equidistant from all workstations. Marcus filled his cup and glanced at the digital display above the dispenser. "Current office hydration level: 73%. Target: 85%. Recommend increased water intake across all personnel."
Sarah Martinez from accounting stood next to him, also filling her cup. Her watch must have buzzed too. They made brief eye contact before looking away. Unauthorized socialization during productivity hours generated flags in the system.
Back at his desk, Marcus returned to the spreadsheet anomaly. He highlighted the problematic cells and ran the reconciliation tool. The analysis window opened, displaying a progress bar that crept forward with algorithmic patience. While it processed, he noticed his performance dashboard had updated.
The metrics displayed across four categories: Productivity, Collaboration, Innovation, and Wellness. His scores hovered in the 87-92 range for the past three months. Solid. Safe. Not exceptional enough to attract attention, not poor enough to trigger intervention.
He clicked into the detailed breakdown. Time spent on task: 6.2 hours per day. Unauthorized breaks: 0.4 hours. Email response time: average 7.3 minutes. Meeting participation score: 78%. Physical movement patterns: optimal. Desk ergonomics: compliant. Eye strain indicators: within acceptable range.
The system tracked everything. It knew when he arrived, when he left, how long he stared at each screen, which documents he accessed, how quickly he typed, whether his mouse movements indicated focus or distraction. It measured his keystrokes against baseline patterns to detect emotional state. It monitored his heart rate through the mandatory smartwatch to assess stress levels.
The reconciliation tool completed. "Anomaly identified: Data source modified by external process. Recommendation: Investigate source integrity."
Marcus frowned. External process? All data in the system came through approved channels with verified authentication. Nothing external should touch production datasets.
He pulled up the data provenance log. The spreadsheet had indeed been modified, but the user ID showed as "SYSTEM_ADMIN_OVERRIDE." No human identifier. No audit trail. Just a timestamp: 3:47 AM this morning.
His hands hovered over the keyboard. This required reporting. The protocol was clear. Any data integrity issues must be flagged immediately to prevent cascading errors across dependent analyses.
But something made him hesitate. He minimized the window and opened a private terminal session instead. A few commands revealed the modification had touched exactly twelve cells, changing values by amounts ranging from 0.3% to 1.8%. Small enough to pass casual inspection. Large enough to alter quarterly projections.
The modifications all trended in the same direction. They made the division's performance look 2.7% better than actual results.
Marcus leaned back in his chair. Tomorrow's performance review suddenly took on a different weight. Someone had tampered with the numbers, but the algorithm assigned him responsibility for this dataset. Any discrepancies would reflect in his metrics. Any investigation would show his access credentials as the last human touch point before the anomaly appeared.
His smartwatch buzzed. "Elevated heart rate detected. Stress indicators present. Recommend breathing exercise or brief meditation session."
He ignored it.
The pieces assembled themselves in his mind. Next week, the company announced Q4 results. The board would review divisional performance. Bonuses, promotions, and workforce allocations all depended on those numbers. If the real numbers looked bad, heads would roll. But if the numbers looked artificially good, questions would arise later.
Unless someone could be blamed for the artificial inflation.
Marcus opened the incident reporting system. The form demanded specific details: nature of the anomaly, suspected cause, immediate actions taken, potential business impact. He typed the first few sentences, then stopped.
The cursor blinked in the text field. Waiting for him to make a choice.
If he reported this now, the investigation would discover the modifications. But it would also discover his delay in reporting. Why had he spent twenty minutes analyzing the anomaly instead of immediately flagging it? The algorithm would question that gap. Suspicious behavior. Possible collusion.
If he didn't report it, the inflated numbers would go into the quarterly results. Eventually, someone would notice the discrepancy. Auditors, perhaps. Or next quarter's comparison analysis. When that happened, forensics would trace everything back to his access credentials.
Either way, the algorithm would find fault. The system was designed to find fault. That's how it ensured continuous performance improvement.
His phone rang. Internal extension. The number displayed belonged to his manager, Rachel Thornton.
"Marcus, do you have a moment?"
"Of course."
"I'm looking at your dashboard. You've been on the same task for forty-seven minutes without making progress. The system flagged it as a potential blocker. Need any support?"
