PerpForge

The anti-hype backtester

A backtest can show +40.3% and still be a coin flip.

PerpForge tests trading strategies on real perpetual-futures data, then runs the one check most backtests skip: is the edge real, or indistinguishable from luck?

See the public leaderboard →Test your own idea — free →

No signup to look. No card to start.

EMA Cross 12/26

Inconclusive

Did it make money?+40.3% in-sample

How consistently?Sharpe 0.66

Barely. The gains were small next to the swings.

Can we rule out luck?Sharpe 0.66 ± 0.76

No. The 95% range still crosses zero. A longer test window would sharpen the answer, toward real or toward luck.

Could you live through it?−14% worst drop

~2.3 years underwater — and counting

You'd have to sit through that to collect the +40.3%.

Real Binance perpetual-futures · Sep 2020 – Jul 2026source ↗

Every result on PerpForge, decomposed like this. See the leaderboard →

The receipts

Thousands tested. Ranked by what's real, losers and busts included.

No cherry-picking. The same board shows the strategies with a real edge, the ones indistinguishable from luck, and the ones that lost money. Sorted by edge significance: whether the edge is real, not the loudest return.

How to read the board

Ranked by edge significance, not return. A strategy that reliably loses is more informative than one that might just be lucky. We sort by what we can trust.
In-sample only. The top row clears the bar on the data we replayed. Not a forward guarantee.
Luck survives at scale. Across thousands of strategies, a few pass a 95% test by chance, so the #1 spot is the most likely to be one of them. Treat the very top of the board with extra skepticism.
Spawned, not curated. Every strategy comes from fixed indicator grids. Never hand-picked, and never ranked first on too few trades.

One of the rows below lost most of the account. We leave them on the board, ranked next to the winners. Hiding them is how a backtest flatters itself.

Most significant · across all families

#	Strategy	Verdict	Returnin-sample	Worst drawdowndepth · recovery
1	EMA Cross 12/26 SOL·1h·long·858 trades	+0.45Real edge	+1.2%	-11%·24mo
2	EMA Cross 9/21 Heavy lossBTC·1m·short·54454 trades	-13.19No edge	-0.2%	-98%·71mo
3	EMA Cross 9/21 ETH·15m·long·4763 trades	-0.72Inconclusive	+0.0%	-13%·63mo

⚑ in-sample · Binance · deterministic replay · as of Jul 5, 2026

See the full leaderboard →No signup to look.

How it works

Here's the math. And the gaps we haven't closed yet.

Every number on PerpForge is a deterministic replay over real perp candles. Trading fees and liquidation included, the same inputs producing the same result every time. No black box, no cherry-picked window. And we publish what we don't simulate yet, because the gaps are part of the receipt.

What you can backtest

Assets

BTC·ETH·SOL

Intervals

1m·5m·15m·30m·1h·4h·1d

Scope

Full history·Named era·Custom range

Full history = Binance perp candles from Jan 2020 (BTC · ETH) · Sept 2020 (SOL). Named eras cut that into bull / bear stretches.

Every strategy in a run replays the same window, so the leaderboard ranks apples-to-apples.

What we model

The rigor behind every verdict.

Real market data

Actual Binance perp candles. The exact range is stated on every result, and differs by interval.

Trading fees

A per-side fee of 5 bps (0.05%), taken on every entry and every exit. Every round-trip costs something, just like live.

Conservative slippage estimate

A calibrated two-term cost. A fixed 5 bps spread moves every fill price against direction (takers cross the spread). A √-law impact term then adds the extra cost of consuming order-book depth, sized to the candle's realized volatility. A conservative lower bound: real fills may be slightly worse still.

Liquidation

Force-close when an adverse move of 1/leverage wipes the margin. A simplified model with no maintenance-margin buffer yet, so if anything it's slightly kinder than a real exchange. When the margin's gone, the account is wiped. We don't pretend it survives.

Deterministic replay

Same inputs, same result, every time. No randomness, no hidden state. We publish the data source, the window, and the fee behind every number.

Not modeled yet

Each one would make a strategy look worse, not better.

Funding rates

The recurring payment between longs and shorts that tethers a perp to spot price. Real results would be slightly worse.

Full order-book depth

Impact decay and resilience after large orders, cross-asset impact, liquidation-cascade slippage, maker/taker fee tiers, and sub-candle path-dependence. Each one would make results look worse, not better.

Every gap here flatters the result. Funding and tiered fees would drag real returns down, so a strategy trades slightly worse live than it tests here. We tell you anyway. That's the difference between a receipt and a sales pitch.

We're not all-knowing, and we say so. If you find a flaw in how we test significance, or a better way to do it, show us. We'll adopt it, credit you, and publish the change.

Hype can't invite scrutiny. We can.

methodology@perpforge.com

The Strategy Lab

You describe the idea. We test it.

You describe the hypothesis in plain conditions. We handle the simulation.

Entry and exit are built from a condition tree: OHLCV values plus configurable indicators like moving averages, RSI, MACD, and Bollinger Bands. Optional take-profit / stop-loss bracket. Long, short, or both. Sweep the same idea across assets, timeframes, or leverage in one run.

See for yourself

Find out before your money does.

We can't promise your strategy makes money. We can tell you, honestly, whether its edge is real or just luck, before you risk a cent. Seeing the board is free. Testing your first idea is too.

See the public leaderboard →Test your own idea — free →

No signup to look. No card to start.