How Buffybot works

In the spirit of Warren Buffett and Charlie Munger: it is far better to own a wonderful business at a fair price than a fair business at a wonderful price. So the bot spends most of its effort deciding what is wonderful — and stays patient on price.

1 · The quality gate (reject most of the market)

Each day the bot scans a rotating batch of the large-cap universe and runs a deterministic quality score (0–100) from the numbers alone — no model involved. It rewards the fingerprints of a durable moat:

• High returns on capital — a moat shows up as ROIC (or ROE for financials) earned well above the cost of capital, consistently, not for one lucky year.
• Consistency & predictability — steady returns, profitable every year, real cash every year.
• Pricing power — fat, stable gross margins.
• An honest balance sheet — low net debt to EBITDA, comfortable interest coverage.
• Cash conversion — reported earnings that actually become free cash flow.
• Capital discipline — a flat or shrinking share count (buybacks), not serial dilution.

Hard disqualifiers — negative equity, sub-hurdle returns, heavy leverage, recurring losses — drop a name straight onto the “too hard” pile no matter how cheap it looks. Most of the market never clears the bar, and that is the point.

2 · The deep dive (the judgment a spreadsheet can't make)

Only names that clear the gate are handed to a frontier reasoning model. It writes the qualitative half Buffett cares about — the moat and whether it is widening or eroding, whether the business sits inside a circle of competence, and the quality of management's capital allocation — and sets one number: a conservative, normalized growth rate for owner earnings. It never sets a price.

3 · Owner earnings & a margin of safety

Value is estimated on Buffett's owner earnings — reported earnings plus depreciation & amortization, minus the maintenance capital expenditure needed to keep the business where it is (approximated as the smaller of capex and D&A). Python capitalizes those owner earnings at a conservative growth rate, discounted at the higher of the CAPM cost of equity and a fixed opportunity-cost hurdle — the question is always “would I rather own a bond?” Net cash on the balance sheet is credited to value (its interest is stripped from owner earnings so nothing is counted twice), and a business the gate scores wonderful earns a longer explicit fade — durability of the moat is exactly what the gate certified, so duration is what it buys.

A reverse DCF keeps the model honest in both directions: every valuation states what growth the current price already pays for, the deep dive judges whether the business can beat that, and credited growth is capped a few points above the market-implied rate so the model can never quietly assume the market is wildly wrong.

The rating starts from price discipline: margin of safety = (value − price) / price. A BUY needs a real discount; a STRONG BUY — a “fat pitch” — needs a fat one. But a business that cleared the quality gate is never rated below HOLD: Buffett doesn't short wonderful businesses, so “expensive wonderful” means wait for your price, not sell. (Our own backtests agree — SELL ratings on quality names carried no information, so we stopped issuing them.) For a full ten-year discounted-cash-flow treatment of the same name, our sibling Vinebot takes the Damodaran route.

Patience is the product

Wonderful businesses accumulate on the watchlist and are re-priced against fresh prices every day (intrinsic value only moves when fundamentals or a new deep dive do) — so a fat pitch rings the day the price breaks, not when the monthly rotation next reaches the name. A cheap name still in a clear downtrend is deferred to HOLD until it stops falling. Most days there is nothing to do — and the site will say so. It exists to wait, and to shout when a wonderful business finally goes on sale.

Honest limits

Owner-earnings maintenance capex is an approximation; banks and insurers are scored on different metrics and remain hard to value; multi-year history from free data is shallow; the model's moat judgment can be wrong. This is a research experiment, not advice.

Buffybot is an automated research experiment that screens for durable-moat 'wonderful businesses' and estimates a conservative owner-earnings value using public data and large language models. It is NOT investment advice, a solicitation, or a recommendation to buy or sell any security. Outputs may be wrong, stale, or incomplete. Do your own research.