You saw a setup online — a moving-average cross, a breakout rule, a “buy the dip” recipe with suspiciously round numbers — and the honest voice in your head asked the only question that matters: would this have actually worked? Answering that used to require Python, a data subscription, and a weekend. It now requires neither code nor money. Here is the whole method, tool-agnostic where it can be, specific where it helps.
Step 1 — Turn the idea into exact rules
This is the step that kills most ideas, and it’s supposed to. “Buy strong stocks on pullbacks” is not a strategy; it’s a mood. A testable rule answers, with no judgment calls left: what exactly triggers an entry? What exits — a reverse signal, a stop, a target? How much per position? How many positions at once?
In our labyou do this by describing the idea in plain English — “buy when the 50-day crosses above the 200-day, sell on the reverse cross, 20% per position” — and the AI writes the exact, machine-readable rules and shows you every assumption it made. Whatever tool you use, do not skip the discipline: if you can’t state the rule precisely, you can’t test it, and if you can’t test it, you’re trading a feeling.
Step 2 — Run it on serious history
A backtest over six months tells you almost nothing — you’re seeing one regime, and every strategy looks smart in the regime it was born in. Run daily-bar strategies over many years, through at least one crash and one boring sideways stretch. (Daily-data backtests are free here on every plan — every template, and custom universes up to 10 symbols — and a full run over two decades takes seconds.)
Step 3 — Read the results in the right order
Most people read a backtest backwards. The order that protects you:
- Max drawdown first.This is the price of admission — the stretch where you’d have watched a third of your account evaporate while the strategy insisted it was fine. If you couldn’t have lived through it, the CAGR is irrelevant, because you’d have quit at the bottom.
- Trade count second. Twelve trades in a decade is an anecdote, not a statistic. Be much more skeptical of beautiful results built on few trades.
- Returns last— and always against a benchmark. “Made money” is not the bar; “beat just buying the index, after the extra drawdown and effort” is.
Our reading-results guide goes deeper on every metric, in plain language.
Step 4 — Try to kill it
A good backtest is a hypothesis that survived one attack. Launch more: move the date window five years earlier; swap the universe; make the stop tighter and looser; add realistic costs if your tool doesn’t already. If small changes crater the results, you never had an edge — you had a coincidence with parameters.
Step 5 — Prove it forward
However well the history reads, a backtest remains an exam with the answers printed in the back — we wrote a whole piece on why edges evaporate between the backtest and live. The final step costs nothing but patience: freeze the exact rules and paper-run them forward on data nobody has seen. On this site that’s one click — deploy to the public ledger— and the record it accrues is timestamped, append-only, and visible to anyone you’d ever need to convince, including future-you.
Total cost of the whole method: an afternoon, zero dollars, no code. The setup you saw online either survives it or it doesn’t — and both answers are worth having before a dollar is at risk.