I was asked to optimize conference room utilization across Arcturus's GLMZ properties. Standard facilities work — identify underused rooms, flag overbooking patterns, recommend reconfigurations. The analytics platform pulls calendar data from the enterprise scheduling system, cross-references with badge-access logs and environmental sensor data (motion, CO2 levels, thermal signatures) to determine actual occupancy versus booked occupancy. Most buildings show the expected pattern: rooms booked at 60-70% capacity, actual occupancy around 40%, the usual gap between what people schedule and what they attend.

Then there are the Ghost Buildings. I didn't know they were Ghost Buildings when I started. The analytics platform doesn't label them. They're just building codes in a database. But the data was immediately wrong. Conference rooms in seven buildings showed 100% booking utilization — every room, every slot, every day, going back as far as the scheduling system retained records, which is four years. One hundred percent. Not 99%. Not 98% with occasional cancellations. Every single room booked for every single available hour of every single business day for 1,460 consecutive days.

The meeting titles are plausible. That's what makes it unsettling rather than merely anomalous. They're not test entries or placeholder text. They're the kinds of meetings that actually happen in corporate environments: "Q3 Revenue Alignment," "Product Roadmap Sync," "Client Onboarding Follow-up," "Cross-functional Sprint Review." The titles follow naming conventions consistent with Arcturus's corporate culture. They use the right abbreviations. They reference real quarterly cycles. Someone — or something — that understands how this company talks has been generating meeting titles at scale for four years.

Each meeting has an attendee list. The attendees are drawn from the pool of 340 employee IDs that the Municipal Authority's HR investigation flagged as anomalous. The IDs resolve in the employee directory to names, titles, and department codes. Click on a name, and you get a profile: photo placeholder (always the default silhouette — no one has uploaded an image), desk assignment (always in a Ghost Building), reporting chain (always terminating in a manager whose profile is also a default silhouette). The organizational chart of ghost employees is complete, hierarchical, and functions exactly like a real org chart, except that no human occupies any position in it.

I pulled the meeting room environmental data. The rooms are empty. Motion sensors have not triggered in years. CO2 levels are ambient — no breathing. Thermal signatures are flat — no body heat. The rooms are booked. The meetings are scheduled. The attendees are listed. Nobody comes. The meetings happen anyway, in whatever sense a meeting can happen without participants. The calendar system records them as completed. No minutes are filed. No action items are generated. The next meeting begins on schedule.

I submitted my utilization report with a recommendation to release the conference rooms in the affected buildings for reallocation. My recommendation was rejected by a system administrator I have never met, working from a terminal in one of the Ghost Buildings. The rejection message was polite, professional, and followed standard Arcturus communication protocols. It said the rooms were in active use and could not be released at this time.