How we run backtests

The value of a backtest is not a pretty equity curve but the fact that anyone can reproduce it and get the same numbers. Here is what our engine runs on, how, and where the honesty line is.

Updated 2026-07-29

Data

Every reproducible run uses one sample: EUR/USD (дневные бары), 400 bars over 2025-06-16 — 2026-07-20. Source — Binance Spot REST API (api.binance.com/api/v3/klines).

The honest caveat about the data

This is a Binance spot EUR/USDT proxy series, not a regulated forex broker feed. For daily momentum signals the gap versus a broker EUR/USD is negligible; for scalping it is critical, which is why we do not test intraday hypotheses.

The sample files are open for download so you can run the logic yourself and check the metrics.

CSV: /data/eurusd-daily.csv (400 rows, OHLCV)
JSON: /data/eurusd-daily.json (with source and period metadata)

Execution costs

A cost-free backtest is an ad, not research. Every run factors in real costs:

Spread: 1.0 pip (0.0001) per round-trip
Commission: folded into the spread
Slippage: fills at bar close, no look-ahead

Overfitting guard

The hypothesis is stated before the result is seen; overfitting is checked on the out-of-sample tail of the window.

We report drawdown, trade count and time-in-market next to return, so it is visible the result is not three lucky trades.

What we do not do

We do not compute a result for an instrument absent from our sample — we describe the limitation instead of inventing numbers.
We do not test intraday, arbitrage, news or pairs hypotheses: they need data the portal does not have.
We do not publish metrics without a methodology to reproduce them.

Principle

Blank space where a hypothesis is untestable is more honest than a fabricated study. We choose the former.