Where the numbers come from

FFP One answers from the FFP One Data Layer, a research database built by FFP One's founder, Randall Helms, from a variety of sources, above all SEC filings. This means that the relevant data is sitting there before you ask, and every answer tells you what it drew on. This page covers what goes into that database, how it's checked, and what to keep in mind. For the shorter version, see how FFP One works.

Figures come from the database

FFP One turns companies' SEC filings and funds' N-PORT holdings into a structured research database, alongside daily market prices and macro data from FRED. Standard measures are calculated there in advance, the same way for every company: margins, P/E, return on equity, the Piotroski, Altman Z and Beneish scores, and a fund's look-through figures. The AI looks figures up with database queries; it doesn't answer from a model's memory.

When the AI does its own arithmetic

Some questions need a calculation nobody stored in advance, like growth over five years or one company's revenue against another's. The AI works those out from the figures it looked up, and Chat checks the numbers in its answer against that data before you see it, sending back an answer that doesn't hold up.

Words come from the filings, and they're checked too

Investment research isn't only numbers, so FFP One builds a second path for what companies say. Each filing is split into its sections (the business, the risk factors, management's discussion, cybersecurity) and each section is summarized with its key points, so the AI can work with strategy, risks and the outlook as readily as with the figures.

That path has its own checks. Before anything is summarized, the filing's text is checked against the original SEC document, and a filing that doesn't match isn't summarized. Every dollar figure and percentage in a summary is checked against the section it came from. When an answer quotes a filing, the passage comes from searching the filing's own text, and a Research Report's quotes are matched word for word against the original.

Every chat answer says what it drew on

Each chat answer carries a label: Figures, AI Summaries or Filing Text, Synthesized when it draws on more than one, System Knowledge when it's about FFP One itself, and General Knowledge when it rests on what the model knows (what a term means, say) rather than on FFP One's data. When an answer uses FFP One's data, it also lists the SEC filings it drew on (the company, the form, the year and the sections) with a link to each filing.

Research Reports, and the web

Chat never searches the web. A Research Report can also bring in recent news for context, and keeps it in its own clearly marked section, apart from the filing figures. Each report ends with a section naming the filings and sections behind its figures, and which figures are market prices or calculations, and it gets a second AI pass that re-checks its figures before it's delivered.

Dated, so you know how current it is

Every report shows when it was generated, and an exported copy shows when it was exported. A report is a snapshot of what FFP One had on that day; it isn't updated afterward.

Fund holdings come from each fund's latest N-PORT filing. Funds publish their holdings quarterly and with a delay, so what you see is usually last quarter's portfolio, not this morning's.

What to keep in mind

• The checks are automated: they're software and AI, not a person, and they catch a lot. A figure can still come from the right filing and be the wrong period or the wrong comparison, so check the numbers a decision rests on; every answer shows where they came from.
• Some calculations can't be confirmed: Chat's check covers the figures it fetched and the standard changes between them, not every kind of calculation (a five-year compound rate, for one), so a figure like that is worth a second look.
• A report's second pass can fail: if it can't finish, the report is delivered as first written, with a note at the top saying its figures weren't re-checked.
• Quarterly figures aren't in yet: FFP One has annual reports (10-K and 20-F) and recent 8-K filings; quarterly (10-Q) figures are on the way.
• Research, not advice: FFP One doesn't know your circumstances, and nothing it writes is investment advice. It's built to say when the filings don't answer a question, rather than fill the gap.

Last updated: October 2026