Tonality + scenario stories

Where the stories that move markets get scored

Fed-Greenbook tonality regression, AI-generated scenario narratives, narrative-risk by topic. Plugs into the US-Impact engine as the narrative channel.

Live tonality — cross-link

Per-topic news tonality lives at /sentiment with a 9-topic table refreshed every 3 minutes from /api/proxy/news/headlines.

Engine narrative channel

--
Politics signal
--
Conflict signal
--
Markets stress

Tonality models in production

Fed-Greenbook tonality

Regress Fed/IMF/CBO narrative tone vs forward GDP / unemployment / inflation. 1-4 quarter lag. R²~0.32.

Per-topic keyword scoring

9 topics × pos/neg keyword sets — politics, war, markets, health, AI, border, drugs, riots, elections. Used on /sentiment.

War-tone (Permutable-style)

Region-tagged conflict tone. Drives the US-Impact pass-through formula.

AI-generated scenario narratives

Local Ollama (gemma3:1b) generates 24h / 7d / 30d scenario stories per category. Cached 30 min on the FastAPI side.

Per-category synthesis

The engine pulls /api/world/predictions?category=<world|conflicts|markets|disasters|cyber|space|local> on demand. Each category gets its own gemma3:1b prompt with the per-category sitrep as context. First-load 5-25s; subsequent loads from 30-min cache.

← Sentiment · World · US-Impact engine