Evelyn Hart had learned to listen to the city in a way other people called noise. On the screens that lined her desk, streams of temperature, vibration, camera metadata and audio levels flowed in a slow, mechanical tide. She could tell which crosswalk sensor would misreport after a thunderstorm and which sewer pump would ratchet up as the morning deliveries arrived. She took pride in records: tidy queries, reproducible joins, margins of error that lived in spreadsheets with her initials on them.
Tonight the tide felt wrong. It had been wrong since Sam Keller didn’t show up for his shift. Sam had called her from a cracked sidewalk three nights ago, voice laughing about the city’s old wiring, promising to swing by with coffee and a ridiculous pastry. He hadn’t answered messages. He hadn’t shown up for the patch-up at a maintenance pit he had flagged. His badge had logged out of the system and then stopped. No one had filed the kind of report that made a desk full of men in uniforms run. The radio had offered polite questions and blank faces.
Evelyn’s screens kept whispering at her in raw numbers. She told herself she was only checking the logs. She told herself she was following procedure. She started by pulling Sam’s access trail: the maintenance grid, a half-hour where his badge pinged an inspection node under the north overpass, the node’s diagnostic dump that had been flagged as ‘corrupted’ by the field software. Field software that had a half-life of patches and duct tape and long weekends.
She threaded his timestamp through camera feeds. The camera over the overpass had a three-second blackout where nothing was recorded; the metadata showed a frame drop flagged as transient. Later a traffic sensor nearby had recorded a microdrop in its ambient amplitude at four successive readings. She logged those drops, annotated them, and then widened the net. If Sam’s disappearance was a data point, Evelyn’s habit was to treat it like algebra—look for constants, for repeating features that could be factored out of noise. She ran a query across three months of sensor telemetry, set to isolate brief, low-amplitude anomalies that coincided with human presence.