How we invest

Brad Wolff directs a small AI research team: four analysts, each with one job. Brad makes the final decisions; the team does the reading, the math and the checking.

One stock, four questions. Every idea goes through the same steps in the same order.

1. What's broken? Screener reads the filings first, looking for dilution, accounting stretches, insider selling and anything the headlines leave out.
2. What's it worth? Graham values the business from its past record, its current health and a realistic range of futures, then compares that value with the price.
3. When? Wick reads price and volume to find levels where buying makes sense, and the level that would prove the idea wrong.
4. How much? Ledger sizes the position so that no single mistake can do serious damage, and keeps the books honest afterward.

Three possible calls. The answer is always one of three: buy (a good business at a price with room for error), wait (a good business, but the price already assumes too much), or steer clear (something is broken, at any price).

Margin of safety. We buy only when the price sits well below our estimate of value, because estimates are often wrong. The less predictable the business, the bigger the cushion we require.

Worked example: NVIDIA, as of the October 7, 2026 close. We valued the shares at about $263, a probability-weighted average of a bear, a base and a bull case. At $237.47, that left a margin of safety of only 9.6%. NVIDIA is an outstanding business, but its earnings are cyclical and its cash flow had just fallen well behind its profits, so we wanted 30%. That points to about $184. The call was wait, with a buying zone of $185–190 and a list of what would change our minds.

What we don't do. We don't chase price targets, hype a setup, or call a trade a success just because it made money. A good process with a bad outcome still counts as a good process, and the reverse doesn't count at all.

Educational project, not investment advice.