01
Quality metrics
Sharpe, Sortino, Calmar, profit factor, drawdown and win rate, every one computed after costs.
Reject if any of the six is missing, or the edge dies once costs are real.
54 of 56 strategies never reached live money.
Six statistical gates sit between a backtest and a dollar of real money. Almost nothing gets through, and that is the point.
96%
Rejection rate
Fifty-four studies were run and never funded. Two carry money today. Anyone can produce a good backtest by running enough of them, which is exactly why the gates below discount a result for how many attempts it took.
Candidate ledger56 studied
Of 56 strategy studies run, 54 were never deployed and 2 are running today.
54 never deployed
2 carrying money
02
Walk-forward analysis
Fit on one window, trade the next one blind, roll forward, repeat.
Reject if fewer than five out-of-sample windows, or a negative worst window.
03
Deflated Sharpe
The headline Sharpe is marked down for how many parameter sets were tried.
Reject if it falls below 0.95, which is not separable from a lucky search.
04
Overfitting probability
How often does a top-half in-sample rank land in the bottom half out of sample?
Reject if the probability reaches 0.50, a coin flip wearing a backtest.
05
Leak audit
Every signal is shifted one bar forward and the whole test is run again.
Reject if the edge survives the shift, because it was reading the future.
06
Registered on the record
Filed with its full parameter sweep, so anyone can rebuild the result.
Reject if the result cannot be reproduced from a registered sweep.
The rejections are the point. A shop that shows only its survivors has told you nothing about how it decides.
See what got through