Entry rule: Bybit crowd long % (op <, pct 40) AND BTC 24h vol percentile (days 30, op <, pctile 30), long, 15m candles, top 100 Binance USDT perpetuals by volume, 2026-09-25 to 2026-10-27. Textbook exits for every template: 2% stop, 4% target, 48 bars max hold, taker fee 0.05% per side, 0.02% slippage, actual funding. Every number below is net of costs and comes with the controls other backtesters hide: a random-entry control with the same exits, a 24h placebo shift, a train/test split and a Bonferroni-adjusted t-statistic.
Net +0.29% per trade, but t(NW) +0.78 below the Bonferroni bar +2.58; sign flips in test or second half. Indistinguishable from luck.
Bybit crowd short-heavy + calm BTC: long fired 107 trades on 100 Binance perps between 2026-09-25 and 2026-10-27 (15m, long). Net result after fees and funding: +0.29% per trade, win rate 40%, profit factor 1.23, max drawdown -2.3%. Random entry with the same stop, target and hold made -0.34%, so the entry rule beats it by 0.62 points, still a different thing from making money; the 24h placebo shift made -0.20%. Newey-West t is 0.78 against a Bonferroni bar of 2.58 for 5 cells, which does not clear the bar; train +0.55% / test -0.25%, halves +0.83% / -0.24%, the sign flips out of sample. Verdict: noise, indistinguishable from luck.
Train is the first 70% of the period by date, test the last 30%. A real edge keeps its sign in test and in the second half.
| segment | n | net % | t | WR |
|---|---|---|---|---|
| train | 72 | +0.55% | 2.48 | 44% |
| test | 35 | -0.25% | -2.64 | 31% |
| first half | 53 | +0.83% | 3.88 | 49% |
| second half | 54 | -0.24% | -2.29 | 31% |
The same rule with the swept parameter at each value. One good cell next to bad ones is curve fitting, not an edge. Green bars clear the Bonferroni bar and beat random.
Open this exact spec in the builder: change the threshold, the stop and target, the universe or the period, add GoldenGlass conditions (open interest, funding, liquidations, order book) and sweep a parameter. The report comes back in seconds with the same controls.
The number here is net of a round-trip taker fee, slippage and actual funding, on 100 coins instead of one cherry-picked chart, with a fixed stop, target and hold. Most textbook examples skip all of that.
The engine fires the same number of entries per coin at random bars and applies the same 2% stop, 4% target and 48-bar hold. If the strategy cannot beat that, the entry rule adds nothing.
Yes. The button above opens this exact spec in Reality Check Lab. After sign-in you get 3 free runs on up to 100 coins and 366 days; larger runs use credits from $7.
Same universe, period, exits and costs for every row. Sorted from the worst net result.
Open any template in the builder, change what you believe is wrong with it and run it against the same controls. 3 free runs after sign-in.