The math check on the home page is arithmetic on the numbers you type. It uses no market data, it runs in your browser, and it never judges a person. It shows what a promise would mean if it were true, and how easily chance could produce the record shown.
An audit, and every case file, is a test of a publicly posted claim on historical market data. It takes time and it is done by us, not by the page.
Payoff is the average win divided by the average loss. Cost is the spread, fees and slippage of one trade, written as a share of what is risked on that trade.
Break-even win rate = (1 + cost) ÷ (1 + payoff).
Edge per trade = win rate × payoff − (1 − win rate) − cost. It is written in units of risk, shown as R.
One year. Trades in a year = trades per week × 50. Each win multiplies the account by 1 + risk × (payoff − cost). Each loss multiplies it by 1 − risk × (1 + cost). The receipt shows the typical outcome with profits reinvested, and on the next line the plain sum without reinvesting.
The same trades with no edge, the second line on the chart, uses the same formula with a win rate of 1 ÷ (1 + payoff), costs included.
For scale. S&P 500 with dividends, 10.4% a year, and Berkshire Hathaway, 19.9% a year, are the compounded annual gains for 1965 to 2024 from Berkshire's 2024 annual letter. Best fund on record is the Medallion fund's average yearly return before fees, 66%, for 1988 to 2018, from Gregory Zuckerman's book The Man Who Solved the Market.
Who else would show this record. Two exact probabilities of seeing at least the shown number of winners in the shown number of trades. First, for a trader with no edge at all, who wins with probability 1 ÷ (1 + payoff); costs are not part of this number. Second, for a trader who only breaks even after costs, who wins at the break-even rate, (1 + cost) ÷ (1 + payoff). With "best of N runs", the chance that at least one of N runs shows it is 1 − (1 − p)N.
Trades to prove a profit is the smallest number of trades at which the claimed win rate could be told apart from break-even with 95% confidence (one-sided, normal approximation). It is an approximation, and it is never shown as fewer than 30.
The stamp follows fixed rules. Loses on paper: the edge after costs is zero or negative, or the account shrinks over the year at the stated risk per trade. Otherwise the stamp says how believable the yearly return is. Realistic: at most 19.9% a year. Unlikely: at most 66%. Doubtful: at most 1,000%. No record: above that.
The receipt is an illustration of this arithmetic. It is not a financial document and it cannot tell you whether the numbers you typed are true.
We test only claims that were posted in public. The rules are coded as they were stated. Before a new audit runs, its rules and its pass mark are written down and time-stamped.
The five tests are: trading costs on every trade, a comparison with simply holding the market over the same days, a comparison with random entries in the same kind of instrument, a second half of the data that the rules never saw, and a higher pass mark when many variations were tried. Each case file says which of these it had and where it falls short.
For every verdict we keep the source of the claim, the rules as coded, the code and the results.
Not supported. The promise did not show up. We tested the rules the way they were described, and the result they promised was not there.
No edge found. No better than luck. It did about as well as picking at random, or the gain was gone after the trading costs.
Unproven. Not enough to say yes. Something showed up, but it is too weak or too shaky to count on yet.
Held up. It passed every test we set before we started. That is rare, and it is still not a promise about tomorrow.
A label describes a claim and a test. It does not describe a person, and it is not a statement about anyone's honesty.
We never tell you what to buy or sell. We do not manage money. Nothing here is tailored to your personal situation, and nothing here promises a future result.