● instant idea backtest — no look-ahead, explained
Type a trade idea. Get a backtest in one second.
Direction + entry, stop, target → the agent compiles it to an event-driven plan, walks the path bar-by-bar (fill at next open, then stop/target/horizon — no look-ahead), and hands you the trade, a tearsheet (R-multiple, expectancy, max adverse excursion) and a ready-to-arm alert. Replayed on real price history only.
entrystoptargetreal price history
NO FILL · no_fill
Hit rate0%
Avg R0.00
Expectancy0.00R
Profit factor0.00
⚑ Derived alert: Alert when NVDA rises through 180 (long entry).
on this path the entry at $180.00 was never reached, so the idea produced no trade. The same levels still arm a live "crosses $180.00" alert for when it is.
Below, Atlas turns a published idea into a runnable rule and an alert, then explains every number — what it is, what it means, and what it suggests. Evaluated on real price bars — your decisions, your risk.
The idea, as a rule a computer can run
Risk per share is $10.00 (entry→stop) and reward is $20.00 (entry→target) — a planned reward:risk of 2:1.
A published idea is just levels until you make it mechanical. We compiled it into: BUY NVDA when price rises through $180.00 (the entry), with a stop at $170.00 and a target at $200.00.
→ The math is favorable before any trade: you risk 1 to make 2+.
No look-ahead: we fill at the NEXT bar's open
In this run the entry level was never crossed, so no trade was taken.
A signal that appears on a bar can only be acted on AFTER that bar closes. So when price reaches the entry level, the fill is taken at the open of the following bar — never inside the same bar. This is what stops a backtest from cheating.
→ If two backtests disagree, the one that fills same-bar is usually the one fooling itself.
Exit priority: stop, then target, then time
No exit to score — there was no fill.
On every bar after the fill we check the STOP first, then the target, then whether we have run out of time (the horizon). Checking the stop first means a bar that touches both your stop and your target on the same candle is scored as a LOSS — the honest, conservative assumption when you cannot see the intrabar path.
→ —
Glossary
- R (risk unit)
- One R is the dollars you risk per share — the distance from entry to stop. Every result is measured in R so trades of any size compare apples-to-apples.
- Look-ahead bias
- Using information a real trader could not have had yet (e.g. acting inside the bar that produced the signal). It inflates results and is the #1 way backtests lie.
- MAE (max adverse excursion)
- The worst drawdown the trade endured before it exited, in R. High MAE on winners means you took a lot of heat — a sign the stop could be tightened.