Build any entry rule from 32 primitives: price and indicators plus GoldenGlass fields nobody else lets you backtest, open interest, top-trader positioning, actual funding, liquidation bursts, order book imbalance, depth and spot premium. Run it on up to 627 perpetuals across four years of history with real fees and funding. Every report ships with the parts other backtesters hide: a random-entry control, a placebo time shift, a train/test split, a parameter landscape and a Bonferroni-adjusted t-statistic. Three free runs after sign-in.
Every popular strategy, run at its textbook defaults on the same universe of perpetuals, real round-trip fees, against a random-entry control with an identical stop, target and hold. Running the numbers
Identical 2% stop / 4% target / 8-hour max hold for every row; the entry rule is the only variable. A strategy "beats random" only if it clears a positive net edge and t at or above 2. Tap any row to load it as a template in the lab above, then widen the universe, add GoldenGlass conditions and sweep a parameter. Full honest backtests of all 30 templates on 100 perps over the last 365 days, with equity curves and train/test splits: strategy results.
Side (long, short, both), timeframe (5m, 15m, 1h), period from 2022 to today, universe (top N by volume, all 627 perps, your own list, a sector, minimum 24h volume, minimum listing age), up to 8 entry conditions with every parameter as its own field, AND/OR logic, hour and weekday windows, trading session, BTC regime, one parameter sweep with your own list of values, stop, target, max hold, trailing stop, ATR stop, partial take-profit, exit hour, exit on opposite signal, fee, slippage mode, funding on or off, max concurrent positions, size and cooldown. The whole spec is also editable as JSON.
Net mean per trade above zero after fees and funding, above the random and placebo controls, the same sign in the test segment and in the second calendar half, and a Newey-West t at or above the Bonferroni threshold for the number of cells in the sweep. Anything positive that fails one of those is labelled noise. Anything net negative is labelled no edge.
Our own continuous market data collection: open interest, top-trader long share, actual funding, liquidations, best bid/ask, Bybit crowd positioning, order book depth, spread and spot flow. GoldenGlass covers a shorter history than the candles; if part of your period has no data the report counts only what exists and lists the gap.
The run goes into a queue and a worker picks it up, usually within seconds. The panel shows the queue position, the stage, the percentage and the elapsed time. If the worker is busy you can close the page: the report is saved to your account and at its public link. A run that ends with an error returns its credit automatically.
No. A report is a research output on past data. It tells you whether a rule would have survived honest controls, nothing about the future. Trade at your own risk.
Most "top strategies" sold in $500 to $2,000 courses lose to a random control after fees. We documented every one with the math and the verdicts, and the handful that survive turn out to be the same old momentum effect wearing a new name.