Why it exists
A company’s filings are the most reliable thing anyone publishes about it, and the hardest to read. A 10-K runs to hundreds of pages. The accounting lives in the footnotes. The most important change from one year to the next can be a single word.
The tools that exist mostly hand you one of two things: a database of numbers with no way to see where they came from, or a chatbot that writes figures from memory and states them with confidence. We wanted a third thing: a reader that does the arithmetic itself, shows its working, and tells you what is new.
How we build it
- Numbers are computed, never generated. Ratios, forensic scores, segments and the footnote comparison come from a fixed engine reading the XBRL data a company itself filed. The same engine runs in the website, the Chrome extension and the backend, so they cannot disagree.
- The AI judges, and says so. A model writes the summary, the findings, the tone and the materiality ratings, and each is labeled as AI. It never produces a figure.
- Every claim shows its source. A finding carries the exact passage it rests on and opens the filing. If the passage cannot be found, the claim is downgraded and says so.
- We say what we cannot do. A score the engine cannot compute for a filing is shown as unavailable, with the reason, not as a guess.
The forensic scores come from published research, and each has a plain-English guide that says how to read it and, just as important, what it does not tell you. Start with the Beneish M-Score and the Sloan ratio.
What it is not
FilingEye is a research tool, not investment advice. It does not recommend any security, and it does not replace reading the parts of a filing that matter to you. It is built to take you to those parts faster and to show you what you would have missed.
Where it stands
FilingEye is not open for sign-ups yet. The analysis you can try on the home page and the worked examples on the product page are computed by the same engine that will run for you, from real filings. To be told the day it opens, ask for early access.
Questions, corrections and ideas are welcome at [email protected], and the answers to common questions are on the questions page.