Signal intelligence
Seven layers.
One clear verdict.
Finsight Core synthesizes seven independent intelligence layers into a single directional signal for every stock we cover. Here's exactly what each layer reads — and why it matters.
News & sentiment analysis
Every news article, earnings call transcript, and press release is parsed using large language models trained to detect tone drift, topic emergence, and sentiment trajectory — not just positive/negative polarity. We track how a company's narrative is evolving over weeks, not just whether today's headline was upbeat. Sustained tone deterioration across earnings calls has historically preceded corrections 7–14 days before the price chart showed any signal.
Organizational behavior
C-suite departures, board composition changes, reporting structure shifts, and sudden hiring or firing sprees. Companies behave differently before significant events — our models capture those behavioral fingerprints and compare them against historical patterns that preceded both corrections and breakouts.
Insider filing patterns
SEC Form 4 filings show when executives buy or sell their own company's stock. But a single transaction is noise. We analyze clustering — when multiple insiders are all net buyers over the same window, especially when it's unusual relative to their historical behavior, the signal becomes far more meaningful.
Institutional flow
13F filings reveal where funds with over $100M AUM are building or trimming positions. We track quarter-over-quarter changes in institutional ownership, identify when large holders are quietly accumulating below current prices, and flag when the divergence between institutional behavior and price action is historically significant.
Earnings call linguistics
The specific language used by executives on earnings calls correlates measurably with subsequent price performance. Phrases that hedge future guidance, deflect analyst questions, or shift blame carry different predictive weights than direct affirmations. We score every call across 40+ linguistic dimensions and compare to each company's own historical baseline.
Alternative data signals
Data sources that don't appear in any financial statement but reliably precede them: job posting velocity on LinkedIn and Indeed, app store rating trends, shipping and logistics satellite data, credit card spend proxies, and web traffic patterns. These signals lead fundamental data by weeks, giving you a view into operating performance before it's reported.
Macro correlation
Company-level signals don't exist in a vacuum. We score every signal against the macro backdrop — sector momentum, broader market risk-off indicators, rates regime, and cross-asset flows — and adjust confidence scores accordingly. A strong individual signal in a deteriorating macro environment gets flagged differently than the same signal in a supportive one.
Confluence model
Layers vote.
Confluence wins.
Each layer independently scores direction and strength. The composite signal is a weighted vote — layers with stronger historical accuracy for a given stock type carry more weight. The final confidence percentage reflects how much of the evidence stack agrees.
A signal firing from a single layer is noise. Four or more layers agreeing in the same direction is a high-confidence event that clears our minimum threshold for surfacing to users.
Example: NVDA signal build
News sentiment
+88
Insider filing
+95
Institutional flow
+76
Earnings tone
+82
Alt data
+71
Composite confidence
92%
See these signals
on your watchlist.
Start your free trial and get live intelligence across all seven layers for 4,000+ stocks today.