How the AI picks are chosen

Before any angle runs, a company must have a real price, a computed moat score and rating, and a data quality score of at least 55, meaning most of its model inputs actually came from filings rather than being guessed. Each company can be picked once. If fewer than six names clear the bar, the leftover slots fill from the overall rating, and if the data is too thin even for that, they stay empty. A blank slot is more honest than an invented reason.

Each pick is the best qualifier on one angle, and each angle is a plain filter over the fundamentals.

Quality on sale. Moat score 60 or better, debt to equity under 1.5, price at least 15% below our fair value estimate. Ranked by the discount.

Income it can afford. Dividend yield between 3% and 10%, debt to equity under 1.2, profit margin above 5%. The cap matters: past 10%, a yield is usually a cut being priced in.

Beaten down, not broken. Price at 65% or less of its 52-week high, moat score 55 or better, debt to equity under 1.5. Ranked by moat, because the fall is only interesting if the business survived it.

High returns, low debt. Return on equity above 18% but capped at 100% to exclude shrunken equity bases, debt to equity under 0.6. The point is returns funded by the business, not by borrowing.

Growth, not fully priced. Revenue growth above 15% on a P/E below 30. A combination the market rarely leaves lying around, which is itself a warning worth reading.

The steady hand. Beta below 0.8, profit margin above 10%, moat score 60 or better. The defensive slot.

The reason printed under each pick is assembled from that company's own numbers, and every pick states its risk in the same breath, because a pick without a bear case is an advert.

Below the six sits a stricter test: names where three independent signals point the same way at once. A name needs at least one member of Congress disclosing a purchase AND at least one tracked fund holding it in its latest 13F; where we hold fundamentals, a price below our fair value estimate and a moat score add weight. Congressional buying is weighted hardest because it is the rarest signal, fund holders are capped so mega-caps cannot dominate, and disclosed sells count against. Each signal alone is noise; the overlap is where it gets interesting.

The six picks are recomputed from the fundamentals every time the data load refreshes, which happens on each site build; the picks page shows the date of that load. The conviction row reads the latest congressional disclosures and 13F holdings on each visit. The same selection function drives the picks widget and the Market Pulse map, so the two can never disagree about which six names made it.

A 13F can already be 45 days old the day it is filed, and congressional disclosures lag the trades they report, so treat those dates as history, not news. The fair value model is an estimate whose assumptions are documented on the main methodology page, along with where it is weakest. The screens read trailing numbers and cannot see why a price fell or what tomorrow's news says. And none of this is a recommendation: the picks are a starting point for your own research, as the disclaimer spells out.