Backtest
Run a strategy across symbols and years of candles, and read what it actually did.
A backtest answers one question: if these rules had been followed exactly, without hesitation or improvisation, what would have happened. It is the only cheap way to find out that an idea which feels obvious on a chart loses money once the costs are charged.
The hard part is not running it. The hard part is running it honestly, because almost every way of doing this accidentally cheats — filling at the close of the bar that triggered the signal, ignoring fees, assuming the target was hit before the stop on a bar that contained both, or quietly tuning the rules on the same data used to judge them.
So this is being built to charge costs on every fill, to fill at the next bar open rather than the signal bar, to assume the pessimistic order when a bar contains stop and target, and to state its sample size and out-of-sample split next to every number it reports.
What it will do
Return, drawdown, win rate, expectancy and trade count — reported together, because any one of them alone can be made to look good.
Measure on a window the rules were not chosen on. A result that survives in-sample and collapses out-of-sample is the single most useful thing a backtest can tell you.
Vary one setting and watch the result. An edge that exists at exactly one value and nowhere near it is a coincidence you found, not an edge.
A rule that works on one asset and fails on four similar ones has told you something. Per-symbol breakdowns sit beside the combined figure.
Not live yet. When it is, remember that a profitable backtest is a description of the past — it says what the rules would have done, never what they will do.