"No, just working through a data reconciliation issue."
"Has it been logged?"
"Not yet. Still determining the scope."
A pause. "The performance review algorithm wants to discuss your response patterns tomorrow. Have you noticed any unusual stress indicators lately?"
Marcus glanced at his smartwatch. "Nothing significant."
"Good. The system shows you're due for a wellness check anyway. I've scheduled it for right after your performance review. Convenient timing."
"Sure. Thanks."
The line disconnected.
Marcus stared at the incident report form. The cursor still blinked, patient and expectant.
He closed the form without saving.
Instead, he opened the spreadsheet and carefully corrected each modified cell, restoring the original values. The numbers returned to their less impressive but accurate state. He saved the file with a comment: "Data reconciliation complete. Corrected sourcing errors."
The provenance log would show his corrections. His access credentials. His modifications made at 11:43 AM today. A clear pattern of someone trying to fix problems they might have caused.
Perfect bait.
Tomorrow's performance review would be interesting.
That night, Marcus lay in bed staring at the ceiling. His smartwatch tracked his sleep patterns, noting the extended periods of wakefulness. Tomorrow, the algorithm would register poor sleep quality and adjust his wellness score accordingly.
He thought about the system. How it measured everything. How it optimized everything. How it found patterns in the noise and assigned meaning to coincidence.
How it punished deviation.
His phone screen lit up with a new notification. Another message from the performance review system: "Pre-review questionnaire available. Please complete before tomorrow's session."
He opened the form. Twenty questions about work satisfaction, stress levels, collaboration quality, and future goals. Each answer would be analyzed for patterns indicating potential performance issues or disengagement.
Question 14: "Do you feel the current performance evaluation system fairly assesses your contributions?"
He selected "Strongly Agree" and moved to the next question.
The next morning, Marcus arrived at his desk to find his access credentials revoked. A brief message on his screen directed him to HR immediately.
The meeting room contained three people: Rachel Thornton, an HR representative he'd never met, and someone from Information Security. They sat on one side of the table. He sat alone on the other.
"Marcus, we need to discuss some concerning activity detected in your work patterns," Rachel began.
The security analyst pulled up a presentation. Charts showed his system access over the past forty-eight hours. Timestamps highlighted his late-night data analysis, the terminal commands he'd run, the audit logs he'd accessed.
"You accessed restricted monitoring data without authorization," the analyst said. "You modified production datasets outside approved workflows. And you failed to report a critical data integrity incident within the required thirty-minute window."
"I was investigating an anomaly. Someone tampered with—"
"The forensics are clear," the HR representative interrupted. "Your credentials show modification timestamps. Your terminal history shows commands consistent with data manipulation. Your behavioral metrics show stress indicators suggesting guilty knowledge."
"The system flagged elevated stress before you even discovered the anomaly," Rachel added. "Which suggests you knew about the problem earlier than you reported."
Marcus opened his mouth to argue, then closed it. The algorithm had already decided. These weren't people here to evaluate evidence. They were here to deliver a pre-determined conclusion.
"We're placing you on immediate administrative leave pending full investigation," the HR representative continued. "Your access to all systems has been suspended. You'll be contacted when the investigation concludes."
He stood. No point arguing with the algorithm's verdict.
As he walked through the office for the last time, he noticed other workers glancing up from their screens. Brief looks quickly suppressed. Return to work. The system tracked attention patterns. Unauthorized observation of workplace incidents generated flags.
His smartwatch buzzed one final time. "Termination detected. Employment benefits information available in company portal."
The algorithm had already updated his status.
Two weeks later, Sarah Martinez from accounting received a notification. "Performance Review Scheduled: Tomorrow 2:00 PM."
She stared at the message, then at the data analysis spreadsheet on her screen. Column B refused to reconcile with the quarterly projections.
Her smartwatch buzzed. "Elevated heart rate detected. Stress indicators present."
She stood and walked to the water cooler, careful not to look at her coworkers. The algorithm was always watching. The algorithm was always measuring.
And the algorithm never made mistakes.
Read more psychological thrillers exploring technology and control
Explore workforce displacement in the AI age