**I built a Buffett-style company checker based on his shareholder letters. Looking for a few people with a finance background to try to break it before it goes public.**
Heads up tho, there's AI in this. If that's an instant no, skip the post, no hard feelings. But 30 seconds first, because it isn't AI slop. As someone with a programming/software background for more then 10 years i can safely say that what is happening in the world right now, is bonkers. & what people say about spending way too much money on Opus and Fable is also true... it is to good not to notice.
**The AI doesn't do the sums.** Owner earnings, return on equity, debt, the dollar test, the prices — plain arithmetic on the figures in the company's own filings, a test on every formula. Where a filing leaves a figure untagged, more common in European reports, Claude can propose the number with the quotation it came from and you approve or reject it; anything unverified stays out of the calculations.
What it *does* do is draft the judgments the figures can't settle, which is what Buffett's filters actually turn on: moat, management, whether current trouble is temporary or structural, how many more years growth can last. You can answer those yourself from the annual report — the check works fully that way, free, no account. Or Claude drafts them in three passes, a cheap model reading the filings and keeping only facts whose quote it can find word for word, a mid-priced one searching an approved list of sources for what the report can't know, and Opus judging on those checked facts alone. Nothing counts until you accept it. Happy to go into that pipeline properly in the comments if anyone wants it.
**The check itself.** Type a US ticker, or an ISIN or LEI for an EU company:
* **Owner earnings** (1986 letter): what the owners could actually take out, not the accounting profit
* **Return on equity with the debt right beside it**, so borrowing can't fake a great return
* **The $1 test** (1984 letter): did every dollar kept inside turn into a dollar of market value? (Needs year-end prices the importer doesn't fill yet, so it shows "unknown" until you type them)
* **Balance sheet and capital allocation**: debt, buybacks, dividends
* **Price**: earnings against the long-term government bond, a margin-of-safety price, and how much growth today's price already assumes
No price targets, no analyst estimates, no beta, no ten-year forecast. The base case assumes no growth at all: what the business is worth if it just keeps earning what it earns. For a company that isn't growing, that's the whole answer. Growth only enters where the company's own record shows it, as a ladder — 5, 10 or 20 more years, then holds steady, paid for out of its own earnings. No growth in the record, no ladder. So there's no single magic "fair value": there's a no-growth price, and one estimate above it for each number of years you're willing to believe in. That's also the one place an AI judgment reaches a figure — accepting Claude's answer on years of growth moves the estimate, by up to 39% on some companies. I'd rather say that plainly than have you find it.