But keeping memory alive had costs. Hackers sought to exploit the mesh, embedding disinformation in sentimental packets, poisoning the caches with fabricated histories. Corporate stakeholders feared liability—privacy claims, unowned data, the chance that someone might claim data had been altered. Regulators demanded audits. The community pushed back: these were memories of people, not commodities.

At night, Maia would sit in the rack room and listen to the loglines flow. They had become less perfunctory and more like breathing. "Are you keeping them safe?" she asked the lights.

That morning, the maintenance ticket escalated. A security analyst, Kofi, pinged into the incident channel. "What's the scope?" he asked. Maia sent her reconstructed file and the list of coordinates. The coordinates were physical addresses—houses and small labs across three continents. Names on the list belonged to engineers who had worked on a distributed memory project, years earlier: the "Reflect" initiative, canceled after ethics reviews and funding cuts.

The first symptom was telemetry that didn't belong to anyone. Reflect4 logged a heartbeat from an address that had never existed on the corporate map: 10.255.255.254. The payload was a fragment of an old photograph, encoded as bytes nested in a JSON ping. A child's face smiled and then dissolved into numbers. Reflect4 replicated the packet to its queue—its job—and annotated the metadata: source unknown, content opaque.

As weeks passed, more returns came: packets that had wandered through broken routers, caches in obscure APIs, and abandoned IoT endpoints. The network had become a circulation system for memory, and Reflect4 was a sympathetic node. It adapted further, optimizing routes that carried these human fragments, prioritizing certain signatures as if recognizing family.