What this is

A machine that reads public financial data about small American companies, scores them by a fixed set of rules, and publishes the twenty-five that score highest. Nobody chooses the companies. Nobody reads about them. The rules are written down, they do not change, and the same rules applied to the same data would produce the same list tomorrow.

What it does, each time it runs

It starts from every company filed with the American financial regulator — about 9,909 of them — and throws almost all of them away. Too big, too small, too cheap, too rarely traded, too little revenue: roughly 1,697 survive as the right size, and about 909 have enough measured about them to be scored at all. Those are ranked on three things, in these proportions:

• Growth (40%) — is revenue rising, and is it rising faster than it was? Compared against other companies in the same industry, so a good year for mining does not make every miner look clever.
• Momentum (40%) — has the share price been rising, divided by how jumpy that price is. A steady climb counts for more than a violent one.
• Quality (20%) — debt, cash, whether margins are improving, and whether the company keeps issuing new shares.

Price is deliberately not part of the score. This looks for companies that are growing, not companies that are cheap — those are different questions and mixing them produces an answer to neither.

The six portfolios, and why there are six

Six imaginary pots of money, each with a notional $100,000 — these are American shares, priced in dollars — all buying the same twenty-five companies and all paying the same trading costs. They differ only in their rules. One holds everything in equal amounts and never sells early — that is the control. The others add one idea each: bigger bets on higher-ranked companies; an automatic sell if a holding falls far enough; holding back cash when the market is falling; and all three together.

The point of the plain one is that it is the thing to beat. Any clever rule has to earn back the extra trading it causes. If the clever books do not finish ahead of the simple one, the cleverness is costing money for nothing. The books have been running for 19 days. The plain one is -0.97%, of which 0.66 percentage points went on the cost of trading alone.

No money is involved anywhere in this. No order has ever been placed. These are simulations, and simulations leave out the things that hurt real traders most: the price moving against you as you buy, tax, and the nerve required to keep following a system through a bad stretch.

What it is trying to find out

One question: does ranking small companies this way beat simply buying the whole small-company market, after paying for all the trading it takes? That last clause is where most systems like this quietly fail.

To answer it, every published list is checked again about a week later and the result is written down and never touched again. 2 of the 12 separate weekly checks needed are done. The answer gets judged on 2026-12-14, against a rule written in advance on 2026-09-23, before anyone knew how it would turn out. That rule can say “stop”, and none of its outcomes allows real money to be risked.

Things worth distrusting

• It is young. A few weeks of results tell you almost nothing, and anyone reading a short run of numbers as skill is fooling themselves.
• All the company data comes from one free supplier. If that supplier is wrong about a company, so is this.
• Nothing here understands any business. It has an industry label, some filed figures and a price history. It cannot tell a genuine growth company from an accounting artefact.
• Buying companies that have already risen is a well-known approach that works until it stops, usually suddenly.
• The trading costs are estimates. Real ones in companies this small are often worse.

None of this is investment advice, and it is not written by anyone qualified to give any. It is a record of what a set of fixed rules did, published so that it can be checked.