Under the hood
How it works
The goal is not to predict the future. It is to narrow thousands of stocks down to the handful where several independent analyses agree, and then to keep a public record of every single call so you can judge for yourself.
- 1
The whole US market gets scanned
Every morning we pull roughly 7,000 stocks listed on the Nasdaq, NYSE and AMEX. Anything under $5 or too thinly traded is dropped, because an illiquid stock cannot be bought and sold cleanly within a single day.
- 2
A numbers-only shortlist, no AI yet
Each stock gets a score built from session-level signals: gap at the open, unusual volume, typical daily range, and news less than 48 hours old. No model is involved here, only reproducible measurements.
- 3
One hundred finalists, via two routes
The top 80 scores go through, joined by 20 stocks picked for fresh news or unusual activity. That second route stops the system from only ever noticing what is already going up.
- 4
Perplexity researches, three AIs vote
Perplexity first digs up recent news on every finalist. GPT-4.1, Claude and Perplexity then each grade the stock, carrying exactly the same weight. It is an election, not a hierarchy.
- 5
Market conditions move the bar
When the market is jumpy — S&P 500 falling, volatility high, few stocks trending — the publication threshold rises. A stock that has already run up hard with no new information gets penalised.
- 6
Up to three picks, each labelled for what it is worth
At most three stocks are published, each from a different sector. The best candidate is named every session, but it carries a conviction level: high when it cleared every bar, moderate when it came close, low when it is simply the best of a weak field. Saying a call is weak is more useful than hiding it.
- 7
Some mornings we still say nothing
Four things stop us naming a stock at all: too few models answered, the underlying data was unreliable, the stock was too illiquid to trade cleanly, or not a single model expected it to rise. Those are not matters of degree, so there is no honest way to publish through them.
- 8
Graded the same evening
After the close, every published stock is measured from the open to the close of its own session, minus simulated trading costs, and compared to the S&P 500 on that same day. Win or lose, it goes on the record.
What it deliberately does not do
- It never tells you when to sell or how much to put in.
- It knows nothing about you — your situation, your horizon, your tolerance for losing money.
- It guarantees nothing. A good chunk of these picks will lose, and the track record shows exactly which ones did.
The exact thresholds and weights are published on the methodology page, and every past call is in the track record.