Break-even, and what a backtest actually reports

Most tools in this category report a survival rate and stop. A survival rate on its own cannot be read, and the reason is the same asymmetry that makes ladders feel safer than they are.

Why a survival rate says nothing by itself

In a ladder, a win and a loss are not the same size. A sequence that completes wins the target profit — capped, by design. A sequence that runs past its last rung loses everything staked across every rung, which is many multiples of that target.

So the rate at which sequences have to survive, simply to break even, is not somewhere near half. It is high, and it climbs with the depth of the ladder. A strategy can complete the overwhelming majority of its sequences and still lose money steadily, because the rare failures are each worth many successes.

The number this site shows in red

On the current data, over three seasons and 180 ladders with every rung priced at the odds captured before its own kick-off, the break-even rate the strategy has to clear is 93.7%. The rate it achieved is 92.2%.

That is 1.5 percentage points short, and it is not a near miss. It is a losing strategy, and it is reported as one — in red, next to the rate it needed to beat, rather than as a 92.2% success figure with the requirement left off. Marketing that contradicts the product is worse than dull, and the product says this out loud.

The three things a backtest should print beside it

Ruin probability. Not a footnote: the share of sequences that lost the entire stake. This is the number a survival rate is hiding.

The full return distribution. Every sequence binned, so a bimodal outcome cannot hide behind a median. A strategy that mostly returns a little and occasionally returns catastrophically is not described by its average.

Closing prices only. Each rung priced from the odds snapshot captured before its own kick-off, so the backtest cannot quietly award itself a price that was not available at the time.

What to do with a losing backtest

Not hide it. A backtester that only ever produces encouraging numbers is a backtester with a bug or a sales function, and either way it is not measuring anything.

The useful response is the boring one: change the parameters and measure again, accept that most configurations do not clear the bar, and treat the ones that appear to as suspects rather than findings. That is what the sandbox is for, and it will tell you when the strategy loses.