News intelligence: the deterministic why-it-moved brief.
News Intelligence answers the question every investor actually asks: why did this move today? It fuses the day's price action (move size, overnight gap, volume versus average) with ranked headline catalysts into a structured, reproducible brief, and flags the macro events worth watching next.
Deterministic, not a black box
The news brain combines price facts with naive, transparent keyword sentiment over headlines to rank likely catalysts, then writes a one-to-three sentence summary sizing the move, noting whether volume confirmed it, and attributing the top catalyst. Sentiment scoring is deliberately explainable rather than a hidden model, and the whole output is reproducible from its inputs. It is the structured input an LLM layer polishes later, so the facts stay auditable.
Headlines plus the calendar
Headlines come from Polygon news when keyed and a labeled mock feed otherwise. A curated economic calendar surfaces upcoming macro events (rate decisions, jobs, inflation prints) with beginner explanations, so the brief covers both what just happened and what could move the name next. Gaps are called out explicitly: a large opening gap is flagged as news that landed outside market hours and was priced in before the bell.
- Why-it-moved brief: move size, overnight gap, volume versus 20-day average
- Ranked headline catalysts with transparent keyword sentiment
- One-to-three sentence plain-English summary, fully reproducible
- Curated macro calendar with beginner explanations of each event
- Explicit gap detection: news that repriced the stock overnight
- Structured output designed as clean input for the AI language layer
Questions
Does an LLM write the brief?
No. The core brief is deterministic and reproducible from price and headlines. An LLM layer only polishes the language afterward, so the underlying facts stay auditable.
How is sentiment scored?
With a transparent keyword approach: positive and negative words are counted per headline and averaged. It is explainable on purpose, good enough to rank catalysts before any model runs.
Are the headlines live?
They come from Polygon news when a key is configured, and a clearly labeled mock feed otherwise, with the provider reported on every response.