Chapter 3: Probability of Impact

A month after the incident at North Science Hall, campus life had absorbed the event and filed it away as something mildly interesting that had happened to other people. Students referred to it the way they referred to snow days, fire drills, and power outages—temporary disruptions to the real business of assignments, gossip, exams, and weekends. Rumors had inflated and collapsed in cycles. Some claimed a chemical leak had nearly poisoned half the science department. Others insisted a professor had heroically shut down an exploding machine. The official statement from the university was deliberately vague, mentioning preventative evacuation measures, equipment malfunction, and a successful emergency response. It contained no names, no causes worth discussing, and certainly no mention of the student who had arrived at the building before the alarm ever sounded.

Adrian preferred that omission. Public recognition was attention, and attention was friction. He wanted neither.

He kept Aegis private, stored behind layers of encryption and disguised beneath ordinary project folders on his laptop. To anyone glancing at his screen, it looked like a dull collection of coursework and technical documents. In reality, the system had grown significantly since that rainy morning. Adrian had rewritten major portions of the inference engine, improved anomaly detection, expanded municipal and campus data feeds, and built better methods for separating meaningful signals from useless noise. It processed faster now. It explained itself better. It made fewer reckless assumptions.

And yet, as its performance improved, its outputs had become stranger.

Originally, Aegis had focused on systems: traffic flow, building failures, crowd compression, infrastructure strain, utility interruptions. Those made sense. Machines broke. Roads jammed. Crowds panicked. Pressure accumulated and then released. But recently the system had begun generating localized risk profiles tied not to structures or neighborhoods, but to individuals moving through environments. It did not claim destiny or inevitability. It merely modeled exposure—how a person’s routine intersected with hazards, timing, stress, weather, and chance. Even so, Adrian disliked the direction immediately.

It felt invasive.

Worse, it felt plausible.

Late on a Thursday night, while rain threatened again beyond the dormitory windows, Adrian was reviewing model outputs when a profile card expanded on the center monitor. He almost dismissed it automatically before reading the name.

MAYA BENNETT Localized Injury Risk: Elevated Time Window: Next 72 Hours Severity Estimate: Moderate to Severe Likely Outcome: Lower Limb Fracture He stared at the screen without moving.