How it works

What the system does each week, what it deliberately doesn't do, and the honest limitations.

The weekly cycle

1. Every Friday, the model scores the universe. It pulls a year of daily price history for roughly 1,600 US stocks and ranks them all against each other.
2. The top 40 are published to anyone with an account. Published Fridays at 15:00 America/New_York.
3. You choose which to buy, enter your amount, and get exact dollar figures.
4. You place the orders yourself in your own brokerage. Nothing here touches your account.
5. Optionally record what you bought and prices are captured after each close so you can see how it's doing.

Buying fractional shares, step by step

The exact wording differs by broker, but every one that supports this has the same toggle somewhere on the order screen.

1. Make sure you have enough settled cash — money from a recent deposit or sale may take a day or two to become available.
2. Search for the ticker.
3. Tap Buy.
4. Switch the order from shares to dollars. Look for "Buy in dollars", an "Amount" field, or a small $ / Shares toggle.
5. Type the dollar amount we calculated.
6. Leave it as a market order unless you specifically want otherwise, review, and submit.
7. Repeat for each ticker.

Where the toggle lives, by broker

• Robinhood"Invest in dollars" on the order screen
• FidelitySet Order Type to "Dollars" in the quantity field
• SchwabUse "Stock Slices" for fractional purchases
• AlpacaEnter Notional rather than Qty
• Public / SoFiDollar amounts are the default

How your money is divided

Evenly, across whatever you selected — with one detail that matters more than it sounds.

A naive split computes "amount ÷ 10" once and tells you to buy that exact figure ten times. Cents don't divide evenly, and whole-share names consume a different amount than planned, so the last line ends up asking for money that isn't there.

Choosing your own

The top 10 score highest, and that's the default. All 40 are published so you can substitute.

• Screening out businesses you won't own. Many people exclude conventional banks and insurers, alcohol, gambling, or defence — on religious or ethical grounds. Deselect and pick further down.
• Small accounts and whole-share names. A few tickers can't be bought fractionally; they're flagged.
• Names you already hold elsewhere and don't want doubled up on.

What the model looks at

A gradient-boosted classifier trained on daily price and volume history. Its features describe momentum across several horizons, position within a trading range, realized volatility, and how strong a move is relative to how choppy the stock has been. Scores are compared across stocks rather than in absolute terms, so the output is always a ranking of the universe against itself.

It was trained on 2017–2022 and evaluated on later periods it never saw in training.

What isn't published, and why

You get the tickers and their order. Not the scores, the feature values, the model file, or the parameters — those are the product. Every endpoint returns at most an ordered list of symbols, and the model never leaves the server.

Honest limitations

• Ten stocks turned over weekly is concentrated and high-variance. Expect drawdowns that would be alarming in a diversified portfolio.
• Backtests flatter. They run on stocks that exist today, which quietly excludes companies that failed along the way.
• Weekly turnover has tax consequences. In a taxable account nearly every gain is short-term, taxed at ordinary income rates. This may matter more than any edge the strategy has.
• A model that worked can stop working. Nothing here adapts to that.
• Trading costs and spreads are real, especially on thinly traded names, and are not reflected in any backtest.
• Performance tracking is self-reported. We compute what your stated positions would be worth at public prices. If you didn't buy exactly what you recorded, the numbers won't match your broker.