About StockXray

A free, non-commercial research tool. AI analysts read public market data and filings and write a report for one stock. Every report anyone runs is published here for everyone to read.

How a report is made

• Understand the business first. The company is matched to one of about 40 business types (for example seat-based software, bank, REIT, cyclical industrial, single-drug biotech). The type decides which metrics matter, which analysts run and which valuation methods apply. Metrics that mislead for a type are switched off (for example accounting-fraud scores for banks).
• Collect data. Prices and estimates (Yahoo Finance), financial statements and segment data straight from SEC filings, the latest 10-K, 10-Q, 8-K and earnings call, rates and economic series (U.S. Treasury, FRED), short interest (FINRA) and options (Cboe). Every number keeps its source and date.
• Business breakdown. Where the money comes from (segments, products, regions), what it costs and where cash goes. Key metrics come from SEC data or from filing and call text; a text number is kept only if its quote is found word for word in the source.
• AI analysts. Specialised AI agents (financials, accounting quality, market, sentiment, guidance, stress test, capital allocation) review the data with instructions for this business type.
• Valuation. Fair value blends several methods chosen for the business type: discounted cash flow with a moat-based competitive-advantage period, peer and own-history multiples, and sector methods (tangible book vs return on equity for banks, FFO for REITs, sum of the parts for conglomerates, drug-pipeline value for biotechs). Growth assumptions are checked against what similar-sized companies actually achieved. Analyst price targets are shown for context only.
• What the price assumes. The price is turned back into the growth it implies and compared with the company's real trend. When a stock looks cheap for a reason (for example a fear that AI replaces the business), the report names the fear, the evidence on both sides and the milestones that would prove it right or wrong, and adds a "fear comes true" scenario.
• Verdict. One rule-based engine combines upside to fair value, business quality and risk checks. The upside needed for a BUY grows with uncertainty. Value-trap, estimate-cut and debt-distress checks can hold a verdict back. A verdict is a model estimate, not advice; there are no position sizes or stop-losses.
• Freshness. Each report shows its date and the age of its data. A new run shows what changed since the last report and how each change moved the fair value. Reports older than 30 days, or made with an older version of this method, are marked "Needs refresh" and left out of the Buy / Hold / Sell filters.

How we check ourselves

• Fair values are compared with analysts in a "blind" mode where the model never sees analyst targets.
• Next-quarter forecasts and milestones are logged and scored against what the company later reports.
• The rule-based core is back-tested on 2012–2022 data using only what was known at each date. The debt-distress check works well (it beats plain debt ratios); the fair value alone has not yet shown a reliable edge, so treat it as one input.

Data sources and attribution

All data below is fetched without a visitor key. Each report lists the sources it actually used at the bottom of the page.

• Yahoo Finance (unofficial, through the open-source yfinance library and Yahoo's public headline RSS feed): quotes, statistics, analyst targets, ownership, options and headlines. Not an official API; quotes may be delayed.
• SEC EDGAR (U.S. Securities and Exchange Commission): company filings, XBRL company facts and frames (quarterly fundamentals, industry peers), SIC industry codes and Form 4 insider transactions. Public U.S. government data, fetched within the SEC's fair-access limits.
• U.S. Department of the Treasury: daily par yield curve rates (the risk-free rate). Public domain U.S. government data.
• FRED and ALFRED (Federal Reserve Bank of St. Louis): macroeconomic series. Some series are owned by third parties and carry their own copyright; see each series' notes on fred.stlouisfed.org.
• FINRA: equity short interest (twice-monthly settlement data) and daily Reg SHO short sale volume. Short sale volume includes market-maker hedging and is not the same as short interest.
• Cboe Global Markets: delayed (about 15 minutes) options data. Only derived figures are shown (implied volatility, expected move, put/call ratios, skew); the raw option chain is never stored or republished.
• Wikipedia: S&P 500 membership and GICS sector from the "List of S&P 500 companies" article, licensed under CC BY-SA 4.0. S&P 500 and GICS are trademarks of S&P Dow Jones Indices and MSCI.
• News headlines from public RSS feeds (Google News, Yahoo Finance and some company press-release feeds): only the headline, publisher, time and a link to the original article are kept. Articles belong to their publishers.
• Earnings call transcripts from the DefeatBeta dataset, licensed under the Open Data Commons Attribution License (ODC-BY 1.0).
• Optional data from Financial Modeling Prep, Finnhub and Alpha Vantage when the person who ran the report supplied keys.
• StockXray is not affiliated with, endorsed by or sponsored by Yahoo, the SEC, the U.S. Treasury, the Federal Reserve, FINRA, Cboe, Wikipedia, Google or any data or AI provider. All trademarks belong to their owners.
• Each report shows the date of its data. Values are a snapshot from the run, not live prices.

How your API key is handled

• To run a report you paste your own AI provider key (for example a Gemini key). It stays in your browser; "remember" stores it in this browser's local storage only.
• The key is sent only with the run you start, in the request body, over HTTPS. The server keeps it in memory for that run and never saves or logs it.
• If you close the page, your run is stopped. Published reports contain market data and analysis only, never your key or who you are.
• You pay your AI provider directly for the requests your run makes (about 20 per report). This site charges nothing.

Limitations

• AI models can misread numbers, miss news or state things with too much confidence. Check important figures yourself.
• Coverage is best for U.S.-listed stocks. Funds, ETFs, very small companies and foreign filers may have gaps.
• Reports do not cover your personal situation, taxes or portfolio, and are not updated after they are published.
• Reports older than 30 days are marked stale. Anyone can run a fresh one.