Signal Backtest — Simulated Results, Costs Charged

These are simulated results, not live performance. Every figure below comes from replaying the signal engine over historical prices. No money was traded and no client account produced these numbers. Simulated results have the benefit of hindsight over the data they are computed on and do not predict future returns. For what the engine has actually done since going live, see the live track record.

Correction (2026-10-22). Earlier versions of this page showed a positive edge. They were produced by a scoring bug: the backtest let trades run past their 30-day holding limit until they hit a target or stop. With the limit enforced, the figures below are materially worse, and we have replaced the old ones rather than keep them up. The figures are for the current engine, whose target was lowered from 2:1 to 1:1 in October 2026 after the 2:1 version lost about two trades for every win. Signals are still generated, published and scored on the site, but we have stopped sending them to Telegram, Discord and email until the live record proves an edge.

Simulated

Backtest results

We publish this because a signal service that shows you only its live record while it is young is telling you very little. The method, the costs and the unflattering numbers are all here.

The engine was replayed over 107 assets across roughly two years of daily candles. Each series was split in half: the first half is in-sample, where any tuning happened, and the second half is out-of-sample, which the tuning never saw. Only the out-of-sample column is evidence, and it is the only one shown in the tables below.

Every trade is charged a realistic round-trip cost — spread, slippage and fees — scaled by asset class and liquidity, from 0.04% on a major currency pair up to 0.5% on a thin altcoin. Backtests that skip costs systematically overstate an edge, and at these trade sizes the cost is a large share of it.

Trades never overlap: one position per asset at a time, opened only after the previous one resolves. A stop and a target inside the same candle is scored a loss, because the candle cannot show which came first and the conservative reading is the honest one. Levels are a 2.5-ATR stop against a 1:1 target, held for up to 30 days.

Overall, out-of-sample

Trades

863

Win rate

53%

Profit factor

1.10

Per trade

+0.63%

A profit factor of 1.10 means the simulated trades that hit their target returned 1.10 times what the trades that hit their stop cost, after trading costs. That is not an edge: this backtest does not show the engine making money. The per-trade figure also counts trades that reached neither level and closed at the holding limit, which is why it can differ in sign from the profit factor; on its own it is not evidence of an edge either.

By playbook

Out-of-sample, net of costs. A playbook is only sent to anyone once its live record proves an edge; every playbook is published on the site and scored so the record stays complete.
Bucket	Trades (out)	Win rate	Profit factor	Per trade
Trend continuation	417	53%	1.01	+0.69%
Breakout	354	55%	1.22	+0.58%
Mean reversion (not delivered)	92	52%	1.06	+0.60%

By asset class

Out-of-sample, net of costs.
Bucket	Trades (out)	Win rate	Profit factor	Per trade
Crypto	320	58%	1.28	+2.37%
Stocks	396	51%	0.88	-0.46%
Forex	147	47%	0.86	-0.20%

What a backtest cannot tell you

Fills are assumed at the published level. A real order may fill worse, or not at all.
The engine is being tested on the same market history that informed its design, and no split fully removes that.
It says nothing about whether the next two years resemble the last two. A strategy measured on a trending period will flatter itself in one.
It is not a forecast, an offer, or advice. Trading carries risk of loss.

Generated 2026-10-22 by npm run backtest -- --json. Regenerating it republishes this page from the same code path the live engine uses.