Our Methodology

The foundation of our analysis is clean, comprehensive data. Unlike simple stock screeners that only look at trailing P/E ratios, our ingestion engines pull from a vast array of sources:

• SEC Filings (10-K, 10-Q, 8-K): We ingest the raw text of corporate disclosures, tracking changes in "Risk Factors" and "Management Discussion & Analysis" quarter over quarter.
• Earnings Transcripts: Natural Language Processing (NLP) is applied to quarterly earnings calls to measure management sentiment, evasiveness, and forward-looking optimism.
• Institutional Order Flow: We analyze historical options chain data, dark pool block trades, and failure-to-deliver (FTD) metrics to gauge "Smart Money" positioning.

Once the raw data is ingested, it passes through our custom-trained Large Language Models (LLMs). These models are specifically tuned on a corpus of historical financial crises, accounting frauds, and massive bull runs.

The AI does not predict the future; rather, it looks for statistical similarities between the current state of a company and historical precedents. If a CEO uses the exact same defensive language that preceded a 40% drop in a similar company three years ago, our model flags it. We extract both a "Bull Case" and a "Bear Case" based purely on these objective findings, ignoring Wall Street analyst upgrades or downgrades.

The final output of our pipeline is the Truth of Market Risk Score (0-100). This score is a weighted composite of several sub-metrics: