Nousvigil
📡
📡 In Dev

Cold Sonar

A cold sensor for what a media stream is actually made of — noise, fear, and division, measured, not felt.

Type
Data Tool
Languages
TypeScript · Python
Platforms
WebSelf-hosted
kenoma nlp media-literacy open-source

Vigilus — Kenoma asks its question through fiction: what if the world you trust isn’t the world you truly see? Cold Sonar asks the same question of an actual, ordinary news feed — and answers it with a number instead of a story.

What it measures

Cold Sonar ingests RSS streams, runs named-entity extraction over the text, and computes a single ratio: bruiaj (noise) — the fraction of a story that isn’t load-bearing information. Everything that’s left after you strip out the entities and facts a story is actually about is fear language, urgency framing, and division rhetoric, quantified rather than sensed.

A topic gate keeps the sensor pointed at what matters — politics, geopolitics, economy, disaster, public health — and out of sports scores and celebrity gossip, where the same rhetorical noise exists but the stakes don’t.

Built to survive its own infrastructure

The domain logic — the black box that computes bruiaj, the topic gate, the use cases — has no vendor in it at all. Cloudflare and a local SQLite/cron setup are just two interchangeable adapters underneath the same core. If the cloud side ever needs to disappear, the migration path is documented as a fact, not a hope: stop Cloudflare, run the same local cron and serve commands on a plain Debian box, same domain logic underneath.

Without spaCy installed, the entity extractor falls back to a heuristic — the same formula, a narrower signal.

Status

The extraction pipeline, topic gate, and both adapters (local SQLite and Cloudflare D1) run today. It shares its conceptual universe with Vigilus — Kenoma but is a working tool, not a story: same question, answered in software instead of narrative.

← Back to Portal